Mind b Language ISSN 0268-1064

Vol. 7 No. 4 Winter 1992 @ Basil Blackwell Ltd.

Article
Visualizing as a Means of Geometrical
Discovery

MARCUS GIAQUINTO

1. Zntmduction

Visualizing seems to play a large part in the actual forming of mathematical

beliefs. Evidence for this comes from one's own case, from mathematicians'
reports of their own routes to discovery, and from work in the psychology
of mathematics education. However, the nature and functions of visualiz-
ing are far from well understood. This paper is concerned with a question
about the epistemic functions of visualizing in mathematics. Must the role
of visualizing be merely suggestive, like analogical thinking in physics, or
could it sometimes be a means of discovery? In order to make some
headway with this difficult topic, this paper will focus on one simple
geometrical example. Even with this restricted focus, the account suggested
here is a mere outline. But some superficially attractive views about the
epistemological role of visualizing in geometry are exposed as implausible;
a novel alternative is put forward; and some reason is given for taking
seriously the idea that visual imagination can be a means of geometrical
discovery.

2. Epistemological Premisses

One discovers a truth by coming to believe it independently in an epistem-

ically acceptable way. You may discover a truth that someone already

I would like to thank Ned Block, W.D. Hart, Paul Hoxwich and especially Kris Kirby for
extensive and helpful written comments on earlier versions of this paper,
Address for correspondence: Department of Philosophy, University College London,
Gower Street, London WClE 6BT.
 Visualizing as a Means of Geometrical Discovery 383
knows: priority is not entailed by discovery, as that term is being used
here. But you must come to believe it by 'your own lights': you cannot
discover something by being told. That is the force of the independence
requirement. What is the requirement of epistemic acceptability? This has
two components. A way of coming to believe a truth is epistemically
acceptable only if, first, that way of acquiring a belief is reliable. Accounts
of the relevant kind of reliability are given by McGinn, Dretske, Goldman
and others.' Secondly, a way of coming to believe a truth is epistemically
acceptable only if it is not undermined or defeated by other beliefs of the
person concerned. (For discussion see Goldman, 1986, Chapter 5.) This
condition is needed to rule out discovery in such cases as the following:
one comes to believe a truth in a way which one wrongly believes to be
unreliable. So the way in which you come to believe a truth is epistemically
acceptable if and only if it is reliable and undefeated by your other beliefs.
Here, then, is the sense in which the word 'discover' and its cognates are
used in this paper: one discovers a truth when and only when one comes
to believe it in a way which is reliable, undefeated and independent.
Is someone who discovers a truth thereby justified in believing it? Unfor-
tunately, 'justification' in the context of epistemology is ambiguous in
several dimensions. (See the opening of Goldman (1988)for elaboration.)
For the purposes at hand we should distinguish between the following
two senses. In one sense a person is justified in believing a proposition
only when able to meet a public challenge to justify it; that is, one must
be able to make explicit some thinking which leads to the belief so that it
can be checked by a third party and found to satisfy appropriate standards.
Let us call this demonstrable justification. Suppose one comes to have a
belief in an epistemically acceptable way but is not able to make explicit
any route to the belief which can be checked by the appropriate standards.
In this case the believer does not have demonstrable justification but is
justified in the sense that no violation of epistemic rationality is involved
in the believer's having the belief. Let us say that the believer in this case
has default justification. On this account someone who discovers a truth
thereby has default justification for believing it but may or may not have
demonstrable justification.
In my view one may know a truth without having demonstrable justifi-
cation for believing it. Default justification and reliability may be enough.
If so, in discovering a truth one comes to know it. Still, in any collective
endeavour to augment knowledge demonstrable justification is extremely
important. If we are to rely on some proposition announced by a Galois
without rediscovering it ourselves, it is not enough that he has discovered
it: we must know that it has been discovered, and this almost always

