Anda di halaman 1dari 352




Founded by Wilfrid S. Sellars and Keith Lehrer

Keith Lehrer, University of Arizona, Tucson

Associate Editor
Stewart Cohen, Arizona State University, Tempe

Board of Consulting Editors

Lynne Rudder Baker, University of Massachusetts at Amherst
Radu Bogdan, Tulane University, New Orleans
Marian David, University of Notre Dame
Allan Gibbard, University of Michigan
Denise Meyerson, Macquarie University
François Recanati, Institut Jean-Nicod, EHESS, Paris
Stuart Silvers, Clemson University
Barry Smith, State University of New York at Buffalo
Nicholas D. Smith, Lewis & Clark College

The titles published in this series are listed at the end of this volume.

Edited by

University of Constance, Germany


A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN 1-4020-1605-0

Published by Kluwer Academic Publishers,

P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Sold and distributed in North, Central and South America

by Kluwer Academic Publishers,
101 Philip Drive, Norwell, MA 02061, U.S.A.

In all other countries, sold and distributed

by Kluwer Academic Publishers,
P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

Cover art: MetaMe, Keith Lehrer, 2002, Oil painting, 16˝ × 20˝

Printed on acid-free paper

All Rights Reserved

© 2003 Kluwer Academic Publishers
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
Table of Contents

PREF ACE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. VB

INTRODUCTION ............................................ .
Erik J. Olsson: "The Epistemology of Keith Lehrer"

EXTERNALISMVS. INTERNALISM ............................ 21

Chapter 1,
Ernest Sosa: "Epistemology: Does it Depend on Independence?" . . . . . . .. 23

Chapter 2
John Greco: "Why Not Reliabilism?" ............................. 31

Chapter 3
Jonathan L. Kvanvig: "Justification and Proper Basing" ............... 43

Chapter 4
Todd Stewart: "Lehrer on Knowledge and Causation" ................ 63

Chapter 5
Volker Halbach: "Can we Grasp Consistency?" ..................... 75

COHERENCE AND PERSONAL JUSTIFICATION. . . . . . . . . . . . . . . .. 89

Chapter 6
Glenn Ross: "Reasonable Acceptance and the Lottery Paradox:
The Case for a More Credulous Consistency" ....................... 91

Chapter 7
Charles B. Cross: "Relational Coherence and Cumulative Reasoning" . .. 109

Chapter 8
Wolfgang Spohn: "Lehrer Meets Ranking Theory" .................. 129

Chapter 9
Carl G. Wagner: "Two Dogmas ofProbabilism" .................... 143

TRUSTWORTHINESS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 153

Chapter 10
James Van Cleve: "Lehrer, Reid, and the First of All Principles" ....... 155

Chapter 11
G. J. Mattey: "Self-Trust and the Reasonableness of Acceptance" ...... 173

Chapter 12
Richard N. Manning: "The Dialectic Illusion ofa Vicious Bootstrap" ... 195


Chapter 13
Hans Rott: "Lehrer's Dynamic Theory of Knowledge" ............... 219

Chapter 14
Gordian Haas: "Some Remarks on the Definition of
Lehrer's Ultrasystem" ......................................... 243

Chapter 15
Jacob Rosenthal: "On Lehrer's Solution to the Gettier Problem" ....... 253

SKEPTICISM .............................................. 261

Chapter 16
John W. Bender: "Skepticism, Justification and the
Trustworthiness Argument" . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 263

Chapter 17
Peter Klein: "Coherence, Knowledge and Skepticism" ............... 281

Chapter 18
David A. Truncellito: "The Ultrasystem and the Conditional Fallacy" ... 299

Chapter 19
Keith Lehrer: "Coherence, Circularity and Consistency:
Lehrer Replies" ............................................. 309

Few have contributed so much to the articulation and defense of internalism

and the coherence theory as Keith Lehrer. Thanks to his persistent efforts, we
are in a better position to appreciate their consequences and assess their
tenability. The authors who have contributed to this book were asked to take a
closer look at Lehrer's epistemology from their own different perspectives.
Many are, of course, critical; that is in the nature of the game. But the reader
will also find several constructive attempts to defend or improve on Lehrer's
theory. In the final essay, Lehrer gives his replies. All articles appear here for
the first time.
I am personally indebted to Keith Lehrer in many different ways. Most
importantly, his Theory of Knowledge was the first book I read on
epistemology, and it made me start thinking seriously about the subject. I
never stopped. In recent years I have had the great pleasure of meeting Keith
and discussing philosophical issues with him on several occasions. I take the
opportunity here to express my gratitude for this and also for his support and
practical advice in connection with this book project. Ann Hickman has done
an excellent job in preparing the book for publication, for which I am
extremely grateful. Thanks also to Christopher von Buelow and Radu Dudau
for their editorial assistance. Finally, my warm thanks goes to all contributing
authors for their dedication and commitment. My own work was financed by
the German Research Council (Deutsche Forschungsgemeinschaft) as a
contribution to the research project Logic in Philosophy (Logik in der
Philosophie). Involved in this research project, among the authors, are also
Wolfgang Spohn, Hans Rott, Volker Halbach and, as associated members,
Gordian Haas and Jacob Rosenthal.

Erik J. Olsson



Erik J. Olsson
University of Constance

Fido sees that there is a bone on the plate, but does Fido know that there
is a bone on the plate? David, a two-year-old, sees that the door to the
refrigerator is open, but does he know that it is open? Examples such as these
prompt very different reactions from philosophers. Some think it is obvious that
Fido and David know, and that they know in the same sense as adult humans do.
Others respond, equally emphatically, that they do not know, at least not in the
same way as adult humans. Philosophers ofthe latter inclination may grant that
someone who believes that Fido knows is allowed to use the term 'know' in any
way he or she wishes and, further, that there might be some point in defining a
concept of knowledge applicable also to Fido; but, they will urge, that concept
will not be of great interest if what we really care about is human knowledge in
its characteristic form.
Keith Lehrer belongs to the second category of epistemologists for whom
the mere possession of correct information is insufficient for human knowledge,
however reliable the source delivering the information may have been. In order
to know one must, in addition, recognize that the information one possesses is
correct. This additional demand for reasons internal to the subject
-characteristic of the position known as internalism-excludes poor old Fido,
and probably also David, neither of whom can plausibly be credited with the
conceptual resources required for such recognition, for "[t]hey lack any
conception of the distinction between veracity and correct information, on the
one hand, and deception and misinformation, on the other" (Lehrer, 2000, p. 11).
Why is it so important to recognize that one's sources are reliable? Why
does it not suffice that they actually are reliable, that the belief was caused in a
reliable way? Lehrer's answer, if! understand him correctly, is that the role of

E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 1-20.

© 2003 Kluwer Academic Publishers.

knowledge in human reasoning is essential to its nature (ibid., p. 6), and one role
of knowledge concerns its employment in reasoning, e.g., in confirming some
hypotheses and refuting others. It is essential to knowledge that it enables us to
"reason about what is true or false, what is real and unreal" and to justify our
knowledge claims "in critical discussion and rational confrontation" (ibid., p.
11). Thus, knowledge, as Lehrer conceives it, is, in its essence, "inextricably
woven into reasoning, justification, confirmation, and refutation" (ibid., p. 6).
An externalist will certainly agree that knowledge plays an important role
in reasoning, but he or she will typically resist the conclusion that this role is
essential to its nature. He or she will concede that it is a good thing to have
reasons for one's beliefs-for instance, in convincing others that we
know-while insisting that having such reasons is not an ingredient in the very
concept of knowledge (see for instance Dretske, 1991). An externalist might
even grant that having reasons is part of the pre-systematic concept of
knowledge, and yet argue that there are good grounds, in this case, to depart
from it in favor of an allegedly more fruitful externalist conception. Several of
the papers in this volume address, directly or indirectly, the
internalism/externalism issue, e.g., the articles by Ernest Sosa, John Greco, and
Volker Halbach. Lehrer's view on justification and causation is discussed in the
papers by Jonathan L. Kvanvig and Todd Stewart. Highly relevant in this
connection is also the article by James Van Cleve.
The main purpose of this introduction is to survey the main ideas in
Lehrer's epistemology, so as to provide the necessary background against which
the other papers in this volume can be more readily appreciated. Another aim is
to point out what might be some difficulties in Lehrer's view. A majority of
these issues are explored in greater detail in the other contributions to this book,
and I have added references to guide the reader to the corresponding places. This
introduction, then, is also intended to serve as a conceptual and argumentative
map of the present book. Unless otherwise indicated, references to Lehrer are to
the 2000 edition of his Theory of Knowledge (abbreviated TK). Theory of
Knowledge, an extended and thoroughly revised version of Lehrer's early book
Knowledge, was published for the first time in 1990. Between the two editions
there are some interesting differences that will play a role later in this


Lehrer subscribes to a traditional post-Gettier analysis of knowledge,

according to which a subject S knows that p if and only if (i) p is true, (ii) S
accepts that p, (iii) S is (personally or subjectively) justified in accepting that p,
and (iv) S is justified in accepting that p in some way that does not depend on

a false statement. The last clause is intended to take care of troublesome Gettier
examples, a topic I will return to later.
The particular interest of Lehrer's theory lies of course in its details.
Taking truth for granted, let us focus first on the condition of acceptance: for S
to know that p, S must accept that p. Why "acceptance" and not "belief', and
what might be the difference between the two? Acceptance, Lehrer writes, is an
attitude defined in terms of some purpose and involves an evaluation of whether
the attitude fulfills the purpose. Moreover the special kind of acceptance
relevant to knowledge is acceptance for the purpose of "attaining truth and
avoiding error with respect to the very thing one accepts" (p. 13): to accept that
p if and only ifp. Belief, on the other hand, is not defined in terms of a purpose.
Belief may happen to serve the purpose of attaining truth and avoiding error but
it is not defined in terms of that, or any other, purpose. We may, to take Lehrer's
example, believe that a loved one is safe because of the comfort of believing
this, and not because of an intrinsic interest in the truth of the matter. Belief,
Lehrer argues, is not the attitude characteristic of genuine knowledge;
acceptance is. Another feature of acceptance that will playa role later is that it
is a functional state, being characterized by the role it plays in thought, inference
and action.
Obviously much hinges on the third condition of personal justification.
In Lehrer's view, such justification amounts to coherence with a background
system. The relevant background system---called the evaluation system
-consists of three parts: the acceptance system, the preference system and the
reasoning system. The acceptance system is the core of the evaluation system
and is defined as the set of states of acceptance of S described by statements of
the form "S accepts that p" attributing to S just those things S accepts at t with
the objective of obtaining truth and avoiding error with respect to the content
accepted, that is, with respect to the content that p (p. 130). In the 1990' sedition
of TK, the background system was equated with the acceptance system.
Suppose, to take an example, that S accepts that Paris is the capital of
France. That might lead one to expect that the statement "Paris is the capital of
France" should be an element of S's acceptance system. But this, as we just saw,
is not how Lehrer defines the notion. Rather, the acceptance system contains the
statement "S accepts that Paris is the capital of France". So, when Lehrer writes
that in personal justification we must start with what we accept (p. 123), this
does not mean that we are allowed to take the truth of "Paris is the capital of
France" and other propositions we accept for granted. It means only that we may
take for granted that we accept those things, i.e., that we take a certain attitude,
that of acceptance, towards those propositions. Lehrer sometimes calls the set
of all propositions p such that "S accepts that p" is in the acceptance system the
content of that system, a practice I will follow here.

The preference system of Sat t over acceptances is defined as the set of

states of preference described by statements of the form "S prefers accepting that
P to accepting that q" attributing to S just those preferences S has at t with the
objective of obtaining truth and avoiding error with respect to the contents of the
acceptances. Finally, the reasoning system of S at t is the set of states of
reasoning described by statements of the form "S reasons from acceptance of the
premises PI' P2' and forth to Pj to acceptance of the conclusion COO with the
objective of obtaining truth and avoiding error with respect to the content of the
What is Lehrer's motivation for introducing the preference and reasoning
systems as part of the evaluation system? As for preferences, he writes the
following in his book Self- Trust (pp. 27-28): "I have said before that a person is
personally justified in accepting something if and only if acceptance of it
coheres with the acceptance system of the person. I now think that will not
suffice, because preferences are also essential to the kind of coherence that
yields justified acceptance." It is also of interest to note that, in Lehrer's view,
the justification of preferences parallels the justification of acceptances: "Thus,
personally justified acceptance, acceptance justified for me, is acceptance that
coheres with an evaluation system including preferences, just as personally
justified preference, preference justified for me, is preference that coheres with
an evaluation system that includes acceptances" (ibid., p. 28). I will return to the
reasoning system in connection with the Gettier problem.
So much for the evaluation system. Justification, we are told, is coherence
with that system. How, then, should we understand coherence? The intuitive
idea is that we can think of all sorts of objections an imaginative critic may raise
to what a person accepts. These objection might be directly incompatible with
what the person accepts or they might, while being compatible with the thing
accepted, threaten to undermine my reliability in making assessments of the kind
in question. For instance, a critic might object to my claim that I see a tree by
suggesting that I am merely hallucinating. That would be an example of the first
sort of objection. As an example of the second sort, we might take a case in
which the critic replies that I cannot tell whether I am hallucinating or not
(Lehrer, 1989, p. 253). Coherence, and personal justification, results when all
objections have been met.
Thus, the process of justifying a claim has the character of a game with
the objections and answers being the different moves the players can make.
Lehrer, fittingly, calls it the justification game. If all the objections raised by the
critic can be met, then the claimant wins the game. If she wins the game, her
original claim coheres with the evaluation system and she is personally or
subjectively justified in accepting her original claim; ifnot, she is not justified
in her acceptance (TK, 1990b, p. 119). Lehrer is careful to point out that the
justification game is only a "heuristic device for understanding the

considerations that make a person justified in accepting something rather than

a psychological model of mental processes" (ibid.).
Leaving heuristic considerations aside, Lehrer's semi-formal definition
ofjustification runs as follows: S is personally justified in acceptingp at t if only
if p coheres with S's evaluation system at t. Further, p coheres with S's
evaluation system at t if and only if all objection to p are answered or neutralized
relative to S's evaluation system at t. This raises the question of how the notion
of an objection should be understood, and what it might mean that an objection
is answered or neutralized relative to an evaluation system. Lehrer defines the
notion of an objection as follows: 0 is an objection to p if and only if it is more
reasonable to accept that p on the assumption that 0 is false than on the
assumption that 0 is true (relative to S's evaluation system and t). An objection
o to p is answered, moreover, if and only if 0 is an objection to p, but it is more
reasonable for S to accept p than to accept 0 (with the appropriate
relativizations). In the 1990 edition of TK, objections were called "competitors"
and answered objection were said to be "beaten".
Before citing Lehrer's definition of neutralization, it might be helpful to
consider an example. Suppose I claim to be seeing a zebra and I am faced with
the objection that I am sleeping and dreaming that I see a zebra. Then I might be
in a position to answer this objection by showing how it follows from my
evaluation system that it is more reasonable that I am actually seeing a zebra
than that I am merely dreaming that I see one. But suppose the critic instead
were to object merely that people sometimes dream that they see zebras. This is
an objection, in Lehrer's technical sense, to my claim that I see a zebra; for it is
less reasonable to accept that I actually see a zebra if people sometimes dream
that they see zebras than if people never dream that they see zebras. And yet it
may be very difficult to answer this objection by showing that it is less
reasonable to accept than my claim that I see a zebra for the simple reason that
it is very reasonable to accept that people sometimes dream they see zebras. In
order to allow the claimant to counter objections of this sort, Lehrer introduces
the notion of neutralization. The idea is that I should be allowed to counter an
objection by pointing out its irrelevance to the issue. In this case, I would be
allowed to reply that the objection that people sometimes dream that they see
zebras is irrelevant because I am not dreaming. Lehrer defines neturalization as
follows: n neutralizes 0 as an objection to p if and only if 0 is an objection to p,
the conjunction of 0 and n is not an objection to p, and it is as reasonable for S
to accept the conjunction of 0 and n as to accept 0 alone (with the appropriate
relativizations). In the example the conjunction that people sometimes dream
they see zebras and that I am not dreaming is not an objection to my seeing a
zebra. Moreover, it is, we grant, as reasonable for me to accept this conjunction
as it is to accept that people sometimes dream they see zebras alone.

It is not evident that this is the correct way of defining neutralization.

Moreover, on pain of triviality, this definition rules out a probabilistic
interpretation of reasonableness. For, disregarding some uninteresting limit
cases, a conjunction is always less probable than its conjuncts. Hence, on a
probabilistic reading of reasonableness a conj unction can never be as reasonable
as one of its conjuncts, which means that no objection can be neutralized. For
more on comparative reasonableness and neutralization, see the contributions to
this volume by Glenn Ross, Charles B. Cross, and Wolfgang Spohn. A related
problem in the theory of probability is explored in Carl G. Wagner's paper. Also
relevant in this connection are the contributions by G. J. Mattey and Hans Rott.
Glenn Ross looks at Lehrer's theory from the point of view of the lottery
paradox. The reader might wish to consult Olsson (1998) for another perspective
on Lehrer's treatment of the paradoxes of justification.
The part of Lehrer's theory that I have outlined so far is intended to
specify under what conditions a person is justified in accepting a proposition
from the point of view of that person's own evaluation system. But it also
suggests a solution to another problem, viz., the problem of specifying rules for
inductive inference. This problem can be stated in the following form: Given a
set of propositions that I accept, what am I entitled to accept in addition?
Obviously I should be allowed to accept everything that follows logically from
what I accept. More interestingly, I might also be entitled to accept some
propositions which, although they do not follow logically from what I accept, are
nevertheless very plausible on that basis. Lehrer's theory suggests the following
solution to this problem: in addition to what I accept, I may accept all
propositions that cohere with what I accept. The problem of inference is not
exactly the same problem as Lehrer's problem of justification, if only because,
for the purposes of inductive inference, we would like to start with what we
accept and not merely with reports about what we accept. There are also
indications that the problem ofjustification, as Lehrer sees it, is one ofjustifying
acceptance based on some sort of input (in the form of reports), whereas the
problem of inductive inference is one of adding things in the absence of input. 1
The fruitfulness of Lehrer' s semi-formal theory of personal justification for the
purposes of inductive inference is explored, in this volume, by Charles B. Cross
and Wolfgang Spohn. See also Hans Rott's contribution.

Where does self-trust or trustworthiness come in? Knowledge, as Lehrer
defines it, requires personal justification with the evaluation system of the
person. But as that system is defined preciously little will be personally justified.
Statements of the form ItS accepts that pIt constitute too meager a basis for

concluding anything interesting about the relative reasonableness of more

substantial claims. They are mere reports of what I accept, and little follows
unless I can somehow conclude that what they report is true or at least plausible.
So how do I get from "I accepts that p" to p itself? Lehrer's answer is to invoke
the principle of trustworthiness. In addition to my other acceptances, I must
accept that I am trustworthy, that is, I must accept the following principle:

(T) I am trustworthy (worthy of my own trust) in what I accept

with the objective of accepting something just in case it is true.

In the 1990 edition of TK, Lehrer employed the principle of trustworthiness as

a principle of detachment, allowing me to detach p from "I accept that p". The
2000 edition is less optimistic in this respect. According to the view stated there,
trustworthiness in combination with "I accept that p" does not allow me to
conclude that p is true but only my being reasonable in accepting that p. The
reasoning leading up to this conclusion-Lehrer's trustworthiness argument-
runs as follows:

(T) I am trustworthy in what 1 accept with the objective of

accepting something just in case it is true.
I accept that p with the objective of accepting that p just in case it
is true.
Therefore, 1 am trustworthy in accepting that p with the objective
of accepting that p just in case it is true.
Therefore, 1 am reasonable in accepting that p with the objective
of accepting that p just in case it is true.

The trustworthiness argument illustrates the alleged explanatory role of

trustworthiness. That I am reasonable in accepting that p may be explained by
my being trustworthy, provided that the premises of the trustworthiness
argument are true. In particular it must be true that I am actually trustworthy,
since a false premise does not explain anything.
Lehrer stresses that the principle (T) should not be seen as saying that I
am always trustworthy. Rather, it is a statement of a capacity or disposition to
be trustworthy and is compatible with my failing now and then. Hence, the
inference from my general trustworthiness to my being reasonable in accepting
p is inductive rather than deductive. Moreover, my trustworthiness is not only
a matter of my past track record in obtaining truth and avoiding error. For "I may
proceed in a manner that is worthy of my trust in what 1 accept but be deceived
through no fault of my own", as would be the case if I were deceived by a
Cartesian demon (p. 139). It is only required that "I am as trustworthy as the

circumstances allow" (p. 140). For more on the intended interpretation of the
principle of trustworthiness, see the first chapter in Lehrer's Self-Trust.
There are corresponding principles of trustworthiness for the other parts
of the evaluations system. Thus I may accept that I am trustworthy in what I
prefer with the objective of preferring to accept somethingjust in case it is true.
By an argument paralleling the trustworthiness argument for acceptance I may
then conclude that I am reasonable in having a given preference. By the same
token, I may accept that I am trustworthy in how I reason, from which I may
conclude that I am reasonable in my reasoning to a given conclusion.
My acceptance of the principle of trustworthiness does not automatically
make my other acceptances justified; but it does make them reasonable and
hence in that way it contributes to their justification. This raises the question of
how to justify trustworthiness itself. Trustworthiness can hardly contribute to the
justification of other acceptances unless it is itself justified or at least
reasonable. Lehrer's basic answer is that trustworthiness applies not only to other
acceptances but also to itself. We recall that, for any proposition p which I
accept for the purpose of obtaining truth and avoiding error, I may reason from
the acceptance of my trustworthiness to the reasonableness of my accepting that
p. As a special case, I may reason from the acceptance of my trustworthiness to
the reasonableness of my accepting that I am trustworthy.
Lehrer has characterized the special role played by self-trust using an
analogy from Thomas Reid: ''just as light, in revealing the illuminated object,
at the same time reveals itself, so the principle, in rendering the acceptance of
other things reasonable, at the same time renders the acceptance of itself
reasonable" (p. 143). Reid's epistemology has had a profound influence on
Lehrer's theorizing about self-trust. On Lehrer's interpretation, one of Reid's
principles of common sense is applicable to all other principles including itself.
See Lehrer (1989) and, for a discussion of Lehrer's Reid interpretation, James
Van Cleve's article in the present volume.
As in the case of other acceptances, the trustworthiness argument falls
short of establishing that I am justified in my acceptance of (1). It only
establishes that I am reasonable in my acceptance (provided that the premises
are true). For me to be justified in my acceptance of (1), all objections to that
claim have to be answered or neutralized. In order to answer or neutralize such
objections I will need to refer to other things I accept. Hence, even in the case
of (1), background information is required in order for me to be personally
justified in accepting it.
The principle (1) does not justify itselfbut depends for its justification on
the background system of other things we accept. Therefore it would be
incorrect to call it a basic belief in the foundationalist sense. Lehrer has
suggested that (1) is more like a keystone in an arch. Without the keystone, the

arch would collapse; at the same time the keystone is supported by the other
stones in the arch.
Is the circularity involved in the argument from my acceptance of my
trustworthiness to the reasonableness of that acceptance a vicious one? Lehrer
has argued that it is not vicious by pointing out that his intention is not to use (T)
as a premise to prove something to a skeptic, but rather to use it for explanatory
purposes, the claim being that the principle of trustworthiness can be used for
explaining why it is reasonable for us to accept what we accept. Indeed, the shift
from justification to explanation makes it less obvious that the circle is vicious.
Lehrer even thinks that circularity in this case may be a virtue rather than a vice.
It is, he notices, preferable to leave as little as possible unexplained. Hence, an
explanation that does not only explain why other acceptances are reasonable but
also why its own acceptance is reasonable is better in this respect that an
explanation which accomplishes the former but not the latter.
Several contributors to this volume express dissatisfaction with Lehrer's
discussion of trustworthiness. As some point out, what Lehrer says about the
possible explanatory merits of self-trust seems irrelevant to the issue of
justification. Lehrer's view on the matter is examined in the contributions by
James Van Cleve, G. J. Mattey and Richard N. Manning. The papers by John W.
Bender, John Greco and Peter Klein contain related material. Manning, for
instance, contends that difficulties arise because of Lehrer's presupposition that
the principle of trustworthiness is something that needs to be argued for in the
first place. In Manning's view, by contrast, self-trust is a transcendental
condition on the possibility of our epistemic practice.



What has been said so far raises the following question: In what sense, if
any, is Lehrer's theory a coherence theory, as Lehrer claims it is (at least in
part)? If one takes it as essential that such a theory make use of a concept of
systematic or global coherence, then Lehrer's theory is clearly not a coherence
theory. For in Lehrer's view, "[c]oherence ... is not a global feature of the
system" (1997, p. 31). Rather, what he calls coherence, as we have seen, is a
relation between an evaluation system and a proposition. This relation,
moreover, "does not depend on global features of the system" (ibid.). This
notwithstanding, Lehrer has said that the content of the acceptance system
should be logically consistent, thus referring to the global feature of consistency
(1991, p. 131). Lehrer has also addressed the issue of consistency in his paper
"Reason and Consistency", reprinted as Chapter 6 in Metamind. The role of

consistency in an internalist epistemology is explored in Volker Halbach's

contribution to this volume.
What reasons, then, are there for calling the relation of meeting objections
to a given claim relative to an evaluation system a relation of coherence? As I
understand Lehrer, his answer is that it is a relation of "fitting together with",
rather than, say, a relation of "being inferable from". He writes, in the 1990
edition of TK: "[i]f it is more reasonable for me to accept one of [several]
conflicting claims than the other on the basis of my acceptance system, then that
claim fits better or coheres better with my acceptance system" (p. 116). He also
contends that "[a] belief may be completely justified for a person because of
some relation ofthe belief to a system to which it belongs, the way it coherence
with the system, just as a nose may be beautiful because of some relation of the
nose to a face, the way it fits with the face" (p. 88). Lehrer is here claiming that
a statement's cohering with a system is analogous with a nose's fitting with a
However, as I have argued elsewhere (Olsson, 1999), this analogy is
incompatible with Lehrer's contention that coherence does not depend on global
features of a system. For when we say that a nose fits with a face, we mean that
combining the two yields a beautiful overall result, so that the nose fits with the
face in virtue of the underlying global property of beauty. If cohering is
analogous to fitting, as Lehrer proposes, then a statement coheres with a system
if combining the two yields a coherent overall result, so that the statement fits
with the system in virtue ofthe underlying global property of coherence. This,
again, clashes with Lehrer's declaration that coherence does not depend on
global features ofthe system.
So Lehrer's relation of coherence with an evaluation system has little to
do with coherence properly so called, being more akin to inference.
Nevertheless, Lehrer's more recent ideas about trustworthiness are arguably
sufficient to turn his theory into a coherence theory after all. As we have
observed, there is a circularity involved in the argument from my acceptance of
my trustworthiness to the reasonableness of that very acceptance. And a salient
feature of a coherence theory is presumably that it licenses circular
reasoning-at least this is one popular way of characterizing such a theory.


Externalist theories allow for a modest psychological makeup of the

knower. Internalist theories, on the other hand, are typically much more
demanding in this respect, insisting that the knowing subject be capable of
various higher level cognitive states. Lehrer's internal ism, as we have seen, is
no exception to this general rule. According to Lehrer, I am not in a position to

know, say, that I see a tree over there, unless I also accept my trustworthiness in
relevant matters. But is it psychologically realistic to suppose that acceptances
about other acceptances-in particular about their trustworthiness-are
necessary for knowledge? It seems that people often make perceptual claims
without ever having thought about the trustworthiness of their perceptual
faculties. Are we then forced to say that they do not know what they claim to
While Lehrer concedes that some "unrealistic theory of belief maintaining
that all beliefs are occurrent states may yield the consequence that we lack such
beliefs [about trustworthiness]" (p. 202), he contends that his own theory of
acceptance avoids this conclusion. For according to the latter the mental state of
acceptance is a functional state, one that plays a role in thought, inference and
action. As a consequence, I may be said to accept that p at a time t even if my
acceptance ofp is not something I am contemplating at t. What is essential to my
acceptance of p is how I think, infer and act. This holds, in particular, for my
acceptance of my trustworthiness, which need not be present to my mind either.
As Lehrer puts it, "[ w]e think, infer, and act, in a way manifesting our trust in
what we accept" (ibid.). He concludes that "it is appropriate and not at all
unrealistic to suppose that, in addition to the other things we accept, we accept
our own trustworthiness and the reliability of it as well" (ibid.).
But does this clever defense against charges of lack of psychological
realism really work? Let us first ask this question: What would be the difference
between the functional role of my acceptingp only and the functional role of my
accepting, in addition, my trustworthiness in matters concerningp? Suppose that
I accept that p without accepting my trustworthiness in matters concerning p.
This means that I am inclined to think, infer and act in a certain way. More
precisely, my acceptance of p amounts to my being "inclined to assent to it
automatically, to draw inferences based on the assumption of it, and, in general,
to act as though it were true" (Lehrer, 1989, p. 270). For instance, suppose that
I am planning to take a walk. My acceptance that it is raining will then manifest
itself in my acting as though it were true, e.g., in my being inclined to bring my
umbrella. My acceptance that it is raining will also manifest itself in my being
inclined to infer, and thus to accept, further propositions as well, e.g., that the
ground will be wet. This new acceptance in turn will make me inclined to put on
my boots rather than my less sturdy Italian shoes, and so on.
Now add to my acceptance that p my further acceptance that I am
trustworthy in matters concerning p. My suspicion is that this addition
contributes nothing to how I will think, infer and act beyond what was already
part of my acceptance of p. To continue the example, my acceptance that it is
raining will manifest itself in my being inclined to bring my umbrella and infer
that the ground will be wet, and so on. Add to this my acceptance of my
trustworthiness in telling whether it is raining or not. This addition has no

further effect on my dispositions to think, infer and act. I will still be inclined to
bring my umbrella and to infer that the ground will be wet, and no new
dispositions to think, infer and act seem to come forth. From the point of view
of how I think, infer and act, there is no difference between, on the one hand, my
accepting that p and, on the other hand, my accepting that p and, in addition, that
I am trustworthy regarding p. Rather, these two descriptions-"my accepting
that p" and "my accepting that p and, in addition, that I am trustworthy regarding
p"-are merely two different characterizations of one and the same functional
state, namely, that in which I accept that p.
The upshot of all this is to confront Lehrer with the following dilemma.
Without a functional theory of acceptance, his theory is indeed vulnerable to
charges of lack of psychological realism. But the same functional theory which
solves this problem so neatly has the unwelcome effect of trivia Ii zing the very
idea of trustwort hi ness, reducing trustworthy acceptance to mere acceptance. A
related objection is raised in John Greco's contribution to this volume.



In a classical paper, Edmund Gettier gave a counter example to defining

knowledge as justified true belief. The variation discussed by Lehrer runs as
follows. A teacher wonders whether any of her students owns a Ferrari. She has
strong evidence that one student, Mr. Nogot, owns one. Mr. Nogot says he owns
a Ferrari, he drives one and has all the papers proving that he owns one, and so
on. The teacher has no reason to think that there be other Ferrari owners among
her students. From her evidence she concludes that Mr. Nogot owns a Ferrari,
from which she draws the further conclusion that someone in her class owns a
Ferrari. Now, as a matter of fact, Mr. Nogot has lied about the Ferrari. He does
not own one and the evidence he has presented was fabricated for the purposes
of improving his social status. This notwithstanding, there is another student,
Mr. Havit, who does own a Ferrari, though the teacher has no evidence of this.
Hence, the teacher's belief that someone in her class owns a Ferrari is true after
all. Moreover, being based on good evidence, her belief is also justified. Yet it
seems intuitively incorrect to say that the teacher knows that someone in her
class owns a Ferrari. Rather, she was just lucky that there was a real Ferrari
owner in her class.
The teacher's justification for believing that someone in her class owns
a Ferrari depends on the false premise that Mr. N ogot owns a Ferrari. Hence, the
obvious reaction is to add a fourth requirement to the definition of knowledge
to the effect that the subject's knowledge claim not be dependent on a false

premise. However, as Lehrer notices, one can construct a similar example in

which no false premise is in play. For the teacher could conceivably conclude
directly from the evidence (Mr. Nogot is a student of my class, he has told me
that he owns a Ferrari, and so on) that someone in her class owns a Ferrari
without first concluding that Mr. Nogot does. And we tend to think that she
would not know that there is a Ferrari owner in her class in this case either.
In the 1990 edition of TK, Lehrer suggested an elegant general approach
to Gettier type cases. The idea was to consider not only the subject's actual
acceptance system but also acceptance systems that can be obtained from that
system by removing any false statement or replacing any such statement by its
negation. The collection of all systems obtained in this manner he called the
ultrasystem of the subject. Lehrer then required that the subject's claim, in order
to count as knowledge, be justified relative to all elements of the ultrasystem, in
which case the subject's justification was said to be undefeated.
It is obvious that this proposal takes care ofthe original Ferrari example.
For one element of the ultrasystem is obtained by replacing the claim that Mr.
Nogot owns a Ferrari by its negation, so that the teacher is no longer justified in
concluding that someone owns a Ferrari. Let us also examine how the proposal
applies to the more difficult example in which the teacher never accepted that
Mr. Nogot owns a Ferrari in the first place, but concluded directly from the
evidence that someone in her class owns one. Lehrer argues that in this case the
teacher would nevertheless subscribe to the conditional "If the evidence I have
that Nogot owns a Ferrari is true, then Nogot owns a Ferrari". We may now
construct an element of the ultrasystem by replacing this conditional statement
by its negation, i.e., by "The evidence I have that Nogot owns a Ferrari is true
but Nogot does not own a Ferrari". Clearly, the teacher is not justified in
accepting that someone in her class owns a Ferrari on the basis ofthe member
of the ultrasystem thus constructed, and therefore she does not know what she
claims to know.
One problem with this intriguing proposal has to do with how to eliminate
and replace statements while retaining logical integrity. Suppose that p is to be
eliminated (or more exactly "I accept that pOl) and that p is implied by q which
the subject also accepts. Then, as Lehrer points out (1990b, p. 141), it is not
sufficient just to remove p but q has to be removed as well; otherwise p can be
derived back from q. However, as Hans Rott observes in his contribution, and
also in his book from 2001, elimination is unexpectedly complicated. The same
goes for replacement which is quite a subtle type of change in need of more
careful attention. There is an interesting connection here between the 1990
version of Lehrer's theory and the area of philosophical logic called belief
revision theory in which problems of the kind just mentioned has been studied
independently by logicians since the seventies. This connection is explored in
Rott's paper.

Promising as Lehrer's 1990 solution may seem, the 2000 edition of TK

features a substantially different approach to Gettier problems. The new solution
is simpler in that only one alternative system is taken into account-the one
obtained from the present system by removing all false judgments-and no
complicated contractions and replacements are called for. The new solution is
more complex in another respect, though, since the system from which
judgments are deleted is now the more complex evaluation system and not the
simpler acceptance system (which, we recall, is but a part of the evaluation
system). Hence not only false acceptances are removed but also preferences for
accepting something false over something true as well as unsound patterns of
reasoning. The result is called the ultrasystem, a term which has thus been given
a new definition? Finally, for the subject to know that p it is necessary that p be
justified relative to the ultrasystem.
In the preface to the 2000 edition of TK, Lehrer writes that the new
solution to Gettier problems is based on a "simplified" and "improved"
conception of an ultrasystem, claims that are not substantiated in the course of
his book. As we have seen, while the new conception is simpler in some
respects, it is more complex in another respect, being based on a more complex
conception of a background system. Moreover, it is not clear that the new
proposal represents an improvement, since, as I will try to make likely, Lehrer's
new treatment of the hard Ferrari example (in which the teacher or claimant
concluded directly from the evidence that someone owns a Ferrari) is not
entirely convincing.
The idea behind what Lehrer calls the ultra justification game is that the
ultracritic should try to disprove the claimant on the basis of the claimant's
ultrasystem. Lehrer envisages the following ultra justification game to illustrate
the outcome of applying his new proposal to the hard Ferrari example:

Claimant: Someone in my class owns a Ferrari. If I have the

evidence that Nogot owns a Ferrari, that he is student in my class,
that he has told me that he owns a Ferrari, that he has shown me
papers stating that he owns a Ferrari and that he drives a Ferrari,
then Nogot owns a Ferrari. I have all that evidence consisting of
true claims.
Ultracritic: You must eliminate your hypothetical claim that if you
have the evidence that Nogot owns a Ferrari, that he is student in
your class, that he has told you that he owns a Ferrari, that he has
shown you papers stating that he owns a Ferrari and that he drives
a Ferrari, then he owns a Ferrari! The evidence you have that
Nogot owns a Ferrari is true, but Nogot does not own a Ferrari. It
is not more reasonable for you to accept that someone in your class

owns a Ferrari than that no one does. No one in your class owns a
Ferrari. (My italics.)

Lehrer now maintains that the claimant, on the basis of her ultrasystem, will not
be able to answer or neutralize the ultracritic's objection.
As I understand it, Lehrer's analysis of the hard Ferrari example is
intended to illustrate the claim that "[t]he proper solution to these [Gettier]
problems may be obtained from the ultrajustification game by extending the role
of the critic in the game to include considerations of preferences and reasonings
that supplement the acceptance system in the evaluation system of the claimant"
(p. 160). Clearly, this particular example must be taken to illustrate
considerations of reasonings.
Now, the ultracritic may instruct the claimant to give up a piece of
reasoning only if that reasoning is unsound. In the example, she instructs her to
give up her reasoning from the evidence about Mr. Nogot's driving a Ferrari etc.
to his owning a Ferrari. But is that reasoning really unsound? I think not. The
fact that the premises are true but the conclusion false in the example only serves
to disprove the deductive validity of the inference, but this is beside the point,
since the claimant is not committed to its deductive validity but at most to its
inductive validity. As an inductive inference it is sound. For the conclusion that
Mr. Nogot owns a Ferrari is highly probable on the evidence of his driving one,
showing papers stating that he owns one etc.
But perhaps Lehrer did not after all intend his new analysis of the hard
Ferrari example to illustrate the elimination of reasonings from the evaluation
system. Perhaps we should think of the hypothetical linking the evidence about
Nogotto his owning a Ferrari not as belonging to the reasoning system butto the
acceptance system (as in the 1990 edition of TK). But in that case it is unclear
what role the reasoning system plays in Lehrer's analysis of Gettier cases. Does
it perform any useful function at all?
Hence, it is not evident that Lehrer's new approach to the Gettier problem
amounts to an improvement in comparison to the old. Of course, some logicians
that were attracted to the 1990 version of Lehrer's theory primarily because of
its interesting relationship to belief revision theory will be quite happy with this
observation. Yet, Lehrer might be able to derive some support for his new
approach from Gordian Haas's article in this volume, in which Haas presents a
formal argument to the effect that the complicated elimination and replacement
operations occurring in the older approach are actually superfluous. Jacob
Rosenthal argues, in his contribution, that both solutions presented by Lehrer are
unsatisfactory. Undefeated justification and the Gettier problem are also
discussed in the papers by John Greco, Peter Klein and Wolfgang Spohn.


What is Lehrer's reply to skepticism? In answering this question I will try
to distillate two different and, as I hope to show, conflicting tendencies in
Lehrer's work. According to Lehrer, the sophisticated common sense
philosopher, there is no need to bother with radical skepticism, and this is so for
purely methodological reasons. We can safely assume that we know most of the
common sense things we think we know, and we do not have to take various
skeptical hypotheses into account. Lehrer, the moderate skeptic, is of a different
opinion. According to him, we have knowledge only if we are not systematically
deceived. If we are deceived, e.g. by a Cartesian demon, then we do not know
anything at all.
Let us start with Lehrer, the common sense philosopher. Lehrer addresses
skepticism already in the first chapter of TK in which he rejects Cartesian
epistemology and its method of pretending to total ignorance in the hope of
arriving at something completely certain (p. 2). He rejects it in favor of what he
calls critical epistemology which is said to be "neither skeptical nor
metaphysical" (ibid.), beginning with "common sense and scientific assumption
about what is real and what is known" (p. 3). These convictions, he adds,
"constitute our data, perhaps even conflicting data if common sense and science
conflict" (ibid.). For example, the critical epistemologist takes for granted the
premise that we have knowledge of the internal world of our ideas from
consciousness and of the external world of matter from observation (p. 4). The
object of philosophical inquiry in general, and critical epistemology in
particular, is to account for the data in a way that is "systematic and critical" (p.
3). That it is critical means that "[s]ometimes we explain the data and sometimes
we explain the data away" (p. 4). The best systematic account of the data may
be one which, "in a few instances" (ibid.), does not classify as knowledge what
was classified as such by common sense.
For instance, such data would include my claims that I know that I am
sitting before my computer now, that Paris is the capital of France, that George
W. Bush is the president of the United States, and so on. It would be the job of
the critical epistemologist to find what is common to all, or most, of these
presumed instances of knowledge, e.g., thatthey have been derived from reliable
sources, or are examples of justified true beliefs, and so on.
A few pages later, Lehrer clarifies the task of the critical epistemologist
further by saying that she should aim at explicating concepts in the sense of
RudolfCarnap, i.e., by generating new philosophically and scientifically useful
concepts. Carnap required of an explication that the explicatum (the concept
resulting from the explication) be (1) similar to the explicandum (the original
pre-systematic concept that is to be explicated), (2) exact, (3) fruitful and (4)
simple. More precisely, the explicatum should be similar to the explicandum in

such a way that the explicatum can be used in most cases in which the
explicandum has so far been used, but the similarity need not be very close. The
explicatum, besides being exact, should also be fruitful in the sense of being
useful for formulating universal statements (empirical laws and logical
theorems). Finally, the explicatum should be as simple as the other more
important desiderata permit.
Lehrer's characterization of critical epistemology conveys the impression
that he is essentially a common sense philosopher in the tradition of G. E.
Moore. Like G. E. Moore, he repudiates the idea that epistemology should
engage in radical doubt concerning common sense knowledge claims. Yet being
based on a general view as to the nature of philosophical and scientific
definitions, Lehrer's approach is more sophisticated than Moore's position,
which, lacking a more solid methodological foundation, amounts to ruling out
skepticism on the sole grounds that it conflicts with common sense.
It is not difficult to appreciate that from the point of view of Lehrer's
critical epistemology, even in the sophisticated form referring to Carnap's notion
of explication, radical skepticism can be dismissed for methodological reasons
alone. For a radical skeptic will want to define knowledge so that most or all of
our common sense knowledge claims turn out false, e.g., by insisting that
anything worthy of the name must be such that there is no logical possibility of
error. Hence, any such attempt will score very poorly as regards Carnap' s
criterion of similarity to the explicandum. To compensate for this, it would have
to do extremely well in other respects, i.e., regarding exactness, fruitfulness and
simplicity. But even ifit can be given an exact and simple definition, a skeptical
conception of knowledge will be a complete failure with respect to the important
desideratum of fruitfulness: a concept under which nothing, or very little, falls
will be utterly useless for articulating laws and theories. We would not accept
a definition of water that makes nothing qualify as water; and we should be
equally discontent with a definition of knowledge that makes nothing qualify as
such. Almost any other proposal one could think of will fare better.
The (sophisticated) common sense position is presented in the
introductory chapter of TK, but its influence seems to decline as the book
develops, and in Chapter 9 radical skepticism makes a surprising comeback as
an epistemological threat that must be "answered". Lehrer's reasoning in
defense of what is, in effect, a moderate skepticism is not easy to follow, and I
will not be able to cover all details here. Nonetheless, the main idea seems to be
the following. We cannot rule out the possibility that we may be in error
regarding what we accept because of the general fallibility of our faculties and
because we may be deceived by a Cartesian demon. However, our fallibility
does not imply a lack of personal justification on our part in accepting what we
do accept. For we may still be able to answer or neutralize all objections on the
basis of our acceptance system, or so Lehrer claims. This is accomplished by our

acceptance of our own trustworthiness which serves to neutralize all objections

referring to our fallibility (p. 220). Recall that, in general, n neutralizes 0 as an
objection to p if accepting the conjunction of 0 and n is not an objection to p and
is as reasonable as to accept 0 alone (relative to a person, an evaluation system
and a point in time). The principle of trustworthiness neutralizes fallibilism as
an objection to any specific knowledge claim since it is, Lehrer contends, as
reasonable to accept both that we are fallible and that we are trustworthy as it is
to accept that we are fallible alone. This last step is clearly one of the weaker
links in Lehrer's argument, and, to the best of my knowledge, Lehrer has
presented no detailed argument in its support.
To continue the argument, personal justification is not sufficient for
knowledge. In order for me to know, my personal justification must be
undefeated by errors in the evaluation system. For instance, it must be true that
I am really trustworthy, or the neutralization of the fallibilistic objection will fail
relative to the ultrasystem. In short, if there is a demon, I will fail to know even
the most basic thing: "[i]fwe were massively mistaken, as we would be if the
Cartesian demon were loose in the land, then we would lack knowledge" (p.
209). On the other hand, if there is no demon, I will presumably know most
common sense things.
Lehrer's argument does not allow us to conclude, on good grounds, that
we have knowledge. On the other hand, we may not conclude that we are
ignorant either. Whether we know or not depends on whether we are
systematically deceived or not. Since, as a matter of principle, we can have no
empirical evidence either for or against systematic deception, I take it that
Lehrer's moderately skeptical concept of knowledge will be useless in
formulating empirical laws and theories. This is, of course, a problematic fact
on the background of Lehrer's methodological reliance on Carnap's notion of
This moderately skeptical view should be contrasted with what I take to
be Lehrer's common sense position, according to which we do have knowledge
whether we are systematically deceived or not (if systematic deception is at all
regarded a serious possibility). It should also be distinguished from radical
skepticism, or at least one form of it, according to which we are ignorant
whether we are systematically deceived or not. Lehrer, the common sense
philosopher, is in conflict with Lehrer, the moderate skeptic. For, as I interpret
the former, he is saying that we have knowledge even if we are systematically
deceived. This is denied by the latter, according to whom systematic deception
implies ignorance.
For reasons already noted, I find the view of Lehrer, the moderate skeptic,
unconvincing. A crucial step relies on the rather vague idea of comparative
reasonableness and, more fundamentally, it is unclear to me what the
methodological basis for his answer to skepticism might be. On the other hand,

I have great sympathy for Lehrer, the sophisticated common sense philosopher,
who has a simple and, in my view, entirely satisfactory answer to skepticism
based on some general methodological considerations concerning the nature of
philosophical and scientific definitions. Lehrer's answer to skepticism is
critically assessed in the contributions by John W. Bender and Peter Klein.
There is also the issue of whether Lehrer's theory implies radical
skepticism concerning facts of acceptance. Robert Shope (1978) has argued that
Lehrer, in his account of undefeated justification, commits the "conditional
fallacy". A similar objection is raised by Peter Klein in this volume. Suppose
that I accept some proposition p which is actually false. While p's falsity
prevents me from knowing that p, it should not prevent me from knowing that
1 accept that p. But Lehrer's theory seems to have just this consequence: if p is
false, 1 cannot know even that 1 accept that p. The reason is that "I accept that
p" will not be in the ultrasystem, making it impossible for me to answer or
neutralize the objection that 1 do not accept that p. Lehrer, as a matter of fact,
has addressed this problem in the latest edition of TK, in which the ultrasystem
is required to acknowledge the existence of the eliminated states of acceptance,
preference and reasoning in the original evaluation system (p. 160). The
conditional fallacy is discussed in David A. Truncellito's contribution to this
volume. Truncellito maintains that the objection rests on a misunderstanding of
Lehrer's account.
The purpose of this introduction has been merely to give a brief overview
of Lehrer's theory of knowledge and justification, and to indicate what might be
possible troubles. It goes without saying that there is much more to be said about
Lehrer's epistemology. My perhaps most serious omission concerns his and Carl
G. Wagner's celebrated theory of rational consensus which they have developed
in a series of papers and an influential book, Rational Consensus in Science and
Society. Wagner's paper in the present volume discusses some probabilistic
issues of importance to this other, no less important, aspect of Lehrer's
epistemological theorizing. 3

1 Lehrer writes on p. 10 in TK: "We shall be concerned with an analysis that will be useful for

explaining how people know that the input (the reports and representations) they receive from
other people, their own senses, and reason is correct information rather than error and
misinformation. A person may receive a representation that p as input without knowing that the
representation is correct and, therefore, without knowing that p."
2 For reasons that that will emerge later, the ultrasystem is also required to acknowledge the
existence ofthe eliminated states of acceptance, preference and reasoning in the original evaluation
system (p. 160).
3 Thanks are due to Gordian Haas for his comments on an earlier version of this introduction.

Carnap, R. (1950), Logical Foundations ofProbability, The University of Chicago Press, Chicago.
Dretske, F. (1991), "Two Conceptions of Knowledge: Rational vs. Reliable Belief', Grazer
Philosophische Studien, Vol. 40: 15-30.
Lehrer, K. (1974), Knowledge, Clarendon Press, Oxford.
Lehrer, K. (1975), "Reason and Consistency", in Analysis and Metaphysics: Essays in Honor of
R, M Chisholm, Lehrer, K. (ed.), Dordrecht.
Lehrer, K. (1989), "Reply to My Critics", in The Current State ofthe Coherence Theory, Bender,
J. W. (ed.), Philosophical Studies Series, Vol. 44, Kluwer, Dordrecht.
Lehrer, K. (I 990a), Metamind, Clarendon Press, Oxford.
Lehrer, K. (1990b), Theory of Knowledge, First Edition, Westview Press, Boulder.
Lehrer, K. (1991), "Reply to Mylan Engel", Grazer Philosophische Studien, Vol. 40: 131-133.
Lehrer, K. (1997), Self-Trust: A Study of Reason, Knowledge and Autonomy, Clarendon Press,
Lehrer, K. (2000), Theory of Knowledge, Second Edition, Westview Press, Boulder.
Lehrer, K, and Wagner, C. G. (1981), Rational Consensus in Science and SOCiety, Reidel,
Olsson, E. 1. (1998), "Competing for Acceptance: Lehrer's Rule and the Paradoxes of
Justification", Theoria, Vol. 64: 34-54.
Olsson, E. J. (1999), "Cohering With", Erkenntnis, Vol. 50, Nos. 2-3: 273-291.
Rott, H. (2001), Change, Choice and Inference, Oxford Logic Guides No. 42, Oxford University
Shope, R. (1978), 'The Conditional Fallacy in Contemporary Philosophy", Journal ofPhilosophy,
Vol. 75: 397-413.
Chapter 1



Ernest Sosa
Brown University and Rutgers University

It is Lehrer's epistemology that most engages me, and that is what I

will focus on here, while acknowledging his more recent extra-
epistemological analogies and comparisons, and the parallel forms of
reasoning that he uncovers concerning ethics, and even wisdom, autonomy,
and love. Perhaps we will go astray in abstracting thus from the broader
context. If so, I hope Keith will set us straight.
Let me begin with a word of explanation about my title. Independence
there is Lehrer's term, by which he means non supervenience. And here is the
question on which I will focus: What is the place of supervenience in
epistemology? Most epistemologists take epistemology to include as a central
concern the study of normative epistemic statuses such as epistemic
justification, warrant, rationality, reasonableness, and knowledge itself. It is
widely held that although the normative is not definable in terms of the
nonnormative, nevertheless it does supervene on the nonnormative. Recall the
notorious naturalistic fallacy. The mistake there is essentially that of trying to
define the normative through the nonnormative. It was G.E. Moore who
brought that fallacy to our attention, but Moore himself also advocated a
thesis of the supervenience of the normative on the nonnormative. As for the
supervenience of the epistemically normative, that would seem just a special
case of the supervenience of the normative generally.
Again, the orthodox stance has it that epistemic justification and other
normative epistemic statuses supervene. This is often just taken for granted as
it silently structures reflection and research in the field. Main controversies,
such as internalism/externalism and foundationalism/coherentism, can often
E.!. Olsson (ed.), The Epistemology of Keith Lehrer, 23-30.
© 2003 Kluwer Academic Publishers.

be seen as disagreements over the general shape of how justification

supervenes. Occasionally this becomes explicit, especially in recent work.
For example, in a recent defense of epistemic internal ism, Conee and
Feldman call their version mentalism, which is the thesis that the epistemic
justification of any belief supervenes on the mental properties of the believer.
I myself favor the boringly orthodox stance. Lehrer, by contrast, has
argued for the more exciting contrarian view, that the epistemically normative
does not supervene. Excitingly contrarian views are bound to meet resistance,
however, and in keeping with my role as commentator I intend in what
follows dutifully to resist.
In epistemology, Lehrer has been a vigorous opponent of externalism
and foundational ism. Together with Bonjour and others influenced by
Wilfrid Sellars, he has rejected simplistic thermometer conceptions of
knowledge as belief with appropriate causal, or tracking, or reliabilist
properties. These properties may all remain outside the relevant grasp of the
believer, after all, in which case they can give him "information" but hardly
knowledge. What more is required? Here Lehrer's internal ism and
coherentism come to the fore.
To know that p, one needs a coherent perspective wherein one "accepts
that p with the goal of gaining truth and avoiding error." Accepting that p is
not just believing it. It is rather an evaluative stance on the question whether
p, to the effect that it is advisable to accept it, with a view to gaining truth and
avoiding error. No matter how coherently and deeply one might defend one's
acceptance of p with further supportive acceptances, however, this will not
ensure that one knows, since one's justification can still be defeated. How so?
Through the falsity of enough of the supportive structure that lends coherence
to one's acceptance ofp.
Either one's justification is defeated, then, or it is not. If one's
justification is defeated, if the removal of falsehood does entail the loss of
sufficiently coherent support, then one's acceptance of p is justified
personally but not also verifically. If one's justification is not defeated, on the
other hand, if one's accepting p is still supported with sufficient coherence
even after the removal of all such falsehood, then one's acceptance of p is
justified both personally and verifically.
It is one's acceptance system in the search for truth (and avoidance of
error) that determines what one is justified in accepting. Only what coheres
with one's system is one justified in accepting. How then does acceptance of
a hypothesis cohere with such a system? This is detailed through a complex
set of definitions, but the key idea is that accepting the hypothesis needs to be
reasonable enough in the light of the acceptance system. (Here I am
simplifying by focusing on the acceptance system, whereas Lehrer's more
recent, fuller view would replace the acceptance system with the so-called

evaluation system, which includes not only acceptances and evaluations of

acceptances of various sorts, but also preferences and evaluations of
preferences of various sorts. My simplified account is thus a caricature, but
one that does, I hope, still convey, perhaps more vividly, key features of the
epistemology proper.)
Crucial to Lehrer's defense of such coherentism is his claim that the
epistemic does not supervene on the nonepistemic and natural. In his view,
only this rejection of supervenience enables him to defend his favored
coherentism from its rival, foundationalism. So how should we understand
the rejected supervenience thesis? What is the supervenience at issue?

Properties X supervene on properties Y if and only if whenever

an X property is exemplified, this follows necessarily from the
exemplification of a Yproperty.

That then is what is required for epistemic properties of beliefs to

supervene on their natural, nonepistemic properties. According to
Lehrer, no such requirement is satisfied by the epistemic relative to the
nonepistemic: there are no nonepistemic, natural properties and
relations, such that whenever one reasonably accepts something, this is
entailed by the fact that one's acceptance exemplifies such natural
properties or enters into such natural relations.
One might take coherentism to offer an advantage over
foundationalism for those who prize explanation. Thus compare the
following two principles: one foundationalist, one coherentist.

(F) IfNBp, then Jp,

where NB is a naturalistic property of basic
beliefs and J is a property of justification. To
the question, why are we justified in
accepting that p ifNB of p, the answer is that
we just are ... 2

(CSR) If S has an evaluation [acceptance] system A which

contains (CSR) and p n-coheres with A, then
S is justified in accepting that p.

If On-coheres' were defined non-epistemically, perhaps in terms

of some naturalistic probability relation such as relative truth
frequency, this principle could be construed as a principle of
supervenience .... 3

Now, it may be thought an advantage of CSR-coherentism that one

could explain one's justification for acceptance of CSR by appeal to CSR
itself, so long as CSR itself n-coheres with one's evaluation [acceptance]
system A. It may be thought that F would be denied this explanatory
advantage. But this is only an illusion. After all, there is no reason why F
could not itself have foundationalist property NB, in which case the
foundationalist could equally explain his justification for accepting F by
appeal to F itself.
Having recognized that no advantage accrues to coherentism through
the reasoning just reviewed, Lehrer reacts as follows:

If we are interested, therefore, in arguing that the coherence

theory has an advantage in explaining why things are justified
that the foundations theory must lack, we must reject the thesis
of supervenience and argue for a coherence theory that makes
justification independent [non supervenient]. Moreover, we must
do this in such a way as to show why coherence leads to the
doctrine of independence and the foundations theory does not.
But how can we accomplish this? What is there about coherence
that yields independence?4

"The answer, in brief, is that an adequate account of coherence must

involve epistemic notions." Lehrer argues for this both with regard to
the undefeated justification requisite for knowledge and also with
regard to the personal justification requisite for undefeated justification.
I confess that the structure of Lehrer's argument for this stance is
not crystal clear to me. So far as I have been able to get it in focus,
however, it seems to depend on a point made for example in the
following two passages:

... I am trustworthy in what I accept only if I can tell whether I

am trustworthy or not. I must be able to tell whether I am
trustworthy or not about something in order to be trustworthy
about it. I must accept that I am trustworthy when I am and not
accept that I am trustworthy when I am not, at least a trustworthy
amount of the time. 5

... if I am asked how often a person must be correct in order to

be trustworthy over time, I can only say, uninformatively, that he
must be right a trustworthy amount of the time. 6

If I have understood him properly, Lehrer reasons that even so much as

personal justification requires that the subject be trustworthy enough, and
there's no specifying in nonepistemic terms how trustworthy that needs to be.
So even the most basic epistemic state, personal justification, will fail to
supervene on the nonepistemic. Equally, therefore, will the more complex
epistemic states, of verific justification and of knowledge, also fail to
supervene, constituted as they are out of personal justification. 7
I have the following three questions about this way of reasoning in
favor of coherentism and against supervenience.

1. Even if the relevant epistemic statuses, of knowledge, say, and of epistemic

reasonability and justification, are to be understood in terms of coherence,
and even if an adequate "account" of coherence must involve epistemic
notions, how does this refute the supervenience of the epistemic? Compare
tallness. It may be that there is no way to define 'tall relative to group G'
without using tallness terminology. Thus it may not be possible to do so
without using the predicate "sufficiently taller than the average" or the like,
which will only raise the question "Sufficiently for what?"-to which the
only acceptable answer may be "Sufficiently to make one tall." Yet, even if
any adequate account of tallness must in that way involve tallness notions, so
that in effect there is no acceptable noncircular definition of that notion,
surely tallness does supervene on the distribution of heights over the relevant
group. Moreover, the point here need not be restricted to relative terms such
as 'tall' and its cognates. It would seem to apply well beyond that, to any
terms that while indefinable in terms of some basis still supervene on that
basis. (And the point would apply not only to definability but also to
reducibility, if the standards for biconditional reducibility are set lower than
those for outright definability. It would still be plausibly arguable that failure
of biconditional reducibility to a certain basis is compatible with
supervenience on that basis.)
Suppose we do need to appeal in our explication of our concept of
tallness, such as it is, to "sufficiently taller than the average to make one tall."
Surely we can still insist quite plausibly that whenever someone S is tall
(relative to group G) this follows necessarily from the distribution of heights
in group G, including S's own height. Moreover, it would still seem plausible
that, relative to any possible counterpart group with the same distribution of
heights, necessarily the counterpart of subject S would also be tall (relative to
the counterpart of group G) as a necessary consequence of that distribution of
Note, moreover, that it does not matter for our purposes how the
people in that counterpart group use the word 'tall' or what contextual
variation there might be to their proper and truthful application of that word.

That does not affect the truth or falsehood of what we say when we say that in
that group what matters to whether the counterpart of S is tall is the
distribution of heights in the group. Here we are of course relying on what we
mean by 'tall' and on whatever contextual constraints there may be on our
present proper and truthful use of such terminology. So it is important to
distinguish our question of whether tallness-relative-to-G supervenes on the
distribution of heights alone, from the question of whether the correctness of
applications of 'tall (for group G)' in group G is entailed simply by the
distribution of heights. (Obviously it is not entailed simply by that; such
correctness is also crucially dependent on the standards operative in group G
for the use of the relevant terminology, which need not coincide with our
standards. )
Our question should also be distinguished, moreover, from the question
whether our attributions of 'tall (for group G)' have their correctness
determined simply by the distribution of heights. Here again, the answer to
our correctness-of-attribution question is clearly in the negative, since here
again a crucial factor determinative of such correctness will be the standards
and meanings for the predicate among the relevant set of speakers, those the
correctness of whose attributions is at issue. But this further factor is surely
not determinative of which of us is actually tall; it is determinative rather of
which of us is now truthfully and properly said to be "tall" (which in turn is
distinct from the question of which of us is now truthfully and properly said
to be tall.)

2. Lehrer's defense of coherentism relies on the idea that coherentism

comports better with the nonsupervenience of the epistemic. And he believes
this to be so in virtue of something else he believes: namely, that there is no
"account" of the epistemic in nonepistemic terms. So far I have questioned
whether nonsupervenience would indeed derive from such lack of an
"account." Be that as it may, there is now this further question: Why should
we suppose that no foundationalist status could be akin to coherence in
failing to have any adequate "account" in nonepistemic terms? Take clarity
and distinctness as the relevant foundationalist status. In fact, Descartes's
proposal is in terms of enough clarity and distinctness. And, in any case, even
if it is not Descartes's own, a foundationalist proposal could be defined in
terms of enough clarity and distinctness. This would of course raise the
question "Enough for what?"-to which the only acceptable answer might
then be "Enough to give one certainty (or justification, or reasonableness,
etc.)." Suppose, only for the sake of argument, that the indefinability in
nonepistemic terms of the key epistemic status used by a theory
(foundationalist or coherentist) would make that epistemic status independent
(nonsupervenient). And suppose, again only for the sake of argument, that

this would give that theory an advantage in competition with its rival theories
of that epistemic status. In that case, it would seem that our foundationalism
of enough clarity and distinctness would share the supposed advantages of

3. Suppose, again only for the sake of argument, that coherence is

inextricably epistemic, and, furthermore independent: that it does not
supervene on the nonepistemic. And consider a foundationalist theory on
which epistemic status does supervene on nonepistemic properties. Why
should this give an advantage to coherent ism? If foundational justification is
in fact supervenient, might this not give an advantage to foundationalism:
namely, that it provides an account of epistemic justification, and of
knowledge, and explains how these are supervenient on the nonepistemic? So
long as the foundational account is able to do as well as the coherentist
account in every other relevant respect, why should the coherentist account
derive any advantage from the fact that it invokes only properties that are
independent and not supervenient?
Having thought hard about Lehrer's work for years, I was not about to
stop doing so for this occasion. Thinking hard about something often leads to
a lot of questions, of course, and much of the fun of philosophy is in the give
and take of question and answer. I have tried to present my questions as
starkly and clearly as I could, so as to avoid side issues and distracting
qualifications. This does incur the risk of overlooking important subtleties.
But if my questions are based on oversight or misunderstanding, they may at
least elicit from Keith a fuller account of just how his anti supervenience
argument for coherentism is supposed to go. It is an honor to have this
opportunity to importune my friend, and I look forward to his response. 8


I This paper originated as a talk in a Symposium in honor of Keith Lehrer held as part of the 2001

Pacific Division meetings of the American Philosophical Association. I remember meeting Keith
at an APA "smoker," as they were called in the early sixties. Keith was then a professor in the
legendary Wayne State department, I a graduate student at Pitt. Little did we know that our paths
would cross so often in the years to follow, always to my benefit, as Keith realized his brilliant
potential, with many publications on issues of main interest to us both, especially in epistemology.
Though we started out far apart, over the ensuing decades our relevant views have converged
steadily. To dwell on agreement would be boring, however, so here I will discuss one main
remaining disagreement.
2 Self-Trust: A Study of Reason, Knowledge, and Autonomy (Oxford University Press, 1997), p.

3 Ibid., p. 65.
4 Ibid., p. 67.

5 Self- Trust, pp. 70-l.

(, "Replies," in a book symposium on Self-Trust in Philosophy and Phenomenological Research

59 (1999): 1065-74; p. 1071.
7 On the question of supervenience there is a gulf between Lehrer and me. Already in "How Do

You Know'?" (APO, 1974; reprinted in my Knowledge in Perspective), r am committed to the

supervenience of the epistemic: Every case of knowledge by a subject S is said there to require a
Tree of knowledge, such that: "Trees display true epistemic propositions concerning S and they
also show what 'makes their propositions true' via epistemic principles .... A tree must do this for
every epistemic proposition that constitutes one of its nodes. That is to say, trees contain no
epistemic terminal nodes. It is in this sense that trees provide complete epistemic explanations .... "
(KIP, p. 33.)
8 For illuminating discussion of epistemic supervenience and of Lehrer's views see James Van

Cleve's "Epistemic Supervenience revisited," Philosophy and Phenomenological Research LIX

(1999): 1049-1055, part of a PPR book symposium on Lehrer's Self-Trust, along with Lehrer's
reply to Van Cleve on pp. 1069-1071. That in turn continues a conversation between the two of
them over many years, continued first in Self-Trust, and then in the book symposium. My present
paper joins that conversation, to which I had also been invited already in Lehrer's book.
Chapter 2


John Greco
Fordham University

When I was a graduate student I would have gone to great lengths to talk
with Keith Lehrer about epistemology. In fact, I did! Hearing that Lehrer would
be talking at my alma mater, Georgetown University, I drove down to
Washington D.C. from Providence, listened to his talk, and got myself invited
to dinner. Much to the Georgetown students' horror, and probably Lehrer's as
well, I proceeded to monopolize his time with my questions and commentary.
(Sorry about that, Keith.) In any case, it is a great honor for me now to be
invited to engage Lehrer more formally and more appropriately. I very much
appreciate this opportunity.
In his Theory ofKnowledge Lehrer tells us that his coherentist theory can
be regarded as a form of reliabilism. 1 His reasoning is that, on his view,
reliability is a necessary condition of both undefeated justification and
knowledge. More specifically, personal justification requires that one accept
both of the following principles:

(T) I am trustworthy (worthy of my own trust) in what I accept with the

objective of accepting something just in case it is true, and

(TR) If! am trustworthy in what I accept, then I am reliable in obtaining

truth and avoiding error in what I accept.

Moreover, undefeated justification and knowledge require that both of these

acceptances be true. Hence undefeated justification and knowledge require
reliability regarding what one accepts. (194) Notice that the reliability that
Lehrer's view requires is reliability of the knower herself, as opposed to
E.1. Olsson (ed.), The Epistemology of Keith Lehrer, 31-41.
© 2003 Kluwer Academic Publishers.

reliability of her methods, or her evidence, or something else. Hence Lehrer's

view is a version of what I have elsewhere called "agent reliabilism"; i.e., he
requires that the knower herself, the cognitive agent, be reliable regarding what
she accepts. 2 In this paper I want to raise the following question: Why isn't
agent reliability enough? Or better, why does knowledge require, in addition to
agent reliability, further conditions regarding coherence? In pursuing this
question, I will consider what I take to be Lehrer's three most important
arguments for thinking that agent reliability is not enough, and that therefore
something further, such as coherence, is required. I will argue that none of
Lehrer's arguments against a non-coherentist version of reliabilism is
Lehrer's three arguments can be summed up as follows.

1. Knowledge has a functional role in science and in practical reasoning, and

this functional role requires abilities involving articulation, reason-giving,
and defense against objections. Hence the grounds of knowledge must be
in an appropriate way available to the knower, as they are in a coherence

2. Externalism in general, and reliabilism in particular, are open to "the

opacity objection." More specifically, they are open to counter-examples
such as that ofMr. Truetemp, in which a) the person in question is reliable
in what he accepts, but b) this reliability is opaque to him. Contra
reliabilism, such persons lack knowledge. However, coherence removes
opacity and thereby issues in knowledge.

3. Gettier problems show that true reliable acceptance is not sufficient for
knowledge. However, coherence requires further conditions, and these
work so as to solve Gettier problems.

In the next section I will consider Lehrer's first argument rather briefly.
I will then tum to the opacity objection in a more extended way. Here my
argument will be this: Depending on how we interpret Lehrer's conditions for
coherence, we must say either a) that Mr. Truetemp satisfies those conditions
and therefore knows, or b) that in ordinary cases of perception, people do not
satisfy those conditions and therefore do not know. If (a), then the counter-
examples employed by the opacity objection fail to distinguish Lehrer's
coherentism from non-coherentist versions of reliabilism. If (b), then Lehrer's
coherentism has unacceptable skeptical results. I will end by considering
Lehrer's treatment ofGettier problems. Here I will argue that a non-coherentist
version of agent reliabilism can adopt a strategy very similar to Lehrer's.



Lehrer has long endorsed a broadly Sellarsian theme: that knowledge properly
so-called involves abilities to articulate and give one's reasons, and to defend
one's knowledge and reasons against relevant objections. In the second edition
of his Theory of Knowledge, for example, Lehrer writes:

It is fundamental to the kind of human knowledge that concerns us in this

book that it is inextricably woven into reasoning, justification,
confirmation, and refutation. It is required both for the ratiocination of
theoretical speculation in science and practical sagacity in everyday life.
To do science-to engage in experimental inquiry and scientific
ratiocination-one must be able to tell whether one has correct
information or not.. .. Engaging in law or commerce requires the same sort
of knowledge, which may be used as the premises of critical reflection or
claimed as the prizes of informed reasoning. (6)

This same theme is the basis for objections to both foundationalism and
externalism. Thus Lehrer objects to a foundationalist account of
perceptual beliefs on the following grounds:

Perceptual beliefs are considered innocent until proven guilty when we

care not the least whether the belief is innocent or guilty. Once we do
care, though, we start to ask serious questions, the ones concerning
justification .... We seek to determine if the person has information that
would enable him to determine whether he actually sees the thing he thinks
he does and render him justified. (74)

A similar objection is lodged against externalism.

As in our refutation offoundationalism, what is missing from the accounts

of externalists is the needed supplementation of background information
and the transparency of it.... Such information is what is needed to
supplement the information contained in the belief alone, and it is
precisely the sort of information required for coherence and irrefutable
justification. (185-6)

The reasoning behind these objections can be reconstructed as follows.

First, knowledge plays an essential role in scientific inquiry and practical
wisdom, which role makes knowledge "inextricably woven into reasoning,
justification, confirmation, and refutation." But these, in turn, issue in the

requirement of relevant background information. In other words, such

background information is needed to make possible the knowledge-related
practices of reasoning, justification, confirmation, and refutation.
This sounds reasonable enough, but it raises the question of psychological
plausibility. Namely, is it plausible that we typically have the background
information that is required? For example, do we typically have relevant
information about our perceptual faculties, the circumstances in which they are
reliable, and whether we are presently situated in such circumstances? Lehrer's
answer to the psychological plausibility objection is to understand background
information in terms of acceptances, and to understand acceptances in terms of
their functional role in reasoning, confirmation, etc. 3 Hence, for S to accept that
her perceptual faculties are reliable in present circumstances does not require
that she explicitly judges that this is so. Rather, it requires only that S act in
certain ways when in certain circumstances. But can we think of this sort of
implicit acceptance as background information, as we would have to for it to be
relevant to coherence and justification? Lehrer's answer is yes:

A person may be said to have information he cannot easily describe and

to employ such information in various ways. For example, suppose I
know the shortest route from Rochester to Buffalo, though I cannot tell
you the name of the highway. Moreover, imagine that I am not very good
at giving directions, so I cannot tell you how to get from Rochester to
Buffalo .... That I make the trip successfully on many occasions shows that
I have the required information, which I might find difficult to articulate,
about the route from Rochester to Buffalo. Similarly, my reliability in
accepting that I see when I do, or even in accepting that I feel or think
when I do, depends on my ability to employ information, which I might
find difficult to articulate, about seeing, feeling, and thinking. (74-5)

The problem is now this: The present objection to externalism, and the
motivation for turning to a coherentist version of reliabilism, is that knowledge
requires background information for the purposes of reasoning, justification,
confirmation, and refutation. But now, it turns out, the background information
that coherence provides need not be available for use in these activities. It
would seem that Sellarsian themes regarding the role of knowledge in science
and practical wisdom cannot be used to motivate Lehrer's coherentism over non-
coherentist reliabilism.


The opacity objection to externalism in general and to reliabilism in particular

is as follows.

There is, however, a general objection to all externalist theories that is as

simple to state as it is fundamental: the external relationship might be
opaque to the subject, who has no idea that her beliefs are produced,
caused, or causally sustained by a reliable belief-forming process or
properly functioning cogntive faculty .... All externalist theories share a
common defect, to wit, that they provide accounts of the possession of
information, which may be opaque to the subject, rather than of the
attainment of transparent knowledge. (185)

The opacity objection is illustrated by the case ofMr. Trutemp.

Suppose a person, Mr. Truetemp, undergoes brain surgery by an

experimental surgeon who invents a small device that is both a very
accurate thermometer and a computational device capable of generating
thoughts .... Assume that the tempucomp is very reliable, and so his
thoughts are correct temperature thoughts. All told, this is a reliable
belief-forming process and a properly functioning cognitive faculty.
Now imagine, finally, that Mr. Truetemp has no idea that the
tempucomp has been inserted in his brain and is only slightly puzzled
about why he thinks so obsessively about the temperature; but he never
checks a thermometer to determine whether these thoughts about the
temperature are correct. He accepts them unrefiectively, another effect of
the tempucomp. Thus, he thinks and accepts that the temperature is 104
degrees. It is. Does he know that it is? Surely not. He has no idea
whether he or his thoughts about the temperature are reliable. (187)

The example ofMr. Truetemp is supposed to serve as a counter-example

to reliabilism. More specifically, it is supposed to be a counterexample to non-
coherentist versions of reliabilism. The idea is that Mr. Truetemp does not
know, although he does fulfill the conditions for knowledge set down by non-
coherentist reliabilism. However, the example does not serve Lehrer's purposes
unless some other things are true as well. First, it must be the case that Mr.
Truetemp does not know on the conditions for knowledge set down by Lehrer's
coherentist theory. Otherwise, the Truetemp example will not distinguish non-
coherentist reliabilism from coherentist reliabilism. Furthermore, it cannot be
the case that Lehrer's theory is overly restrictive, so that it turns out that, on his
view, there is no knowledge even in ordinary cases of perception. Otherwise, the
fact that Lehrer's theory rules that Mr. Truetemp does not know will not indicate
a virtue of his theory. Accordingly, for the Truetemp example to work for
Lehrer against non-coherentist reliabilism, all of the following three conditions
must be fulfilled:

1. Mr. Truetemp knows on non-coherentist reliabilism,


2. Mr. Truetemp does not know on Lehrer's coherentist

reliabilism, and

3. Ordinary cases of perception count as knowledge on

Lehrer's coherentist reliabilism.

I will now argue that there is no way to interpret Lehrer's theory so that
all of these conditions are fulfilled. More specifically, on some interpretations
of Lehrer's theory, Mr. Truetemp knows. On other interpretations, ordinary
cases of perception do not count as knowledge.
First, it is not clear why Mr. Truetemp does not know on Lehrer's
account. Specifically, it is not clear why Truetemp is not in the same position,
vis-a.-vis coherence, as knowers are in ordinary cases. Consider, for example,
Lehrer's remarks concerning knowledge and skepticism.

Although we thank the skeptic for reminding us that...we sometimes err

and are ever fallible in our judgment, we may at the same time neutralize
her objection .... The objection based on our fallibility is neutralized by our
trustworthiness. It is as reasonable to accept both that we are fallible and
that we are trustworthy in a truth-connected way as it is to accept only that
we are fallible. (220)

In general, Lehrer thinks, we can invoke our trustworthiness to reason that some
acceptance is reasonable.

A consequence of adding principle (T) to my evaluation system is that I

may reason from it and the acceptance of some target acceptance that p to
the conclusion that the target acceptance is reasonable. My reasoning
would be as follows.

T. I am trustworthy (worthy of my own trust) in what I accept with the

objective of accepting something just in case it is true.

I accept that p with the objective of accepting that p just in case it is true.

Therefore, I am trustworthy in accepting that p with the objective of

accepting that p just in case it is true.

Therefore, I am reasonable in accepting that p with the objective of

accepting that p just in case it is true.

The argument from trustworthiness to reasonableness, which I shall refer

to as the trustworthiness argument, assumes that my trustworthiness may
explain why it is reasonable for me to accept what I do. (139)

The relevant question is, why can't Mr. Truetemp reason in exactly the same
way? Put differently, why isn't the reasonableness of Mr. Truetemp's
acceptance about the temperature explained in exactly the same way?
Moreover, why can't Mr. Truetemp invoke principle (TR) to reason even
further, and to conclude that his acceptances about the temperature are not only
trustworthy, but reliable?
Again, consider how, according to Lehrer, the dialectic with the skeptic
proceeds in ordinary cases.

Critic (or skeptic): Let us admit that your are intellectually trustworthy and
intellectually virtuous, as you claim. You are, nevertheless, in error
because such trustworthiness and virtue fail to achieve their purpose. What
you accept in this trustworthy and virtuous way is not reliably connected
with the truth ....

What should the claimant reply? The reply must be that the critic or
skeptic is wrong! ...

Claimant: What I accept in this trustworthy and virtuous way is reliably

connected with truth in a successful way. The way of virtue is also the
way of truth. (211-2)

Again, why can't Mr. Truetemp follow the same dialectic? If he can, then
Truetemp knows on Lehrer's account, and so the Truetemp case does not
distinguish non-coherentist reliabilism from Lehrer's coherentism.
I assume that Lehrer thinks that Mr. Truetemp cannot follow the same
dialectic. However, it is not clear why. One reason that suggests itself is that
Mr. Truetemp accepts his thoughts about the temperature "unreflectively." But
what does this mean? One thing it might mean is that Truetemp has no further
acceptances regarding his trustworthiness and reliability in this regard, and so
does not have the requisite materials in his evaluation system to sustain a
relevant dialectic. However, it is not at all clear that Mr. Truetemp does not
have the relevant acceptances in his system. Remember, Lehrer defines
acceptances in terms of their functionality in action and thought. But according
to the example, Mr. Truetemp does act and think as if he thinks that he is
trustworthy and reliable regarding his ability to determine the temperature. He
acts and thinks more or less the way that normal people do regarding their ability
to determine simple mathematical truths, or regarding their ability to determine
colors in good light. That is, he acts and thinks as if he thinks that he is
trustworthy and reliable, although he makes no conscious judgements in this
regard. Moreover, Mr. Truetemp is reliable in the relevant regard, and therefore
his acceptances are true and convert personal justification to undefeated
justification and knowledge.

Lehrer might object that Mr. Truetemp cannot explain his ability to tell
the temperature in any detailed way; i.e. in a way more specific than referring
to principles (T) and (TR). But neither could most people defend their ability
to tell colors in any detailed way. Most people, I assume, would simply insist
that they can tell, or else very quickly get into a quite inadequate (and probably
false) explanation as to how they can tell.
Let me be clear: The point I am making here is not that Mr. Truetemp
knows. On the contrary, I think that he does not know, and below I will offer
my own explanation regarding why he does not know. My point, rather, is that
Lehrer's coherence theory is faced with the dilemma set out above. Depending
on how one interprets Lehrer's conditions for coherence and knowledge, either
a) Truetemp has the requisite true acceptances in his evaluation system, in which
case he knows, or b) he does not, in which case neither do ordinary perceivers,
and so they don't know.
Let us consider one more reply on Lehrer's behalf, however. Lehrer
might argue that there are true objections (or competitors) in the Truetemp case,
which defeat Truetemp's personal justification, and which are not present in
cases of ordinary perception. For example, Lehrer might argue, Truetemp
cannot answer the following objecton: (0) Someone has implanted a device in
your brain, and that is why you believe what you do about the temperature.
Notice, however, that (0) is misleading: (0) counts as an objection on Lehrer's
view, since it is less reasonable for Truetemp to accept what he does about the
temperature on the assumption that (0) is true than it is on the assumption that
(0) is false on the basis ofTruetemp's evaluation system. However, Truetemp
is perfectly reliable in matters regarding the temperature, and so the objection
is misleading insofar as it suggests a reason against Truetemp's belief. But now
consider that there will also be misleading objections available against S's
acceptances in ordinary cases of perception. For example: (0') You are reliable
in your perceptual beliefs in conditions C but not C', and it is possible that you
are presently in conditions C'. Or consider: (0") There are experts in
philosophy and science who believe that apples are not red and that the sky is
not blue. How are these sorts of objections to be answered in ordinary cases of
perception? Why should we think that such objections always can be answered?
For example, consider that C' in (0') might describe conditions about which S
is wholly ignorant, or perhaps conditions that S lacks the conceptual capacities
even to consider. Similarly, S might have no idea why anyone would accept that
(0") is true, and no idea how to reply to such considerations if she did have an
idea of them.
Perhaps S can neutralize (0') and (0") by referring to principle (T)
and/or principle (TR) above. Or maybe she can neutralize the objections in
some other way. But if she can, why can't Mr. Truetemp neutralize (0) in the
same way? Again, depending on how Lehrer's theory is supposed to handle

misleading objections, it would seem that either a) Truetemp has the requisite
true acceptances in his evaluation system to neutralize such objections, in which
case he knows, or b) he does not, in which case neither do ordinary perceivers,
and so they don't know. Either way, the Truetemp case does not serve Lehrer's
purposes in the opacity objection against externalism and reliabilism.


I now turn to Lehrer's third argument against non-coherentist reliabilism: that

true reliable acceptance is not sufficient for knowledge.

Another equally important reason why reliabilism will not suffice [for the
sort ofjustification required for knowledge] is that global reliability might
be irrelevant locally. Consider again the case of Mr. Goodsumer, who
sums reliably but not in the particular case. He is not trustworthy in the
way he sums in this case. Trustworthiness must be connected with truth
in order for personal justification to convert to knowledge in the particular
case. (224)

What is the nature ofthe required connection?

What does it mean to say that trustworthiness in what one accepts is

successfully connected with truth in what one accepts in the particular
case? It cannot mean, as we have noted in the case of Goodsumer, that
being trustworthy in what one accepts is generally or reliably successful.
It means that the person is successful in accepting what is true because she
accepts what she does in a trustworthy way in the particular case. Her
trustworthiness explains her success in accepting what is true .... Her
trustworthiness and the reliability of it explains her success in the
particular case. (223)

This is a very nice idea. In Gettier cases, the person has true acceptance but it
is only a matter of luck that the acceptance is true. In cases of knowledge, the
person has true acceptance because she is trustworthy (or virtuous) in the way
that she accepts what she does. This idea does not give the advantage to
coherentism over non-coherentist reliabilism, however, because the reliabilist
can say the same thing. More specifically, the agent reliabilist can say the same
thing: in cases of knowledge, S accepts what is true because she is reliable. 4
This move is particularly attractive from the perspective of agent
reliabilism, since in general credit is closely tied to agent reliability. For
example, we give an athlete credit when we think that she accomplishes her feat
because she has great abilities. According to Aristotle, actions deserving moral
credit "proceed from a firm and unchangeable character."s The present

suggestion is that knowledge involves a kind of intellectual credit. In cases of

knowledge, it is to S's credit that she accepts what is true; i.e. she accepts what
is true because she is reliable in the relevant way.
Finally, the present perspective gives us an explanation as to why Mr.
Truetemp does not know. Namely, the tempucomp in Truetemp's brain
undermines the notion that he accepts the truth because he is reliable. In other
words, it is not to Truetemp's credit that he accepts the truth. Consider the
following remarks by Joel Feinberg, concerning the requirements for moral

should like to suggest that the more important to a full and

comprehensive understanding of a triggered action the actor's own
dispositions and thresholds are, the more likely we are to consider the act
truly his, provided that those dispositions and thresholds are not
biologically or psychologically abnormal and that they were not imposed
on him by manipulation. 6

In other words, we do not consider Truetemp's acceptances about the

temperature to be truly his, in the sense required for intellectual credit, precisely
because they are the result of dispositions and thresholds that are "biologically
and psychologically abnormal" and "imposed on him by manipulation."
Truetemp accepts the truth because the tempucomp is reliable, not because he
is reliable. 7
Hence agent reliabilism can endorse Lehrer's strategy for solving Gettier
problems, and can place this strategy within a general theory of credit.
Moreover, this strategy gives agent reliabilism resources for explain ing why Mr.
Truetemp does not know. None of this has anything to do with coherentism,
however. I conclude, then, that Lehrer does not give us good reasons to prefer
coherentism to non-coherentist versions of agent reliabilism. 8

IKeith Lehrer, Theory 0/ Knowledge, 2nd edition (Boulder, CO: Westview Press, 2000), p. 172.
Below, all page numbers in the text refer to this volume.
2. See my "Agent Reliabilism," Philosophical Perspectives. J3, Epistemology, James Tomberlin,
ed. (Atascadero. CA: Ridgeview Press, 1999); and Putting Skeptics in Their Place (Cambridge:
Cambridge University Press, 2000).
3. See Lehrer's "Replies" in The Current State 0/ the Coherence Theory, John Bender, ed.
(Dordrecht: Kluwer Academic Publishers, 1989); and Theory o/Knowledge, esp. pp. 39-40.
4. I argue for this account ofknowledge in "Knowledge as Credit for True Belief," in Intellectual

Virtue. Perspectives from Ethics and Epistemology, Michael DePaul and Linda Zagzebski, eds.
(Oxford: Oxford University Press, 2002).

5. Nicomachean Ethics, II.4. Joel Feinberg suggests that all attributions of moral credit imply that
the action in question proceeds from character. This is probably too strong. Nevertheless, there
is a special sort of moral credit, with which Aristotle is concerned, that does imply this. See .Toe I
Feinberg, DOing and Deserving: Essays in the Theory of Responsibility (Princeton: Princeton
University Press, 1970), p. 126.
6. Doing and Deserving, p. 171. Emphasis added.

7 More needs to be said, of course. I try to say some of it in "Knowledge as Credit for True
Belief." On my own view, abnormality and manipulation do not always undermine positive
epistemic status. But they do in cases where they undermine credit.
x. I would like to thank Keith Lehrer and Ernest Sosa for helpful conversations on relevant issues.
Chapter 3


Jonathan L. Kvanvig
University of Missouri

Some thirty or so years ago, Keith Lehrer attacked the idea that causation
has much to do with knowledge or justification with the case of the gypsy
lawyer, and has more recently endorsed the same kind of attack with the case of
the racist scientist. 2 These cases threaten not only causal theories of knowledge
but also theories of knowledge or justification which require that one's evidence
be at least a partial cause of one's belief. They threaten, that is, the view that
causation is at the heart of the distinction between propositional justification, the
justification one has for the content of one's belief, and doxastic justification,
the justification which attaches to the believing itself. When justification
attaches to the believing itself rather than merely to the content of what is
believed, it is because one holds the belief in question on the basis of the
evidence. When justification only attaches to propositional contents, there is a
failure of such basing, e.g., one may have the evidence but believe for different
This distinction is important for at least two reasons. First, only
doxastically justified beliefs are candidates for knowledge, on any theory which
requires justification for knowledge. Propositional justification is a step in the
right direction, but if one's believing itself is not justified, one cannot have met
the justificatory requirements for knowledge. Second, only doxastically justified
beliefs satisfy any purely intellectual requirement to believe claims that are
justified, for any such requirement will surely include the requirement that we
hold such beliefs for the right reason, i.e., on the basis of that which justifies
them. So a proper account of the distinction between doxastic and propositional
justification is important for a complete theory of justification and for a
complete theory of knowledge.
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 43-62.
© 2003 Kluwer Academic Publishers.

Though Lehrer's arguments played a role in the abandonment of causal

theories of knowledge, the epistemological community has not endorsed his
view that a causal account of the basing relation is defective. In some cases,
epistemologists have simply found Lehrer's examples unpersuasive,3 and in
other cases, they have argued against the conclusion Lehrer draws from his
examples. I want to look at this issue here, for I think that there is more to be
said on behalf of Lehrer's view than has been appreciated. Causality is
ubiquitous in nearly all of our experience of the world, but it is not conceptually
involved in the concepts of knowledge or justification. In particular, Lehrer is
right that the basing relation is not a species of causal relation.
A terminological point is in order before beginning. There is a sense in
which, when someone denies that evidence needs to be causally responsible for
belief in order for it to be doxastically justified or to count as knowledge, that
person is denying that belief needs to be based on that evidence. After all, in
such cases, belief is not causally grounded in, nor explained by, awareness of the
evidence. That is not the concept of basing that is relevant here, however. What
is central here is the distinction between propositional and doxastic justification.
Two jurors, for example, can be presented with precisely the same evidence and
both believe the defendant is innocent. One might believe this claim by
attending to the evidence, and the other because his horoscope said, "you will
need courage today to make a negative judgment about a very bad person." In
the second case, the juror does not attend to the evidence or even consider the
question of whether the evidence confirms the guilt of the defendant. He hears
the evidence and forms beliefs about it, but this experience is not connected at J
all with his belief. So, whereas the first juror is doxastically justified and may "
know, on the basis of the evidence, that the defendant is guilty, the secondjuror
only has propositional justification and fails to know on the basis of the evidence
that the defendant is guilty. The concept of basing that is relevant here is that
concept central to this distinction. Even ifthere are other concepts of basing on
which a person can be said to know without basing belief on evidence (because
the evidence doesn't cause the belief), the concept of concern here is that
concept in virtue of which we distinguish candidates for knowledge in terms of
whether the person has doxastic or merely propositional justification for belief.


We can begin with Lehrer's examples. The first case concerns a gypsy
lawyer of a client accused of eight murders. The lawyer consults the Tarot
cards, and they say that the client is guilty of committing all but the eighth
murder. The lawyer believes what the cards say, and his conviction of the
innocence of his client regarding the eighth murder leads him to reconsider the

evidence, which he now comes to see conclusively establishes that his client is
innocent of the eighth murder. Lehrer then goes on to claim:

He freely admits, however, that the evidence which he claims shows that
he knows his client to be innocent of that crime is not what convinced him
of the innocence of his client, and, indeed, would not convince him now
were he not already convinced by the cards ... His conviction could not be
increased by his consideration of the evidence because he was already
completely convinced. On the other hand, were his faith in the cards to
collapse, then emotional factors which influence others would sway him
too. Therefore the evidence which completely justifies his belief does not
explain why he believes as he does, his faith in the cards explains that, and
the evidence in no way supports ... or partially explains why he believes as
he does. 4

Lehrer thus claims that the lawyer knows that his client is innocent even though
the evidence which justifies his belief does not either prompt his original
acquisition of the belief, nor does the evidence lend increased confidence in the
belief once it is discovered, nor does the evidence at present sustain the belief.
This last point is true in virtue of the fact that, if the influence of the cards were
removed, emotional factors would hold sway and the lawyer would no longer
believe that his client was innocent. Nonetheless the lawyer now knows and
justifiably believes that his client is innocent of the eighth murder.
More recently, Lehrer has forwarded the case of Ms. Prejudice:

Imagine the case of Ms. Prejudice, who out of prejudice against a race
believes that the members of the race who have a certain disease get the
disease because of their genetic makeup. Of course, she, being very racist,
believes this is a sign oftheir racial inferiority, and she is totally convinced
because of her racism that the disease is the result of the genetic
constitution of the race. Now imagine that Ms. Prejudice becomes a
medical student and learns, to her pleasure, of the medical evidence that
supports her prejudiced conviction. She becomes, however, a medical
expert of the highest quality, fully capable of separating her prejudices
from her scientific studies. As luck would have it, she becomes part of a
research team assigned the task of checking on the genetic basis of the
disease .... She wants to make sure it [the disease] really is [genetically
caused], and she will force them [her co-investigators] to investigate with
the greatest care every reason for doubting that the disease is genetically
caused. She wants to make absolutely sure that she cannot be charged
with concluding on the basis of the scientific evidence that the disease is
genetically caused because of prejudice. Of course, her belief that the
disease is genetically caused is the result of her still very intense prejudice,
but her scientific evaluation ofthe evidence in favor of this belief must be
rigorously tested.

Every objection to the claim is considered and refuted by the team, all
of whom, except for Ms. Prejudice, who plays devil's advocate, are
completely without prejudice. They all conclude that the scientific
evidence shows conclusively that the disease is genetically caused, as Ms.
Prejudice has believed all along. After the investigation, she knows that
the disease is genetically caused. She has the same evidence ... for
believing this as the other members of her team, and if they know that the
disease is genetically caused, so does she. But her belief is the product of
an improperly functioning system of racial prejudice. 5

Just as in the case of the gypsy lawyer, the racist scientist initially comes to a
belief on the basis of suspect motivations. According to Lehrer, these
motivations remain the basis for the belief even after learning of the evidence
that confirms the belief. Still, Lehrer claims, the racist scientist has knowledge
(and, by implication, justification) if her colleagues do. After all, they have
worked on the project together, knowing the intellectual character and prejudices
of each member of the team. It is obvious that, if asked who on the team knows
and who doesn't, they'd be perplexed at the idea of having to draw such a
distinction. They'd surely say that either everyone knows or no one does.
Since I will be focusing on the issue of the proper construal ofthe basing
relation, it will be useful to recast the discussion in terms of it. The lawyer and
the scientist each have adequate evidence for thinking what they do, but they do
not initially base their beliefs on that evidence. Later on, each comes to see that
the evidence confirms their belief, even though the evidence is not even a partial
cause of their beliefs. Since there is nonetheless a distinction between merely
having sufficient evidence for a belief and having that evidence justifY one's
believing of the claim in question, we can characterize the two cases as follows.
The lawyer and the scientist come to be doxastically justified in believing what
they were originally not doxastically justified in believing, and this claim
implies that they satisfY whatever basing relation between evidence and belief
is appropriate for capturing the distinction between doxastic justification and
propositional justification. Their beliefs are nonetheless not caused, not even
partially, by their awareness of the evidence; their deficient intellectual
characters result in still-true regrettable causal stories about their beliefs. Thus,
if we accept Lehrer's account, a causal account of the basing relation is


Causal theorists have gone in two different directions in response to such

cases. Some have admitted the existence of knowledge and proper basing after

investigation by the principals involved, and have attempted to save the heart of
the causal theory by giving up the claim that any actual causal relations are
required. Instead, they have attempted to salvage some causality in knowledge
by finding a true counterfactual involving a causal relation between evidence
and belief which is true and hence is compatible with these admissions, claiming
that this causality-embedded counterfactual is necessary for knowledge. 6 I have
argued against such theories elsewhere/ but will bypass the issues involved
here. For such emendations of a causal requirement on the basing relation do
not amount to a defense of a causal theory of basing. Instead, they agree with
Lehrer that a causal requirement is mistaken, so they pose no threat to the
conclusion Lehrer wishes to draw from his examples.
The other direction is to argue against the claim that the players in
Lehrer's examples come to have doxastically justified beliefs and hence against
the claim that they have knowledge. Robert Audi presents the only sustained
defense in the literature of this position, and his discussion focuses on the case
of the gypsy lawyer. He says,

Let us first consider some consequences of Lehrer's interpretation of the

example. Recall the assumption that the cards are not actually relevant to
p. Thus, even though S (here the gypsy) has (objectively) good evidence
for p, given a contrary verdict from the cards he would (other things equal)
have had the false belief that not-po Second, given his faith in the cards,
he would have believed p even if it had been false, indeed, even if, on the
basis of the cards, it had not been rendered so much as objectively likely
(to any degree) to be true, i.e., very roughly, likely to some degree given
the actual facts relevant to p... Third, S would have believed other
falsehoods about the crime, had the cards pointed to them, e.g., that the
client's spouse committed it. These points, especially the second, strongly
suggest that S does not justifiably believe p. It is, after all, simply good
fortune (because the cards happened to be right) that S did not believe
something false in place of p. Surely if one's belief that p is justified by
good evidence, it cannot be simply good fortune that one did not believe
something false instead. 8

Audi here worries about a certain sort of epistemic luck in believing a proposi-
tion. This sort of luck is present when one is indiscriminate with respect to the
truth. In the case of the lawyer, his reliance on the cards makes him
indiscriminate with respect to the claim that his client is innocent. As Audi
claims, "given his faith in the cards, he would have believed p even if it had been
false, indeed, even if, on the basis of the cards, it had not been rendered so much
as objectively likely (to any degree) to be true." This feature of the gypsy lawyer
case runs counter to what Audi sees as a reliable indicator connection es-
tablished by the presence of personal justification. That connection is that

personal justification implies that if the proposition in question were not true,
the person in question would not believe it. Audi thus claims that because the
lawyer's belief is just a matter of this sort of epistemic luck, or "good fortune"
as he puts it, it is wrong to think that the lawyer really knows, or justifiably
believes, that his client is innocent of the eighth murder.
Audi seems to recognize some weakness in this defense of the causal
requirement, however. Consider his formulation of the reliable indicator

[T]here is also an important (and far less widely recognized) connection

between personal justification and truth. The latter connection is our main
concern here. In the most common cases where Ss belief that p is justified
by good evidence, q, S would not have believed p if p were not true or at
least objectively likely to be true. For here S would not have believed p
ifhe had not believed q, in virtue of which-since q is (objectively) good
evidence for p-p is at least objectively likely. There is, I suggest, a
measure of protection from believing falsehoods which justification by
good evidence provides. 9

Audi claims that there is a connection between personal justification (what I

have been calling doxastic justification) and a measure of protection from
believing falsehoods. The protection arises from the reliable indicator
requirement, to the effect that if the claim had not been true, the person in
question would not have believed it. This protective clause is supposed to
support a rejection of Lehrer's account of his examples, but notice that Audi
does not give a complete endorsement of the clause. Rather, he claims that the
requirement holds "in the most common cases ... " A response on behalf of
Lehrer is easy if this qualifier is taken seriously, for gypsy lawyer cases need not
be all that common in order to undermine the causal view.
If it is granted that the reliable indicator connection only holds in the most
common cases, however, it is not clear how the reliable indicator connection can
form a strong enough reason for maintaining that there is an exceptionless causal
sustaining requirement. Yet that is just the structure of Audi's argument: he
holds that there is a reliable indicator connection which holds in the most
common cases and uses this information to conclude that there is an
exceptionless causal sustaining requirement. That argument is not telling
against Lehrer's view, for it would be at least as plausible to conclude from
Audi's premise that there is a causal sustenance element in the most common
cases of knowledge.
Furthermore, Audi's worries about the presence of luck are not telling,
either. It is well-known that it takes a cooperative environment for knowledge
to occur. Think, for example of Goldman's fake barn case, in which the
residents of a certain area undermine our claim to knowledge by planting a huge

number offake barns in a landscape in which we happen to focus on the one real
barn. Suppose, though, that they were only giddily considering the fun they'd
have doing so, but decide at the last minute not to play such ajoke on us. We'd
have knowledge in such a case, but be lucky to have it. Or suppose that they
decide to play the joke on us, but we serendipitously happen to take a different
road and happen to look at a different landscape with no fake barns on it.
Perhaps a black cat crossed the road, and our superstitions led us to turn to the
right rather than the left in order to avoid the horror of having the cat cross the
fork in the road that we traveled. Then we'd still have knowledge, but be lucky
to have it.
Audi might insist that these are not the kinds of luck that knowledge rules
out, whereas the kind he is speaking of is that kind. To defend such a position,
we would need an account of the distinction between these two kinds of luck,
and Audi recognizes that no such account will be forthcoming by appeal to the
reliable indicator connection (since it can only be said to hold in the most
common cases). In addition, progress on the dispute between Audi and Lehrer
is not going to be made by relying on some delineation of the various kinds of
luck that might infect belief. As I see it, the only way to determine what kind of
luck knowledge rules out is to forward an acceptable account of knowledge, and
then classify instances ofluck in terms of that account. That is, I see no grounds
for thinking that we could sort instances of luck into kinds apart from our
interest in the nature of knowledge, and have one of these kinds be just the right
kind to solve the Gettier problem. In the absence of the plausibility of this
approach to the Gettier problem, appeals to the concept of luck will end up as
unconvincing as relevant alternative responses to skepticism, which claim that
skepticism is false because one need only be immune from error in relevant
alternatives, among which are not found skeptical alternatives. For these
reasons, Audi's arguments based on the concepts ofluck and reliable indication
against Lehrer's examples must be determined to be unsuccessful.


In terms of persuading the reader to accept the alternative conclusion that

a causal theory of the basing relation is mistaken, the discussion to this point has
done little more than the presentation of the gypsy lawyer case itself would have
done. We have seen that the arguments against Lehrer's examples are
unconvincing, but it remains the case that this example has not moved many to
reject causal accounts. It seems, then, that some further argument would be
useful, and I believe there is a strong argument to buttress Lehrer's examples
from the general nature of justification. I say "general nature," because
justification is not only a property of beliefs but also of actions. It is plausible,

therefore, to assume that a proper account of justified beliefwill be extendable

to an account of justified behavior. I want to argue that a theoretically unified
approach to the concept of justification, one that attends not only to the nature
of justification as a feature of belief but also to its nature as a feature of action,
provides a strong argument against a causal view of the basing relation.
In the arena of action, the causal view maintains that there is a distinction
between reasons which only justify the action performed and reasons which
justify the performance of the action. On the causal view, if the reasons are
causally responsible (in the right way) for the action in question, then those
reasons justify the performance of the action as well as the action itself;
otherwise they can only justify the action itself. In this way, the causal account
of rational action claims to have explained the difference between a sort of
personal justification, where the performance of an action is justified, and a
more impersonal justification, where only the action performed, but not the
performing of it, is justified. This distinction mirrors the distinction we have in
the arena of belief between propositional justification and doxastic justification.
I believe this causal view is not adequate and I shall argue that it cannot
make sense of one feature of justification. Most of us discover, at one time or
another, that our motivations for behavior are less than desirable. I shall argue
that one response to this awareness, a response in which justification is created
where it did not exist before, is one for which the causal view cannot account.
Let us begin by considering a bit of behavior which is to be assumed to lack full
justification. Suppose Jim is running for Congress, where this behavior is to be
explained by an irrational desire to prove his critics wrong. This desire arose
because, being from the hill country in Texas, he believes that people make fun
of him and ridicule him because there is no possibility of his ever "amounting
to much." As a matter of fact, Jim has no reason to believe this and is just
showing paranoid tendencies, but he does hold the beliefs in question and
because he does, he forms the desire to prove his critics wrong. So, Jim is
running for Congress to show that he has amounted to something.
Regardless of how common or understandable such motivations are, they
present an explanation for Jim's actions which does not amount to ajustification
of them. Whatever reasons there might be which could justify Jim's running for
Congress, this irrational desire to prove his critics wrong (irrational because his
beliefthat there are such critics is unfounded) is not among them.
Further, suppose that Jim has not detected his real motivations and that
he has given reasons both to himself and others for running which have been
quite different from the "real" reasons for which he runs. But now, having
advanced in years and self-understanding, Jim has come to realize his true
motives. He has come to realize that the reasons he has given for running are
not what brings him to run for Congress.

In evaluating the plausibility of this causal view, I will be employing two

kinds of claims as a guide to which causal connections exist in a given case.
First, I will employ counterfactuals. One theory of causation is a counterfactual
theory, on which to say that one event or state is the cause of another is to say
(roughly) that, were the first event or state absent, the second would be absent
as well. 10 There are many problems with this theory of causation, II one ofwhich
is especially pertinent here. For, at most, the counterfactual theory of causation
is only adequate for deterministic causation, and is wholly implausible regarding
probabilistic causality. In order to address this concern, I will also look at the
issue of probability enhancement by proposed causal factors, in addition to the
counterfactuals to be examined. In each case, I do not presume that the concept
of causality can be analyzed or explicated in terms of these notions. I only
presuppose that the existence or lack of such is good evidence regarding the
existence of causal connections. In the case we are considering, we have
sufficient evidence for thinking that Jim's running for Congress is the result of
his irrational desire because, if Jim's irrational desire were absent, he would quit
the campaign. Perhaps he would become a beach bum instead.
When Jim's self-awareness increases, Jim comes to realize thatthe reasons
he has been offering for his behavior (i) did not originally prompt the behavior,
(ii) have not, in the past, sustained the behavior, and (iii) do not now sustain the
behavior. Regarding this third fact, what Jim realizes is that he is so constituted
at present that the reasons he has offered do not even enhance the probability of
his running, even if we were to control for the causal force of his irrational
desire. Upon confronting these rather disturbing facts, Jim then reasons as
follows: "the inadequate motivations both past and present are regrettable and
everything possible ought to be done to alter them; but, until this alteration can
be accomplished, everything possible ought to be done to maintain some
motivation or other to keep running for Congress since, after all, it is nonetheless
true that I am extraordinarily good at convincing others of correct policy, that
I am best qualified to serve the constituents of this district, and if persons were
to attempt to quit doing everything which is done for inadequate reasons, not
(as) much good would be done." So, Jim concludes, he ought to do all in his
power to keep the race for Congress alive in spite of his bad motivations.
There is an epistemic fact about Jim which needs to be explained by a
satisfactory account of justification. How to express this fact is a bit
troublesome, but we might try to put the point as follows. Jim has made
progress of a purely theoretical sort toward the goal of being perfectly rational,
of achieving the justificatory ideal. He has achieved a level of coherence
between thought and action which, though not ideal, is closer to the ideal than
what his prior self had achieved, first in ignorance of his true motivations and
next in a quandary about what to make of the tension between his behavior and

his awareness of his inadequate motivations. The fact which must be explained
is this progress.
The explanation I will argue for is that the progress achieved is that sort
attaching to the performance of the action in question which, when it obtains,
justifies the performance of that action. Of course, this explanation cannot be
employed by the causal theory, because the reasoning process above does not
have any causal impact on the particular behavior with which we are concerned.
Call the reasoning process in question "R". The occurrence of R is compatible
with the following truths: (a) if Jim did not have his irrational desire, he would
no longer be running for Congress; (b) even if R had not occurred, Jim would
still be running for Congress; and (c) if R had occurred and Jim's irrational
desire disappeared, Jim would no longer run for Congress. Moreover, if we
were to find a way to control for the effect of Jim's irrational desire, R would
have no probabilistic effect on Jim's behavior. Surely it is reasonable to assume
that R will affect his overall behavior at some point, for that is exactly what the
reasoning process shows Jim's intentions to be; but it need not have any im-
mediate causal impact on his running for Congress. We might capture this point
by calling the reasoning process a meta-motivational or meta-causal reasoning
process. If the reasoning process is a motivational one, i.e., if it is even
minimally causally efficacious with respect to the action in question, it would
be quite confused; for it includes the recognition that Jim cannot, simply by wil-
ling it or thinking about it, alter his motivation at present. I shall assume,
though, that Jim is not so confused.
What needs to be explained, to repeat, is the progress Jim makes in the
above case. What I shall argue is that any of the explanations open to a causal
theory do not sufficiently explain this progress, and thus that the causal theory
is shown to be inadequate by cases involving meta-motivational reasoning
processes. In order to defend these claims, we need some understanding of the
components of the justificatory ideal so we will be able to ascertain the variety
of alternative explanations open to the causal theory. This ideal requires, first,
that all actions, beliefs, desires, intentions, etc., be fully justified, i.e., all actions
be justified ones to perform, and all mental states be justified states in which to
be. Further, it requires that one justifiably perform all of one's actions, and
justifiably be in all the relevant mental states in question. Finally, this ideal re-
quires that the person in question have certain character traits which I shall refer
to loosely using the notion of being fully rational. The first two requirements
of the ideal relate, first, to the actions and states themselves, and then to the
relation between the person and the actions and states. This final requirement
relates to the person alone: he/she must be fully, or ideally, rational as well. So,
we are looking for an explanation in one of these three areas for the progress that
Jim has made in reconciling his running with his inadequate motivations.

Consider first the implications of the causal theory which give no

explanation of Jim's progress. The causal theory denies that Jim's running is
justified either before or after his discovery and resolution of his real motives,
i.e., prior to Jim's discovery his running is not justified, after discovering his
motivations and prior to his formulation of R his running is not justified, and
after formulating R his running is not justified. In what follows, I will refer to
these three stages of this case as stages one, two, and three, respectively. A
causal theorist may say that there is an impersonal justification in all three stages
for the action which Jim performs, but he must deny at each stage that Jim's
performing ofthat action is justified.
As we have just seen, this answer can be maintained only if the causal
theory can also explain the progress Jim makes toward the justificatory ideal in
some other way than by claiming he comes to have a justification for the
performance of an action where he previously only had a justification for the
content ofthe action. We can categorize the options which a causal theory can
appeal to by which to explain the movement toward the ideal as follows. First,
the theory can appeal either to some present difference in Jim or to a difference
there will be in the future. If the appeal is to some present difference in Jim, the
options are several: (i) the appeal may be to some internal mental state of Jim
which becomes (more) justified, (ii) the appeal may be to certain acts or
omissions, or aspects of certain acts or omissions, which undergo an increase in
their level of justification, or (iii) the appeal may be that Jim, himself becomes
more rational.
As far as I can tell, there are no other options to which a causal theorist
can appeal. In order to show that the causal theory is inadequate, what needs to
be done is to show that Jim's progress is compatible with no increases of the sort
described in the last paragraph to which a causal theorist might appeal. I shall
begin this extended argument to eliminate the available options by considering
the appeal to features of the future. We shall see that this option is the least
attractive, and thus we will devote the major portion of this elimination
argument to features of the present and features of Jim himself to which a causal
theorist might appeal in explaining Jim's progress.


The causal theorist might attempt to account for the case by claiming that
the only difference is a future difference. So, for example, the causal theorist
might hold that, given R, it is now true that in the future, Jim will justifiably run
for Congress (perhaps when he comes up for re-election); whereas withoutR, he
will not. On this option, no present difference exists except for the fact that it
is now true that the future will be different than it otherwise would have been.

This claim seems straightforwardly wrong, however. There is clearly a

present difference between stages one and three, and any view which cannot
explain that difference is inadequate. Once we have recognized our inadequate
motivations and seen that it is better that our behavior not change, we are closer
to the justificatory ideal, at that instant, than we were before. Were we to die
in the next instant, our legacy should include our having made such an advance
during the final moments of our lives. If the above explanation of the difference
is all that the causal theorist can offer, all he can say is that we would have made
such an advance had we not expired. Hence, the causal theorist must look at
features of the present for an account of Jim's progress.


A more promising attempt looks for certain mental states to which a

causal theorist might appeal to explain Jim's progress, mental states which in
virtue of their enhanced justificatory state can explain Jim's progress toward
ideal rationality. The answer to the question whether there are such mental
states is, I think, no. First, the progress is not merely a matter of increased
self-awareness. Jim is no more aware of himself after reconciling his inadequate
motivations with his continued playing than he was before that reconciliation,
i.e., no increase in self-awareness occurs between stages two and three. Yet,
progress is clearly achieved between stages two and three, so we cannot explain
his progress by appeal to some increased self-awareness.
Also, we cannot explain the progress made by an appeal to an increase in
the rationality of any of Jim's beliefs or desires. First, consider his desire to run
for Congress. If Jim's desire is to run in order to show his critics wrong, this
desire has the same irrational status both before and after constructing R. On the
other hand, it might be claimed that Jim has the desire to run, after constructing
R, in order to use his abilities for his constituents, thus making the desire to run
a rational one where no such desire was rational before constructing R. The
problem with this view is that Jim need not have any such desire. He may only
intend to come to have that sort of desire, knowing (regrettably) that his only
present desire is to show his critics wrong. So, in either case, the causal theory
cannot be rescued by appeal to Jim's desire to run for Congress.
An appeal to Jim's desire or intention to alter his motivational structure
is of no use either. Jim could have added this desire or intention (and had either
be justified) prior to constructing R and hence prior to resolving his motivations
with the continuation of his campaign. In other words, such a desire or intention
could have been added before stage three arises, and hence before the progress
which needs to be explained had been achieved. In fact, if Jim is like the rest of
us, he probably added this desire or intention in stage two prior to resolving his

behavior with his motivations. And yet his distinctive progress does not occur
until stage three. Hence, neither of these internal states can be the only
explanation of the difference in Jim's state before constructing R and after con-
structing R.
The final internal state to which a causal theorist might appeal is Jim's
belief that he should (continue to) run for Congress. A causal theorist might
claim that, before constructing R, this belief was not doxastically justified, for
it was sustained by Jim's desire to prove his critics wrong; after constructing R,
R comes (at least in part) to sustain that belief. Hence, the progress Jim makes
is to be accounted for by noting that, before constructing R, Jim may have had
ajustification for thinking that he should continue to run, but he did notjustifi-
ably believe that claim. After constructing R, he comes to justifiably believe
that he should continue the campaign.
The difficulty with this attempt is that there is no reason to think that R
must sustain in part the belief in question. The irrational desire to prove his
critics wrong may be responsible both for his behavior and for his belief that he
should continue the campaign. Jim may know this sad fact about himself, and
try to sever the causal connection between his desire and this belief in addition
to trying to sever the causal connection between his desire and his running for
Congress. Nonetheless, it is still intuitively obvious that Jim has made progress;
thus, appeal to the belief in question will not explain this progress.
Moreover, there is something unsatisfying about the general approach
suggested here, that Jim's progress can be explained by citing some additional
mental states that are justified. Mere numbers of unrelated beliefs, desires,
intentions, or other mental states would add to Jim's overall epistemic condition,
but would not explain the unique kind of progress Jim has made. Furthermore,
the mere addition of further justified mental states about the particular issue of
running for progress won't explain his progress either. For note that such
changes are bound to occur between stages one and two, for the simple reason
that new experience can be counted on to provide additional evidence for new
mental attitudes. Yet, the sense of progress that Jim makes between stages two
and three is simply not there in the transition from stage one, where he is
unaware of his inadequate motivations, to stage two, where he becomes aware
of them. With this new awareness comes a host of new mental attitudes about
himself and his situation, attitudes which we may presume to be justified. Some
progress in terms ofjustification or rationality has been achieved because of this
change, but it is not the distinctive kind achieved in the transition from stage two
to stage three. This fact suggests that it is not in virtue of addingj Llstified mental
states that explains the progress Jim has made.
So, it would seem, appeal to internal states cannot salvage the causal
theory from the case of the ill-motivated politician. Let us turn then to a
different area to see if it offers more hope for the causal theory, for if internal

states cannot explain what needs to be explained, perhaps external features, i.e.,
that which counts as overt behavior by Jim, can.


These external features include the acts, omissions, and aspects of each
which characterize or might characterize Jim at present. One such external
feature which cannot be of any use to the causal theorist is the act of running for
Congress itself. That act had a justification for it before Jim discovered his
motivations for performing it, and presumably the act itself is also justified after
Jim constructs R. Further, the causal theorist cannot hold that Jim justifiably
performs the act either before or after constructing R, for in neither stage does
that which justifies the act causally sustain his so acting.
A causal theorist might claim, though, that the act in question comes to
have a greater justification, or perhaps a justification all things considered (of
which Jim is aware), after constructing R. The appeal to a greater justification,
though, does not explain the advance toward the justificatory ideal. For one can
acquire a greater justification for believing, e.g., that all ravens are black just by
seeing another black raven, without making any such advance. The intuitively
obvious point about Jim is that he has made that sort of progress, and since
greater justification can occur without such progress, citing it in this case does
not explain that progress. Nor does the appeal to justification all things
considered explain the progress. The only way for this appeal to explain the
progress would be for the act to fail to be justified, all things considered, before
Jim undergoes the reasoning process in question, for otherwise there would be
no difference (on these grounds) between the first and second stages. Yet, when
Jim was not aware of his poor motivations for running, he had ajustification, all
things considered, for his campaign; so if we are to accept this explanation, we
must hold that after finding out about his poor motivations, Jim loses this
justification, and then regains it after engaging in the reasoning process
described. This explanation is inadequate precisely because the process from
lack of awareness to reconciliation through the reasoning process involves
progress toward an ideal, whereas the proposed explanation leaves Jim at the
same level after the total process as before. Hence, this attempt fails to free the
causal theory of the counterexample of the ill-motivated politician.
Could the causal theorist claim that Jim has greater justification than he
had before for running for Congress? I think the answer is no. On the proposal
above, Jim loses justification for running when he becomes aware of his
inadequate motivations. This information defeats whatever justified his running
in the first place. The force of the reasoning process is, then, to override the
defeating information. When such overriding occurs, the force of the original

justification is reinstated, but it is not enhanced. For enhancement to occur,

some other explanation of the stages would have to be forthcoming.
Perhaps a causal theorist might appeal to however Jim is proceeding in
attempting to alter his motivational structure. Perhaps this act is justified and
can explain Jim's progress.
It cannot. Jim may be doing nothing at present to even attempt to alter his
motivational structure. His intention may be simply to take any actions he can
find to alter it, but he may not have found any as yet.
Since there are no other actions which Jim must be performing at the time
in question, perhaps a causal theorist will appeal to acts which Jim omits to
perform to explain his advance in rationality. Perhaps one might appeal to Jim's
not taking steps that will alter his desires and cause him to quit the
Congressional race, or to Jim's not looking for acts to undermine the effects of
his inadequate motivational structure.
Any such explanation is adequate only if it is the omitting which is
justified, and not just the content of the omission. If we suppose that only the
omission (and not Jim's omitting) is justified, no progress will have been
explained. Rather, such instances present an even greater need for progress
toward the ideal in question because, in the case at hand, if only the omission
were justified, we would have a case in which an omission occurs but the
omitting is not justified. Such an explanation would add to Jim's problems
regarding justification, not explain how he is eliminating them.
So, any omission cited needs to be justifiably omitted. Regarding steps
which would alter his desires and cause him to quit the race, it is hard to see why
one would think Jim justifiably omits such steps. Perhaps ifhe had some steps
in mind, he could justifiably omit them; but Jim may have no idea of how to go
about altering his desires and thereby cause himself to quit the race. There
simply is no good reason to suppose that Jim justifiably omits such actions.
This argument does not affect Jim's omitting to look for such steps. Jim's
not looking may be justified, but it is not clear that this fact helps the causal
theory. When an absence of action is justified, at least in certain types of cases,
such justification obtains only because some other action (which is not an
omission) is justified. The sorts of cases in which this is so are cases where the
action, the omission of which is justified, has a primafacie justification for it.
F or example, if a doctor's failing to stop and help an accident victim is justified,
it is so justified only in virtue of some other action overriding the importance of
the first (such as being needed at a more serious operation at the hospital). In
Jim's case, looking for a way to alter his motivational structure isprimafacie
justified--at least in part because it is either a part of, or inextricably linked to,
Jim's performing an action which will alter his motivational structure (an action
which Jim correctly believes to be prima facie justified). So to say that Jim
justifiably fails to look for such an action cannot be the end of the story. For to

say that an omission of a prima facie justified act is justified requires the
justification of some other action which implies that the primafacie justification
for the omission is overridden.
Thus, the difference in levels of rationality before and after having
reconciled his bad motivations with continuing to run cannot be explained
merely by claiming that certain omissions are justified. That may be true, but
if it is, it is only true in virtue of Jim's justifiably doing something else. One
obvious option here is that it is justified in virtue of its understood relationship
to Jim's running for Congress in spite of his inadequate motivations, but that
explanation is not open to the causal theory. For it would first have to be
granted that Jim s running is itself personally justified since its impersonal

justification is no different from stage one to stage three. Without appeal to this
action and its justification, however, it is hard to see where to find an action
whose justification confers justification on the omission in question. So we
must conclude that this attempt at an explanation is only as good as some other
one as yet forthcoming.
The only remaining alternative is to claim that some of Jim's actions, or
some aspects of his action, are justified while some others are not. The causal
theorist might hold that Jim's running is not justified, though his attempting to
help his constituents is. Or he might claim that Jim's running is not justified
though his acting so as to alter his motivational structure is.
This approach does not work either. For the aspects of the action, or the
different acts (however one chooses to individuate actions), are inextricably
linked. Perhaps some of the elements are primafacie unjustified whereas others
are prima facie justified. Given the inextricable linkage that occurs, however,
none of the elements can achieve actualjustificatory status without all the others
achieving the same status. There may be possible circumstances in which the
linkage is dissolved so that, say, Jim's running is not justified and yet his
attempting to help his constituents is. But it would be quite regrettable if this
possibility were taken to imply that the justificatory status actually diverges. For
the only way to make sense of the transition from prima facie justification to
actual justification is with reference to the entire set of elements of the actual
circumstances--to say that a prima facie justified action is actually justified is
to say that there is nothing else in the actual circumstances that overrides the
primafacie justification. And to say that a prima facie unjustified action is not
on the whole justified is to say that no other action has a primafacie justification
strong enough to override the prima facie lack of justification in such a way as
to justify the second action. But that is equivalent to requiring that the jus-
tificatory status of the elements stand or fall together, given that they are all
parts of the same situation. Thus, the causal theorist cannot hold that Jim's
playing is not justified whereas his using his talents is, and hence we must look

elsewhere if the causal theory is to escape the case of the ill-motivated



The final option open to the causal theorist is to claim that, whereas no
progress is made in the areas considered above, progress is made in that Jim,
himself is more rational after constructing R than before. It is, on this option,
progress regarding the rationality of the person in question (rather than his
(present or future) mental states, acts or omissions) which explains Jim's
In order to evaluate this attempt to rescue a causal theory, we must
consider what it is for a person to be rational. When we claim that a person is
rational, there are two things we might be claiming: first, we may be saying
something about the collection of actual beliefs, actions, etc., of the person and
noting that the collection is constituted by a sufficiently high degree of rational
beliefs, actions, etc., to warrant calling the person a rational one; or,
alternatively, we may be claiming that a positive evaluation applies to the
character of that person. The first option is ruled out by the above discussion
of the internal and external features of the case ofthe ill-motivated politician,
for all the same points can be made about the presence of rational belief and
action as were made about the presence of justified belief. So let us concentrate
on the second option. On it, to say that a person is rational is to say something
about the way in which that person determines what to do and how to do it, what
to believe, what and how to change, or what and how not to change. In other
words, we are saying something about the dispositions of the person in question
to proceed rationally or justifiably in forming and holding beliefs, choosing
actions, etc.
To return to the case of Jim, our ill-motivated politician, the causal theory
is claiming that Jim has better dispositions with regard to his actions, beliefs,
desires, etc., after constructing R than he had before constructing R. There is
something to be said in favor of the view that, through the process of
self-discovery, Jim himself becomes more rational. Perhaps prior to learning of
his poor motivations, he was disposed to form beliefs and perform actions in line
with poor motivations; after becoming aware of his poor motivations, he is less
inclined to do so. After learning that his irrational desire can prompt his
behavior, perhaps he is less likely to form a belief or perform an action when all
it has in its favor is that it satisfies such an irrational desire.
This difference is not sufficient for the causal theorist. In the case of the
politician, we have three separate stages: the first stage is where he is unaware
of his poor motivations, the second is where he becomes aware of his poor

motivations, and the third stage involves his reconciliation of his poor
motivations with his continuing to play professionally. Any advance in the
rationality of Jim's character, as clarified in the last paragraph, occurs in the
second stage, for his awareness of his poor motivations and the role his irrational
desire can play in his behavior have already been perceived prior to the
reconciliation in question. The progress Jim makes for which we are seeking an
explanation, however, occurs most obviously at stage three. Hence, this
explanation fails to account for the data.
It may be thought that Jim adds an intention in the third stage, so that this
stage involves two distinctive elements: first, the reconciliation about the
particular bit of behavior under question and second, an intention to alter his
motivational structure if he can. It might then be claimed that the progress to be
explained is that the additional intention is justified, and given its presence Jim
is both better off now and will be better off in the future.
The intention in question could just as easily have been added during the
second stage, however, and if it were added at that point, it would still be the
case that Jim makes progress toward ideal rationality in the third stage. Thus,
the case does not depend on the additional intention at the third stage. Further,
even if the added intention were central to the third stage, that would not explain
the progress in question, for, as we saw earlier, appeals to desires, intentions,
and increased self-awareness on their own do not explain the distinctive advance
characteristic of stage three, as opposed to stage two, which involves increased
self-awareness and new mental attitudes which can be presumed to bejustified.
It appears, then, that the causal theorist has no resources whatsoever by which
to explain the progress Jim makes toward ideal rationality.
It is important to notice that none of my arguments against the causal view
here presupposes a deterministic theory of causation. These arguments work
just as well against the proposal that Jim's reasoning provides some degree of
causal support, however minute. It is just as possible for this reasoning to occur
and have no probabilistic causal impact on his behavior as it is for it to occur
without having a deterministic impact. A denial of this possibility could be
sustained only by arguing that reasons must always be causes, but such a claim
wreaks havoc with the idea that Jim's running failed initially to be personally
justified. For in stage one, Jim has reasons for what he is doing (he gives these
reasons to himself and to others when asked), and hence if reasons are always
causally active in some way, these reasons would have to be partial causes even
in stage one. Such a maneuver would make a defense of the causal view even
more difficult than it is on the assumption that reasons need not always be
causes, for if reasons are always causes, then the causal view has no interesting
story to tell to distinguish doxastic from propositional justification.
No matter how causal theorists wish to view such a problem, however, the
point remains that nothing argued here presumes a non-probabilistic account of

causation. So there is nowhere for a causal theorist to turn for rescue from the
case of the ill-motivated politician.


The most natural account ofthis case is just this: before constructing R,
Jim did not possess the kind of rationality which implies that he is justifiably
running for Congress; but after constructing R, his running is rational in a way
which implies that it is perfectly justified. The advance Jim makes is to be
explained by his moving from performing a justified action to justifiably
performing that action, from performing an impersonally justified action to
performing a personally justified one.
Such a conclusion gives us an account of the nature of justification that
covers the variety of things to which it applies and one which fits well with
Lehrer's intuitions about the gypsy lawyer and the racist scientist. The only
relevant difference between the racist scientist and the ill-motivated politician,
in this view, is that the scientist has no remorse for her racism whereas the
politician has regrets for his motivational structure. In that way, the politician
is better offthan the scientist. Still, the relevant beliefs and actions are on par,
being justified by the reasons available to each person, contrary to the demands
of the causal account of basing.
One may worry here that in accepting the non-causal account of the case
given above, we have lost the obvious point that there is something lacking
about Jim and his relation to his Congressional campaign. That is not so. We
can readily grant that Jim has not achieved ideal rationality, even if it turns out
that the lack of rationality in his running for Congress is the last vestige of ir-
rationality or lack of justification remaining among any of his acts, beliefs,
intentions, desires, valuations, etc. In order to be ideally rational, Jim must have
a proper character; and having a proper character requires being disposed toward
rationality and justification in the arenas of action, belief, desire, intention,
valuation, etc. Resolving his inadequate motivations with his continued running
surely does not eliminate the character flaw he possesses, even if this event is
part of the process toward character perfection.
I close with one last word of speculation on the attractions of the causal
theory. Perhaps the more fundamental kinds of knowledge, such as perceptual
knowledge, are kinds for which meta-motivational reasoning processes such as
those discussed here are not possible. Perhaps, that is, perceptual beliefs rely
essentially for their justification on a causal element in perception, and no
circuitous route through a meta-motivational reasoning process could make up
for a faulty causal story about such beliefs. Given the central place that
perceptual knowledge plays in the construction and defense of theories of

knowledge, it would be understandable for theorists to generalize from what may

be true of such fundamental kinds of knowledge to all knowledge of any type.
It would still be a mistake, but an understandable one.

I. This essay is written in honor of Lehrer's luminary career in philosophy, out of deep respect for

his intellectual achievements and gratitude for his acquaintance and gracious assistance throughout
my career. It is a privilege to know him, and I dedicate this essay to a splendid human being and
brilliant philosopher.
2 Keith Lehrer, Knowledge, (Oxford: Oxford University Press. 1974), pp. 124-125; "Proper
Function versus Systematic Coherence," in Jonathan L. Kvanvig, ed., Warrant in Contemporary
Epistemology, (Lanham, Maryland: Rowman & Littlefield Publishers, Inc., 1996), pp. 25-46.
3 The common attitude is represented in discussion of the issue by John Pollock, who footnotes
his claim that the basing relation is in part a causal relation as follows: "Lehrer has argued against
this, but I do not find his counterexample persuasive," (Contemporary Theories of Knowledge,
(Totowa, NJ: Rowman and Littlefield, 1986), p. 81). He offers no argument, supplies no
4 Keith Lehrer, Knowledge, pp. 124-125.
5 Keith Lehrer, 'Proper Function versus Systematic Coherence,' pp. 33-34.
6 See, for example, Marshall Swain, Reasons and Knowledge, (Ithaca: Cornell University Press,
1981), chapter 3.
7 See "Swain on the Basing Relation," Analysis, Vol. 45, No.3, June 1985, pp. 153-158. Lory
Lemke criticizes my arguments in "Kvanvig and Swain on the Basing Relation," Analysis, Vol. 46,
No.3, June 1986, pp. 138-144; I reply to his objections in "On Lemke's Defense of a Causal
Basing Requirement," Analysis, Vol. 47, No.3, June 1987, pp. 162-167.
8 Robert Audi, "The Causal Structure of Indirect Justification," Journal of Philosophy, 1983, p.
9 ibid., p. 407.
10 For a developed defense of the counterfactual theory of causation, see David Lewis,
Countelfactuals, (Oxford: Basil Blackwell, 1973).
II See, e.g., Jaegwon Kim, "Causes and Counterfactuals," in Ernest Sosa, ed., Causation and
Conditionals, (Oxford: Oxford University Press, 1975), pp. 192-195.
Chapter 4


T odd Stewart
University ofArizona

Keith Lehrer has argued that it is possible to have knowledge of a

proposition despite the fact that one's belief in that proposition is causally
unrelated to one's epistemic reasons. This is an interesting and contentious
claim, but an assessment of it is not the task of this paper. Instead, I will argue
that Lehrer, and perhaps one critic as well, both take Lehrer's theory to entail
that knowledge is possible without causation by reasons, but that this is
mistaken. Lehrer's theory as it stands actually delivers no verdict on the issue
of whether proper causation is necessary for knowledge or a belief's being
justified. While this means that the coherence of Lehrer's theory can be saved
should his treatment of these issues yield counterintuitive results, it also means
that the theory as it currently stands is incomplete. The theory simply does not
make any predictions about certain kinds of examples, and this reveals that
Lehrer's theory is unfinished in an important respect.
Let us turn to the racist Mr. Raco, which is where Lehrer asserts that an
agent can be completely justified and yet that agent's reasons need play no
causal role in the explanation ofthe belief that is completely justified. Here is
Lehrer's statement ofthe case:

It is easy to imagine the case of someone who comes to believe

something for the wrong reason and, consequently, cannot be said
to be justified in his belief, but who, as a result of his belief,
uncovers some evidence which completely justifies his belief.
Suppose that a man, Mr. Raco, is racially prejudiced and, as a
result, believes that the members of some race are sllsceptible to
some disease to which members of his race are not susceptible.
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 63-74.
© 2003 Kluwer Academic Publishers.

This belief, we may imagine, is an unshakable conviction. It is so

strong a conviction that no evidence to the contrary would weaken
his prejudiced conviction, and no evidence in favor would
strengthen it. Now imagine that Mr. Raco becomes a doctor and
begins to study the disease in question. Imagine that he reads all
that is known about the disease and discovers that the evidence,
which is quite conclusive, confirms his conviction. The scientific
evidence shows that only members of the race in question are
susceptible to the disease. We may imagine as well that Mr. Raco
has become a medical expert perfectly capable of understanding
the canons of scientific evidence, though, unfortunately, he
becomes no less prejudiced as a result of this. Nevertheless, he
understands and appreciates the evidence as well as any medical
expert and, as a result, has reason for his belief that justifies it. He
has discovered that his conviction is confirmed by the scientific
evidence. He knows that only members of the other race are
susceptible to the disease in question. Yet, the reasons that justify
him in this belief do not causally explain the belief. The belief is
the result of prejudice, not reason, but it is confirmed by reason
which provides the justification for the belief. Prejudice gives Mr.
Raco conviction, but reason gives him justification. I

I find the case a bit underdescribed by Lehrer. Two critical pieces of the
example are left unstated: (1) whether or not Mr. Raco, when asked about his
belief regarding the susceptibility of a particular minority to some disorder, lists
off the relevant scientific studies as his reasons, and (2) if Mr. Raco himself
takes these scientific studies to be his reasons of ifhe simply gives these reasons
to others to stop them from attacking his racism. I think that the case is meant
to be one in which Mr. Raco does in fact respond to challenges by citing the
scientific studies and in which he does sincerely take his reasons to be those
studies. Mr. Raco does not realize that the root of his belief is his racism. What
Lehrer intends is to put forward a case where the agent honestly takes his
reasons to be ones which are not the actual cause of the belief or acceptance?
The important thing to notice about the example is that Lehrer does
clearly claim that Mr. Raco knows and is completely justified in believing that
members of the race in question are susceptible to the disease. A brief sketch
of the relevant pieces of Lehrer's theory is warranted at this point. Knowledge
according to Lehrer is, roughly, completely justified true belief (acceptance).3
A belief is completely justified when it is both personally and verifically
justified. 4 Lehrer uses complete justification as a technical term, so it is
important that he explicitly claims that Mr. Raco is completely justified. This
means that Lehrer thinks that Mr. Raco is both personally and verifically

justified according to his theory. Lehrer's analyses of personal and verific

justification will be discussed below when we turn to whether or not Mr. Raco
is in fact completely justified in his racist belief according to the theory as it
Dirk Koppelberg has criticized Lehrer's claim that appropriate causation
is not necessary for justification and knowledge. While Koppelberg never
explicitly claims that Lehrer's own theory has this consequence, he does say
some things that make it plausible that he thinks this. For example, he opens his
"Justification and Causation" with the following:

Coherence theorists subscribe to the thesis that epistemic

justification consists in appropriately specified inferential relations
among beliefs. In contrast many reliabilists hold that it is the
causal origin or the causal sustenance of a belief which is
responsible for epistemic justification ... Keith Lehrer and Thomas
Bartelborth have vigorously argued that any theory ofjustification
which allows for causal considerations is deeply mistaken. 5

From this, we can infer that since Lehrer is a coherence theorist and further
thinks that any correct theory excludes causal factors as irrelevant to whether or
not a belief is justified, Lehrer's own theory must exclude these causal factors,
since otherwise Lehrer's own theory would (a) not be a coherence theory, and
(b) be incorrect. This is such an obvious inference that I suspect that
Koppelberg would endorse it. Again he never actually addresses the issue of
whether or not Mr. Raco is justified according to Lehrer's theory since he is
more interested in showing that Mr. Raco is not justified according to the
various notions of justification that Koppelberg thinks to be correct. 6
Koppelberg ends his paper by making similar general remarks about
causation and coherence theories. He writes, "A genuine coherence theory
claims that epistemic justification consists in adequately specified inferential
relations among belief contents--causal relations among belief states do not
matter.,,7 Again, the inference to the claim that Lehrer's theory, insofar as it is
a coherence theory, will entail that a belief may be justified even if it is not
appropriately related to its justifying reasons is obvious. So, I think it fair to
claim that Koppelberg would agree that Lehrer's theory is committed to the view
that appropriate causation is not necessary for a belief's being justified. Of
course, Koppelberg would use this to claim that Lehrer's theory is false, but this
is not important at present.
Interestingly enough, the arguments later in this paper will show that
Koppelberg (and perhaps Lehrer as well if he sees coherentism as being
logically incompatible with giving a role to causation) is simply incorrect to
suppose that coherence theories must by their nature exclude the causal origin

or sustenance of belief from consideration. For, while Koppelberg is correct that

coherentism considers only inferential relations between beliefs, if the beliefs
themselves have partially causal content, the inferential relations themselves will
be impacted by this. While it is still presumably true that the facts of causation
themselves do not matter when it comes to being personally justified (as I will
explain below), when we add whatever Gettier conditions, etc., are necessary in
order to create an account of knowledge, the facts about the causation of a belief
may become relevant. And we should always remember that Lehrer is
ultimately interested in an analysis of knowledge, and hence Gettier conditions
are an important component of his theory. In his theory, verific justification
plays the role of bridging the gap between personal justification and knowledge.
And, as I hope to show, verific justification leaves room for appropriate
causation as a necessary condition for knowledge, as it is a possibility that
claims about what is a reason for what have a causal component built into their
content. 8 So, while it is perhaps true that bare coherence theories do not make
room for causation, when we add whatever is necessary to these theories for an
account of knowledge, causation may become relevant once again.
Now that we have seen that both Lehrer and a critic take Mr. Raco to be
completely justified according to Lehrer's theory, we must assess whether this
is correct. Recall that, according to Lehrer, a belief is completely justified if and
only if that belief is both personally and verifically justified. Let me now briefly
sketch Lehrer's basic theory of personal justification, and explain why Lehrer's
theory says that Mr. Raco is personally justified (assuming that he meets all of
Lehrer's other conditions). According to Lehrer's final account of personal

S is personally justified in accepting that p at t if and only if

everything that competes with p for S on the basis of the
acceptance system of Sat t is beaten or neutralized on the basis of
the acceptance system of Sat t. 9

A proposition C competes with p for an agent (given her acceptance system) if

it is less reasonable for that agent to accept p on the assumption that c is true
than to accept that p on the assumption that c is false.1O In other words, c
competes with p iff Rs(P/c) < Rs(P/~c).1l Proposition p beats c for S iff c
competes with p but it is more reasonable for S to accept that p than to accept
that c on the basis of the agent's acceptance system (all at a time). Again, we
might express this as:p beats c iffRs(P/c) <Rs(P/~c) and Rs(P) > Rs(c). Finally,
n neutralizes c as a competitor of p for S iff c competes with p for S, but the
conjunction of nand c does not compete with p for S, and it is as reasonable for
S to accept this conjunction as to accept c alone according to S's acceptance

system. This can be represented as: n neutralizes c as a competitor of p iff

(Rs(P/c) < Rs(P/~c) and Rs(P/(c !\ n)) ~ Rs(P/~(c !\ n)) and Rs(c !\ n) ~ RS(C).12
In the case at hand, Mr. Raco does meet all of these conditions. In terms
of the scientific evidence which Mr. Raco believes, all competitors to Mr.
Raco's beliefs are beaten or neutralized. For example, if the skeptic suggests (as
per the justification game) that one of the scientific studies could be
inappropriate, Mr. Raco can respond that given his knowledge of how the
studies were conducted, etc., it is more reasonable for him to believe that the
studies are fair and accurate (this competitor is hence beaten). But, suppose the
skeptic offers the following competitor:

c. You (Mr. Raco) are not causally influenced in your belief by

the scientific evidence.

Mr. Raco can neutralize this competitor by claiming something like the

n. The scientific evidence I possess shows that what I believe

is true, and it really doesn't matter whether or not the
possession of the evidence in this case causally influences
my belief. I am concerned in believing what is true about
who suffers from this disease, so why should I care about
the causation of my belief?

Proposition n can neutralize c, assuming that c does in fact compete with the
racist belief, because the conjunction of c and n does not compete with p given
Mr. Raco's beliefs and it is at least as reasonable for Mr. Raco to accept this
conjunction as to accept c alone given his acceptance system.13 For, we can
stipulate that Mr. Raco does in fact accept various things that do appropriately
support n. Hence, c is neutralized and Mr. Raco is personally justified in his
belief despite the evidentially-illicit causation of his belief.
So far, we have seen that Lehrer's theory does entail that Mr. Raco is
personally justified. But, is Mr. Raco verifically justified? Unfortunately, I
think that Lehrer's theory as it stands does not deliver any verdict on this issue.
Further, this failure can be traced to the fact that Lehrer's theory actually takes
no stand at all on whether appropriate causation by reasons is necessary for a
belief to be completely justified or an instance of knowledge. Here we should
separate Lehrer from his theory; clearly, Lehrer thinks that causation is not
required for justification. But, surprisingly, Lehrer's theory as it stands does not
commit itself on this question.
Let us now turn to the argument for this claim. An agent's verific
justification system is what remains of her personal j ustification system but with

all false beliefs deleted. If belief in a given proposition can be justified on the
basis of what remains (and these propositions can all be verifically justified as
well, and so on until a special kind of circle emerges), then that belief is
verifically justified. Thus, if we wanted to show that Mr. Raco is not verifically
justified in his racist belief, we would need to locate false beliefs that will be
used in support of the belief in question (either directly or in the handling of
competitors suggested by the skeptic). And, this is the heart of the problem:
Lehrer needs to explicitly claim that certain propositions are true or false in
order for his theory to entail that Mr. Raco has or lacks verific justification. But,
Lehrer's theory as it stands simply does not do this. Because of this, it is
impossible to say whether Mr. Raco's racist belief is verifically justified
according to the theory.
To see this, suppose that the skeptic makes the following claim during the
ultra-justification game: 14

c. Mr. Raco, the reasons you give for your belief that the
minority suffers from a disease are good reasons but they
are not your reasons because they have no causal influence
on your belief. As this is the case, despite the fact that they
are good reasons, since they are not your reasons, they do
not justify your bel ief.

Before moving to Mr. Raco's reply to competitor c, a brief discussion of a

reason vs. one's own reason is in order. I think that there is a fairly intuitive
difference between the two that can be brought out by considering cases where
one has lied about why one believes something but has nonetheless provided a
good reason for the belief. Suppose that Wendy is trying to convince her friend
Greg that a certain mathematical theorem is true. Knowing that Greg is a bit
slow when it comes to math, she claims that Mr. Mathright (their high-school
calculus teacher who was renowned for his care, professionalism, and precision)
said the theorem was true. Actually, Wendy holds the theorem to be true
because she proved it herself one afternoon in a fit of inspiration. IS Mr.
Mathright did say that the theorem is true. But, Wendy is a bit of a skeptic about
trusting others; she refuses to take anyone's authority as a justification for a
mathematical claim until she can establish it for herself. Awareness of Mr.
Mathright's assertions and reputation alone would suffice to justify most
people's acceptance of the theorem. 16 So, as it happens Wendy has offered a
good reason for accepting the theorem to be true. But, it is not her reason. She
lied to Greg to simplify matters. Wendy believes the theorem on the basis of her
When one gives one's own reason, one is tacitly making a claim about the
causation of one's beliefs; this explains how it is possible to lie about one's

reasons while still giving what is a good reason. Causation is actually part of the
content of certain reasons claims, and hence relevant to their truth or falsity. So,
when Wendy claims that she accepts a certain mathematical proposition on the
basis of Mr. Mathright's testimony, a necessary condition for the truth of her
claim is that Mr. Mathright's testimony itself or her belief that there was such
testimony was in some part causally responsible for producing or sustaining the
belief in the mathematical proposition. Because this causal component is false
in Wendy's case (and further we may assume that Wendy believes it to be false
as well - this must be added so that Wendy counts as actually having lied),
Wendy has lied about her reasons for accepting the mathematical proposition. I?
This, I hope, helps to make sense of c above; causation can and does enter the
picture where one's own reasons are concerned. The skeptic then claims that
only one's own reasons are relevant to justification. This is the skeptic's
objection to Mr. Raco.
So, what can Mr. Raco say in response to c? He could try to beat this
competitor by saying the following:

b. Justifying reasons need not be my own reasons (in the sense

just clarified). The reasons I just gave are the ones that
justify me (Raco) in believing what I do whether or not
these reasons have any causal influence on my belief in this
case. For example, the valid deduction of a conclusion from
justified premises (when I know that these premises are
justified and the deduction valid) is enough to justify my
belief in the conclusion even though the these facts are not
the causes of my belief in the conclusion. IS

But, here it matters very much whether b is true or false. One can only
beat or neutralize a competitor in the ultra-justification game with a proposition
if that proposition is true. Otherwise, the skeptic is allowed to delete the belief
and make any other necessary adjustments to the agent's doxastic system in light
of this deletion. So, if b is true, and Mr. Raco responds to c with b, then b beats
c and Mr. Raco is verifically as well as personally justified in his belief
(assuming that b is in fact both personally and verifically justified as well).
Hence, he knows thatthe minority suffers from some peculiar and otherwise rare
disease. However, if b is false, then the skeptic can say this, excluding b from
use in the ultra-justification game. This means that b would not beat c, and
therefore Mr. Raco would not have knowledge according to Lehrer's theory as
it is not the case that all competitors have been beaten or neutralized. 19 Thus, we
can appreciate the critical importance of the truth or falsity of b in Lehrer's

The problem is that Lehrer's theory does not actually tell us whether b is
true or false. The core of Lehrer's theory-all of Lehrer's definitions, formal
principles, etC.-is compatible with conflicting judgements about Mr. Raco; if
b is true, then Mr. Raco is verifically justified in his racist belief, but if b is false
then Mr. Raco is not verifically justified. One way of better seeing this is by
noticing that one could adjust Lehrer's theory to require appropriate causation
merely by understanding all reasons claims as about one's own reasons. Mr.
Raco's own reasons are not the scientific studies but rather his racism (if we
count this as a reason at all). Therefore, when Mr. Raco replies to the skeptic
with b and the skeptic rebuts him and deletes the proposition from Mr. Raco's
acceptance system, Mr. Raco is left without good reasons for his racist belief.
Thus, this belief will fail to be verifically justified and will not be an instance
of knowledge. To secure this result, however, no changes were made anywhere
to Lehrer's stated theory of knowledge and justification. No definitions were
adjusted, nor were any principles rewritten. Even the spirit of Lehrer's theory
is preserved here; only our conception of how we are related to a fact, as
encoded in the statement of our reasons for acceptance, and truth and falsity (as
the externalist bridge between personal justification and knowledge) are relevant
to justification. 20 This in itself helps make clear that Lehrer's theory simply
does not entail a verdict except in light of the truth or falsity of various
propositions which the theory itself does not entail.
The deeper significance of this sort of concern is that Lehrer's theory is
incomplete as it stands. In particular, it will only deliver verdicts about certain
kinds of cases when it is supplemented by specific claims as to the truth or
falsity of certain propositions that might be used to beat or neutralize various
skeptical objections. For, an agent can be personally justified in accepting that
a proposition beats or neutralizes a competitor or objection and yet not be
verifically justified in believing this. This reveals a more troubling problem for
Lehrer's theory: all of the definitions and principles do little to actuaIly pin
down exactly when an agent is verifically justified. These principles wiIl need
to be supplemented with various contentful claims about what is true and false
before it will be possible to derive results about verific justification. 21
This leaves us with the question as to whether it matters that Lehrer's
theory is incomplete. On the one hand, the incompleteness of the theory means
that it may be adjusted in all sorts of ways without changing the existing theory
itself. So, as I argued above, Lehrer's theory can be made to account for the
intuition that Mr. Raco's racist beliefis not an instance of knowledge by adding
a claim about the falsity of a certain proposition that will be employed by the
agent when she attempts to neutralize or beat the skeptic's objections in the
ultra-justification game. And, it might be considered a considerable virtue of a
theory that it remains silent about, yet compatible with, all sorts of contentious
claims about the relationship of causation to justification, etc. At the very least,

the theory is weaker than it perhaps appears, but hence more flexible because of
In some sense, this is as it should be since Lehrer's theory is fallibilist and
allows that an agent can be personally justified and yet lack knowledge. But, it
also makes clear that the externalist components of Lehrer's theory- namely
the creation of an ultra-system by the deletion of all false beliefs-play such an
important role that we can only assess knowledge claims according to Lehrer's
theory if we have an indication of exactly what propositions are true and false
and hence which propositions an agent may use to neutralize or beat objections,
remembering that all false propositions will be deleted by the skeptic in the
ultra-justification game. Lehrer's treatment of the Mr. Raco case makes clear
what Lehrer thinks, namely that knowledge is possible without causation by
justifying reasons. But, Lehrer would need to explicitly add claims about the
truth or falsity of various kinds of propositions likely to be employed to meet the
skeptic's objections in the ultra-justification game to his theory for it to entail
this result.
On the other hand, this incompleteness makes Lehrer's theory very
difficult to assess, and might even make the theory nearly unfalsifiable. Testing
our intuitions against the theory as it stands may be impossible in some cases,
as with Mr. Raco. Critics will have to be extremely careful to make sure that
Lehrer's theory really does deliver certain results before they can present
compelling counterexamples. Because of this, it might be best to focus on the
general structural features of the theory and refrain from the usual sort of
counterexamples until the theory is modified so as to have clear consequences
about various kinds of cases, but this severely limits our ability to determine
whether or not Lehrer's theory is a good one. The flexibility of the theory is
here purchased at the cost of our ability to test the theory in various respects.
I have argued that Keith Lehrer's theory of knowledge does not in fact
have some of the consequences that Lehrer and others have taken it to have. In
particular, Lehrer's theory as it stands remains silent about the relationship of
causation to justification. Indeed, it seems possible to easily accommodate the
view that causation is necessary for justification by simply understanding
reasons claims as always being partly causal. Causation would enter the picture
here because it is explicitly represented in reasons claims, and improper
causation can make such reasons claims false. Of course, Lehrer's theory is also
compatible with the view that justifying reasons need not be causal. Whether
the incompleteness of Lehrer's theory is a virtue or a vice, and both the
flexibility and difficulty of assessment this incompleteness reveals, I leave to the
reader to decide. 22

I Lehrer (1990), pp. 169-70. The Mr. Raco case also appears in Lehrer's more recent (2000)

second edition in an essentially unchanged form on p. 196.

2 Dirk Koppelberg understands the case in basically the same way. See his (1999), p. 452.
3 Hereafter I will speak in terms of belief rather than acceptance merely for ease of exegesis.
Acceptance, however, is not the same as belief according to Lehrer; rather, acceptance is something
like belief with the goals of believing truths and avoiding falsehoods.
4 See Lehrer (1990), p. 149.
5 Koppelberg (1999), p. 447.

(, In some sense, Koppelberg never takes Lehrer's theory of justification very seriously at all.
Koppelberg never assesses Lehrer's theory but rather simply argues that Mr. Raco is not justified
according to various other conceptions of justification. Insofar as there is any direct contact
between Koppelberg and Lehrer, it is that Koppelberg would deny that Mr. Raco has knowledge
of the proposition belief in which is sustained or caused by prejudice. See fn. 21, p. 460.
7 Koppelberg, p. 459.

8 One might argue that what this shows is that Lehrer's theory is not really a coherence theory after

all. This taxonomic maneuver strikes me as unsatisfying, though. Lehrer's theory strikes me as
an exemplar of a certain type of coherence theory (with certain externalist components of the sort
needed to solve the Gettier problem), and it would do injustice to our taxonomy to classify it
otherwise. In particular, I think that we will always have to add non-coherence factors - most
obviously truth - to our account of knowledge, and so as long as one's basic theory of justification
is coherentist, one should be classified as a coherentist. This does seem true of Lehrer's theory.
9 Lehrer, p. 126.

10 For convenience, I have dropped reference to a specific time t t!'om the definitions.

II I intend Rs to be interpreted as reasonableness given acceptance system S. Obviously,

reasonableness is not the same thing as probability on many interpretations of probability, and the
notation I use is strikingly similar to that of the probability calculus. According to Lehrer, as
reasonableness is a primitive normative notion, it seems unlikely that any view of probability
which is non-normative could adequately cxpress Lehrer's position. But, the point of my use of
this notation here is simply to help make clear some of Lehrer's definitions in a condensed format.
12 All of these definitions can be found in Lehrer, pp. 117-126.

13 It is not actually clear to me that c as I state it does compete with Mr. Raco's racist belief. But,

if it does not, the skeptic can presumably work c into something that is a competitor that can be
neutralized (or perhaps beaten) with something approximately like n. For example, the skeptic
might claim that it unreasonable to hold a belief if it is not causally responsive to evidence. We
could then understand n as rebutting this further claim which would act as a competitor to the
racist belief.
14 In the ultra-justification game, the skeptic is allowed to simply delete any false propositions (and

all propositions that logically imply this proposition) used in the defense of some other
proposition, disqualifying such stricken propositions from use as justifiers. This is core of the
difference between the justification and ultra-justification games. See Lehrer, pp. 141-44.
15 Read the "because" in this sentence as both causal and justificatory.

16 I wish to remain neutral on how it is that testimony justifies what the content of what the testifier

says (viz., whether testimony proceeds on the basis of perception and induction, or whether it has
some sort of intrinsic a priori authority as a basic source of belief, etc.). I use testimony here only
because it helped make for a clear example.
17 I am being intentionally vague here about exactly what is to count as a reason. Is it Wendy's

belief that Mr. Mathright testified, or Mr. Mathright's testimony, or something else entirely? I do
not think that my argument depends on any specific construal of a reason, so I leave it to the reader
to supply their own preferred notion. What I am suggesting, however, is that one could very easily

build an explicit causal component into the truth conditions for at least some of these conceptions
ofa reason.
18 I think we need to treat this proposition as beating the skeptic's objection rather than neutralizing

it because one cannot consistently believe both c and b. Second, I will admit that I am a bit
concerned that it is far from likely that any average person, even one with a bit of philosophical
savvy, would respond to the skeptic's challenge in this way. I fear that perhaps Lehrer's account
may exclude many intuitive cases of knowledge. But, this is an argument for another time.
19 Interestingly enough, one might try to explain differences in intuition (which a small poll has
revealed do seem to diverge) about the Mr. Raco case by focusing on one's attitude towards b. If
one thinks that b is true, then one will likely have the intuition that Mr. Raco does have knowledge.
On the other hand, if one sees b as false, then one will be inclined to deny that Mr. Raco has
20 See Lehrer (1990), p. 153.

21 I suspect that this problem is not limited to various causal principles of justification. For
example, the skeptic might challenge certain people on the grounds that their justifications are too
short. Instead of presenting any content-dependent reasons for accepting thatp, suppose the agent
simply argues that she is trustworthy in what she accepts and hence, given that she accepts that p,
p is likely to be true (this argument will work for any proposition which the agent accepts).
Suppose that the skeptic objects that not everything that the agent believes can be justified with
this universal argument. Suppose that the agent replies by simply reapplying the simple argument
again: since the agent accepts that his justifications are long enough to justify his beliefs, and what
he accepts he accepts in a trustworthy manner, it is probably true that his arguments are long
enough to justify his belief. Is this an effective way of beating the skeptic's objection? This will
depend where verific justification is concerned on whether or not the agent's reply is in fact true.
But, does Lehrer's theory as it stands tell us whether or not we can use this simple universal
argument to completely justify all our beliefs? This is not clear to me.
22. Thanks to Keith Lehrer for his support and comments on drafts of this paper. Thanks also to

Erik Olsson and to Paul Thorn for their extremely helpful suggestions.