For references and discussion see Goldman (1986),Chapter 3. We need not give a
favoured account of reliability here: our ability to recognise instances is surer than
our grasp of the global considerations needed to construct a precise general account.
384 Mind b Language
requires that someone has seen and checked a proof of it. This is why
proof is so important in mathematics.
The obvious importance of proof in mathematics should not lead us to
overlook the possibility of mathematical discovery without demonstrable
justification. Children and savants seem sometimes to have mathematical
knowledge which goes beyond what they have been told without having
any proof, and mathematicians commonly announce theorems before hav-
ing proved them. Fermat, for example, announced many arithmetical the-
orems for which he never gave a proof. The best mathematicians of the
following century struggled to prove Fermat’s arithmetical claims; all but
two of these turned out to be correct, and one of the exceptions is the
undecided ‘Last Theorem’ for which the indications are all favourable. The
historian of mathematics M o m s Nine (1972, p. 275) writes of Fermat,
‘There is no doubt that he had great intuition, but it is unlikely that he
had proofs for all his affirmations.’ In this case we have some inductive
evidence that Fermat discovered the truths he announced as theorems but
did not have demonstrable justification for believing them. This and other
cases give us some reason to take seriously the possibility of mathematical
discovery without demonstrable justification.
Cases like that of Fermat are doubly intriguing, because they suggest
not only that it is possible to make discoveries without being able to prove
them but also that the thinking involved in making such a discovery would
not constitute a proof; otherwise the mathematician need only make his
thinking explicit in order to prove it. Even the mathematical experience
of a graduate student can point to this possibility: often something new
seems clear, but one struggles to prove it and the proof one finally con-
structs is quite circuitous and is in no way a rendering of the thinking
that led one to believe the proposition in the first place. So it would be
premature to rule out the possibility that one can discover a mathematical
truth by means of thinking which does not constitute an implicit proof.
The upshot of these considerations for this paper is as follows. The
proposition under investigation is that visualizing can be a means of
discovering a geometrical truth. This proposition entails neither that vis-
ualizing can enable one to prove a geometric theorem nor even that it can
furnish an implicit proof. So we have no need to consider the question
whether what one visualizes can constitute or be re-presented as a proof
or part of a proof. The central question is one of reliability. In the next
section an example is presented which will serve as the focus of the
paper. After presenting and assessing some views about visualizing in this
example, we turn to the question of reliability in the penultimate section
of the paper. In the final section some conclusions are drawn.
 Visualizing as a Means of Geometrical Discovery 385
3. The Example2

Imagine a square. Each of its four sides has a midpoint. Now visualize
the square whose comer-points coincide with these four midpoints. If you
visualize the original square with a horizontal base, the new square should
seem to be tilted, standing on one of its comers, ‘like a diamond’ some
people say. Clearly, the original square is bigger than the tilted square
contained within it. How much bigger? By means of visual imagination
plus some simple reasoning one can find the answer very quickly.
By visualizing this figure, it should be clear that the original square is
composed precisely of the tilted square plus four comer triangles, each
side of the tilted square being the base of a comer triangle. One can now
visualize the comer triangles folding over, with creases along the sides of
the tilted square.
Many people conclude that the comer triangles can be arranged to cover
the tilted inner square exactly, without any gap or overlap. If you are in
doubt, imagine the original square with lines running between midpoints
of opposite sides, dividing the square into square quarters, its quadrants.
The sides of the tilted inner square should seem to be diagonals of the
quadrants. Now imagine the comer triangles folding onto the tilted inner
square, with creases along these diagonals.
Assuming that this leads you to the belief that the comer triangles can
be arranged to cover the inner square exactly, you will infer that the area
of the original square is twice the size of the tilted inner square. (You will
reason, let us assume, somewhat as follows. The comer triangles can be
arranged to cover the inner square exactly; so the total area of the comer
triangles equals the area of the inner square; the area of the original square
equals that of the inner square plus the total area of the comer triangles;
so the area of the original square equals twice the area of the inner square.)
Here is a way of coming to believe a true general proposition about
squares in uncurved space: the proposition that any square x is twice as
big as the square whose comer-points are midpoints of x’s sides. You may
have known this already; you may have acquired this belief by having
followed a proof of it from certain other beliefs, or by being told, or in
some other way. But this should not prevent you from seeing that a person
could have acquired this belief by visualizing in the way suggested. The
route to belief described above is mixed: part was valid verbal reasoning;
part was the act of visualizing which led to one of the premisses of the
verbal reasoning, namely, the true belief that if y (’the inner square’) is a
square whose vertices are midpoints of the sides of a square x (‘the original
square’), then the parts of x beyond y (‘the comer triangles’) can be

I came across this example in Kosslyn (1983),who got it from Jill Larkin. It is not a
million miles away from the example used by Plato in The Meno.
No such squares are physically instantiated, because physical space is curved. The
squares we have in mind are geometrical forms that would have been instantiated had
physical space been uncurved, unperforated, etc. I omit the qualification henceforth.
386 Mind 6 Language
arranged to fit exactly into y, without overlap or gap, and without any
changes of size or shape.4 Let this proposition be called 'B. The rest of
this paper will focus on the way of arriving at belief B suggested above.