Koppelberg, Dirk. "Justification and Causation." Erkenntnis 50, 447-462 (1990).
Lehrer, Keith. Theory of Knowledge. Boulder, co: Westview Press, 1990.
Lehrer, Keith. Theory of Knowledge, Second Edition. Boulder, CO: Westview Press, 2000.
Chapter 5


Volker Halbach
University of Constance

In this paper, I will scrutinize a family of arguments which are supposed to show
that consistency is not accessible in an epistemically relevant sense. If these ar-
guments were sound, then consistency would be on a par with external facts:
neither are directly accessible to us. External factors are not directly accessible
because they need to be mediated in some way. Consistency would be inaccessi-
ble, according to these arguments, because proving or determining consistency
in the relevant cases exceeds our intellectual capabilities. Thus, if consistency
were actually inaccessible, then consistency could hardly playa role in an inter-
nalist account of epistemology. Traditionally (for instance, in Schlick's account
(1934)), however, consistency has been seen as a main ingredient of coherence.
Modern epistemologists, like Bonjour (1985), have also used consistency as a
criterion of coherence, and they have even used consistency in order to define
The arguments against the accessibility of consistency jeopardizes this
central role of consistency in any internalist account of epistemic justification. I
Thus, epistemic justification cannot imply the consistency of the belief system.
If consistency is not directly accessible, then it cannot be used as a criterion for
consistency on an internalist account.
Keith Lehrer's epistemology is not internalist. Internalist justification,
however, is an important ingredient of Lehrers's account since internalist justi-
fication is a necessary condition for 'full' justification. Because of the Gettier
problem, external factors enter the definition of one form of epistemic justifica-
tion on Lehrer's account. Lehrer (1990b), for instance, distinguishes between
personal and undefeated justification. External factors form part of the defi-
nition of the latter. Undefeated justification is then used in the definition of
E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 75-87.
© 2003 Kluwer Academic Publishers.

knowledge. As we will show, this account of internalism makes Lehrer's the-

ory as vulnerable to the arguments mentioned above as an actually internalist
theory-if consistency is employed as a condition for internal justification.
Lehrer's opinion on the relationship of consistency and justification has
not been constant, as I will show below. In (1991), he explicitly endorses the
view that consistency is a necessary condition for justification: 2
Our acceptance systems may not be deductively closed, but the deductive
closure of an acceptance system fully articulates the logical content of
the system. The content of an inconsistent system would, therefore, be
useless for the purposes of justifying anything and would, consequently,
fail to serve as the basis of knowledge.

Thus justification implies the consistency of the acceptance system. 3 The quot-
ed passage does not imply that we have a direct grasp of the consistency of our
acceptance system, and Lehrer avoided, as far as I can see, such strong claims.
The reason for avoiding such a commitment may be, in part, the argument I will
discuss in this paper.
Implicitly, however, consistency is also central for Lehrer's theory, for
an inconsistent theory logically refutes any given belief. One would like to
conclude that an inconsistent belief system therefore also defeats any belief. It
seems hard to see how Lehrer can avoid that consequence, although he must
deny this when claiming that beliefs can be justified on the basis of an incon-
sistent belief system. Therefore, one has to be, at least, very careful about in-
consistency. The reader is sent to the above mentioned papers (Lehrer 1990a)
and (Lehrer 1999) for a discussion of this topic. At any rate, it seems interesting
to see to what extent consistency can be used as a criterion for justification and
coherence in a framework like Lehrer's.
In the following, I will discuss the arguments against the accessibility of
consistency in a general setting in order to see whether consistency can be used
as a criterion for epistemic justification. I will not explore what consequences
are implied for particular theories like Lehrer's or Bonjour's.
The point of this paper is modest: I will show that some of these argu-
ments have been accepted rashly and that their scope is more limited than is
often assumed. I hope that this is interesting enough. For these arguments have
had a major impact, and they have influenced many epistemologists.
Thus, I will examine whether the arguments actually demonstrate that
consistency is not accessible and, in this respect, is much like an external factor.
The problem is vague, of course. In the first place, it has to be specified whether
one is dealing with the consistency of arbitrary sets of beliefs or only with the
consistency of one or more specific sets like one's belief system, (i.e., the set
of beliefs one accepts). Moreover, I have to specify for whom consistency is or

is not accessible. Is consistency accessible only to intellectually perfect beings

with no limits of their computational powers, or also to human beings with their
imperfections? The arguments differ in their scope. Some purport to show that
consistency is inaccessible to agents with limited computational abilities, while
other are supposed to apply to perfect agents too. Furthermore, accessibility
itself is a vague notion. Must we have a 'direct grasp' of consistency, or are we
allowed to employ auxiliary resources in order to prove or check consistency?
For instance, may we calculate with the aid of paper and pencil, or even with
the aid of a computer?
Rather than giving here at the beginning a precise account of accessi-
bility, consistency, and so on, I will instead examine what is achieved by the
respective arguments, since they differ widely in their scope. As we shall see,
some arguments actually prove that consistency is not accessible in a very strong
sense; but accessibility in this strong sense hardly matters for the purposes of
The arguments against the attainability of consistency for epistemological
purposes rely on an assumption about belief systems; viz., that belief systems
can be represented by sets of sentences. Something can be said for the view
that belief systems are more than just pure sets of sentences. However, for
our purposes considering sets of sentences and their consistency is sufficient: a
belief system is consistent if and only if the corresponding set of sentences is
consistent. If this assumption were denied, then the arguments I will scrutinize
would not get off the ground at all, and there would be no need to refute them.
In the following, I will consider the arguments. I start with the most
radical and basic arguments, and then move to more sophisticated arguments.
In particular, I will move on to arguments that rely on more and more technically
sophisticated arguments.


The following is the most radical argument I will consider. According to this
argument, consistency cannot be checked by human beings for more than a very
limited number of beliefs. Thus, not even reference to the complexity of truth
tables, or similar metalogical aspects, are envoked.
In his presentation ofChemiak's (1986) account, Hooker (1994) writes:

In practice, humans can only juggle a few beliefs simultaneously in con-

sciousness to check for consistency, to make and check inferences, and
so forth-less than ten for most of us, and only so long as they are sim-
ple in structure. [ ... ] It is simply not possible to carry out even partial
global consistency checks, [ ... ]

An even stronger claim is made by Kornblith (1989, p. 211):

It is not simply that human beings have difficulty in determining the con-
sistency of large sets of sentences. It is simply beyond the powers of any
possible computational device to determine the consistency of a large set
of sentences.

I agree that human beings are not able to determine the consistency of arbitrary
large sets of sentences for their consistency. But in order to attain justification
a subject only has to prove that its own belief system is consistent. Thus, Korn-
blith seems to think that humans are not able to establish the consistency of a
set of sentences that is as large as our belief systems.
However, it is not in general impossible for human beings to establish the
consistency of certain large and even infinite sets. Kornblith's claim taken as
a universal statement is simply false. It is not generally beyond the powers of
any possible computational device to determine the consistency of large sets of
Logicians prove consistency of infinite sets of axioms all the time. Peano
Arithmetic PA, for instance, is given by infinitely many axioms and cannot be
axiomatized by a finite set of axioms. Nevertheless, it does not exceed the ca-
pabilities of the average logician to show the consistency of PA by providing a
model, or by proving its consistency in Gentzen's style (cut elimination). Fur-
ther examples can be found in abundance. In general, a high cardinality of an
axiom set does not prevent logicians from providing consistency proofs.
In general, consistency proofs do not require that we juggle all sentences
of a set simultaneously in consciousness; this is not necessary for carrying out
consistency proofs. In a consistency argument, we do not have to use the sen-
tences of the set in question, which is impossible; rather we only have to talk
about them. The latter is no problem; we can say that all sentences in a set have
a certain property. In particular, we can ascribe certain formal properties to all
sentences in a set and finally arrive at the conclusion that a contradiction cannot
be derived from these sentences.
Therefore, the argument does not show, by any means, that we cannot
determine the consistency of sets with more than ten sentences.


I now turn to more elaborate arguments. These arguments, it is claimed, show

that consistency is inaccessible to human epistemic subjects for metamathemat-
ical reasons.

There are, at least, three results in metamathematics that have been em-
ployed in order to prove the inaccessibility of consistency. The arguments fall
into three classes:

• Complexity of truth tables. This point concerns consistency only with re-
spect to propositional logic. Propositional logic is decidable; a straight-
forward method for checking it is provided by the truth table method,
which is taught in almost any introductory course to logic. There are 2n
many different assignments of truth values to n propositional variables.
Thus, a propositional sentence with, for instance, 12 propositional vari-
ables yields a truth table with 212 = 4096 many lines. It is easily seen
that even very simple sentences are hard to check for consistency with
the truth table method and practically cannot be checked for consistency
by the truth table method.

• Undecidability of predicate logic (Church's Theorem). In contrast to

propositional logic, predicate logic is not decidable. There is no deci-
sion procedure to determine whether a given (recursive or even finite) set
of sentences is or is not consistent. This argument is essentially differ-
ent from the preceding because it shows that there are general recursion-
theoretic problems in checking a set of sentences for consistency, not just
"practical" problems.

• Unprovability of consistency (G6del's Second Theorem). No system sat-

isfying some minimal requirements can prove its own consistency. There
is not only no decision method for consistency, but we also cannot even
stumble by chance over a proof of consistency. We cannot show in our
system of beliefs that the set of all beliefs is consistent. Only this ar-
gument needs the requirement that the consistency of the set of beliefs
must be proved within the belief system. Thus, it is different from both
preceding arguments.

How conclusive are the three arguments? Do they show that consistency
is like an external factor in being out of our control and in being (in the relevant
cases) inaccessible to us?

2.1. Truth Table Complexity

According to this argument, checking belief systems for consistency may be
feasible for computational devices exceeding our capabilites, but human beings
are not able to carry out computations required for consistency checks of 'nor-
mal' belief systems. While consistency might be accessible to ideal epistemic

agents with less restrictions on their computational capabilities, human beings

are precluded from checking or establishing the consistency of their belief sys-
Cherniak (1986) has emphasized the relevance of the complexity of truth
tables for consistency checks in epistemology. Even comparably short sentences
yield many lines in their respective truth tables. Does this imply that consistency
is not verifiable in cases that matter for the coherence theory of knowledge?
Arguments involving the complexity of truth tables should not impress
the epistemologist. In the first place, there are methods other than truth tables for
checking a set of sentences for its propositional consistency. In fact, these other
procedures may terminate very quickly. These tricks are taught in introductory
logic courses.
Furthermore, it is not unnatural to assume that the elements of a belief
system have a not too complicated propositional structure. Even if the belief
system of a person is built up from many propositional atoms, they will hardly
be combined in one complicated sentence. In practice, humans are unable to
grasp sentences with excessively many propositional atoms. 4 At a propositional
level, a typical belief system will consist of many unrelated sentences; they will
stand alongside each other without many connections. And it is easy and not
complex to check a set {PI, P2, P3, ... } for consistency because we do not have
to go through all lines of the truth table. In order to instantiate the claim that
belief systems are usually very simple at the propositional level, I could present
physical and mathematical theories formalized in propositional logic. It should
be obvious that the resulting propositional sentences would be very simple.
Thus, the arguments relying on the complexity of truth tables for check-
ing consistency do not threaten the claim that consistency is accessible to us.
From these arguments, it cannot be seen why a person should not have grasp of
the consistency of his or her system.
Truth tables are a very bad tool for checking consistency. I do not want to
challenge their value for didactical purposes, but they are hardly ever employed
in consistency proofs outside of the toy world of propositional logic.
Since the argument concerning truth tables does not show that the consis-
tency ofhis/her belief system is not accessible to a subject, I leave propositional
logic and tum to predicate logic. Propositional consistency does not matter
anyway. What matters is general consistency, which is much better parsed as
consistency in quantificationallogic.

2.2. Church's Theorem

The arguments of the second class, relying on Church's Theorem, provide a
harder challenge. Church's Theorem does not claim that consistency of a given
system is hard to compute, rather it states that there is no general method for
checking consistency, at least not for epistemic agents whose computational
capabilities do not exceed those of Turing machines. Therefore, it cannot be
argued as above that there are somehow better methods to decide whether a
belief system is consistent. Church's Theorem concerns all possible methods of
checking consistency of arbitrary systems without regard to their computational
complexity. Church's Theorem states that there are no such methods at all.
First, I will show that a certain direct retort does not succeed. It has been
argued that the computational capabilities of human agents are not limited to
those of ideal Turing machines. Obviously, the argument involving Church's
Theorem relies on the assumption that epistemic agents do not have means to
check consistency that go beyond those of ideal computer programs. But, ob-
viously, this assumption also obtains: real humans have only finite resources,
which are much less than a Turing machine has. The latter has unlimited mem-
ory, no temporal limitations for computations (as long as they are finite), etc.
What ideal epistemic agents can or can not do may, naturally, be controversial.
It can be claimed consistently that ideal epistemic agents are not bound to recur-
sive methods. God, it may be said, is an ideal agent and he/she knows how to
solve any problem whether it is computable (in the sense of recursion theory) or
not. Although there is no inherent mistake in the concept of such an ideal epis-
temic agent, it is not interesting for the present topic. In the following, I will
not consider agents who have justified divine intuitions about the consistency of
belief sets.
Nevertheless, I do not think that Church's Theorem shows that consis-
tency is inaccessible. It shows only that we do not have a general method for
deciding whether a set of sentences is consistent or not. It does not show that
we cannot decide and even prove consistency in all relevant cases.
Here it is useful to take a closer look at the proof of Church's Theorem.
Using G6del's techniques, it is shown that there is an undecidable finitely ax-
iomatized theory Q, called Robinson's arithmetic. 5 In particular, there is no
general recursive method for deciding whether a given sentence A is provable
in Q or not. This is a consequence of G6del's First Theorem for Q. If fA Q is
the conjunction of all axioms of Q, then there is also, consequently, no recur-
sive procedure for deciding whether fA Q ----> A is logically valid or not. Hence,
there is no recursive procedure for deciding whether any arbitrary sentence is
logically valid or not.
From this it can also be deduced that consistency of a given sentence in

predicate logic is not decidable. A sentence A is consistent with Q if and only

if Q does not prove ....,A, that is, if and only if fA Q - t ....,A is not logically
valid. But, as shown above, there is no recursive method to check whether a
sentence of the form fA Q - t ....,A is logically valid or not. Thus, there is no
effective procedure for deciding whether a sentence A is consistent with Q, or
for deciding whether fA Q 1\ A is consistent.
The most plausible strategy for defending the accessibility of consistency
consists in the denial that a general method for checking consistency has to be
applied. This means that the description above of how we should revise our
belief system is faulty. There is no simple check for consistency; rather, we
sometimes have to try hard to establish consistency. But still we can say that a
set of beliefs is only justified if its consistency has been demonstrated.
Of course, there is no general procedure for checking consistency, but
why should an epistemic agent be required to have a general procedure for de-
ciding whether an arbitrary set of sentences (belief system) is consistent or not?
For obtaining epistemic justification, he needs a consistency proof for his very
own belief system, but not for the belief system of other people or any other
belief system different from his own.
This description is compatible with Church's Theorem. It does not follow
from this theorem that we cannot prove the consistency of the system Q. That
is, we might be able to refute that fA Q - t 1- is logically valid, although there
is no general test for the validity of sentences of the form fA Q - t A. Actually,
there are straightforward proofs of the consistency of Q. 6 It may well be that
we can prove the consistency of our belief system, although we lack a general
procedure for determining whether a given set of sentences is consistent or not.
Thus consistency would be accessible in all relevant cases, but inaccessible in
some others.
Church's Theorem actually raises a problem for certain accounts of be-
lief revision. There are accounts that describe how we should deal with new
information as follows. Assume that S is the set of beliefs you have already ac-
quired, and suppose further that there is a new input A. Now check first whether
the new belief A is consistent with the old stock S of beliefs. If it is consistent,
simply add A to the set S of old beliefs and you obtain the new belief system
S U { A }, which incorporates A. If A is not consistent with the beliefs in S, then
do something else.
Many formal approaches to belief revision are concerned with this "some-
thing else". But here I am not interested in the problem of how our belief system
can be revised in such a way that a new (consistent) belief set is obtained in the
case that A is not consistent with S. Here I am interested in the earlier step
where it is required to check whether A is consistent with S. Most formal ac-

counts of belief revision deal only with propositional logic. Thus "consistency"
is to be understood in terms of propositional logic. Since propositional logic is
decidable, the check for consistency can be carried out in principle (although it
may require fairly complex computations).
But ifthis picture of how new beliefs are incorporated in our belief system
is applied to predicate logic, a fundamental difficulty arises. Epistemic agents
do not know how to obey the command "check first whether the new belief A
is consistent with the old stock S of beliefs!" According to Church's Theorem,
there is no general recursive method to carry out this check.
So there might be a general problem for the applicability of formal ac-
counts of belief revision. But I do not take it as a conclusive argument against
the claim that consistency is not accessible in a sense relevant for epistemology.

2.3. Godel's Second Theorem

It can be granted that consistency of very complex belief systems can be shown,
and that the complexity of truth tables and Church's Theorem do not contradict
this observation. In fact, consistency proofs of even very strong systems can
be very simple in terms of length of proofs. Thus, I conclude that the above
arguments cannot disprove conclusively that consistency is accessible.
Here is an interesting twist. What tools do we have at our disposal in
order to establish the consistency of our belief system?
Obviously, we must carry out the consistency proof of our belief system
on the basis of what we believe, that is, on the basis of our belief system. For we
can use in our reasoning only assumptions we already believe. The consistency
of the belief system must be established within that very belief system.
Godel's Second Incompleteness Theorem seems to tell us that such a
consistency proof cannot work, according to the slogan "No system can prove
its own consistency." So, proving consistency of one's belief system does not
produce problems because the proofs are too complex or because there is no
uniform procedure to check consistency, but rather because the resources for
the proof are limited in the case we are interested in, namely where we try to
prove consistency of a belief system within that very same belief system. In all
interesting cases, the consistency of a belief system can only be established in a
more comprehensive belief system, or so it seems.
In the discussion of the argument based on Church's Theorem, I claimed
that consistency proofs can be carried out although we lack a decision procedure
for consistency. These consistency proofs, however, require means that go be-
yond the resources available in the system for which consistency is shown. The
consistency of Peano Arithmetic, for instance, can easily be shown within set

theory, or in Peano Arithmetic plus some additional principles like transfinite

induction up to co, but not within Peano Arithmetic, or so it is claimed.
I grant that G6del's Theorem on the unprovability of consistency poses
the hardest challenge for consistency as an epistemically valuable criterion of
coherence. However, the case is far from being clear because of some nontriv-
ial metamathematical problems that have some bearing on the problem. As is
well known, it is not so easy to state G6del's Second Theorem in a precise and
general form. The slogan "No system can prove its own consistency" is at least
misleading if not simply false.
In this paper, I do not discuss all ins and outs ofG6del's Second Theorem.
I will only mention some points that seem relevant to our discussion.
In the first place, most formulations of G6del 's theorems pertain to arith-
metical languages. In particular, consistency is expressed in this arithmetical
language via some coding. Epistemologists, however, did not claim that coher-
ence requires provability of certain arithmetical sentences that somehow 'ex-
press' consistency relative to some coding. Thus, if G6dels theorems are rele-
vant at all, they must be generalized beyond the realm of arithmetical languages.
Even if it is assumed that the belief system is entirely formulated in an
arithmetical language, it is by no means clear that a system cannot prove its own
consistency. I will hint at only a few well-known technical details.
What is the consistency statement for a formal system S? Usually one
provides a formula Bew (x) that represents (numerates) provability within the
system. That is, Bew( n) is provable (relative to a system to be specified) if and
only if n is the code of a formula provable in S (where n is a numeral for the
number n). Then, the consistency statement is defined as -.Bew(#(O =I- 0)).
However, there are many formulas representing provability in S. For some,
G6del's Second Incompleteness Theorem holds; for others, it does not.
There is a strong feeling among logicians that those consistency state-
ments for which the second incompleteness theorem does not hold are somehow
unnatural. But so far no-one has been able to give a clear account of 'natural-
Feferman (1960) has succeeded in providing certain technical conditions
on Bew( x) that allow for a proof of the incompleteness theorem, but these con-
ditions fall short of being 'natural.'
Therefore, the second incompleteness theorem is inconclusive as an ar-
gument against consistency as a necessary condition for coherence. Popular
simplified formulations of the second incompleteness theorem like "No system
can prove its own consistency" may indeed pose a problem for somebody claim-
ing that consistency is accessible to us. However, the mathematical facts do not
support these formulations. Therefore, I believe that the burden of proof is with

those who claim that consistency is not attainable because of Gbdel 's theorem;
they have to provide versions of the theorem that yield their claim that consis-
tency is not within our reach.


I have shown that the arguments that are purported to prove that consistency is as
little accessible to us as purely external facts are not conclusive. Of course, con-
sistency is not accessible, if 'accessible' is understood in a very strong way, for
we do not have general procedures for determining whether an arbitrary given
set of sentences is consistent. But this kind of accessibility is not required for
most epistemological purposes. For those purposes, we only need to establish
the consistency of our own belief system or the consistency of our belief system
with another belief.
Although I have shown that the discussed arguments are not conclusive, I
do not claim that consistency actually is accessible in a relevant sense. Church's
Theorem and Gbdel's Second Theorem will make it hard to give a general ac-
count of how justification is obtained if consistency is to be used as a partial
criterion for justification. The arena is open again.

Acknowledgements. The paper was presented to the Belgian Society for Logic
and Philosophy of Science, Brussels, Belgium, on 23 January 1999. The helpful
suggestions of the audience are gratefully acknowledged. The work on this
paper was carried out while the author was a member of the research group
Logic in Philosophy financed by the Deutsche Forschungsgemeinschaft. I thank
David McCarty and the members of the research group and, in particular, Erik
Olsson for useful suggestions and discussions.

1 Moreover, these arguments have also been employed against extemalist coherence theories. See

Komblith's (1989) attack on Hmman's (1973) theory.

2 After the quote below, Lehrer also discusses the prospects of working with inconsistent belief

3 According to this quote, an inconsistent acceptance system does not justify any belief. In the

more recent paper (1999), Lehrer explicitly maintains on p. 252 that beliefs can be justified on
the basis of inconsistent belief systems. I quote another passage where he seems to suggest that
inconsistency might not be fatal to our efforts to obtain justified beliefs. He writes in (l990a,
p. 166):
Hence one who values consistency as an end in itself should recognize that a person with
equally pure intellectual concerns may reasonably accept an inconsistent set of sentences.
What he accepts may be suited to the objectives of obtaining new information and eschew-
ing error.

4 At this place it makes a difference how belief is defined. If, linguistically speaking, the limited
performance of a subject is neglected and only his or her competence is considered, then a person
may believe even very long sentences and in fact sentences of arbitrary length. However, if there
is no limit imposed on the perfonnance of a subject, then there is no limit on the complexity of
truth tables that can be calculated.
5 There is also another system called Robinson's arithmetic which is usually designated by R.

6 Just how valuable the consistency proofs for Q are is a different problem. In the light of Go del's
Second Incompleteness Theorem, the value of these consistency proofs may be doubted. I will
return to this point below.

Bonjour, Laurence. 1985. The Structure C?f Empirical Knowledge. Cambridge, Massachusetts:
Harvard University Press.
Chemiak, Christopher. 1986. Minimal Rationality. Cambridge, Massachusetts: MIT Press.
Feferman, Solomon. 1960. "Arithmetization of metamathematics in a general setting." Funda-
menta Mathematicae XLIX:35-91.
Harman, Gilbert. 1973. Thought. Princeton: Princeton University Press.
Hooker, Cliff A. 1994. "Idealization, Naturalism, and Rationality: Some Lessons from Minimal
Rationality." Synthese 99:181-231.
Komblith, Hilary. 1989. "The Unattainability of Goherence." In The Current State C?fthe Coher-
ence Theory, edited by John Bender, 207-214. Dordrecht: Kluwer.
Lehrer, Keith. 1990a. "Reason and Consistency." In Metamind, 148-166. Oxford: Clarendon
Press .
.- - . 1990b. Theory of Knowledge. Boulder: Westview Press.
- - - . 1991. "Reply to Mylan Engel." Grazer Philosophische Studien 40:131-133.
- - - . 1999. "Justification, Coherence and Knowledge." Erkenntnis 50:243-258.
Schlick, Moritz. 1934. "Uber das Fundament der Erkenntnis." Erkenntnis 4:79-99.
Chapter 6



Glenn Ross
Franklin and Marshall College

In his formulation of coherentist theories of knowledge and epistemic

justification, Keith Lehrer has often returned to the lottery paradox to draw
important lessons. Some of these lessons are about knowledge. Lehrer has
maintained that if a lottery has lots of tickets, only one of which will win, one
cannot know, simply on probabilistic grounds, that any particular ticket will not
win. Lehrer also defends a less obvious lesson: that it is not even reasonable to
accept that one's ticket will not win. It is not clear, however, that Lehrer's theory
of personal justification has this consequence. It is even less clear that it should.
The lottery paradox for reasonable acceptance can be seen as a dilemma,
and Lehrer's resolution amounts to swallowing a skeptical horn of that dilemma.
Showing that this resolution is too skeptical requires that we canvas several
similar approaches to the lottery paradox. Since I agree with Lehrer and Iike-
minded philosophers that the other horn of the dilemma, an inconsistency
resolution, is wholly unacceptable, I endorse slipping between the horns.
Nonetheless, I find that there is much in Lehrer's general theory of reasonable
acceptance that is congenial to my view.

In a fair lottery with very many tickets, and only one winning ticket, I
should be very confident that my ticket would lose. My reasons for being highly
confident would seem to make it epistemically permissible for me to accept that
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 91-107.
© 2003 Kluwer Academic Publishers.

my ticket will lose. Yet, since my reasons for accepting that my ticket will lose
are qualitatively identical to my reasons for accepting that any other ticket will
lose, I do not have any epistemic reason to prefer accepting that my ticket will
lose to accepting that some other ticket will lose. It seems epistemically wrong
and arbitrary to accept that which I recognize I have no better reason for
accepting than that which I choose not to accept. So, either I should not accept,
of any ticket, that it will lose, or I should accept, of each and every ticket, that
it wi\l lose. To refuse to accept of any ticket that it will lose in a very large
lottery seems immoderately skeptical. To accept that a given ticket will lose is
attractively less cautious. Yet, if it would be arbitrary and epistemically
unreasonable to accept that one ticket will lose and not to accept the same of the
other tickets, then one cannot accept of one ticket that it will lose without being
obliged to accept the same of all the rest. If I accept of each ticket that it will
lose, but also accept that some ticket will win, my acceptances will be self-
recognizably inconsistent. That seems just as bad. ,
Keith Lehrer 1 and many others (e.g., Mark Kaplan 2 , John Pollocko, Sharon
Ryan4, and Dana K. Nelkin 5) adopt the epistemically cautious resolution and
insist that considerations of consistency demand that one should uniformly
withhold judgment on each proposition that a particular lottery ticket will lose.
Richard Fole/ and Peter Klein7 opt for the much more credulous
recommendation that one can be recognizably inconsistent: accepting of each
ticket that it will lose while also accepting that some ticket will win. I propose
that we can slip through the horns by adopting a consistently credulous
approach 8 : one can rationally accept that one's ticket will lose, while not
accepting that of many of the others, even though one is no less confident that
they too willlose. 9
Both the skeptical and the inconsistency solutions presuppose a principle
of symmetry: that if! have equally good reason to accept of any ticket that it will
lose as I have to accept that any other ticket will lose, I should adopt the same
attitude uniformly. I should either accept of each ticket that it will lose or not
accept of any ticket that it will lose.



Lehrer accepts the principle of symmetry and argues that considerations

of consistency show it is not reasonable to accept that any lottery ticket will lose.
He then proceeds to show how he can get this result from his theory of rational
acceptance. His theory involves a few technical terms that must first be defined.

Lehrer analyzes personal justification in terms of coherence in an acceptance

system. Whether a statement is reasonable to accept depends upon what
statements compete with it. The notion of competition is defined as follows:

c competes with p for S on the basis of the acceptance system of S

at t if and only if it is less reasonable for S to accept that p on the
assumption that c is true than on the assumption that c is false on
the basis of the acceptance system of Sat t.10

Naturally, statements beat their competition for reasonab Ie acceptance when they
are more reasonable to accept.

p beats c for S on [the acceptance system of S] at t if and only if c

competes with p for S [on the acceptance system of S] at t and it is
more reasonable for S to accept that p than to accept that c on [the
acceptance system of S] at t.

Lehrer provides a heuristic device to aid in our understanding of these

conditions: a scenario in which a claimant to reasonable acceptance plays a game
with a skeptic. In this justification game, the skeptic produces challenges that
compete with the statements the claimant accepts. The claimant then responds
to these challenges by defending her claims, for example, by showing that the
competing claim can be beaten. If all the challenges raised by the skeptic are
beaten, then the claimant wins the justification game and is personally justified.

Claimant: I see a snake.

Skeptic: You are dreaming you are seeing a snake.
Claimant: It is more reasonable to accept that I see a snake than that
I am dreaming that I am seeing a snake.

The point of the game is not to refute the skeptic, but to exhibit those elements
of one's acceptance system that justify the original claim.
Still, a statement need not beat all of its competitors to merit reasonable
acceptance. It is sufficient to neutralize the competition. The notion of
neutralization is defined thusly:

n neutralizes c as a competitor ofp for S on [the acceptance system

of S] at t if and only if c competes with p for S on [the acceptance
system of S] at t, but the conjunction of c and n does not compete
with p for S on [the acceptance system of S] at t, and it is as
reasonable for S to accept the conjunction of c and n as to accept c
alone on [the acceptance system of S] at t.ll

In the following justification game, a challenge is neutralized, not beaten:

Claimant: I see a snake.

Skeptic: People sometimes dream that they see snakes.
Claimant: I am not now dreaming.

The reason that the claimant's response neutralizes the skeptical challenge is that
it is as reasonable to accept the conjunction that I am not now dreaming even
though people sometimes dream that they see snakes as it is to accept that people
sometimes dream that they see snakes. Though a conjunction is less probable
than its conjunct, the additional informational content of a conjunction can make
it as reasonable to accept as the conjunct inasmuch as it can have greater
epistemic utility.
Now we are in a position to consider how Lehrer utilizes his theory to
handle the lottery paradox:

The definition of justification given above, when combined with

the formula for reasonable acceptance, yields the correct result that
I am not justified in accepting that the number one ticket has not
won. Consider the following move in the justification game:

Claimant: The number one ticket has not won.

Skeptic: The number two ticket has not won.

The skeptic has produced a competitor to my claim because, by

definition, c competes with p just in case it is more reasonable to
accept that p on the assumption that c is false than on the
assumption that c is true. If what she has claimed is false and the
number two ticket has won, then my claim must be true. On the
other hand, on the assumption that what she has claimed is true, the
probability of my claim is reduced to 98/99 because the number of
potential winners is reduced to 99. In this case, the utilities of
accepting the two claims, the skeptic's, and mine are obviously the
same, and, therefore, the comparative reasonableness of the two
claims is the same. Consequently, the skeptic's claim is not beaten,
it is as reasonable as mine, and it cannot be neutralized either. 12

The very last claim, however, demands an argument. Why should we think that
the skeptic's claim could not be neutralized? Could it not be that the conjunctive
claim that both the number one ticket and the number two ticket have not won,
while less probable than the claim that the number one ticket has not won, is at
least as reasonable in virtue of its additional informational content? It is not
obvious that the rate of increase in the utility, as we conjoin these lottery

statements, will never be as great as the rate of decrease in the probability. So,
it is not obvious that the skeptic's objection cannot be neutralized.
Whether or not Lehrer's theory is adequate to the task of giving him the
result he seeks for resolving the lottery paradox, the fundamental question is
whether he is seeking the right result. Do considerations of consistency preclude
the rational acceptance of any lottery proposition? Let us turn to an argument
that withholding on such lottery statements is analogous to withholding on many
other statements about the future, statements that we have adequate inductive
reasons to accept.



The fact that my lottery ticket is very probably a loser constitutes a good
reason for accepting that it will lose as long as merely probabilistic grounds can
suffice for reasonable acceptance that my ticket is not going to win. Such
probabilistic reasons are adequate for reasonable acceptance if lottery situations
are analogous to situations in which one has merely probabilistic, but adequate,
grounds for various empirical acceptances. There do seem to be many lottery-
like empirical claims about that which is unobserved, where our evidence,
though statistical, seems perfectly adequate for reasonable acceptance. Ifso, then
by analogy, it can be reasonable to accept that one's lottery ticket will lose.
Jonathan Vogel has suggested that lottery-like empirical statements
abound. One is the proposition that a meteorite will not hit the place where I am
sitting right now. I have no better reason to accept this than I have to accept of
any other place of comparable size on the surface of the globe that a meteorite
will not hit there at some particular time. Moreover, ifI consider a long enough
span of time, I have high confidence that a meteorite will hit somewhere and
sometime. Yet, even with no better reason to deny that the meteorite will now hit
here than I have to deny that it will hit somewhere else at some other time, I
confidently assert and accept that a meteorite will not hit me here and now.
Likewise it is reasonable for me to accept that a satellite due to fall out of orbit
will not hit my house, though pieces of it are expected to hit somewhere. Further,
it is reasonable for me to accept that I will not be burned to death by a pyroclastic
flow of hot ash from Mount Rainier while I spend a few days at a conference in
Seattle, though I fully expect that eventually such events will occur around active
volcanoes. I can reasonably accept an earthquake will not kill me during my next
visit to Los Angeles, despite my having good reason to expect there will be such
earthquakes, and resultant fatalities, sometime or other. 13
It would seem immoderately cautious and skeptical to deny that I can
reasonably accept that a meteorite will not hit me, or that the satellite falling out

of orbit will not hit my house, or that I will not be killed by a volcano while in
Seattle or an earthquake while in Los Angeles.
The skepticism will be extreme if we endorse a restricted principle of
closure for reasonable acceptance. Suppose I am subject to epistemic criticism
for failing to accept what I recognize to follow, in a deductively immediate
manner, from that which I accept. Then, if it is unreasonable to accept that a
meteorite will not imminently crush me, it is unreasonable to accept anything
that entails this negative fact, including any fact about my future life. Yet, even
if we do not accept such a limited principle of closure, it still seems
immoderately skeptical and cautious to maintain that an epistemically scrupulous
individual should refrain from accepting that a meteorite will not hit here and
now, despite reasons that should make one highly confident that this proposition
is true.
Yet we should also not follow Foley and Klein and contend that it is
rational to accept of each ticket that it will lose, despite the recognition that these
acceptances are not logically compatible with accepting that one of the tickets
will win. We should resist this position, as Kaplan and Pollock have argued,
because if recognized inconsistency does not leave one vulnerable to epistemic
criticism, then there is no dialectical point in providing a reductio ad absurdum
of someone else's position. That reductios do have epistemic weight shows that
recognized inconsistency is an epistemological vice.


It is incumbent on those who reject the above argument from analogy,
who think we can rationally accept that we will survive the next five minutes but
not that our lottery ticket will lose, to try to find some relevant disanalogy to
distinguish the lotter~ and the lottery-like cases.
Keith DeRose 4 offers an interesting proposal when considering what
differences might account for why a lottery-like proposition is knowable or
assertible while a lottery proposition is not. Since it is plausible to connect what
is assertible with what is rational to accept, one might think that his proposals
could plausibly carry over to our problem and provide those wishing to reject the
analogical argument with a relevant disanalogy.
DeRose considers a case in which I accept that the Bulls won last night
despite the fact that I know that the paper sometimes misprints the scores. I have
no better reason to think the Bulls won than I have when I read a misprinted
score that has them winning when they did not. DeRose perceives a difference
between this case and that of the lottery in the truth-values of corresponding
counterfactuals. In the lottery case, but not the score-reading case, I would have
accepted the relevant claim even had it been false. That is, I would accept that
my ticket would lose, even if my ticket were to be the one that will win. On the

other hand, it is false that I would still accept that the Bulls had won even ifthey
had not. For, had the Bulls not won, I very well might have read an accurate
newspaper account of the game and consequently have accepted that they did not
win. 15
Yet, whether an accepted statement passes or fails DeRose's subjunctive
conditional test cannot make the difference between whether or not it is rational
to accept it. Consider a revised newspaper case proposed by DeRose, in which
one knows that each day there is one, and only one, copy of the newspaper in
which all of the sports scores are inverted. Using DeRose's criterion as a
requirement for rational acceptance on this case yields the consequence that one
can rationally accept that the Bulls won on the basis of the reported score in the
paper. Yet, one cannot rationally accept that one's paper is not the defective
copy. Nonetheless, one's grounds for accepting that the Bulls won is the score
that is reported in one's copy of the paper. It would seem that if one cannot
reasonably accept that one does not have a non-defective paper, and one
recognizes that fact, then one cannot reasonably accept that the Bulls won. Since
it is epistemically reasonable to accept in these circumstances that the Bulls won,
I conclude that DeRose's test, though possibly demarcating cases of knowledge
and assertibility, will not work as a criterion for demarcating cases of rational
Another unique feature of the lottery case that might be exploited to
provide a disanalogy with the lottery-like cases is the fact that in the lottery case
one knows that there is a winning ticket. Since one knows that there is a winning
ticket of the lottery, one knows that there is some objective probability of one's
winning. So one might contend that though it may be reasonable to accept that
one will probably lose, it is not reasonable to accept that one will lose. On the
other hand, despite the probability that not all parts of the satellite will burn up
before reaching the earth's surface, one cannot take it for granted that some parts
will reach the ground. 16 Since one does not know that any parts of the satellite
will hit the ground, one cannot be certain that there is any objective probability
of my house being hit. So, perhaps this difference could allow one to claim that
in the lottery case, one cannot rationally accept that one's ticket is a loser, while
in the case of the satellite, one can reasonably accept that one's house will not
be hit.
That this disanalogy is not adequate to the task at hand is apparent,
however, if we consider a revised lottery case. Suppose that one ticket among the
million or so tickets is designated the "No Winner" ticket and is assigned to no
one, so that if that ticket is selected, no ticket wins. If it is reasonable for you to
accept that the satellite will not hit your house because you cannot be sure that
any part of the satellite will crash anywhere, then it is similarly reasonable for me
to accept that my ticket is not a winner. For, one cannot take it for granted that
any ticket is a winner, and it is very unlikely that any particular ticket will win.
Yet, since one knows that one ticket will be selected, then according to this line,

it is still unreasonable for me to accept that my ticket was not selected. So, I can
reasonably accept that my ticket will lose, but not reasonably accept that it was
not selected. This cannot be right. For, I know, by an immediate inference, that
if my ticket is not the winner then it was not selected, since I know that it is not
the No Winner ticket. So, if it is reasonable to accept that my ticket will lose,
then it is reasonable to accept that my ticket was not selected. It follows, that
whether or not I know that some lottery ticket will lose cannot matter in the
reasonability of accepting whether a particular ticket will lose.
Good analogies cut both ways. If you are convinced you do not have good
reason to accept that your lottery ticket is a loser, you could use the analogy to
lottery-like cases to argue that you similarly have no good reason to accept that
your house will fail to be hit by the satellite. Dana K Nelkin, in seeking to find
a difference between lottery situations and cases in which inductive reasoning
based on perception is justified, argues that the feature of lottery inferences that
makes them epistemically unjustified lies precisely in the purely statistical nature
of the evidence. Thus, on Nelkin's account, there is no difference between the
lottery situations and the lottery-like situations, so long as one's evidence in both
cases is purely statistical. Nelkin argues that statistical inferences are only
acceptable when grounded in presuppositions of a causal explanation of the
statistical evidence.
The intuitive costs of this account are evident when Nelkin considers a
case proposed by Gilbert Harman. I? If Mary will be in Trenton only if she wins
the lottery, and in New York otherwise, then if one cannot rationally accept that
Mary will not win the lottery, it would seem that one could not rationally accept
that Mary will be in New York. Nelkin responds by conceding this implausible
consequence, and responds that such situations are relatively rare.
Nelkin's position is immoderately skeptical. If it is not rationally
permissible to accept that your house will not be hit by a satellite or a meteorite,
on purely statistical grounds, then if a limited closure of recognized immediate
inferences holds, then it would seem to be correspondingly impermissible to
accept that which presupposes that we will not be hit by a meteorite, and then it
is difficult to imagine much of anything in our future about which one could
form a reasonable expectation.
Moreover, even if limited closure fails, apparently justified statistical
inferences are not all that rare, and not confined to the future. If you were to tell
me that you randomly flipped twenty heads in a row yesterday. I would infer that
you were flipping for a long time. I do so on the basis of statistical
considerations, and not because I accept that there is a causal connection between
flipping a coin for a long time and getting twenty heads in a row. It would seem
that such purely statistical inferences are both common and reasonable.
Using a generalization of Keith Lehrer's Racehorse Paradox,18 John
Pollock has provided a technical argument to show how one can turn any
statistical syllogistic inference from high probability into a lottery-like form. 19

Since Pollock incorporates a principle of symmetry into his analysis of the

lottery to show that any lottery inference to accepting that any particular ticket
will lose is "collectively defeated," and thus unreasonable, he needs some
difference between lottery situations and lottery-like situations to avoid
immoderate inductive skepticism. He suggests that the difference lies in the
projectibility or non-projectibility of disjunctive predicates. Nonetheless, he
admits that this is only to label the difference, and not to provide an account of
it, since projectibility is understood in terms of which inferences are
epistemically acceptable to draw. 20 Moreover to adopt his analysis of the lottery
case in terms of "collective defeat" would be to presuppose the very issue at
stake in this paper: the principle of symmetry.
Absent any other account of why we have reasonable acceptance in the
case of the meteorites, earthquakes, and volcanoes, but not in the case of the
lottery, the lottery and the lottery-like cases would seem analogous. It is
epistemically reasonable for me to accept that my lottery ticket will lose, just as
it is epistemically reasonable for you to accept that a satellite will not hit your
house tonight.


Both horns of the lottery dilemma presuppose the symmetry principle: If
we have no better reason to accept a statement than to accept another, then our
attitude toward each should be the same. We should either accept both or accept

Symmetry Principle (for Reasonable Acceptance):

If S recognizes at t that p and q are equally reasonable for S to
accept at t on the basis of the acceptance system of Sat t then it is
reasonable for S to accept that p at t on the basis of the acceptance
system of S at t only if it is reasonable for S both to accept that p
and to accept that q on the basis of the acceptance system of S at t.

In the lottery, this throws us back onto our two horns: either we do not accept of
any particular ticket that it will lose, or we accept of all tickets that they will lose
despite our recognizing the inconsistency with our accepting that some ticket will
Gilbert Harman considers a via media:

... To say one can infer this of any ticket is not to say one can infer
it of all. Given that one has inferred ticket number 1 will not win,
then one must suppose the odds against ticket number 2 are no
longer 999,999 to 1, but only 999,998 to 1. And after one infers

ticket number 2 won't win, one must change the odds on ticket
number 3 to 999,997 to 1, and so on. If one could get to ticket
number 999,999, one would have to suppose the odds were even,
1 to 1, so at that point the hypothesis that this ticket will not win
would be no better than the hypothesis that it will win, and one
could infer no further. (Presumably one would have to have
stopped before this point.)21

There is thus no inconsistency in accepting that some tickets will lose while not
accepting that of others. Yet, that is not to say that such arbitrary choices are
epistemically reasonable. Indeed, they will not be, if the symmetry principle
Harman seems to endorse a symmetry principle for epistemic acceptance,
for he takes non-arbitrariness to be a mark that distinguishes theoretical from
practical choices:

Theoretical and practical reasoning differ in this respect. In

practical reasoning one can be justified in satisficing even in
choosing among competing plans at the same level. In fact, often
this is just what one should do-make an arbitrary choice of a
satisfactory plan to accomplish one's goals. But in theoretical
reasoning one would not be justified in making an arbitrary choice
of what to believe among competing hypotheses at the same level. 22

John Pollock maintains that confusion between practical and epistemic

reasoning underlies any temptation to give up a symmetry principle:

The preceding considerations suggest that the controversy over

skeptical and credulous reasoning stems from a confusion of
epistemic reasoning (reasoning about what to believe) with
practical reasoning (reasoning about what to do). In practical
reasoning, if one has no basis for choosing between two alternative
plans, one should choose at random. The classical illustration of
this is the medieval tale of Buridan's ass who starved to death
standing midway between two equally succulent bales of hay
because he could not decide from which to eat. This marks an
important difference between practical reasoning and epistemic
reasoning. An agent making practical decisions must first decide
what to believe and then use that in deciding what to do, but these
are two different matters. If the evidence favoring two alternative
hypotheses is equally good, the agent should record that fact and
withhold belief.23

In arguing against the acceptability of reasoning in the credulous manner

described by Harman, Dana Nelkin elides this sharp distinction between practical
and theoretical reasoning and offers a pragmatic consideration in defense of

Suppose, with Harman, that Jim can rationally infer that t1 through
t999,000, say, will lose. Jim is also rational in believing that one of
t1 through 1,000,000 will not lose. It would seem to follow that Jim
could rationally infer a logical consequence of these beliefs,
namely, that one oft999,001-tl,000,000 will not lose. But this is
strongly counterintuitive. If Jim were rational in believing that one
of those 1,000 tickets will win, then, depending on the order of his
inferences, he should either try to get his hands on one of those
1,000 tickets or feel fortunate to be holding one of them already!
But there is no reason for Jim to do either of these things. Thus,
Harman's argument fails ... 24

Plausibly, a rejection of the principle of symmetry should not provide this

much license. For one could reject symmetry, and accept of several tickets that
they will lose without practicing a form of epistemic brinksmanship and
accepting that all but a particular 1,000 will lose.
More importantly, however, it is to misunderstand the acceptance system
of such a reasoner to think that there need be any acceptance of a proposition to
the effect that the tickets one accepts to be losers are objectively different from
the tickets one does not accept to be losers. Obviously, not to accept that a ticket
will lose is not to accept that it will not lose. Less obviously, not to accept that
a ticket will lose is not to have less confidence that it will lose. To treat the
difference between accepting that one lottery ticket will lose and not accepting
that another will lose as implying a distinction in the reasons for accepting that
the one or the other will lose, i.e., in the comparative confidence levels one has
that one or the other will lose, is tantamount to presupposing the symmetry
principle itselfl
Let us return, then, to the charge that to defend the reasonability of
rejecting the symmetry principle is to confuse pragmatic and epistemic reasons.
Can such arbitrariness be justified only pragmatically and never epistemically?
The answer depends on the nature of epistemic justification. Those who take a
deontological approach to epistemic normativity might insist that the principle
of symmetry has a priori plausibility.25
It is less clear why those who take a more consequentialist approach to
epistemic justification should be similarly moved to accept symmetry at the price
of foregoing the chance to accept highly probable and plausible lottery
propositions (and perhaps also lottery-like propositions). It is a familiar theme
of Keith Lehrer's work on epistemic acceptability that "the goal of acceptance

is to obtain truth and avoid error.,,26 If truth is our ultimate aim, why can we not
have purely epistemic reasons to reject symmetry and accept some lottery
Are there philosophers who question symmetry for what appear to be
purely epistemological reasons? It would seem so. In "Sellars on Induction
Reconsidered," Lehrer displays a new interpretation of Wilfrid Sellars's view of
empirical acceptance rules, a view that violates a non-arbitrariness condition that
is implied by the Symmetry Principle. Here is how Lehrer describes Sellars's

If one is only concerned with getting a maximum of true

observation statements, then it is reasonable to accept all of the
statements' a] is B,' 'a2is B,' and so forth to 'a" is B.' In the interest
of accepting true observation statements, this policy is warranted
when the probability that an a] is B is greater than 112. But, and this
is the heart of the new interpretation, such acceptance is only prima
facie reasonable. When we ask, not what is prima facie reasonable,
but what is reasonable sans phrase, the answer will be different.
For then we must be concerned, not only with accepting as many
true observation statements as possible, but also with maintaining
coherence among statements of different types, for example,
observation statements and generalizations about the population,
and this will preclude us from accepting all the observation
statements mentioned so as to avoid inconsistency.

Thus, according to the new interpretation, this way to combine the objectives of
accepting true observation statements and at the same time maintaining
explanatory coherence is to accept the percentage ofthe observation statements,
provided it is greater than 50%, that corresponds to the percentage of members
ofK that are known to be B in the total population. For example, if we know that
3/4 K are Band K satisfies the pertinent relevance conditions, then we accept
75% of the statements of the form' a] is B. ,27
While Lehrer maintains his own commitment to symmetry, he does allow
that Sellars's position is "justified by systematic epistemic objectives and is in
no way ad hoc ... ,,28 I agree. Sellars's position demonstrates that rejection of
symmetry can be grounded in purely epistemological, not pragmatic,
In rejecting symmetry, we need not endorse a position akin to Sellars's.
If we did, we would license accepting of all but one ticket that it lose! That is
obviously too permissive. Yet, is there not good reason to accept of a few tickets
that they will lose, while not getting carried away? The only barrier in our way
is the Symmetry Principle.

Yet, if it is permissible to accept of any lottery ticket that it will lose, is it

not permissible to accept that they all are losers? If this is not just another appeal
to the Symmetry Principle, then there is a confusion underlying the question.
Permissibility is not adjunctive. I may be permitted to have any fruit in the fruit
basket but not be permitted to have them all. My being permitted to have the pear
may depend upon whether I have already exercised my permission to have the
banana. Similarly, what is reasonable to accept is contingent on what is already
Since Lehrer's analysis of personal justification does not presuppose the
Symmetry Principle, it has the flexibility to provide an analysis of personal
justification for the more credulous reasoner. This, by my lights, is a virtue of his
analysis. 29 To see how we can use Lehrer's analysis for less skeptical purposes,
consider again the justification game, this time played between a skeptic and a
credulous reasoner:

Claimant: The number one ticket has not won.

Skeptic: The number two ticket has not won.
Claimant: It is as reasonable for me to accept both that the number
two ticket has not won and that the number one ticket has not won
as to accept the former alone. (While the conjunction is less
probable than either conjunct, the added informational content
outweighs the negligible additional risk of error.)

The claimant has won this round by neutralizing the skeptic's challenge. The
claimant's original claim, 'The number one ticket has not won', neutralizes the
skeptic's challenge, inasmuch as the conjunction of their two claims is at least
as reasonable as the claimant's original claim. Of course it could happen that this
conjunction of the objection with the original claim does not neutralize the
objection. This will happen if the increase in informational content due to
conjoining does not outweigh the decrease in probability. For example, if the
lottery is not very large, then the probability of these two tickets losing could be
substantially less than the probability of one ticket losing. In a three-ticket
lottery, the probability is thereby halved. It follows that in a small lottery one is
not personally justified in accepting that one holds a losing ticket. That is as it
should be.
The game could continue with the claimant winning round after round by
neutralizing the opposition:

Claimant: The number one ticket has not won.

Skeptic: The number two and number three tickets have not won.
Claimant: It is as reasonable for me to accept that the tickets
numbered one, two and three have not won as to accept just that the
number one ticket has won. (While the conjunction is less probable

than either conjunct, the added informational content outweighs the

negligible additional risk of error.)

For some large n, the increase in informational content gained by

accepting that every ticket among the n tickets will lose is offset by the
corresponding decreased probability. Consider this continuation of the game:

Skeptic: The number two through number n tickets (for some large
n) have not won.
Claimant: It is more reasonable for me to accept that the number
one ticket has not won than to accept that the number two through
n tickets have not won.

The claimant wins this round by beating the skeptic's challenge.

What happens in the transition from neutralizing to beating? Suppose the
expected utility of accepting that ticket one will lose is equal to the expected
utility of accepting that tickets one through m will lose. Suppose further that the
skeptic attempts the following objection:

Skeptic: The number two through m tickets have not won.

On our first supposition, it is no less reasonable to accept that ticket one will lose
on the assumption that ticket two through m will all lose than on the assumption
that they will not all lose. Consequently, this objection of the skeptic does not
compete with the original claim and is thus not a legal move for the skeptic in the
justification game.
We might worry about what happens on precise boundaries. We should
not. Given the limitations of human reasoners to make very fine epistemic
distinctions, the supposition of one ticket's losing will only make the relevant
epistemic difference if the lottery is too small for one to reasonably accept that
a particular ticket will lose. In a very large lottery, if we can successfully judge
that an objection by the skeptic is indeed a competitor, then we will be able to
neutralize or beat it. If one cannot judge whether an objection competes, then it
should not defeat one's personal justification. 30
Notice that up to this point, the Symmetry Principle has not come into
play. So, let us consider these moves:

Skeptic: You do not accept that the number two ticket has not won
but you have no better reason to accept that the number one ticket
has not won.
Claimant: The Symmetry Principle is false.

So long as the claimant is personally justified in rejecting the Symmetry

Principle, the claimant has successfully neutralized the skeptic's last challenge.
There are good theoretical reasons to reject the Symmetry Principle. We
accept in order to gain truth and avoid error. Basing acceptance on purely
statistical grounds can be epistemically reasonable when the prospects for gain
are great and the risks of loss small. A concomitant commitment to coherence
provides the epistemological basis for our avoiding recognized inconsistency.
Thus, I commend a position of credulous consistency as the most plausible of the
three resolutions of the lottery paradox for reasonable acceptance. The lesson of
the lottery is that we can sometimes have no better reason to accept, but reason

I Keith Lehrer, Metamind, Oxford, Clarendon Press, 1990, pp. 235-6 and Theory of Knowledge,

Boulder: Westview Press, 1990, pp. 129-30.

2 Mark Kaplan, "Believing the Improbable, " Philosophical Studies 77 (1995), pp. 117-46,

particularly pp. 136-7 and Decision Theory as Philosophy, Cambridge: Cambridge University,
1996, pp. 139-40.
lJohn Pollock, Cognitive Carpentry: a Blueprintfor How to Build a Person, Cambridge, Mass.:
MIT/Bradford, 1995, chapter 2.
4Sharon Ryan, 'The Epistemic Virtues of Consistency," Synthese 109 (1996), pp. 121-41,
particularly p. 126. Ryan is working with a strong notion of justified belie( i. e., whatever
justification knowledge requires. In the lottery paradox I am considering, the relevant concept of
justification ranges from a weak notion of epistemic rational permissibility to a concept of
justification no stronger than Lehrer's notion of personal justification.
S DanaK. Nelkin, "The Lottery Paradox, Knowledge, and Rationality," Philosophical Review 109
(2000) pp. 373-409.
6Richard Foley, Working Without a Net, New York: Oxford, 1993, pp. 164-5.
7Peter Klein, "The Virtues of Inconsistency," Monist 68 (1985), pp. 105-35, particularly p 108.
Note that Klein, like Ryan, is presuming a notion ofjustified beliefthat is required for knowledge.
R I borrow the labels' skeptical' and 'credu lous' for the two positions one can take on symmetry

from John Pollock, "Justification and Defeat," Artificial Intelligence 67 (1994), 377-407, who in
turn acknowledges borrowing from D. S. Touretzky, J. F. Horty, and R. H. Thomason, who speak
of credulous and skeptical reasoners.
9 Bayesians who reject the notion of acceptance in favor of quantitative measures of confidence
would reject all three positions. I will not rehearse the standard replies to such Bayesians.
10 Theory of Knowledge, p. 118.
II Theory of Knowledge, p. 125. These definitions of competition, neutralization, and personal
justification are essentially preserved in Lehrer, Self Trust: A Study of Reason, Knowledge, and
Autonomy, Oxford: Clarendon Press, 1997, p. 30.
12 Theory of Knowledge, p. 130.
13 See Jonathan Vogel in "Are There Counterexamples to the Closure Principle':>" in Doubting,
Michael D. Roth and Glenn Ross (eds.), pp. 13-27, for several such cases where we would
ordinarily claim to have knowledge though they appear to be analogous to lottery situations.
14 Keith DeRose, "Knowledge, Assertion and Lotteries," Australasian Journal of Philosophy 74

(1996) pp. 568-580.


15 For this account to succeed, we must reject Robert Stalnaker's principle of conditional excluded
middle for counterfactuals.
16 It is just this disanalogy that Sharon Ryan exploits in (1996).

17 Gilbert Harman, Change in View: Principles 0/ Reasoning, Cambridge, Massachusetts:

Bradford/MIT, 1986, p. 71.

18 Keith Lehrer, "Coherence and the Racehorse Paradox," Midwest Studies in Philosophy, (1980)
5, pp. 183-192.
19 John L. Pollock, "How to Use Probabilities in Reasoning," Philosophical Studies 64 (1991), pp.

65-85. Also see Pollock's "A Solution to the Problem ofInduction," Noiis 18 (1984), pp. 423-461
and "Justification and Defeat," Artificial Intelligence 67 (1994), pp. 377-407.
20 Pollock (1991), pp. 80-1.
21 Change in View, p. 71.
22 Change in View, p. 68.
23 Pollock (1994), p. 383.
24 Nelkin (2000), pp. 377-8.

25 Epistemological deontologists might also demur from such a defense of symmetry, seeing
considerations of high probability and coherent acceptance as trumping aprima/acie reasonability
behind symmetry.
26 Theory o/Knowledge, p. 121.
27 Lehrer (1983), p. 471.
28 Lehrer (1983), p. 469.
29 Note that this flexibility to accommodate more credulous intuitions on the lottery is not shared
by Pollock's (1994) notion of justification, inasmuch as his "Principle of Collective Defeat" (p.
383) incorporates a symmetry principle.
30 The argument can be fully spelled out as follows: Suppose the skeptic offers this challenge:
Skeptic: The number two through m tickets have not won.
There are three cases to consider:
(a) The expected epistemic utility of accepting that ticket one will
lose is less than the expected epistemic utility of accepting that all
of the tickets one through m will lose. In this case, the statement
that ticket number one will lose neutralizes the challenge, as in
the first two rounds above.
(b) The expected epistemic utility of accepting that ticket one will
lose is equal to the expected epistemic utility of accepting that all
of the tickets 1 through m will lose. In this case, the statement that
tickets numbered two through m will all lose does not compete
with the original claim ofthe claimant. It is no less reasonable to
accept that ticket one will lose on the assumption that ticket two
through m will all lose than on the assumption that they will not
all lose.
©) The expected epistemic utility of accepting that ticket one will
lose is greater than the expected epistemic utility of accepting that
tickets one through m will lose. There are two sub-cases:
(c-i) The expected utility of accepting that ticket one will lose is greater than the
expected epistemic utility of accepting that tickets two through m will lose. In this
case, the statement that ticket number one will lose beats the challenge, as in the
above case for large n.
(c-ii) The expected utility of accepting that ticket one will lose is less than or
equal to the expected utility of accepting that tickets two through m will lose.
Given the limitations in our abilities to make fine discriminations in the expected
utility for the acceptance of statements, these cases will not arise unless the lottery
is small enough for one ticket to make such a difference. So, in a very large
lottery, these cases will not arise for human reasoners. If we can successfully

judge that an objection by the skeptic is indeed a competitor, then we will be able
to neutralize or beat it. If we cannot judge whether an objection competes, then
it does not defeat our personal justification.
Chapter 7



Charles B. Cross*
University of Georgia

According to Keith Lehrer's (1974, 1986, 1988, and 2000) theory of knowledge,
coherence is the basis of epistemic justification. In Lehrer's theory, coherence
is a relation between a proposition and what Lehrer (2000: 170) defines as an
evaluation system:

D 1. A system X is an evaluation system of S if and only if X contains

(a) states expressed by statements of the form, 'S accepts that p',
attributing to S just those things that S accepts with the objec-
tive of accepting that p if and only if p (the acceptance system
of S), (b) states expressed by statements of the form, 'S prefers
accepting p to accepting q', attributing to S just those things that
S prefers accepting with the same objective concerning accep-
tance, (the preference system of S), and (c) states expressed by
statements of the form, 'S reasons from p, q, r, and so forth to con-
clusion c', attributing to S just those states of reasoning with the
objective of being sound (having true premises and being valid).
Based on the notion of an evaluation system, Lehrer (2000: 170) defines justifi-
cation in terms of coherence as follows:
D2. S is justified in accepting p at t on system X of S at t if and only
if p coheres with X of S at t.
Now, to the question, "What conclusions should one draw from the propositions
one accepts?" it would seem sensible to answer, "One should draw those con-
clusions whose acceptance would be justified on the basis of what one accepts."
E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 109-127.
© 2003 Kluwer Academic Publishers.

But if this is right, then Lehrer's account of justification should be interpretable

not only as a theory of justification but as a theory of inductive inference, too.
In what follows I shall investigate the consequences of interpreting Lehrer's
account of system-relative justification as a theory of inductive inference. In
Section 1, I set out Lehrer's theory of coherence. In Section 2, I present the ele-
ments of the theory of cumulative reasoning, I a widely known and well worked-
out proof theory for inductive reasoning. In Section 3, I formulate a definition
of inductive inference in terms of a simplified version of Lehrer's analysis of
coherence, and I discuss what assumptions about coherence would be sufficient
to make the account of inductive inference derivable from his theory of justifica-
tion conform to a series of widely discussed general principles, including those
constitutive of cumulative reasoning. In Section 4, I consider the epistemologi-
cal significance of the theory of inductive reasoning developed in Section 3.


Having defined justification in terms of coherence, Lehrer (2000: 170) fleshes

out the notion of coherence in the following reformulation of D2:

D3. S is justified in accepting pat t on system X of Sat t if and only

if all objections to p are answered or neutralized for S on X at t.

Thus, on Lehrer's account, p coheres with X of Sat t if and only if all objections
to p are answered or neutralized for S on X at t. Lehrer (2000: 170) defines the
notions of objection, answering, and neutralizing as follows:

D4. 0 is an objection to p for S on X at t if and only if it is more

reasonable for S to accept that p on the assumption that 0 is false
than on the assumption that 0 is true, on X at t.

D5. An objection 0 to p is answered for S on X at t if and only if 0 is

an objection to p for S on X at t, and it is more reasonable for S
to accept p than to accept 0 on X at t.

D6. n neutralizes 0 as an objection to p for S on X at t if and only if

o is an objection to p for S on X at t, the conjunction of 0 and n is
not an objection to p for S on X at t, and it is as reasonable for S
to accept the conjunction of 0 and n as to accept 0 alone on X at t.

D4, D5, and D6 appeal to an undefined notion of relative reasonableness that

Lehrer takes as primitive. 2 The following example, which adapts the example
in Lehrer 1988:342 to Lehrer's (2000) new terminology, applies this notion of
reasonableness in an illustration of D5.

Wendy tells Keith that Fred Feldman's book is now in print. Is

Keith justified in believing this? Keith's background information
includes that he edits a series in which the book appears and that
Wendy works for the publisher and is trustworthy about when
books appear in the series. Now suppose that some skeptic pro-
poses to Keith that Wendy is lying. The claim that Wendy is lying
is an objection to the claim that the book is in print, but, on the
basis of Keith's background information, it is more reasonable for
Keith to accept that Wendy is telling the truth and that the book
has been published than to accept that Wendy is lying.

In this example, the claim Wendy is lying is an objection to the claim Feldman's
book is in print because it is more reasonable for the subject to accept Feldman's
book is in print on the assumption Wendy is not lying than on the assumption
Wendy is lying. Since it is more reasonable for the subject to accept The book
has been published than to accept Wendy is lying, the claim Feldman's book is in
print answers the objection Wendy is lying. But is every objection to Feldman's
book is in print answered? Consider this example, again adapted (with updated
terminology) from the example in Lehrer 1988:342:

Imagine the skeptic persists and says, "Well, you know people
sometimes lie about when books are in print." Now this skeptical
innuendo does count as an objection to the claim Keith believes in
a sort of indirect way. It would be more reasonable for Keith to ac-
cept that Feldman's book is in print on the assumption that people
do not ever lie about when books are in print than on the assump-
tion that they do sometimes lie. Moreover, it is quite reasonable
for Keith to accept that people do sometimes lie about these mat-
ters. But this skeptical innuendo, though it cannot be answered,
can be neutralized by conjoining the reply that Wendy is not lying.
Of course, the reasonableness of accepting the latter depends on
Keith's background information about Wendy.

The claim Wendy is not lying neutralizes the claim People sometimes lie about
when books are in print as an objection to Feldman's book is in print because
People sometimes lie about when books are in print, but Wendy is not lying is
not an objection to Feldman's book is in print, and because it is as reasonable to
accept People sometimes lie, but Wendy is not lying as to accept People some-
times lie by itself, given the subject's background evaluation system. If every
objection to Feldman's book is in print is either answered (like the claim Wendy
is lying) or neutralized (like the claim People sometimes lie about when books

are in print), then the subject is justified in accepting that Feldman's book is in
This example illustrates how natural it would be to interpret Lehrer's nec-
essary and sufficient condition for justification relative to an evaluation system
as a sufficient condition for the permissibility of an inference. The subject Keith
in the previous example who asks, "Am I justified in believing that Feldman's
book is in print?", may be using certain background information and skeptical
alternatives to evaluate the epistemological status of a belief he already holds,
or he may be considering whether to accept Feldman's book is in print as a new
belief on the basis of that background information. But to consider whether to
accept a given claim on the basis of given background information is to consider
whether to infer that claim from the background information. This would be an
inductive inference, of course, since the claim Feldman's book is in print does
not follow deductively from the subject's background information, but the fact
that the subject would be justified in believing this claim surely indicates that
it would be a good inductive inference. What are the consequences of reinter-
preting Lehrer's account of justification relative to an evaluation system as an
account of inductive inference? In order to answer this question we will first
need an introduction to the theory of inductive inference in its most general
form: the theory of cumulative reasoning.


Systems of deductive inference (for example, classical first-order logic) satisfy

three well-known conditions often labeled Reflexivity, Monotonicity, and Cut 3

Reflexivity If p E X then X f- p.

Monotonicity If X f- p and X r;;:; Y, then Y f- p.

Cut If X U {p} f- q and X f- p then X f- q.

But most of the reasoning subjects actually do on a day-to-day basis (such as in-
ference to the best explanation) is inductive. Despite its nondeductive character,
it is possible to theorize in a proof-theoretic way about inductive reasoning, and
formal approaches to inductive reasoning abound in the artificial intelligence
literature. 4 Formally, the most conspicuous way in which inductive inference
differs from deductive inference is the fact that inductive inference is nonmono-
tonic. Where 'X ~ p' means that p is an inductive consequence of X (under
some given conception of inductive inference represented by '~'), it is possible
for it to be the case that X ~ p and X r;;:; Y even though Y If p. For example,

let X, Y, andp be the following:

X = {'This match was struck', 'Oxygen was present'},

Y = X U {'This match was wet when struck'},
p = 'This match lit'.

Knowing that the match was struck and that oxygen was present, we can rea-
sonably infer that the match lit. But if our evidence is expanded to include
information that the match was wet, then the inference to the conclusion that
the match lit is not licensed given the expanded evidence.
If in the above list of conditions Monotonicity is replaced by the weaker
condition of Cautious Monotony, the result is a version of what Gabbay (1985)
calls cumulative inference:

Reflexivity If p E X then X ~ p.

Cautious Monotony If X ~ p and X ~ q, then X U {p} ~ q.

Cut If Xu {p} ~ q and X ~ p then X ~ q.

Makinson (1989) and Kraus, Lehmann, and Magidor (1990) present semantical
approaches to cumulative inference based in part on Shoham's (1988) notion
of preferential entailment. By investigating how Lehrer's notion of relational
coherence can be interpreted as an account of inductive reasoning we will be
developing an alternative to the received preferential model semantics for cu-
mulative inference.



In our application of Lehrer's conception of justification below we will assume

that the time variable t and epistemic subject variable S are fixed and so can be
Since in the context of inference-making what will matter about an evalu-
ation system is the set of propositions accepted in the acceptance system compo-
nent, we will depart from Lehrer's use of the variable X. Specifically, instead
of assuming that X represents a triple consisting of an acceptance system, a
preference system, and a reasoning system, we will assume that X represents
the sentences which formulate the claims that are accepted in the acceptance
system component of an evaluation system. In other words, X will range over
the sets of sentences expressing the p's that S accepts in a range of possible

evaluation systems associated with some fixed S and t. Now, different evalua-
tion systems in Lehrer's theory can share the same acceptance system, hence a
subject's evaluation system at a given time is not strictly a function of his or her
acceptance system. But in the coherence-based theory of inductive consequence
to be explored below, the inductive consequences associated with a given eval-
uation system will be a function of the acceptance system component alone.
Accordingly, the theory of inductive reasoning presented below may appear to
assume that a subject's reasoning and preference systems are a function of his
or her acceptance system. How can this be reconciled with the spirit of Lehrer's
theory? The conflict is only apparent. When two evaluation systems incorporate
the same acceptance system but not the same preference and reasoning systems,
this means that the two evaluation systems in question are associated with dif-
ferent relations of comparative reasonableness relative to the given acceptance
system. In the theory of inductive consequence developed below, the inductive
consequences of X are determined in part by comparative reasonableness rela-
tive to X. When two evaluation systems incorporate the same acceptance sys-
tem but not the same preference and reasoning systems, this means that the two
evaluation systems in question are associated with different notions of inductive
consequence. Lehrer's theory of evaluation systems is therefore, potentially, a
theory of the dynamics of the inductive inference concept-a theory of induc-
tive logic revision. We shall not offer a dynamic theory here. The theory to be
presented below will concern the statics of coherence-based inductive inference.
Since we are taking the variables S and t as fixed, Lehrer's seven-place
relation of comparative reasonableness ("it is more reasonable for S to accept
that p on the assumption that q than to accept r on the assumption that s on the
basis of X at t") will become a five-place relation ("it is more reasonable to
accept p on the assumption q than to accept r on the assumption s on the basis
of X,,).5 These modifications allow us to define 'X ~ p' as in C1 below and
work with C2-C6 instead of Lehrer's D2-D6. We will assume a background
deductive logic whose consequence relation is represented by the straight turn-
stile 'f-'. 6 We will assume that this background logic is a compact, consistent
extension of classical tmth-functionallogic and that an expressively complete
set of boolean connectives is available in the language:

C 1. Where p is a sentence and X a set of sentences, X ~ p iff either

X f- or p is justified on the basis of X.

C2. p is justified on the basis of X if and only if p coheres with X.

C3. p coheres with X if and only if all objections to p are answered or

neutralized on X.

C4. 0 is an objection to P on X if and only if it is more reasonable to

accept p on the assumption of --'0 than on the assumption of 0 on
the basis of X.

C5. p answers 0 on X if and only if 0 is an objection to p on X, and it

is more reasonable to accept p than to accept 0 on X.

C6. n neutralizes 0 as an objection to p on X if and only if 0 is an

objection to p on X; and 0 & n is not an objection to p on X; and
it is as reasonable to accept 0 & n as to accept 0 alone on X.

Since the account of cumulative inference to be given below is developed with

classical logic as a background, Cl is formulated in such a way that if X is
logically inconsistent then X f-v p for all p. Coherence figures in whether X f-v
p only if X is logically consistent. It would be interesting to see a paraconsistent
version of the logic developed here, but that is a project for another occasion.

3.1. Basic Postulates aud Related Results

Given C 1-C6, the key to the logic of 'f-v' will of course be the logic of the com-
parative reasonableness relation. Lehrer (2000:144) states that he takes com-
parative reasonableness as undefined, but he identifies both probability (Lehrer
2000:144) and epistemic utility (Lehrer 2000:146) as factors in the determina-
tion of the degree of reasonableness of a claim. Lehrer (2000: 146) even provides
an expected utility equation for calculating degrees of reasonableness. Our ap-
proach to comparative reasonableness will be entirely nonquantitative. 7 We will
treat comparative reasonableness relative to a set X of sentences as a binary re-
lation >- x relating one pair (p, q) of sentences to another such pair. A locution
of the form 'It is more reasonable to accept p on the assumption of q than to
accept r on the assumption of s on the basis of X' will thus be abbreviated
'(p, q) >- x (r, s)'. With comparative reasonableness understood qualitatively,
the following all seem plausible as principles of comparative reasonableness:

Noutriviality If (p, q) >- x (r, s) then X }L .

Asymmetry If(p,q) >-x (r,s) then (r,s) 'Ix (p,q).

Irreftexivity If X }L, then (p, q) 'I x (p, q).

Negative Transitivity If (PI, ql) 'I X (p2, q2) and (p2, q2) 'I X (p3, q3) then
(PI, ql) 'I x (P3, q3).
Equivalence If p -Jf- p' and q -Jf- q' and r -Jf- r' and s -Jf- s', then (p, q) >- X
(r,s) iff (p',q') >-x (r',s').

Superiority If X f- p and X .J.L, then (p, T) »- x (q, T) for all objections q to p


Enlargement for Objections If X ~ p, then every objection to q on X is an

objection to q on X U {p}.

Enlargement for Answering If X ~ p and 0 is an objection to q on both

X and Xu {p}, then q answers 0 on.X only if q answers 0 on Xu {p}.

Enlargement for Neutralizing If X ~ p and 0 is an objection to q on both X

and X U {p}, then for all n, n neutralizes 0 as an objection to q on X
only if n neutralizes 0 as an objection to q on X U {p}.

Reduction for Objections If X ~ p, then every objection to q on X U {p} is

an objection to q on X.

Reduction for Answering If X ~ p and 0 is an objection to q on both X and

X U {p}, then q answers 0 on X U {p} only if q answers 0 on X.

Reduction for Neutralizing If X ~ p and 0 is an objection to q on both X and

X U {p}, then for all n, n neutralizes 0 as an objection to q on X U {p}
only if n neutralizes 0 as an objection to q on X.

Nontriviality conveys the idea that distinctions of comparative reasonableness

cannot be made if X is inconsistent. 8 Asymmetry and Irreflexivity are supported
intuitively by the more in more reasonable than. To expose the significance of
Negative Transitivity, let us next make the following definition:

(p,q) ~x (r,s) ifandonlyif (p,q) 'l-x (r,s) and (r,s) 'l-x (p,q).
The following result follows immediately by Nontriviality, Irreflexivity, Asym-
metry, and Negative Transitivity:

Theorem 1 For any fixed X, ~x is an equivalence relation on the set ofpairs

of the form (p, q).

That is, ~ x is reflexive, symmetric, and transitive for any fixed X. The fact that
~ x is an equivalence relation makes it acceptable to interpret' (p, q) ~ x (r, s)'
as meaning it is as reasonable to accept p given q as to accept r given s on the
basis of X. Where q and s are both logical truths, '(p, q) ~ x (r, s)' means
p is as reasonable as r on X, and it is this notion of equal reasonableness that
we appeal to in C6. The fact that ~x is an equivalence relation also yields the
following useful substitution principle.

Theorem 2 If (PI , qI) '::::.X (p2, q2) and (rIl 81) '::::.x (r2' 82), then (PI, qr) >- x
(rI' 8r) iff(p2, q2) >- X (r2' 82).
Proof: Assume that (PI, qI) '::::.x (p2, q2) and (rI, 8r) '::::.x (r2, 82). Now for re-
ductio suppose that (PI, qr) >- X (rI' 8r), but (p2, q2) >I- X (r2' 82). By hypothe-
sis, (PI, qI) >I- X (p2, q2) and (p2, q2) >I- X (r2' 82) and (r2' 82) >I- X (rI' 81). By
two applications of Negative Transitivity, (PI, qr) >I- X (rI, 81), which is con-
trary to hypothesis. So by reductio if (PI, qI) >- X (rr, 81) then (p2, q2) >- X
(r2' 82)' An exactly similar argument shows that if (p2, q2) >- X (r2, 82) then
(PI, qI) >- X (rI' 8r). (QED)
Equivalence has this direct consequence:

Theorem 3 If X f-v P and P -1r q, then X f-v q.

Proof: Suppose that X f-v P and P -1r q. If X r then X f-v q follows imme-
diately, so suppose X j,L. Then every objection to P on X is either answered or
neutralized. Let 0 be an obj ection to q on X. Then (q, -'0) >- X (q, 0), but since
P -1r q it follows by Equivalence that (p, -'0) >- X (p, 0). Hence 0 is an objection
to p, so 0 is either answered or neutralized as an objection to P on X. Suppose
that 0 is answered as an objection to P on X. Then (p, T) >- x (0, T), but since
P -1r q, it follows by Equivalence that (q, T) >- x (0, T). Hence 0 is answered
as an objection to q on X. Alternatively, suppose that n neutralizes 0 as an ob-
jection to P onX. Then (p, -,(o&n)) >l-x (p, o&n) and (o&n, T) '::::.x (0, T).
Since P -1r q, it follows by Equivalence that (q, -, (0 & n)) >I- x (q, 0 & n), so
n neutralizes 0 as an objection to q. So 0 is either answered or neutralized as an
objection to q on X. Generalizing on 0, it follows that X f-v q. (QED)
Superiority expresses the idea that the logical consequences of a consistent
set X answer all of their respective objections. The role of Superiority in
the logic of coherence is to ensure that the principle of Supraclassicality holds
for f-v:

Theorem 4 If X r P then X f-v p.

Proof: Suppose that X r p. If X r, then X f-v P by C1 and we are done. Al-
ternatively, suppose X j,L. Then, by Superiority, P answers all of its objections
on X. Hence X f-v p. (QED)
Theorem 4 has Reflexivity as a direct consequence:

Theorem 5 IfP E X then X f-v p.

As far as coherence is concerned, Enlargement and Reduction for Objections
together entail that if P coheres with X and X is logically consistent, then P has

the same objections on X as on X U {p}. If P coheres with X and X is logically

consistent, then Enlargement and Reduction for Answering entail that the same
pairs of claims stand in the answering relation on X as on X U {p}, whereas
Enlargement and Reduction for Neutralizing entail that the same objections to
a given claim are neutralized by the same neutralizers on X as on X U {p}.
Thus, expanding a given set of claims by adding something that coheres with
the given set neither increases nor decreases the set of claims that cohere. As far
as ~ is concerned, the principles of Enlargement and Reduction for Objections,
Answering, and Neutralizing serve to ensure that Cautious Monotony and Cut

Theorem 6 If X ~ p and X ~ q, then X U {p} ~ q.

Proof: Suppose that X ~ pandX ~ q. If XU{p} 1-, thenXU{p} ~ q by CI,

so suppose Xu {p} }L , and let a be an objection to q on X U {p}. By Reduction
for Objections, a is an objection to q on X. Since X ~ q, a is answered or
neutralized as an objection to q on X. By Enlargement for Answering and
Enlargement for Neutralizing, a is answered or neutralized as an objection to q
on X U {p}. Generalizing on 0, X U {p} ~ q. (QED)

Theorem 7 If X U {p} ~ q and X ~ p, then X ~ q.

Proof: Suppose that Xu {p} ~ q and X ~ p. If X 1-, then X ~ q by CI,

so suppose X }L, and let a be an objection to q on X. By Enlargement for
Objections, a is an objection to q on Xu {p}. Since Xu {p} ~ q, a is answered
or neutralized as an objection to q on X U {p}. By Reduction for Answering
and Reduction for Neutralizing, 0 is answered or neutralized as an objection to q
on X. Generalizing on 0, X ~ q. (QED)
Of course, the six Enlargement and Reduction principles can be assumed
to be lemmas whose basis lies in further assumptions about comparative rea-
sonableness. What further assumptions are needed? The following invariance
principle for comparative reasonableness turns out to suffice:

Invariance Under Inductive Expansion (IUIE) If X }L and X ~ p, then for

all q, r, s, and t, (q, r) >- X (s, t) iff (q, r) >- xu{p} (s, t).

Theorem 8 IUIE implies Enlargement for Objections.

Proof: Assume IUIE and X ~ p, and let a be an objection to q on X. Then

(q,--,o) >-x (q,o). ByIUIE, (q,--,o) >-xu{p} (q,o),henceoisanobjectiontoq
on Xu {p}. (QED)

Theorem 9 IUIE implies Reduction for Objections.

Proof: Similar to the proof of Theorem 8. (QED)

Theorem 10 IUIE implies Enlargementfor Answering.

Proof: Assume that X f-v p and 0 is an objection to q on both X and X U {p}

and q answers 0 onX. Then (q, T) >- x (0, T). By IUIE, (q, T) >- xU{p} (0, T).
Hence q answers 0 on Xu {p}. (QED)

Theorem 11 IUIE implies Reduction for Answering.

Proof: Similar to the proof of Theorem 10. (QED)

Theorem 12 IUIE implies Enlargement for Neutralizing.

Proof: Assume that X f-v p and 0 is an objection to q on both X and X U {p}.

Let n neutralize 0 as an objection to q on X. Then (q, -'(0 & n)) 'I- x (q, 0 & n)
and (0 & n, T) ~x (0, T). By IUIE, (q, -'(0 & n)) 'I- xu{p} (q,o & n) and
(o&n, T) ~xu{p} (0, T). HencenneutralizesoasanobjectiontoqonXU{p}.

Theorem 13 IUIE implies Reduction for Neutralizing.

Proof: Similar to the proof of Theorem 12. (QED)

3.2. Some Postulates aud Results for Negation

Since the language in which our logic is formulated includes an expressively

complete set of truth-functional connectives, every sentence (including every
logical truth) has a boolean negation. The following postulates seem reasonable
as principles relating comparative reasonableness to sentences and their nega-

Inconsistency If X J.L and {p, q} f--- and {p} J.L and {q} J.L, then (p, -,q) >- x
(p, q).
Top/Bottom If X J.L and {p} J.L, then (T, p) >- x Cl, p).
Bottom/Top If X J.L, then el,.1) >- x (.1, T).
Exclusive Objection for.1 If (.1, -,p) >- x (.1,p), then f--- p.
Exclusive Equivalence If X J.L and (.1, T) ~x (p, T), then {p} f---.

Top Level If X J.L and (T, T) ~x (p, T), then f--- p.


The upshot of Inconsistency is that a pair of individually consistent but jointly

inconsistent statements are always objections to one another. Top/Bottom en-
tails that a logical truth is always more reasonable, given a consistent assump-
tion p, than any contradiction given that same assumption. Assuming Equiva-
lence, Bottom/Top entails that every logical truth is an objection to every con-
tradiction, and Exclusive Objection for ..l entails that every objection to a con-
tradiction is a logical truth. Exclusive Equivalence entails that anything which
is exactly as reasonable as a contradiction is itself a contradiction. Top Level
entails that anything which is exactly as reasonable as a logical truth is itself a
logical truth.
An important consequence of this collection of principles is the thesis
that f-v has the property of Consistency Preservation:

Theorem 14 If X ¥ then there is no p such that X f-v p and X f-v -'p.

Proof: Suppose that X ¥. For reductio, suppose that there is a p such that
X f-v p and X f-v -'p.
Case 1: Suppose that I- p. Then -,p -II- ..l, so by Theorem 3, X f-v ..l.
Since X ¥, it follows that every objection to ..l is either answered or neu-
tralized on X. By Bottom/Top and Equivalence, (..l, -, T) 'r x (..l, T), so
T is an objection to ..l on X. Since ¥ -, T, we have by Top/Bottom that
(T, T) 'r x (..l, T), so by Asymmetry, (..l, T) >f x (T, T). By Exclusive
Objection for ..l, every objection to ..l on X is logically equivalent to T, so
by Equivalence, (..l, T) >f x (0, T) for every objection 0 to ..l on X. Hence
..l does not answer any of its objections on X. Let n neutralize 0 as an ob-
jection to ..l on X. Then 0 is an objection to ..l on X and 0 & n is not an
objection to ..l on X and (0 & n, T) r:::=.x (0, T). Exclusive Objection for ..l
implies that 0 -II- T, so by Equivalence we have that T & n is not an objection
to ..l on X, i.e. (..l, -,(T & n)) >f x (..l, T & n), and T & n is as reasonable
as T on X, i.e. (T & n, T) r:::=.x (T, T). But T & n -II- n, so by Equivalence,
(i) (..l, -,n) >f x (..l, n) and (ii) (n, T) r:::=.x (T, T). By Top Level, (ii) implies
n -II- T, from which it follows by (i) and Equivalence that (..l,..l) >f x (..l, T),
which contradicts Top/Bottom. So a contradiction follows from the hypothesis
that I- p.
Case 2: Suppose that I- -'p. This hypothesis leads to a contradiction by
an argument similar to Case 1.
Case 3: Suppose that ¥ p and ¥ -'p. By Inconsistency, (p, -,-,p) 'r x
(p, -,p) and (-,p, -,p) 'r X (-,p, p). Hence -,p is an objection to p on X and p is
an objection to -'p on X. Since X f-v p and X f-v -,p and X ¥, -,p is either
answered or neutralized as an objection to p on X, and p is either answered
or neutralized as an objection to -'p on X. Asymmetry implies that p and -,p

cannot answer each other. Hence either -,p is neutralized as an objection to p

on X, or p is neutralized as an objection to -,p on X. But neither of these
alternatives is possible. Let n be any sentence, and suppose for reductio that
n neutralizes -,p as an objection to p on X. Then -,p & n is not an objection
toponX,i.e. (p,-,(-,p&n)) >l-x (p,-,p&n),and(-,p&n,T) ~x (-,p,T).
Since {-,p} J.L, it follows by Exclusive Equivalence that { -,p & n} J.L. Since,
in addition, (p, -, (-,p & n)) >I- x (p, -,p & n) and {p} J.L and {p, -,p & n} f-,
Inconsistency is contradicted. By reductio, -,p is not neutralized as an objection
to p on X. Exchanging p and -'p, an exactly similar argument shows that pis
not neutralized as an objection to -,p on X. The hypothesis of Case 3, that J.L p
and J.L -'p, therefore leads to a contradiction. (QED)
Consistency Preservation is a significant property because any consistency-
preserving inference relation that allows inferences which go beyond the de-
ductive consequences of a set must be nonmonotonic: 9

Theorem 15 Suppose that ~ is a relation on pairs consisting of sets of sen-

tences and sentences, and suppose that ~ satisfies Reflexivity, Consistency Pre-
servation, and Superclassicality:

Consistency Preservation: If X ¥ then there is no p such that

X ~ p and X ~ -'p.
Superciassicality: For some p, X ~ p but X J.L p.

Then ~ is nonmonotonic, i.e. there are X, Y, and p such that X C Y and

X ~ p but Y J76 p.

Proof: Assume Reflexivity, Consistency Preservation and Superclassicality. By

Superclassicality, find a p such that X ~ p but X J.L p. Since X J.L p, it
follows that X U { -,p } J.L . By Reflexivity, X U { -,p} ~ -'p, so by Consistency
Preservation, Xu {-,p} J76 p. It therefore suffices to let Y = Xu {-,p}. (QED)

3.3. Postulates and Results for Disjunction

Since Boolean disjunction is present in the object language, it is reasonable to

ask how disjunction interacts with our coherence-based inference relation' r-' .
Kraus, Lehmann, and Magidor (1990:190) and Makinson (1989:14) consider
analogs of the following "introduction on the left" rule for disjunction:

Or If X U {p} r- r and X U {q} r- r, then X U {p V q} r- r.

The Or postulate follows if we make the following reasonable assumptions
about objections, answering, and neutralizing.

Left Replacement If X U {p} r-- r and X U {p} f- q and X U {q} f- p, then

XU{q} r--r.
Objection for Disjunctions If Xu {p} J.L and Xu {q} J.L and X U {p} r-- r
and Xu { q} r-- r, then every objection to r on Xu {p V q} is an objection
to r on both X U {p} and X U {q}.

Answering and Neutralizing for Disjunctions If Xu {p} J.L and Xu {q} Y

and X U {p} r-- r and X U {q} r-- r, and if 0 is an objection to r on
X U {p V q} which is answered or neutralized as an objection to r on
X U {p} and as an objection to r on X U {q}, then 0 is answered or
neutralized as an objection to r on Xu {p V q}.

Theorem 16 If Left Replacement, Objection for Disjunctions, and Answering

and Neutralizingfor Disjunctions hold, then the Or postulate holds.

Proof: Assume Left Replacement, Objection for Disjunctions, and Answering

and Neutralizing for Disjunctions, and suppose that XU{p} r-- rand XU{ q} r--
r. If Xu {p V q} f-, then Xu {p V q} r-- r by C 1, so suppose that X U {p V q} J.L .
If X U {p} f- , then X U {p V q} f- q, in which case, since X U { q} f- p V q and
Xu { q} r-- r, it follows by Left Replacement that X U {p V q} r-- r. Similarly, if
X U {q} f-, then X U {p V q} r-- r, so suppose that X U {p} J.L and X U {q } J.L .
Let 0 be an objection to r on X U {p V q}. By Objection for Disjunctions,
o is an objection to r on X U {p} and on X U {q}. Hence 0 is answered or
neutralized as an objection to r on X U {p} and 0 is answered or neutralized as
an objection to r on Xu {q}. By Answering and Neutralizing for Disjunction,
o is answered or neutralized as an objection to r on Xu {p V q}. Generalizing
on 0, X U {p V q} r-- r. (QED)
Left Replacement follows from the following plausible principle on comparative
reasonableness, where Cn(X) = {p: X f- p}:

Invariance Under Reformulation If Cn(X) = Cn(Y), then for all p, q, r,

ands, (p,q) >--x (r,s) iff(p,q) >--y (r,s).

Theorem 17 Invariance under Reformulation implies that if X U {p} r-- rand

Xu {p} f- q and Xu {q} f- p, then X U {q} r-- r.

Proof: Assume Invariance under Reformulation and suppose that X U {p} r-- r
and X U {p} f- q and X U { q} f- p. If X U { q} f-, then X U { q} r-- r by C 1, so
suppose that X U {q} J.L. Since X U {p} f- q and X U {q} f- p, it follows that
Cn(X U {p}) = Cn(X U {q}). Let 0 be an objection to r on Xu {q}. Then
(r, -'0) >-- xu{ q} (r, 0). By Invariance under Reformulation, (r, -'0) >-- xu{p}

(r, 0). Hence 0 is an objection to r on X U {p}. Since X U {p} f"v r, it follows

that 0 is answered or neutralized on X U {p}.
Case 1: Assume thatr answers 0 on XU{p}. Then (r, T) >- xU{p} (0, T).
By Invariance under Reformulation, (r, T) >- xU{q} (0, T). Hence r answers 0
onX U {q}.
Case 2: Assume that there is an n such that n neutralizes 0 as an objection
to r on Xu {p}. Then (0 & n, -,r) tXU{p} (0 & n, r) and (0 & n, T) ~xu{p}
(0, T). By Invariance under Reformulation, (0 & n, -,r) t xU{q} (0 & n, r)
and (0 & n, T) ~XU{q} (0, T). Since 0 is also an objection to r on Xu {q},
n neutralizes 0 as an objection to r on Xu {q}.
By Cases 1 and 2, it follows that 0 is answered or neutralized as an ob-
jection to r on Xu {q}. Hence Xu {q} f"v r, as required. (QED)
The following principle is sufficient to derive Objection for Disjunctions:

Distribution for Disjunction If X U {p} j,L and X U {q} j,L, then for all r,
s, t, and u, (r,s) >-xu{pvq} (t,u) only if (r,s) >-xu{p} (t,u) and
(r,s) >-xu{q} (t,u).

Theorem 18 Distribution for Disjunction implies Objection for Disjunctions.

Proof: Assume Distribution for Disjunction, and suppose that X U {p} j,L and
X U {q} j,L and X U {p} f"v r and X U {q} f"v r. Let 0 be an objection to r on
Xu {p V q}. Then (r, -'0) >- xu{pVq} (r,o). By Distribution for Disjunction,
(r, -'0) >- xU{p} (r,o) and (r, -'0) >- xU{q} (r,o). Hence 0 is an objection to r
on both X U {p} and X U {q}. (QED)
The question of how to secure Answering and Neutralizing for Disjunctions via
a reasonable principle of comparative reasonability remains an open problem.

3.4. Deductive Closure for Inductive Consequences

When, as in this paper, cumulative reasoning is formalized in the context of a
deductive background logic, the question arises whether the inductive conse-
quences of a set form a deductively closed set. The following proof-theoretic
postulates jointly suffice as a definition of deductive closure for 'f"v':

Special Supraclassicality If r- p then X f"v p.

Right Modus Ponens If X f"v p:J q and X f"v p, then X f"v q.
Special Supraclassicality follows from Supraclassicality and so is a consequence
of Superiority, but Special Supraclassicality also follows immediately from Top/
Bottom, Equivalence, and the following two principles:

Top/Top (T, T) )- x (T, -.1).

Exclusive Objectiou for T If X J.L and (T, ...,p) )- x (T, p), then {p} f-.

Theorem 19 .if Top/Bottom, Top/Top, Equivalence, and Exclusive Objection

for T hold, then if f- p then X f-v p.

Proof: Assume Top/Bottom, Top/Top, Equivalence, and Exclusive Objection

for T, and let f- p. If X f-, then X f-v p by Cl, so assume that X J.L. Since
(T, T) )- x (T, -.1) by Top/Top, it follows by Equivalence and Exclusive Objec-
tion for T that q is an objection to T on X iff q -1f- -.1. Since {T} J.L , it follows
by Top/Bottom that (T, T) )- x (-.1, T). Since -.1 -1f- q for every objection q to
T on X, it follows by Equivalence that T answers all of its objections on X.
Hence X f-v T. Since f- p, p -1f- T; hence by Theorem 3 we have that X f-v p.
Right Modus Ponens follows if we assume the following three principles:
Objection for Conditionals If X J.L and X f-v p, then q and p ::J q have the
same objections on X.

Answering for Conditionals If X J.L and X f-v p and p :J q answers a on X,

then q answers a on X.

Neutralizing for Conditionals If X J.L and X f-v p and n neutralizes 0 as an

objection to p ::J q on X, then n neutralizes a as an objection to q on X.

Theorem 20 .if Objection, Answering, and Neutralizing for Conditionals hold,

then if X f-v P ::J q and X f-v p, then X f-v q.

Proof: Assume Objection, Answering, and Neutralizing for Conditionals, and

suppose that X f-v p::J q and X f-v p. If X f-, then X f-v q by Cl, so suppose
that X J.L . Let r be an objection to q on X. Then by Objection for Conditionals,
r is an objection to p ::J q on X. Since X f-v p ::J q, r is answered or neutralized
as an objection to p ::J q on X. If p ::J q answers r on X, then, by Answering for
Conditionals, q answers r on X. If there is an n such that n neutralizes r as an
objection to p ::J q on X, then, by Neutralizing for Conditionals, n neutralizes r
as an objection to q on X. So r is answered or neutralized as an objection to q
on X. Generalizing on r, X f-v q. (QED)
Objection, Answering, and Neutralizing for Conditionals in turn follow from
the following simple requirement on comparative reasonableness:
Conditional Invariance If X J.L and X f-v p, then for all q and r, (q, r) ~x

Theorem 21 Conditional Invariance implies ObjectionJor Conditionals.

Proof: Assume that X r and X [-v p. Let 0 be an objection to q on X. Then

(q, -'0) ;.- x (q,o). By Conditional Invariance, (q, -'0) ~x (p:=J q, -'0) and
(q, 0) ~x (p:=J q, 0). By Theorem 2, (p :=J q, -'0) ;.- x (p:=J q, 0). Hence 0 is
an objection to p :=J q on X. So every objection to q on X is an objection to
p:=J q on X. A similar argument shows that every objection to p :=J q on X is an
objection to q on X. (QED)

Theorem 22 Conditional Invariance implies AnsweringJor Conditionals.

Proof: Assume that X r and X [-v p and p :=J q answers T on X. Then T is

an objection to p :=J q on X and (p :=J q, T) ;.- X (T, T). By Theorem 21, T is
an objection to q on X. By Conditional Invariance, (q, T) ~x (p:=J q, T),
and, since ~x is an equivalence relation, (T, T) ~x (T, T). By Theorem 2, it
follows that (q, T) ;.- X (T, T). Hence q answers T on X. (QED)

Theorem 23 Conditional Invariance implies NeutralizingJor Conditionals.

Proof: Assume that X r and X [-v p and n neutralizes T as an objection to

p :=J q on X. Then T is an obj ection to p :=J q on X, and T & n is not an obj ection
to p:=J q on X, and (T & n, T) ~x (T, T). By Theorem 21, T is an objection to
q on X and T & n is not an objection to q on X. Hence n neutralizes T as an
objection to q on X. (QED)


We have seen that it is possible to interpret a slightly simplified version of

Lehrer's (2000) theory of relational coherence as a species of inductive rea-
soning, indeed as a species of cumulative reasoning, and we have seen that the
cumulativity of this species of inductive reasoning can be derived from certain
very plausible assumptions about comparative reasonableness. The additional
properties of Consistency Preservation and deductive closure for inductive con-
sequences can also be derived from plausible principles of comparative reason-
ableness. It remains to be seen whether Or and Answering and Neutralizing for
Disjunctions can be derived from plausible principles of comparative reason-
As an exercise in logic, the hypothesis that relational coherence is a
species of inductive reasoning appears to be a success. But does this way of
looking at relational coherence have any epistemological significance? Clearly
it does. The criterion of coherence with one's evaluation system is intended by
Lehrer, in the first instance, as a means of evaluating the personal justification

status of claims that a subject accepts against a larger background consisting of

acceptance, reasoning, and preference systems. But any criterion of epistemic
justification has to be more than this. Regardless which theory of knowledge is
correct, it is surely true that any truth-seeking, error-averse subject who seeks to
expand what he or she accepts should add only beliefs which he or she would be
justified in holding if he or she held them. And the process of adding beliefs in
this fashion is nothing ifnot inductive reasoning. Accordingly, if Lehrer's coher-
ence theory of knowledge is correct, then any truth-seeking, error-averse subject
wishing to reason inductively should do so by expanding his or her acceptance
system to include claims which would cohere with his or her evaluation system.
Nothing, therefore, could be more natural in the context of Lehrer's theory of
knowledge than an account of inductive reasoning in terms of relational coher-

* I am pleased and honored to participate in celebrating the career of so brilliant a philosopher
as Keith Lehrer. Lehrer's status as a leading figure in the field is richly deserved, and, without a
doubt, his work will continue to stand the test of time.
1 Cumulativity was defined by Gabbay (1985), but see also Kraus et al. 1990, Makinson 1989,
and Makinson 1994.
2 But see Lehrer 2000:144-146 for a discussion of relative reasonableness, probability, and ex-
pected value.
3 In this and later sections, the variables p, q, r, and s range over statements in some given
language, and the variables X and Y with and without subscripts range over sets of statements.
The expression 'X f- p' means that p can be inferred from X in the inference system defined
4 See Makinson 1994 for a survey.
5 In parallel with Lehrer's definitions, "it is more reasonable to accept p than to accept r on the
basis of X" will be assumed to mean "it is more reasonable to accept p on the assumption T than
to accept r on the assumption T on the basis of X", where T is some fixed logical truth. 1. is
defined as -, T.
6 'p -jf- q' will mean that {p} f- q and {q} f- p. Where X is a set of sentences, 'X f- ' will mean
that X f- p and X f- -,p for some p.
7 In Lehrer's theory, comparative reasonableness is conditional as well as comparative ("it is more
reasonable to accept p on the assumption q than to accept r on the assumption s"). For a well
worked-out example of a nonquantitative theory of unconditional comparative reasonableness,
see Chisholm and Keirn 1972.
8 It might be argued that distinctions of comparative reasonableness can be made in such cases.
If this is right, then our adoption of Nontriviality should be considered a simplifying assumption
that reflects a decision not to address comparative reasonableness for inconsistent X.
9 See Cross 1990 for an application of a version of this result to the question of the tenability of
the Ramsey test for conditionals.

Chisholm, R. and R. G. Keirn. 1972. "A system ofepistemic logic." Ratio 14:99-115.
Cross, C. 1990. "Belief revision, nonmonotonic reasoning, and the Ramsey test." In Knowledge
Representation and Defeasible Reasoning, edited by H. Kyburg, R. Loui, and G. Carlson,
223-244. Dordrecht: Kluwer.
Gabbay, D. 1985. "Theoretical foundations for non-monotonic reasoning in expert systems."
In Logics and Models of Concurrent Systems, edited by K. R. Apt, 439-457. Berlin:
Kraus, S., D. Lehmann, and M. Magidor. 1990. "Nonmonotonic reasoning, preferential models,
and cumulative logics." Artificial Intelligence 44:167-207.
Lehrer, K. 1974. Knowledge. Oxford: Oxford University Press.
~~-. 1986. "The coherence theory of knowledge." Philosophical Topics 14:5-25.
~~-. 1988. "Metaknowledge: Undefeated justification." Synthese 74:329-347.
~~-. 2000. Theory ofKnowledge. 2nd ed. Boulder: Westview Press.
Makinson, D. 1989. "General theory of cumulative inference." In Lecture Notes in Artificial
Intelligence 346: Nonmonotonic Reasoning, edited by M. Reinfrank, 1. de Kleer, M. L.
Ginsberg, and E. Sandewall, 1-18. Berlin: Springer-Verlag.
Makinson, D. 1994. "General patterns in nonmonotonic reasoning." In Handbook of Logic in
Artificial Intelligence and Logic Programming, Volume 3: Nonmonotonic Reasoning and
Uncertain Reasoning, edited by D. M. Gabbay, C. 1. Hogger, and 1. A. Robinson, 35-110.
Oxford: Oxford University Press.
Shoham, Y. 1988. Reasoning about Change. Cambridge: MIT Press.
Chapter 8


Wolfgang Spohn
University of Konstanz

Meets what? Ranking theory is, as far as I know, the only existing theory suited
for underpinning Keith Lehrer's account of knowledge and justification. If this
is true, it's high time to bring both together. This is what I shall do in this paper.!
However, the result of defining Lehrer's primitive notions in terms of
ranking theory will be disappointing: justified acceptance will, depending on the
interpretation, either have an unintelligible structure or reduce to mere
acceptance, and in the latter interpretation knowledge will reduce to true belief.
Of course, this result will require a discussion of who should be disappointed.
So, the plan of the paper is simple: In section 1 I shall briefly state what
is required for underpinning Lehrer's account and why most of the familiar
theories fail to do so. In section 2 I shall briefly motivate and introduce ranking
theory. Basing Lehrer's account on it will be entirely straightforward. Section 3
proves the above-mentioned results. Section 4, finally, discusses the possible



I shall base my considerations on Lehrer (2000), the most recent

presentation of his theory. It indeed adds simplifications and clarifications to the
first edition. For instance, the basic notions on which his theory of knowledge
and justification builds stand out more clearly. They are summarized in his
definition of an evaluation system in Lehrer (2000, p. 170, D 1)2 which consists
of three components: (a) an acceptance system, i.e., a set of accepted statements
or propositions, (b) a preference system, i.e., a four-place relation among all
E.!. Olsson (ed.), The Epistemology of Keith Lehrer, 129-142.
© 2003 Kluwer Academic Publishers.

statements or propositions saying for all A, B, e, D whether A is more reasonable

to accept given or on the assumption that ethan B given or on the assumption
that D (I symbolize it here as Ale >- BID),3 and (c) a reasoning system, i.e., a set
of inferences each consisting of premises and a conclusion.
The idea behind (c) is that what a person accepts, and justifiedly accepts,
depends also on the inferences she carries out. Here I understand Lehrer as
referring to deductive inferences or rather to the inferences taken by the person
to be deductively valid and sound. 4 The idea behind (b), by contrast, is to take
care of inductive reasoning in the widest sense which is always a matter of
weighing reasons and objections on the basis of some such preference system.
We shall look in detail at Lehrer's specific proposal for this weighing ofreasons.
The first step I would like to take in my present discussion is to fix the
reasoning system once for all. The reason is that otherwise my comparative
business could not even start. Of course, we shall have to discuss in the final
section whether this is already the first misrepresentation of Lehrer that will
entail all the other ones. How do I fix that system? Since Lehrer emphasizes
again and again that he is interested in acceptance only insofar as it is governed
by the aim of truth, I propose to extend this attitude to the objects of belief or
acceptance and to conceive of them only insofar they can be true or false, i.e., as
truth conditions or propositions. Thereby, I ignore all questions of syntactic
structure, of logical equivalence, and of logical entailment, and assume that the
rationality constraint of consistency and deductive closure of the acceptance
system is trivially satisfied. This entails in particular the assumption that the
reasoning system is maximal and has no independent role to play. I am well
aware that I am taking this step very swiftly. My excuse is that I am convinced
that a lengthy treatment of the issue would not reveal a viable constructive
Having taken this step the task of underpinning reduces to accounting for
the acceptance and the preference system. Concerning the latter, the first idea is,
of course, to appeal to a probability measure P and to define that Ale >- BID iff
P(AIc) > P(BID). However, the relation between probability and acceptance is
problematic, as is highlighted by the famous lottery paradox. I am not rejecting
all attempts to solve this paradox out of hand, but the mere fact that they are
debated heatedly and that all ofthem are contested shows that probability theory
is, presently, not a good foundation for Lehrer's theory. Moreover, as I shall
point out below, there is a particular feature in Lehrer's notion of neutralizing an
objection which prevents any probabilistic interpretation. Olsson (1998a)
discusses further difficulties of a purely probabilistic construal of justified
For similar reasons Lehrer, too, has given up on finding purely
probabilistic foundations, which he still hoped to build in Lehrer (1971, 1974).
There he suggests, moreover, that the foundations may be construed as some
kind of epistemic decision theory. The hint is still found in Lehrer (2000,

pp.145ff.) and also used for a solution of the lottery paradox. However, I am
doubtful because epistemic decision theory has remained a promise that has
never been redeemed in a satisfying way in the last 35 years.
The basic difficulty, I believe, is this: Probability theory may claim, in a
way, to offer a complete epistemology. If so, it is hard to complement it or to
merge it with other epistemological ideas like acceptance, epistemic decisions,
or whatever, and radical probabilism as Jeffrey (1992) has defended it seems
Hence, we should put probability theory aside and rather look at theories
dealing directly with acceptance or belief. A large variety of such theories-such
as default logic, AGM belief revision theory, Pollock's and other accounts of
defeasible and non-monotonic reasoning, etc.-has been developed in the past
25 years. Maybe they provide an account of Lehrer's preference system as well.
Alas, they don't. At least, I claim this with confidence with respect to AGM
belief revision theory. There it is shown that the behavior of belief revisions is
equivalent to the behavior of so-called entrenchment relations. 5 These could
indeed fill the role of Lehrer's unconditional preference relation. Maybe
entrenchment relations can be generalized so as to capture the special case A IC
>- BIC of Lehrer's preference relation which refers twice to the same condition.
However, Lehrer requires the full conditional relation, which cannot be
accounted for in AGM belief revision theory. I suspect that essentially the same
is true of all accounts of defeasible reasoning implicitly or explicitly appealing
to some kind of epistemic ordering.
There is only one theory that is about belief or acceptance and provides
a sufficiently powerful preference system: ranking theory. That's why I said it
is the only existing theory suited for underpinning Lehrer's account. What does
it look like?


The basics are quickly told. Originally, ranking theory was developed 6 in
order to overcome essential restrictions of AGM belief revision theory. As it
turns out, AGM theory generally accounts only for one step of belief revision
and thereafter returns to a static picture. But, of course, a full dynamics has to
account for several or iterated belief changes. The problem has been around since
Harper (1976), and there have been quite a number of attempts to solve it within
the confines of AGM theorizing.? However, I find these proposals inferior to the
one provided by ranking theory.
Iterated belief revision is not our concern here. However, there exists a
close connection between iterated belief change and full conditional epistemic
preference. 8 It is for this reason that ranking theory, though addressed to the
former, can also provide for the latter. So let's take a look.

Let's start with an exhaustive set Wofpossibilities (possible worlds, first-

order valuations, or whatever). Subsets of Ware propositions (let's not worry
about their algebraic structure), W itself is the logically true and 0 the logically
false proposition. As explained above, I take such propositions as objects of
epistemic attitudes.
Ranks, then, are grades of disbelief (where I find it natural to take non-
negative integers as grades, but other numbers would do so as well). These
grades obey some fundamental laws summarized in

Definition 1: K is a ranking function iff K is a function from the

power set of W into Nu{ rf)} such that K(W) = 0, K(A) = rf) iff A =
0, and K(AuB) = min{K(A), K(B)}. K(A) is called the rank of A.
The rank ofB given A is defined as K(BIA) = K(AnB) - K(A).

Hence, K(A) > 0 says that A is disbelieved (to some degree), and K(A) > 0 says
that A is believed. 9 K(A) = 0 only expresses that A is not disbelieved and leaves
open the possibility that A is not disbelieved as well. Since there is no point here
in distinguishing between belief and acceptance, we thus have

Definition 2: A is accepted by K iffK(A) > O. {A IA is accepted by

K} is the acceptance system ofK.

What are the fundamental laws according to Definition I? K(W) = 0 says

that the logically true proposition is not disbelieved. The condition that K(A) =
rf) iff A = 0 says that it is most strongly believed, i.e., that the logically false

proposition is more strongly disbelieved than any other. More substantial is the
law ofdisjunction that K(AuB) = min {K(A), K(B)}. Clearly, the disjunction AuB
cannot be more firmly disbelieved than either of its disjuncts. Nor can it be less
firmly disbelieved than both disjuncts, since this would entail the absurdity that
given AuB both, A and B, are disbelieved, though AuB is not. An immediate
consequence is the law ofnegation that either K(A) = 0 or K(A) = 0 (or both). A
and A cannot be both disbelieved. Perhaps the most important law is the law of
conjunction that K(AnB) = K(A) + K(BIA) which follows trivially from the
definition of conditional ranks. It says that in order to arrive at the degree of
disbelief in AnB one has to sum up the degree of disbelief in A and the
additional degree of disbelief in B given A. This, I believe, agrees with intuition.
There is a surprisingly well working translation from probabilistic into
ranking terms which almost automatically generates a large number of ranking
theorems from probability theorems. This applies also to the account of belief
change. The basic rule for probabilistic belief change is simple conditionalization
according to which one moves to the probabilities conditional on the information
received. This is generalized by Jeffrey's conditionalization lO which is

unrestrictedly performable and thus defines a full dynamics within the realm of
strictly positive probability measures. In the corresponding way, such
conditionalization with respect to ranks offers a full dynamics of belief or
acceptance. 11
This informal hint at belief revision may suffice. However, we should
formally introduce belief contraction because Lehrer makes explicit use of it in
what he calls the ultrasystem. Belief contraction is the operation of giving up
some belief without adding new ones. It is extensively discussed in AGM belief
revision theory because it is interchangeable with belief revision.!2 Within
ranking theory it is easily defined as well (and turns out then to have all the
properties described in AGM theory!3):

Definition 3: The contraction K - A of K by A is defined by K - A

= K, if K(A) = O. If not, it is defined by (K - A) (B) = K(B) for B <;:;;;
A and (K - A) (B) = K(B) - K(A) for B <;:;;; A; for other B the rank
may be inferred from these conditions by applying the law of

Hence, if A is not believed, anyway, the contraction by A does not have

any effect at all. And if A is believed, i.e., A is disbelieved, then again (K - A)
(A) = 0, i.e., A is no longer disbelieved after the contraction. However, the ranks
conditional on A and on A are unaffected by the contraction.
So much for ranking theory as such. Let us now apply it to Lehrer's



Conditional ranks do not only allow us to account for (iterated) belief

revision and contraction. They also offer what we are directly aiming at, namely
an explanation of Lehrer's preference system, i.e., of his primitive four-place
relation >- :

Definition 4: A is more reasonable to accept given C than B given

D relative to K, i.e., AIC >- BID, iff K(AIC) > K( BID). Moreover, A
is more reasonable to accept than B, A >- B, iff AIW >- BIW, i.e.,
K(A) > K( B ). Similar notions are defined correspondingly.

Hence, a ranking function K comprises all components of what Lehrer

calls an evaluation system: an acceptance system (see Definition 2), a preference
system (see Definition 4), and a reasoning system (as trivialized by me above).

One may be tempted to say that Lehrer's account is wedded to ranking theory.
The wedding is unhappy, however. On the basis of Definition 4, Lehrer's
account ofjustification is translated in a straightforward way, and the fatal results
are inescapable:

Definition 5 (= 04, p. 170): B is an objection to A iff AI B >- AlB.

Is being an objection a relation among accepted propositions or among

propositions in general? The latter according to Definition 5, though Lehrer may
intend the former. However, there is no issue here, since we shall have to
stipulate that justified acceptance entails acceptance and shall thus consider only
objections to accepted propositions. Otherwise, we would get nonsensical results.

Definition 6 (= 05, p. 170): The objection B to A is answered iff B

is an objection to A, i.e., AI B >- AlB, and A >- B.

Definition 7 (= 06, p. 170): C neutralizes the objection B to A iff

B is an objection to A, i.e., AI B >- AlB, BnC is not an objection to
- -
A, i.e, AI B u C ::5 AIBnC, and BnC is at least as acceptable as B,
i.e., BnC c:: B.
Indeed, it is this last condition that prevents a probabilistic interpretation
of Lehrer's preference system because it would be empty in this interpretation;
no objection could then be neutralized. In ranking terms, however, this
consequence need not be feared.
Thus, we arrive at

Definition 8 (= 03 and 07, pp. 170 f.): A isjustifiedly accepted, or

the acceptance of A is personally justified, relative to K in the
strong sense iff A is accepted and each objection to A is answered
or neutralized by some C.

This is the literal translation of Lehrer's definition. My qualification "in

the strong sense" indicates, however, that we shall have to consider a weaker
sense as well.
Unfortunately, this chain of definitions yields strange results. What is
justifiedly accepted according to Definition 8 is an unintelligible selection from
the accepted propositions. In order to define this selection let Em = U{D I K(D)
;:> m}; hence, Em is the logically weakest proposition having at least rank m. Then
we have:

Theorem 1: A is justifiedly accepted relative to K in the strong sense

iff, given K(A) = n > 0, for all m ?: n with Em - Em+1 =f- 0
A n Em - En1m+! =f- 0 holds true (or, equivalently, iff for all m ?: n
with K(Em) = m there is aD (;;; A with m :<:: K(D) :<:: n+m).

Proof Suppose there is an m for which this condition does not

hold. Clearly, m = n cannot be the exception. Hence, m > n. Now
take any B such that K(S) = m. Since A n Em - En+m+1 = 0,
K(An B ) ?: n+m+ 1. In any case K(AnB) = n. So we have K(AI B )
?: n+ 1 > K(A IB), i.e., B is an unanswered objection to A. How

could a C neutralize this objection? It should satisfy K( B u C) =

m and thus K( C) ?: m. Hence, still K(An( B u C)) ?: n+m+ 1 and
K(AnBnC) = n, and so K(AI B u C) > K(AIBn C). Thus, there can
be no neutralizer, and A is not justifiedly accepted relative to K.

Suppose conversely that the condition holds true for all m ?: n. This
unfolds into three cases:

First, it may be that K(A) = 00, i.e., A = W. Since for all m ~ 00 Em

= 0, the condition is satisfied. However, in this case there can be
no objection to A, and so A is justifiedly accepted.

Second, it may be that K(A) = n < 00 and indeed A = En' Suppose

B is an objection to A. This requires at least K(AI B ) > 0. Suppose
further that B is not answered. This requires K(A) :<:: K( B). But
since A = En' this entails B (;;; A and thus K(AI B) = 0.
Contradiction. Hence, there is no unanswered objection to A, and
A is justifiedly accepted.

Third, it may be that K(A) = n < 00 and A c En' Then there is a B

with K(B) = m and B (;;; (En - A) u E n+m+!, and exactly those B
are unanswered objections to A, since exactly those B satisfy both,
m ?: n, i.e., K( B) ?: K(A), and K(An B) ~ n+m+ 1, which is
tantamount to K(AI B ) > K(AIB), since K(AIB) = n. But we have
supposed that A n Em - En+m+l =f- 0. This secures that B can be
neutralized. Take any C such that A n Em - E n+m+l (;;; C (;;; Em.
Hence m :<:: K( C) < n+m+ 1. SO, K( B u C) = m = K( B ), and K(A
n( B u C)) < n+m+ 1, and hence K(AI B u C) :<:: n = K(AIBnC).

Therefore, any unanswered objection to A can be neutralized in this

case, and, again, A is justifiedly accepted.

Now, we can see why we had to restrict justified acceptance to accepted

propositions in Definition 8. If we had omitted this proviso, the proof of
Theorem 1 would have to consider the case where K(A) = O. However, for n =
oevery bit of the proof would go through in the same way. This would allow for
many justifiedly accepted propositions not previously accepted. Indeed, even 0,
the logically false proposition, would turn out as justifiedly accepted! Simply
because there can be no objections to 0 according to Definition 5; 0 is
maximally disbelieved under all conditions. So, this had to be avoided.
Still, Theorem 1 is awkward. I have tried to express the necessary and
sufficient condition for justified acceptance as perspicuously as possible (but I
cannot circumvent the mathematical facts). However, this condition just makes
no intuitive sense. If Lehrer's definitions force us to distinguish between
justifiedly and unjustifiedly accepted propositions in this way, then there is
something wrong with the definitions.
Indeed, there may be cause for suspicion. For instance, objections may be
restricted to inductive objections, where B is an inductive objection to A iff
K(AIB) < K(AI Ii) < 00. Or I was wondering whether Lehrer really meant
Definition 7 as stated. Perhaps, the idea of neutralization is better expressed by
the condition that given the neutralizer C the objection B to A is no longer an
objection to A ( i.e., AI Ii nC ::5 AIBnC). However, as far as I have checked, this
leads to nowhere. Theorem 1 thereby changes considerably, but does not
No, a better cure is revealed by noticing that Theorem 1 implies that
justified acceptance is not even deductively closed: if A is justifiedly accepted
and logically implies B, B still need not be justifiedly accepted. This seems to be
a flaw in Definition 8 as it stands. But then, of course, we have to correct
Definition 8. Thinking about what one is personally justified to accept means
also working through one's reasoning system. Given my fixation of the
reasoning system, this leads us to the following weaker and more adequate sense
of justified acceptance:

Definition 9: A is justifiedly accepted relative to K in the weak

sense iff A is logically implied by propositions justifiedly accepted
in the strong sense. 14

What is justifiedly accepted in this sense? This is answered by

Theorem 2: A is justifiedly accepted relative to K in the weak sense

iff A is accepted by K.

Proof The proof of Theorem 1 shows that £1 is justifiedly

accepted. If A is accepted by K, i.e., K(A) > 0, then A <:;;;; E 1, i.e., £\
<:;;;; A. Thus A is logically implied by somethingjustifiedly accepted.

In a sense, this looks much nicer, but it empties Lehrer's theory of

justification. So, it looks undesirable as well. Theorem 2 would not change under
the modifications of Lehrer's defin itions mentioned above.
These results also affect Lehrer's theory of knowledge. In order to see how
we must first turn to undefeated justification and the ultrasystem. A person's
ultrasystem is generated from her evaluation system by deleting from the latter
all acceptances of, preferences for, and reasonings from falsehoods. This is
immediately explicable in ranking terms.
Let w* E W be the true or actual possibility in W, and let us consider any
evaluation system, i.e., ranking function K. If everything accepted in K is true,
then no falsehoods intrude into the justifications with respect to K, anyway. In
this case, K is its own ultrasystem. That's the rare case, though. Usually, some
false proposition will be believed in K. This entails K( {w*}) > 0.
Hence, the logically weakest falsehood accepted by K is (w*). Now, if we
contract K by (w*) , not only this falsehood, but also all other, stronger
falsehoods must go (since the contracted acceptance system is again deductively
closed). Hence, in the resulting ranking function K* exactly the true among the

propositions accepted by K are accepted. Moreover, if A is true and B is false,
then K*(A) = :<:; K*(B), and hence in the preference system provided by K* no
false proposition is ever preferred to a true one, as required by Lehrer in D8, p.
171. Finally, we do not worry about the reasoning system of1.c*, for the familiar
reason. All in all, the explication is remarkably smooth, and we may conclude

Definition 10 (= D8, p. 171): The ultrafunction K* ofK is K- (w*) ,

the contraction ofK by (w*).

This covers also the case where no falsehood is accepted in K, i.e., where
K* =K.
On this basis, the rest of Lehrer's definitions is immediately translated:

Definition 11 (= D9, p. 171): The justification for accepting A in K

is undefeated (or irrefutable) iff A is justifiedly accepted in K*.

Definition 12 (= DK, pp. 169f.): A is known in K iff (i) A is

accepted in K, (ii) A is true, i.e., w* E A, (iii) A is justifiedly

accepted in K, and (iv) the justification for accepting A in K is


Lehrer's claim that knowledge reduces to undefeatedly justified

acceptance, i.e., to condition (iv) is now easily confirmed. But stronger results
obtain. If knowledge is based on justified acceptance in the strong sense, just
those propositions are known which satisfY the condition of Theorem 1 relative
to K*. Unpalatable knowledge! If knowledge is based on justified acceptance in
the more adequate weak sense, we get

Theorem 3: A is known in K iff A is true and A is accepted in K.

Proof Theorem 2 reduces condition (iii) to (i). But, Theorem 2

applies also to K*. Hence, (iv) reduces to acceptance in K*, and
hence to (i) and (ii).

Thus, again, the sophisticated considerations relating to the ultrasystem

seem empty, and knowledge reduces to true belief, a conclusion Lehrer definitely
wants to avoid.


It is not clear what to think of these results. I find that the structure of what
I called justified acceptance in the strong sense is too weird to be worth
discussing. Hence, I proceed on the assumption that it is the weak sense that is
relevant. I have already indicated why I believe to agree with Lehrer on this
point. But Theorems 2 and 3 look troublesome as well, though it is not clear
where to locate the trouble. There are several ways, and more than one good
way, to respond.
(1) One may point to various flaws in my translation. For instance, I have
carelessly interchanged belief and acceptance, whereas Lehrer (pp. 12f.)
emphasizes their difference. Likewise, I have defined an acceptance system as
a set of accepted propositions, whereas for Lehrer it is a set of propositions ofthe
form "I accept that A". But these kinds of flaws are insignificant. Also, I have
neglected Lehrer's own foundations in terms of trustworthiness (pp.138ff.), but
taking them into account would have no effect on the present considerations.
However, my fixation of Lehrer's reasoning system by rightaway
assuming deductive closure is doubtlessly a major deviation from Lehrer. But,
again, I don't think it does any harm. The acceptance system to start with could
as well consist in an arbitrary, not deductively closed set of accepted statements,
as long as it is consistent. Working out what is justified on the basis of such an

acceptance system means working out the preference system, which is a more
permanent disposition of the epistemic subject extending to all statements. And
it means, as we have observed above, working out the reasoning system, i.e.,
accepting the logical consequences of what one has justifiedly accepted. Hence,
we are back at deductive closure, at least as far as justified acceptance is
concerned. Theorem 2 then says that the deductive closure of an acceptance
system is justifiedly accepted relative to this system, and this is equally
No, Theorems I and 2 are generated by the characteristics of the
preference system as specified by a ranking function. The trouble lies there and
not in my light way of dealing with deductive relations.
(2) This suggests another conclusion. If one proceeds from preference
systems generated by ranking functions and arrives at such undesired results,
then the two theories do not fit together, and one has to look for other preference
systems. After all, nobody claims that ranking theory delivers the only legitimate
kind of preference systems. So, the conclusion would be that ranking theory may
have useful applications, but Lehrer's account ofjustification does not belong to
(3) However, I tend to the reverse conclusion. The more offers for an
underpinning of Lehrer's account are rejected, the stronger the obligation to
come up with some sound theoretical foundation. As for my part, I doubt that
there is any better offer than the one made here. In any case, as long as the
foundation is missing, there is not really any theory of justification and
Perhaps this quest for a theoretical underpinning is too strong, though.
Concerning the acceptance system the picture rather seems to be that it collects
all the variegated items of information and inference, with the aim of truth, but
without critical standards. Any arbitrary set of statements may be formed in this
way. Then, it seems, there can be no theory about acceptance systems, and
asking for one is asking too much.
This attitude, however, does not carry over to the preference system. We
need not entertain the illusion that it is uniquely determined by rationality alone.
Carnap was under this illusion with inductive logic, but soon woke up. The
preference system and hence the standards ofjustification may well be subjective
to some extent. However, this is not to say that anything goes. This would mean
anarchy and throwing away the idea of rationality altogether. Some rationality
standards should be set up and defended. This is just what ranking theory
attempts to do, though perhaps in a debatable way. In any case, one cannot
simply be silent on the structure of preference systems. Hence, if one thinks that
the theorems above are undesirable, then, I think, Lehrer's account of
justification is really in trouble.
(4) Perhaps, though, one need not think that the theorems are undesirable.
When comparing ranking theory with Pollock's defeasible reasoning in Spohn

(2002) I concluded, among other things, that ranking functions are best
compared with what Pollock calls ideal warrant, which is, so to speak, the end
product of his defeasible reasoning machinery. IS Though Pollock's theory is
quite different from Lehrer's, this suggests that acceptance according to ranking
functions is already justified acceptance. This suggestion is supported by the fact
that ranking acceptance cannot yield any arbitrary acceptance system, but is
deductively closed, due to a correctly and maximally executed reasoning system.
In this perspective, then, Theorem 2 would not at all be surprising, it should be
expected. Accordingly, ranking theory and Lehrer's account ofjustification may
indeed be seen as mutually supporting each other, since Theorem 2 shows that
they reach the same result via entirely different considerations.
Too much harmony? Yes, I think so. First, the talk of ideal warrant is
Pollock's, and it was helpful in the above-mentioned comparison. But there is
nothing in ranking theory by itself forcing this comparison. Ranking theory is,
as I prefer to say more neutrally, about rational belief and its dynamics. So,
Theorem 2 rather shows either that ranking theory is about ideal warrant, despite
my disclaimer, or that ideal warrant reduces to rational belief. This would be my
preferred conclusion, but it reopens, it seems, a difference to Lehrer.
Secondly, even if Theorem 2 lends support to Lehrer's account of
justification, it does so in an unfriendly way, because it renders it vacuous at the
same time. The sophisticated considerations about answering and neutralizing
objections do not do any real work. This holds also when one starts, as Lehrer
does, from an arbitrary, unorderly acceptance system, because it is simply its
deductive closure that is justified relative to it. 16 This vacuity is certainly against
Lehrer's intentions.
Thirdly, there remains a problem with Theorem 3. If, as stated above,
ranking functions represent rational belief and nothing stronger, then the
reduction of knowledge to true belief as represented by ranking functions appears
doubtful, even though it has been defended by von Kutschera (1982, sect. 1.3),
and Sartwell (1991, 1992). In any case, it is unacceptable to Lehrer. Atthis point,
hence, harmony ceases at the latest.
So, as I said, it is not really clear what to conclude, though (3) would be
my preferred conclusion. The case shows once more how difficult it often is to
square different epistemological approaches.
However, there is, I think, a general lesson to learn. Justification is a
central notion in epistemology, and hence it is rightly scrutinized in many
discussions, from many perspectives, and with many examples and arguments.
In all that literature, though, I find very little rigorous theory. However, the
amenability to rigorous theorizing provides an important test. This test is usually
not even sought. I? But it is useful as a critical and as a constructive authority. At
least I hope to have shown this here with respect to the paradigm of Lehrer's

I I am indebted to Gordian Haas for various valuable remarks and for discovering bad faults in a
first draft which required major corrections and to Erik Olsson for further helpful suggestions.
2 In the sequel, mere page numbers are always meant to refer to Lehrer (2000).
3 In D 1, p. 170, Lehrer mentions only the unconditional preference relation, but it is clear that he
requires the conditional one.
4 On p. 127, when introducing the reasoning system, Lehrer refers only to "cogent" reasoning, and
generally the notion of validity makes clear sense only relative to deductive inference.
5 See Al chourr6n et al. (1985), which is considered as the foundation of AGM belief revision
theory, or Gardenfors (1988, ch. 4).
6 In Spohn (1983, sect, 5.3) and (1988).
7 Cf., e.g., Nayak (1994).
8 The connection is not obvious and perhaps not cogent. In order to really understand it we would
have to go much more deeply into issues of belief revision than we can and need to do here.
9 A is the complement of A with respect to W. The thinking in negations will continue, though I

am fully aware that it is cumbersome.

10 Cf. Jeffrey (1965, ch. 11).
II Cf. Spohn (1988, sect. 5).

12 Cf. Gardenfors (1988, ch. 3).

13 Cf. Spohn (1988, p. 133, footnote 20).

14 Lehrer (1971, p.221) took recourse to the same move when observing that the rule of induction
he proposed is not deductively closed.
15 Cf. Pollock (1995, sect. 3.10).
16 If the initial acceptance system should be inconsistent, the theoretical situation changes
drastically. Then paraconsistent logic may help, or the theory of consolidation (cf. Hansson 1994
and Olsson 1998b), or whatever. But there is no evidence in Lehrer's writings that this is his
17 For instance, it springs to one's eyes that at least three ofthe five conditions Bonjour (1985, pp.
95-99) offers for coherence are theoretically hardly explicable.

Alchourr6n, C.E., P. Gardenfors, D. Makinson (1985), "On the Logic of Theory Change: Partial
Meet Functions for Contraction and Revision", Journal of Symbolic Logic 30, 510-530.
BonJour, L. (1985), The Structure of Empirical Knowledge, Cambridge, Mass.: Harvard
University Press.
Gardenfors, P. (1988), Knowledge in Flux, Cambridge, Mass.: MIT Press.
Hansson, S.O. (1994), "Taking Belief Bases Seriously", in: D. Prawitz, D. Westerstahl (eds.),
Logic and Philosophy of Science in Uppsala, Dordrecht, Kluwer, pp. 13-28.
Harper, W.L. (1976), "Rational Belief Change, Popper Functions, and Counterfactuals", in: W.L.
Harper, c.A. Hooker (eds.), Foundations ofProbability Theory, Statistical Inference, and
Statistical Theories of Science, vol. I, Dordrecht: Reidel, pp. 73-115.
Jeffrey, R.C. (1965), The Logic of Decision, Chicago: The University of Chicago Press, 2nd ed.
Jeffrey, R.C. (1992), Probability and the Art of Judgment, Cambridge: Cambridge University
Lehrer, K. (1971), "Induction and Conceptual Change", Synthese 23, 206-225.
Lehrer, K. (1974), "Truth, Evidence, and Inference", American Philosophical Quarterly 11,79-92.
Lehrer, K. (2000), Theory of Knowledge, Boulder, Colorado: Westview Press, 2nd edition.
Nayak, A.c. (1994), "Iterated Belief Change Based on Epistemic Entrenchment", Erkenntnis 41,
Olsson, E. (1998a), "Competing for Acceptance: Lehrer's Rule and the Paradoxes of
Justification", Theoria 64, 34-54.
Olsson, E. (1998b), "Making Beliefs Coherent", Journal ofLogic, Language, and information 7,
Pollock, J.L. (1995), Cognitive Carpentry, Cambridge, Mass.: MIT Press.
Sartwell, C. (1991), "Knowledge is Merely True Belief" American Philosophical Quarterly 28,
Sartwell, C. (1992), "Why Knowledge is Merely True Belief", Journal ofPhilosophy 89, 167-180.
Spohn, W. (1983), Eine Theorie der Kausalitat, unpublished Habilitationsschrift, Munich.
Spohn, W. (1988), "Ordinal Conditional Functions. A Dynamic Theory of Epistemic States", in:
W.L. Harper, B. Skyrms (eds.), Causation in Decision, BeliefChange, and Statistics, vol.
II, Dordrecht: Kluwer, pp. 105-134.
Spohn, W. (2002) "A Brief Comparison of Pollock's Defeasible Reasoning and Ranking
Functions", Synthese 31, 39-56.
von Kutschera, F. (1982), Grundfragen der Erkenntnistheorie, Berlin: de Gruyter.
Chapter 9


Carl G. Wagner*
The University of Tennessee


Lehrer's epistemology, as articulated, for example, in Rational Consensus in

Science and Society (Lehrer and Wagner 1981), has always emphasized that ra-
tional decision making must take account of the total available evidence. Yet
dogmatic restrictions on the representation of uncertain judgment, or on the
way in which such judgment may be revised, undermine the goal of faithfully
representing the evidence. In this paper we discuss two such restrictions, dog-
matic Bayesianism and the dogma ofprecision, and outline some ways in which
probabilism has begun to be liberated from their grip.
Dogmatic Bayesianism, which asserts that the only acceptable method of
revising a probability distribution is by conditionalizing on an event E that one
has come to regard as certain, has already been substantially weakened by the
discovery of principled ways of updating probabilities when conditionalization
is simply inapplicable. Since descriptions of these alternative revision methods
are accessible and clear, we shall simply mention

1) revising one's probability distribution by means of a weighted average of

that distribution along with those of other informed individuals (Lehrer
and Wagner 1981);

2) probability kinematics (Jeffrey 1965, 1983, 1988), a generalization of

conditionalization in which new evidence alters the probabilities of a dis-
joint family of events; and

3) reparation (Jeffrey 1991, 1995; Wagner 1997, 1999), a revision method

E.l. Olsson (ed.), The Epistemology of Keith Lehrer, 143-152.
© 2003 Kluwer Academic Publishers.

that raises the probability of hypothesis H when it is discovered that

H implies previously known evidence E. 1

Probabilism, which allows the representation of degrees of belief by subjective

probabilities taking any values in the interval [0, 1], itself offers a profound ex-
pansion of the classical dogmatic epistemological categories "accept," "reject,"
and "suspend judgment." Yet the expressive resources ofprobabilism need to be
further enlarged. Recall that on the standard account (Ramsey 1990) subjective
probability is a measure of one's degree of confidence in the truth of a proposi-
tion or the occurrence of an event, as reflected in one's willingness to take either
side of certain bets. It is supposed that one is always capable of articulating
the precise odds governing such bets. That this dogma of precision, as Walley
(1991) has called it, is both unrealistic and unnecessary, is gaining acceptance
among students of the foundations of probability. In what follows we outline
the elementary parts of the theory of upper and lower probabilities, with the aim
of giving these ideas wider currency among epistemologists.



For the sake of simplicity it is assumed in what follows that your frame of dis-
cernment n regarding possible states of the world is finite. 2 It is also supposed
that there is an infinitely divisible unit of utility. Suppose that you are able to
assign to each event A c n a real number p(A) such that

1° You are willing to pay anything less thanp(A) units of utility in exchange
for receiving one such unit if A occurs, and nothing if A fails to occur;

2° In exchange for receiving anything more than p(A) units of utility, you
are willing to obligate yourself to pay one such unit if A occurs, and
nothing if A fails to occur.

The set function p is your subjective probability on events in n, with p( A) being

your threshold price/or A.3 Standard Dutch book arguments (see, e.g., Earman
1992, pp. 38-40) show that in order to avoid a sure loss, p must be coherent,
i.e., satisfy the usual axioms for a probability measure,

p(A) ~ 0, for all A en, (2.1)

p(n) = 1, and (2.2)
p(A U B) = p(A) + p(B), for all A, Ben such that A n B = 0. (2.3)

In section 4 we show as a simple corollary of a much more general result that

coherence is also sufficient to avoid a sure loss.
The demands placed on probability assessors by the dogma of precision
are stringent, and unrealistic. How, for example, is one to assess the probability
of getting a white ball in a random selection from an urn containing red and
white balls in unknown proportion?4 An additional such example, the case of
incompletely specified contingency tables, is described in the next section. As
we shall see, an elegant analysis of this case, due to Strassen (1964), leads
naturally to a simple, intuitively appealing general account of upper and lower


Imagine a collection of objects, consisting of spheres, cylinders, cubes, and

cones, each of which is colored red, white, or blue. Suppose that 50 % of the
objects are red, 30 % are white, and 20 % are blue. There are no red cubes or red
cones, no white cylinders or white cubes, and no blue spheres. An object is cho-
sen at random from this collection. What is the probability that it is 1. a sphere,
2. a sphere or cylinder? This problem involves the incomplete contingency table
sphere cylinder cube cone

cod I

and furnishes another example in which information is insufficient to assess

precise probabilities. On the other hand it is fairly easy to see, e.g., that no more
than 80 % of the objects can be spheres, and that at least 50 % of the objects are
spheres or cylinders.
Strassen (1964) furnished the following elegant analysis of the general
problem of this type: Let 0 and 8 be frames of discernment regarding the state
of the world, let p be a probability on events in 0, and suppose that for each
w E 0 the set T(w), comprising those outcomes E 8 compatible with the
outcome w, is nonempty. For all A c 8, let
A* := {w EO: T(w) c A}, and (3.1)
A* := {w EO: T(w) n A =I 0}, (3.2)
and let
p(A) := p(A*), and (3.3)
a(A) := p(A*). (3.4)

The set functions (3 and a are, respectively, the Strassenian lower and upper
probabilities 5 on events in 8 induced by p and the compatibility relation T.
The reasons for this terminology will soon be made clear, but let us first note
some basic properties of these set functions.
Theorem 3.1. The set functions (3 and a defined by (3.3) and (3.4) have the
following properties:
(i) 0 ~ (3(A) ~ a(A) ~ l,for all A c 8.
(ii) (3(0) = a(0) = 0 and (3(8) = a(8) = 1.

(iii) (3 and a are monotone, i.e., if Al C A 2, then (3(AI) ~ (3(A 2) and

a(At) ~ a(A2)'

(iv) (3 and a are conjugates, i.e., (3(A) + a(A) = lfor all A C 8.

(v) For every positive integer r, (3 is r-monotone and a is r-alternating, i.e.

(3(AI U ... U AT) ~ L (3(Ai)


- L(3(Ai n Aj) + ... + (-lr- I (3(A I n .. · nAT) (3.5)


a(AI n ... n AT) ~ L a(Ai)


- L a(A U Aj) + ... + (-lr-Ia(AI u··· U AT), (3.6)

from which it follows that

(vi) (3 is superadditive and a is subadditive, i.e., if Al nA 2 = 0, then (3(AI U

A 2) ~ (3(At) + (3(A2) and a(AI U A 2) ~ a(At) + a(A2)'
Proof The proofs of (i)-(iii) are straightforward. The proof of (iv) follows from
the fact that A* and (A)* are set-theoretic complements. To prove (3.5) note that
(AI U ... U A T)* => (At)* u ... U (AT)* and that for every I C {I, ... ,r}

Then apply monotonicity of p and the principle of inclusion and exclusion for p.
The inequality (3.6) follows from (3.5) and (iv). Assertion (vi) follows from the
case r = 2 of(3.5) and (3.6). 0

Theorem 3.2. The following are equivalent:

(i) If p( w) > 0, then w is compatible with exactly one outcome e E 8.

(ii) {3 is a probability measure.

(iii) 0: is a probability measure.

(iv) {3 = 0:.

Proof Straightforward. o
The above theorem simply confirms what one would expect, namely, that
when T is in effect a 8-valued random variable, then 0: and {3 coincide with the
usual probability measure induced on events in 8 by p and T.
We now demonstrate the appropriateness of calling 0: and (3 upper and
lower probabilities. In what follows, fA denotes the indicator of the event A,
i.e., fA(e) = 1 ife E A and fA(e) = 0 ife E A, and we routinely omit the
phrase "units of utility." The betting commitments described in 10 and 2 0 of
section 2 above will be tersely characterized, as a willingness to "buy fA for
p(A) - c, for all c > 0," and "sell fA for p(A) + c, for all c > 0."
Suppose that p is your subjective probability on events in D, with T, 0:,
and (3 as above. Except in the case described in Theorem 3.2, you have inad-
equate information to ground assessment of a subjective probability on events
in 8. There are, however, identifiable constraints on any such probability.

Theorem 3.3. Any coherent probability q that might be a candidate for repre-
senting your threshold prices for events in 8 must satisfy

(3(A) ~ q(A) ~ o:(A), for all A c 0), (3.7)

for if (3.7) is violated, you will suffer a sure loss.

Proof Suppose that q(A) < (3(A) for some A c 8, with (3(A)-q(A) = c > O.
You'll sell fA for q(A) + c/4 and buy fA. for p(A*) - c/4 = (3(A) - c/4.
Suppose that w is the true D-state and that is the true 8-state. If w E A*, then
e E A, and so your net gain is (q(A) +c/ 4-1) + (1- (3(A) +c/ 4) = -15/2 < O.
If w 1: A*, mayor may not be an element of A. If E A, your net gain is e
(q(A) + c /4- 1) + (0 - (3(A) + c / 4) = -1 - c /2 < 0, and if 1: A, your net e
gain is (q(A) + c/4 - 0) + (0 - (3(A) + c/4) = -c/2 < O.
Supposethatq(A) > o:(A). Thenq(A) = 1-q(A) < l-o:(A) = (3(A),
by Theorem 3.1 (iv), which leads, as above, to a sure loss. 0

As noted above, you are perfectly justified in the above situation in refus-
ing to announce threshold prices for events in 8. But there are some additional
bets that you ought to be willing to make:
Theorem 3.4. If f3 and a are defined by (3.3) and (3.4) you ought to be willing,
for each A c 8, and for all c > 0, to
1° buy lAfor f3(A) - c, and

2° sell lAfor a(A) + c.

Proof The argument here is not that you will otherwise suffer a sure loss (no
one can make a Dutch book against someone unwilling to bet), but, rather, that
1° and 2° are at least as good as bets you are willing to make. In the case of 10,
you'll buy lA. for p(A*) - c = f3(A) - c, so you ought to be willing to buy lA
for that price, since if the true D- and 8-states are, respectively, wand e, and
w E A*, then e E A, so lA pays off for you in every case that lA* does, and
possibly other cases as well.
In the case of 2°, you'll sell lA* for p(A*) + c = a(A) + c, so you ought
to be willing to sell I A for that price. With wand e as above, if w t/: A *, then
T(w) n A = 0, i.e., T(w) c A. Since e E T(w), e t/: A. So in every case in
which you avoid paying off on IA* (and perhaps in other cases as well) you will
avoid paying off on IA. 0


Theorem 3.4 leads naturally to the following generalization of classical subjec-

tive probability: Suppose that for each event A c D you are able to assign real
numbers A(A) and v(A) such that, for all c > 0, you are willing to
1° buy lA for A(A) - c, and

2° sell IA for v(A) + c.

The set functions A and v are then, respectively, your lower and upper subjective
probabilities on events in D. Whereas in section 2 above your threshold prices
as bettor and bookie were required to be identical, here they may be distinct,
with obvious gains in realism and expressive possibility. The following theorem
gives necessary and sufficient conditions for avoiding a sure loss in the above
Theorem 4.1. Given the betting commitments 1° and 2° above, you will avoid
a sure loss if and only if there exists a coherent probability q on events in D such
A(A) ~ q(A) ~ v(A), for all A c D. (4.1)

Proof See Walley 1981, p. 15. o

It follows immediately from Theorem 4.1 that coherence of subjective
probabilities, as defined in section 2 above, is not only necessary, but also suffi-
cient to avoid a sure loss (cf. Kemeny 1955 and Lehman 1955).
Note that (4.1) is a rather weak condition, which, for example, does not
even imply monotonicity of A and u. Buehler (1976) has argued for much more
stringent restrictions. Indeed, he claims inter alia that the lower probability A
must be additive! Buehler's argument is based on the following theorem.

Theorem 4.2. Suppose that A(A) ~ q(A)for all A c n, where q is a coherent

probability, and that A is self-conjugate, i.e., A(A) + A(A) = 1 for all A c D.
Then A = q.

Proof Suppose that A(A) < q(A) for some A. Then, by self-conjugacy of A,
A(A) = 1- A(A) > 1-q(A) = q(A), contradicting thefact that A is dominated
byq. 0

It follows from Theorems 4.1 and 4.2 that if A avoids a sure loss, and is
self-conjugate, then A is in fact a coherent probability. Buehler's argument that
A must be self-conjugate goes as follows: Clearly, A(A) + A(A) ~ 1; otherwise
you will suffer a sure loss. Suppose that A(A) + A(A) < l. Let a and b be such
that A(A) < a, A(A) < b, and a + b < l. Then you'll reject buying fA for a
and fA for b, even though by accepting both bets you would be guaranteed of
the net gain 1 - a - b > O. So you will miss out on a sure gain. Apart from the
fact that missing a sure gain is considerably less serious than suffering a sure
loss, this argument is further weakened by its dependence on your being offered
fA and fA one at a time, with no knowledge that both will be offered. If you
were offered these bets simultaneously, you would recognize immediately that
you were being offered a certain payoff of 1, and would clearly agree to pay any
price less than 1 in exchange.
Here is a simple way in which lower and upper probabilities satisfy-
ing (4.1) arise: Let P be a nonempty family of coherent probabilities on events
in D and let

Ap(A) := iuf {p(A)}, and (4.2)


vp(A) := sup{p(A)}. (4.3)


Then Ap and Up satisfy (4.1) as well as the conjugacy relation Ap(A)+up(A) =

l. The set functions Ap and Up are, respectively, the lower and upper envelopes

Examples of naturally occurring families 'P (called probasitions in Jef-

frey 2001) include such things as 1. the family of all additive representations
of some comparative probability relation (Roberts 1976) and 2. the family of
all probabilities with respect to which a fixed random variable has a fixed ex-
pected value. The Strassenian lower and upper probabilities (3 and a are also
envelopes, with 'P = the set of all marginalizations to e of all probabilities Q
on 0 x e that are compatible with p and T in the sense that the marginalization
of Q to 0 is p and Q(w, e) = 0 if e tt T(w) (Wagner 1992).


Dogmatic restrictions on the representation of uncertain judgment, or on the way

in which such judgment is revised, undermine the goal offaithfully representing
the evidence regarding the state of the world. While Bayesian dogmatism has
begun to yield to other principled methods of probability revision, the dogma of
precision is still dominant.
One source of resistance to working with non-additive upper and lower
probabilities is the fear that such measures must necessarily be mathematically
intractable. This greatly exaggerates the true state of affairs. While space does
not permit a detailed account, we mention that there is a useful theory of upper
and lower expectation (see Dempster 1967 and Walley 1981, 1991), as well as a
generalization of probability kinematics in which new evidence places bounds
on possible revisions of prior in the form of Strassenian upper and lower prob-
abilities (Wagner 1992). Finally, to conclude this paper on the same note on
which it began, we remark that there is a theory of consensus for upper and
lower probabilities (Wagner 1989) which is remarkably similar to that in Lehrer
and Wagner 1981.


* Research supported by the National Science Foundation (SES-9984005)

1 Reparation thus provides a solution to the old evidence problem, first posed by Glymour (1980).
2 The term "frame of discernment" is due to Shafer (1976). The elements of 0 are mutually
exclusive and exhaustive, i.e., precisely one element of 0, though typically unknown, represents
the true state of the world. Multiple frames of discernment may, however, be brought to bear on
a single problem, as, for example, when we classify the possible outcomes of selecting an object
at random from a set of colored shapes by frames delineating I) the possible colors and 2) the
possible shapes. Mathematicians typically refer to 0 as a "sample space."
3 It is implicit here that you are unwilling to pay more than p( A) in 10 and unwilling to take less
than p(A) in 2°. In the usual treatment, you must be willing to pay p(A) in 10 and to take p(A)
in 2°. We do not require this. A virtue of our treatment is that one can assign proper subsets of 0
(i.e., contingent events) probability one without being in the position of having no prospect of

positive gain, and the possibility of a loss (cf. Eannan 1992, p. 41). Our treatment also allows for
a natural segue to the account of upper and lower probabilities in section 4.
4 Devotees of the principle of insufficient reason would adopt the unifonn distribution here, thus
employing the same distribution in the case of complete ignorance that they would given reliable
infonnation that exactly half the balls in the urn are white.
5 In 1967 Dempster, unaware of Strassen's 1964 paper, published a similar analysis. Shafer
(1976) offered a sui generis account of set functions having the monotonicity properties of the
lower probability (3, regarding such set functions, which he called beliejjimctions, as directly
assessable measures of degrees of belief.

Buehler, R. 1976. "Coherent Preferences." Annals ojStatistics 4: 1051--1064.
Dempster, A. 1967. "Upper and Lower Probabilities Induced by a Multivalucd Mapping." Annals
of Mathematical Statistics 38:325-339.
Earman,1. 1992. Bayes or Bust? Cambridge: MIT Press.
Glymour, C. 1980. Theory and Evidence. Princeton: Princeton University Press.
Jeffrey, R. 1965. The Logic of Decision. New York: McGraw-Hill; 1983, 2nd ed., Chicago:
University of Chicago Press.
----. 1988. "Conditioning, Kinematics, and Exchangeability." In B. Skyrms and W. Harper
(eds.), Causation, Chance, and Credence, vol. I, Dordrecht: Kluwer,221-255.
- - - . 1991. "Postscript 1991: New Explanation Revisited." In Jeffrey 1992, 103-107.
- - - . 1992. Probability and the Art ofJudgment. Cambridge: Cambridge University Press.
'---. 1995. "Probability Reparation: The Problem of New Explanation." Philosophical Stud-
ies 77:97-102.
2001. "Petrus Hispanus Lectures, II: Radical Probabilism." Actas da Socieda-
de Portuguesa da Filosofia (forthcoming; and currently available in http://www .
Kemeny, 1. 1955. "Fair Bets and Inductive Probabilities." Journal o/Symbolic Logic 20:263-273.
Lehman, R. 1955. "On Confinnation and Rational Betting." Journal ojSymbolic Logic 20:251-
Lehrer, K. and Wagner, C. 1981. Rational Consensus in Science and Society. Dordrecht: Reidel.
Ramsey, F. 1990. "Truth and Probability." In Philosophical Papers ofF P Ramsey, D. H. Mellor,
ed., Cambridge: Cambridge University Press.
Roberts, F. 1976. Discrete Mathematical Models. Englewood Cliffs: Prentice-Hall.
Shafer, G. 1976. A Mathematical Theory oj Evidence. Princeton: Princeton University Press.
Strassen, V. 1964. "MeBfehler und Information." Zeitschri/i fur Wahrscheinlichkeitstheorie
Wagner, C. 1989. "Consensus for Belief Functions and Related Uncertainty Measures." TheOlY
and Decision 26:295-304.
1992. "Generalized Probability Kinematics." Erkenntnis 36:245-257.
----. 1997. "Old Evidence and New Explanation." Philosophy ojScience 64:677-691.
- - - . 1999. "Old Evidence and New Explanation II." Philosophy oj Science 66:283-288.
Walley, P. 1981. "Coherent Lower (and Upper) Probabilities." Technical Report, Department of
Statistics, University of Warwick, Coventry, England.
- ---'-. 1991. Statistical Reasoning with Imprecise Probabilities. London: Chapman and Hall.
Chapter 10



James Van Cleve

Brown University

Weare indebted to Keith Lehrer for his groundbreaking work on the

philosophy of Thomas Reid. He has done a great deal to make the interest and
importance of Reid's philosophy clear. I would like to thank him for this, and
also to raise certain questions concerning his interpretation of Reid's
epistemology. I shall be especially concerned with Lehrer's view that one
among Reid's principles of common sense stands out as an indispensably
important metaprinciple.
In Essay VI, Chapter 5 of the Essays on the Intellectual Powers of Man,
Reid presents us with a list of "first principles of contingent truths." Some of
these principles are purely or primarily metaphysical; for example, Principle 2
tells us that thoughts require a thinker. But others are plainly intended to have
epistemological significance, proclaiming the trustworthiness of consciousness
(Principle 1), memory (Principle 3), perception (Principle 5), our faculties in
general (Principle 7), our beliefs concerning the minds of others (Principles 8
and 9), testimony (Principle 10), and induction (Principle 12). My concern here
is with the proper understanding of the epistemological principles, about which
I shall discuss three issues in particular. First, are the epistemological principles
on Reid's list principles of truth or principles of evidence? Second, are they
principles that must themselves be known if knowledge is to arise in accordance
with them? Given Lehrer's interpretation of the principles, this amounts to the
question whether knowers must have knowledge of the reliability of their own
cognitive faculties. Third, what special role is played by Reid's Principle 7,
which says that our natural faculties are not fallacious? This is the principle that
E.1. Olsson (ed.), The Epistemology of Keith Lehrer, 155-172.
© 2003 Kluwer Academic Publishers.

Lehrer calls the keystone of Reid's system, of crucial importance for the way in
which it supports the other elements and itself as well.



According to Lehrer, Reid's principles are in the first instance principles

of truth rather than principles of evidence. He writes, "When one considers the
first principles Reid articulates, one finds beliefs telling us about truth instead
of evidence.'" He holds that in the end Reid's principles do add up to a theory
of evidence, but that is because "evidence for Reid just is information about
what is true or false.,,2 He elaborates as follows:

The first principles state that the convictions of our faculties are
true rather than evident, but the information that our convictions
are true is the evidence that grounds them. These first principles
are, therefore, principles of evidence as well as principles oftruth.3

To anticipate Lehrer's answer to the second of my three questions, I believe he

thinks Reid's principles are principles of evidence only in so far as the subject
knows them to be true. I am going to propose an alternative interpretation
according to which Reid's principles are principles of evidence in their own
right, regardless of whether the subject has any knowledge of them.
We must begin by answering the question, What is a first principle? As
the name implies, it is a principle that comes first in all reasoning or inquiry; that
is to say, it is a principle on which we base other beliefs, but which is based on
nothing in turn. In company with tradition going back to Aristotle, Reid thinks
that a principle is fit to play this role only if it is self-evident:

It is demonstrable, and was long ago demonstrated by Aristotle,

that every proposition to which we give a rational assent, must
either have its evidence in itself, or derive it from some antecedent
proposition. And the same thing may be said of the antecedent
proposition. As, therefore, we cannot go back to antecedent
propositions without end, the evidence must at last rest upon
propositions, one or more, which have their evidence III
themselves, that is, upon first principles. (EIP 6.7, p. 685)4

Reid affirms the self-evidence of first principles in many other passages

as well, including this one:

There is no searching for evidence; no weighing of arguments; the

proposition is not deduced or inferred from another; it has the light
of truth in itself, and has no occasion to borrow it from another.
Propositions of the last kind, when they are used in matters of
science, have commonly been called axioms; and on whatever
occasion they are used, are called first principles, principles of
common sense, common notions, self-evident truths. (ElP 6.4, p.
593, with a paragraph break omitted)

I shall take it, then, that the central trait of first principles is their self-evidence.
A cautionary note: I do not necessarily equate Reid's epistemological
principles with first principles. As already noted, there are some first principles
(e.g., axioms of mathematics or metaphysics) that are not epistemological
principles. More significantly, I believe one should be open as well to the
converse possibility that Reid's epistemological principles are not first
principles themselves-that they specify first principles without being first
principles. So the questions I raise in this section and the next about Reid's
epistemological principles are not necessarily questions about principles he
regards as first principles.
With this in mind, let us tum to the question whether the epistemological
principles in Reid's list are principles of truth or principles of evidence. I think
the answer depends on a crucial but little noted scope ambiguity in the wording
of Principle 1 and several other of the principles as well. Here is how Reid
formulates Principle 1:

First, then, I hold, as a first principle, the existence of every thing

of which I am conscious. (EIP 6.5, p. 617)

Two preliminary observations are necessary. First, by 'consciousness' Reid

means that by which we have knowledge ofthe operations of our own minds; it
is roughly equivalent to introspection. Second, when Reid says "if I am
conscious of a certain thought, it exists," we may just as well say "if I am
conscious that I am thinking thus and so, then I am thinking thus and so." That
will facilitate symbolization without distorting Reid's views. The ambiguity to
which I wish to call attention may now be brought out by the following two
ways of symbolizing Principle 1 (where 'Cp' is short for 'I am conscious that

lA. It is a first principle that (P)(Cp -> p).

lB. (p)(Cp -> it is a first principle thatp).

Notice that what 1A specifies as a first principle is a principle oftruth-a single

principle laying down that all the deliverances of consciousness are true. By
contrast, 1B is a principle laying down that each of the deliverances of
consciousness is itself a first principle. Unlike lA, which gives us one general
first principle, 1B gives us many particular first principles
I believe the same ambiguity characterizes Reid's other epistemological
principles as well. The principle about perception, for example, with 'Pp'
abbreviating 'I perceive thatp', could be understood in either of the following

SA. It is a first principle that (P)(Pp -> p).