4. An Inference Fmm Sense Experience?

Suppose that someone not already having belief B acquired it by visualiz-

ing in the manner suggested above. Could this have been a genuine
discovery? To answer this, we must get a clearer view of what might be
happening when belief B is acquired this way. An obvious hypothesis is
that this way of acquiring belief B is a case of inferring from sense
experience. In this section I will argue that it does not have to be so, and
that when it is so, it is not a case of discovery.
Visualizing the comer triangles of the original square folding over onto
the tilted inner square does not seem to involve any inference from sense-
experience, still less to constitute an inference. But this could be illusory.
Sure, the experience of visualizing is quite distinct from the experience of
sentential reasoning; but this leaves open the possibility that in some uses
of visual imagination we are drawing upon past experiences of seeing as
evidence. Suppose you are in a second-hand furniture store looking for a
desk to go in the room where you study; you see an attractive desk,
somewhat larger than you had been looking for. Would it fit? At this point
you visualize the desk in various positions in the room to discover the
possibilities. In visualizing thus you surely would be drawing on your
experience of the room and its current furniture as evidence for the judge-
ments you make about where the desk wiII fit, even though sentential
reasoning is no part of the process.
Is it not the same when through visualizing one concludes that the four
comer triangles of a square would fit exactly onto the tilted inner square
without overlap? Having seen a few Christmases come and go, your
attempts to wrap gifts and fold paper provide you with experiences which
might be drawn upon as evidence for the belief B. When you visualize
the comer triangles folding over onto the tilted inner square, are you not
generalising from past experiences of this kind which have been fused in
the memory? If so, you would be using these past experiences as inductive
evidence, and the visualizing would constitute an inductive inference in
function if not in form.
Several considerations indicate that it does not have to be so. Here
is one. Beliefs acquired and sustained by inductive inference alone are
accompanied by the feeling that there might turn out to be a counter-
example. If you believe that all snow is naturally white, your evidence for
that belief, if it was reached solely by induction from your experience of

Z.e. there is a rigid transformation, etc.

 Visualizing as a Means of Geometrical Discovery 387
snow, is compatible with there being decisive evidence against the belief
in the future. You express an awareness of this when you say of the belief
that it might turn out to be false, that there might tum out to be a
counterexample to it. However, it is typical that when proposition B is
arrived at by visualizing in the way suggested, the believer does not feel
that there might tum out to be a counterexample; on the contrary, typically
one feels that a counterexample is an epistemic imp~ssibility.~ The pres-
ence of this feeling (that a counterexample is an epistemic impossibility)
would be a further indication that the belief was not reached by inferring
from sense experience.
A second consideration is this. If the act of visualizing were a way of
drawing on sense experience as evidence, that experience would have to
include fairly extensive confirmation of proposition B, otherwise it would
not produce any conviction in B, let alone strong conviction. But when
you consider the relevant experience, does it not seem a little meagre?
How many times, very roughly, have you managed to fold the triangular
comers of a piece of paper which seems perfectly square, so that they
appear to fit flush over the tilted inner square? A completely successful
performance is probably very rare. Perhaps, though, the evidence is more
indirect. Maybe it is relevant that you have managed to fold two adjacent
comers of an apparently rectangular sheet so that the comers appear to fit
flush. But how many times have you done this? The point is that one’s
successes, those paper-folding experiences which one might have taken as
confirming cases, may not be very numerous and may even be outweighed
by the failures, and yet by visualizing one may acquire a firm belief in
proposition B. This makes it quite implausible that the visualizing must
be a way of inferring proposition B from one’s paper-folding experiences.
A third consideration against the idea that arriving at belief B by vis-
ualizing (in the manner suggested) must be a case of inferring from one’s
paper-folding experiences is this. Sometimes, when we fail to get the
folded comers to fit exactly, there are visible circumstances which account
for the failure: the line of one of the folds is not at the correct angle with
respect to the sides or it was not at the right perpendicular distance from
the comer point or the sheet was not rectangular to begin with o r . . . .On
other occasions when we fail, there are no visible circumstances to account
for the failure. Typically, if not always, we do not find such would-be
counterexamples a threat to our geometrical beliefs, belief B in particular.
Instead we suppose that visible failure is the effect of one or more invisibly
small imperfections of the kind already mentioned. But what makes us so
sure? Is there some bias in operation, preventing us from seeing failures
as evidence against our geometrical beliefs? No: we exhibit a parallel
attitude to our paper-folding successes. When our physical experiments

This is a feeling about one’s evidential situation with respect to B, not about the
metaphysical status of B.
388 Mind b Language
appear to provide confirming instances of our geometrical beliefs, we
suppose that in fact they do not; we suppose that our successful paper-
folding attempts merely approximate the geometrical possibilities. For
example, we take it that the edges of a sheet of paper are rough, that
paper-folds do not crease along perfectly straight lines, that the surface
area of the paper is not preserved with the folding, and so on. This we
believe holds generally: having fuzzy and inconstant surfaces, physical
objects cannot instantiate a theory of perfectly smooth shapes and perfectly
rigid transformations.
However, this attitude towards our paper-folding experiences (viz. that
they could not provide evidence to warrant belief in a geometrical prop-
osition like B) need not stop a person from coming to believe proposition
B by visualizing in the way suggested; nor, if this attitude to sense
experience were amved at on reflection after belief B was amved at (in
the manner suggested), need the attitude undermine the belief. But if in
arriving at B by visualizing one had to be drawing on past experience as
evidence (like the person trying to imagine how the desk in the store
would fit in the study), the thought that one's experience does not warrant
the conclusion very probably would obstruct the process of amving at the
belief, or undermine the belief (if the potentially damaging thought comes
after)."
To summarise: one may amve at belief B by visualizing in the suggested
manner, when (a) one feels that a future counterexample is not even an
epistemic possibility, (b) the putative evidence of sense experience is
meagre at best and (c) one believes that the putative evidence is of a kind
which could not warrant belief in proposition B; but if one arrives at belief
B by visualizing in the manner suggested under circumstances (a), (b) and
(c), it is very unlikely that the process constitutes an inductive (or
abductive) inference; so it is at least very likely that arriving at B by
visualizing in the manner suggested need not constitute an inference from
sense experience.
Let me make two disclaimers. First, I am not saying that one could
acquire belief B by visualizing without having had some sense experience
of physical objects. On the contrary, I think it very likely that such experi-
ence has at least two roles in generating a geometrical belief by visualiz-
ation, both quite distinct from providing evidence for the belief. First,
the mind may draw upon sense experience in forming basic geometrical
concepts. This is not to deny that there are innate capacities for discriminat-
ing shapes. Concepts are not (or not merely) capacities for distinguishing
instances from non-instances. They are constituents of possible thoughts

One might respond that in arriving at B by visualizing one has to be unconsciously

drawing on past experience as evidence for B, so that the process and the resulting
belief would be protected from a conscious thought that the experience would be
evidentially insufficient.Maybe; but I feel that this move is too desperate to be worth
pursuing.
 Visualizing as a Means of Geometrical Discovery 389
which enable us to reason about their instances. Moreover, basic geometri-
cal concepts seem to be idealisations of experience. You probably got your
concept of a geometrical, i.e. perfect, sphere from your experience of objects
which were more or less spherical. So it i s - o r might be-with concepts
of other shapes. Here, then, is one possible role for sense experience:
together with certain innate mental propensities it results in our forming
the geometrical concepts involved in the belief. A second role is this:
memories of visual experiences provide the components on which the
mind operates in producing a visualizing experience. On this account,
sense experience does enter into the causal pre-history of the belief, but
not as evidence; rather it is the raw material from which the mind forms
our geometrical concepts and our visualizing capacities.
The second disclaimer: I am not saying that it is never the case that in
acquiring a geometrical belief by visualizing one draws on sense experi-
ence as evidence. Perhaps a child would draw on visual memories of
drawings of triangles to answer the question whether there is a triangle
with an internal angle greater than a right angle. I am not even saying
that in acquiring belief B by visualizing in the manner suggested, one
could not be drawing on sense experience as evidence (although I think
that it is typically not so).
However, if one were drawing on sense experience as evidence in
acquiring belief B by visualizing, the result would not be a discovery. This
is because B is the proposition that the comer triangles of a square can be
arranged to fit exactly into its tilted inner square, whereas we have good
reason to believe that there could be apparently triangular physical pieces
which fit inexactly but well enough to seem exact to the senses. Thus even
if there were genuine physical instantiations of geometrical theorems such
as 8, and even if we find what seems to be an instantiation of B, we could
not reliably infer that it is an instantiation of B. Hence the reliability
requirement for discovery would not be met.
For this reason, if amving at B by visualizing in the manner suggested
can be a way of discovering a truth, it must be possible for the visualizing
to serve in some way other than as an inference from sense experience.