5B. (P)(Pp -> it is a first principle thatp).

The relevance of this ambiguity to our question in this section should be

obvious. If the first style offormulation of each principle is correct, then Reid's
epistemological principles are in the first instance principles of truth, affirming
the reliability of our various faculties. They would amount to principles of
evidence only in the company of further assumptions connecting evidence with
truth, such as Lehrer's assumption that evidence is information concerning what
is true and false. s But if the second style of formulation is correct, Reid's
epistemological principles are principles of evidence in their own right (despite
not overtly containing the term 'evidence '). They are epistemological principles
attributing first-principle status to the propositions in various classes (those
attested by consciousness, memory, perception, and so on), and as we have seen,
such status is to be explicated in terms of self-evidence. So Reid's
epistemological principles would be principles specifying what count as first
principles, and it would be a further question whether they are first principles
Although Lehrer does not explicitly consider the difference between 1A
and 1B, I believe his discussion of Reid's epistemological principles
presupposes a reading of them in the A style. He takes Reid's epistemological
principles to be first principles, and he says that these principles "state that the
convictions of our faculties are true rather than evident.,,6 Both of these things
are in keeping with the A style. My own view (for which I have argued at length
elsewhere) is that although Reid himself may not have been entirely clear about
the difference between the A and the B styles, he says many things that favor a
reading of his epistemological principles in the B style. 7 My purpose here is not
to discuss further the relative merits of the A and B readings, but to explore
some of the facets of Lehrer's interpretation that go naturally with his A reading
of the principles. Suffice it to say that although both Lehrer and I think that
Reid's epistemological principles amount in the end to principles of evidence,
I take them to be self-sufficient principles of evidence in a way that he does not.

This difference between us is connected with a difference on another point, to

which I now turn.



If Reid's epistemological principles are construed in the way I suggest, there is

a sense in which he is an epistemological externalist. It is this: the mere fact
that a proposition is a deliverance of perception, memory, or consciousness
suffices to make that proposition evident. In order to know that there is a tree
in front of one, for example, one need only perceive that there is. Nothing else
is necessary. In particular, it is not necessary that the subject know anything
about the reliability of sense perception. He need take no thought of that. For
the externalist Reid, consciousness, memory, and perception are knowledge-
conferring (or at least evidence-conferring) factors that do their job regardless
of whether the subject knows anything about their justificatory power.
There are many places in Reid where I believe the externalist strain in his
thinking is prominent. For example, he says that when light from an object
strikes the lower portion of one's retina, that gives one knowledge that the object
lies in an upward direction from the eye (Inq.6.12, p. 124-25).8 More generally,
he says that signs produce "knowledge and belief' of the things they signify
(Inq. 6.24, pp. 190ff.). For example, by a law of our constitution, certain tactile
sensations produce in us the belief in a hard, extended object; this belief can be
knowledge even for one who has no knowledge that such sensations are reliably
connected with such objects.
Lehrer has always been opposed to externalist views in epistemology,
however, and he finds Reid to be no less opposed:

Reid is not a reliabilist of the sort Goldman describes. It would be

possible for a belief to be a product of a reliable belief-forming
process without our having any idea that this was so, without, that
is, our having any information about the trustworthiness of the

The theory of evidence is based on the innate first principles of the

mind. It is, however, not sufficient for a belief to be evident that
it be a product of an innate principle, even a trustworthy and
reliable one. A belief could be the product of such a principle and

not be evident for the person because the person had no idea
whether the belief originated in a way that is trustworthy or
deceptive. In fact, the first principles of our nature not only yield
beliefs but also information about those beliefs, to wit, that they
are trustworthy and not fallacious in origin. 1o

Lehrer is implying (in disagreement with what I suggested in the opening

paragraph of this section) that in order to be justified in believing that there is
a tree over there, it is not enough to perceive that there is a tree there, even if we
do so in accordance with a reliable innate tendency of our constitution. We must
in addition have the idea (or the information) that perception is a reliable source
of belief.
What is it to have such information? Is it merely to believe that
perception is reliable? Or is it something stronger-to believe with evidence
and truth, and thus to know, that perception is reliable? Let us consider each of
these requirements in turn.
First, in order to have knowledge from a source, must we believe that the
source is reliable? If so, then the generality of mankind do not know very much,
for they are seldom as reflective as that. A child or an unreflective adult may
take little thought about the faculties whose deliverances she accepts. I believe
that this is Reid's own position, as shown in the following passage:

We may here take notice of a property of the principle under

consideration [that our faculties are not fallacious] that seems to be
common to it with many other first principles ... and that is, that
in most men it produces its effect without ever being attended to,
or made an object of thought. No man ever thinks of this principle,
unless when he considers the grounds of skepticism; yet it
invariably governs his opinions. When a man in the common
course of life gives credit to the testimony of his senses, his
memory, or his reason, he does not put the question to himself,
whether these faculties may deceive him; yet the trust he reposes
in them supposes an inward conviction, that in this instance at
least, they do not deceive him. (EIP 6.5, p. 632)

It is another property of this and of many first principles that they

force assent in particular instances more powerfully than when
they are turned into a general proposition. (Ibid.)

Mindful of passages such as these, Lehrer qualifies the requirement that

subjects must believe in the reliability of their own faculties. He says that
principles such as 'What I perceive to be the case is generally so' must be

"operative" in us even if we do not explicitly believe them, II and that general

principles must be "in us" in the sense that they cause us to believe their
particular instances. 12
In what sense do we have general beliefs that cause us to have beliefs in
their particular instances? Two observations will help to clarify this matter.
First, what Reid and Lehrer mean by an "instance" of a general belief is really
an instance of its consequent, obtainable by subsumption (i.e., universal
instantiation and modus ponens). If the general proposition is' All deliverances
of perception are true', symbolizable as '(P)(Pp -> p)', then an instance of it
would be 'there is a tree over there' (if that is what I am now perceiving)Y
Second, there is a sense in which one believes implicitly that all Fs are Gs
simply by having the disposition to believe, concerning anything one believes
to be F, that it is G. In symbols, we could say that one has an implicit belief in
'(x)(Fx -> Gx)' if one has the disposition expressed by 'BFx -> BGX.'I4 Ifone
has such an implicit belief, one will be caused to believe instances of the
consequent of'(x)(Fx -> Gx)' whenever one believes instances of its antecedent.
Even if a man never framed in his mind the proposition that perceptual beliefs
are generally true, he might still be so constituted that whenever he believes he
perceives that p, he also believes that p. (Alternatively, if we wish to
accommodate the possibility that sometimes we perceive things without
believing that we do, we could say that a general belief in the reliability of
perception is operative in S if S is so constituted that whenever he is conscious
of perceiving that p (e.g., that there is a tree over there), he believes that pys
We have now identified one plausible sense in which ordinary subjects
believe in the reliability of their faculties. Even if they do not have general
propositions such as 'When I perceive something to be the case, it is the case'
explicitly before their minds, their opinions are governed by them, in the sense
that whenever they are aware offalling under the antecedent, they will believe
the consequent.
Let us now tum to the stronger requirement envisioned above-that to
know something through a faculty, you must know that the faculty is reliable.
I believe that Lehrer accepts this requirement and that he thinks Reid does, too.
Of course, if the sense in which we believe in reliability is only the implicit
sense just identified, the sense in which we have knowledge of reliability would
presumably be implicit in a corresponding sense. I will ignore this complication
Following Stewart Cohen, the requirement we are now considering may
be put thus: I6

(KR) A potential knowledge source K can yield knowledge for a

subject S only if S knows K is reliable.

Given the way Lehrer understands Reid's epistemological principles (as

principles affirming that the deliverances of our faculties are generally true), th is
would mean that Reid's epistemological principles contribute to a subject's
knowledge only if they are themselves known.
It takes but little reflection to see that KR is actually quite a stringent
requirement-one that threatens to lead to skepticism. It would lead indeed to
skepticism if we made two further assumptions. The first assumption is that
what KR lays down as a necessary condition of knowledge through K is a prior
condition of such knowledge, that is, that S can know through K that p only if S
first knows that K is reliable. The second assumption is that knowledge of the
reliabiJity of a source can come about in one way only, namely, by inference
from premises obtained from that very source. To illustrate, Descartes tried to
establish the reliability of clear and distinct perception by proving the existence
and veracity of God, but could obtain the premises for the proof only through
clear and distinct perception itself. Others have envisioned proving that a
faculty is reliable by appeal to the reliable operation of that faculty in the
past-a so-called track record argument-but obtaining the premises for the
track-record argument requires using the faculty on whose behalf it is
conducted. 17 In sum, it seems that we must know that various of the particular
deliverances of a source are true before we can know that the source is reliable.
Putting these points together, we would find ourselves in the following

(1) We can know that a deliverance of K is true only if we first

know that K is reliable.
(2) We can know that K is reliable only if we first know,
concerning certain of its deliverances, that they are true.

If (1) and (2) are both true, skepticism is the inevitable consequence. Clearly,
if we cannot know either of two things without knowing the other first, we must
remain ignorant of both of them.
How would Reid, the great foe of the skeptic, respond to this threat? If
Reid is an externalist, as I have suggested he is at least some of the time, he
would deny KR and with it proposition (l). Consciousness, memory,
perception, and our other faculties give us knowledge of their deliverances even
if we are initially ignorant of their reliability. But Lehrer's Reid is not an
externalist, and it must be admitted that there are places where Reid seems to
accept a requirement like KR. Here is a notable passage:

If any truth can be said to be prior to all others in the order of

nature, this [that our faculties are not fallacious] seems to have the
best claim; because in every instance of assent, whether upon

intuitive, demonstrative, or probable evidence, the truth of our

faculties is taken for granted, and is, as it were, one of the premises
on which our assent is grounded. (EIP 6.5, p. 631-32).

Knowledge of anything whatever, Reid seems to be saying, depends on

knowledge of a special fact-that our faculties are reliable. Is he saying that
knowledge of anything requires knowledge that all of our faculties are reliable,
or only that knowledge through a given faculty requires knowledge that that
faculty is reliable?18 Either way, the question arises how we are to acquire
knowledge of the special fact. If Reid accepts proposition (2) alongside
proposition (1), he will have made such knowledge impossible. But Lehrer's
Reid, as we shall see, evades this difficulty by rejecting proposition (2). I shall
return to this matter after discussing Lehrer's interpretation of Principle 7.


Here is how Reid formulates the seventh of his principles:

7thly, Another first principle is, that the natural faculties, by which
we distinguish truth from error, are not fallacious. (EIP 6.5, p.

According to Lehrer, Principle 7 is the most important principle of al1. 19 It is

special in two ways. First, it is a metaprinciple, affirming the truth of all the
others.zo Second, it is a looping principle. "The principle vouches for itself. It
loops around and supports itself."2l For these reasons, Lehrer calls this principle
"the keystone principle of the first principles."n He also calls it "the first first
But if Principle 7 is really the first in importance, why does it occur
seventh in a list of twelve? One would have expected that if Principle 7 plays
a pre-eminent role, Reid would have put it either at the top or bottom of his list
or perhaps outside the list altogether, rather than burying it in the middle.
There are ways of understanding Principle 7 as co-ordinate with the other
principles, treating of just certain faculties among others, and thus deserving of
its place in the middle ofthe list. Philip de Bary has noted that the clause "by
which we distinguish truth from error" may have been intended by Reid as a
restrictive clause, singling out a subset of the faculties, rather than as a
parenthetical gloss on what all faculties do. 24 And yet do not all faculties
enable us to distinguish truth from error, informing us that some things are true
and others false? So what restriction could Principle 7 be introducing?

One suggestion for a restricted Principle 7 would be to take the faculties

"by which we distinguish truth from error" to be second-order faculties, that is,
faculties whereby we judge of other faculties. 25 Ifwe do this, Principle 7 would
be a principle about meta-faculties, but it would not necessarily be a meta-
principle. Nor would it be a principle of supreme generality, for it would
concern some faculties among others, even if those faculties are of special
importance. Against this suggestion, however, is the fact that Reid twice uses
the phrase "until God give us new faculties to sit in judgment upon the old" (6.5,
p. 631; 6.7, p. 678), implying that we have nothing to judge of our faculties but
those faculties themselves.
Another suggestion would be to take Principle 7 as concerned with
reasoning, which is not mentioned elsewhere in Reid's list of principles. 26
Reasoning, Reid tells us, is "the process by which we pass from one judgment
to another" (p. 475 in Hamilton). There is some support for this suggestion in
the fact that many of Reid's illustrations of Principle 7 are cases of reasoning.
For example, immediately following his enunciation of Principle 7, he says, "If
any man should demand a proof of this it is impossible to satisfy him ... because
to judge of a demonstration, a man must trust his faculties" (EIP 6.5, p. 630;
emphasis mine). He follows this up by observing that the very point in question
in Principle 7 is whether reasoning may be trusted.
These remarks are not decisive, however. If Principle 7 concerned all our
faculties, as Lehrer holds, it would concern reasoning, too, so Reid could still
make his point that it would beg the question to offer reasoning in support of
Principle 7. Moreover, as we read further in Reid's commentary on Principle 7,
we find him saying that it is concerned with "all our reasoning and judging
powers" (pp. 630 and 632), which suggests a considerably broader scope for
Principle 7.
But how much broader? De Bary suggests a broader but still restricted
scope for Principle 7. Noting that Reid devotes separate Essays in the
Intellectual Powers to perception, memory,judging, and reasoning, he proposes
that the scope of Principle 7 is just judging and reasoning, not our faculties in
general. Against this suggestion, however, is the fact that Reid views perception
and memory as special cases of judgment. He says, for example, "There is no
more reason to account our senses fallacious, than our reason, our memory, or
any other faculty ofjudging which nature has given us" (EIP 2.22, emphasis
mine). Confirming this point, Essay 6, "Of Judgment," makes points applicable
to all our faculties; indeed, it is in that Essay that we find the list of principles
that is our topic in this essay. Moreover, as we read Reid's thirteen paragraphs
of commentary on Principle 7, I think it becomes fairly clear that Principle 7 is
meant to cover all our faculties. The trust we repose in our senses, our memory,
our consciousness, and our reason are all said to be instances ofthe general trust
that is affirmed in Principle 7.27

I shall assume with Lehrer, then, that Principle 7 does indeed concern all
of our cognitive faculties-not just some specially delineated faculty or
And yet the question remains: In what way does Principle 7 go beyond
the other principles on Reid's list? Why is it not merely redundant-perhaps
simply a device enabling Reid to make points concerning all the principles by
discussing a single one?
To this question, Lehrer's answer is that Principle 7 is a metaprinciple,
affirming the truth of the other principles, and a looping principle, affirming the
truth of itself as well. In this last way it goes beyond the other principles.
If we take Reid's formulations at face value, however, there does not
appear to be anything particularly "meta" about Principle 7. The sequence

Consciousness is reliable.
Memory is reliable.
Perception is reliable.
All our faculties are reliable.

is comparable with the sequence

My dogs are friendly.

My cats are friendly.
My birds are friendly.
All my pets are friendly.

The last item on each list does not seem to be much more than a summary of
what has gone before. If Principle 7 goes beyond the other principles, it is only
by virtue of implying either that I have no faculties not already mentioned or, if
I do, that they are reliable, toO.28
Nonetheless, the "meta" and "looping" character of Principle 7 may be
brought out more clearly if we rewrite Principle 7 in a way Lehrer has proposed.
Principle 7 may be recast, he suggests, as a principle affirming that all first
principles are true. 29 So construed, Principle 7 does indeed convey what the
other principles convey (that consciousness is reliable because Principle I is
true, memory reliable because Principle 3 is true, and so on). But Principle 7
also conveys more, because it is itself a first principle. It therefore implies its
own truth by way of self-subsumption:

All first principles are true (= Principle 7).

Principle 7 is a first principle (i.e., it is a first principle that all first
principles are true).

Therefore, Principle 7 is true.

The self-subsuming character of Principle 7 may not have been obvious in

Reid's formulation, because 'all our faculties are reliable' talks of faculties
rather than principles. 30 But Lehrer's rewrite makes the self-subsuming feature
manifest, and it is precisely this feature, he thinks, that enables Principle 7 to
play its keystone role.
I have two questions about Lehrer's rewrite. First, is it a legitimate
transcription of Reid's original? 'All first principles are true' would be
equivalent to' All our faculties are reliable' provided the following biconditional
were true: P is a first principle iff P is a principle affirming the reliability of
some faculty or faculties of ours. The right-to-left half of this biconditional
depends on whether it can be self-evident that a faculty is reliable-a debatable
question, but one to which Lehrer clearly thinks Reid would answer yes, as we
shall see further below. The left-to-right half ofthe biconditional is problematic.
There are first principles that do not affirm reliability (e.g., particular
propositions such as there is a tree in front of me, and metaphysical principles
such as the thoughts of which one is conscious must be thoughts of a thinker).
But I shall assume that we may qualify Lehrer's rewrite of Principle 7 so as to
get around this problem.
The more interesting question is this: What do we gain by the rewrite?
Lehrer's answer is that we obtain thereby a principle that explains its own truth.
He is worried by the prospect of epistemic surds-that is, epistemological
principles for which there is no explanation. Given a choice between a surd and
a loop-that is, between an unexplained principle and a principle that explains
itself-he would always prefer the 100p.3l
By making Principle 7 self-subsuming, do we really enable it to play this
special explanatory role? A qualm about this may be engendered by considering
the following list:

2 + 2 = 4.
Aberdeen is northeast of Glasgow.
Water boils at 212 degrees F.
All the sentences in this list are true.

The concern is that the final sentence is semantically ungrounded, just like the
truth-teller sentence, 'this sentence is true'. If one of the sentences preceding it
were false, that would make the final sentence false. But if all others on the list
are true, whether the last is true comes down to whether it is true, and there
seems nothing to determine that. There is nothing to make it true or false, so
arguably it is neither. The situation seems similar with Principle 7. If the other
first principles are all true, then Principle 7 goes beyond them just to the extent

that it ventures out over the void, with nothing to sustain a truth value for it. So
the worry is that in trying to come up with a keystone principle that explains the
others and itself in the bargain, we obtain a principle that explains nothing
because it lacks truth value.
I do not wish to suggest that there is anything automatically defective
about general principles that subsume themselves. All necessary propositions
are true is itself a necessary proposition, and therefore may be subsumed under
itself. But there is no question about its truth value; it is an accepted axiom of
modal logic. Similarly, everything that God believes is true would be both self-
subsuming and true (indeed, necessarily true) if there were an essentially
omniscient and infallible God who believed it. But it is significant that these
examples of self-subsuming propositions that are unproblematically true are
necessary truths. There is a necessary connection between the properties that
figure in antecedent and consequent, and that is what makes them true. Lehrer,
however, does not think that Principle 7 is a necessary truth. He thinks that
Reid's first principles of contingent truths are themselves contingent. 32 So he
does not have this way of alleviating the worry I have raised that under his
construal of it, Principle 7 is semantically ungrounded.
I shall return to this worry at the end of the next section. It is time to see
how two aspects of Lehrer's interpretation of Reid-the KR requirement on
knowledge and the endorsement of looping principles-are connected.


The problem we left hanging at the end of section 2 was this: if we cannot know
anything through a faculty without knowing that that faculty is reliable (as
required by KR), how can we know anything at all? The problem can be
formulated as a pair of premises that jointly entail skepticism:

(1) We can know that a deliverance of K is true only if

we first know that K is reliable.
(2) We can know that K is reliable only if we first know,
concerning certain of its deliverances, that they are

When the problem is posed this way, it is clear that we can avoid
skepticism only by denying (1) or (2).
An externalist would deny (1), but Lehrer's Reid is no externalist. So let
us consider the option of denying (2)-of denying that knowledge of the
reliability of a source must be collected from various of the particular
deliverances of that source. If not derived from knowledge of its own particular

deliverances, from what other knowledge could knowledge of K's reliability be

derived? One answer is: from no other knowledge. Knowledge of the
reliability of our faculties is epistemically basic.
That is the answer given by Lehrer's Reid: we know that our faculties are
reliable not by deriving this knowledge from any other knowledge, but simply
because the reliability of our faculties is self-evident. 33 It is a self-evident first
principle that consciousness is reliable; it is likewise a self-evident first principle
that perception is reliable; and so on for all our other faculties. So we have a
way of breaking the skeptical impasse set up by propositions (1) and (2). We
can accept the condition that KR lays down for all knowledge, but affirm that
that condition is thankfully met, owing to the first principles of human
Let us look more closely at the implications of this. For Lehrer's Reid,
we are enabled to know that the particular deliverances of sense perception are
reliable because it is a piece of basic knowledge, inferred from no other, that
sense perception is reliable. We are still entitled to ask: what is the source of
this basic knowledge? Presumably, it is not perception itself, which does not
yield truths of such generality.34 If only to give the source a name, let us call it
intuition. 35 By KR, intuition yields knowledge only if we know that intuition is
reliable. What is the source of that knowledge? This time we may answer that
it is intuition itself-intuition intuits its own reliability. Indeed, it seems
plausible that we must give an answer of that form sooner or later-KR can
admit basic knowledge of the reliability of a source only if there is at least one
source that gives basic knowledge of its own reliability.
But now let's go back to the principle that perception is reliable. How can
that be a first principle if knowledge of it depends on the knowledge that some
other faculty, namely, intuition, is reliable? Perhaps the answer is something
like this: although it would not be evident (or known) that perception is reliable
unless it were evident (or known) that intuition is reliable, that is not because the
former proposition derives its evidence from the latter. Rather, intuition confers
evidence simultaneously on its primary object (in this case, that perception is
reliable) and also on its own reliability.36
The Reidian view that is emerging evidently requires that there be certain
faculties or sources that deliver knowledge not only of their primary objects, but
also of their own reliability. This may be what Reid himself is getting at in a
striking passage comparing evidence to light, which follows immediately upon
his suggestion that if any truth be prior to all others, it is Principle 7:

How then come we to be assured of this fundamental truth on

which all others rest? Perhaps evidence, as in many other respects
it resembles light, so in this also, that as light, which is the
discoverer of all visible objects, discovers itself at the same time:

so evidence, which is the voucher for all truth, vouches for itself at
the same time. (EIP 6.5, p. 632) 37

The idea would be that as light discloses features both of visible objects and
itself, so our cognitive faculties disclose features both of their primary objects
and ofthemselves-in particular, their own reliability.
With this answer, we have drawn close to Lehrer's interpretation of
Principle 7. Recall that for Lehrer, Principle 7 undergirds all the others and
itself at the same time. It is a principle that affirms its own truth along with the
truth of the other principles. We are now suggesting that there are faculties that
apprehend their own reliability along with the reliability of other faculties. We
have thus arrived at a view structurally similar to Lehrer's, incorporating the
same looping strategy he favors. And we have been led to do so in the attempt
to show that knowledge is possible even under the tough demands imposed by
KR. It is no accident, then, that a philosopher who, like Lehrer's Reid, holds
that there is no knowledge through a faculty without knowledge of its reliability,
also holds that there are faculties that vouch for their own reliability.
It is now time to return to the worry left unresolved at the end of the
previous section-that self-subsuming principles of the sort Lehrer favors would
be semantically ungrounded. Is there not an analogous worry about self-
authenticating faculties? Suppose that intuition intuits its own reliability-that
is, that I have an intuiting whose content is that all intuitings are true. If all other
intuitings are true, what would make that one true? As before, I think it would
have to be some sort of necessary connection between the properties of being an
intuiting and being true. (If it were simply a matter of the individual truth values
of all intuitings, ungroundedness would threaten.) Is there plausibly such a
connection, and could Lehrer accept it? I noted above that he does not believe
that Reid's principles are metaphysically necessary truths; he would deny that
there is any metaphysically necessary connection between being an intuiting and
being true. But I expect he might allow that there is a nomologically necessary
connection between being an intuiting and being true, and perhaps such a
connection would suffice as a truth-maker for the intuiting that all intuitings are
true. I leave further exploration ofthat possibility for another occasion.


There is a side of Lehrer's Reid I have so far left out of account. As we have
seen, Lehrer's Reid holds that general principles affirming the reliability of our
faculties are evident in themselves and known immediately. But Lehrer also
says that particular deliverances of our faculties, such as the beliefthat there is
a hard object in my hand, are evident in themselves and known immediately.

"Our knowledge of both the first principles and the particular beliefs to which
they give rise are both immediate .... The [general] principles and [particular]
beliefs of common sense fit together like links in a chain, and he that is not fit
to pick up the whole should not attempt to lift up any of the parts.,,38
Here Lehrer seems to be telling us that general principles and particular
beliefs are both self-evident, but at the same time, that neither generals nor
particulars would be evident unless both were part of our system of belief. Is
that a consistent combination?
Lehrer's Reid is beginning to sound like a coherentist, for whom
particular beliefs and general beliefs become evident together once a wide
enough body of mutually supporting beliefs is in place. 39 Yet coherence theories
are normally taken to repudiate the category of the self-evident, while Lehrer's
Reid asserts that plenty of things are self-evident, general principles and
particular beliefs alike. What are we to make ofthis?
Suppose we say that one belief is epistemically prior to another (or that
beliefs in one class are epistemically prior to those in another) iff the latter
belief(s) derive their evidence from the evidence of the former, in such fashion
that the latter beliefs could not be evident unless the former beliefs were already
evident. Suppose we then say that a belief is self-evident iff it is evident, but
there is nothing epistemically prior to it. We could then say that a coherence
theory that rejects the very idea of epistemic priority is a theory that, rather than
repudiating the category of the self-evident, makes every evident belief self-
evident. Reid, while not going this far, could hold that particular beliefs and
general beliefs depend on each other for their evidence without beliefs of either
sort being prior to beliefs of the other sort; thus particulars and generals would
both be self-evident. This could be one way in which Reid, as Lehrer says,
transcends the dichotomy between foundational ism and coherentism. 40

1 Keith Lehrer, Thomas Reid (London: Routledge, 1989), p. 197. This book is cited hereafter as

2 Reid, p. 198.

) Reid, p. 157.
4 This abbreviates Essay 6, Chapter 7, of Reid's Essays on the Intellectual Powers of Man. My
page references are to the edition edited by Baruch Brody (Cambridge, Mass.: MIT Press, 1969).
5 Another assumption that would convert lA and the other A-style principles to principles of
evidence is a reliability theory of justification, but Lehrer and Lehrer's Reid both repudiate
reliability theories.
(, Reid, p. 157.
7 James Van Cleve, "Reid on the First Principles of Contingent Truths," Reid Studies, 3 (1999),


8 This abbreviates Chapter 6, Section 12, of Reid's Inquiry into the Human Mind. My page
references are to the volume edited by Derek R. Brookes (Edinburgh: Edinburgh University Press,
9 Reid, p. 198.
11 Lecture at the NEH Seminar on Thomas Reid, Brown University, August, 2000.

12 Keith Lehrer, "Chisholm, Reid, and the Problem ofthe Epistemic Surd," Philosophical Studies,

60 (1990), 39-45, at p. 40. This article is cited hereafter as "Surd."

13 A substitution instance of the generalization '(x)(Fx -> Gx)' would be the conditional 'Fa->

Ga'; a confirmation instance of it would be the conjunction 'Fa & Ga' (or perhaps an object that
is both F and G). What Reid and Lehrer mean by an instance is neither ofthese things, but simply
14 Ernest Sosa offers an analysis of "implicit commitments" along these lines in Ernest Sosa and

James Van Cleve, "Thomas Reid," in The Blackwell Guide to the Modern Philosophers from
Descartes to Nietzsche, edited by Steven M. Emmanuel (Oxford: Blackwell, 2001), pp. 179-200,
beginning on p. 190. For the more radical view that general belief is never anything over and
above such BFx -> BGx dispositions, see David Armstrong, Belief, Truth, and Knowledge
(Cambridge: Cambridge University Press, 1973), ch. 6.
15 Reid may be committed to rejecting this possibility. In his account ofthe system ofLeibniz, he

explicitly disapproves of Leibniz's distinction between perception and apperception. "As far as
we can discover, every operation of our mind is attended with consciousness, and particularly that
which we call the perception of external objects .... No man can perceive an object, without being
conscious that he perceives it" (EIP 2.15, pp. 236-37 in Brody). In another place (EIP 3.1, p.
325), he says that consciousness is always attended with belief of that whereof we are conscious.
Putting these together, it follows that we never perceive without believing that we perceive.
16 Stewart Cohen, "Basic Knowledge and the Problem of Easy Knowledge," forthcoming in

Philosophy and Phenomenological Research.

17 See William P. Alston, "Epistemic Circularity," Philosophy and Phenomenological Research,
47 (1986), 1-30.
18 As Ernest Sosa points out to me, there is a further ambiguity in the first disjunct: is it the de
dicto knowledge that all my faculties are reliable, or the de re knowledge, concerning each faculty,
that it is reliable?
19 Reid, p. 162.

20 Reid, pp. 144, 157, 162, and 187.

21 "Surd," p. 43.

22 "Surd," p. 42.

23 Keith Lehrer, "Reid, Hume, and Common Sense," Reid Studies, 2 (1998),15-25, on p. 15 and
24 Philip de Bary, "Thomas Reid's Metaprinciple," American Catholic Philosophical Quarterly,

74 (2000), 373-83. He reports in note 22 on p. 380 that in Reid's manuscript of the Intellectual
Powers at Aberdeen University Library, there are no commas surrounding "by which we
distinguish truth from error."
25 I thank Gideon Yaffe and Sue Cox for this suggestion.
26 I thank Alan Hazlett, Nick Treanor, and Ali Eslami for this suggestion.
27 I have in mind the twelfth paragraph of commentary on Principle 7. This paragraph may not be

decisive, however, since Reid says he is noting a property that Principle 7 has in common with
other principles, and it is just possible that the trust we repose in our senses, our memory, and our
reason may illustrate the implicit belief we have in principles other than Principle 7.
28 So that Principle 7 not convey less information than the preceding principles, I am assuming
that a reference to the faculties already mentioned is implicit-'the above-mentioned faculties, as
well as any others that I possess, are reliable'.

29 Lecture at the NEH Seminar on Thomas Reid, Brown University, August, 2000. Lehrer's actual

formulation was 'all first principles are trustworthy,' where being trustworthy implies both being
evident and being true. The fact that trustworthiness implies truth generates the worry about
ungroundedness that I raise in the text.
30 Ifwe render 'all our faculties are reliable' as 'all deliverances of our faculties are true', however,
we would obtain a principle that is self-subsuming provided it is itself a deliverance of our
31 Keith Lehrer, Self-Trust (Oxford: Clarendon, 1997), pp. 22-23.

32 The "first principles of contingent truths" are so called because they make knowledge of

contingent truths possible; this leaves it open whether they are themselves necessary or contingent.
It is clear from Reid, p. 157, however, that Lehrer takes them to be contingent. Moreover, to the
extent that Lehrer construes the principles as attributing trustworthiness and not just truth, he has
a principled reason for regarding them as contingent: he believes that epistemic properties never
supervene with metaphysical necessity on nonepistemic properties. On this point, see Chapter 3
of Self-Trust.
33 "Surd," p. 40, and "Reid, Hume, and Common Sense," pp. 22-23.

34 Here I may disagree with Lehrer, who says at p. 22 of "Reid, Hume, and Common Sense" that

the first principle about perception comes from the faculty of perception itself.
35 Perhaps a good Reidian name for it would be 'common sense', for Reid tells us that "the sole
province of common sense" is "to judge of things self-evident" (EIP 6.2, 567).
36 We see here, by the way, that the Reid who gets around the skeptical impasse by denying (2)
must also deny (1). Though he accepts KR, he denies that to have knowledge through K you must
have knowledge of K's reliability first. Otherwise, there could be no such thing as basic
knowledge of the reliability of a source. Rather, there are cases in which knowledge through K
and knowledge ofK's reliability arise simultaneously.
37 Lehrer has drawn on this passage for a somewhat different purpose. He poses a question about

Reid from Chisholm: "How can we tell that a belief is evident if not by appeal to a general
principle?" He cites the paragraph about light as Reid's answer, glossing it as "the evidence of
some beliefs is itself evident.... Ifthere are some beliefs whose evidence is evident to us, we have
no need for a criterion to pick them out as evident" ("Surd," p. 41).
38 "Surd," p. 40; see also Reid, pp. 199-200.

39 In the article cited in footnote 16, a coherence theory is Cohen's way around the impasse created

by (1) and (2). In effect, he denies both of them, holding that knowledge of the reliability of a
source and knowledge of the truth of its particular deliverances depend on each other without
either being prior to the other.
40 "Reid, Hume, and Common Sense," p. 16.
Chapter 11



G. J. Mattey
University a/California, Davis

Keith Lehrer's theory of knowledge has undergone considerable

transformation since the original version he presented in his 1974 book
Knowledge [2]. Among the original elements of the theory, belief has been
replaced by acceptance, subjective probability by reasonableness, the doxastic
system by the acceptance system, and beating competitors by answering
objections. New elements, such as the preference system and the reasoning
system, have been added. These changes have enhanced the depth and
plausibility of the theory.
A feature added in the first edition of Theory of Knowledge [3], the
"principle of the trustworthiness of acceptance," also known as "(T)", has by
contrast been treated by Lehrer in a way that arouses suspicion. The most recent
formulation appears in the second edition of Theory of Knowledge: "I am
trustworthy (worthy of my own trust) in what I accept with the objective of
accepting something just in case it is true" ([6], p. 138). Lehrer makes a case,
which will be examined below, that one's acceptance of (T) contributes to the
reasonableness of everything that one accepts.
By virtue of its form, if principle (T) is accepted with the objective of
accepting it just in case it is true, it applies to itself. Then, given its general
contribution to the reasonableness of what one accepts, accepting (T) contributes
to the reasonableness of accepting (T): "If a person accepts (T), then her
acceptance of (T) itself will have the result that it is reasonable for her to accept
(T)" ([6], p. 142). Lehrer regards such direct self-application of (T) to be both
natural and illuminating. He recognizes that it generates a circle, or "loop," but
he claims that the circularity is not vicious, because the loop is explanatory rather
than argumentative ([5], p. 136). This paper will examine the role of the principle
E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 173-194.
© 2003 Kluwer Academic Publishers.

of trustworthiness in making acceptance reasonable and the way in which it

might make itself reasonable.


Lehrer intends his theory of knowledge to provide an account of an

intellectual sort of know Iedge, one that presupposes a healthy degree of cognitive
sophistication. In particular, this kind of knowledge is more than the mere
possession of correct information, requiring in addition a recognition of the
information as being correct. "It is information that we recognize to be correct
that yields the characteristically human sort of knowledge that distinguishes us
as adult cognizers from machines, other animals, and even our infant selves"
([6], p. 7). Information recognized as correct "is inextricably woven into
reasoning, justification, confirmation, and refutation" ([6], p. 6).
A person who possesses correct information must, in order to have
knowledge of the type Lehrer is trying to analyze, take the information to be
correct. But recognizing information to be correct involves more than this. A
person may take information to be correct without any purpose. 1 Purposive
recognition of information as correct is what Lehrer calls "acceptance." It is the
taking of information to be true in order to satisfY some specific objective. This
requires evaluating how well the act of taking the information to be true furthers
the objective. Lehrer claims that such evaluation can take place without
reflection. "Positive evaluation may occur without reflection when reflection
would be otiose and would leave unchanged our intellectual and practical
attitudes concerning what we accept" ([4], p. 4. Cf. [6], p. 40.).
The kind of acceptance that can be knowledge of the sort to be captured
by Lehrer's theory is one based on an evaluation in terms of "the epistemic
purpose" of obtaining true information and rejecting false information ([6], p.
14). (We shall call this kind of acceptance "epistemic acceptance.") The purpose
is in general to maximize the possession of true information and minimize the
possession of false information. The most obvious way in which the evaluation
would occur is through reflection on whether acceptance helps to fulfill the
epistemic purpose. But since Lehrer claims that acceptance may not require
reflection, it appears that he needs to postulate a default mechanism for
acceptance in mundane matters so that reflection is called for only when use of
that mechanism is inappropriate. 2
If reflection is involved, there is a decision to be made by an epistemic
agent whether or not to accept a given piece of information as being true, in
order to fulfill the epistemic purpose. 3

When I consider accepting something, I have two options,

acceptance and non-acceptance. When I accept something, I have,
G. J. MATTEY 175

in effect, raised the question, to accept or not to accept, and

answered the question with a positive evaluation. ([4], p. 10)

The evaluative criterion governing epistemic acceptance is that of

"reasonableness." It can be more or less reasonable to accept epistemically a
given piece of information. The minimal degree of reasonableness required for
a positive evaluation would be that it be more reasonable to accept the
information as being true than not to accept the information. 4 Acceptance might
be anywhere from barely to massively more reasonable than withholding
acceptance. If an epistemic agent is to know that the information he accepts is
true, then the reasonableness of accepting as opposed to withholding should be
very high. Otherwise, the correctness of his decision to accept would be
fortuitous. 5
It is tempting to say that if the reasonableness of acceptance meets a
certain threshold level, then the acceptance is justified and thus meets a condition
for knowledge of the type under consideration. Lehrer realized from the
beginning that such a simple condition for justification is subject to the lottery
paradox. 6 To avoid this problem, he considered other pieces of information
whose acceptance would make the acceptance of a given piece of information
less reasonable. These competing pieces of information he now calls
"objections" to the information whose acceptance is at issue. Lehrer uses this
device to base his definition of justification on the notion of reasonableness
while avoiding the lottery paradox. A piece of information is subjectively or
"personally" justified just in case the agent has a way of dealing with all
objections. 7 If the information is also true and the acceptance of it is objectively
justified, it amounts to knowledge. 8
The idea that justification consists in the ability to deal with all objections
has a certain appeal, especially with respect to the kind of knowledge which is
the target of Lehrer's theory. Paradigmatically, knowledge is the outcome of
critical inquiry; it is what emerges, or at least would emerge, from the crucible
of intensive dialectical engagement with objections. 9 If an actual examination of
objections is required in each case of acceptance, the range of information that
is accepted, and therefore could count as knowledge would be severely limited.
On Lehrer's view such an examination is not required. The act of
epistemic acceptance does not require any reflection at all, so it does not require
that objections be taken into account. Ordinarily, the decision to accept is based
on positive evidence for the truth of the information, and objections are
considered only when there is some reason to think the information is false or
when one is being extra-cautious. 1o Note that from a practical standpoint,
consideration of myriad objections would thwart the goal of accepting as many
truths as possible. Even when reflection is called for in non-routine cases,
generally not all objections are taken into account when making a decision to
accept a piece of information as true to help fulfill the epistemic purpose.

Then the question arises as to how an acceptance can be justified, given

that all objections have to be dealt with. The answer is that one must have the
resources to deal with objections, whether or not one has taken them into account
in the evaluation leading to the act of acceptance. These resources make the
acceptance "reasonable," perhaps reasonable enough to count as knowledge.
This means that it can be asked post hoc to what extent an acceptance is
reasonable, where the answer may involve resources that were not drawn upon
in the act of acceptance.
So we need to make a distinction between the act of accepting and the
ongoing commitment to truth that can also be called acceptance. At one point in
time, acceptance is a mental act of committing to the truth of a piece of
information, in order to help fulfill the epistemic purpose. Reasonableness plays
the role of a criterion for making the commitment. At a later point in time,
acceptance is a commitment already made. As such, it is a candidate for
knowledge. The reasonableness of already-made acceptance might be
understood in terms of whether it is permissible to retain it or perhaps whether
the act of acceptance would be called for given the information one has at the
time. In discussing the reasonableness of acceptance, Lehrer draws on both of
these aspects without clearly indicating which one is in play.

What does it mean to say that it is reasonable, to some degree, for a person
to accept the information that p to fulfill the epistemic purpose of obtaining truth
and avoiding error? Lehrer treats reasonableness as a primitive notion, though
he does note a relation between reasonableness and the epistemic purpose. For
the information thatp to be reasonable (to some degree) to be accepted, it must
be subjectively probable to a certain degree, which promotes the goal of avoiding
error. Conversely, accepting information that p is made more reasonable as it is
more informative, which promotes the goal of obtaining truth ([6], p. 144-145).
Typical of somewhat risky, but highly informative, information are "major
scientific claims, those concerning galaxies, genes, and electrons" ([6], p. 145).
How reasonable it is to accept epistemically a piece of information would
seem on the face of it to be a complex matter, which is perhaps not easily
determined. Lehrer sidesteps this issue by simply assuming that "we are able to
tell, at least intuitively, when it is more reasonable to accept one thing than the
other" ([6], p. 128).11 This allows him to make reasonableness the determining
factor in any evaluation that results in the acceptance of the information that p
to fulfill the epistemic purpose. "I confront the question of whether or not to
accept some information that I receive," and I answer the question on the basis
of "how reasonable it is to accept the information in comparison to other
competing considerations" ([6], p. 126).
G. J. MATTEY 177

The sole source of reasonableness, on Lehrer's account, is the agent's

"evaluation system.,,12 Several components together make up the evaluation
system, of which one, the "acceptance system," is relevant to the present
discussion.13 This is the repository of information a person has already taken to
be true for the purposes of obtaining truth and avoiding error. As the evaluation
system is internal to the agent, no external factors contribute to the
reasonableness of acceptance.
Lehrer describes the role of the acceptance system in terms of its
contribution to the reasonableness of the act of accepting. "In deciding whether
to accept something or not at the present moment, reason requires the use of
relevant information I have accumulated in the quest for truth. That information
is contained in my acceptance system" ([6], p. 125).14 The evaluation system
enables the evaluation to take place by "informing" or "telling" the agent the
extent to which the information available to him is reasonable and how
reasonable the information under evaluation is relative to it ([6], p. 125).15
Lehrer's account of how the evaluation system makes acceptance reasonable
does not describe what makes an already-held acceptance reasonable. The
account can be applied in a couple of different ways to an acceptance one has
already made. The evaluation system might inform the agent about the
reasonableness of retaining the acceptance, or it might inform him about how
reasonable a fresh acceptance would be in light of the information he now has.
When the acceptance system makes it reasonable for a person to accept
some information p to fulfill the epistemic purpose, it can be described as
providing evidential support for the acceptance. Because Lehrer makes the
acceptance system the sole means by which support is conferred, the relation of
support is mutual or reciprocal. The accepted information p is supported by the
rest of the acceptance system. The reasonableness of accepting any information
q contained in the rest of the acceptance system is supported by the remainder
of the acceptance system, which includes the acceptance ofp. The mutual "fit"
of information within an acceptance system will henceforth be referred to as
"concurrence.,,16 Concurrence is not the same as what Lehrer calls "coherence."
Coherence is a relation that is defined in terms of the already-established
reasonableness of accepting that p in the face of objections to that acceptance,
and so it is a condition of justification rather than of reasonableness. 17
Concurrence and coherence are closely related in that both are based on the
evaluation system. So the acceptance ofp may be invoked to answer an objection
to the acceptance of q, and vice versa.
It is open to foundationalists to incorporate mutual support into their
systems. Chisholm allows that concurrence can add to the reasonableness of what
is in itself reasonable to some extent. 18 He illustrates the role of concurrence
using Meinong's analogy of cards tilted up against one another so as to provide
mutual support.

Each of the propositions in our concurrent set must be acceptable

on its own if we are to derive reasonability from concurrence, just
as each of the members of a house of cards must have its own
degree of substance and rigidity if the house is not to collapse. ([1],
p. 123)

In general, for foundationalists there are some acceptances whose reasonableness

can be accounted for, but which need no other acceptances to make them
reasonable. They might be made reasonable by themselves or by some external
As it has been described thus far, Lehrer's account of concurrence is non-
foundationalist. There appears to be nothing in it that can confer any degree of
"substance and rigidity" except for other acceptances. Lehrer, as is well-known,
rejects foundationalist accounts of justification. One of his central anti-
foundationalist arguments helps to flesh out the ways in which acceptances
support each other reciprocally. The justification of particular beliefs usually
rests on an appeal to general beliefs, e.g., those concerning how successful one
is in making judgments based on perceptual evidence (one's "track record").
Lehrer makes the case that such general beliefs are not basic but are justified by
particular beliefs about individual cases of success, and vice versa, which
involves "arguing in a circle" ([6], p. 93).

This suggests, contrary to the foundation theory, that the

justification of both kinds of statements may be reciprocal, that
each justifies the other as the result of cohering with a system of
beliefs containing particular beliefs about what we experience, as
well as general beliefs about our competence to discern truth from
error and the frequency of our success in so doing. To concede
this, however, is to give up the foundation theory and embrace the
coherence theory instead. ([6], pp. 93-4)

This account of coherence in justification can be straightforwardly extrapolated

to the way in which acceptances are made reasonable by other acceptances
through concurrence. What makes it reasonable for an epistemic agent to accept
that p is what he accepts about his competence and previous success, and it is
reasonable for the agent to accept this general information about himself because
of what he accepts about features of himself that make him competent and about
individual instances of success. There is no independent source of
reasonableness, nor is any acceptance reasonable in itself. This can be called a
"broad concurrence" account of reasonableness. In the balance of the paper, it
will be argued that this account is preferable to another account proposed by
Lehrer, one which is very suggestive offoundationalism.
G. J. MATTEY 179


The reasonableness of acceptance is said by Lehrer to depend on

acceptances about one's competence and record of success. It is convenient to
say that in that case, reasonableness depends on acceptance of one's
"trustworthiness." It is more difficult to say, however, what exactly
trustworthiness is for Lehrer. At times, he seems to equate it with competence in
accepting information successfully. An example is the acceptance that I see a
zebra. In order to be justified, the acceptance must be reasonable to some extent.
For it to be reasonable to accept, "I must have reason to think that I can tell a
zebra when I see one in circumstances like those I am in at the moment, and
consequently that I am trustworthy in such matters" ([ 6], p. 138). If! accept that
I am competent in evaluating information that I have accepted, I can be said to
accept that I am trustworthy in fulfilling the epistemic purpose in the present
case. Note that one need not actually be successful in the present case, or even
in a large number of cases, to qualify as being competent. So trustworthiness
need not be a function of "my current rate of success in obtaining truth and
avoiding error" ([6], p. 139). We may grant that someone is competent in
fulfilling the epistemic goal but has run into a streak of bad luck.
Even if competence does not require a successful track record, success
does have a role to play in making acceptance reasonable. Specifically, it
provides evidential support of the acceptance of one's competence.

The claim that I am trustworthy in any particular matter under any

special set of circumstances may be justified on the basis of the
other things that I accept; I accept that I have had success in
reaching the truth about similar matters in similar circumstances in
the past and that the present circumstances do not differ in any
relevant way from past circumstances when I was correct. ([6], p.

This approach might be generalized beyond particular cases to one's competence

in acceptance overall. A generalized view of one's own competence seems to be
what is codified in principle (1), which states in an unqualified way that I am
trustworthy in what I accept. Lehrer claims that this principle must be accepted
in order for any acceptance to be justified ([6], p. 138), and it plays a crucial role
in conferring reasonableness. If one did not accept that one is trustworthy in
general, then one would be unable to respond to an objection that casts doubt on
competence in accepting in general. And since the acceptance system is the basis
for responding to objections, its use would be indefensible. By extension, since
the acceptance system also supports the reasonableness of acceptance, it would

not be very reasonable to accept anything unless one accepted that one is
trustworthy in what one accepts in general.
On the interpretation of trustworthiness as competence, the principle
means that in accepting what I do in general, I exercise competence in fulfilling
the epistemic purpose of acceptance. In that case, the reasonableness of principle
(T) would be supported by acceptances about one's overall record of one's
success in everything one accepts. It is reasonable to accept that I am generally
worthy of my own trust to the extent that I accept that I have earned that trust,
so to speak.
Lehrer has a second way of understanding trustworthiness: as a
deontological notion, "an irreducible element of epistemic value" ([5], p. 138).
He describes it as "a notion of what is worth accepting and what methods are
worth using" ([5], p. 138). In his account of the normative dimension of
trustworthiness, he divorces it entirely from considerations of actual competence
and success. His purpose in so doing is to accommodate the intuition that it is
reasonable to accept what one does even if one is the victim of massive
deception. Though Lehrer does not make this point, it is clear that such an agent
would then be completely incompetent, not merely unsuccessful, in fulfilling the
epistemic purpose. Since, reasonableness requires acceptance of trustworthiness,
Lehrer wants to say that a victim of deception may nonetheless be trustworthy.
"I am worthy of my trust in what I accept though I am deceived. I am as
trustworthy as the circumstances allow" ([6], p. 140).
If worthiness of one's trust in acceptance does not require actual
competence in fulfilling the epistemic purpose, what does it require? Lehrer
casts himself in the role of a hypothetical demon-victim and describes himself
as being deceived "through no fault of my own" ([6], p. 139). Being worthy of
one's own trust, on this deontological construal, is a matter of having followed
certain standards in searching for the truth. As Lehrer puts it regarding the
demon case, "I seek to obtain truth and avoid error with the greatest intellectual
integrity" ([6], p. 140). Similarly, one is trustworthy when one is "circumspect
and seeks to detect every error" ([6], p. 192).
Trustworthiness, viewed deontologically, is the result of the use of a
general method of approaching acceptance, in the exercise of which one takes on
objections forthrightly, meeting them when one can and changing one's view
when one must. It also requires the willingness to change one's methods of
getting at the truth if need be. In general, Lehrer says that his trustworthiness
"rests on a dynamic process of evaluation and amalgamation of information I
receive from others and from my own experience" ([6], p. 140).
Lehrer's example of trustworthy but unreliable acceptance is described in
a way that suggests that the agent has, in general, done her best to fulfill the
epistemic purpose. But it is psychologically unrealistic to assume that there is
a constant level of circumspection applied in every act of acceptance, so in
general, one's acceptances will fall somewhat short of this standard. If
G. J. MATTEY 181

acceptance is restricted to cases of taking information to be true in which one has

done one's utmost to avoid error, then there is little, if anything, that people
accept. We simply do not go through exercises like Descartes' Meditations in our
ordinary lives.
It would seem that trustworthiness in practice requires a lower standard of
circumspection. Moreover, it would be extraordinary if anyone applied a single
standard consistently. Given that this is the case, it is best to look at a range of
degrees of circumspection, beginning with some point at which one is, so to
speak, "circumspect enough" in trying to fulfill the epistemic purpose. To put
it another way, one's methods for arriving at the truth are good enough as means
to fulfill the epistemic purpose.
Competence and methodological circumspection should be closely related
in a plausible account of reasonableness. If an epistemic agent accepts that his
methods for fulfilling the epistemic purpose are good enough, this seems to
imply that he accepts that he is competent in accepting what he does. A method
is not adequate for the fulfillment of a purpose unless it confers competence on
the agent exercising it. If the virtue of the method is circumspection, then
circumspection should not be divorced from competence. A good-enough
method, then, is one which involves both the normative element of
circumspection and the descriptive element of competence.
To get a feel for why this is so, suppose the demon victim accepts that she
has done her very best to fulfill the epistemic purpose. Should she, on that basis
alone, accept that she is trustworthy? It would seem not, but rather that she
should also accept that her most circumspect efforts are the sort of thing that will
help her fulfill the epistemic goal: in short, she needs to accept that she is
competent in accepting what she does. One must not isolate the acceptance of
one's trustworthiness from one's other acceptances. 19
If trustworthiness requires competence as well as circumspection, we
should concede, pace Lehrer, that the victim is not trustworthy but only falsely
accepts that she is. This seems a superior way to handle the case, in that it
accords more closely with our ordinary notion of trustworthiness. All Lehrer
really needs to say is that the victim is epistemically blameless in a way that can
allow her acceptance to be reasonable, if not justified. And this can be handled
if it is allowed that she reasonably, albeit falsely, accepts that she is trustworthy
in what she accepts.
There is one further complication in understanding principle (1). The
principle is simple enough on the surface, stating that I am trustworthy in what
I accept to fulfill the epistemic purpose. But Lehrer considers it as "a statement
ofa capacity and disposition to be trustworthy" ([6], p. 139). This qualification
is due to the fact that one may fail to follow good-enough procedures in specific
cases of acceptance, though one is generally disposed to follow them. In what
follows, then, we shall take it that it is the disposition to be trustworthy that is
supposed to account for the reasonableness of acceptance.

How does it do so? Our original account of reasonableness was based on

the "broad concurrence" approach, which regards the reasonableness of any
acceptance to be a function of the reasonableness of all other acceptances. It can
now be expanded to incorporate the elements that contribute to trustworthiness.
A crucial class of acceptances in this regard is that of acceptances about our
competence to accept all and only what is true in specific areas of investigation
and in general. These acceptances about our competence are supported by
acceptances about our success in fulfilling the epistemic purpose, and these in
turn are supported by our particular acceptances.
Another crucial class of acceptances is that of acceptances about our
integrity and circumspection in accepting what we do in specific areas of
investigation and in general. Such acceptances will be supported by observation
of the way in which we go about accepting what we do, as well as acceptances
about what constitutes the best means to fulfill the epistemic goal. Most
importantly, they will be based on what we accept about the way we respond to
objections and to new information.
Principle (T) should be taken as summarizing these acceptances about
many facets of the acceptance system. Trustworthiness helps to make other
acceptances reasonable only because of its concurrence with many elements of
the acceptance system. Lehrer does not, however, always describe the relation
between principle (T) and reasonableness in terms of "broad concurrence."
Instead, he relies mainly on what he calls the "trustworthiness argument" to
make the connection in a way that appears to be more direct. This, it will be seen,
leads Lehrer in the direction of foundationalism.


The "trustworthiness argument" consists of two inferences. The first

consists of two premises and a conclusion, and the second is a direct inference
from the first conclusion. The first premise asserts my trustworthiness and the
second my acceptance of some information as true. It runs verbatim as follows. 20

T. I am trustworthy (worthy of my own trust) in what I accept with

the objective of accepting something just in case it is true.
I accept that p with the objective of accepting that p just in case it
is true.
Therefore, I am trustworthy in accepting that p with the objective
of accepting that p just in case it is true.
Therefore, I am reasonable in accepting that p with the objective of
accepting thatp just in case it is true. ([6], p. 139)
G. J. MATTEY 183

Since there is no restriction on the value ofp, the conclusion must be taken to be
generalizable to all acceptances.
Lehrer notes that the first inference is meant not to be deductive, but rather
inductive. That is, the first premise is not intended to be a universal
generalization "to the effect that I am always trustworthy in what I accept" ([6],
p. 139). Instead, it is supposed to be taken as a claim to the effect that I am
generally trustworthy in what I accept.

It is like the inference from the premise that my lawyer is

trustworthy to the conclusion that he is trustworthy in the way he
has constructed my will or from the premise that a city water
supply is trustworthy to the conclusion that the water supplied in
my glass is trustworthy. ([6], p. 139)

This is why he understands the principle of trustworthiness to be about a capacity

or disposition. His lawyer may be disposed to act in a trustworthy way but fail
to do so, perhaps due to weakness of will. Similarly, epistemic agents can be
subject to "doxastic akrasia" ([6], p. 142).
The second inference is an enthymeme. It depends on the conditional: if
I am trustworthy in accepting that p with the objective of accepting that p just in
case it is true, then I am reasonable in accepting that p with the objective of
accepting that p just in case it is true. In [4], a variant of the implicit conditional
is made explicit: "If I am reasonable to trust my acceptance of p, then I am
reasonable to accept thatp" ([4], p. 7).
Having elucidated the structure of the argument, we may now examine its
premises. As already noted, the first premise is not to be taken as strictly
universal, but only as a description of a "capacity and disposition to be
trustworthy." In terms of the way Lehrer understands trustworthiness, to say that
I am generally trustworthy is to say that in accepting what I accept, I generally,
though perhaps not always, proceed according to good-enough methods. In the
same way, a city's water supply might be generally trustworthy because its
operators generally follow, well-enough, standard methods to keep the water
safe, even though they may fail to follow those methods from time to time.
The second premise, I accept that p, is ambiguous and could describe the
act of accepting that p or the fact that p is already being accepted. The most
plausible reading is that it describes what is already accepted. Otherwise, the
argument would be limited in its scope to what is presently being accepted.
Lehrer's goal in advancing the argument is clearly to provide support for the
reasonableness of everything that a person has accepted. Moreover, the argument
itself can apply only to what a person has already accepted, not to a person's act
of accepting that p. If p has not yet been accepted, then the second premise is

The premise might be taken as describing an act of acceptance because of

a remark Lehrer makes in the second edition of Theory ofKnowledge, just after
introducing principle (1) and before giving the "trustworthiness argument."

If someone else accepts that I am trustworthy in this way, then my

accepting something will be a reason for her to accept it. Similarly,
if I accept that I am trustworthy in this way, then my accepting
something will be a reason for my accepting it. ([6], p. 138)

The most plausible construal of the description of the other person's acceptance
of my trustworthiness is that she uses it, along with the fact that I accept some
information, as a factor in evaluating that information and making a decision to
accept it. But in that case, the analogy breaks down, since I cannot make what
I already accept a factor in my deciding to accept it. My acceptance of my own
trustworthiness can playa role in my deciding what to accept, in that without it,
I might be disposed to withhold judgment rather than accept any information at
all. It will be argued below that this generic way in which trustworthiness makes
accepting reasonable is also the only way in which it makes what is already
accepted reasonable.
Given the interpretations ofthe two premises, the first conclusion must be
read in this way: I have accepted that p on the basis of good-enough methods for
obtaining truth and avoiding error in the acceptance of p. In the context of the
trustworthiness argument, what those methods are is immaterial. Since I am
trustworthy, the fact that I accept that p means that I have (most likely) relied on
those methods I deem fit to make my acceptance a correct one.
Now we can see how the first conclusion supports the second one: why
trustworthiness entails reasonableness. It has been noted that Lehrer assumes
that one can tell how reasonable it is to accept a given piece of information,
relative to one's evaluation system. Presumably this means that one can
determine how suitable its acceptance is to advance the epistemic purpose. Then
the idea would be that if I use the methods I deem to be good enough for
fulfilling the epistemic purpose in accepting that p, then I should regard the
purpose as being fulfilled. Lehrer states the relation this way: "My
trustworthiness serves the objectives of reason, and if I am trustworthy in the
way I serve the objectives of reason in what I accept, then I am reasonable to
accept what I do" ([5], p. 136).
So the thrust of the whole argument is this. IfI am disposed generally to
use good-enough methods in accepting what I do, and I accept some piece of
information p, then I can conclude inductively that I have used good-enough
methods in accepting that p. If I have used good-enough methods for accepting
that p, then my acceptance that p is reasonable, to some degree.
Ordinarily, when we evaluate the reasonableness of acceptance, we take
into account the specific methods which are applicable to the specific
G. J. MATTEY 185

information in question. The statement "I am reasonable in accepting that p" can
be understood in two very different ways. It can be, and ordinarily is, read as a
statement about the reasonableness of accepting the specific information p. Or
it can be read as a statement about accepting any information at all, regardless
of its specific content. It is only the latter sense that could possibly be
established by the "trustworthiness argument." It is only in this sense that Lehrer
could be entitled to assert that, "A consequence of adding principle (T) to my
evaluation system is that I may reason from it and the acceptance of some target
acceptance that p to the conclusion that the target acceptance is reasonable" ([6],
p. 139).
A more specific counterpart to the generic "trustworthiness argument"
might be one to the effect that one is disposed to be circumspect one's
investigations and that those methods of investigation sanction the acceptance
of the specific information p, so that it is reasonable to make a commitment to
the truth of p. One would expect that the first premise would be established
inductively. The second premise would be established by appeal to the specific
evidence in favor of accepting that p. The original "trustworthiness argument,"
on the other hand, says nothing about what makes it reasonable specifically to
accept that p rather than some other information. So whatever degree of
reasonableness it establishes is minimal compared to that established by the
counterpart argument.
Suppose an ordinary person were to ask me why it is reasonable for me to
accept that the water in my glass is safe to drink. If I were to respond, "Because
it is something that I accept, and I use good-enough methods to accept what I
do," my response would most likely be met with bewilderment. On the other
hand, if a foundationalist like Chisholm were to ask this question in the context
of his epistemological investigations, the answer would make sense, since he
then would be concerned with the source of reasonableness as such.
The "trustworthiness argument" is really appropriate only in the context
in which Lehrer raises it, i.e. as a response to a foundationalist objection.
Justification depends on the reasonableness of what one accepts, and
reasonableness depends on the acceptance of trustworthiness. "The
foundationalist will surely note that everything now depends on the claim that
my acceptance is a trustworthy guide to truth and that I am trustworthy, as I aver.
She will inquire how that claim is itself justified" ([6], p. 138).
The foundationalist inquiry can be extended to the issue of reasonableness:
what makes it reasonable for me to accept that I am trustworthy? What gives that
acceptance what Chisholm called "substance and rigidity?" Though appeal may
be made to competence and success, the following response has to be given:
"when I accept something, that is a good enough reason for thinking it to be true,
so that it is reasonable for me to accept it" ([6], p. 138). Again, "If! accept that
I am trustworthy in this way, then my accepting something will be a reason for
me to accept it" ([6], p. 138).

This generic approach has the advantage of being able to confer

reasonableness in one fell swoop, rather than requiring that each acceptance be
shown to be reasonable on its own. In that case, it looks as though Lehrer is
making a concession to the foundationalist by not resting with the "broad
concurrence" approach outlined above. That is, he is singling out one particular
acceptance as supporting the reasonableness of all the others. Moreover, he
holds that principle (1) makes itself reasonable, since it applies to itself if it is
accepted. This is structurally akin to self-justifYing acceptances in a
foundationalist theory of justification, where there is a narrow, rather than a
broad, circularity. The self-application of principle (1) draws on the
foundationalist model of self-justifYing acceptances, such as "I accept
something," which exploits the self-referential character of what is accepted. 2 !
Lehrer acknowledges a measure of foundationalism in his account of
justification: "To be personally justified one must accept some principle of
trustworthiness that is in part self-justified" ([6], p. 202). This is on the grounds
that, "Part, but not all, of what makes us personally justified in accepting that we
are trustworthy is that we do accept that we are" ([6], p. 202). Reasonableness
is treated in a parallel way, though not explicitly. The question that remains is
whether this foundationalist turn is well motivated.



Lehrer offers two accounts of what makes principle (1) reasonable, both
of which require that it contribute in some way to its own reasonableness. On one
account, the contribution is indirect; on the other it is direct. The first account
will not be plausible to someone who rejects all circularity in the relation of
evidential support. The second account is burdened with its own variety of
circularity and has additional problems of its own.
In the first account, the claim that one is generally disposed to be
trustworthy in acceptance is supported by an inductive generalization. The
starting-point is the trustworthiness of most of one's specific acceptances and the
conclusion is that one is generally disposed to be trustworthy in acceptance.

What defense should [a person] offer in favor of (1) itself? She

may, of course, appeal to the character of what she accepts, to the
various things she accepts, and reason inductively from premises
concerning the trustworthiness of individual acceptances in support
of her conclusion that (1). She might reflect on what she has
accepted and her fine track record of mostly accepting what was
worthy of her trust to accept. This argument would establish that
G. J. MATTEY 187

the trustworthiness of her acceptances manifests her disposition to

be trustworthy in what she accepts. ([6], p. 142)

The reasonableness of principle (7) in that case depends on the reasonableness

of the acceptances that comprise the information invoked in the defense. For
example, it must be reasonable, to some degree, for the person to accept in any
given case that she has accepted what she has in a trustworthy way.
What makes these acceptances reasonable to the degree they are will have
to involve the acceptance of principle (7), as was noted above. Lehrer claims
that it is the "trustworthiness argument" that connects (7) with the more specific
acceptances. So the principle makes an indirect contribution to its own
reasonableness, engendering a circle.

There is obviously a circularity in the trustworthiness argument

when we use the principle (7) as a premise to support the
conclusion that other acceptances are reasonable and then use those
acceptances and the principle itself to conclude that it is reasonable
to accept it. ([6], p. 143)

The circularity to which Lehrer refers is a version of "broad concurrence," with

principle (7) playing a crucial role by conferring on everything one accepts the
reasonableness it has.
Lehrer recognizes that principle (7) cannot be used to defend its own
reasonableness in the face of a skeptical objection.

But to explain why it is reasonable to accept what we do, the circle

may be virtuous. If we have a principle that explains why it is
reasonable to accept what we do, it is a virtue rather than a vice that
it should at the same time explain why it is reasonable to accept the
principle itself. ([6], p. 143)

The crucial difference between responding to a skeptical objection and giving an

explanation is that in the latter case, the datum is taken for granted. So we
assume that it is reasonable to accept principle (7) and ask why this is the case.
Since it is not meant as a response to a skeptic, at most it shows to a person
already committed to the reasonableness of what he accepts what it is that is
supposed to make those acceptances reasonable. It does so by appeal to a notion
of mutual support that many would find suspect.
Perhaps the appeal to mutual support can be avoided if the reasonableness
of principle (7) is explained by a direct application of (7) to itself. "If a person
accepts (7), then her acceptance of (7) itself will have the result that it is
reasonable for her to accept (7) by the application of the trustworthiness
argument to (7) itself as the target acceptance (P)" ([6], p. 142). He takes this

direct self-application to be "natural," apparently since (T) is applicable to all

other acceptances ([6], pp. 142_3).22
But the direct self-application of (T) in the "trustworthiness argument"
once more opens up the issue of circularity, where the circle is now as small as
it can be. As Lehrer describes them, foundationalist theories of justification
appeal to self-justifying acceptances. 23 These acceptances are said by the
foundationalists to guarantee their own truth. The use of principle (T) to explain
its own reasonableness appears to allow a similar sort of "bootstrapping"
operation. But the circularity here is different, since the acceptance of one's
trustworthiness does not in any way guarantee the truth of that acceptance in the
way that the acceptance of one's existence guarantees that one exists.
In his 1999 article "Knowledge, Scepticism and Coherence," Lehrer gives
the following account of the explanatory role played by principle (T).

I accept that I am trustworthy in what I accept, and if I am

trustworthy in what I accept, then I am reasonable in accepting that
I am trustworthy in what I accept. My trustworthiness in what I
accept explains why I am reasonable in accepting that I am
trustworthy in what I accept. ([5], p. 136)

He opts for this small circle over the larger circle because it allows him to
avoid a regress in explanation. "I could argue for my trustworthiness by
consideration of other things I accept and my success in attaining truth, but that
way a regress threatens, whatever the merits of such arguments in supporting the
principle" ([5], p. 136). Butthere is no regress when the other things one accepts
are made reasonable in part by the acceptance of one's trustworthiness, so we are
not forced to apply the principle to itself to account for the reasonableness of
acceptance. Whether the small circle actually explains anything remains to be
In the second edition of Theory ofKnowledge, published in 2000, Lehrer
does not mention the regress argument, and as seen above, he explains the
reasonableness of accepting (T) by appeal to its concurrence with the whole
acceptance system. Still, though Lehrer does not defend the use of principle (T)
to explain its own reasonableness without appeal to any other information, he
does endorse the direct application of (T) to itself. What reason is there for doing
this, except as a formal exercise? Lehrer notes that the argument is "more direct"
than one using an inductive argument from individual cases of trustworthy
acceptance ([6], p. 142). This directness has the advantage of economy, but it
gains this advantage at the expense of content.
It is useful to note that Lehrer recognizes that he is not making the
stronger claim that principle (T) is completely self-justifying, though "the
principle of our own trustworthiness contributes to its own personaljustification"
([6], p. 202). It does not justify itself fully because it "must cohere with what we
G. J. MATTEY 189

accept about our successes and failures in past epistemic employment" ([6], p.
202). In that case, one must ask why Lehrer restricts this requirement to
justification. Can we plausibly say that it is reasonable to accept a piece of
information without regard to whether it coheres (or "concurs") with information
we have about our past record of success, among other things? The fact that
reasonableness (to some unspecified degree) need not meet the standard of
justification does not exempt it from the need for a comprehensive base of
This consideration raises the more general question of what kind of
explanation could be provided by the direct self-application of principle (1). I
want to know why it is reasonable for me to accept that I am trustworthy in what
I accept. In terms of the interpretation of trustworthiness developed thus far, the
question is why it is reasonable for me to accept that I have a disposition to use
good-enough methods in accepting what I accept. The obvious indirect
explanation for this is on the basis of what I accept about how I have used good-
enough methods in the past. The indirect loop is generated by adding that part
of what makes those acceptances reasonable is the acceptance of my own
The direct explanation is simply a vastly diminished version of the indirect
explanation, one which omits all the details that enlighten me as to why it is
reasonable to accept that I am trustworthy. The result is hardly edifYing.
Suppose I were testifYing to a jury and averred that I am trustworthy in my
testimony. When asked to explain why it is reasonable to accept that I am
trustworthy, I answer that my original statement that I am a trustworthy witness
is what explains why it is reasonable for me to accept that I am. Would I have
explained, to anyone's satisfaction, anything at all?
Another way to put the point is by noting what kind of reasonableness is
supposed to be explained. As was noted above, we can ask, for any given piece
of information (P), whether it is reasonable to accept that (P) as having a given
content, or whether it is reasonable to accept that (P) in the sense that it is
reasonable to accept what we accept in general. And as has been argued earlier,
the kind of reasonableness established by the "trustworthiness argument" is
generic. So when principle (1) is the target acceptance, the most the application
of the trustworthiness argument to (1) can explain is that it is reasonable for me
to accept that (1) insofar as it is reasonable for me to accept anything at all. So
the direct self-application of principle (1) does nothing to explain why it is
reasonable to accept information with the specific content that I am trustworthy
in what I accept.
Lehrer might respond to this description ofthe thinness of the explanation
by claiming that any explanation is better than none. The principle itself is part
of the account of the reasonableness of any acceptance, and so if it were not
reasonable to accept principle (1), there would be no explanation at all. In that
case, principle (1) "should be a kind of unexplained explainer that explains why

it is reasonable for us to accept other things we accept and then falls

mysteriously silent when asked why it is reasonable to accept the principle itself'
([6], pp. 143-4). Lehrer states that he seeks to maximize explanation and leave
nothing unexplained ([5], p. 137). If (1) does not explain itself, then it is a "kind
of explanatory surd" ([5], pp. 136-7). Preference for maximizing explanation
and avoiding the surd is "one I act upon in developing my philosophy" ([5], p.
The surd can, however, be avoided with the broad account that explains
specifically why it is reasonable to accept that I am trustworthy. It can also be
said that this account provides a vastly more thorough explanation, and so it
helps to maximize explanation. It explains something that would be left
unexplained by the mere direct self-application of (1), namely, why my
acceptance of the specific information that I am trustworthy is reasonable.
As stated above, Lehrer's version of the broad account of reasonableness
places principle (1) in the key role of explaining reasonableness on all
acceptances. In view of the present discussion, it seems that this role is not as
crucial as it first appeared. The principle can only explain the reasonableness of
acceptances qua acceptances. The bulk, so to speak, of their reasonableness is
explained by the specific concurring information that supports them. And as
argued above, principle (1) itself is a summary of a complex of information
about the methods one uses in fulfilling the epistemic purpose. All, or nearly all,
of the reasonableness of accepting principle (1) itself stems from specific
concurring information. So while it might be granted that (1) plays an important
role in conferring reasonableness, that role is not foundational.
A final consideration Lehrer advances in favor of the direct self-
application of principle (1) is an appeal to analogies. The first of these is due to
Reid: "just as light, in revealing the illuminated object, at the same time reveals
itself, so the principle, in rendering the acceptance of other things reasonable, at
the same time renders the acceptance of itself reasonable." ([6], p. 143). Notice
that, in Reid's image, light "illuminates" other objects but "reveals" itself: it
makes both other objects and itself visible. To make the analogy work, light
would have to make itself visible in the same way that it makes the other objects
visible. But it does not make itself visible by illuminating itself in the way it
illuminates other objects. So this is not a good way of illustrating how the
application of principle (1) explains its own trustworthiness.
The second analogy is that of a keystone.

The keystone is a triangular stone inserted in the top of an arch. It

supports the arch, for the arch would collapse were it removed; at
the same time, it is, of course, supported by the other stones in the
arch. We may think of the stones in the arch as the acceptances in
the acceptance system and the principle (1) as the keystone. ([6],
p. 143)
G. J. MATTEY 191

Lehrer might also have noted that the keystone would fall to the ground if the
other stones were removed. The keystone supports itself only through its support
of the other stones. A keystone is not a foundation stone. So this analogy, if it
has any value at all, favors an account of the reasonableness of accepting the
principle of trustworthiness in which the principle supports itself indirectly. In
general, it favors the "broad concurrence" account of the reasonableness of
In summary, there seems to be nothing favoring the direct application of
(T) to itself other than the fact that it can be made and that it simplifies the
response to a skeptical objection to the reasonableness of what one accepts. But
in fact it is no response to a skeptic, and if it explains anything at all, it explains
only how it is reasonable to accept to some extent, merely as something that is
accepted in general. Even this explanation is largely incomplete and can only be
completed by appeal to a large number of other acceptances. Finally, there is no
explanation of why the specific content of (7) is reasonable to accept.
All of these deficiencies disappear when "broad concurrence" is invoked
to explain what makes principle (T) reasonable to accept epistemically.24 So the
direct self-application of (T) appears to be a useless exercise. It might even do
some harm by engendering the illusion that the principle of trustworthiness is
foundational rather than a "first among equals." Given the argument of this
paper, Lehrer has no reason to make his "ecumenical" concession to
foundationalism ([6], pp. 201-3).


Lehrer's doctrine that reasonableness is based solely on acceptance leaves

him open to a charge of broad circularity, a charge avoided by foundationalist
accounts of reasonableness. It is only through a relation of mutual support that
acceptances can make one another reasonable. Lehrer singles out a special
acceptance, that I am trustworthy in what I accept, as playing a key role in
providing that support. It has been argued here that acceptance of the principle
makes the mere acceptance of a piece of information, including itself, reasonable
to some extent, though in an entirely generic way. It does so in the context ofthe
acceptance system as a whole.
The principle of trustworthiness might also make itself reasonable by
applying to itself directly, in which case it seems to be foundational and
potentially to avoid the problem of broad circularity. But this direct application
is narrowly circular and so holds no advantage in this respect over the indirect
application. Because a direct application explains nothing that is not explained
by the indirect approach, and indeed omits what ought to be included in any
explanation of reasonableness, there is no reason to concede anything to the
foundationalist. The essential ingredients in the explanation of reasonableness

are to be found in the acceptance system as a whole, as is consonant with

Lehrer's coherence approach to justification. The narrowly circular application
of the principle of trustworthiness to itself is an aberration.*

*1 am grateful to Lenny Clapp for several excellent suggestions for

improving the presentation in this paper.


1 This seems to be what Lehrer calls mere belief. which "arise[s] in us naturally without our
bidding and often against our will" ([6], p. 40).
2 Lehrer acknowledges in his "bear print" example that sometimes circumstances dictate a greater
degree of scrutiny than in normal circumstances ([6], p. 73). Complex or highly general
information would also call for reflection.
] Ifreflection is not involved, there is no "decision" strictly speaking, but a commitment must be
made in a manner analogous to the making of a decision.
4 This kind of comparison was made by Chisholm in [1].

5 It would be fortuitous from the point of view ofthe agent. There might be some sort of external
factor that makes the correctness of the decision non-accidental, as in the case of a device
implanted in the brain that brings about correct acceptances. See [6], p. 186-8.
6 See [2], pp. 192-197.

7 Objections must either be "answered" or "neutralized" ([6], pp. 134-136).

8 Objective justification is described in Chapter 7 of [6].

9 The "justification game" illustrates the way in which critical objections might be handled ifthey
were to arise. See [6], pp. 132-128.
]0 This feature of acceptance is highlighted by theories of prima facie justification.

11 Presumably, one would be able to tell as well whether it is more reasonable to accept than to
withhold acceptance.
12 This is less evident with respect to informativeness than with respect to probability, but very

general information can be completely uninformative to a person who does not have the
conceptual resources to integrate it into his view of the world.
13 The other components are the "preference system" and the "reasoning system." See [6], pp.
126-127. In [5], Lehrer notes that only the acceptance system is relevant to the issues that he raises
there, and these are the issues discussed in the present paper. See p. 138, note 2.
14 It appears to be psychologically unrealistic to assume that an epistemic agent can properly

distinguish between what he already believes and what he has already accepted. One reason is that
we often forget how we came to take a given piece of information to be true.
15 The information contained in the acceptance system is also the basis for the determination of
16 This term is taken from Chisholm, who attributes the concept to the ancient academic skeptic
Carneades ([1], p. 43).
17 It may be that in many, most, or all cases, concurrence involves dealing with objections, in
which case it is closely related to justification.
18 See [1], Chapter 3.
19 Since following good-enough methods requires evaluation of our own competence, we will

hereinafter describe trustworthiness in terms of methods only.

20 Similar versions appear in [4], pp. 6-7 (called "the acceptance argument" there) and [5], p. 136.
G. J. MATTEY 193

21 Lehrer allows that it is plausible that "I believe something" is a self-justified belief. See [6], p.
54. See also p. 67-8, where he writes that "fallibility infects almost all our beliefs" (emphasis
22 In the first edition of Theory of Knowledge, Lehrer had called the self-application of (T) "more
natural" than trying to avoid the self-application for fear of self-referential paradox (p. 123).
23 See [6], Chapter 3.