5. An Inner Experiment?

On an alternative account, the process does involve drawing on experience

as evidence, but this is the visualizing experience itself rather than past
sense experiences of paper folding, etc. The idea we want to consider is
that the experience of visualizing serves as direct and immediate evidence
for a judgement about what is visualized, just as the experience of seeing
can be immediate evidence for a judgement about what is seen. But
visualizing is not simply seeing what is internal to the mind, since what
is represented in the experience of visualizing is produced by an intention
to visualize such-and-such, whereas what is represented in an experience
390 Mind 6 Language
of seeing is not dependent in this way on the see-er‘s intentions. Accord-
ingly, visualizing is thought of as an internal experiment, i.e. a process that
incorporates both performing the experiment and observing its outcome.
This idea entails that in visualizing the comer triangles of the original
square folding over onto the inner square, one is observing (some feature
of) that very experience of visualizing. This might seem quite puzzling.
How can the experience that results from observing the visualizing experi-
ence differ from merely having the visualizing experience? If it does not
differ, what does observing the experience consist in, over and above
merely having it? Yet people clearly do sometimes observe features of their
own visual experience. Suppose, for example, that you are having an eye
test. The optometrist asks you ‘Which seems clearer, the cross on the green
background or the cross on the red background? or ‘Do you see two
horizontal lines or one? It is understood that you are not being asked to
make judgements about nearby objects. You are being asked to observe
and report on features of your visual experience-something you can
probably do without too much trouble. In other circumstances and in other
sense modalities we sometimes observe features of our own experience:
we can observe the shade of a visual after-image, the timbre of a sound
from an unknown source, the pattern of variation in the intensity of a
pain and so on. In all these cases of inner observation there is something
over and above merely having the experience: there is the sense of
directing one’s attention and noticing something as a result, both of which
add something to the feel of the experience.
Now what about visualizing in the way suggested earlier? If that is not
a way of inferring proposition B from past sense experience, must it be a
case of observing something in the experience of visualizing, thence infer-
ring the geometric belief? Must it be an inner experiment? The examples
above suggest not. For the difference in feel that one gets from concentrat-
ing one‘s attention and noticing something in the experience is sometimes
(perhaps always) missing. When you visualized the corner triangles folding
over onto the inner square, did you scrutinise the end-state of your folding,
for example, and as a result notice that the inner square was covered
exactly? Surely this would be to misdescribe the phenomenology of the
experience.
A second difficulty for the view that the visualizing must be, if not an
inference from sense experience, an ‘inner experiment’, arises from one’s
certainty in the truth of 8: by visualizing in the manner suggested we
may come not merely to believe B but to feel certain of it. Would we feel
this certainty if it depended on observing precisely the fleeting experience
of visual imagining? Of course people do become certain of things on the
flimsiest of grounds; but in these cases they do not maintain their certainty
on reflection, unless their attachment to the belief is enforced by some
psychological disposition having nothing to do with genuine curiosity or
the need for correct information.
We are aware of the fallibility of visual observation of the physical
 Visualizing as a Means of Geometrical Discovery 391
world; optical illusions are common. But there is some temptation to think
that we are infallible observers of our own experience, whether this be the
experience of seeing or visualizing. However, the difficulty of accurately
observing our o w n visual experience, a difficulty known to many who
have undergone extensive eye tests as well as to those learning to draw
or paint what they see as they see it, brings home to us the weakness
of our capacities for internal observation. Added to this is the extreme
elusiveness of experiences of visualizing, in contrast to the relatively clear
and stable character of experiences of seeing. Now if our certainty in the
truth of B survives reflecting on these points without reinforcement (by
e.8. knowledge of a proof), that is some evidence that the belief is not
produced by observing our visualizing experience.
To summarise: one may acquire belief B by visualizing in the manner
suggested and, in addition to circumstances (a), (b) and (c) of the previous
section, it can happen that (d) the phenomenology of looking and noticing
is absent, and (e) one has a feeling of certainty in B which is not under-
mined by recognising the fallibility of inner observation; just as the con-
junction of (a), (b) and (c) makes it extremely unlikely that, in getting
belief B by visualizing in the manner suggested, one is inferring from
sense experience, so the addition of (d) and (e) makes it very unlikely that
one is performing an experiment in one’s visual imagination and inferring
from the visualizing experience; so I think it is possible to acquire belief
B by visualizing in the way suggested without drawing on inner or outer
experience as evidence.
Let me make a disclaimer at this point. I am not suggesting that we
never arrive at geometrical beliefs by experiments of visual imagination.
O n the contrary, I think we sometimes do. For example, one can try to
discover how many edges an octahedron has by imagining a wire model
of an octahedron, the straight parts of wire representing its edges, and
. ~ am I saying that in
’counting the straight parts’ to get an a n ~ w e r Nor
getting belief B by visualizing as suggested, one could not be performing
an inner experiment and accepting the truth of B as a result (although I
think it unlikely).
However, if in getting B by visualizing one were performing an exper-
iment of visual imagination and drawing on the experience as evidence
for B, this could not be a case of genuinely discovering the truth of B. If
we observe the visualizing experience, we would observe, not the perfect
geometrical forms represented by phenomenal features of the visualizing
experience, but the representing features themselves; we have no reason
to believe that such features instantiate perfect geometrical forms, no
reason to think that squares are represented by perfectly square imagess

’ I am grateful to a member of the Cambridge Moral Sciences Club for this example.
Images here are phenomenal items of experience, not structures such as patterns of
filled cells in the visual buffer as in Kosslyn (1980, Chapter 5), which represent images
in a computational theory of imagery.
 Visualizing as a Means of Geometrical Discovery 393
it is a necessary condition of x’s tacitly believing p that x would assent to
p upon entertaining p.’O People have come to believe B by visualizing in
the manner described, with the same phenomenology, who were unable
to say whether B is correct prior to visualizing, even though they under-
stood the question perfectly well. Such people, therefore, did qot tacitly
believe B, and so in their case the role of visualizing could not be to trigger
retrieval. But we want an account that covers this case.
An alternative hypothesis that comes readily to mind is that the visualiz-
ing initiates a process of unconscious sentential reasoning which ends
with a sentence expressing B. Sometimes when we visualize something in
mathematics we get the idea for a proof; then somehow we use this
idea to construct a proof. Could not something like this be happening
unconsciously in the case we are considering? If the steps in the reasoning
are all of a kind regarded by the thinker as valid, we have the core of an
explanation of the sense of certainty about B. But what about immediacy
and the sense of clarity? This account does not explain them, but it may
be consistent with them. If there are not too many steps and they are all
of a kind familiar to the thinker, the unconscious reasoning might be
quick; though valid sentential reasoning does not always preserve a sense
of clarity, it need not always destroy it. Perhaps this view (that the
visualizing initiates a chain of unconscious sentential reasoning) can be
elaborated so as to give an explanation of these phenomena.
But this view is not very plausible. If you now try to write out a proof
of B using the idea of the corner triangles folding over, you will not find
it a simple and straightforward translation from the visual to the verbal.
For a start, what is the initial premiss? You may well be able to whip
through the sequence of steps in a proof once you have constructed it, but
the process of constructing it will not be so quick. So, the hypothesis in
question (unconscious sentential reasoning) has the improbable conse-
quence that a mental task which takes a while to perform consciously can
be performed unconsciously in an instant.
There is a second consideration here. Having visualized the comer
triangles folding over onto the inner square so that belief B is produced
or reinforced, you would have unconsciously performed the sentential
reasoning a short while ago. If it now takes you some mental effort and
several minutes to write out the reasoning, the question naturally arises
why it takes time and effort to retrieve and reproduce reasoning that was
performed such a short while ago? Is there some Freudian mechanism of
repression at work? This points to a further puzzle for the view in question:
if the role of visualizing here is to initiate a process of sentential reasoning,
why is the reasoning unconscious? Why not conscious reasoning?
Then there is the feeling that ‘it’s obvious!’, the sense of clarity men-

lo Lycan mentions an objection, raised by Audi, which he states and refutes in footnote
7 of Lycan (1986).
394 Mind 6 Language
tioned above. It is true that valid sentential reasoning does not always
destroy a sense of clarity. But once it is admitted that the sense of clarity
comes with the visualizing and precedes the putative sentential reasoning,
the game is up. For if the truth of B seems clear to the visualizer before
the reasoning occurs, the reasoning is redundant. We want to know how
the visualizing can produce belief 8; sentential reasoning cannot be part
of the answer if the belief is produced prior to the reasoning.
This prompts a modification of the hypothesis. Perhaps the belief is
produced by unconscious sentential reasoning, but this reasoning is not
initiated by the visualizing; rather reasoning and visualizing are contem-
poraneous; the visualizing has no epistemic role here; it is epiphenomenal.
In this case it is the visualizing and not the reasoning that would be
redundant. But this still leaves us with some uncomfortable questions.
Why is the reasoning unconscious? How come it is so fast? How can it
produce the sense of clarity? And now there is a new puzzle: What initiates
the reasoning and what determines its direction? On the unmodified view
the reasoning is initiated by the visualizing and the direction of the
reasoning is guided by some process of transition from the visual to the
verbal. On the modified view it looks as though the reasoning is initiated
by the same thing that initiates the visualizing, namely one's understand-
ing and accepting the instructions to visualize ('Imagine a square. Now
visualize the square whose vertices are midpoints of. . .'). This is implaus-
ible. Do instructions to visualize, once understood and accepted, always
initiate sentential reasoning? If so, why? If not, why so in this case?
Moreover, on the modified view there is no saying what determines the
direction of reasoning; it is a complete mystery."

7. The Non-evidential Role of Visualizing

What alternative hypothesis is open? Recall the problem. We visualize the

comer triangles folding over exactly onto the inner square. It immediately
appears clear and certain that this is a way in which the comer triangles
could be arranged to fit exactly onto the inner square. But why? Why are
we immediately sure that geometric reality is as imagined? Rather than
postulate an additional hidden process to explain the onset of belief, we
could allow that in this case the process of visualizing itself generates
belief and seek the explanation of this fact in the prior cognitive state of
the visualizer.
What features of the prior cognitive state could provide an explanation?
Perhaps we should look to the ways in which the relevant geometrical
entities were thought of prior to the visualizing, the visualizer's prior