24 This is not to say that these advantages make "broad concurrence" a convincing alternative to

Roderick Chisholm. Theory o/Knowledge. Prentice Hall, 1966.
Keith Lehrer. Knowledge. Oxford University Press, 1974.
Keith Lehrer. A Theory 0/ Knowledge. Westview Press, first edition, 1990.
Keith Lehrer. Self Trust: A Study o/Reason, Knowledge and Autonomy. Oxford University Press.
Keith Lehrer. Knowledge, scepticism, and coherence. Philosophical Perspectives, 13: 131-139,
Keith Lehrer. Theory o/Knowledge. Westview Press, 2000.
Chapter 12



Richard N. Manning
Carleton University

We can take it as given that at least one ineliminable goal of epistemic

justification is the acquisition of true beliefs. From this it follows that no
account of justification which fails to show that justification, as conceived in
that account, conduces to truth, can properly be termed epistemic. Call this issue
arising from the basic idea of what epistemic practice is all about 'the truth
conduciveness problem'. In this paper I will discuss Keith Lehrer's approach
to the truth conduciveness problem, in the context of his coherence theory of
knowledge as undefeated justification. Lehrer's coherentist approach centrally
involves an appeal to our own self-trust, which self-trust is itself purportedly
warranted in part by appeal to itself. I will, in due course, argue that Lehrer's
attempt to solve the truth conduciveness problem by appeal in this way to self-
trust fails, leading to logical circles, abysses and blind alleys. But this failure is
highly instructive. Self-trust is indeed crucial, not just to coherentist
epistemology, but to epistemic practice as such. For this reason, self-trust I will
suggest, neither can or need be argued for at all. Lehrer's strategy fails, then,
not by placing too much reliance on self-trust, but by seeing self-trust as a
candidate for justification.
The truth conduciveness problem is, in its most general form, a problem
for all theories of epistemic justification. The structural strategy for solving this
problem invoked by foundational accounts of justification, in both their
traditional and externalist forms, is clear enough: a claim is justified via a chain
of epistemically acceptable inferential relations ultimately terminating in some
claim or claims with a privileged epistemic status. Such claims, in virtue of their
content or relations to the world, are prima facie likely to be true, and hence do
E.l. Olsson (ed.), The Epistemology of Keith Lehrer, 195-216.
© 2003 Kluwer Academic Publishers.

not require justificatory support from other claims. That is, according to
foundationalism, there are basic claims which are non-inferentially justified.
To suppose that there are such non-inferentially justified claims is not, of
course, to suppose that agents do or even could have some good reasons for
believing of any particular claim, or any class of claims, that it is so justified.
Indeed, the foundational status of a belief means that no such belief could owe
its justificatory status to anything the agent possesses as a reason. To suppose
otherwise would reinstate the impotent regress of justification that
foundationalism is in part designed to avoid. On the foundationalist picture,
then, not every belief that isjustified need be justified by an agent's reasons. As
William Alston notes, all that is needed to stop the regress of justification is
belief which is immediately justified, whether or not the agent has reason to
believe that it is so justified. l Faced with the natural objection that this would
mean that ajustificatory chain could terminate in a claim which the agent has no
reason to hold true, that from the agent's point of view this claim would amount
to a brute assumption, Alston argues that there is no bar to adding the additional
requirement that the agent have a justification for supposing the foundational
claim immediately justified, but that this justification need not (and cannot)
itself be immediate, but must be inferential. As it is a reason for holding the
putative basic claim justified, it is moreover an epistemic reason in the sense of
being a second order claim about the justificatory status of a first order claim.
(More generally, a claim of order n about the justificatory status of some claim
or claims oflevels ml, m2, ... mi<n.) But the inferential grounds an agent has or
needs for supposing a claim immediately justified-his epistemic reasons-
cannot form an essential part of the justification of the claim, lest they
compromise its foundational status. Thus the epistemic reasons a person might
have for holding a belief are on this view distinct from the justification of that
belief, which is non-inferential.
This divorce of justification and reason is precisely what the
epistemological coherentist rejects. For the coherentist, an agent's reasons for
supposing a claim to be justified just are that agent's justification for holding the
claim. As these reasons must be inferential, the claim's status is at best one of
being inferentially, mediately supported. There is no room in the coherentist
picture, then, for claims which are justified for an agent by anything other than
the inferential reasons the agent has for holding the claims to be justified. To
accept this is to accept the core of coherentism. It is also to reject the possibility
of exploiting the foundational structure that at least made the truth-
conduciveness problem seem solvable.
With this in mind we can characterize the central tenets of epistemic
coherentism as follows. First, all justification is inferential. Second, nothing
can confer justification without itselfbeingjustified. Third,justification cannot
be achieved by an indefinitely long inferential regress. Fourth, the only things

that can justify a belief are further beliefs? As we have seen, coherence accounts
of justification do not number among their virtues a logical structure obviously
suited to the solution of the epistemic regress, and hence the truth conduciveness
problem. Moreover, given the fourth condition, which holds that belief-belief
relations are exhaustive of justificatory relations, the truth conduciveness
problem has an immediate intuitive grip. Truth, after all, is a relation (or so we
tend to think) between beliefs, among other things, and the world, not between
beliefs and other beliefs. And why should we suppose that the first kind of
relation is guaranteed on the basis of the second, on any particular account ofthe
belief-belief relations that comprise coherence? Absent a good answer to this
question, it is hard to see how belief-belief relations, so conceived, can be
epistemic at all. So, despite the considerable intuitive appeal of the notions that
it is the business of rationality to seek reasons for all claims, and to seek the kind
of systematicity, unity, comprehensiveness, and explanatory power implied in
the notion of coherence, coherence accounts of justification have continually
foundered against the truth conduciveness problem. We can see this from the
fact that the other main objections usually raised against coherence accounts are
such that a solution to the truth conduciveness problem would go a long way
toward mooting their force. Consider, for example, the objection that a fully
coherent system of claims (including what seem to be empirical claims) might
be utterly cut off from the reality it purports to capture; it is hard to see how this
so called 'isolation objection' could get a grip, if it were satisfactorily
demonstrated that the contents of such a system are likely to be true. Surely the
demand for a connection between world and system is motivated by the sense
that, absent such a connection, the system's contents could not be true of the
world. So a demonstration of truth conduciveness, whether it invokes such a
connection or not, disarms the threat that motivates the objection. Consider also
the objection that there might be many possible coherent systems, but only one
world for them to be about, or only one true such system. An adequate solution
to the truth conduciveness problem would moot this objection too, either by
showing that only one such system is possible, if truth is univocal, or by showing
that truth is not univocal, if many equally coherent systems can be constructed
by truth conducive means? The task for any coherentism, then, is to show that
beliefs which cohere with one another in some specified sense are likely to be
true, in a way that makes one whose beliefs are coherent in that sense
epistemically justified in her beliefs.
To understand the approach Lehrer takes to this task, it will help to
contrast it with alternative coherentist approaches to the same difficulty.
Coherence would obviously conduce to truth if truth were identified with
coherence. But, as Laurence Bonjour points out, the truth conduciveness of
justification on a particular conception cannot count in its favor as compared
with other such conceptions, if our view of truth has been tailored, ad hoc, to

guarantee that result. The same move would be open to proponents of any view
of the nature ofjustification, and cannot help us to decide between them. Given
this constraint, Bonjour despairs of the kind of demonstration of truth
conduciveness I discuss here, one made from entirely from within the coherent
system. His own approach to the problem is to attempt an a priori demonstration
that coherent systems of belief meeting specific conditions are more probably
true than false. This approach cannot work in the context of a pure coherentism,
since it depends upon recourse to a priori judgments of comparative probability
which are not themselves justified on coherence grounds. Bonjour recognizes
this and consequently presents a coherence theory of empirical knowledge only.4
But perhaps it is possible, after all, to find an independent, non-trivial
basis for identifying truth with coherence. This was the route taken by the
classical coherentists, on whose view reality is a unified whole and claims-
beliefs contents or propositions-are abstracted parts of that whole which
necessarily capture it in a merely partial way. The more comprehensive and
coherent a set of such abstacta is, on such a view, the more closely it captures
concrete reality itself, since the unification of comprehensive abstracta through
coherence is just the reverse of the process of abstraction itself. Such views are
out of favor in this predominantly atomistic age. Be that as it may, the general
program of attempting to provide non-trivial grounds for identifying truth and
coherence is still live. In his A System of Pragmatic Idealism S, Nicholas
Rescher offers such an argument. Truth conceived as adequation to fact, is,
according to Rescher, identifiable with the notion of optimal coherence with a
perfected data base, where a data base is conceived as a set of candidates for
truth. Whatever else may be said about this argument 6 , the highly idealized
character of the conditions under which it concludes that coherence yields truth
stand out. There is no pretending that we now have belief systems which are
optimally coherent in Rescher's sense, and we are of course never presented
with a perfected data base. Ironically, Rescher's approach is on this score
similar to Bonjour's own effort to show that the stable coherence of empirical
belief systems meeting certain requirements is in the very long run destined to
yield truth. Whether or not idealization of this kind is appropriate in this context,
Lehrer's approach makes no such appeals to the ideal, intending instead to show
that concrete agents have a reason to think their current, certainly less than
comprehensive, beliefs nonetheless I ikely to be true in virtue oftheir coherence,
again less than ideal, with one another.
One way to attempt to show that an agent's coherent beliefs are likely to
be true under the normal, less than ideal conditions with which we situated
humans are faced is to approach the matter through an analysis of meaning. On
such a view, the truth of belief is built in through the relations with the world in
virtue of which the very content of belief is fixed. The approach at which I am
hinting is Davidson's. It invokes its own idealizations, thought in an entirely

different spirit, and is not without its difficulties, which I will not pursue here. 7
Lehrer's approach, however, declines to see epistemology in the mirror of
meaning, taking the contents of belief to be fixed, apparently, independently of
any relations to the world which would ensure their likely truth. The contents
of belief are what they are; the truth of belief is another, independent matter.
If strictly all of the beliefs in a given system must be justified on the basis
of coherence, and if the external connections that fix the contents of belief are
put out of play in addressing questions of epistemic warrant, then there is
nothing left to do but to seek the reasons for supposing coherence truth
conducive in the contents of the members of the belief system themselves. We
are left with the hope that beliefs about the probable truth of our beliefs can
themselves be justified by their coherence with the very beliefs whose probable
truth they assert. As our reasons will have to be reasons for holding a belief
justified, they will be epistemic in the sense mentioned above, higher-order
beliefs about the justificatory status of lower-order beliefs. We are to pick
ourselves up to truth by our own epistemic bootstraps. Such is Lehrer's
For Lehrer, knowledge is undefeated justification, where justification is
defined in terms of coherence with a background system. Lehrer offers the
following definition:

DK: S knows that p at t if and only if (i) S accepts that p at t, (ii)

it is true that p, (iii) S is personally justified in accepting that p at
t, and (iv) S is justified in accepting that p at t in a way that is
undefeated. 8

Acceptance, for Lehrer "is a sort of positive attitude toward a content, resulting
in employment of the content as background information in thought and
inference."9 Acceptance is thus epistemically oriented belief, individuated
Condition (iii) is concerned with the general structure ofjustification and
the conditions under which a person is personally-subjectively- justified in
accepting some content. For Lehrer, justification is coherence with a
background system of acceptances. A subject S's acceptance system is the set
of all statements of the form "S accepts that ... ," attributing to S just those
contents S accepts in Lehrer's sense. S's acceptance that p is justified if and
only if p coheres with her acceptance system.
Lehrer takes the notion of comparative reasonableness as a fundamental
primitive. A content coheres with an acceptance system if it is comparatively
more reasonable for a subject to accept that content than to accept any
competing claims, relative to that acceptance system. A claim c competes with
a content p just in case it is more reasonable for S to accept that p on the

assumption that c is false than on the assumption that c is true. That content p
coheres with S's acceptance system if it beats or neutralizes all competitors. p
beats c if it is more reasonable for S to accept p than c on the basis of S's
acceptance system. c is neutralized for S if there is some n such that the
conjunction (c&n) is as reasonable to accept on the basis of S's system as c
itself, and (c&n) does not compete with p. Consider the following abstract
Suppose a subject S avers some content q. An interlocutor, call him D,
contests this claim, saying "not-q". Obviously it is more reasonable for S to
accept that q on the assumption that this counterclaim is false than on the
assumption that it is true. Thus, D's claim competes with S's. Is S justified in
his averral? This depends on whether the counterclaim is beaten or neutralized
on S's acceptance system. Suppose S's acceptance system also contains the
following: "S accepts that r", and: "S accepts that r implies q". It is tempting to
conclude that since D's claim that not-q is logically inconsistent with the
entailments of these contents of S's acceptances, S's claim beats D's and is
justified. But this would be premature. Nothing present in S's acceptance
system bears on the reasonableness of accepting q over its competitor not-q. To
see that this is so, simply note that the members of S' s acceptance system are not
the contents accepted by S but rather statements of the form "S accepts that ..
" And it is relative to members of S' s acceptance system rather than to
contents accepted by S that competitors must be beaten or neutralized. Indeed,
this point is important for Lehrer. If competitors had to be beaten relative to the
contents accepted by S, rather than to the members of the acceptance system of
S, then accepted contents would trivially beat competitors. For it is always more
reasonable to accept some p on the basis of p than it is to accept some
competitor to p (e.g., not-p) on the basis of p. Moreover, it is precisely the
presence of the acceptance operator that distinguishes the relata of the coherence
relation as doxastic contents, as opposed to, say, facts or propositional contents
not believed; it is what makes Lehrer's coherentism doxastic.'o Now the
members of S's acceptance system so described, "S accepts that r" and "S
accepts that r implies q", bear no inferential or other rational relation to either
q or not-q. How, then, is D's competitor to be beaten relative to S's acceptance
system? Acceptance has the goal of obtaining truth and avoiding error. Hence
the presence of"S accepts that q" in S's system makes it less reasonable for S
to accept that not-q only if S' s acceptance of those contents is a reason for him
to suppose them more likely true than not. That S accepts some contents,
without more, is not a reason for him to suppose them true. Thus subjective
justification is, to this point, unable to get off the ground.
Before completing Lehrer's account of how justification might get
airborne, let us first briefly turn to his fourth condition. This condition, which
requires that a person's justification be "undefeated", is designed to help avoid

counter-examples of Gettier and related sorts. Simplifying slightly, a

justification is undefeated, for Lehrer, if it is undefeated relative to every set of
acceptances that can be created by correcting errors in the agent's actual
acceptance system. This ensures both that no essential falsehood is involved in
the agent's justification, and that the claim at issue is true, since if it were not,
the true claim that it is not would replace the false claim that it is true on some
corrected version ofS's acceptance system, on which S 's justification would be
defeated. Hence an undefeated justification of a claim is clearly epistemic.
But whether an agent's justification is undefeated is simply not
something which can be seen or determined from the doxastic point of view,
here from within the agent's acceptance system. I I The agent holds all of her
accepted contents to be true. Hence, from the doxastic point of view, there is no
saying whether ajustification is merely personally justified, or in fact ultimately
undefeated (hence not flawed) and yielding truth. The facts that make for
veridical justification are, as it stands, unreflected in subjective justification. So
to this point we have two problems. First, justification cannot get started.
Second, the facts that show that justification is not defeated, hence is likely to
yield truth, are not reflected in the agent's acceptances. Lehrer's bootstrapping
strategy seeks to solve both problems at once.
According to Lehrer, what is required to get justification started, and
moreover to give an agent reason to suppose her justifications ultimately
undefeated, is a kind of doxastic ascent. "One must have some information that
such acceptance is a trustworthy guide to truth.,,12 "1 must accept ... that when
1 accept something, that is good enough reason for thinking it to be true, so that
it is at least as reasonable for me to accept it than to accept its denial.,,13 Lehrer
regards an inquirer's self-trust in pursuit of truth as a special principle of an
acceptance system:

T. I am a trustworthy evaluator ofthe truth.

According to Lehrer, when a person accepts a principle such as T-when she

accepts that she is trustworthy in the evaluation of truth-the crucial link
between acceptance and epistemics, likely truth, is established. On his view, if
S accepts Principle T, she can conclude not just that D's claim that not-q
conflicts with her own views, but that it conflicts with what is likely to be true.
Since it is more reasonable to accept what is likely to be true than what conflicts
with what is likely to be true, S's claim that q would defeat D's competing
claim. Thus, S' s claim would be justified.
Principle T purports to make accepting something itself a basis for
supposing it likely to be true. Moreover, since justifications that are defeated
are not trustworthy, accepting that one's acceptances are trustworthy amounts
to accepting that one's justification for a claim is not defeated. 14 One's personal

acceptance ofT is supposed, then, to give one a reason for supposing that one's
warrant is veridical as well as personal, and thus personal warrant is to be made
epistemic. So if Lehrer's principle T works at all, it bootstraps acceptance to
likely truth, turning coherence with what is accepted into epistemic warrant.
But does it work? Recall that an acceptance system is comprised of
statements of the form "S accepts that p." Lehrer calls Principle T a principle
of detachment: he holds that one who accepts T is thereby entitled to detach the
contents of her acceptances from the fact that she accepts them. This in turn
permits those contents themselves to justify one another and to beat competitors
in appropriate cases. But this detachment process fails, as the following
argument shows.
Let x, y, ... be variables ranging over propositional contents P, Q, R, ..
. Introduce operators "As", such that "AsP" means "S accepts that P," and
"reass", such that "reassP" means "S's acceptance that P is reasonable." Let P
be some arbitrary content accepted by S.

(1) AsP

Assume that S accepts Principle T:

(2) AsT

We may render T formally as

T: (X)(Asx -7 reassx).

Assuming that the acceptance operator does not create an opaque context 15, then

(3) As(x)(Asx -7 reassx).

It is tempting at this point to instantiate the universally quantified statement

within S's acceptance in (3) with the content accepted in (1). But (3) is not a
universally quantified statement; it is rather an acceptance, just like (1). In fact,
no acceptance-no statement of the form "s accepts that ... "-follows logically
from any other. Clearly, if the detachment process is ever to get started, we must
introduce some logical apparatus that permits the desired instantiation.
Moreover, the need for a logic of acceptance is not limited to detachment.
Without a logic of acceptance, individual acceptances are isolated entities,
bearing no rational relations, coherent or otherwise, at all (save bare
consistency) to other acceptances. For coherence relations to be effective within
an acceptance system, a fairly powerful logical apparatus will be required. Let
us introduce the following three schemata.

(a) Asx ~ AsAsx; i.e. acceptance is iterative.

(b) (Asx & Asy) ~ As(x & y); i.e. acceptance is closed
under conjunction.
(c) (Asx & x 1- y) ~ Asy; i.e. acceptance is closed under
deductive entailment.

Think of these as inference rules a given subject mai6 employ in

maneuvering around her acceptance system. Since coherence is a matter of

explanatory and inductive relations as well as deductive ones, the foregoing,
though necessary, are not sufficient logical principles for Lehrer's theory. They
do nonetheless permit the agent, despite the presence of the acceptance operator
on each member of the acceptance system, to exploit some of the relations
among the contents of her acceptances, e.g., by generating new acceptances by
inference from others. But the question is whether this will help to detach any
contents from the acceptance operator. We may proceed.

(4) AsAsP (I), (a)

(5) As[(x)(Asx ~ reassx) & AsP] (3), (4), (b)
(6) As[(AsP ~ reassP) & AsP] (5), (c)UI

We can now detach the consequent ofthe instantiated Principle T.

(7) AsreassP (6), (c), MP

(7) says that S accepts that his acceptance ofP is reliable. But this is not
what S needs. That S accepts that P is reliable is not sufficient to show that his
acceptance of P is reliable. Even with the strong logical apparatus we have
supplied, acceptance of Principle T has failed to help detach P from its
acceptance operator. Thus P is no more reasonable to accept than its denial
relative to S' s acceptance system. As Davis and Bender put it, information
about what S accepts is not "adequate to drive the engine of justification." 17
What is required is access to the content p itself. Clearly, if T itself
appeared in the acceptance system of S, then the necessary detachment would
be possible. But T does not appear; the acceptance of T appears, and this
acceptance is as impotent to detach itself as it is to detach other contents:

(8) AsAsT (2),( a)

(9) As[(x)(Asx ~ reassx) & AsT] (3), (8), (b)
(10) As[(AsT ~ reassT) & AsT] (9), (c), UI
(11) AsreassT (10), (c), MP

That is, S accepts his acceptance of Principle T is reasonable. This, again, is no

In order to break this cycle of acceptances, S needs recourse to Principle
T itself.

(12) (x)(Asx -7 reassx) Assumption

Instantiating with any content, say P, we get:

(13) AsP -7 reassP (12),UI.

And then given (1), we get:

(14) reassP (1), (13), MP.

That is, S' s acceptance of P is reasonable.

This of course applies to the accepted content T as well.

(15) As T -7 reass T (12), UI

(16) reassT (2), (15), MP

That is, S' s acceptance ofT is reasonable. But a glance at (12) shows that
it is not of the form "S accepts that P" and thus cannot be a member of S' s
acceptance system. The engine has no spark.
The foregoing objection may appear overly technical, relying as it does
on the precise definition of the acceptance system. This reply is inapt, for the
exclusive use of doxastic states qua doxastic states in justification is of the
essence for coherentism, and on Lehrer's view, it is the presence of the
acceptance operator which marks a content as a member of an agent's
acceptance system. Moreover, as we saw, justification for accepted contents
would be trivial matter but for the presence ofthat operator. Be that as it may,
it is clear that if justification is ever to get started, Lehrer needs some ground for
including (12) or Principle T itself in S's acceptance system, free of any
acceptance operator.
Though Lehrer does not recognize that T is impotent as a principle of
detachment, and thus does not directly address whether it could appear in an
acceptance system without the acceptance operator, he does emphasize that T
is a "special," "make or break" principle. Perhaps what he says in this context
may show that T merits this special status.

What are we to say with respect to T itself? Am I intrinsically

justified in accepting T? Does the attempt to justify T take us to

an even higher level of doxastic ascent generating a regress? Or

does the attempt to justify T lead us in a short circle in which T
provides us with the reason for accepting itself? There is a sense
in which the answer to all three ofthese questions is positive,'S

Now obviously, if Lehrer is serious that T is justified as a

foundation-one acceptance among others but not in need of justification by
appeal to the rest-he has just given up on coherentism itself, at least in any
pure form.19 Nonetheless, it is interesting to note that T itself is not a
particularly good candidate for foundational status. Perceptual, introspective
and memory foundationalisms are preferable to a foundational ism of general
trustworthiness for the simple reason that they discriminate between those sorts
of beliefs that are highly trustworthy and those that are not. 1fT helps to justify
anything one believes, it helps to justify everything one believes injust the same
way. But we are of course much better at truly perceiving tables than at truly
theorizing about quarks. It is absurd to suggest that common perceptual beliefs
and speculative scientific claims are primafacie equally trustworthy in virtue of
their mere acceptance. (That T would apply indiscriminately to all acceptances
is a clue to the special status T has, I think, and is a point to which I will return.)
Lehrer has two reasons for thinking T a good candidate for foundational
status. First, T can be confirmed. That is, the more one learns about the world,
the more support she has for believing she is trustworthy. This point can hardly
count in T's favor as a foundation, for if it is a foundation it needs no further
support, and if it needs confirmation it cannot, on pain of circularity, provide
ground for that which confirms it. Furthermore, Principle T is by no means
special among foundations in its confirmability. Our perceptual beliefs,
introspective beliefs and memories are regularly confirmed by more ofthe same
and by other beliefs as well. T gains no advantage over such foundations here.
Second, Lehrer claims that principle T is preferable to other foundations because
it "tells me my accepting something is a guide to truth." In particular, its content
"provides reason for considering acceptance of it to be justified.,,20 That is, it
indicates why it is itself a good foundation. But if T is a foundation, then it
needs no extra support, from itself or elsewhere, as a foundation. If it is a
foundation, then it can groundjustification. If it is not, then it cannot on its own
ground its own justification or that of anything else.
How is the justification of Principle T regressive? Any attempt to justify
Principle T by reference to other acceptances depends upon the trustworthiness
of those acceptances, and thus ultimately upon T itself. In addition to a
circularity problem, to which I will turn presently, we face then, a case in which
Principle T is applied to itself. Such self-reference is notoriously problematic,
and the way to avoid it is also a way to avoid circularity. All we need do is
ascend another level on our doxastic staircase, positing a meta-acceptance to the

effect that one is a trustworthy evaluator of whether one is a trustworthy

evaluator of truth. On pain of the same circularity and self-reference in the
metalevel that we seek to avoid at ground level, the justification of the meta-
acceptance demands yet another T -like principle at yet a higher meta-level. And
the regress ensues. Lehrer claims that this regress is not vicious. "The regress
does not commit us to carrying out an infinite series of acts. On the contrary, it
simply shows us that an infinite series ofT principles at various levels are each
such that, when accepted, one is justified in accepting them all."21 Now, the
mere fact that a regress is not vicious in the sense of requiring an impossible
infinity of distinct acts hardly shows that the regress can actually confer
justification. The idea here is that each Principle T in the series is justified by
the one at the higher level, and that when one accepts the series as a whole, each
member of it is justified. But the intelligibility of such an infinite series of meta-
acceptances is doubtful. Are we really to suppose that we understand, and can
employ in reasoning, acceptances of the form AsTn, where Tn is to be unpacked
as "I am a trustworthy evaluator of the truth of T n-l ", and where Tn-l is in turn
to be unpacked as "I am a trustworthy evaluator of the truth Tn-2", for arbitrarily
large, let alone infinite, n? I think the need to suppose so should call into
questions the very project of attempting to justifying according T a special
status as an acceptance among others. And even supposing we could make sense
of this regressive justification we would still face the question "what justifies
acceptance of the series?" Itself? That is self-referential and circular. Another
meta-acceptance, of a yet higher order altogether? Another series of meta-
acceptances? We're off to the races again.
We come finally to the circular justification of Principle T. The argument
is supposed to run as follows: I accept that T; according to T my acceptances are
trustworthy; therefore my acceptance ofT is trustworthy. But ajustification of
T is supposed to tell us that it is reasonable to accept T in the first place, and in
order to determine this we must put T out of play in our justification, just as we
refuse to allow any claim we accept to figure in our justification for it when we
are challenged, at least insofar as the claim is used directly to support itself.
But while Lehrer remarks at one point that "even in the case of principle (T), we
require some background information in order to be personally justified in
accepting the principle,,22, this does not mitigate against the charge that the circle
here is vicious. Clearly, any background information upon which the
justification for T depends will itself depend vitally upon T for its own
justification. Not only must T itself appear in its own justification, it must
appear in the justification of everything else that appears in that justification.
A less benign circle is hard to imagine.
Moreover, as a functional matter-and remember that acceptance is a
functional matter-reasonable people accept something that would appear to
bar, or at a minimum conflict with, this circular justification: we generally

accept that such circular justifications do not increase the antecedent likelihood
or reasonableness of the justificandum. This conflict might be resolved by
recourse to the idea that this specific acceptance is a competitor to the claim that
T justifies itself, and that this competitor can be neutralized by being conjoined
with the claim that this case of circular justification is benign. That is, while

c: It is unreasonable to accept circular justifications

competes with

TR: My acceptance of T makes it reasonable for me to accept T,

c&n: Circular justifications are generally unreasonable, and TR

does not. But in order for c&n to neutralize c it must be as reasonable to accept
c&n as it is to accept c itself. But an acceptance can only be reasonable if the
content is more probable than its denial, relative to the acceptance system. This
is precisely what circular justification, generally and in the case of Principle T,
cannot show. Since the general prohibition on viciously circular justification
clearly serves the interest in obtaining truth and avoiding error, c&n can work
to neutralize conly ifTR is treated as an exception to the general prohibition on
circular reasoning. But to grant an acceptance such an exemption seems utterly
ad hoc. Now it is clear that, as a functional matter, we do trust ourselves as
evaluators oftruth. But that is not to say that we in fact exempt the acceptance
of our reliability from the standard proscription against circular justification. It
would only follow that we exempt self-trust from this proscription if we
appealed to self-trust as another acceptance exploited in an inference or
argument for its own validity; and this is exactly what this neutralization
involves. If that is how T achieved its special status, it would be at the cost of
making us ad hoc, bad reasoners. But, as we shall see, general self-trust need
not be conceived as an acceptance exploited in justificatory inference.
Since Principle T is not justified as a foundation, and the regressive and
circular justifications are unacceptably vicious, there is no reasonable ground for
permitting Principle T to appear in one's acceptance system free of the
acceptance operator. Consequently, the principle cannot operate as it must on
Lehrer's account if we are ever to be justified in our acceptances. That we think
we are trustworthy does not make what we think worthy of trust. Consequently,
Lehrer's coherentist solution to truth conduciveness problem fails. Including an
epistemic bootstrap among the relata of the coherence relation does not render
that relation epistemic.
In Self Trust, a comparatively recent work, Lehrer seems more aware of
the futility of a bootstrapping demonstration of the reasonableness of our

coherent acceptances. Conceding that "circular reasoning justifies nothing,m,

Lehrer recasts his goal in appealing to self-trust as explanatory, rather than
justificatory. He declares: "my intention is not to prove conclusions by circular
reasoning, but to explain the reasonableness of what we accept ... in terms of
a principle of our own trustworthiness." 24 The idea now is that, while the
principle of self-trust cannot be used to prove that we are justified in accepting
the principle, the truth of the principle can explain our reasonableness. But it
remains the case that the explanatory strategy appeals to the truth of the
principle of self-trust in order to explain the reasonableness of accepting that
very same principle. So, as Lehrer concedes, the circularity has not gone away.
But he argues that circularity is not impermissible in explanation. He claims that
explanation, at least insofar as finite, must either end in some unexplained
explainer, or in some principle which explains both itself and other explananda.
Thus, "we must choose between the surd and the loop". 25 Seeing no reason to
prefer explanations that terminate in surds, he opts for to admit circular
explanations (though he is not dogmatic about this-we may opt for the surd if
we like), on the ground that, if we do so, "nothing need be left unexplained,,26.
This putative retreat from justification to explanation cannot cure the
logical ills of the appeal to self-trust. First, in explanation, as oppose to
justification, the truth of the explanandum is taken for granted. Here, at least
part of the explanandum is the justified status - the reasonableness -- of
acceptances. Thus the reasonableness of what we in fact accept is taken for
granted in this explanatory project. But in taking this for granted, Lehrer has
certainly changed the game we have been playing. If we were willing to take
this reasonableness for granted all along, then the truth conduciveness of our
acceptance generating processes would not, it seems, have merited such fuss.
Indeed, one wonders why a natural, psychologistic description of our acceptance
formation processes would not have sufficed to describe that whose
reasonableness needs to be explained (but not proven). Now Lehrer might
respond that there could be no explanation of the reasonableness of our belief
forming processes described psycho logistically. After all, it is reasonableness
we are trying to explain, and, given Lehrer's staunch anti-naturalism 27 , there can
be nothing reasonable about a merely natural psychological process. But this
shows, I think, that the distinction between explanation and justification here is
not so sharp as Lehrer must suppose. For surely the reason there can be nothing
reasonable in merely natural psychological processes is that it cannot be
explained how such processes yield justified, reasonable acceptances. Part of
explaining how a process can yield justified, reasonable acceptance must, then,
involve displaying that the process is justified, and that involves justifying it.
So if circularity is impermissible in justification generally, then it is
impermissible in the explanatory project as Lehrer conceives it here, where part

of the explanandum is our acceptances' entitlement to justificatory status, and

hence where explanation involves justification.
Second, despite Lehrer's confidence, it is not at all clear that explanation
can be circular. The very idea of explanation only makes sense, one might
suppose (and many have argued), against a backdrop of information taken for
granted, and available for service in explanans. Relative to a given explanatory
context, this information is taken as surd. But it is this very status that makes
explanation, as an attempt to come to understand what is not understood by
appeal to what is understood, possible. This also entails that not everything can
be explained at once. But that is as it should be. From the fact that appeals to
the surd do not explain everything, it does not follow that they explain nothing.
Thus the fact that explanation must inevitably appeal to the contextual surd does
not mark a kind offlaw in explanation, which would form a suitable second horn
of an unhappy dilemma with circularity as the other, and thereby make the circle
look no less appealing than the surd. For circular explanations do not explain
anything. Rather, they incoherently treat the very same thing as in need of
explanation (even if true) and as surd, so not in such need. For this reason,
rejecting circular explanation is not merely a matter, as Lehrer suggests, of
taking philosophical offence.
So I conclude that the retreat from justification to explanation, in the
context of showing the reasonableness of coherent justification, fails. If the
move to explanation in fact marked more than a merely apparent retreat from
justification, it would amount to giving up on the goal of solving the truth-
conduciveness problem, by assuming it away. Moreover, any purported
explanation would be circular, and circular reasoning is no more explanatory
than it is probative. And in fact the retreat is only apparent, for explaining how
a process merits normative status involves in part displaying its entitlement to
that status, by justifying it. But the justification would have to be circular, and
Lehrer has certainly not been unaware of the sorts of logical worries with
arguing for the reasonableness of our self-trust that I have been belaboring in
this paper. Over the years he has been candid and dogged in their face. As we
have seen, he has tried on each of the strategies of Agrippa's ancient trilemma-
regress, surd foundation, and circle-in defense of his appeal to our own
trustworthiness in the context of demonstrating the truth-conduciveness of
coherence. He has even tried these concurrently. Lately he has opted for
circularity, seeking to avoid the implicit vice by recasting his project as
explanatory rather than justificatory. I have been harsh in my assessment of
these moves, and have not had to resort to much subtlety in my criticism.
Moreover, I am far from their only critic. Yet Lehrer will, no doubt, as he has,
persist in his defense of the essential role of self-trust in epistemology. And so
he should. For Lehrer is right to recognize the absolutely fundamental

importance of self trust to the epistemological enterprise, and indeed, to

epistemic practice generally. Where he is mistaken is in thinking that an
entitlement to self-trust could possibly be argued for at all. It cannot be.
Moreover, in a crucial sense, it is not something that can be doubted. 28
As I noted at the outset, epistemic practices are essentially directed at
truth. But we should not let the focus on truth blind us to the practice. For a
practice to be directed at truth, it must involve the treatment of faculties that
generate contentful states as potentially truth disclosing, and the treatment of
contentful states as premises in reasoning, which reasoning itself is treated as
truth generating (or as at least probabilifying). Now to treat a content generating
faculty as potentially truth disclosing in practice is potentially to exploit those
contents in reasoning and further inquiry. For a faculty of content generation to
be a component of epistemic practice proper, the contents it generates must be
treated as truth candidates, since otherwise they would not be involved in a truth
directed practice. To involve such contents in a truth directed practice is to
exploit them in reasoning, again directed at truth. But to exploit contents in such
a manner is to trust that they are true, or at least presumptively so. Moreover,
to exploit them in inference is to trust not only them, but also one's competence
in inferential matters.
This is so no matter how we conceive the psychology ofbeliefformation
and reasoning. Suppose the progress of our thoughts were a matter ofHumean
associationism, without that progress tracing any demonstrative relations of
ideas. Still, despite the lack of any necessary or even clearly probabilifying
relations among contents in such an associationist doxastic economy, the links
in the associationist chain would be taken by the subject, in practice, to
constitute epistemic links. This is revealed in the mind's movement from
believed content to believed content, as opposed to the movement from believed
or entertained content to entertained content. Indeed the very idea of taking a
contentful state to be true implies a self-trust in so taking. Self-trust is also
implicit in the authority we have over the contents of our thoughts. And of
course the idea that we can reason and evaluate arguments presumes that we can
maintain that authority through time. Even doubting a claim presumes that the
doubter trusts his own grasp of what he doubts. Nobody who engages in
reasoning can at the same time really fail to trust their evaluation of the truth of
their second order thoughts. Self-trust is thus at play, and cannot be suspended,
in all thinking, even in epistemological or skeptical thinking.
For these reasons, self-trust is a condition on the possibility-a
transcendental condition, if you will--of epistemic practice. But any attempt
to justify self-trust as a blanket matter will involve self-trust, and hence be either
circular or involve an infinite regress. But to call self-trust into question, which
the idea that it demands justification does, is of course to call into doubt an
essential precondition of the practices called upon to justify it, and hence those

practices themselves. The result is predictable. The practices break down, in

the sense that they seem unable to function, and lead to aporia. This shows not
that it is unreasonable to trust ourselves, but that self-trust is not a candidate for
rational support.
Now we do often, in the course of a skeptical challenge to some claim or
other, appeal to our own self-trust. But when we do this, we are doing one of
two things. We might be citing a limited principle, involving our
trustworthiness in this or that context or domain of inquiry, or respecting this or
that kind of content. Such limited self-trust can be justified without circularity
or regress. Or we might be appealing to a general principle of self-trust, which
as we have seen, cannot be justified. But such an appeal should not be construed
as a matter of giving a further reason of the same sort as others. Rather, it is a
matter of the respondent affirming her status as an epistemic agent, by citing a
condition on the entire enterprise of giving and asking for reasons. She is, as it
were, citing the rules that make the practice intelligible. And she is citing a rule
that all participants in the game perforce accept, in their own case. But of
course, to cite the rules of a game is not to make a move in the game. (It is not
as if she couldn't give further reasons of the ordinary sort. This suggests that the
move really is of a different sort).
On this way of thinking, a principle of general self-trust is not a member
of one's acceptance system, but a kind of hand-stamper at the door, whose effect
is to mark contents as accepted, thereby making them available in truth directed
reasoning and justification. Self-trust operates, not as a principle of detachment
of an acceptance operator, but to distinguish wh ich among those contents we can
think are to be exploited in epistemic inquiry.
From the essential ubiquity of self-trust, it follows that all arguments,
including all efforts to prove or explain why our coherent, undefeated thinking
is truth conducive, will rely on self-trust. But this need not be a problem, for no
explicit appeal need be made to general self-trust, from which circularity or
regress would have to follow. There is a difference between presuming self-trust
and arguing from or for it. The former must be done, the latter cannot. A proper
appreciation of the epistemic significance of self-trust, to which Lehrer comes
so very close, would shield it from the embarrassments that it suffers when we
try to show that we have earned it. But the way Lehrer conceives of self-
trust-as a principle which is special, but special among acceptances to be
exploited in bootstrapping-forces him to attempt such a showing. Since none
can be made, we tug at those bootstraps in vain.
This should not be surprising. Bootstrapping is in its nature a futile act.
So what is it that has driven Lehrer to such extremes? As we saw, Lehrer gives
accounts of justification on two levels, the first subjective and doxastic, the
second objective and veridical. The express purpose behind the objective
accounts is in each case the avoidance of Gettier and Gettier related

counterexamples. The basic idea seems to be that personal justification is

actually unproblematic excepting in the light of those putative counterexamples.
Generating counterexample-proof conditions, and then making sure that personal
justification appeals to these measures at a general level, should accordingly
restore an individual's knowledge based on her justification. I think the lesson
of Gettier is deeper than that. What these examples show is that, in any given
case, for justification to yield truth, some external help, the cooperation of the
world, is required. Whether we get that needed help in any given case is always
a contingent matter, depending on factors outside of the agent's space of
reasons. The lesson of Gettier and other cases may be that a certain kind of
skepticism, the skepticism which claims that in any given case our justified
empirical beliefs may fall short of knowledge, is unavoidable. But that would
lend no support, on its own, to the more radical skeptical claim that we might
therefor be massively mistaken in what we believe. The gap between personal,
doxastic justification and veridical epistemic warrant will seem to need closing,
by hook or by crook or by bootstrap, only if we fail to separate these two threats.
And we will fail to separate them only if we think that the only way to ensure
our grip on reality is through the identification of a justificatory procedure that
will give us the ability to weed out the false from the true in our particular
beliefs. The coherentist who feels the need for bootstraps shares at least that
much with the foundational Cartesian: the urge for a principle the satisfaction
of which by any given case of believing will show that belief to be true. While
the coherentist has a different kind of principle in mind, it is to serve the same
purpose. (They tend to share more than this with the Cartesian as well,
including a pernicious subjectivism and a strict internal ism about mental
contents, though Descartes himself may not have been a Cartesian in this latter
regard). Ifwe reject the demand for such a principle, how then can we reply to
the general skeptic? We can say that while the connection between any given
truth and belief is contingent, and hence any particular belief is fallible, the
connection between belief and truth in general is not contingent. This will not
necessarily mean that justification is out as an interesting epistemological
analysandum. Surely all the cooperation the world can provide will not make
you knowledgeable if you lack the epistemic virtues embodied in the concept(s)
of justification. And in this same way we can answer the question which
motivated Lehrer's bootstrap in the first place, of how justification, in the light
of these results, is epistemic at all. Justification is epistemic because of the
intrinsically epistemic nature of what gets justified and does the justifying. But
this is an argument for another day.
I want to conclude, in this light, with a couple of observations from other
contexts in which one speaks of bootstrapping. The image it calls to mind is of
some poor supine soul's pitiable and failed attempt to yank himself into standing
without pushing off of solid ground, or to achieve some leverage even without

benefit of Archimedes' one fixed point. In the economic sphere, however, some
people are credited with pulling themselves up from destitution "by their own
bootstraps", and we admire them for turning the trick. But they do not really do
so. They use their labor, skills, wits and guts, each of which can help them rise
only if the surrounding world, social and other, complies by valuing and
rewarding them. I think a lesson can be drawn from each of these images
associated with the notion of bootstraps. First, the perceived need for such
desperate epistemic efforts only arises when we think of ourselves as utterly
unmoored, unanchored, unconnected to what our beliefs are about. But this
supposition is false, as the phenomenon of empirical belief itself shows. Second,
we can expect that our own efforts in the epistemic realm, however sincere,
coherent, and strenuous they might be, will not achieve their valuable ends
without the help of what is outside us: the world and other inquirers. As the
related phenomena oflanguage and thought show, we are not without such help.

* Thanks to Arthur Fine and Meredith Williams for especially valuable comments on earlier
versions of this paper.
1 William P. Alston, "Two Types of Foundationalism", pp. 76-94, in Empirical Knowledge, ed.

P. Moser ((Rowman and Littlefield 1986).

2 This fourth thesis is widely held by coherentists and thought by some writers to be central to the

view. So Donald Davidson, a coherentist, more or less identifies coherentism with the fourth
thesis, and John Pollock rejects coherentism precisely because, in his view, the fourth thesis must
be false. But it is worth asking whether the only possible coherentism is doxastic coherentism.
It is perhaps possible that some factive states with empirical content, such as takings-in of
perceptual, memory, and perhaps even testimonial contents (i.e., seeing, remembering, and being
told that p), are involved in inferential justification. The obvious problem with this line is that, by
the second thesis, in order for such factive states to provide justification, they must themselves be
justified inferentially, and it is hard to see how factive states, qua factive, could need or bejustitied
by inference. So what is clear is that the contents of such states could not appear in justifications
qua [active, but only qua accepted contents. Indeed, in this sense, the way factive states might play
a role in justification fully respects the rationale behind the fourth thesis.
3 It is a further objection to coherence accounts that the notion of coherence has been less than

fully worked out. But in this regard, coherence is innocent by association; whether or not there
are any non-inferentially justified claims to serve as foundations, still it is pretty well agreed upon
that some justification is inferential, and that strictly deductive inference is insufficient to support
anything like what we regard as justified in our accounts ofthe world. Thus the sorts of relations
which figure heavily in coherence accounts, explanatory, unifying and inductive, figure in
competing accounts as well.
4 See Bonjour's The Structure of Empirical Knowledge, (Harvard 1985), especially ch. 8.

5 Princeton University Press, 1992. See Part Three, and especially the appendix to Chapter 12, at
pp. 216-222.
6 One problem for Rescher is his notion of a datum. A datum is not simply a possible truth; it has

a stronger claim on veridicality than that, such that it is "to be classed as true, provided that doing
so creates no anomalies." Gp. Cit., at p. 166. One may well ask in virtue of what an otherwise
mere claim is entitled to this more presumptively alethic status. Rescher has much to say about
this concedely crucial issue which I must pass over here. I can say, however, that a metaphysical
view like that of the idealists, for whom all claims are connected to the real as abstracta, could be
pressed into service here. Likewise the pragmatics of praxis might be invoked on the behalf of the
treatment of claims as data. Neither ofthese moves is obviously out of bounds in Rescher's system
which is, after all, a pragmatic idealism.
7 In my "Interpreting Davidson's Omniscient Interpreter", Canadian Journal of Philosophy,

1995:335-374, I offer a sympathetic reading of Davidson's anti-skeptical arguments which

nonetheless seeks to isolate a central question-begging move in his account. A different, deeper
difficulty for Davidson's account is that it treats the external relation in virtue of which beliefs get
their contents as purely causal. But as John McDowell has urged, the phenomenon of empirical
belief is hard to explain unless mental contents are connected to the world not just causally, but
rationally. Absent a rational constraint imposed by the world, it is hard to see how a belief can
properly be thought to be about the world, to take a stand with respect to how the world is. Recall
here the discussion from footnote two above, involving the possibility that factive states, taken in
perceptually or through testimony, for example, can playa role in justification, though not qua
factive. Taking in facts about the world in such ways would make our connection to the world a
rational one, for facts, unlike causal events, have the form of reasons. In recognizing thatjustifiers
may be factive, though notjustifiers qua factive, I am merely resisting what McDowell calls the
'highest common factor conception of experience', which holds that two states which are
indistinguishable from the point of view of the agent, and his epistemic justification, are for this
reason metaphysically indistinct. In this connection, see McDowell's Mind and World (Harvard
University Press 1994) and his "Criteria, Defeasibility, and Knowledge", Proceedings of the
British Academy (Oxford University Press 1982) especially section Ill.
x Keith Lehrer, "Coherence and the Truth Connection: A Reply to My Critics" [hereinafter CTC],
The Current State of the Coherence Theory, ed. John W. Bender (Boston: Kluwer, 1989) 255.
Other versions of Lehrer's theory and definitions appear in "Knowledge Reconsidered"
[hereinafter KRJ, in Knowledge and Skepticism, ed. Marjorie Clay and Keith Lehrer (Boulder, Co:
Westview, 1989) pp. 131-154; and Theory of Knowledge (Boulder, CO: Westview 1990)
[hereinafter TK].
9 KR, p. 136.
10 Towards the end of this paper, I argue that self-trust should be conceived as discriminating
which of the contents we can think are to be made available for epistemic practice; that is, it
determines which contents we accept, in Lehrer's sense. The members of an acceptance system
would not, on my view, have an acceptance operator on them, but would rather simply be contents
accepted. But this would not have the effect of making justification trivial, for in justifying
something, we put it out of play as ajustifier. Thus, a given accepted content p cannot be appealed
to in defeating its own competitors. Moreover, that the particular contents used in epistemic
justification are accepted is sufficient to secure the doxastic nature of justification, without the
need for an acceptance operator in each case.
II Indeed, given that an acceptance's status as personally justified depends upon whether it beats

all competitors, and that the agent's acceptance system does not plausibly reflect all the these
competitors, it is clear that even personaljustification on Lehrer's account fails to be doxastic. See
my "Justified Acceptance, Information and Knowledge" Philosophical Forum, XXV, 1994:212-
230, for the details of this argument.
12 CTC, p. 253.

13 TK, p. 122.
14 Lehrer is happy to include in an agent's acceptance system specific acceptances whose contents
are denials of specific defeaters, where that is required to preserve intuitions about knowledge in
particular contexts, including those not involving essential reliance on falsehoods. He can do this
because of the functional character of acceptance. In accepting the things I do, I evidently also
accept, with respect to specific possible defeaters, that they do not obtain. Ifmy being cut off from
the world, for example, would defeat my claim to true belief about it, then, in accepting that I have
such true belief, I evidently accept that I am not cut off from the world. In my "Justified
Acceptance, Information, and Knowledge", I urge that this has the consequence of obliterating the
distinction between justified and unjustifed beliefs.
15 It is an interesting question whether opacity problems could arise in the context of acceptance
for Lehrer. Since acceptance is a functional notion, acceptance being a matter of how contents are
actually used in inference, it might be though purely extensional. On the other hand, to the extent
that one's behavior in assenting to and dissenting from statements is functional evidence of
acceptance, opacity problems might arise. I can assent to "Venus is the morning star" while
dissenting from "Venus is the evening star" and can treat these claims differently from a functional
standpoint notwithstanding that they are coextensive.
16 Given the functional notion of acceptance, the general iterative principle seems reasonable, and

the conjunctive principle is not counter-intuitive. The deductive closure principle is certainly false
if taken in its generality. This is why I have used the permissive "may employ" in characterizing
the operation of the rules. Thanks, by the way, to Myles Brand, for bringing to my attention a
premise missing from a previous formulation of my argument here.
17 Wayne A. Davis and John W. Bender, "Technical Flaws in the Coherence Theory", Synthese
79.2 (1989): 271.
18 KR, p. 143.

19 Lehrer has in fact described his account as an ecumenical reconsideration of externalism,

foundationalism, and coherence (TK 173). He acknowledges that he is a sort of a metareliabilist

(CTC 263). He admits that Principle T "has many ofthe features ofa basic belief in foundational
epistemology," and he has "no objection to the suggestion that ... T ... is a first principle" (KR
145). While this insulates Lehrer's theory as he conceives it from the charge that it fails as
coherentism, I am here concerned with it as a coherence theory. Moreover, Lehrer's candor about
the impurity of his account does not insulate him from the charge that he, like other ecumenicists,
rather than taking the good from the positions he reconciles, has taken the bad from each and lost
the benefit of anyone of them.
20 KR, p. 145.

21 KR, p. 146. A grammatical error makes this hard to interpret.

22 TK, p. 124.

23 K. Lehrer, Self-Trust: A Study o/Reason, Knowledge, and Automony (Oxford 1997), p. 22.
24 Ibid.
25 Ibid.
26 Ibid.

27 This anti-naturalism is made manifest in many places. A particularly clear and worked out
argument for the claim that epistemic properties are not natural can be found in chapter 3 of Self-
Trust. There Lehrer argues that knowledge does not supervene on any natural properties, since my
actual trustworthiness as an acceptor is crucial to knowledge, and "nature is silent about what has
worth, what is worthy of our trust, and what is justified" (p. 72).
28 The situation is indeed quite like that which faced Descartes as he tried to ground our

knowledge. He recognized that what was needed was a way of defusing the worry that our beliefs
might be mistaken, despite our most careful use of our reasoning powers. His method was to
attempt to instill this self-confidence by causing us to reflect, first, on our inability to doubt our
own existence, and second, on the general features - the clarity and distinctness- of the "I exist"

that make it a practical impossibility to doubt.

Descartes' methodology, and the success of his strategy, have of course been subject to repeated
attack and refutation. The point I want to make is that Descartes' strategy is of course doomed to
fail if it is taken as argument. For any argument for the claim that we can trust ourselves - that
what seems clearly to be true to us is true - will of course involve us in trusting ourselves. As
Spinoza pointed out, taking the Cogito as an argument would be a serious mistake. Rather, it is
a self-standing proposition, "I exist as a thinker", which we are asked to contemplate, to see that
it cannot be doubted. The same must be true of Descartes' third meditation "proof' of the
principle of clear and distinct perception. If it is taken as a genuine piece of argument, it is
manifestly circular both in relying on clearly and distinctly perceived premises and on inferences
whose validity is clear and distinct. But unlike the Cogito, this "proof' is hard to treat as a self-
standing proposition for similar contemplation. Perhaps for this reason, Spinoza rejects the need
for an epistemic criterion of truth along the lines of Descartes' principle, and holds instead that
truth is its own standard.


Chapter 13


Hans Rott
University ofRegensburg


Philosophers must not be allowed to confuse epistemic and doxastic concepts.

It is their duty to clarify the subtle interconnections between knowledge and
belief. As this is too formidable a task for a single paper, I will not develop an
epistemological theory of my own, but rather focus on Keith Lehrer's influential
theory of knowledge as elaborated in his 1990 book Theory ofKnowledge. This
book represents only one stage of the development of Lehrer's epistemology. It
is the successor of, and shows considerable overlap with, a book with the title
Knowledge published by the same author in 1974. The basic structure of the
1990 definition of 'knowledge' is later retained in Lehrer 1997, Chapter 2, and
duplicated in an analogous definition of 'wisdom'. The recent second edition
of Theory of Knowledge (Lehrer 2000) presents a concept of knowledge that is
much simplified as compared to the one of the first edition. I The present paper,
however, is based mainly on the more 'dynamic' 1990 version of Lehrer's book.
Its purpose is, first, to draw attention to some problematic features of Lehrer's
account and, second, to argue that a proper understanding of knowledge does not
only require an understanding of belief simpliciter, but in addition a thorough
understanding of the dynamics of belief
The main point ofthis paper is to show that a good theory of belief rev i-
sion is necessary for a proper development of a theory of knowledge. I shall later
also argue that the theory of belief revision can profit from a study of the con-
cepts that have evolved in recent epistemology. We ought to have a basic picture
of what knowledge is, and of how knowledge can be obtained, when we search
for principled solutions of the problems of belief change. In Rott 2001, I exploit
E.1. Olsson (ed.), The Epistemology of Keith Lehrer, 219-242.
© 2003 Kluwer Academic Publishers.

the fundamental distinction between foundationalist and coherentist accounts

of knowledge that has played a central role in the epistemological literature of
this century. 2 I want to argue that this distinction applies more properly to the-
ories of doxastic states than to theories of epistemic states. It is unfortunate
that philosophers have tended to focus on knowledge without attending equally
closely to the-seemingly less problematic-notion of belief.
Another distinction is not going to be treated in this paper. Externalist
accounts of knowledge argue that people or animals need not be able to justify
their knowledge, but rather hold that there must be some reliable (causal, coun-
terfactual, nomological) connection between the knower and the things known.
Indeed it makes perfect sense to say that, for instance, the dog knows where
in the garden its favourite bone is buried, but I want to focus on an internalist
and what Lehrer (1990, pp. 4, 36) calls 'characteristically human sort of know 1-
edge'. To my mind, the verb 'to know' is ambiguous and its variant meanings
reflect different and conflicting intuitions-a fact that has caused much unnec-
essary dispute in recent epistemology. I do not claim to cover all uses of 'to



Arguably, the representation of pieces of knowledge should not differ from the
representation of beliefs. From a first-person perspective there is little if any-
thing that allows one to distinguish between mere belief and real knowledge,
and it is doubtful whether we should expect the representation of belief and
knowledge to represent more than what is accessible to the reasoner or reason-
ing system itself.
Now let us suppose, for the sake of argument, that questions concerning
the representation of knowledge and belief have been answered to everybody's
satisfaction. Many interesting questions are still left open. Solutions regarding
the concept and the representation of knowledge do not automatically give an-
swers to questions concerning the dynamics of belief and knowledge. How is
'knowledge', or better: alleged knowledge, revised in the light of new evidence?
How should it be revised? Agents are fallible, and what they consider to be
knowledge quite often turns out to be false-and hence not to be proper knowl-
edge at all. At this stage the focus of attention gets shifted from knowledge
to belief, and normative questions are seen to become increasingly important.
Epistemology, the philosophy of mind and psychology analyse and describe
what knowledge is and how it is obtained and represented in human beings.
Knowledge representation has to do with normative issues in so far as there are

many different approaches to representing information, some of which are 'bet-

ter' and some of which are 'worse'-with respect to what we demand from the
relevant 'knowledge systems' (e.g., computational tractability, efficiency, com-
prehensiveness, reliability, transparency). The change of alleged knowledge,
that is, the change of belief and acceptance systems, is intrinsically beset by
normative problems as well. It surely is reasonable to ask for an 'ethics' of be-
lief and acceptance. But it seems that this question can be set aside when we
talk about knowledge. In so far as 'knowledge' is perceived to be in need of
revision, however, it is perceived to be inferior to real knowledge, and should
rather be accorded the status of belief, opinion, prejudice or some similar sort of
doxastic (rather than epistemic) term. Problems of the ethics of belief arise in so
far-and perhaps only in so far-as problems of a genuinely dynamic sort arise.
The ethics of belief in the traditional understanding is primarily concerned with
the question when it is rational or justified to adopt some new belief, thus in-
volving an act of belief acquisition. 3 The additional problem of belief change
is that we have to face the question when to eliminate or replace which of the
previously held beliefs, thus involving acts of belief dislodgement.
Primajacie, it appears that the questions posed by epistemology, knowl-
edge representation and 'knowledge' revision are, though related, clearly sep-
arable. In any case, there does not seem to be a close connection between the
theory of knowledge and the theory of belief revision. It is the thesis of the
present paper that precisely this is illusory. Even if we do not contemplate any
problems of the 'intermediate' field of knowledge representation, we can find
very close interdependencies of epistemology and belief revision. More pre-
cisely, I want to illustrate that

(i) the analysis of knowledge requires a proper solution of the problem of

belief revision;

(ii) the analysis of belief revision should not be conducted without a proper
understanding of the concepts and categories that have been used in study
of knowledge.

We must actually restrict claim (i) considerably because the following consid-
erations will be based on the particular epistemological theory of Keith Lehrer.
I take Lehrer to be following in great detail a trail that was started by Plato, in a
beautiful passage in one of his earlier dialogues:

True opinions too are a fine thing and altogether good in their effects so
long as they stay with one, but they won't willingly stay long and instead
run away from a person's soul, so they're not worth much until one ties
them down by reasoning out the explanation .... And when they've been

tied down, then for one thing they become items of knowledge, and for
another, pennanent. And that's what makes knowledge more valuable
than right opinion, and the way knowledge differs from right opinion is
by being tied down. (Meno 97e-98a, Plato 1994, p. 69)

Most of the following considerations will not depend on the details of Lehrer's
theory; in fact I shall offer a few non-trivial improvements on some of his defi-
nitions. But I want to base my arguments on the overall architecture of Lehrer's
undertaking. If Lehrer were completely misguided, then what I say about the
relation between epistemology and the theory of belief revision might equally
well be mistaken.
Claim (ii) needs to be qualified as well. The analysis of belief revision is
not dependent on features that distinguish genuine knowledge from mere belief.
It is rather dependent on the structure and the fonnation of beliefs as they are
relevant in the theory of knowledge. What I have in mind above all is the funda-
mental distinction between foundations and coherence theories of knowledge.
This distinction happens to have come to the fore in the theory of knowledge,
but it may just as well be placed in a theory ofbelief. 4 It is primarily concerned
with the inferential relations between various beliefs, that is to say, with the
internal structure of our belief systems. The contrast lies in the answer to the
question whether there is such a thing as a belief base, that is, a distinguished
set of beliefs that are not in need of an inferential justification by other beliefs
and that taken together inferentially justify all the remaining ('derived') beliefs.
All of this can be dealt with in a general theory of belief; nothing requires to
refer this topic to the theory of knowledge. Claim (ii) can still be upheld, given
the fact that many relevant aspects of the structure of beliefs have as a matter of
fact come out most clearly in epistemological discussions.



In this section we unroll the central parts of Lehrer's theory of knowledge and
show how they are rooted in problems and questions which belong to the theory
of belief change. The following presentation is based on the summary in Lehrer
1990, pp. 147-149. The time parameter t which does not play any interesting
role in Lehrer's theory will be removed.
For a long time it has been thought in philosophy that knowledge is jus-
tified true belief. The short and famous article by Gettier published in 1963
has made it clear that this analysis is inadequate. Let us have a look at one of
Gettier's counterexamples.
Suppose that Smith has very strong evidence for

cp: Jones owns a Ford.

Suppose further that Smith is totally ignorant of Brown's whereabouts. Still he

can (and does) correctly infer from cp that

7/J : Either Jones owns a Ford, or Brown is in Boston.

X : Either Jones owns a Ford, or Brown is in Barcelona.

~: Either Jones owns a Ford, or Brown is in Brest-Litovsk.

By pure coincidence, and entirely unknown to Smith, Barcelona happens

to be the place where Brown actually is. However, Jones does not own a Ford.
Now, X is a true justified belief, since cp is a justified belief and and X may
be logically derived from cp, and the second disjunct of X is true. However, it
would be utterly counterintuitive to say that Smith knows that X is true, because
he believes that X is true 'for the wrong reasons'. He has just been lucky that
the right belief occurred to him. 5
Gettier's example has led many epistemologists to the conclusion that
knowledge is more than justified true belief. They do not discard this vener-
able definition of knowledge, but supplement it by a fourth clause. Lehrer's
suggestion is the following:
DK S knows that cp if and only if

(i) S accepts that cp,

(ii) it is true that cp,
(iii) S is completely justified in accepting that cp, and
(iv) S is completely justified in accepting that cp in a way that does not
depend on any false statement.

Although Lehrer (1990, pp. 10-11) separates acceptance from belief, I

do not think that it would make a big difference if we substituted belief for
acceptance in clause (i). 6 Similarly, clause (ii) is not very controversial in the
theory of knowledge. The essential parts of the definition are the last two clauses
which appeal to the problematic notions of justification and dependence. I have
substituted in clause (iv) a formulation taken from Lehrer 1990, p. 18, for the
formulation in his official summary in Lehrer 1990, p. 147. In the latter he
requires S to be completely justified in accepting that cp 'in a way that is not
defeated by any false statement'. This statement seems to be slightly screwed
up. What Lehrer means, I think, is that S is completely justified in accepting
that cp in a way that cannot be defeated by pointing out that the justification

relies on a false statement. The above phrasing of clause (iv) expresses this
more accurately than Lehrer's official formulation. 7
Lehrer further characterizes the concept of knowledge in a two-layered
strategy.8 First he develops the notion of personal justification which is based on
an agent's subjective acceptance system. In a second step the purely subjective
standpoint gets transcended by several operations on the agent's current belief
set that could help him to approximate the whole truth. 9

3.1. Personal Justification and the Comparative Reasonableness of


All of Lehrer's considerations are based on the notion of an acceptance system

which is defined as follows. \0 We slightly adapt the notation.
D1 A system X is an acceptance system of S if and only if X contains just
statements of the form, 'S accepts that cp', attributing to S just those
things that S accepts with the objective of accepting that cp if and only
if cpo
This definition could be amended a little by taking into account that on Lehrer's
account, there is only one acceptance system for each agent at a certain time.
So it seems that D1 should better start like that: 'A system X is the acceptance
system of S if and only if ... '
Lehrer is one of the main advocates of a coherence theory of know ledge.
According to this approach, all justification comes from coherence with a given
acceptance system X. There is no justification simpliciter, only justification on
the basis of some X.
D2 S is justified in accepting cp on the basis of system X of S if and only if
cp coheres with X of S.
What is needed now is of course an elucidation of 'coherence-with-a-
system'. Lehrer defines it in terms of the relations of competing, beating and
neutralizing of propositions.
D3 cp coheres with X of S if and only if all competitors of cp are beaten or
neutralized for S on X. 11
Interestingly, Lehrer does not seem to require that cp belongs to X for
D2 and D3. The function that takes an acceptance system X and yields back
the set of all sentences cohering with X, or equivalently, of all sentences S is
justified in accepting on the basis of X, may be interpreted as an inference
operation in the sense of Rott 2001.
D4 1/J competes with cp for S on X 12 if and only if it is more reasonable for S
to accept that cp on the assumption that 1/J is false than on the assumption

that 'IjJ is true, on the basis of X.

D5 cp beats 'IjJ for Son X13 if and only if'IjJ competes with cp for S on X,
and it is more reasonable for S to accept cp than to accept 'IjJ on X.
D6 X neutralizes 'IjJ as a competitor of cp for S on X if and only if'IjJ competes
with cp for S on X, but 'IjJ 1\ X does not compete with cp for S on X, and
it is as reasonable for S to accept 'IjJ 1\ X as to accept 'IjJ alone on X.
Intuitively, definitions D4 and D5 are not beyond reproach. Let us as-
sume, for instance, that it is extremely reasonable to accept that cp, and that
'IjJ weakens the reasonableness of accepting cp just a little bit, by a more or less
negligible degree. Should we say that 'IjJ 'competes with' cp? In what sense
would we speak of a competition? The proposition 'IjJ may be about quite an-
other subject matter than cp, so there cannot be much rivalry or conflict between
the two. And again, if it is still a little more reasonable to accept cp than to ac-
cept 'IjJ, should we say that cp 'beats' 'IjJ? Surely in the situation just described it
would make good sense to accept both cp and 'IjJ (assuming that it is reasonable
to accept 'IjJ in the first place), even if cp beats 'IjJ in Lehrer's sense. We are ready
to accept sentences that weaken each other a little, and we tend to connect such
sentences by 'although'. Ifwe say, 'They get along together very well, although
they have no interests in common', in symbols 'cp although 'IjJ', we do accept
both cp and 'IjJ. It does not matter that 'IjJ 'competes with' cp in Lehrer's sense,
and it does not matter whether cp 'beats' 'IjJ or not.
The last three definitions, D4-D6, all lead us to a comparative concept
'reasonable-to-accept', relativized to a given acceptance system X. Before turn-
ing to a brief discussion of that concept, we finish off the first, subjective part of
Lehrer's analysis of knowledge. For personal justification, it is just the agent's
current acceptance system which is the basis for judgements of coherence.
D7 S is personally justified in accepting that cp if and only if S is justified
in accepting that cp on the basis of the acceptance system of S.
Now of course everything hinges on what is meant by 'reasonable-to-
accept' -again, relativized to a given acceptance system. Surprisingly, Lehrer
does not say very much about this, but he considers it as an advantage that his
primitive term of reasonableness is open to many different interpretations. In
the few pages he devotes to the topic (Lehrer 1990, pp. 127-131), however, he
recommends to employ cognitive decision theory.14 Lehrer suggests to iden-
tify the degree of reasonableness of accepting a hypothesis cp with its expected
epistemic utility:

r(cp) p(cp) . Ut(cp) + p(...,cp). Uf(cp),


where p( cp) and p( ---,cp) = 1 - p( cp) are the probabilities of cp being true or false
respectively, and Ut( cp) and Ufi cp) are the positive or negative utilities of ac-
cepting the hypothesis cp, when cp is true or false, respectively. The utility Ut( cp)
is supposed to reflect the informativeness of cp, and possibly other virtues such
as cp's explanatory power, simplicity, or pragmatic value, and the advantage of
conserving existing beliefs (Lehrer 1990, p. 131).
It should be noted that this absolute degree of reasonableness of accept-
ing is not quite sufficient for what we need in order to understand the fore-
going definitions. In the definition of competition, Lehrer appeals to the rea-
sonableness of accepting cp on the assumption that?j; is true or false. What we
need, then, is something like expected conditional epistemic utilities r( cp!?j;) and
r(cp!---,?j;), and it is left unspecified how we can get them. There is no problem
with the well-known concept of conditional probabilities,15 but it is not quite
clear whether the utilities of accepting a hypothesis cp conditional on accepting
?j; or ---,?j; should be thought of as different from the plain, unconditional utility of
accepting cpo But if the utility of accepting a sentence is dependent on accepting
some other sentence, should not the utility of accepting a sentence be sensitive
to the context of acceptance, that is, to the acceptance system X as a whole?
This question leads us to a problem that is both more general and more
important. As indicated in definitions D4-D6, the reasonableness of accepting a
statement may be-and probably should be-relative to the acceptance system
of the agent. But the above definition of r( cp) does not reflect this. It is rather
an absolute measure of reasonableness. This is quite contrary to the coher-
entist's aim of evaluating systems of hypotheses rather than single hypotheses
taken in isolation. Lehrer can counter this objection by saying that the utility
functions Ut and Uf depend on the current acceptance system X. But then one
may ask whether it is illuminating to base an analysis of 'coherence of cp with
a system X' on an unexplained notion of 'utility of accepting cp on the basis of
Appealing to cognitive decision theory suggests that the acceptance of a
proposition is a matter of decision. Such an assumption is at least controversial.
But in this respect Lehrer's replacement of 'belief' by 'acceptance' is a prudent
move. It is certainly much more plausible to say that an agent decides to accept
something than that he decides to believe something. Another potential point
of criticism is that it need not be (objective? subjective?) probability that is
taken into account when the reasonableness of accepting certain hypotheses gets
assessed. Perhaps plausibility, a notion with different formal characteristics,
is an equally suitable candidate. This objection, too, could be countered, by
pointing out that decision theory is simply a theory based on probabilities. 17

3.2. Varying Imperatives for Coherence

The most serious questions for Lehrer's account seem to be the following. How
can we be sure that coherence in his sense, explained by means of a compli-
cated mechanism of competing, beating and neutralizing, which in turn is based
on a decision-theoretic criterion, will give us a coherent acceptance set? 18 If
every element of an acceptance set coheres with that very set (as it presumably
should), can we be sure that the acceptance set is consistent? 19 Will the accep-
tance set be closed under logical consequences? Both consistency and closure
are themselves requirements of-inferential-coherence. If all of Lehrer's cen-
tral definitions are finally based on a numerical degree of reasonableness, why
not take that very same degree as the sole arbiter of acceptance, without the
taxing detour via definitions D2-D7?
All these questions call for a fresh and systematic look at the principles
that are involved in Lehrer's concept of coherence.
First and foremost, we have coherence according to Lehrer's own the-
ory. The corresponding imperative is this: Accept precisely those sentences the
competitors ofwhich are beaten or neutralized!
Second, we have seen that Lehrer's theory is grafted on top of cognitive
decision theory which of course brings along its own standards of coherence.
Lehrer does not address this point. An attempt to make it explicit reveals that
there are several ways to go. The following idea seems initially plausible: Ac-
cept those sentences that promise the greatest expected cognitive utility! A mo-
ment's reflection, however, shows that this idea is premature. Why should we
reject sentences of positive (or at least non-negative) expected utility for the sole
reason that there are still more useful ones? Don't all sentences with positive
r-values contribute to the overall expected utility? Thus the right imperative
seems to be this: Accept all sentences with positive (or non-negative) expected
cognitive utilitypo But perhaps this is again mistaken. It may be wrong to
suppose that the utilities of individual sentences simply add up to yield the util-
ity of the whole body of beliefs. This would mean that we cannot rely on the
above imperatives, because conditional utilities (see above) may be very differ-
ent from the unconditional ones. For instance, I may entertain two alternative
hypotheses, both with high expected epistemic utility, which contradict each
other. Accepting either one of them seems reasonable, but accepting both of
them would lead to inconsistency which is not particularly useful. Thus one
should take into account interactions between the individual beliefs, and regard
beliefs as constituting a system that may be assessed only holistically. This is
very much in the spirit of coherentists anyway, who argue that only the corpus
of belief or knowledge taken as a whole is a proper unit of epistemic appraisal.
The degree r of reasonableness of acceptance then must not be applied to in-

dividual sentences, but to sets of sentences, and the corresponding imperative

reads thus: Accept those sets of sentences that promise the greatest expected
cognitive utility!
A third concept of coherence is the logical or inferential one. I shall
discuss its implications in detail later, but it is expedient to anticipate the main
points already here. A set of sentences is inferentially coherent if it is consistent
and closed under consequences. The corresponding imperative is: Accept all the
logical consequences of what you accept, but avoid accepting contradictions!
In Rott 2001, Chapters 7 and 8, it is shown that the second and the third
concepts of coherence are compatible with one another. But even if one is ready
to grant Lehrer's mechanics of justification in the sense of his definitions D3-D6
some plausibility in itself, it is doubtful whether it can be made to cohere with
other concepts of coherence. One has to face the fact that different coherence
criteria may conflict with one another, and decide which of these criteria are
the most justified ones. As of 1990, Lehrer's theory is a hybrid of at least three
different intuitions.

3.3. Undefeated Justification and Justification Games

In order to arrive at knowledge one has to go beyond the actual acceptance

system of an individual agent. In a sense very close to the passage of Plato's
Meno quoted in Section 2 above, knowledge must be stable under criticism.
In Lehrer's 'justification games' the part of the critic is taken over by an om-
niscient 'sceptic'. Let us now have a look at the formal definitions that take
Lehrer from merely personal justification to complete and indeed indefeasible
D8 A system V is a verific system of S if and only if V is a subsystem of
the acceptance system of S resulting from eliminating all statements of
the form, 'S accepts that cp', when cp is false.
As in the case of definition D 1, it would be preferable to say that the V
mentioned in D8 is the verific system of S, since it results from the unique
acceptance system of S (at a given time) by just cutting out the false beliefs.
The verific system is the basis for verific and complete justification:
D9 S is verifically justified in accepting that cp if and only if S is justified in
accepting that cp on the basis of the verific system of S.
D lOS is completely justified in accepting that cp if and only if S is personally
justified in accepting cp and S is verifically justified in accepting cpo
Complete justification in this sense, however, does not solve the Gettier
problem. Smith's belief that Jones owns a Ford need not depend on any false

belief. Let us suppose that Jones told Smith that he has a Ford, showed him
papers stating that he, Jones, owns a Ford, and always drives a Ford on his
way from his home to his office. All this is believed and known (in some pre-
theoretical sense) by Smith, and justifies his belief that Smith does in fact own
a Ford. So the reason for Smith's not knowing that sentence X above is true
is not that he accepts false sentences, but rather that he is not aware of all true
sentences that are relevant to the case. Notice that it is not enough to know some
relevant facts since a biased selection of true facts may be utterly misleading
and tum the agent away from some other truths. Lehrer suggests to solve this
difficulty be looking at what he calls the 'ultrasystem' of an agent. Here are his
D 11 S is justified in accepting that cp in a way that is undefeated if and only
if S is justified in accepting cp on the basis of every system that is a
member of the ultrasystem of S.
D12 A system M is a member of the ultrasystem of S if and only if either
M is the acceptance system of S or results from

- eliminating one or more statements of the form, 'S accepts that 1/)' ,
when 'ljJ is false,
- replacing one or more statements of the form, 'S accepts that 1jJ' ,
with a statement of the form'S accepts that not 'ljJ', when .1jJ is
- or any combination of such eliminations and replacements in the
acceptance system of S,

with the constraint that if'ljJ logically entails X which is false and also
accepted, then'S accepts that X' must also be eliminated or replaced
just as'S accepts that 'ljJ' was.
Before moving on to the criticism of Lehrer's, we should mention his key
result: 'Knowledge reduces to undefeated justification, a just reward for our ar-
duous analytical efforts.' (Lehrer 1990, p. 149) Clearly, undefeated justification
implies personal justification: the actual acceptance system of S is a member
of the ultrasystem; it also implies verific justification: the sceptic can make S
eliminate all his false beliefs, thus effecting a transition to the agent's verific
system;21 finally, it also implies truth: if cp were false, the sceptic could make S
eliminate cpo
Lehrer's theory of knowledge can be called a dynamic one because we
can think of the repeated operations mentioned in D 12 as (potential) steps in a
journey through the space of belief states.

One may wonder why Lehrer requires for the undefeated justification
of ip that every member of the ultrasystem must support ip. It would seem suffi-
cient that S is ultimately justified. By this we mean that for every member M
of the ultrasystem of S there is another member M' of the ultrasystem of S
which improves on M and on the basis of which S is justified in accepting ip.
That NI' improves on M means, of course, that M' can be reached from M
by some combination of 'truth-conducive' eliminations and replacements of the
kind specified in definition D12. This concept seems more adequate since even
if S knows that ip, pre-theoretically understood, a mischievous sceptic may well
advance an impressive battery of true facts speaking against the truth of ip, so
that S looses his confidence that ip is true. Only later in his conversation with
the omniscient sceptic, when S comes to know more about the truth, will he
regain his old true and justified belief. Although the correct belief would be
dropped on the receipt of true but misleading information, we may consider it
to constitute knowledge, since one can later learn that this information has in
fact been misleading. Temporary doubts about ip should not count, so it seems,
as long as all potential paths of the ultra justification game finally lead to ip'S ac-
ceptance. 22 Lehrer himself seems to agree with that when discussing his Grabit
example: 23
Suppose I see a man, Tom Grabit, with whom I am acquainted and have
seen often before, standing a few yards from me in the library. I observe
him take a book off the shelf and leave the library. I am justified in
accepting that Tom Grabit took a book, and, assuming he did take it, I
know that he did. Imagine, however, that Tom Grabit's father has, quite
unknown to me, told someone that Tom was not in town today, but his
identical twin brother, John, who he himself often confuses with Tom, is
in town at the library getting a book. Had I known that Tom's father said
this, I would not have been justified in accepting that I saw Tom Grabit
take the book, for if Mr. Grabit confuses Tom for John, as he says, then I
might surely have done so, too. (Lehrer 1990, p. 139)

Lehrer summarizes the lesson to be drawn from this example as follows:

'I may be said to know that Tom Grabit took the book despite the fact that, had
I known what his father said without knowing about his [the father's] madness,
I would not know whether it was Tom who took it.'
In contrast to Lehrer's undefeated justification, ultimate justification no
longer implies that S is justified on the basis of his current personal or verific
acceptance systems. But what we have been looking for is an objectified no-
tion of justification, which can be conjunctively added to the (at least partially)
subjective notions of personal and verific justification. In particular, we cannot
dispense with verific justification if we want to end up with the right analysis
of Gettier-type examples. In the example discussed on page 222, Smith is both

personally and ultimately justified in believing that either Jones owns a Ford or
Brown is in Barcelona (0. What prevents him from knowing this is that his
belief that Jones owns a Ford is false, so he is not verifically justified to believe
that ~.24 Ultimate justification is not sufficient for knowledge.
Since every theory can be improved by a transition to the one true and
complete theory about the world,25 ultimate justification reduces to justification
on the basis of that theory. On the one hand, it does not seem objectionable to
call for that theory as the final arbiter of knowledge. On the other hand, I cannot
see why the whole truth must be coherent, in Lehrer's or in any other but the
purely logical sense. Shouldn't we try to avoid stipulating that the one true and
complete theory is coherent, because that would mean basing epistemology on
a questionable metaphysics? Similarly, I do not see any intuitive reason why
every truth should be justified, on the basis of the true and complete theory. If
this is right, then it is impossible, by Lehrer's own definition of knowledge as
well as by the definition using ultimate justification, that an agent will know
the whole truth. Shouldn't we try to avoid this conclusion? This suggests that
undefeated justification may not be necessary for knowledge, and even ultimate
justification may not be.
But let us stop with these cosmic speculations now and return to more
definite matters again. In Lehrer 2000, especially pp. 153-154, 168-169, there
are no replacements any more, and there is no talk of strong corrections. In this
new account, the ultrasystem is closer in essence to what was called the verific
system earlier,26 and undefeated justification is similar to verific justification,
i.e., justification on what remains when everything false is eliminated from the
person's acceptance system.
Unfortunately, Lehrer does not tell the reader why he has changed his
earlier definitions and given up on the idea that not only eliminations, but re-
placements, too, may be prompted by the sceptic. It is not clear whether he
just considers it as a simplification of his former account or whether he thinks
that the new edition of his book actually corrects an inadequacy of his former
account. I presume that the reason lies in the problem of misleading informa-
tion. In cases like the Grabit example, receipt of information about what Tom
Grabit's father said (without information about the father's mental state) would
probably have done away my acceptance that Tom Grabit stole the book, even
though this seems to be a bit of genuine knowledge. So it appears that accord-
ing to Lehrer, we should not admit replacements or additions to our stock of
accepted propositions when testing for knowledge. This is an interesting ar-
gument, but its validity may well be doubted. Misleading effects cannot only
be achieved by adding truths but also by removing falsehoods. There are other
cases of a similar structure where in fact no genuine knowledge seems to be in-

volved. Lehrer (2000, p. 160) discusses a 'newspaper example' originally due to

Harman (1973). I do not think, however, that the distinguishing criterion offered
does the job that it has been assigned by Lehrer. In the newspaper example, the
subject's justification is said to depend on "unstated" facts about the newspa-
per's trustworthiness. But then, why are we not to assume, for instance, that
the justification of our belief that Grabit stole the book depends on the unstated
fact that no-one has offered any evidence about look-alike suspects? Ifwe can't
rule out this, we don't seem to know that Grabit stole the book, according to
Lehrer's definition, and the strategy of using only eliminations, rather than both
eliminations and replacements, does not help.
In sum, then, I cannot see that there is an epistemologically significant
difference between the sceptic's removing errors and his supplying new truthful
information. Sometimes misleading evidence that one does not possess may
block the claim to knowledge, just as wrong beliefs that one does possess may
do.27 Whenever the omniscient sceptic succeeds in making us abandon a belief
as a result of an improvement of our belief set, this is strong indication that the
belief has not been a piece of knowledge. Knowledge, so it seems, should be
stable in any kind of critical, truth-directed dialogue. For this reason I will stick
to the richer 1990 definition of ultrasystems.