In the example (Section 3) there is sentential reasoning from B; these remarks against
an account in terms of unconscious sententid reasoning concern the route to B.
 Visualizing as a Means of Geometrical Discovery 395
concepts of square, triangle, etc. For this route to be at all promising, we
must suppose that the prior ways of thinking of the relevant geometrical
entities include visual ways. This is very much in accord with empirical
research in the psychology of mathematics education. Vinner and Hersh-
kowitz (1983) identify three components in the way we think of entities
of a given category. These are, in their terms, the concept image, the concept
definition, and a group of operations to which they give no title. A person's
concept image of e.g. the rectangle will be a set of 'mental pictures' of
things which the person takes to be rectangles.'? A person's concept
definition is a sentence which the person treats as a definition of the
rectangle. The group of operations includes visualizable transformations
of things represented in the concept image, some logical operations, maybe
some compass-and-ruler constructions and so on. For our purposes it is
useful to focus on a subset of these operations, namely, those visualizable
operations which the person treats as preserving the size and shape of
things in the category concerned. For example, rotations about a point in
the plane might be regarded as preserving the size and shape of rectangles,
whereas expansions and ruptures would not. Let us call this set of oper-
ations the person's concept group of e.g. the rectangle.
Visualization can interact with prior beliefs to produce new beliefs if
the entities and operations visualized are drawn from the concept image
and the concept group of the visualizer. For example, suppose that Meno's
slave boy visualizes a non-square rectangle with a long horizontal base
and shorter vertical sides and that this reproduces an item in his concept
image of the rectangle. Suppose he believes that the area of a horizontal
rectangle is given by the length of its horizontal base multiplied by either
one of its vertical sides; and suppose that this thought is fresh in his
mind. If he now visualizes the rectangle rotate 90" in the plane and if this
rotation is in his concept group for the rectangle, he will end up visualizing
a rectangle which he takes to have the same area as the original but a
shorter horizontal base and longer vertical sides; and this will bring him
to realise that the longer length multiplied by the shorter equals the shorter
multiplied by the longer.
One's concept images tend to grow with mathematical experience and
will sometimes overlap. For example, if you visualize a square with one
of its diagonals, this may reproduce something in your concept image of
a square and your concept image of a pair of congruent triangles. If so,
not only will what you visualize appear to you to be a square composed
of two congruent triangles with one diagonal as a common base; you will
also believe that a square can in fact be thus composed. Beliefs of this sort
are special; they are accompanied by a sense of certainty and a sense of
clarity because they issue from one's concept images of the categories
involved.

l2 This account of concepf image is not explicit in Vinner & Hershkowitz (1983), but
seems to be what the text demands. A broader account is given in Vinner (1983).
396 Mind 6 Language
The suggestion here is that visualizing can take us from beliefs of this
kind to a new belief in a way which preserves the sense of certainty and
clarity, provided that the content of the visualizing stays within the ambit
of the concept images and the concept groups of the relevant categories.
For example, if you visualize one of the triangles composing the square
fold over, i.e. rotate about the diagonal through 180", this makes it seem
clear and certain to you that the folded triangle would come to rest where
it appeared to, viz. exactly coincident with the other triangle, only if a
rotation of that kind belongs to your concept group for the triangle.
So now let us retum to the original question. One visualizes the comer
triangles fold over onto the inner square and immediately comes to believe
that they could fit onto the inner square just as visualized. The question
is: why is the visualizing immediately followed by this result? It was
suggested that the visualizing itself could be the proximate cause of the
believing, and that this could be explained in terms of features of the prior
cognitive state of the believer. The present suggestion is that these features
are (i) beliefs which issue from the concept images of the categories
involved and (ii) those concept images and concept groups supplying
items which figure in the content of the visualizing experience. The details
of the process (the beliefs involved and the way the visualizing interacts
with them to produce a new belief) remain to be worked out-a project
that awaits empirical research of a finer grain.
But the guiding idea is clear: we start off viewing items in certain concept
images as related in a certain way; we visualize a certain transformation of
these items, a transformation in the relevant concept group; as a result we
end up viewing them as related in a new way. If so, the role of the
visualizing would be to take one from conceptual beliefs to a new belief
by means sanctioned by some of the concepts involved. The present
hypothesis is that, in the case under discussion, the visualizing can have
this role.13
The account given here stands in opposition to the accounts of previous
sections, according to which the visualizing must be a way of delivering
inner or outer experience as evidence for B, from which one then infers
B. Those accounts, it was pointed out, have some weaknesses. In particular,
they do not square with the possibility that circumstances (a)-(e) occur
together. Does the account proposed in this section fare any better? Let
us review (a)-(e).