We have seen that eliminations and replacements of accepted propositions are

of paramount importance for Lehrer's approach to the theory of knowledge.
Eliminations and replacements are respectively called 'weak corrections' and
'strong corrections' in Lehrer 1997, pp. 45-49. These operations are very close
to the operations of contraction and revision as they are known in the theory of
belief revision. 28 As Lehrer places no constraints on the structure of acceptance
systems (see Definition D1), there seems to be no need for him to apply non-
trivial change operations to acceptance systems. 29 May we not just eliminate a
false sentence 1jJ by simply dropping the statement' 5 accepts that 1jJ' from the
system X, and similarly, may we not simply substitute the statement' 5 accepts
that -,1jJ' for the statement '5 accepts that 1jJ' in X, when a false sentence 1jJ is
to be replaced by its negation?
The answer is, 'No'. Effortless eliminations and replacements are ex-
cluded by the logical constraint stated in Lehrer's definition D 12: 'if 1jJ logically
entails X which is false and also accepted, then "5 accepts that X" must also be
eliminated or replaced just as "5 accepts that 1jJ" was.,30

These are the only constraints Lehrer enters into his official definitions,
but in the running text he acknowledges more constraints of a similar kind.
They are in a sense complementary to the ones we just mentioned. Let 7/J again
be the false sentence to be eliminated or replaced by its negation -,7/J. While the
first group of constraints concerns (false) sentences implied by 7/J, the second
group deals with (necessarily false) sentences implying 7/J. The first group of
constraints is forward-looking, the second group is backward-looking. Here is
the quotation from Lehrer 1990, p. 141:

The sceptic ... may require the claimant to eliminate anything the claim-
ant accepts that is false, and the claimant must eliminate the specified
item from his acceptance system and at the same time eliminate anything
he accepts that 10gicaIIy implies the eliminated item. Or the sceptic may
require the claimant to replace anything the claimant accepts that is false
with the acceptance of its denial and at the same time replace anything
that 10gicaIIy implies the replaced item with acceptance of its denial. [My

In order to make the discussion of Lehrer's logical constraints more eas-

ily surveyable, I will now give shorter, semi-formalized formulations.
(FE) Forward Elimination. If 7/J is to be eliminated and 7/J f- X, then item X,
if false, must be eliminated.
(FR) Forward Replacement. If 7/J is to be replaced by -'7/J and 7/J f- X, then
item X, iffalse, must be replaced by -'X.
(BE) Backward Elimination. If 7/J is to be eliminated and X f- 7/J, then
item X (which is false) must be eliminated.
(BE) Backward Replacement. If 7/J is to be replaced by -'7/J and X f- 7/J, then
item X (which is false) must be replaced by -,x.
I have not mentioned here that only X's that are accepted should possibly
be eliminated or replaced. This should be self-evident.
I am not going to discuss the falsity conditions in (FE) and (FR), which
presume an impartial, objective, omniscient supervisor for the eliminations and
replacements in the 'ultra justification game'. I want to reflect on the logical
constraints only in so far as they are accessible to the agent himself. In doing
so, I want to avoid the assumption that the agent is or should be omniscient;
however, I do embrace the idealizing assumption that he or she is capable of
drawing all and only valid logical inferences that can be drawn on the basis of
their acceptance systems.
The point I want to make is that Lehrer's logical constraints are too sim-
plistic. Being based on consequence relations between single sentences, they in

effect consider the accepted items in isolation rather than as items in an accep-
tance system. The constraints do not really address the logical coherence of a
belief with all its surrounding beliefs. It is important to take into account the
context of the remaining accepted items when formulating logical constraints
for eliminations and replacements. Without any claim that these are 'the right'
constraints, the following ones are certainly more adequate in that they show
some sensitivity to the context in which beliefs are situated. We keep on using
the variable '1jJ' for the false sentence that is to be eliminated or replaced by its
negation, and give both a formulation that is close to Lehrer's own statements
and a slightly more formalized version.

(FE-)/(FR -) Forward Elimination/Forward Replacement.

If 1jJ is an essential premise for the logical derivation of X and X is false,
then'S accepts that X' must also be eliminated. In symbols:
If 1jJ is to be eliminated or replaced by -,1jJ, and F U {1jJ} I- X for some
set F of accepted items which are not eliminated, but F }L X for any
such F, then item X (if false) must be eliminated.
(FR +) Forward Replacement.
If the addition of -,1jJ creates a new implication of ~ (and ~ is true) then
'S accepts that t must be added to the acceptance system. In sym-
If 1jJ is to be replaced by -,1jJ, and F U {-,1jJ} I- ~ for some set F of accepted
items which are not eliminated, then item ~ (if true) must be added, if
it is not already accepted.
(BE-)l(BR -) Backward Elimination/Backward Replacement.
If F is a set of accepted premises that logically implies the eliminated or re-
placed item 1jJ, then for at least one (false) member X of F, 'S accepts
that X' must be eliminated. In symbols:
If 1jJ is to be eliminated or replaced by -,1jJ, and F I- 1jJ for some set F of
items, then at least one member of F (which is false) must be elimi-

These conditions correct a number of counterintuitive features of Lehrer's

constraints. In the case of a replacement of 1jJ by -,1jJ, there is no reason to
replace by their negations all the X's that are either critically implying 1jJ or
essentially implied by 1jJ. It is enough that such X's are eliminated. Let us
temporarily employ the following concepts of critical and essential implication.
A sentence X critically implies another sentence 1jJ in the context of a set F,
if F implies 1jJ, F \ {X} does not imply 1jJ, and X is one of 'the weakest' or

'most vulnerable' elements of F. A sentence X is essentially implied by another

sentence 1/J in the context of a set F, if F implies X, but F \ {1/J} does not
imply X. Critically implying and essentially implied X's are to be eliminated
when 1/J is just eliminated without being replaced by its negation. And in this
respect, there is no reason at all to think that replacements present problems any
different from those presented by contractions.
However, a replacement occasions one kind of adjustment of the accep-
tance system which cannot be initiated by an elimination. Since there is an
addition of -'1/J to the old acceptance system, it is reasonable to require that any
new derivations that are made possible by this new item should in fact be made,
and the results be added to the acceptance system. This is condition (FR+)
which has no counterpart in Lehrer's definitions.
The most important respect in which the above constraints improve upon
Lehrer's constraints is that they pay attention to the fact that 1/J or any of the X's
mentioned are part of a system. The F's mentioned in the constraints repre-
sent the contexts in which the respective tests for implications have to be made.
The contexts are subsystems of the original acceptance system of the agent. By
making the contexts explicit, we raise important new questions: Which items
may be included in those F's that figure in the forward-looking constraints?
Which items should be excluded from those F's that figure in the backward-
looking constraint? More generally, how do we know which items in an accep-
tance system survive the process of elimination or replacement? Lehrer should
give answers to these questions if his theory of knowledge is to be considered
complete, but he fails to do so. It is precisely the theory of belief change that
addresses these questions and offers a variety of ways to answer them.
The constraints (BE-) and (BR-), which concern sets'F that imply the
false belief 1/J that has to be given up or replaced, leave much room for choices.
It requires the agent to give up at least one (false) belief in F, but it does not
tell us which belief or beliefs ought to be given up. The constraint does not
fully determine what to do, but leaves us the freedom to choose. But how are
the choices which members to eliminate from the acceptance set to be made?
Even if one cannot enumerate concretely the cognitive values that govern our
decisions what to give up, it is possible to inquire into the logic of 'coherent' or
'rational' choices involved in belief change (Rott 2001),
A choice-theoretic perspective has already come up in Lehrer's sugges-
tion to explicate the notion of personal justification. We saw that he ends up with
a decision-theoretic explication ofthe comparative notion of reasonableness-of-
acceptance. According to this approach, it is more reasonable to accept 'P than
to accept 1/J just in case 'P has greater expected epistemic utility than 1/J, But this
is a comparison between two single items of belief only, and it is not clear how

it can be extended to a criterion for assessing whole systems of belief. In fact it

is not at all evident that the logical constraints that Lehrer mentions on his way
from personal to undefeated justification mix well with the decision-theoretic
advice given in the case of personal justifications. It has to be shown that a
combination of logical and choice-theoretic constraints is indeed possible. In
Rott 2001, I have tried to show this independently of Lehrer's particular idea
that go for maximal or at least non-negative expected epistemic utility. Using
choice-theoretic methods, one can maintain the idea that acceptance systems re-
sulting after some change ought to be inferentially coherent, that is, consistent
and closed with respect to a given logic. Choice and logic can indeed coexist
in harmonyY It is unclear whether Lehrer's coherence mechanics in terms of
competing, beating and neutralizing is compatible with such an account, but
what seems to be clear is that both logic and the theories of choice and decision
are better motivated and more principled than Lehrer's coherence mechanics. In
case of conflict, they might therefore take priority.



In this paper, I have used concepts and ideas borrowed from belief revision the-
ory to elaborate on an important contemporary account in the theory of knowl-
edge. But the symbiosis between the two research areas may equally well be
viewed from the opposite perspective. In this concluding section, I want to give
an indication of how concepts and ideas developed in epistemology can help to
structure and interpret much of the work that has been done in belief revision
One of the most relevant distinctions for belief revision is that between
foundationalist and coherentist approaches in epistemology. Lehrer (1990,
p. 13) characterizes the fundamental difference between foundationalist and co-
herentist views of knowledge as follows.
According to foundationalists, knowledge and justification are based on
some sort of foundation, the first premises of justification. These prem-
ises provide us with basic beliefs that are justified in themselves, or self-
justified beliefs, upon which the justification for all other beliefs rests.
Coherentists argue that justification must be distinguished from argu-
mentation and reasoning. For them, there need not be any basic beliefs
because all beliefs may be justified by their relation to others by mutual
support. [My italics, HR]

Ernest Sosa (1980, pp. 23-24) makes essentially the same point in more meta-
phorical terms:

For the foundationalist, every piece of knowledge stands at the apex of a

pyramid that rests on stable and secure foundations whose stability and
security does not derive from the upper stories or sections.
For the coherentist a body ofknowledge is a free-floating raft every plank
of which helps directly or indirectly to keep all the others in place, and
no plank of which would retain its status with no help from the others.
[My italics]

It is important to see that the categorical distinction between foundationalism

and coherentism can more properly be applied to theories of belief than to theo-
ries of knowledge, since no reference is made in the drawing of this distinction
to the question of whether the beliefs in question are actually true. 32 There is
also a second, more 'dynamic' sense in which the distinction becomes relevant,
and that is when it comes to the modelling of the dynamics of belief. 33 So there
are two questions that separate foundationalists from coherentists.

(1) Can a distinction between basic beliefs and derived beliefs be validly

(2) And if so, are changes of beliefs made primarily on the base level or on
the level of 'coherent theories'?

It is clear what the foundationalist's and the coherentist's answers will look like.
The former af'finns while the latter denies the first question. In response to the
second question, the former would say' on the base level', while the latter, lack-
ing a distinguished base level, must opt for the level of coherent theories. It is
necessary to work out in greater detail the concepts and distinctions on which
these answers are based. In Rott 2001 I have tried to provide a framework that
helps us to understand the issues involved and to characterize two fundamen-
tally different perspectives on the process of belief revision. These perspectives
turned out to be related to, but not identical with, the dichotomy between foun-
dationalism and coherentism. I did not go as far, however, as Hansson and Ols-
son (1999) who argue that the coherence theory in the epistemologist's sense is
trivialized in the context of the coherence perspective on belief revision.
When transferring the idea of 'foundations' of knowledge to the area of
belief change, it is not particularly important whether the basic beliefs are true,
let alone infallibly true. Neither is it important that they are justified to such
a high degree that they may be regarded as certain. In the current theories of
belief change, belief bases are not supposed to carry any of these connotations.
Basic beliefs are distinguished from derived beliefs only by the fact that they
are somehow 'given', either explicitly or as things that are taken for granted.
Givenness is not at all supposed to imply indefeasibility here. Still, as I tried to

show in Rott 2001, these categories coming from epistemology can be exploited
for an illuminating analysis of the dynamics of belief.

Acknowledgements. This paper is a revised version of the second chapter of

Rott 2001. I want to thank Gordian Haas, Keith Lehrer, Erik Olsson, Wolfgang
Spohn and all the other participants of the 2000 Konstanz Workshop on Lehrer's
epistemology for many helpful comments, even though they will certainly be
dissatisfied with what I made out of them.


1 Cf. the remarks made at the beginning of Section 4 below.

2 See, for instance, Lehrer 1990, Plantinga 1990 or Sosa 1980.
3 For the exciting ethics of belief debate that took place in the nineteenth century, see the anthol-
ogy edited by McCarthy (1986).
4 Wolfgang Spohn has pointed out to me that this point is fully explicit in Bonjour 1985.
5 Actually luck is not even needed. If we do not require that a belief manifests itself in an
occurrent event in the believer's mind, but may reside hidden in an implicit theory of his or
hers, we may assume that the beliefs of an agent are logically closed. Then any justified false
belief cp gives rise to an infinite set of justified true beliefs-all disjunctions cp V cpt where cpt is
any arbitrary truth. So Smith is not just lucky, but his false belief that Jones owns a Ford will
automatically generate infinitely many justified true beliefs like x.
6 According to Lehrer, we sometimes believe cp for the sake of felicity or the pleasure of believing
so, but we would not accept cp for these reasons. Lehrer says that there is always a potential
conflict between the 'ancient system of perceptual belief' which is the 'yield of habit, instinct,
and need' and the 'truth-seeking ... scientific system of acceptance'. While the former is an
'automatic input system', the latter is 'capable of ratiocination' (Lehrer 1990, pp. 113-114). In
this paper, I work on the simplifYing assumption that all doxastic attitudes are 'aiming at' truth:
If an agent either believes or accepts that cp, this means that he holds cp true.
7 Lehrer (1990, p. 138) apparently thinks that the two formulations are 'equivalent'.
8 Compare in particular Lehrer 1990, pp. 141-152, and 1997,28-45.
9 These operations are carried out only hypothetically, 'for the sake of argument' in fictitious test
dialogues with an omniscient sceptic. In reality, the agent's state of mind remains unchanged.
10 This is the term used in Lehrer 1990; in Lehrer 1997, pp. 25-29, and Lehrer 2000, the leading
part is taken by the 'evaluation system' which includes not only a person's accepted propositions
(geared to truth) but also her preferred propositions (geared to merit).
11 Definition 03 on p. 148 of Lehrer 1990 actually starts as follows: '5 is justified in accepting cp
on the basis of system X of 5 if and only if ... ' From definition 02 and the surrounding text,
however, it is obvious that this is a misprint and the formulation given above is the intended one.
12 Some renaming of Lehrer 2000: ''ljJ is an objection to cp for 5 on X'.
13 Lehrer 2000: 'objection'ljJ to cp is answered for 5 on X'.
14 The most eminent advocate of cognitive decision theory is Isaac Levi (1967, 1984). For a
critical voice, see Weintraub (1990). In Lehrer 1997, in particular pp. 30-35, the contribution of
cognitive decision theory to personal justification has vanished and its role is taken over by the
principle of trustworthiness, a kind of rationality principle that is geared solely to the acquisition
of truth. Cognitive decision theory is still present, however, in the second edition of Theory of
Knowledge (Lehrer 2000, pp. 144-148).

15 Let us assume that p( 1j;) and p( ~1j;) are positive, so that conditionalizing by either 1j; or ~1j; is
not beset with the problem of an ill-defined division by zero.
16 In his brief discussion of expected epistemic utilities, Lehrer shows little awareness of the fact
that Ut and Uf mayor should depend on X, if reasonableness of acceptance is to be relative to the
current acceptance system. Lehrer contrasts Ut( 'P) with 'P's (objective or subjective) probability
and links it to 'P's truth, but not to the acceptance of other sentences. He does not say anything
about Ul
17 Theories of plausibility of the kind I have in mind are offered, amongst others, by Grove
(1988), Rescher (1976), Shackle (1961), and Spohn (1988). They would have to be supplemented
by a qualitative decision theory.
18 The distinction between relational coherence (coherence as a relation) and systemic coherence
(coherence as a property of a system) is discussed in connection with Lehrer's theory by Olsson
19 This problem for Lehrer's theory has also been treated by Olsson (1998).
20 This precept is plausible only if one assumes, as Lehrer apparently does, that the expected
utility of rejecting a hypothesis is zero. An alternative idea would be to accept just those proposi-
tions 'P which have a degree of reasonableness that exceeds a contextually fixed threshold value.
21 We neglect the possibility that the removal of a false belief may tear along some true beliefs.

22 Could there be an eternal wiggling of the acceptance value of'P in response to the sceptic's

challenges? Not if we neglect the possibility that a truth-conducive change can make 5 drop
truths (as we decided to do in footnote 21) and if we set aside questions of infinity. In such a
context, the sceptic has the means to make 5 accept the true and complete theory about the world
which, of course, cannot be further improved.
23 A similar example about barns and papier-mache facsimiles originally due to Carl Ginet is
discussed in Nozick 1981, pp. 174-175, and Bach 1984, pp. 40--41.
24 This argument depends on the assumption that if ~ is not part of an acceptance system, then
it cannot be justified on the basis of that system. Strictly speaking, Lehrer does not make that
assumption; compare Lehrer's definitions D2-D6 and especially my comment on D3 above.
25 Saying this actually steps beyond Lehrer's account, which does not provide for the possibil-
ity of knowledge expansion through the sceptic-which marks an important difference between
Lehrer's concept of replacements and the usual understanding of the concept of revision. The
uniqueness involved in talking about 'the one true and complete theory about the world' is of
course relative to the language used, which I assume as given.
26 I neglect here, perhaps uncharitably, the fact that Lehrer's second edition uses richer 'evalua-
tion systems' instead of the 'acceptance systems' of the first edition. The presentation of Lehrer
2000 follows that of Lehrer 1990 more closely than is warranted by its contents. While the first
edition has the ultrasystem as a genuinely new and complex system, in the second edition the
ultra system is nothing more than the pair consisting of the original system and the verific system.
Introducing the term 'ultrasystem' for this entity seems a bit pompuous, but it is just a reflection
of how the book came into being.
27 This is reminiscent of the idea that the justification for 'P does not only consist in the presence
of reasons for 'P, but also in the absence of reasons against 'P. The point is given pride of place in
nonmonotonic reasoning in the tradition of Doyle (1979).
28 I have mentioned an important difference between 'replacements' and 'revisions' in foot-
note 25: replacements substitute ~'P for some previous belief 'P, while revisions include expan-
sions of belief sets by sentences about which there had not been any opinion before.
29 If this were true, then Lehrer would tum out to be a foundationalist (at least, a foundationalist
in the sense of Chapter 3 of Rott 200 I).
30 My italics. Compare footnote lion p. 194 of Lehrer 1990: ' ... with the constraint that

if'lj; logically entails X, which is false and also accepted, then "3 accepts that X" must also
be eliminated or replaced in the same way as "3 accepts that 'Ij;" was.' (Again, my italics) I
am reading the phrases "in the same way as" (in Lehrer's footnote I I) and 'just as" (in D12)
as indicating that a replacement of 'Ij; by its negation should enforce a replacement of X by its
31 Another fundamental idea which is not mentioned in Lehrer's account is that doxastic changes
should be conservative, i.e., that they should incur only minimal changes to the previous accep-
tance system. On this idea, and its role in belief change theories, compare Rott 2000.
32 As Lehrer realizes very clearly, this objective question has to be linked to the subjective ques-
tion of coherence in a separate step.
33 There has been an ongoing controversy over the coherentism-vs.-foundationalism issue in the
belief revision literature for more than a decade, see Nebel 1989, Giirdenfors 1990, Doyle 1992,
Nayak 1994, del Val 1994, del Val 1997, Hansson and Olsson 1999, Bochman 2000, Bochman
2001 and Rott 2001. Particularly influential as a mediator between epistemology and belief
change was Harman (I 986).

Bach, Kent. 1984. "Default Reasoning: Jumping to Conclusions and Knowing When to Think
Twice." Pacific Philosophical Quarterly 65:37-58.
Bochman, Alexander. 2000. "A Foundationalist View of the AGM Theory of Belief Change."
Artificial Intelligence 116:237-263.
---.2001. A Logical Theory ofNon monotonic Inference and BeliefChange. Springer Verlag.
Bonjour, Laurence. 1985. The Structure of Empirical Knowledge. Cambridge, Mass.: Harvard
University Press.
del Val, Alvaro. 1994. "On the Relation Between the Coherence and Foundations Theories of
Belief Revision." AAAI'94 - Proceedings of the National Conference of the American
Association on Artificial Intelligence, pp. 909-914.
- - - . 1997. "Non-Monotonic Reasoning and Belief Revision: Syntactic, Semantic, Founda-
tional, and Coherence Approaches." Journal ofApplied Non-Classical Logics 7:213-240.
Doyle, Jon. 1979. "A Truth Maintenance System." Artificial Intelligence 12:231-272.
- - - . 1992. "Reason Maintenance and Belief Revision: Foundations vs. Coherence Theo-
ries." In Peter Gardenfors (ed.), BeliefRevision, Cambridge University Press, Cambridge,
Gardenfors, Peter. 1990. "The Dynamics of Belief Systems: Foundations vs. Coherence Theo-
ries." Revue Internationale de Philosophie 44:24-46.
Grove, Adam. 1988. "Two Modellings for Theory Change." Journal of Philosophical Logic
17: 157-170.
Hansson, Sven 0., and Erik 1. Olsson. 1999. "Providing Foundations for Coherentism." Erkennt-
nis 51:243-265.
Harman, Gilbert. 1973. Thought. Princeton: Princeton University Press.
- - - . 1986. Change in View. Cambridge, Mass.: Bradford Books, MIT Press.
Lehrer, Keith. 1990. Theory of Knowledge. London: Routledge.
- - - . 1997. Self-Trust: A Study of Reason, Knowledge, and Autonomy. Oxford: Oxford
University Press.
- - - . 2000. Theory of Knowledge. Second, substantially revised edition. Boulder: Westview
Levi, Isaac. 1967. Gambling with Truth. New York: Knopf.
- - - . 1984. Decisions and Revisions: Philosophical Essays on Knowledge and Value. Cam-
bridge: Cambridge University Press.
McCarthy, Gerald D. (ed.). 1986. The Ethics of BeliefDebate. Atlanta: Atlanta Scholars Press.
Nayak, Abhaya C. 1994. "Foundational Belief Change." Journal ofPhilosophical Logic 23:495-
Nebel, Bernhard. 1989. "A Knowledge Level Analysis of Belief Revision." In Ronald Brachman,
Hector Levesque and Ray Reiter (eds.), Proceedings of the 1st International Conference
on Principles ofKnowledge Representation and Reasoning, Morgan Kaufmann, San Ma-
teo, Cal., pp. 301-311.
Nozick, Robert. 1981. Philosophical Explanations. Cambridge, Mass.: Belknap Press, Harvard
University Press.
Olsson, Erik J. 1998. "Competing for Acceptance: Lehrer's Rule and the Paradoxes of Justifica-
tion." Theoria 64:34-54 .
..- ...~. 1999. "Cohering With." Erkenntnis 50:273-291.
Plantinga, Alvin. 1990. "Justification in the 20th Century." Philosophy and Phenomenological
Research 50, Supplement Vol., 45-71.
Plato. 1994. Meno. In Jane M. Day (transl. and ed.), Plato's Meno in Focus, Routledge, London.

Rescher, Nicholas. 1976. Plausible Reasoning. Amsterdam: Van Gorcum, Assen.

Rott, Hans. 2000. "Two Dogmas of Belief Revision." Journal o/Philosophy 97:503-522.
- - - . 200 I. Change, Choice and Inference. Oxford: Oxford University Press.
Shackle, G. L. S. 1961. Decision, Order and Time in Human Aflairs. Cambridge: Cambridge
University Press.
Sosa, Ernest. 1980. "The Raft und the Pyramid: Coherence Versus Foundations in the The-
ory of Knowledge." Midwest Studies in Philosophy 5:3-25. Reprinted in Ernest Sosa,
Knowledge in Perspective, Cambridge University Press, Cambridge 1991, pp. 165-191.
Spohn, Wolfgang. 1988. "Ordinal Conditional Functions." In William L. Harper and Brian
Skynns (eds.), Causation in Decision, Belief Change, and Statistics, Vol. II, Reidel, Dor-
drecht, pp. 105-134.
Weintraub, Ruth. 1990. "Decision-Theoretic Epistemology." Synthese 83:159-177.
Chapter 14

Gordian Haas
Universitat Konstanz

According to the analysis of knowledge proposed by Lehrer, knowledge
equals undefeated justified acceptance. Undefeated justification is then spelled
out as coherence with every element of the ultrasystem, as Lehrer calls it. The
question arises how this key-notion should be defined. Two definitions of the
ultrasystem which have been proposed by Lehrer are investigated. An
argument is presented that both definitions are flawed. A formal proof is given
that a third and simpler way to define the ultrasystem is preferable.

According to the analysis of knowledge proposed by Lehrer,

knowledge equals undefeated justified acceptance. Undefeated justification is
then spelled out as coherence with every element of the ultrasystem, as Lehrer
calls it. So, the notion of an ultrasystem is central to Lehrer's theory of
knowledge. The question arises how this notion should be defined. Assuming
some familiarity with Lehrer's theory, this paper is concerned with just this
question. Three definitions of the ultrasystem, differing only in the additional
constraint they either state or do not state, will be investigated. Let us first
look at the following, relatively simple, way of defining the ultrasystem:

(DefUltra) A system M is a member of the ultrasystem of Sat t

if and only if either M is the acceptance system of S at t or results
from eliminating one or more statements of the form, 'S accepts
that q', when q is false, replacing one or more statements of the
E.1. Olsson (ed.), The Epistemology of Keith Lehrer, 243-252.
© 2003 Kluwer Academic Publishers.
form, 'S accepts that q', with a statement of the form, 'S accepts
that not q' when q is false, or any combination of such
eliminations and replacements in the acceptance system of Sat t.

This is not, however, the way in which Lehrer defines the ultrasystem. In his
1990 book Theory of Knowledge, he instead gives the following slightly
different definition: 2

(DefUltra ') A system M is a member of the ultrasystem of S at t

if and only if either M is the acceptance system of Sat t or results
from eliminating one or more statements of the form, 'S accepts
that q', when q is false, replacing one or more statements of the
form, 'S accepts that q', with a statement of the form, 'S accepts
that not q' when q is false, or any combination of such
eliminations and replacements in the acceptance system of S at t
with the constraint that if q logically entails r which is false and
also accepted, then'S accepts that r' must also be eliminated or
replaced just as 'S accepts that q' was.

This definition differs from (DefUltra) only in the additional requirement that
every statement r which is logically entailed by q and is also accepted and
false should be treated just the way q was treated. In his 1997 book Self- Trust,
Lehrer also gives basically the same definition of an ultrasystem. 3 A
somewhat more informal characterization of the ultrasystem that is equivalent
to the more formal definition (DefUltra') can already be found in Lehrer's
1988 paper Metaknowledge: Undefeated Justification. 4 But in Theory of
Knowledge, Lehrer also describes the ultrasystem in terms of the ultra
justification game, which is a heuristic to illustrate the ultrasystem: 5

She [the skeptic] may require the claimant to eliminate anything the claimant
accepts that is false, and the claimant must eliminate the specified item from his
acceptance system and at the same time eliminate anything he accepts that
logically implies the eliminated item. Or the skeptic may require the claimant to
replace anything the claimant accepts that is false with the acceptance of its
denial and at the same time replace anything that logically implies the replaced
item with acceptance of its denial. [italics GH]

Here Lehrer seems to favor a different definition of the ultrasystem, which

again differs from (DefUltra) only with respect to the additional constraint it

(DefUltra' ') A system M is a member of the ultrasystem of S at

t if and only if either M is the acceptance system of S at t or
results from eliminating one or more statements of the form, 'S
accepts that q', when q is false, replacing one or more statements
of the form, 'S accepts that q', with a statement of the form, 'S
accepts that not q' when q is false, or any combination of such
eliminations and replacements in the acceptance system of S at t
with the constraint that if q is logically entailed by r which is
also accepted, then'S accepts that r' must also be eliminated or
replaced just as'S accepts that q' was.

Since (DefUltra') and (DefUltra' ') are obviously not equivalent by definition,
it seems odd that Lehrer gives these two different characterizations of the
ultrasystem in the same book without presenting any philosophical argument
for their equivalence. Worse still, it seems to be pretty obvious that both
definitions are in fact not equivalent except for some trivial cases. Although
there is certainly, therefore, a serious tension between these two
characterizations of the ultrasystem given by Lehrer, I shall not inquire which
one could or should be regarded as the authorized one. Instead, I shall argue
against both (DefUltra') and (DefUltra"). My claim is that one should adopt
the most simple definition of the ultrasystem, that is (DefUltra). I will thereby
argue against all the characterizations of the ultrasystem which Lehrer gave in
[1], [2] and [3]. It is worth mentioning that in the second edition of his Theory
of Knowledge, Lehrer gives a simplified characterization of the ultrasystem 6
that is not affected by the criticism in this paper, and to which this author is
actually quite sympathetic. Also, one can understand the following argument
for (DefUltra) as a partial motivation for Lehrer's new characterization of the
But before I will argue for (DefUltra), let us first consider the merits of
(DefUltra') and (DefUltra"). A motivation can be given for both of them.
Although Lehrer has not, to my knowledge, published a motivation for the
introduction of the additional constraints of (DefUltra') and (DefUltra"), he
has orally reported to this author that the reason for doing so has been a worry
that could be called 'the problem of the creation of artificial justifications'.
Let me try to explain what this means as well as I can before arguing that this
worry is groundless. Let us first consider an example supporting (DefUltra)
Suppose someone accepts p and also accepts P/\q, but for some reason fails to
accept q. Now suppose further that p is false and hence that P/\q is also false.
According to (DefUltra) there would be an element M of the ultrasystem
where p, -,(p/\q) E M.7 According to the system M, one would be justified in
accepting -,q since this is a logical consequence of p and -,(p/\q). Since the
person initially did not have any epistemic attitude toward q or -,q, this
creation of a justification for -,q seems to be artificial. A more drastic case
would be one in which the person initially accepts q as well. Then, according
to (De fUltra) , there would be an element M of the ultrasystem with
p, q, -,(p/\q) E M. Since M is logically inconsistent, every proposition would
be justified relative to M So, according to (DefUltra), it is possible that in
some elements of the ultra system artificial justifications are created. At least,
prima facie, one can conceive of a worst-case scenario in which a person
originally has a faulty justification for some statement p that gives rise to
Gettier-type problems and in which in every element of the ultrasystem as
defined by (DefUltra), there will be created an artificial justification for p as
well, so that the justification for p will remain undefeated. According to the
analysis of knowledge as undefeated justified acceptance, one would therefore
know that p, contrary to what an adequate definition of the ultrasystem should
The additional constraint of (DefUltra ') seems to be appropriate to
avoid this unwanted effect. Considering the initial example, the system M
would not be an element of the ultrasystem defined by (DefUltra') because
P/\q implies p and p is false and also accepted. Therefore, the constraint of
(DefUltra') would require us to form a system M, instead of M, where we
replace not only P/\q by -,(p/\q) but also replace p by ---,po So we would have
---,p, -,(p/\q) E M. Now -,q is no longer deducible and therefore no artificial
justification has been created. In the modified example from above,
(DefUltra') would similarly require us to form a system M, instead of M,
which is not inconsistent.
A similar consideration can be given to motivate (DefUltra"). Suppose
there is a person who accepts both p and P/\q. Suppose further that p is false
and therefore P/\q is false as well. According to (DefUltra), there is an
element M of the ultrasystem where p ~ M and P/\q E M Though we have
eliminated p in M, p would still be justifiable in M on grounds of accepting
P/\q. Again we would have some unwanted artificial justification. (DefUltra")
seems to be appropriate in dealing with this case. Since p is logically entailed
by P/\q and P/\q is also accepted, the constraint of (DefUltra") would require
us to form a system M' instead of M, where we not only eliminate p but also
eliminate P/\q. Thus on M' we no longer have an unwanted justification for
p. 8
So there seems to be some quite good reasons to define the ultrasystem
by (DefUltra') or (DefUltra") rather than by (DefUltra). Despite this, I will
argue for (DefUltra) because I believe that the goals (DefUltra') and
(DefUltra") are aiming at are already brought into effect by (DefUltra). Ifthis
is true, all one would need is the more simple (DefUltra). Although the proof!
want to give to support my claim is somewhat technical, the underlying idea
is very simple. Therefore, it might be helpful to give a sketch of the idea first,
before we turn to the detailed proof. In the foregoing examples, (DefUltra)
gave rise to systems M of the ultrasystem on which too many beliefs are
justified. By stating additional constraints, (DefUltra') and (DefUltra")
required us to form a system M (M '), instead of M, on which these unwanted
justifications are no longer available. But I will try to demonstrate that, in
addition to M, the system M (M ') is also a member of the ultrasystem
defined by (DefUltra). Since it is necessary for a justification to be undefeated
not to be defeated by any member of the ultrasystem only those statements
that are justified on both systems M and M (M ') will have an undefeated
justification. Therefore, the unwanted justifications relative to M do not
increase these justifications that will remain undefeated. The claim that an
arbitrary member of the ultrasystem defined by (DefUltra') or (DefUltra") is
also a member of the ultrasystem defined by (DefUltra) of course needs to be
proven. I will first prove that we should prefer (DefUltra) over (DefUltra');
the proof concerning (DefUltra") will be perfectly analogous. For such a
proof, we need to give a formalization of the somewhat informal definitions
of the ultrasystems stated above. Obviously, a recursive approach seems to be
appropriate. But first, some remarks on notation are in order:

The acceptance system (of Sat t) will be denoted: A.

The ultrasystem and its members (of S at t) according to
(DefUltra) will be denoted: U= {MJ. M 2 ,oo.}.
The ultrasystem and its members (of S at t) according to
(DefUltra') will be denoted: U' = {MI " M 2',oo.}.

Def 1.1: Definition of an ultrasystem according to (DefUltra):

2. If M E U and q E A is false then M-q := M\{q} E U.
3. If ME Uand q E A is false then M*q:= (M\{q}) u {.q} E U.
4. Uhas no other elements than those mentioned by 1. - 3.

Def 1.2: Definition of an ultra system according to

1. A E U'.
2. If ME U' and q E A is false then
M-' q := M \ {r IrE A and q f- rand r is false} E U'.
3. If ME U' and q E A is false then
M*' q := [M\ {r IrE A and q f- rand r is false}]
u {r I .r E A and q f- .r and .r is false} E U'. (replacement)
4. U' has no other elements than those mentioned by 1. - 3.
The reader will notice that these formalizations are adequate counterparts of
the more informal definitions (DetUltra) and (DetUltra'). To be able to prove
some properties of all members of the set U' by induction, we need to define
the degree of a member of U'. We will do this again recursively.

Def 2: Definition of the degree of M E U':

1. degree(A) = O.
2. degree(M-'q) = degree(M) + l.
3. degree(M*' q) = degree(M) + l.

Now we have got all the clues together to prove:

Lemma 1: U'S;;; U.
Proof: We have to show that if M E U' then M E U. So let us
suppose that M E U'. We will show by induction on the degree
of Mthat M E U.
degree(M) = 0: It follows M = A and therefore M E U (by clause
1 of Def 1.1).
degree(M) = n: o.k.
degree(M) = n+l:
Case 1: M = N-' q for some N, q where N E U', degree(N) = n,
q E A, q is false.
Because N E U' and degree(N) = n, we have by the
induction hypothesis that N E U.
We have to show that M= N\ {r IrE A and q I- rand r
is false} is also an element of U.
Let {r IrE A and q I- rand r is false} = {rl> .. .,rk}.9
We can then rewrite Mas:
M= (... «M{rJ})\{r2})\... )\{rk}'
Note further that all the rJ, .. .,rk are trivially elements of
A that are false. Because of this, and because N E U, we
have by clause 2 of Def 1.1 that M{rJ} E U. By the
same reasoning, we have (M{rd)\{r2} E U etc.
Therefore we have ME U.
Case 2: M = N*' q for some N, q where N E U', degree(N) = n,
q E A, q is false.
Because N E U' and degree(N) = n, we have by the
induction hypothesis that N E U.
We have to show that M is also an element of U where
M = [N \ {r IrE A and q I- rand r is false} ]
U {r I-,r E A and q f- -,r and -,r is false}.
Let {r IrE A and q f- rand r is false} = {rl> .. .,rk} 10, we
then have {r I -,r E A and q f- -,r and -,r is false}
= {-,rl> ... ,-,rk}' We can then rewrite Mas:
M = (( ... ((M{rj} )u{-,rj} )\u ... )\{rk} )u{-,rk}'
Note further that all the rj , ... h are trivially elements of
A that are false. Because of this, and because N E U, we
have by clause 3 of Def1.l that (M{rj})u{-,rj} E U.
By the same reasoning, we have
(((M{rd)u{---,rd)\{r2})u{-,r2} E U etc. Therefore we
have ME U.
Now we need to introduce some new notations. If M is an element of the
ultrasystem then Js(.M) shall denote the set of all beliefs justified on M. That

Js(.M) = {p Ip coheres with M}.

If U is an ultrasystem, then Js(U) shall denote the set of beliefs that are
justified on all members of U.
That is:

If U = U{Mi} then
Js(U) = {p Ip coheres with every Mi E U} or more precise:

Js(U) = nJS(Mi).

For ultrasystems U', according to (DefUltra') and their elements M Js(M')

and Js(U') should be understood in a perfectly analogous way.

Lemma 2: If U j ~ U2, then JS(U j );;2 Js(U2).

Proof: Since U1 ~ U2 , we can assume without loss of generality
that for some index sets I, J we have:

Ul = U{Mi} and U2 = U{Mi}.

We then have:

Js(U2) = n [n
JS(Mi) =
JS(Mi)] n [n
JS(Mi)] =

= Js(UI) n[nJS(Mi)] ~ Js(UI)


Theorem 1: Js(U) ~ Js(U').
Proof: By specializing Lemma 2 on UI = U' and U2 = U we get:
If U' ~ Uthen Js(U) ~ Js(U').
Since by Lemma 1 we have U' ~ U, we get by modus ponens:
Js(U) ~ Js(U').
What is the upshot of all this? The interpretation of Theorem 1 is this: all
justifications that remain undefeated if one defines the ultrasystem by
(DefUltra) also remain undefeated if one defines it by (DefUltra'). The worry
that (DefUltra) could create unwanted justifications, thereby leaving too many
justifications undefeated, was without good reason. Every justification we
wanted to exclude by the additional constraint of (DefUltra') is already
excluded by (DefUltra)! Therefore, if the only reason for introducing the
additional constraint of (DefUltra') is to avoid unwanted justifications, then
we can just abandon this constraint from our definition of an ultrasystem. That
is, we should adopt (DefUltra) because we are benefiting from this in three
ways: (1) We do not have to introduce a constraint for which the motivation is
somewhat controversial; (2) We have exactly what we intended to get by
introducing the additional constraint; (3) We can even simplify the theory by
so doing!
It remains to be demonstrated that (DefUltra) is also preferable over
(DefUltra"). If we give a recursive definition of U" = {MI ", M2", ... }, and
the degree of its members that is analogous to the one given for U and U',
then we are able to prove:

Lemma 3: U" ~ U.

Since the proof for this is also perfectly analogous to the one of Lemma 1, it
would be unforgivable pedantry to present it in detail. But it should be noted
that every element of the set {r IrE A and r f- q} is trivially an element of A
that is false if q is false. Therefore, all elements of this set fulfill the
requirements of clause 2. and 3. of Def 1.1. This is the reason why we can
treat the case of (DefUltra' ') parallel to the one of (DefUltra') in the induction
step. Using Lemma 2 again, we can immediately infer from Lemma 3:

Theorem 2: Js(U) ~ Js(U").

This Theorem should obviously be interpreted as analogous to Theorem 1,

and leads us to the conclusion that we should prefer the simple definition
(DefUltra) not only over (DefUltra') but also over (DefUltra' ').


1 An earlier draft of this paper has been presented to Keith Lehrer. I am indebted to him for his
helpful comments. It is worth mentioning that Lehrer welcomed the proposal of this paper. An
earlier version of this paper was also read at the Lehrer-workshop in Konstanz in 2000. I wish
to express my indebtedness to the participants of the workshop for the helpful discussion, in
particular to Wolfgang Spohn and Erik Olsson who also read a first version of the paper.
2 See [2], p. 149 and also p. 194 (n.!l) where Lehrer defines the ultrasystem exactly as in
(DefUltra') [where I added the italics].
3 See [3], p. 44. There are some minor differences between the formulations in [2] and [3]
including the usage of the term 'evaluation system' instead of 'acceptance system' but the
crucial point is the same. In both definitions of the ultrasystem Lehrer states the additional
constraint of (DefUltra').
4 See [1], p. 343.
5 See [2], p. 14l.
6 See [4], p. 171, definition D8.
7 Here and further expressions like 'p EO M' are to be understood as shorthand for
"S accepts that p' EO M'. This notation simplifies our formulas considerably.
8 This case parallels the case of a contraction discussed in the theory of belief revision. If one
contracts ones belief set ('" acceptance system) by p, one not only has to eliminate p but one
also has to take care that p is no longer deducible from the belief set. Therefore everything that
entails p and is also an element of the belief set must be eliminated too just as (DefUltra")
9 In order to write {r I r EO A and q f- rand r is false} as {rl' .. .,rk} we made the assumption that

{r I r EO A and q f- rand r is false} contains only finitely many elements. This assumption has
been made to simplify the proof. But note that this assumption is not essential. If one abandons
this assumption one has to make a sub-induction on the number of elements in
{r I r EO A and q I- rand r is false}, in addition to the main induction on degree(M).
10 Cf. last footnote.

[1] Lehrer, K.: Metaknowledge: Undefeated Justification. In: Synthese 74 (1988). Reprinted in
his: Metamind. New York 1990.

[2] Lehrer, K.: Theory of Knowledge. 1st ed. Boulder 1990.

[3] Lehrer, K.: Self-Trust. A Study of Reason, Knowledge, and Autonomy. Oxford 1997.
[4] Lehrer, K.: Theory of Knowledge. 2 nd ed. Boulder 2000.
Chapter 15


Jacob Rosenthal
University of Bonn

The classical analysis of the concept of knowledge is as follows. Let Sbe

an epistemic subject and p a proposition. S knows that p if and only if

(1) p is true,
(2) S believes that p, and
(3) S is justified in believing that p.

It seems clear that the conditions (1 )-(3) are indeed necessary for knowledge.
But as some well-known examples (by Edmund Gettier, among others) show, the
stated conditions are not sufficient. The problem is that S's justification for his
belief that p may involve a false belief in an essential way. In such cases we
typically do not speak of knowledge, although the stated conditions are fulfilled.
So the task is to discover a fourth condition such that (1 )-(4) together are
necessary and sufficient for knowledge.
The following is Keith Lehrer's solution of the problem. 2 Take the set of
all beliefs of S. This is called the acceptance system of S. Condition (3) can be
read as stating that the belief that p is justified relative to this system, or, as
Lehrer says, on the basis of this system. Now, what Lehrer demands in addition
to justification relative to the acceptance system is justification relative to
certain modifications of it. Namely, if the acceptance system contains false
beliefs, and some of them are deleted from the system or even replaced by the
corresponding true beliefto the contrary, the belief that p must still be justified
relative to this new system to count as knowledge.
E.J. Olsson (ed.), The Epistemology of Keith Lehrer, 253-259.
© 2003 Kluwer Academic Publishers.

More exactly: Suppose the acceptance system ofS contains n false beliefs.
Now you decide for each of them independently ifit is to be (a) deleted from the
system, (b) replaced by the corresponding true belief, or (c) left untouched. In
this way you can obviously get 3n modifications of the acceptance system (one
of which is the acceptance system itself). S's belief that p is knowledge if and
only if it is justified relative to each of these 3n belief systems.
The idea behind this is that, to count as knowledge, S's belief that p must
survive corrections in the system of all beliefs of S. Make any corrections you
want, either weak (delete a false belief) or strong (replace a false belief by the
corresponding true one }-you always get a system relative to which the belief
that p is still justified. Then, and only then, S's belief that p is knowledge.
This proposal for solving the Gettier problem has great aesthetic appeal,
which is, however, somewhat diminished by a complication introduced by
Lehrer I didn't mention in order to keep things simple. Namely, the deletion or
replacement of false beliefs in the acceptance system is not entirely
unconstrained. If q and r are false propositions, and both are believed by S, and
q logically entails r, then, if you delete the belief that q from the acceptance
system, you must also delete the belief that r, and if you replace the belief that
q by the belief that not-q, you must also replace the belief that r by the belief that
not-r. So, your decisions what to do with the false beliefs are not totally free and
independent from each other. You have to respect relations oflogical entailment
in the indicated way. But this constraint is the only one and, having mentioned
it, the presentation of Lehrer's account of knowledge is complete. 3
I think that Lehrer's conception of knowledge is too demanding. Take the
following example: A reliable person has told me that the senate of my
university has elected Cohen for rector, which is indeed the case. So I know that
Cohen is rector. Most of what we know we get to know in more or less this way.
Now I remember a clause in the constitution of the university to the effect that
the rector is also the chairman of the Research Committee. I conclude that
Cohen is chairman of the Research Committee. But this is, in fact, wrong. The
senate of the university has, on the very same meeting, changed the constitution
and separated the positions. Cohen was only prepared to become rector if he
need not also be chairman of the Research Committee. My source of information
has told me nothing about this (nor should he have). This has the effect that if
I got to know that Cohen was definitely not the chairman of the committee, I
would also doubt his being rector and no longer believe it. There are two false
beliefs in my acceptance system: first, the belief that the constitution of the
university still contains the rule that the rector is also the chairman of the
Research Committee, and second, the belief that Cohen is chairman of this
committee. If the second belief is replaced by the corresponding true belief,
while the first false belief is left unchanged, the belief that Cohen is rector is no
longer justified, i.e., not justified on the basis of this modification of the

acceptance system. (At least you can construe the case in this way.) So Lehrer's
criterion is not fulfilled, and my opinion that Cohen is rector would not count as
knowledge. This is clearly counterintuitive. I know that Cohen is rector. All I
have done is to draw from this true proposition as one premiss and a false
proposition as second premiss a false conclusion. Normally such an act should
not destroy the knowledge status of the true belief. But Lehrer's condition is
such that this is regularly the case.
I think that Lehrer's idea to consider modifications of the acceptance
system of Sin order to decide whether S's beliefthatp is knowledge is the right
idea, but his condition is too strong. It is sufficient, but not necessary for
knowledge. The problem in the example arises because in modifying the
acceptance system, you are allowed to correct just some of the false beliefs,
while leaving others untouched. In my opinion, Lehrer should have said the
following: the modifications of the acceptance system of S relative to which the
belief that p must be justified must not contain any false beliefs any longer. So,
ifthere are n false beliefs in the acceptance system, you have to decide for each
of them whether it is deleted from the system or replaced by its true counterpart.
There are 2n possibilities to do so. A true belief of S is knowledge if and only if
it is justified relative to the acceptance system of S and relative to these 2n
corrections of it. I think this is a better proposal for solving the Gettier problem.
(And the above-mentioned constraint is now in any case superfluous, because
the modified systems contain only true beliefs.)
Is it really necessary to take into account so many different belief
systems? I don't really know, but with criteria of the Lehrer type it is definitely
not enough to consider just one modification of the acceptance system of S. It
would be much easier, of course, if you could say that S's true belief was
knowledge iff it was justified, first, relative to S's acceptance system, and
second, relative to a certain correction of it. But conditions ofthis type turn out
too weak. I consider the two most natural proposals along this line.

(a) S's true belief that p is knowledge if and only if it is justified relative to the
acceptance system of S and relative to the system that results from deleting all
false beliefs from the acceptance system.

That this condition is not sufficient for knowledge is shown by an

example that comes from Bertrand Russel1. 4 A pedestrian is walking down the
street wondering what time it is. He looks at a clock on a church tower which
shows ten minutes past three, from which fact he concludes that it is ten minutes
past three. And indeed this is true. What the pedestrian does not realize is that
the hands of the clock do not move. The clock has stopped a long time ago and
the pedestrian just happens to be looking at it at a moment when it shows the
right time. So, intuitively, the pedestrian does not know that it is ten minutes

past three, but has just happened to acquire a true belief to this effect. This belief
is justified on the basis of the pedestrian's acceptance system. The acceptance
system contains the false belief that the clock is moving and working in the usual
reliable manner. But this false belief is no indispensable part of a justification
for the pedestrian's opinion about the time, and therefore his opinion is still
justified if the false beliefis deleted from the acceptance system. Ajustification
might run as follows: "This is a clock. It shows ten minutes past three. Most
clocks work properly most of the time and therefore show the right time most of
the time. So I conclude that this clock shows the right time right now and believe
that it is ten minutes past three." The false belief that this clock works properly
is not involved in this reasoning, and so this reasoning is not blocked by merely
deleting the false belief from the acceptance system. Therefore the stated
criterion is fulfilled and gives the false result that the pedestrian's belief about
the time is knowledge.
The remedy seems obvious. If you not only delete the false belief that this
clock works properly from the acceptance system, but replace it with the true
belief that this clock does not work properly, then the justification just sketched
is blocked or, as Lehrer says, defeated. So, what about the following condition?

(b) S's true belief that p is knowledge if and only if it is justified relative to the
acceptance system of S and relative to the system that results from replacing in
the acceptance system all false beliefs by their true counterparts.

That this condition also fails can be shown by an example invented by

Roderick Chisholm. 5 A wanderer reaches a meadow on which there are two
animals. The first one looks like a sheep, but in fact it is not: it is a Bedlington
Terrier. Dogs of that race are easily confused with sheep. The wanderer does not
know about this and considers the first animal to be a sheep. He therefore comes
to the conviction that there is a sheep on the meadow. And this is true, because
the second animal is a sheep, although it does not look like one at all. Now, the
wanderer has the true belief that there is a sheep on the meadow. This belief is
justified by the fact that there is an animal looking like a sheep, namely, the first
one. There are two false beliefs in the wanderer's acceptance system: that the
first animal is a sheep, and that the second animal is not. Ifboth false beliefs are
replaced by their true counterparts the belief that there is a sheep on the meadow
is still justified. So the wanderer's conviction would count as knowledge, which
is clearly counterintuitive.
Let's look at this example more closely. Why does the proposed criterion
fail? The problem is that the wanderer's belief is justified before and after the
false beliefs in his acceptance system are replaced by their respective true
counterparts. But how can that be? Isn't the wanderer's justification for his true
belief, namely, that the first animal is (or looks like) a sheep, defeated by the

correction of his acceptance system? Indeed it is, but the correction gives the
wanderer another way to justify his belief, namely, that the second animal is a
sheep, although it does not look like one. So the conviction that there is a sheep
on the meadow is justified before and after the correction of the wanderer's
acceptance system. But it is justified in two different ways.
Now we have hit on the reason why the original and the modified Lehrer
criterion have to take into account so many different modifications of the
acceptance system of the subject S. If you consider just one or a few of these
modifications, one can always dream up examples where S's original
justification for his true belief is defeated, but where the defeating modifications
open up other ways to justify the belief that p. In such cases we typically do not
count S's belief as knowledge. It is merely a lucky coincidence that the subject,
although his original justification is defeated, has now other, new ways to justify
the true belief in question. In order to make such examples impossible, one has
to put up criteria which refer to a multitude of corrections of the acceptance
system, in a sense, to all possible corrections of it.
But obviously there is another possibility to deal with the problem such
examples pose. You just have to demand that the subject's original justification
is preserved when the acceptance system is corrected. 6 Instead of demanding that
S's belief that p be justified relative to very many different belief systems, you
demand that the belief is justified relative to just a few, but always in the same
way. I propose the following condition as a solution to the Gettier problem along
the indicated lines:

S's true belief that p is knowledge if and only if among the reasons S has for his
belief that p there are reasons r l , r z, ... ,rm with the following properties:

a) r l , r z, ... , rm are true,

b) together they are sufficient to justify the belief that p
relative to S's acceptance system,
c) together they are sufficient to justify the belief that p
relative to the system that results when in S's acceptance
system all false beliefs are replaced with their true

In short, a true belief of a subject is knowledge iffthe subject has ajustification

for the belief that remains a justification when in the subject's acceptance
system all false beliefs are replaced with the corresponding true ones. It is not
required that every justification of the subject has this property - a belief may
be justified in many different ways, and it is no harm when some of them are
faulty. But at least one possible justification, a justification that the subj ect could
use if asked, must be able to survive the mentioned strong correction of the

acceptance system. Then, and only then, is the true belief knowledge. In
comparison with Lehrer-type proposals, this proposal for solving the Gettier
problem has the advantage of involving just two belief systems, whereas the
former have the advantage of using merely the concept of a justified belief
(relative to a system of beliefs) and not the more demanding concept of
(sufficient) reasons for a belief (relative to a system of beliefs).
So we have arrived at two proposals for solving the Gettier problem: first,
the modified Lehrer proposal, and second, the one just mentioned. I am not sure
whether they are equivalent. But they could both be satisfactory solutions to the
Gettier problem and yet not be equivalent, as long as they agree in all clear
cases. There are borderline cases of belief in which one does not know whether
to call the belief in question knowledge, because the intuitions are unclear or
divided. No proposed criterion can be dismissed just because it decides a
borderline case in this or that way. As long as it gets the clear cases right, it may
count as a solution of the Gettier problem, and so there may be many
nonequivalent solutions. But I am afraid that sooner or later a clear example will
come up for which the two proposals considered here fail, as was the fate of so
many of their predecessors.

I These considerations were first presented on a workshop on Keith Lehrer's epistemology and

related topics, held at the University of Constance on June 16th , 2000. I am grateful to Keith Lehrer
for a discussion of these topics, to Wolfgang Spohn for several valuable remarks, and to
Christopher von Bi1Iow for improving my English.
2The presentation follows Lehrer's Theory o/Knowledge, Boulder 1990. His account has remained
essentially the same since the middle of the 80s and can be said to be the most prominent
internalistic proposal to solve the Gettier problem. (Compare also his book Self-Trust, Oxford
1997.) But recently Lehrer has changed his mind, as can be seen in the second edition of Theory
o/Knowledge, Boulder 2000. His new proposal is similar to proposal (a), discussed below, which
is also unsatisfactory.
3 Actually, I have some difficulties with this constraint. I would expect it to be the other way round.
If q logically entails r, and you delete the beliefthat r from the acceptance system, you should also
delete the beliefthat q, because otherwise the beliefthat r remains in the system in an implicit way.
After all, r is logically entailed by q. The same holds for the case of replacement. So the constraint
should be that in case you delete or replace the belief that r you must do the same with the belief
that q. But that is not important here, because the constraint, whatever it is, will play no role in
what follows. A constraint ofthis type is a half-hearted step into the direction ofthe AGM-theory
of belief revision (see Peter Gardenfors: Knowledge in Flux, Cambridge (Mass.) 1988), and
therefore unsatisfactory anyway. Either you should accept the whole AGM-apparatus (or
something similar), or you should try to make do without any proviso of this form.
4 Bertrand Russell: Human Knowledge. Its Scope and Limits, London 1948, p. 170.

5 Roderick Chisholm: Theory o/Knowledge, New Jersey 1966,21977, p. 105.


6 As Volker Halbach pointed out to me, John Pollock makes a similar proposal in the Appendix

of his book Contemporary Theories of Knowledge, Totowa 1986. The difference is that Pollock
believes (mistakenly, I think) that there are examples ofthe Gettier type in which the subject does
not believe anything false. So he does not speak of a correction ofthe subject's acceptance system,
but of adding truths to it. (In the clock example the subject does believe something false: namely,
that the clock is working properly. This false belief is no indispensable part of the subject's
justification for his opinion about the time, but it is nevertheless connected with this opinion. That
the subject's justification for his belief does not, or need not, include any false beliefs in Gettier
problem examples does not mean that there is no false belief involved at all.)
Chapter 16



John W. Bender
Ohio University, Athens

We must, therefore, in every reasoning form a new

judgement, as a check or controul on our first judgement or belief;
and must enlarge our view to comprehend a kind of history of all
the instances, wherein our understanding has deceived us,
compar'd with those, wherein its testimony was just and true.
Having thus found in every probability, beside the original
uncertainty inherent in the subject, a new uncertainty deriv'd from
the weakness of that faculty, which judges, and having adjusted
these two together, we are obliged by our reason to add a new
doubt deriv'd from the possibility of error in the estimation we
make of the truth and fidelity of our faculties.
When I reflect on the natural fallibility of my judgment, I
have less confidence in my opinions, than when I only consider the
objects concerning which I reason; and when I proceed still farther,
to turn the scrutiny against every successive estimation I make of
my faculties, all the rules of logic require a continual diminution,
and at last a total extinction of the belief and evidence.
-David Hume A Treatise of Human Nature, Book I


Let's begin with some old Lehrerian themes that continue to be central to
the latest version of his theory of knowledge. I Epistemic justification is a matter
E.1. Olsson (ed.), The Epistemology of Keith Lehrer, 263-280.
© 2003 Kluwer Academic Publishers.

of coherence for Lehrer, and knowledge is a matter of undefeated justification.

Personal justification in accepting a proposition, p, is a matter of whether
accepting p coheres with one's personal evaluation system, i.e., whether the
evaluation system of the person implies that it is more reasonable to accept that
p than to accept any objection that a critic might raise to p (Lehrer [2000a], 130-
132).2 Knowledge requires that these objections continue to be met even when
one's evaluation system is purged of whatever false acceptances might be
lending their support to one's acceptance of p.
Many such objections are skeptical in nature. In fact, it would not be an
unfair characterization of Lehrer' s complex concept of undefeated justification
to say that it is ultimately a matter of our ability (or the capability of our
evaluation system) to answer skeptical questions of various sorts. The theory
envisages two "games" in which a claimant asserts what he thinks he knows, and
a critic puts forth an objection to the claim. The first game tests for coherence
of the claim to the personal evaluation system of the claimant, which includes
all of the acceptances, true or false, that the claimant holds, given his goal of
accepting truths and avoiding falsehoods. The second game, the
"uitrajustification game," whose outcome indicates whether the claimant's
personal justification is "undefeated" and hence whether he knows the target
proposition, p, allows the critic to call for the elimination of anything false that
the claimant is offering in defense of the skeptical critic's objections.
It is surprisingly incongruous, then, that a theory whose structure appears
so concerned with skeptical objections should, in fact, be guilty of skirting and
obfuscating when it comes to the skeptic's main concern. Or so I will argue.
I take the skeptic's central query to be something neo-Humean like this:
"Of course you believe that you are justified in accepting many of the things you
do about the world around you, but are you really, and philosophically-speaking,
justified? Do you truly know what you think you know?" So, at the heart of
things, the skeptic challenges our claims to be epistemically justified, and
thereby, challenges our claims to know.
I want to suggest that, for all its references to the skeptic, Lehrer's theory
is in danger of doing end-runs around her because the two games the theory
plays with the skeptic or critic are aimed respectively at internal coherence,
which may be too weak a notion to establish that a belief is really justified, and
at undefeatedjustification or uitrajustification, which is really about making sure
that our claims to knowledge do not depend upon error and falsehood. But the
skeptic can jovially admit that most of us have an internally (reasonably)
coherent system of beliefs, and also that skeptical possibilities do not have to be
true to perform their philosophical function and therefore, conversely, our
everyday beliefs do not have to be false to be unjustified.
So can the fact that it is claimed that the skeptical critic loses the two
justification games really constitute a satisfying rejoinder to skepticism? Lehrer

admits he cannot prove the skeptic wrong, but he does think that he has a
convincing response to her and argues that ultimately we not only have
knowledge but that we can (and must, since discursive knowledge is
metaknowledge) know that we know (Lehrer [2000a), 205, 229)· His desire to
establish this makes his recent discussions return to the skeptic again and again,
as ifhe senses the inherent weakness of the theory's two games to really answer
The "keystone" of this effort is the concept of self-trust. And this
keystone is put into place in our defenses against skepticism by the deployment
of an argument Lehrer labels the "trustworthiness argument." The
trustworthiness argument is touted and alleged to have the epistemic ability to
loop our justification back on itself, guaranteeing coherence and saving us from
the surd beliefs of foundational ism and the infinite regress of reasons which
threatens if there is no loop back (Lehrer [2000b), 643). Lehrer now apparently
admits that coherence within an acceptance system that fails to contain an
acceptance that one is trustworthy in what one accepts will not be able to be
elaborated in any way that will yield knowledge, or at least, discursive
knowledge (Lehrer [2000b), 638.) This paper will examine the trustworthiness
argument and Lehrer's more wide-ranging discussion of skepticism and
conclude that the argument is impotent and the discussion ultimately
unsatisfying. Although Hume's objection is, as stated, a type of infinite regress
argument, its spirit still applies to Lehrer's theory: there is a serious doubt
whether the "new judgement" of trustworthiness is, in fact, sufficiently justified
to support our claims of knowledge.



To be personally justified in accepting that one sees, remembers,

or introspects something, one must, therefore, accept that these are
trustworthy sources of information of the truth ... (Lehrer [2000a],

It is not enough that one accept something for it to be more

reasonable than the objections to it on the basis of one's evaluation
system. One must have some information that such acceptance is
a trustworthy guide to the truth. (Lehrer [2000a], 137).

As I already intimated, there is a question whether internal coherence in

an individual's personal evaluation system constitutes any degree of epistemic

justification. But I do not want to enter a general and probably useless verbal
discussion about the use of 'justification.' However, there is a related
substantive objection that foundationalists have urged against coherentists, and
which is one that any good skeptic would urge as well. Coherence with a
background system of beliefs will justify a target belief only if the background
beliefs themselves are justified. But a personal evaluation system contains all
acceptances the person holds (with the objective of accepting somethingjust in
case it is true); it is not a system of justified beliefs alone. So the skeptical
question is, "Are we really justified in accepting those other things we need to
accept if we are to be able to meet a critic's objections?"
More clearly now than previously, Lehrer admits the force of this line of
argument. His answer, given in the passages above, is that personal evaluation
systems, if they are ultimately to provide us with discursive knowledge, must
contain meta-acceptances of the form, "My acceptance that p is a trustworthy
guide to truth." It is easy to see the necessity of these acceptances in the
justification game. For example, if, on the basis of taste, I accept that the wine
I am drinking is a Bordeaux, I must be able to meet (answer or neutralize)
objections such as, "You have no ability to distinguish Bordeaux from
Californian or Italian Cabernet," or "People often confuse California Cabernet
and Bordeaux." And, intuitively, it seems correct that unless I believe that I can
trust my taste in these types of matters, I will not be justified in accepting that
this wine is a Bordeaux.
But the obvious problem is that the necessary meta-acceptance is now a
member of the evaluation system, and therefore subject to the same doubt that
it itself may not be justified. Lehrer deploys two responses to this latest doubt.
In both Lehrer [2000a] and [2000b] a straightforward inductive argument from
successful past acceptances of the same or similar sort is thought to support the
meta-acceptance that my acceptance of p is trustworthy. Although it seems to
me that this is the crucial and natural argument, there is little more than a
mention of it in Lehrer's discussions. (Cf. [2000a], 138, 142 and [2000b], 642f.)
His real interest seems to lie with a different kind of argument, based upon an
even more fundamental principle of self-trust:

I may accept that my faculties, perception, memory, reasoning, and

so forth are trustworthy guides to truth in circumstances of the sort
that I find myself in when I accept what I do. I must accept,
moreover, that I am trustworthy as well: that when I accept
something, that is a good enough reason for thinking it to be true,
so that it is reasonable for me to accept it ([2000a], l38).

Hence, an evaluation system that ultimately supports discursive

knowledge requires acceptance of "one special principle of an evaluation
system" ([2000a], 138), viz., Principle T:

T: I am trustworthy (worthy of my own trust) in what I accept with

the objective of accepting something just in case it is true
([2000a], 138).

Acceptance of(T) explains why my particular acceptances are trustworthy

and hence why it is reasonable for me to accept them. Lehrer says:

[I]f I accept that I am trustworthy in this way, then my accepting

something will be a reason for me to accept it ([2000a], 138).

We can wonder, at this point, how accepting something can be a reason

for accepting it, since nothing can be a reason for itself. Apparently, Lehrer is
suggesting that the fact of my acceptance of p, given that I also accept (T),
constitutes a reason or evidence supporting the truth ofp. But it seems clear that
such a line of reasoning has credence only if the acceptance of (T) is justified.
We must not lose sight of this fact as we turn to Lehrer's trustworthiness
argument and the work it is given to do.


Principle (T) acts as a guarantor of my acceptance of the trustworthiness

of my target acceptance that p. In other words, when (T) is included in my
evaluation system, I am able to reason from my acceptance of p to the
reasonableness of that acceptance. This is the trustworthiness argument. Lehrer
formulates the reasoning in the following way:

(T) I am trustworthy (worthy of my own trust) in what I accept

with the objective of accepting something just in case it is true.

I accept that p with the objective of accepting that p just in case it

is true.

Therefore, I am trustworthy in accepting that p with the objective

of accepting that p just in case it is true.

Therefore, I am reasonable in accepting that p with the objective

of accepting that p just in case it is true ( [2000a], 139).

A few clarificatory observations about the trustworthiness argument are

necessary. (T) is meant to be understood not as a universal deductive principle,
but as an expression of a fallibilistic disposition or capacity to be trustworthy.
Also, and relatedly, the inference from (T) to the conclusion that I am
trustworthy in accepting that p is meant as an inductive inference, not a
deduction. Third, and most importantly, it should be noted that the conclusion
of the argument is not equivalent to the conclusion that I am personally justified
in accepting that p. This is because personal justification requires that all
objections that the critic might present to the conclusion of argument must be
answered or neutralized (cf. [2000a], 143). It must be more reasonable for me
to accept the conclusion than to accept any objection against it for me to be
personally justified in accepting the conclusion.
Now, the trustworthiness argument is a bit of reasoning for the conclusion
that I am reasonable in accepting that p. How justified the conclusion of this
reasoning is for me depends on how justified I am in accepting the argument's
premises, especially (T). As we saw from the passage quoted in the opening of
section II above, Lehrer himself believes that mere acceptance of something
does not suffice to make it more reasonable than objections to it on the basis of
one's evaluation system (137). Clearly, the skeptical critic will object that (T)
is not more reasonable to accept than, e.g., the hypothesis that we are deceived
in thinking we are trustworthy. Hence, the conclusion of the trustworthiness
argument would be seen as unjustified. This objection must be adequately dealt
with, and the trustworthiness argument does nothing in and of itself to provide
a response.
So it would be a mistake to think that the trustworthiness argument
constitutes a personally justifying argument for the reasonableness of what we
accept. If (T) is now an important member of my evaluation system, then, just
as we saw earlier that my meta-acceptance that my acceptance that p is
trustworthy, (T) also must bejustifiedly accepted if the trustworthiness argument
is to successfully perform its intended justificatory function. But what is the
argument that my acceptance of (T) is justified?


Although inductive support for (T) is again mentioned (142), Lehrer

focuses, as he did in his book, Self Trust, (Lehrer [1996]), on a different, "more
direct" (2000a,142) argument to the effect that the trustworthiness argument
exemplarizes itself (Cf. [2000b], 641-2), and, in a sense, therefore, is self-
justifying. The idea is that the trustworthiness argument can be applied to (T)
itself, by taking (T) as the value for 'p' in the argument. Lehrer says:

If a person accepts (T), then her acceptance of (T) itself will have
the result that it is reasonable for her to accept that (T) by an
application of the trustworthiness argument to (T) itself as the
target acceptance p. The principle applies to itself. It yields the
results that ifshe accepts (T) with the objective of accepting it just
in case it is true, then she is trustworthy in accepting it, and by the
trustworthiness argument, to the conclusion that she is reasonable
in accepting it (Lehrer [2000a], 142).

Something quite peculiar is going on here. If the trustworthiness

argument is a bit of reasoning that the subject puts forward for justificatory
purposes, then each premise of the argument is an expression of something the
subject accepts. But, then, ifI accept the first premise (T), the second premise
adds nothing to the argument. So it is supposed to follow from the fact that I
accept that I am trustworthy, that I am trustworthy in accepting that I am
trustworthy, and hence reasonable in accepting that I am trustworthy in what I
accept. But how can it follow from the fact that I accept that I am trustworthy,
that it is reasonable that I am trustworthy? When (T) replaces p in the
trustworthiness argument, the conclusion is that I am reasonable in accepting the
trustworthiness principle. But this conclusion is only justified for me to the
degree that the premises are justified. Hence, I must be antecedently justified
in believing premise (T), and therefore, I must already be reasonable in
accepting (T). Consequently, the trustworthiness argument is moot by
presupposing its own conclusion. I see no virtue in the supposed "virtuous loop
of reason" (142).
Another way of seeing that the trustworthiness argument does not provide
a looping justification for principle (T) is to notice that any unjustified (T*)
might be offered as the value of p. Imagine that I accept every statement
occurring in the Bible as true, and that I think this while I am motivated to
believe things just in case they are true. Let (T*) be the principle that I am
trustworthy in my Biblical acceptances. The trustworthiness argument, then
runs as follows:

T: I am trustworthy (worthy of my own trust) in what I accept

with the objective of accepting something just in case it is true.

Therefore, I am trustworthy in accepting (T*): I am trustworthy in

my Biblical acceptances, with the objective of accepting this just
in case it is true.

Therefore, I am reasonable in accepting (T*): I am trustworthy in

my Biblical acceptances, with the objective of accepting this just
in case it is true.

If acceptance of (T) has the power to make (T) reasonable, it has the power to
make (T*) reasonable. But, that (T*) is reasonable on the basis of the
trustworthiness argument is completely absurd. The mere acceptance of (T),
even assuming (T)'s truth, has no inherent justificatory power, and no amount
of "looping" through the trustworthiness argument can generate what is not



If the virtuous loop of reason cannot be relied upon to support the

acceptance that I am reasonable in accepting that p (whether p is principle (T)
or is some other acceptance), it becomes a question whether I can go on to be
personally justified in accepting that p. If the trustworthiness argument fails to
establish that I am reasonable in accepting (T) then the conclusion of the
trustworthiness argument (viz., that it is reasonable for me to accept that p)
seems open to the critic's objection that this conclusion is based upon a premise
(viz., (T)) which is not justified since it is not more reasonable to accept than the
skeptical hypothesis that I am systematically deceived, say, by an all-powerful
demon rendering all my first-order acceptances false. Lehrer's answer to this is
that I do accept that it is more reasonable for me to accept that p than that I am
generally deceived. ([2000a], 133). This may well be true, but we have now
seen that the trustworthiness argument does nothing to establish that it is more
reasonable to accept this. I must accept, apparently on inductive grounds, that
my evidence for p is trustworthy and renders it very improbable that I am
deceived (134). I may, in fact, accept this, and, therefore, there is little argument
that I am not personally justified in accepting p. But we are still far from
knowing anything against the skeptic's objections.


I have charged the trustworthiness argument with circularity. Another,

equally pernicious, circle seems to be lurking in this discussion. Lehrer says
that, "my reasonableness is explained in the argument by my trustworthiness"
(139). And, of course, this is Lehrer's whole intent: to support claims of what

it is reasonable to accept-claims that can be used in the justification

games-by relying on one's acceptance that one is trustworthy. But then,
trustworthiness itself had better not turn out to be the idea of what it is
reasonable for me to accept, or what I am justified in accepting. Yet, Lehrer's
characterization of trustworthiness leads to the suspicion that it either
presupposes, or is itself, a veiled concept of justification.
For instance, Lehrer says that to be trustworthy in such matters as
accepting that I see a zebra, I must have "reason to think" that I can tell that
something is a zebra when I see one (138). Trustworthiness implies a high level
of success in reaching truth, but it does not entail reliability in every instance.
I can be worthy of my trust but be deceived. Lehrer makes it clear that he now
believes that a person deceived by the Cartesian demon may still be trustworthy
in what she accepts because she proceeds in a virtuous and epistemically
faultless manner (140, 211). But notice that, although the deceived person's
acceptances are not, in the normal sense of the word, a trustworthy guide to the
truth in her circumstances, since they are all false, what is left unmolested by the
demon is her justification for her acceptances. She seems to be "trustworthy"
only in the sense that her acceptances have been based on justified methods and
grounds. So, I strongly suspect that trustworthiness in Lehrer's theory is or
involves something like the disposition to form justified acceptances.
Consequently, the crucial move/rom trustworthiness to reasonableness may be
no real epistemic move at all, but only a question-begging entailment.


Yes, perhaps we do win the personal justification game against the

skeptical critic. In Lehrer's example, a claimant accepts that he sees a zebra.
Among other objections offered, the critic proposes that the claimant is generally
deceived in a systematic way and sees nothing. But this objection is answered
in the personal justification game because the claimant in fact also accepts that
it is more reasonable for him to accept that he sees a zebra than that he is
systematically deceived. Lehrer suggests that this acceptance is itself based
upon the claimant's further acceptances that his evidence is trustworthy, and that
the hypothesis of systematic deception is totally improbable on that evidence
([2000a], 133-4).
But the uitrajustification game allows the skeptic to eliminate from a
person's evaluation system anything false. Presumably, the skeptic takes it to
be false that it is more reasonable for the subject to accept that the world is the
way he accepts rather than that he is systematically deceived. The skeptic
doubts that our evidence philosophically justifies our conclusions. Now, the
subject, of course, believes the contrary.

How does the ultrajustification game go forward, then? We are to

envision the critic commanding the claimant to remove some falsehood from his
evaluation system, and the removal to be legitimate when, in fact, the acceptance
is false. But neither the claimant nor the skeptic knows whether the objection
is true or false, i.e., neither knows the answer to the deadlock. And, apparently,
neither Lehrer nor we know any more, because, at this point, Lehrer is only able
to consider the two possibilities and note their consequences: if it is true that the
subject is more reasonable in accepting that he sees a zebra than that he is
deceived, then he has undefeated justification, and knows that he sees a zebra;
if the skeptic is correct and the claimant's acceptance is not more reasonable
than the objection, the subject's justification is defeated, and he fails to have
Notice that this is no response or reply to the skeptical challenge at all; it
is merely a statement of a consequence of Lehrer's theory, that the subject
cannot know something unless the skeptic's denial of the subject's
reasonableness is false. This is tantamount to saying that Lehrer's so-called
"transformation argument" (163) provides no succor against the skeptic. We
mustjustifiedly believe that it is more reasonable for us to accept what we see
than to accept that we are deceived in order to have an actual response to
What we might call "Lehrer's fork," the observation that there are two
prongs to be considered, one epistemic and one skeptical, does nothing to bolster
our belief that our acceptances are more reasonable than the skeptic's
challenges. Ifwe do not know, or at leastjustifiedly believe that we are more
reasonable, we can't say who wins the ultrajustification game. The fact that the
game seems stale-mated at this point, and that Lehrer's fork simply presents the
alternatives, is indicative of the fact that we do not know whether we know or
not. Nonetheless, in his discussion of skepticism, Lehrer claims the opposite,
viz., that we do know that we know. I will turn to that discussion after one final
point about the ultrasystem.


We have already seen that trustworthiness implies high success in gaining

truth, but not in every case. A person is supposedly trustworthy in the Cartesian
demon world, even though the reliability of his acceptances in that world is nil.
The truth of principle (T) does not guarantee a win in the ultrajustification game.
The truth of another principle connecting trustworthiness and rei iability also is
a necessity. Lehrer says:

[T]he conversion of personal justification to irrefutable or

undefeated justification requires the truth of the following

(TR) If I am trustworthy in what I accept, then I am reliable in

obtaining truth and avoiding error in what I accept.

So personaljustification requires acceptance of(TR) or preference

for accepting (TR) over the negation of it. Otherwise the critic will
win the justification game. ... For the justification to be
undefeated, acceptance of (TR), or preference for accepting (TR)
over the negation of it, must be contained in the ultrasystem
([2000a], 194).

Hence, in the uitrasystem, (TR) is necessary. But, in the demon world

scenario, (TR) is false, and therefore, eliminated. This results in the correct
conclusion about knowledge in that world, since the demon's machinations keep
us from knowing, but it appears to give the wrong conclusion about our status
as justified believers. Our non-skeptical intuition is that, since we know in the
actual world, our justification is epistemically sufficient to support knowledge.
This justification is unimpeded by the demon, and therefore, even if we do not
know in the demon world, since our beliefs are false, our justification in that
world is undiminished. However, Lehrer takes uitrajustification, i.e., undefeated
justification, to be the justification necessary to support knowledge. But in the
demon world, I do not win the uitrajustification game because (TR) is false, and
therefore, deleted. Hence, on Lehrer's view, I do not have sufficient epistemic
justification in the demon world. This is the wrong answer, intuitively: I should
be equally justified in the actual and the demon world. It appears that
ultrajustification does not accurately track epistemic justification.


We offer no proof that the skeptic is wrong .... We may,

nonetheless, know that she is wrong ... and indeed we know that
we know (213).

Coherence within one's personal evaluation system, plus the truth of the
members of that system, such that they ascend to the ultrasystem, is supposed
to yield justification, according to Lehrer. I have been arguing that personal
coherence plus truth cannot equal epistemic justification, because the skeptical
question whether the personal acceptances are truly justified has been side-

stepped. We have seen that principles that Lehrer requires in any justificatory
system, such as principles (T) and (TR) are problematic. (T) seems unjustified
by the trustworthiness argument, and (TR) appears to be false in the demon
situation, even though its truth is required for epistemic justfication in that case
on Lehrer's theory. The defenses against skepticism appear to be depleted.
A problem also arises with the claim that, even though no proof can be
offered that skepticism is false, we nevertheless know things and know that we
know those things. The problem has deep roots in the very structure of the
ultrasystem. To see this, consider that everything that a subject accepts with the
objective of accepting it just in case it is true, i.e., everything in his personal
evaluational system, is something he will believe to be a member of his
ultrasystem, or, more precisely, a member of his "T-system," (the subset of the
ultrasystem containing only those states, acceptances, preferences and
reasonings that are true) (168). Since no subject is omniscient in regard to the
possible errors he may have made, he cannot know the membership of his T-
system. (If he did know that his evidence was in error, then, given that he is
motivated to accept things if and only if they are true, he would expunge the
error from his evaluation system.) Yet, Lehrer says,

A person lifting her hand before her eyes accepts that she has a
hand, and she also knows that her justification for accepting this
does not depend on any error of hers. She might not know what
the members of her ultrasystem are, but she knows that whatever
they are, they will leave her justified in accepting that she has a
hand. So, she may know and know that she knows ([2000a], 169).

But how can this be? Precisely because of the fallibility of our quest for truth,
we sometimes accept things that are false or based upon error even when we are
doing our epistemic best. In the famous Gettier-style example, I justifiedly
believe that Mr. Nogot owns a Ferrari because of the evidence I possess for that
acceptance. But I do not know that my conclusion is based upon an error,
because I do not know that Nogot is playing a practical joke on me. Hence,
although I accept and justifiedly accept that my evidence is not deceptive, I am
wrong. So, I do not know the membership of my T-system. Whenever the
skeptical critic commands me to eliminate an acceptance in the ultrajustification
game, it will come as a surprise to me, since, of course, I hold that acceptance
with the objective of accepting something just in case it is true.
Now, if the situation is that the necessary principles (T) and (TR) are
either unjustified or false, and the person does not know the membership of his
ultrasystem, how can it be that the person knows, let alone knows that he knows
that the skeptical objections are less reasonable than his own knowledge claims?
It seems that he cannot.

Even to know that I know that I have a hand would require, on Lehrer's
theory, that I know that my acceptance that I have a hand is invincible in the
ultrasystem. But how can I know this, if! do not have access to the membership
of my T-system? The conclusion can only be that Lehrer has not made out his
claim that knowledge and meta-knowledge are possible even in the face of
skeptical objections.
It is easy to see the crucial role played by principles (T) and (TR) in this
context. Without a justified acceptance of these principles, it is difficult to
imagine any defense against the skeptical objection that it is not more reasonable
for a person to accept what he does rather than to accept that he is systematically
deceived, or at least in error, in what he accepts. If one's evidence is completely
consistent with the skeptical scenario, how can it be thought to positively
confirm the person's preferred acceptance? To cite some independent
implausibility ofthe skeptical hypothesis appears, again, to be question-begging.
Other acceptances of ours may imply such an implausibility, but they, in their
turn, are open to the skeptical criticism. Lehrer admits that reliance upon
principle (T) is question-begging in any attempt to prove that the skeptic is in
error. But he believes that relying upon (T) in explaining why it is reasonable
to accept what we do is not vicious but virtuous. He says,

To use a premise to prove something to a skeptic who challenges

it violates the rules of rhetoric. But to explain why it is reasonable
to accept what we do, the circle may be virtuous (143).

But vicious circularity occurs in argumentative contexts other than proof.

Lehrer's reason for thinking he cannot prove that the skeptic is wrong is that he
cannot prove the truth of principle (T) (227). Even if (T)'s best support is
straightforwardly inductive and therefore, no question of proof arises, (rather
than being based on the dubious loop of trustworthiness) it is still a "rhetorical
violation" to assume (T) in any answer to the skeptic who objects that it is not
more reasonable for us to believe that we are epistemically connected to the
world than to believe that we are systematically in error. That (T) is in our
evaluation system and is, perhaps, true is simply not enough to avoid the charge
of circularity.
To respond effectively to the skeptic, it must be shown that our
acceptance of our trustworthiness is not just held but is justified. Nothing
Lehrer has offered seems to be up to this challenge. Moreover, it should be
remembered that Lehrer's entire picture of what it is to "explain why it is
reasonable to accept what we do" (143) is that the subject is able to win the
personal and ultra-justification games against the skeptical critic. This may not
constitute proving that the skeptic is wrong, but it surely involves being able to

answer the skeptic's objections in a non-question-begging manner. We have

seen that this is a more difficult task than Lehrer has admitted.


In the preceding section, I asked the question how evidence completely
consistent with the skeptical hypothesis could be thought to positively confirm
our natural belief that we are in contact with the world in the way we accept.
Lehrer takes up this question in his discussion of skepticism, and says this:

[H)ow can we know that we are not deceived when the reasons we accept
for concluding we are not deceived are exactly the same reasons we would
accept for concluding we are not deceived if we were deceived?

The reply is that though the content of the reasons we accept would be the
same, the reasons we accept would not be the same. In the one case, in the
actual world, the reasons would be true and in the other case, in the demon
world, they would be false. This is a crucial difference ([2000a),212).

But, first of all, my reasons for believing what I do in the demon world are
not false, if those reasons are construed as acceptances of the sort, "It appears
to me now that there is a zebra in front of me." The demon makes it false that
there is a zebra in front of me, but does so while giving me the misleading but
true evidence that there appears to be a zebra here. Secondly, and far more
importantly, Lehrer's suggestion that we individuate reasons by considering
their truth-values is highly counter-intuitive, and leads immediately to
unacceptable results. My reasons for believing that Nogot owns a Ferrari are the
same whether or not he owns the car or is playing a practical joke. In either
case, they have to do with the likelihood that someone would have official-
looking papers, assert that he owns a Ferrari, and be seen driving a Ferrari ifhe
did not own the car. My reasons are, as in the demon-world scenario, actually
true: I did see official-looking papers, hear Nogot assert that he owns the Ferrari,
and see him driving the car.
If Lehrer means that it is not the evidence that is false, but, rather, the
conclusion they support, then we are still left with unacceptable results. I take
it that in the ultrasystem, we are to eliminate any false reason I have for my
ultimate acceptance. But then, eliminating my acceptance that Nogot owns a
Ferrari as my reason for believing that someone in my office owns a Ferrari is,
allegedly, not to eliminate the same reason I have for believing that someone
owns a Ferrari if, in fact, Nogot does own the car. This seems, simply, wildly
counter-intuitive. Reasons must be individuated on their content alone, and not

on their truth-value. To do otherwise is to confuse their evidential value with

their truth value. Lehrer's view has always been that if a personally justified
reason remains undefeated in the ultrajustification game, then the subject may
know what he accepts on the basis of that reason, while if it is defeated, then
knowledge is not achieved. It is one and the same reason that is being judged
for its invincibility in the ultra-game; it is not one reason if true and another if


I mentioned above that our fallibility regarding our evidence makes it

implausible, on Lehrer's theory, that we know that we know, and more
particularly, makes it unclear how it is more reasonable for us to accept that we
are connected to the world in the ways we assume rather than accepting that we
are not connected and are, perhaps, systematically in error. Lehrer takes up
these matters in one final argument, and this will bring us full circle back to our
original observations about the trustworthiness principle. Here is Lehrer's

The objection based on fallibility is neutralized by our

trustworthiness. It is as reasonable to accept both that we are
fallible and that we are trustworthy in a truth-connected way as it
is to accept only that we are fallible. It is then as reasonable for us
to add that we are trustworthy in this way to the objection and
accept both the possibility of error and our trustworthiness in
avoiding error in a truth-connected way as to accept the skeptical
worry alone ([2000a], 220).

This sounds like an answer to skepticism: we are trustworthy in a truth-

connected way, therefore the skeptical objection that we are not connected to
world as we accept is neutralized. But what Lehrer says next is telling:

In those instances in which we are trustworthy in a truth-connected

way our justification may be undefeated by local errors and convert
to knowledge. In those instances in which we accept that we are
trustworthy when we are not, on the other hand, the neutralization
fails in the ultrasystem and our justification is defeated (220).

This passage, however, says only that ifwe are trustworthy in a truth-connected
way, then we can neutralize and defeat any skeptical challenge, and ifwe are not
trustworthy and reliable, then our justification will be defeated.

One cannot establish that it is reasonable for us to accept that we are

trustworthy in many cases of our acceptances by pointing out that in cases in
which we are trustworthy, we defeat the skeptical objection. The latter claim is
a theorem of the theory, while the former is a substantive epistemic assertion.
The substantive claim requires justification: we cannot merely accept that we are
trustworthy, and wait for that claim to turn out true and, therefore undefeated,
in the ultrasystem. I might irresponsibly accept that I am clairvoyant, and, if it
turns out that I am, I still do not know what I accept on the basis of my
clairvoyance, without a good reason to believe that I am clairvoyant. I have, as
Lehrer himself puts it, no reason in such a case to believe this information is
We have seen that the most "direct" argument for the justification of my
acceptance that I am trustworthy is supposed, by Lehrer, to be the "loop of
trustworthiness," but we have found this argument wanting. We have also
acknowledged that Lehrer believes that there is good inductive support for
principle (T), but very little in the way of a detailed argument for this has been
provided. We are, once again, left with the old Pyrrhonian, and neo-Humean
worry that Lehrer has begged the question against the skeptic.


Lehrer sees that an evaluation system dereft of principles of

trustworthiness and reliability will not produce coherence that supports
justification or knowledge. He, therefore, requires them as part of one's
personal evaluation system and believes they carry through as truths to the
ultrasystem. Knowledge requires this. But, for all his talk of loops-loops of
trustworthiness, and loops of explanation ([2000a], 227)-we have been hard-
pressed to see how those loops are not circles in the epistemically vicious sense
in which no justification is generated. Skepticism is acknowledged by Lehrer
as a serious and coherent threat to knowledge. This paper has been an attempt
to show that Lehrer's recent theory lacks the theoretical agility to loop around
the skeptic and lead us to knowledge and meta-knowledge.

1 Latest expressions of Lehrer's view can be found in Lehrer [1996], Lehrer [2000a] and Lehrer
[2000b]. My attention will focus most upon chapters 6-9 of Lehrer [2000a].

2A person's evaluation system is a newer, more robust conception of what Lehrer used to call the
acceptance system of a person. The acceptance system contained only acceptances (belief-like
states), but the evaluation system now contains not only acceptances, but preferences to accept one
thing over another, as well as reasoning structures or strategies. The additions have been made to
allow doxastic preferences to count towards judgments of comparative reasonableness, and to meet
objections that the acceptance system did not reflect the inferential structure of a person's
reasoning and therefore could not reflect whether the person was justified in her beliefs.

Lehrer, K: 1996, Self Trust, A Study of Reason, Knowledge, and Autonomy, Oxford University
---- :2000a, Theory of Knowledge, Second Edition, Westview Press.
----: 2000b, "Discursive Knowledge," Philosophy and Phenomenological Research, Vol. LK, No.
Chapter 17


Peter Klein
Rutgers University


About twenty years ago Keith Lehrer wrote this about how he does

... In philosophy, we do not have experiments, only the light of

reason to guide us ... Anyone who obscures that light is defeating
the enterprise . . . Should I ever reach the point at which I am
disinclined to seek criticism and amend my views in light of it, I
shall take to writing fiction . . . and give up writing philosophy
altogether. Criticism is the touchstone of philosophical inquiry,
and those who shunt it aside ... are phonies and beguilers. I

Now, we could all say such things. But Lehrer exemplifies this ideal. I have
always admired him for that. My comments about his most recent account of
knowledge should be taken as part of a long series of discussions we have had
over the years. I have always learned a great deal from them. My bet is that his
reply to my comments will amply demonstrate that he is not about to take to
writing fiction.
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 281-297.
© 2003 Kluwer Academic Publishers.



Lehrer's account of knowledge has evolved over the years but there have
always been two central concepts designed to grapple with two central problems
in epistemology. 2 Justification is held to be a matter of coherence and knowledge
is taken to be true, undefeated, justified belief. Those two central concepts have
provided evolving responses to the Gettier Problem and skepticism. 3
Those two problems are not unrelated. The Gettier Problem pointed to the
felicitous, coincidental fulfillment of the justification and truth conditions of
knowledge in rather mundane situations. Skepticism-at least of the academic
as opposed to the pyrrhonian variety-is motivated by envisioning scenarios in
which we have fulfilled our epistemic responsibilities to accept only those beliefs
that pass the requirements for justification but nevertheless fall short of
knowledge because even if the beliefs are true, it is a lucky accident because,
given what justified our beliefs, it could easily be that our beliefs are false. If
the demon were sleeping, perhaps there would be hands on some occasion when
we are justified in believing that there are hands. But that would be a lucky
break for us.
I want to consider the adequacy of Lehrer's most recent treatment of the
Gettier Problem and skepticism, but it is first necessary and useful to sketch the
main outlines of Lehrer's accounts of knowledge and justification. There are
places where I might have misunderstood them and laying them out will give
him an opportunity to teach me once again. Let me say, first, that I agree with
Lehrer that knowledge is true, undefeated justified belief. But since the devil is
in the details, we must look at some of the more important features of the
account. Here are some of them [TK 129-137]:

1.1 a proposition p is personally justified for S iff P

coheres with S's evaluation system, E;

1.2 P coheres with E iff everything that is an objection to

p is either answered or neutralized;

1.3 0 is an objection to p for S on E iff it is less

reasonable for S to accept that p on the assumption
that 0 is true than on the assumption that 0 is false;

1.4 an objection, 0, to p is answered for S on X iff it is

more reasonable for S to accept that p than to accept
that 0;

1.5 an objection, 0, is neutralized by n iff (0 & n) is not

an objection to p and it is as reasonable for S to
accept (0 & n) as to accept 0 alone. 4

A key notion here is the evaluation system. An evaluation system for S

is made up of the set of states of S's acceptances, preferences and adopted
patterns of reasoning - those are states that S has, given only the interest of
pursuing truth and avoiding falsehoods. So, for example, if S believes that h,
someone owns a Ford, as a result of his interest in pursuing the truth, then "S
accepts that h" is a member of the acceptance set.
The step to knowledge is relatively easy. Roughly, undefeated justification
can be defined as personal justification that is not based on error. More
precisely: Let us define S's ultrasystem as the subset of the evaluation system
of S obtained by removing all those acceptances whose contents are false,
removing any preference for anything false over anything true, and removing all
those reasonings that fail to exemplify cogent or valid inferences. 5 Whatever still
remains justified in the uitrasystem is knowledge. In other words, if S's
ultrasystem has the power to answer or neutralize any objection to S's belief that
h, then S knows that h. For example, if S' s belief that h, someone owns a Ford,
were based upon a false belief, f, Nogot owns a Ford, then the justification for
h would be defeated because fhas been removed from the evaluation system and
S no longer has a basis for answering or neutralizing objections to h.
Lehrer's treatment of the distinction between the misleading Grabit case,
in which S does have knowledge, and Harman's Civil Rights Worker case, in
which S lacks knowledge, illustrates how his suggestion is intended to work. 6
Recall that in misleading Grabit case, S sees Tom steal a book and accepts that
he did. But Tom's demented father says that it wasn't Tom, it was Tom's twin,
Buck. In the newspaper case, S reads of the assassination of a civil rights
worker, but the paper later retracts the claim and everyone around S accepts the
revised report.
In the misleading Grabit Case, no change is required in the uitrasystem
because the reasoning did not include any false belief about what the demented
father said. The uitrasystem has the power to at least answer if not neutralize the
objection that Tom's father said that Buck did it. In the newspaper case, S would
have used (at least implicitly) the background belief that the newspaper was
reliable and once that is removed from the evaluation system there is no way to
answer the objection that the newspaper was unreliable in its reporting(s) about
the assassination. 7
The primary points I wish to raise about this account are these:

1. Lehrer's new solution to the Gettier problem appears

not to succeed.

2. Lehrer's response to the skeptic appears not to

succeed; and there is an alternative available.

Before turning to those points, I have a general question I wish to ask

about the account of knowledge. It is prompted by considering a case I will calI
the "Medical Diagnosis on-the-Cheap Case." The case goes like this: 8

I come to accept that Mr. D. Pendable will attend the next Pacific
AP A meeting roughly a year from now because I accept that he is
scheduled to give a paper at the meeting, I accept that he says he is
going, and I accept that he has kept his commitments in the past.

So, it looks like the proposition Mr. D. Pendable will attend the Pacific APA
meeting is at least prima facie justified for me. But is it personally justified
according to Lehrer's view and, if so, could it also be knowledge? If it is
knowledge, then if Closure holds, do I know that between now and next year Mr.
D. Pendable will not have a fatal heart attack?
The answers appear to be,jirst, that on Lehrer's account I am personally
justified in accepting the proposition that D. Pendable will go to the meeting a
year from now because it coheres with my E-system. I could answer (in Lehrer's
terms) the obvious objection that some people do not fulfill their commitments
even when they intend to by invoking the same strategy that Lehrer invokes in
replying to the skeptic to be discussed later, namely, by claiming correctly that
it is more reasonable for me to accept that D. Pendable will go to the APA than
to accept that he will have a fatal heart attack. Second, on the assumption that
there is no relevant false belief in my acceptance set, the ultrasystem would
sanction the personally justified belief as knowledge. And if so, unless Closure
is false, D. Pendable and I would know that he won't have a fatal heart attack (or
that he won't die in a car crash or that he won't be struck by a meteorite or ... ).
I will assume that closure holds for Lehrer, since a coherentist seems almost
honor-bound to accept it.
I think this is just a specific instance of what seems to be a general
problem with many forms of coherence theories of justification. It is this: On
this view, it appears that whether some proposition, say q, is justified, Le.,
reasonable to believe in the face of objections, will depend on the epistemic
relationships that q has to the other elements of my current evaluation system,
E - including that system's resources in answering or neutralizing objections to
q. So, whether I am permitted to or should accept a new proposition into E
depends solely upon what is currently in E. E-systems can expand incrementally
always using the immediately prior E-system. And I can use the existing E-
System to answer objections to a new, candidate proposition to be added to the
existing E-System. But what worries me about this is that it would seem to
sanction the addition of a proposition to some distant ancestor E-system in a

step-by-step process as the E-systems evolve that would not have been
permissible in one giant step from the ancestor system. But that shouldn't be.
I take it that this is the lesson from considering the Sorites situations. It could
be reasonable to add p to some system, E, and to add q to the (E + p) system, but
not reasonable to add q to E. The objection that it would not have been
reasonable for you to accept q on the basis of the original E-system can be
neutralized by the proposition that it is now reasonable on the basis of the (E +
p) system.
The case of Mr. D. Pendable and the more general possible problem with
coherence theories were presented just to get the theory before us. I will return
to them briefly at the end of the next section. Now, I want to turn to the two
main items I wish to discuss: The Gettier Problem and Skepticism.


First, to show that the conditions are not necessary: Suppose there is
some false proposition,/, that I accept. Further, suppose that I know that I accept
/ on the basis, no doubt, of introspecting my acceptance set. I am personally
justified in accepting that I accept! But since the proposition "I accept thatf'
would not be in the ultrasystem (since/is false), I would have no way to answer
or neutralize the objection that I don't accept/and, hence, "I acceptf' would not
be justified in the ultrasystem. 9
I am quite sure that with a little chisholming that problem can be fixed,
although I don't quite see how to do it because, if I understand the account
correctly, "I accept thatf' cannot remain in the ultrasystem and it seems that it
must in order for me to be able to answer the objection that I don't accept!
Indeed, the proposition "I accept f' would seem to ill-cohere with what else
would remain in my ultrasystem since it would seem to include a proposition like
the following: The best way, and a very good way at that, to determine what I
accept is to introspect my acceptance set of beliefs.
Second, the more difficult issue, and the one directly related to the
traditional Gettier Problem narrowly construed, has to do with sufficiency.
Consider a case we can call the Closed-Box Case:

Suppose I am looking at a closed box and I (think) I hear Sally tell

me there is a drawing of a triangle in the box. On the assumption
that I have good reasons for believing Sally, I would seem to be
personally justified in believing that there is a drawing of a triangle
in the box. I infer from that proposition that there is a drawing of
a plane, closed figure in the box the sum of whose interior angles
is 180 degrees.

The following propositions are in my acceptance set and are personally justified
for me:

3. Sally said that there is a drawing of a triangle in the


4. There is a drawing of a triangle in the box.

5. There is a drawing of a plane, closed figure in the box

whose interior angles sum to 180.

Now add to the story that Sally didn't say that there was a drawing of a
triangle in the box. I seriously misheard her. She actually said that there wasn't
a drawing of a triangle in the box. Presumably, then, I didn't know that there
was a drawing of a triangle in the box, even though it was personally justified for
me. But, now add one more twist to the case. Sally was wrong. There really
was a drawing of a triangle in the box!
The following propositions are in my ultrasystem:

2. There is a drawing of a triangle in the box.

3. There is a drawing of a plane, closed figure in the box

whose interior angles sum to 180.

The problem is that 2 is justified or coheres with the evaluation system because
3 provides a perfectly good, indeed an entailing, reason for 2!
Lehrer might perhaps suggest that in the acceptance set and ultrasystem
there will be such meta-evaluative propositions as: I actually (formerly)
accepted both that there was a drawing of a triangle in the box and that there is
a drawing of a plane, closed figure whose interior angles sum to 180 degrees
because I actually (formerly) accepted that Sally said there was a triangle in the
box. I might note that I no longer accept that Sally did say that and hence, I
might conclude that I am no longer justified in accepting the propositions about
what is in the box.
There are four replies to this. First, I might not have such meta-beliefs
about the reasons for my actual acceptances. If it is required that we must have
such beliefs in order to possess knowledge, too much of our knowledge would
be lost. Think of the person who believes that the Battle of Hastings was in 1066
but has no recollection of the basis on which she believes it. Does she forfeit
knowledge?lo A second and related point is that this requires that we have meta-
beliefs about acceptances and their evidential relationships in order to have
knowledge. So, in order to have knowledge that there is drawing of a triangle
in the box, we would all have to be epistemologists and even employ the notion

of "acceptances" (or an equivalent) in our epistemology. That seems a bit much

to ask. Third, if in order to know that p I have to correctly identify its evidential
basis (if it has one), it seems that calling Lehrer's view a "coherence view" is
seriously misleading. For ifthere is such a requirement, then propositions can
be ordered by evidential priority and either that ordering stops at some basic
proposition or it doesn't. If it does, this is really a disguised form of
foundationalism. If it doesn't, then this is really a disguised form of infinitism.ll
Andfinally, it simply is not true that I am no longer justified in believing that the
box contains a drawing of a triangle - at least according to Lehrer's coherence
account ofjustification. For I do have a proposition in my ultrasystem that gives
me the ability to answer or neutralize any objection to the proposition that there
is a drawing of a triangle in the box, namely: There is a drawing of a plane
closed figure whose angles sum to 180 degrees.
So it appears to me that the traditional Gettier Problem, narrowly
conceived, has re-emerged because the Case of the Closed Box is one in which
there is a true, justified acceptance which is not knowledge. The reason is
exactly the same one that led to the two original Gettier cases (Jones and his
coins; Brown and his trip to Barcelona), namely that there are situations in which
our acceptances fall short of knowledge because it is a felicitous coincidence that
we have arrived at a true, justified acceptance. 12
The problem strikes me as a general one with Lehrer's account that is not
fixed by a bit of chisholming. The reason is that a false proposition or
conjunction of false propositions can often provide very good reasons for
believing a whole set of internally coherent propositions that happen to be true.
Think ofa paranoid who happens to be on some one's hit-list. The basis for his
acceptances is false, but he might just by accident have come to accept an
internally coherent and true set of beliefs.
Note that this problem is related to the first questions we considered,
namely those connected with the case of Mr. D. Pendable and the more general
possible problem with coherence theories. For in all of these cases the evidential
ancestry ofthe propositions in question is not correctly taken into consideration.
In the case we just considered, the fact that all of the evidential support for the
internally coherent propositions rests upon a significant falsehood is lost. In the
case ofMr. D. Pendable and in the general case we have no way of keeping track
of the evidential ancestry of our acceptances. As we move further and further
from the original basis of our beliefs, although the degree of coherence of our
beliefs might increase, they actually become less and less justified. That strikes
me as a very general problem with this type of coherence theory. Evidential
priority-something that a foundationalist takes seriously-seems to be left out
of the picture. 13

Let us turn to skepticism. Once upon a time, Lehrer defended the
plausibility of academic skepticism by suggesting that when we are doing
philosophy no hypothesis should be epistemically privileged over another unless
some reason, presumably a non-question begging reason, could be provided.
Here is what he said then:

... generally arguments about where the burden of proof lies are
unproductive. It is more reasonable to suppose that such questions
are best left to courts of law where they have suitable application.
In philosophy [emphasis added] a different principle of agnoiology
[the study of ignorance] is appropriate, to wit, that no hypothesis
should be rejected as unjustified without argument against it.
Consequently, if the sceptic puts forth a hypothesis inconsistent
with the hypothesis of common sense, then there is no burden of
proof on either side ... [WNS, 53]

The passage is a bit ambiguous, but suppose that it means that when doing
philosophy there is no presumption in favor of any two conflicting hypotheses
and that in order to accept one of them one has to first have good evidence
against the other. If that is what was meant (and I'm not at all sure it was),
Lehrer no longer seems to hold this view since he now believes that we can use
"It is more reasonable for me to accept that I see a zebra than that I am asleep
and dreaming that I see a zebra" in our response to the skeptic/critic who poses
the objection, "You are asleep and (merely) dreaming that you see a zebra." In
other words, I don't have to first eliminate the hypothesis that I am asleep and
merely dreaming that there is a zebra in order to be justified in believing that I
see a zebra.
I think rejecting the requirement that there is some obligation to eliminate
all contrary hypotheses to some proposition, say h, prior to our being justified
in believing h is surely correct. For if it were a requirement-even for doing
philosophy-the road to skepticism would be way too short to be worth traveling.
Consider any two contraries, C 1 and c2 • In order to be justified in believing c l , S
would first have to eliminate c2 • And in order to be justified in believing that c2,
S would first have to eliminate c l . So, of course, S could never be justified in
believing either c 1 or be justified in believing c2. And since every contingent
proposition has a contrary, if this requirement were accepted, knowledge of all
contingent propositions would be automatically prohibited. 14 In so far as
skepticism remains a disputable and interesting philosophical position, the
skeptic cannot impose such an outrageous departure from our ordinary epistemic

So, Lehrer was right in rejecting that stringent requirement. Here is what
he now thinks we can say in response to the skeptic:

I can tell that I am awake and not asleep and dreaming now. My
experience does not feel at all like a dream and I have a distinct
memory of what preceded my present experience, leaving my hotel,
taking a cab to the zoo, buying a ticket, all of which is trustworthy
information that I am now at the zoo looking at a zebra and not
asleep and dreaming. (TK, 133)

Thus, in response to the skeptic we appeal to two types of beliefs: (i) particular
beliefs about our experiences and (ii) a general belief that such information is
trustworthy. This strategy will put off the skeptic for a while, but not for long.
The skeptic will quicky note that in answer to all of her objections, the general
proposition that my information is trustworthy keeps recurring. And she will see
that her better objection is this: "What makes you think that you are trustworthy
in the way that you evaluate information?" Or more simply: "What makes you
think that your belief acquisition methods are trustworthy?"
This should be reminiscent of Descartes' strategy on behalf of the skeptic
in the "First Meditation." There, he considers and rejects many grounds for
doubting the objection that he has trustworthy information about the world. For
example, he considers the objection "your senses have misled you in the past"
and, in Lehrer's terminology, neutralizes it by adding to it the proposition that
he can distinguish those occasions when his senses are trustworthy from those
when they aren't. Later Descartes considers the objection that perhaps he is
dreaming now and neutralizes it by responding that dream images are like
"painted representations" and have been formed as the "counterparts of what is
real and true." So, if we limit our acceptances to those "simple" properties
shared by dream-images and waking images, we still can arrive at knowledge.
Thus, Descartes answers these initial objections by neutralizing them in the way
recommended by Lehrer.
But then, the Cartesian dialectical skeptic raises this doubt:

... nevertheless in whatever way they suppose that I have arrived

at the state of being that I have reached ... since to err and deceive
oneself is a defect, it is clear that the greater will be the probability
of my being so imperfect as to deceive myself ever, as is the
Author to whom they assign my origin the less powerful. 15

In other words, the skeptical objection now is this: "You do not have trustworthy
belief acquisition methods."
This is the heart of academic skepticism. The academic skeptic requires
that there be some way within the E-System to answer or neutralize this

overarching objection. Famously, Lehrer seeks to provide the answer to the

academic skeptic in this manner: He identifies a proposition (T) namely, I am
trustworthy (worthy of my own trust) in what I accept with the objective of
accepting something just in case it is true. And, he thinks that it is more
reasonable to accept T than to deny it or withhold acceptance. In other words,
T is in the E-system. The following is a very close paraphrase (including direct
quotations) that presents why he thinks T is acceptable. 16

Any argument for (T) would involve a kind of circularity for I

would have to appeal to (T) in the argument. So such an argument
will not succeed in answering the skeptic, because to use a premise
to prove something to a skeptic who challenges it violates the rules
of rhetoric. But to explain why it is reasonable to accept what we
do, the circle may be virtuous. Ifwe have a principle that explains
why it is reasonable to accept what we do, it is a virtue rather than
a vice that it should at the same time explain why it is reasonable
to accept the principle itself. The other alternative is that the
principle should be a kind of unexplained explainer that explains
why it is reasonable for us to accept other things we accept and
then falls mysteriously silent when asked why it is reasonable to
accept the principle itself. The loop by which the principle
explains why it is reasonable to accept the principle as well as other
things is worthy of epistemic praise rather than rhetorical
condemnation (TK, 143).

Well, loops could be just as vicious as circles, so that metaphor won't

carry the day. The issue here is to get clear about what the reply to the skeptic
actually consists in. If asked by the skeptic why I accept T, I reply that by
accepting T my acceptance system is more coherent than it would be were I to
deny T or withhold acceptance T. If I deny T, I would be saying that I accept
some proposition, say p, but that I accept that I wasn't trustworthy in doing so.
If I withhold T, I would be saying that I accept p but that I remain neutral about
whether I am trustworthy in accepting that p. There is one other way to be
coherent, namely, not to accept any proposition at all. With some important
caveats that is exactly what the Pyrrhonians did, and it is the view I want to
To begin, note that Lehrer says that "any argument for (T) would involve
a kind of circularity for I would have to appeal to (T) in the argument. So such
an argument will not succeed in answering the skeptic, because to use a premise
to prove something to a skeptic who challenges it violates the rules of rhetoric .I
agree that if an argument for T employed T as a premise, it would "violate the
rules of rhetoric." But as I will argue shortly, I do not think that an argument for
Twill, of necessity, employ T. Nevertheless, what is crucial here is that Lehrer

says that the "loop" explains why it is "reasonable" to accept T, and hence the
viciousness associated with an argument for T does not arise. Thus, a central
question becomes this: What is the crucial difference between an explanation
that shows why it is reasonable to accept a proposition and an argument for the
truth of the proposition?
Lehrer is correct that there is a crucial difference between explaining p and
giving an argument for p. Doing the first is completely consistent with assuming
that p. Both I and my critic agree that p is true. In asking for an explanation of
p, the critic is asking why p is true. The rhetorical situation is such that p's truth
is not at stake. So, explaining that p will not beg the question-since the
question is not whether p is true, but why p is true.
But that is not parallel to the worries concocted by the skeptic. The
skeptic is not willing to grant that T is true. She wants to know whether T is
true; not why T is true. In other words, giving an explanation of why T is true
will neither answer nor neutralize the skeptic's challenge. It will not answer it,
because it presupposes that T is true. It doesn't neutralize it because it does not
provide a basis for accepting ~ T along with something else that overrides the
impact of ~T alone. In short, the distinction between explaining and giving an
argument for will not help here. The skeptic isn't asking for an explanation of
T. The skeptic is challenging the truth of T.
That being said, it still might seem that the more satisfyingly coherent
thing to do is to accept T. And coherence is the essence of justification,
according to Lehrer. After all, wouldn't there be something deeply incoherent
in thinking that my belief formation methods are unreliable or even just that they
might not be reliable. Ernest Sosa has argued for just that point. Here is what he
says (this is a close paraphrase):
But now suppose that by using way W of forming beliefs ... we
arrive at the conviction that W is our way of forming beliefs. Now,
so long as we do not go back on that conviction, does that not
restrict our coherent combinations of attitudes? Take: (e) Believe
that W is my overall way of forming beliefs. And compare (f)
Believe that W is reliable, (g) Deny that W is reliable and (h)
Withhold that W is reliable. It is not evident that (e) & (f) would
be more satisfyingly coherent than either of (e) & (g) or (e) & (h)? 17

My primary question is this: Is it necessarily more satisfyingly coherent to

accept T rather than withhold T? My primary answer is that it is not, and by
sketching the answers to three secondary questions, I hope to show that the
primary answer is at least plausible. Those three questions are:

1. Assuming that we accept some propositions, must a

response to the skeptic involve answering or
neutralizing the academic skeptic's objection?

2. Ifwe do choose to respond by giving an argument for

T, must that argument beg the question?

3. Must we be forced into providing any response at all

to this type of skeptic?

Let us take them up in order.

1. Assuming that we accept some propositions, must a

response to the skeptic involve answering or
neutralizing the academic skeptic's objection?

Even if we assume that we accept some propositions that the skeptic challenges,
we are not forced either to answer or to neutralize the skeptic's objection. There
is another, I think better, strategy available, namely to question the basis on
which the skeptic believes or even raises the objection. The skeptic asserts that,
or at least seems to be questioning whether, our methods of belief acquisition are
unreliable. She dares us to come up with an argument for T that doesn't
presuppose T. But why are we limited to either answering the objection or
neutralizing it on the basis of something else we believe? Why can't we simply
ask the skeptic for her reasons for thinking that T is false?
Suppose that we thought that all stated arguments about what is a subject
of genuine controversy were either question-begging or based upon arbitrary
premises-ones which cry out for a defense before the conclusion is accepted.
That was what the Pyrrhonians thought and their strategy in dealing with the
dogmatists-whether they were the type who accepted that we had knowledge
or the Academic Skeptic type who accepted that we didn't have
knowledge-was to ask for the reasons for the accepted proposition in order to
show in case after case that either the offered reasons begged the question or
begged for further defense.
The Pyrrhonians would point out that either the skeptic will or will not
have some argument like the following to present:

1.1 1fT, then some condition, C, holds.

1.2 C does not hold.

:. Not T

Consider the first alternative: If the skeptic has no argument with this
form, why should we take the challenge seriously? After all, we have already
said that we need not eliminate all contrary hypotheses before we are justified in
believing a given hypothesis. And if! don't have to eliminate all of them before

arriving at the view that the animals are zebras, why would I have to eliminate
all of them after I have arrived at the belief that the animals are zebras?
So, there must be something special about this challenge that requires that
we answer it. It cannot merely be, as some contextualists have suggested, that
a hypothesis becomes a relevant alternative to h (one which must be eliminated
prior to being justified in believing that h), whenever it is raised-even raised
with a serious tone. IS Consider the zebras again. Suppose the skeptic says,
"Well how do you know that those things are not cleverly disguised aliens from
a recently discovered planet outside our solar system? Or that they are not newly
invented super-robots? Or that they are not members of the lost tribe of Israel
who have been practicing the zebra-disguise since the 8th century BC, and
consequently, have gotten very good at it?" Those objections are so far-fetched
so that even if someone advancing those alternatives happens to believe them,
there appears to be no reason why one should have to rise to the bait and
eliminate those alternatives either before or after one has arrived at the belief that
the animals are zebras. And finally, isn't the skeptical hypothesis-that we are
not in the actual world but rather in one which just seems identical to it-just as
far fetched? What is so special about that far-fetched hypothesis?19
Now to the second alternative: If the skeptic does have such an argument,
we can examine it. And we might just find that the Pyrrhonians were right, at
least about this argument. 20 Thus, I think we are not forced either to answer or
to neutralize the skeptic's claim on the basis of some of our other acceptances.
Just because the challenge is raised imposes no burden on us to answer or
neutralize it.
This is the first step in sketching the alternative response alluded to earlier.

2. Ifwe do choose to respond by giving an argument for

T, must that argument beg the question?

Suppose we prefer to answer the skeptic. After all, Lehrer thinks what we are
justified in believing depends, in part, on what we prefer. Consider this
argument schema:

1. My methods of belief formation are M-type methods.

2. M-type methods are reliable.

My methods of belief formation are reliable.

An argument instantiating that schema need not beg the question. Of

course any person giving such a defense of their belief forming mechanism will,
on grounds of consistency in practice, use methods M in coming to believe that
her methods of belief formation are reliable. But what's the alternative? To use

some methods other than M which she thinks are not reliable? That would make
her practice inconsistent with her beliefs.
The fallacy of begging the question in reasoning consists in using a
conclusion of an argument as a basic premise-whether overt or suppressed.
Question-begging arguments can provide no additional reason for accepting the
conclusion beyond those already available for the basic premises. But it is not
a fallacy of reasoning to employ what one takes to be the reliable belief
formation rules in arguing for the claim that those very belieffonnation rules are
reliable. Indeed, to do otherwise would be to fail to practice what one preaches.
Now, the academic skeptic might deny one or both of the premises of the
argument. And if the pyrrhonian skeptic is correct, any such argument will
either beg the question or rely upon premises that require further defense.
Nevertheless, the point here is that the argument for the reliability of methods M
need not beg the question.

3. Must we be forced into providing any response at all

to this type of skeptic?

I have already partially addressed that question by arguing that even if we accept
some proposition, say p, we need not answer or neutralize the skeptic's objection
to p. We can challenge the skeptic's basis for accepting the objection. But, here,
I mean to be asking whether we need ever be placed in the position of defending
acceptances at all.
Of course, we have that burden only if we accept some propositions that
could be challenged by the skeptic. My question is this: Why is it assumed that
we must accept anything-at least anything that requires inferential support?
It is absolutely crucial to note that for Lehrer to accept p is not the same
as to believe p. Knowledge entails acceptance not belief. To accept something
is to hold it true precisely because doing so promotes our supposed goal of
accepting a proposition iff it is true. According to Lehrer we can even know
something to be true which we believe is false (again paraphrase and quotation):

A politician may convince you of the truth of what he says when

you know he is untrustworthy . . . He is warm, human, and
comforting, whereas the data are cold, mathematical, and
distressing. How can you resist? You do believe him but you
know that the economy is slipping. This conflict between belief
and knowledge is explained by the fact that we are "divided into
two systems." One system is truth-seeking. The other is the yield
of habit, instinct, and need ... We do not accept what the pol itician
tells us as the bona fide truth even if we cannot help but believe
him. (TK, 124-5)

But what if we thought that our epistemic equipment was not trustworthy enough
to arrive at unobjectionable (in Lehrer's sense of that term) truths in matters as
complex as evaluating the weight of reasons on each side of a complex issue?
What if we thought that we are only trustworthy enough to identifY what seems
true? By "seeming true" I do not mean some sort of hesitating endorsement of
the proposition's being true. Rather, I mean that the proposition has the look of
truth about it. It seems to satisfY what appears to make a proposition true.
Further, what if it seems to us that our goal is not and should not be
unobjectionable propositions but rather provisionally justified belief in what
seems true? That is a goal that seems reachable and does not bring with it the
dangerous flirtation with dogmatism that easily accompanies the preference for
being in a situation in which we are able to answer or neutralize all objections on
the basis of what we currently accept.
Recall part of what I cited that Lehrer wrote twenty years ago: "Should J
ever reach the point at which I am disinclined to seek criticism and amend my
views in light of it, I shall take to writing fiction . . . and give up writing
philosophy altogether." Now I suppose that one could seek criticism in the way
that an arrogant gunslinger would seek out opponents believing that victory will
be his. Such macho philosophic gun fighters have a virtue of their own. But this
isn't Lehrer's motivation. He seeks criticism because he expects that it will
improve the epistemic worth of his beliefs. That sounds like a person who does
not accept that he is ever in a position to answer or neutralize all objections to
what he believes. Such a person would never accept the meta-view that he had
personally justified acceptances.


Let me briefly summarize where the answers to the three subsidiary

questions posed above have left us. We need not "answer"or "neutralize" (in
Lehrer's sense of those expressions) the academic skeptic's challenge to our
acceptances. We can show that the arguments employed by such a skeptic do not
succeed. Further, we can coherently not accept T. But abstaining from
accepting T does not necessarily lead to embracing academic skepticism. Such
skeptics accept that T is false. The third alternative-the pyrrhonian alternative
-is to withhold accepting T. That would bring with it withholding any
proposition whose defense against objections depended upon accepting T. It
strikes me that such a view is satisfYingly coherent and, indeed, better fits
Lehrer's own preference system than the model of justification he has
suggested. 21

* I want to thank Anne Ashbaugh and especially Keith Lehrer for their help with this paper. Keith
and I have been talking and emailing about some ofthese issues for many years. In particular, an
ancestor of this paper was given at the Pacific AP A meeting (March, 200 I). He replied to it there
and also kindly sent me some further comments later. Keith has taught me a lot over the years and
I expect that his replies to this version of the paper will continue to make me think more about the
issues. That's the fun of it all.
I. Keith Lehrer: Profiles (Dordrecht: D. Reidel Publishing Company, 1981), ed. Radu Bogdan, p.
2. In this paper I will be relying primarily on these works by Lehrer:

Theory 0/ Knowledge (Boulder:Westview Press, 2000), second edition [TK]

"Why Not Scepticism?" originally published in The Philosophical Forum, 2.3
(1971), 283-298. It has been anthologized in many places. Citations to this
article are to The Theory 0/ Knowledge (Belmont, CA: Wadsworth Publishing
Company, 1993), edited by Pojman [WNS].

Self-Trust: A Study o/Reason, Knowledge and Autonomy (Oxford and New York:
Oxford University Press, 1997) [ST]
3. By "skepticism" I will be referring to Academic Skepticism unless otherwise noted. Using
Lehrer's terminology, the Academic Skeptics thought that it is more reasonable to accept that we
don't (or can't) have knowledge in those areas in which it is generally thought that we do have
knowledge than to deny or withhold such a judgement about the scope of our knowledge. This is
to be contrasted with the Pyrrhonian Skeptic who would withhold such an acceptance (and its
4. One might worry here that it could never be more reasonable for S to accept a conjunction than

it is to accept one of the conjuncts (unless it entails the other) because the probability of the
conjunction will be less that the probability ofthe conjunct. But I think the answer is this: Since
the goal is to gain truth and avoid falsehood, by accepting the conjunction one has a greater chance
of accepting more truths than one would by merely accepting the one conjunct.
j I am using "valid" in the standard way to indicate an argument whose pattern is such that any
argument with that pattern not possible for the premises to be true and the conclusion false. I am
using "cogent" in the way defined by Richard Feldman in Reason and Argument (Upper Saddle
River, New Jersey: Prentice Hall, 1999 second edition) to refer to good patterns of inductive
6. The best source for a full and illuminating discussion ofthe various counterexamples connected

with the early Gettier Problem literature is Robert Shope's, The Analysis o/Knowledge (Princeton,
New Jersey: Princeton University Press, 1983).
7 It was unreliable at least in this case because too many of its reports were false.
8. This example, or one very similar to it, was suggested to me as a possible counterexample to

some forms of closure by John Hawthorne. I don't think it works against closure because I think
a proper account of justification would have it that you are not justified in believing that D.
Pendable will attend the meeting. But this is not the place to present that account.
9. This example, or one very much like it, was developed by Robert Shope, op cit, pp. 52, 71, 102.
It was also used by Alvin Plantinga in Warrant: The Current Debate (New York and Oxford:
Oxford University Press, 1993, pp. 220-1.) I argued that this counterexample would not work
against my version of the defeasibility theory in "Warrant, Proper Function, Reliabilism, and
Defeasibility" in Warrant in Contemporary Epistemology (London: Rowman & Littlefield
Publishers, 1996), pp. 121-2.
10 Examples like this one were given by Colin Radford, "Knowledge - By Example," Analysis,

11. Of course, I would be very pleased if some epistemologist other than I thought infinitism is the

correct account of the structure of justification! (See fn. 13.)

12 See my paper "A Proposed Definition of Propositional Knowledge," Journal of Philosophy

13. I think neither a coherentist nor foundationalist approach to justification is correct. The third

alternative solution to the regress of reasons problem-namely infinitism-strikes me as the best

solution and I have argued as much in several papers. See for example, "Human Knowledge and
the Infinite Regress of Reasons," Philosophical Perspectives, 13, J. Tomberlin (ed.), 1999,297-
325; "Why Not Infinitism?" in Epistemology: Proceedings of the Twentieth World Congress in
Philosophy, Richard Cobb-Stevens (ed.), 2000, vol 5, 199-208; and "How a Pyrrhonian Skeptic
Might Respond to Academic Skepticism," The Skeptics: Contemporary Essays, (ed. Steven Luper
Ashgate Press, forthcoming.)
14. Take any contingent proposition, h. The proposition (~h & x) is a contrary, where x is any

proposition at all, including (h v ~h).

IS. Descartes, "First Meditation" in The Philosophical Works of Descartes (Dover Publications,

193 I) trans. Haldane and Ross, p. 147.

16. The cited text is TK. But a similar argument occurs in ST, chapter I.

17. Ernest Sosa, "Philosophical Skepticism and Epistemic Circularity," Proceedings of the

Aristotelian SOCiety 68 (1994), p.107.

18. This is essentially the view put forth by Stewart Cohen in "Knowledge, Context and Social

Standards," Synthese, 73 (1987), 3-26 and his "How to be aFallibilist" Philosophical Perspectives,
2 (1988), 91-123, David Lewis in "Elusive Knowledge," Australasian Journal ofPhilosophy, 74
(1996),549-67 and Keith DeRose in "Solving the Skeptical Problem," Philosophical Review, 104
(1995), I-52 and his "Contextualism and Knowledge Attributions," Philosophy and
Phenomenological Research, 52 (1992), 913-929.
19 Let me add that Descartes does not simply raise the hypothesis-he gives a reason for thinking

that it is at least plausible; namely, that the degree of perfection of his epistemic equipment cannot
be higher than the degree of whatever is its cause and, at that point in the Meditations, he has no
reason to think that its cause is sufficiently perfect. Thus, in order to see whether we should take
this challenge seriously, we would have to examine the presuppositions ofthis skeptical challenge.
This is obviously not the place for that.
20. See my paper "Skepticism and Closure: Why the Evil Genius Argument Fails," Philosophical

Topics, 23.1, Spring 1995,213-236, and "How a Pyrrhonian Skeptic Might Respond to Academic
Skepticism," op. cit.
21. This does have the consequence that neither foundationalism nor coherentism is the correct

account of reasoning. As mentioned in fn. 13, I have argued for infinitism in several papers.
Chapter 18



David A. Truncellito
Arkansas State University


Keith Lehrer's account of justification and knowledge is unique and

brilliant. Lehrer's notion ofjustification in terms of coherence can be construed
as a matter of beating or neutralizing competitors 1, and his account of knowledge
can be construed as a matter ofundefeatedjustification. 2 And competition and
defeat can, in turn, be understood via the rubric of the justification game. This
is a picture that has persisted throughout Lehrer's epistemological
writings-from his earliest efforts to respond to the Gettier problem to his
current works-in-progress-although it has evolved over the years.
However, although his theory exhibits these unique features, Robert
Shope has grouped Lehrer together with a wide range of other contemporary
philosophers by accusing him of committing the conditional fallacy. While
Shope seems to have argued convincingly that many philosophers are indeed
guilty of this fallacy, I shall argue in what follows that Lehrer escapes this



Let us begin by briefly reviewing Lehrer's account of justification and

knowledge. For Lehrer, justification comes in two levels: personal justification
E.J. Olsson (ed.). The Epistemology of Keith Lehrer, 299-308.
© 2003 Kluwer Academic Publishers.

and undefeated justification; knowledge reduces to undefeated justification. Let

us first consider personal justification, which Lehrer has defined as follows:

S is personally justified in accepting that p at t if and only if

everything that is an objection to p for S on the basis of the
acceptance system of S at t is answered or neutralized on the basis
of the acceptance system of S at t. 3

That is, personal justification is a matter of answering or neutralizing all

objections to one's acceptance that p; in other words, p must cohere with the rest
of one's acceptance system. Note that it is S's acceptance system that is the
relevant set of propositions with which p must cohere. Also, note that to accept
that p is not (merely) to believe that p4, but rather to accept that p with the aim
of attaining truth and avoiding error with respect to p.5 S' s acceptance system,
then, comprises everything S accepts:

The acceptance system of S at t is by definition the set of states of

acceptance of S described by statements of the form-S accepts
that p-attributing to S just those things S accepts at t with the
objective of obtaining truth and avoiding error with respect to the
content accepted, that is, with respect to the content that p.6

Personal justification, then, arises when one has the resources within one's
acceptance system to reply to any challenge a skeptic might raise. This
dialectical formulation of personal justification suggests the justification game. 7
In short, the game is played between S (the claimant), who claims to have
knowledge, and a critic (or skeptic), who challenges S' s claims to knowledge by
raising objections. S wins a round by adequately answering or neutralizing the
critic's objection; S wins the game, and achieves personal justification, by
winning every round.
Notice, though, that victory in the justification game does not entail that
S knows that p. This is for the obvious reason that while S' s acceptances were
formed with the goal of attaining truth and avoiding error, he might have failed
to meet that goal in some cases; indeed, given that S, like the rest of us, is
fallible, it is likely that he did fail to meet this goal in some cases. But since
victory in the justification game, and thus personal justification, are based on S' s
acceptance system, S's belief that p, although personally justified, might be
based on falsehoods. And, while personal justification is a noteworthy
achievement, belief based on falsehoods cannot constitute knowledge. This is
the motivation behind the' isolation objection' against coherentist-and, indeed,
any purely internalistic-accounts of justification. 8

Lehrer's remedy to this possible problem is unique and elegant. It begins

with the simple observation that "complete justification is personal justification
that is not based on error.,,9 However, the unique flavor of Lehrer's account of
personal justification suggests a unique way of incorporating this relatively
straightforward observation. To avoid reliance upon error, let us imagine a new
system, which is obtained by retaining all truths and eliminating all falsehoods
from S's acceptance system. Call this new system S's ultrasystem. IO
We can now playa modified version of the justification game, the ultra
justification game. I I The game is played as before, except that the players are
now the claimant and the uItracritic. Since the ultra justification game is based
on S's uItrasystem rather than S's acceptance system, the uItracritic can make a
new kind of move: she may prohibit S from appealing to any of her false
acceptances in her efforts to reply to objections. 12 We now see the second level
of justification, viz. undefeated justification: S wins a round of the ultra
justification game by adequately answering or neutralizing the ultracritic's
objection; S wins the game, and achieves undefeated justification and
knowledge, by winning every round. 13


Lehrer's notion of the ultrasystem clearly serves to avoid the worry that
accompanied personal justification, namely the possibility of acceptance being
based on error. However, it exposes him to a new form of objection-one which
might perhaps seem to be less central to the matter of knowledge, but an
important objection nonetheless. Namely, it runs the risk of jeopardizing one's
knowledge of one's own mental states, which surely no account of knowledge
should do.
Robert Shope has suggested a common sort of mistake made by a number
of contemporary philosophers, which he calls the conditional fallacy and
characterizes as follows:

A mistake one makes in analyzing or defining a statement p by

presenting its truth as dependent, in at least some specified
situations, upon the truth (falsity) of a subjunctive conditional 0
of the form: 'If state of affairs a were to occur, then state of affairs
b would occur', when one has overlooked the fact that, in some
specified situations, statement p is actually true, but, if a were to
occur, then it would be at least a partial cause of something that
would make b fail to occur (make b occur).14

Shope then claims that Lehrer is among those many who are guilty
of committing this fallacy. IS Others, including Peter Klein l6 , Alvin
Plantinga and John Pollock, have offered similar objections to Lehrer's
theory. Let me attempt to unify these objections by describing the
purportedly offensive sort of case.
Let us suppose that S accepts some p, such that p is false. Then, of
course, it follows that S does not know that p. But let us suppose further
that S accepts that S accepts that p. This latter acceptance (call it S's
acceptance that q), intuitively, constitutes knowledge. 17 However, because
ofthe way in which Lehrer has characterized undefeated justification, his
account has the consequence that S does not know that q, which is a
serious mark against the view. Recall that undefeated justification-and
knowledge, which is equivalent -requires the ability to win the ultra
justification game. We could, then, characterize knowledge in terms of a
subjunctive conditional analysis, so as to see just how it is meant to be
vulnerable to the conditional fallacy:

S knows that p iff if all falsehoods were to be eliminated from S's

acceptance system, then S would be able to answer or neutralize
any objections to p.

But in the case we are considering, we see a problem for Lehrer's account. For,
p is among the falsehoods that are removed from S's acceptance system in
constructing S's ultrasystem! It follows, then, that S does not know that q (i.e.
does not know that he accepts that p).


My reply is simple and straightforward: the objection appears to rest on
a misunderstanding of Lehrer's account. I shall endeavor, then, to expose this
misunderstanding and thus dissolve the objection. I shall do this by sketching
a few rounds of the ultra justification game as they would be played in the case
under discussion.
To show that Lehrer's account has the consequence that S does indeed
know that q, I shall show, in turn, that his account has the consequences that S
accepts that q, that q is true, and that S is completely justified in accepting that

4.1. S accepts that q

The question we are facing is whether S's acceptance that q is to count as

knowledge on Lehrer's theory. As such, it is relatively uncontroversial that S
does indeed accept that q; the question regards the status of that acceptance. So,
while'S accepts that p' is not part of the description ofS's ultrasystem, 'S accepts
that S accepts that p' is part of that description. We need to be careful on this
point, though. Note that Lehrer's characterization of the ultrasystem is such that
S's acceptances themselves, and not the contents of those acceptances, are
described. So, while'S accepts that p' does not accurately describe any part of
the ultrasystem, 'S accepts that S accepts that p' does; that is, the acceptance that
S accepts that p is among S's acceptances which have not been eliminated in
forming the ultrasystem.

4.2. Q is true

In order to evaluate the truth of S's acceptances about the world, we need
to note whether they hold true of the actual world. In particular, in order to
evaluate the truth of S's acceptances about S's acceptance system, we need to
note whether they hold true ofS's actual acceptance system. But q is indeed true
(i.e. S does accept that p), for p is a member of S's acceptance system, even
though it is not a member of S's ultrasystem. We need to recall that the
ultrasystem is a new system, and that falsehoods have been removed from the
ultrasystem, but have not actually been removed from the acceptance system.
Indeed, Lehrer explicitly makes note of this point in his latest book (although
not, to my knowledge, in any of his earlier work): "The ultrasystem,
nevertheless, is required to acknowledge the existence of the eliminated states
of acceptance ... in the original evaluation system, for it is true that they are states
of the person.,,19 In other words, the ultrasystem is nothing more than a
theoretical construct, designed with the purpose of helping us to evaluate the
justificatory status of S's acceptances. But even while we are conducting this
evaluation, S's acceptance system remains unchanged: it continues to include
any falsehoods which might have been in it previously.

4.3. S's acceptance that q is completely justified

Let us suppose that S's acceptance that p is personally justified. This is,
I take it uncontroversial, since personal justification depends upon the
acceptance system but not the ultrasystem. However, it will be informative to

ask why S's acceptance is personally justified. Such an acceptance normally

derives its justification on the basis on introspection. 20 To make this explicit,
then, let us suppose that S's justification for believing that g is supported by his
following acceptances:

gl: I introspect an acceptance that p.

g2: Introspection is a trustworthy source of acceptances about my mind.

The exact details might be different in the case of a particular agent. The
crucial point to note, though, is that however the correct account might
proceed, S's acceptance that q is not justified by p.
Now, let us imagine a round of the ultra justification game. The
ultra system will contain q, ql' and q2' but not p, since all falsehoods have
been removed in constructing it.

Claimant: g (i.e. I accept that p.).

Critic: p is false. Not-g (i.e. You do not accept that p).
Claimant: It is more reasonable for me to accept g than to accept
not-g, on the grounds that gl and g2'

Note that the claimant's response is exactly as it would be in the ordinary

justification game, since the critic cannot make the claimant remove ql or
q2 from the ultrasystem. But then, it appears that the critic's move does
not constitute an unsuperable objection; indeed, the critic's objection is
entirely irrelevant to S's acceptance that q. For, S's acceptance that q is
based upon ql and q2' and these remain in the ultrasystem. S's acceptance
is not based upon p, so the critic's request that S remove p (and, perhaps,
replace it with not-p) doesn't generate an objection which can't be
answered or neutralized.
Consider another round of the ultra justification game:

Claimant: g (i. e. I accept that p).

Critic: Not-g (i.e. you do not accept that p).

The critic's move is illegitimate, since not-g is false (see section IV.2 above),
and thus not available to the critic as an objection.
So, the only possible objections one might imagine that the critic would
raise are not-p, which does not constitute an objection to S's acceptance that g,
and not-g, which is false and thus not a legitimate objection. S does, after all,
have undefeated justification and knowledge that p.

I should pause at this point to consider an alternate construal of this

objection against Lehrer's theory, which was suggested to me by an ancestor of
Klein's paper.21 Note that S's ultrasystem contains the acceptance that S accepts
that p, but not the acceptance that p. It appears, then, that S's ultrasystem
contains a practical contradiction along the lines of Moore's paradox. However,
there is no reason for this to worry us. Since the ultrasystem is a mere
theoretical construct, there is no need for it to be free of inconsistencies. If S's
acceptance system contained such an inconsistency, that might be troublesome;
but this would be a problem for S, not for Lehrer. 22


There may be other problems with Lehrer's account of justification and

knowledge, as several ofthe other contributors to this volume suggest. I do not
claim to have shown that it is the correct theory, but merely to have defended it
against Shope's objection.
Perhaps Lehrer's description of the ultrasystem was not sufficiently
perspicuous-he might, for instance, have described the members of the
acceptance system in terms of their content (e.g. as 'p' rather than as'S accepts
that p'), and he might have been more clear that the ultrasystem is distinct from,
albeit related to, the acceptance system-and thus led some readers to overlook
the fact that it is a new system, and a theoretical system at that, rather than an
actual modification to the acceptance system. I hope that I have helped to clarify
the matter. 23


'Or, in the terminology of Lehrer (2000), answering or neutralizing objections.

2 Lehrer (2000), pp.169ff.
3 Lehrer (2000), p.137. See also Lehrer (1997), p.30; Lehrer (1990), p.126; Lehrer (1974), p.198.
In Lehrer (1974), this notion was referred to as complete justification, but it is clearly the ancestor
of personal justification. Indeed, Lehrer concedes that the term 'complete justification' should
perhaps be reserved for the kind of justification that is sufficient for knowledge, and that this
notion might be better called 'subjective justification'. (Lehrer (1974), p.214)
4 Lehrer vacillates as to whether acceptance is a kind of belief (Lehrer (1990), p.ll) or is a kind
of mental state distinct from belief (Lehrer (2000), p.14). We might want to say that acceptances
are those beliefs which have been positively evaluated with regard to epistemic purposes, but we
might also want to allow that S accepts that p without believing that p. This is an interesting
matter, and raises issues of doxastic voluntarism (among others). However, for Lehrer's and our
purposes, we need only note that it is acceptances and not beliefs that are relevant to justification.
s Lehrer (2000), p.13, p.130. See also Lehrer (1997), p.3; Lehrer (1990), p.11.

6 Lehrer (2000), p.130; see also Lehrer (1990), p.117. In Lehrer (1974), this was referred to as the
corrected doxastic system (p.190).
7 For a detailed description, and several rounds, of the justification game, see Lehrer (2000),
pp.132ff; Lehrer (1990), pp.119ff. I'll rehearse a few rounds of the game in section IV below.
S Lehrer is explicit that, although his account is often taken to be an internalistic theory of
justification, it is not; it is what he prefers to call a 'match theory'. "There must be a match
between what one accepts as a trustworthy guide to truth and what really is a trustworthy guide to
truth ... To obtain knowledge one needs the right mix of internal and external factors." (Lehrer
(2000), p.I72) However, personal justification is a purely internalistic component of Lehrer's
account, in that it is indexed to a subset of S's internal states (viz. S's acceptance system), and that
is why it is vulnerable to the worry raised above.
9 Lehrer (2000), p.l53. See also Lehrer (1974), p.215.

10 Lehrer (2000), pp.l53-4. In Lehrer (1974), this was referred to as the verific alternative to the

corrected doxastic system. (p.224)

11 Lehrer (2000), pp.154ff; Lehrer (1990), pp.14lff.

12 She may also disallow unsound reasonings (Lehrer 2000), p.160), although that will not concern
us here.
13 Lehrer actually added a third, intermediary stage, in Lehrer (1990), viz. verific justification, with

its attendant notions of the verific system and the verific justification game. Also, in that work,
the ultrasystem was a set of systems rather than a single system. Given that this was not the case
in Lehrer (1974), and no longer is the case in Lehrer (2000), I shall ignore it for the purposes of
my discussion. In any event, the spirit of the proposal is consistent throughout the development
of Lehrer's theory, and that should not be obscured by this wrinkle.
14 Shope (1978), pp.399-400. This is actually Version 2 of the fallacy, which is the version Shope
accuses Lehrer of committing.
15 See Shope (1983), pp.48ff, pp.53ff, and especially pp. 73-74.
16 Most recently in his "Coherence, Knowledge and Skepticism", in this volume.

17 The exact details of the way in which S comes to be justified in his acceptance that he accepts
that p can be ignored here. Presumably, though, the explanation will have to do with the fact that
S introspected such an acceptance and that introspection is a trustworthy source of knowledge of
one's own mental states. If one were to deny that S knows that S accepts that p, then of course the
objection is rendered irrelevant. I shall discuss the details ofthis case somewhat more thoroughly
in the following section.
18 Strictly speaking, since knowledge reduces to undefeated justified acceptance, I do not need to
show that q is true. However, I shall do so nonetheless for the sake of clarity and perspicacity.
Since the objection of the previous section could be interpreted so that the undesirable
consequence of Lehrer's theory is that q is false or that q is not completely justified, I shall respond
to both possibilities.
19 Lehrer (2000), p.160; see also Lehrer (2000), p.168. Note that the evaluation system is the

system composed of all of S's acceptances, as well as his preferences and reasonings. What is
relevant for our purposes, though, is that the acceptance system is a subset of the evaluation
20 Klein, in the very paragraph in which he raises his objection in Klein (forthcoming), agrees that

this must be the case, so this is a point of agreement between the objector and myself.
21 Presented, under the same title, at the occasion of the 2001 Pacific APA session honoring

22 And, given that Lehrer does not insist that coherence requires consistency, it might not even be

that significant a problem for S.

23 Thanks are due to Keith Lehrer for discussion on this topic, and for providing two necessary

conditions for this paper: namely, developing the coherence theory of knowledge and helping me
to develop my philosophical abilities in general, and the particular thoughts which led to this

paper, via his guidance as my advisor and mentor. I consider myself extremely fortunate to have
been Keith's student, and am extremely happy to have him as a friend.
I might note that as I was working on this paper, Keith came to the Mid-South to give a
talk at my University, and we spent several days talking about philosophy, walking along the
Mississippi River, listening to blues, and generally enjoying each other's company. The pleasant
memories of his visit are stilI foremost in my mind as I type these words.

Klein, Peter, forthcoming. "Coherence, Knowledge and Skepticism." (to appear in this volume)
Lehrer, Keith, 1974. Knowledge. Oxford: Clarendon Press.
Lehrer, Keith, 1990. Theory of Knowledge. Boulder, CO: Westview Press.
Lehrer, Keith, 1997. Self-Trust: A Study of Reason, Knowledge, and Autonomy. Oxford:
Clarendon Press.
Lehrer, Keith, 2000. Theory of Knowledge (2nd ed.). Boulder, CO: Westview Press.
Plantinga, Alvin, 1993. Warrant: The Current Debate. Oxford: Oxford University Press.
Shope, Robert, 1978. 'The Conditional Fallacy in Contemporary Philosophy." Journal of
Philosophy 75: 397-413.
Shope, Robert, 1983. The Analysis of Knowing: A Decade of Research. Princeton: Princeton
University Press.
Chapter 19



Keith Lehrer
University ofArizona

The essays in this volume on my epistemology are a testimony to the way

in which first rate philosophers have involved themselves with my published
work in the theory of knowledge stretching over four decades. It is extremely
gratifying to me, for the contributors to this volume are among the most talented
philosophers of the present day. The contributors to this volume have
concentrated on both the general coherence theory I have presented and the
analytical details of it. Though they often disagree with the theory and raise
objections to it, sometimes in a quite critical manner. I consider their careful
examination of the theory as an important contribution to the study of what I
have done. Where they disagree most avidly is where I suspect from my own
study of the history of philosophy, especially the criticism of Reid against Hume,
I have probably made the most important contribution to epistemology even if
they are successful in refuting what I have written. Moreover, if they are
definitive in their refutation, I shall claim part of the credit in formulating a
theory with sufficient clarity and precision so that it admits of definitive
refutation. Having expressed my admiration, my genuine admiration, for the
philosophical acumen of my critics, I do wish to reply to what they have said. I
hope that I shall reply fairly and do justice to their insights. I am afraid that in the
fray of debate, it is difficult to decide whether one has been fair or not, but I shall
do my best.
Part of the problem of replying is that I have changed my views over the
last four decades, which is, I hope, a fact that will be looked upon with favor. It
would, after all, be a bit disheartening if! simply repeated what I had previously
said from year to year for forty years. On the other hand, in considering criticism
E.!. Olsson (ed.), The Episterrwlogy of Keith Lehrer, 309-356.
© 2003 Kluwer Academic Publishers.

of views which I no longer hold, particularly views concerning the details of

analysis, I am not inclined to engage in dispute about whether some criticism of
my now discarded past is well taken. Moreover, in some cases I must admit that
I do not consider myself to be the best judge of what Keith Lehrer meant by what
he wrote some ten years or more ago, at least I consider myself to be no better
judge than some of my interpreters in this volume. So, in some cases, I shall say
nothing in reply, though the criticism is important and may even be decisive,
simply because I no longer hold the view discussed. In other cases, where the
criticism seems based on an interpretation which I had explicitly repudiated, I
shall note the fact without pretending to special authority about what Keith
Lehrer meant but only noting what he said. Sometimes, moreover, I shall say
nothing in reply simply because I have nothing of value to say. My critic should
not consider silence on my part as disregard but rather as indication that I prefer
to say nothing when I have nothing to say however much inclined I am to think
I am right. Silence should not be taken as concession but as a quiet expression
of respect.
I have done my best to articulate a coherence theory of knowledge
elsewhere. As part of replying to my critics, I would like to make another attempt
to make it clearer what I now advocate, though I shall not restate the technical
details I have presented elsewhere. Here are the main features of the theory as
I now see it.


These are the views I hold which constitute my coherence theory of

knowledge. S knows that p if and only if S accepts that p, p, S is personally
justified in accepting that p, and Ss personal justification for accepting that p is
undefeated or irrefutable by errors of S. This account reduces to the theory that
S knows that p if and only if Ss personal justification for accepting that p is
undefeated or irrefutable by errors of S. Personal justification is a relation of
coherence between a target acceptance and an evaluation system of a subject that
enables the subject to meet objections to the target acceptance. An evaluation
system consists of states of acceptance, preferences over acceptances and
reasonings concerning acceptances. Undefeated or irrefutable justification of a
person is a relation of coherence between a target acceptance and an ultrasystem
constituting a system in which the errors of acceptance, preference and reasoning
in the evaluation system of a person are disallowed for the purpose of answering
objections to the target acceptance. The ultrasystem consists of truth states, t-
acceptances, t-preferences and t-reasonings, of the original evaluation system
together with acknowledgment of the other states of the original evaluation
system. The ultrasystem is intended to provide a test for undefeated justification
disallowing the use of errors in the original evaluation system.

The test for undefeated justification proceeds as follows. We allow

conditional reasoning, something analogous to conditional proof, from the
system of acceptances that are acceptances of true propositions. All acceptances
of propositions that are true may be called t-acceptances. Though the acceptance
system of the person remains what it was in the original evaluation system, only
the content of those acceptances that are t-acceptances may be used to meet
objections to what is accepted for the purposes of testing for coherence and, thus,
justification on the basis of the t-acceptance system. This restriction to the use
of the content oft-acceptances for meeting objections has the effect of blocking
the use of the content of acceptances of things that are false in reasoning while
at the same time acknowledging their existence. The t-acceptance system is used
instead of the acceptance system in the evaluation system for the purpose of
testing for coherence and justification without eliminating or replacing the
acceptance system.
We must, of course, introduce a set of conditional preferences and
reasonings to be used to meet objections and test for undefeated justification as
well as introducing the t-acceptances. The set of conditional reasonings or
t-reasonings is restricted to the subset of original reasonings that contain true
premises. The set of conditional preferences or t-preferences is restricted to the
set of preferences for acceptingp to accepting q in the interests of truth in which
it is not the case that p is false and q is true. (This holds for weak t-preference as
well.) The t-reasonings and t-preferences are subsets of the original preferences
and reasonings in the evaluation system. The original preferences and reasonings
are not eliminated or replaced in the evaluation system, but the use of the content
of preferences and reasonings to meet objections and test for coherence and
undefeated justification is restricted to the content of t-preferences and
t-reasonings while acknowledging the existence of all states of the original
evaluation system. The system consisting of t-acceptances, t-reasonings and
t-preferences is the t-system. The system consisting ofthe t-system together with
acknowledgment of the states of the original evaluation system is the
ultrasystem. This is the basis of undefeated justification. Undefeatedjustification
is coherence and justification on the basis of the contents of states of the t-system
while acknowledging the existence of all states ofthe original evaluation system.

That is my current theory articulated in 2000. I have attempted to explain

reasonableness in terms of trustworthiness which leads to a loop in the system
which I have argued is a virtuous loop rather than a vicious circle. My defense
of this as well as the details is contained in the replies below but I should like to
add some remarks on the subject here, even at the cost of repeating myself
below, because the objection is one that often arises.

There are a number of philosophers who have objected to the circularity
in my account of the application of the principle of trustworthiness and other
principles in the explanation of justification. These principles are not part of my
analysis or definition of knowledge and do not render the analysis or definition
circular. I contend that the explanatory loop is a virtue and not a vice. I have
drawn a distinction between explanation and proof. I have conceded that you
cannot prove something to someone who doubts your conclusion by using it as
a premise. That is not legitimate and all the rules of proof disallow it. But
reasoning has different purposes, and only one of those is proof Another is
explanation. If I seek a complete theory of justification for the purpose of
explaining why I am justified in accepting each thing that I am, in fact, justified
in accepting, then the theory, if! am justified in accepting it, must explain why
I am justified in accepting it.
I have no proof that one should seek for such a complete explanation of
justification. I affirm only that it is my goal. Now some, Bender, for example,
seem to think that I should try to prove to myselfthat I am justified, and he notes
that I cannot do that with an argument that is circular, nor can I present any
circular argument to myself that ought to convince me that I am justified I agree
with him. I hope he will not be disappointed to find me so agreeable, but I agree
with him that just as I should not expect to argue another out his doubts by a
circular argument, so I should not expect to alleviate my own doubts by a
circular argument. But suppose, as is the case, that I have no doubts about
whether I am justified in accepting some obvious claims of common sense and
only seek to explain why I am justified. Then I allege that some explanation is
obtained by arguments that are circular and that circularity is necessary in a
complete theory.
The general point might be conceded and the particular forms I have used
repudiated. Some of my critics suggest that rather than circularity of explanation
one might be better advised to affirm that some beliefs or claims or whatnot are
immediately justified. That is the strategy of the foundationalist. I have affirmed
that some beliefs are justified without argument, or if you prefer, they are
non inferentially justified. So where does a coherence theorist disagree with a
foundationalist who claims that some beliefs are immediately justified? A
coherence theorist disagrees about whether immediacy is a satisfactory
explanation of why the belief is justified. Those who claim to be satisfied with
such an explanation often have a reason for thinking special beliefs are
immediately justified. That is a kind of background theory about immediate
justifications. They have background beliefs, exactly what they are may vary,
that explain why some beliefs, the foundations, are justified in themselves. But
once that explanation is made explicit, they will be revealed as closet coherence
theorists. They will assume that such beliefs are ones that are worthy of a

person's trust without argument because they are successfully truth connected or
for some related reason. Once that assumption is made explicit and the role of it
in explanation of why we are justified in accepting those immediately justified
beliefs is acknowledged, they will emerge in their true glory as philosophers who
allege that justification can always be explained in terms of the background
system of the subject, that is, they will emerge as coherence theorists. If they
eschew that noble title, I would not dispute over the word. Call themselves what
they may, their explanations, if traced far enough, will tie their explanation
together in some virtuous loop rather than leaving immediate justification
unexplained. If not, if they choose a theory of justification in which the
immediacy of justification is unexplained, then we differ in our philosophical
objectives. They are satisfied with unexplained first explainers, and I am not. I
do not claim that they should have my goals of explanation, only that ifthey do,
they will be led to the virtuous loop.


3.1 Reply to Olsson

I am greatly indebted to Erik Olsson for his exceptional contribution to the

study of my epistemology in this volume and wish to take this opportunity to
thank him for arranging a conference in Constance on my work and for putting
together this volume. I begin my reply to my discussants with Olsson's questions
and contributions. Olsson raises the question of whether coherence depends on
global features of the background system. It is useful to clarify my view on the
matter. Coherence is a relation between a background system and a target
acceptance. So, the global features of the background system are not part of the
definition of coherence. However, one should not conclude from this definition
of coherence as a relation to a system that no global feature of the system is
relevant to whether a target acceptance coheres with a system. The relation of
coherence of a target acceptance with an evaluation system consists of the system
being able to meet objections to the target acceptance. That definition does not
contain any mention of global features of the system, but it would be hasty to
conclude from this that global features of the system are irrelevant to coherence.
It may be that the capacity of a system to meet objections to a target acceptance
depends on global features, and, indeed, that seems to me to be the case. The
system cannot be so contradictory that it is useless for meeting objections to an
acceptance. It would be useless, for example, if it were a deductively closed
inconsistent system over the contents of acceptances. It would then contain the

acceptance of the negation of every proposition whose acceptance was contained

in the system.
Nevertheless, an inconsistent system over contents of acceptances which
is not deductively closed over contents may enable us to meet objections. This
is so even though the deductive closure over contents of such as system would
be useless. In a natural environment of acceptance, where even the most brilliant
logicians are prone to having acceptance systems that are inconsistent over
contents, it is essential not to impose deductive closure over contents. For such
systems, but not their deductive closures, remain useful for providing coherence,
justification and knowledge of the truth of some target acceptances.
Another interesting question concerns whether my acceptance that p and
my acceptance that I accept that p in a trustworthy way are functionally
equivalent. I do not think that they are functionally equivalent. The point is that
if a person accepts that he is trustworthy in what he accepts, then he will be ready
to infer from his acceptance that p that he is trustworthy in his acceptance that
p, but that is due to his acceptance of the proposition that he is trustworthy in
what he accepts in addition to his acceptance of p. Suppose a person accepts that
he is not trustworthy. That is something that he accepts. But having accepted
that, he should not infer that he is trustworthy in accepting it. Would this really
be acceptance, however? I think it might be. If I conclude that I am not