(a) One feels that it is not the case that there might turn out to be a
counterexample, and this feeling is not weakened by recognising
the fallibility of induction.

l3 Talk of 'items in a concept image' can be misleading, as they are not images, i.e. not
phenomenal items of the visualizing experience. It is not part of the proposal that
belief B arises from or is derived from these images.
 Visualizing as a Means of Geometrical Discovery 397
(b) The putative evidence of sense experience is meagre, but conviction
is strong.
(c) The belief in B is not undermind by recognising that the putative
evidence of sense experience is of a kind which could not warrant
that belief.
(d) The phenomenology of scrutinising one’s experience and noticing
some feature of it is absent.
(e) One has a feeling of certainty in B which is not undermined by
recognising the great fallibility of inner observation.
Does the account proposed in this section square with the conjunction
of (aHe)? It is clear that this account is consistent with (d) because on
this account the process does not involve scrutinising one’s inner experi-
ence. Nor on this account does the process involve inference from the
evidence of the senses; hence the account does not entail that a belief
resulting from the process as described is liable to be undermined by
recognising that sensory evidence could not warrant that belief. So the
account is consistent with (c) as well as (d).
What (a), (b) and (e) bring to our attention is the strength of conviction
in B, the feeling of certainty, the feeling that a counterexample is not a
genuine epistemological possibility. So far from being inconsistent with
this sense of certainty, the story proposed above actually paves the way
for an outline explanation of it. Let us say that a person’s belief x issues
from certain concepts if and only if x is produced (in the person) by the
very processes by which the concepts were initially formed or if x is drawn
from beliefs which have already issued from those concepts in ways which
are sanctioned by those concepts. Then we can say: a belief is felt to be
certain whenever it issues from the believer‘s concepts of the kinds of
things the belief is about; and belief B issues from the believer’s concepts
of the kinds of things B is about, when belief B is produced as described
in previous paragraphs. This is very schematic, and we would need a
deeper understanding in order to say why beliefs which issue from con-
cepts are felt to be certain. But the very possibility that this is right as far
as it goes means that the proposed account is consistent with the feeling
of certainty and sibling phenomena.
So the account proposed in this section does square with the conjunction
of (a)-(e). As (a)-(e) do sometimes hold simultaneously when belief B is
acquired by visualizing in the manner suggested, the role of visualizing
here may sometimes be as outlined in this section. At present I can see no
other role compatible with (a)-(e) for the process of visualizing described in
the third section.

8. Discovery?
Not every way of coming to believe a truth is a way of discovering that
truth: one must come to believe it independently and in an epistemically
398 Mind 6 Language
acceptable way, and then one has made a discovery. I take it for granted
that it is possible to arrive at belief B by visualizing in the manner
described in Section 3 without help from elsewhere, so that the indepen-
dence condition is fulfilled. Recall that a way of arriving at a belief is
epistemically acceptable if and only if it is reliable and not undermined
by the belief-state of the believer. Again I take it for granted that one can
arrive at belief B by visualizing as described without this being under-
mined by one's belief-state. So the central question is one of reliability. If
one were to arrive at belief B by visualizing, and if the process of visualiz-
ing had the non-evidential role outlined in the last section, would that be
a reliable way of arriving at B? Well, why not? Visualizing is certainly
fallible, but that hardly discredits this act of visualizing. Seeing and calcu-
lating are also fallible, yet we can make discoveries by those means.
Reliability does not entail infallibility.
To get to grips with the matter, let us look at the ways in which
visualizing can lead to error. First, there are mistakes which arise from a
mismatch between the intention and the content of visualizing. Let me
introduce an example by way of a puzzle. Suppose you were looking at a
cube, not face on, but with one of its comers pointing straight at you.
How many of its comers would be in view? To answer this people naturally
try to visualize a cube as it would appear comer on; but often what is
actually visualized is a three or four sided pyramid from above and so the
answer given is four or five, which is ~ r 0 n g . Iln~this case people simply
fail to visualize what they intend to visualize.
This type of error is important; but it does not destroy the possibility
of discovery by visualizing in the manner described here. One may intend
to visualize the comer triangles of a square folding over onto the tilted
inner square and fail while thinking that one is succeeding; but our
question is whether, if one does succeed, that could be a reliable way of
coming to believe proposition B. This does raise the further question
whether we know that we succeed, and if we do know this how we know,
but this is beyond our present concerns; for it is not a general truth that
to make a discovery one must know that one has done so, nor that one
must know that one has succeeded in performing intended operations.
A second type of error: one succeeds in visualizing as intended, but has
a false belief about what one has visualized. For example, suppose one
tries to visualize a triangle with a horizontal base and a vertical side, and
then a line from its upper vertex to the midpoint of its base. One may
succeed but falsely believe that one is imagining a triangle with the altitude
to its base, simply because one's grasp of the concept of altitude is askew.15
This type of error is not relevant: in the case that we are considering there
is no false belief about what one visualizes.

The answer is seven. The example is known as Hinton's cube. See Hinton (1979).
See the constructions of student B in Hershkowitz (1987).
 Visualizing as a Means of Geometrical Discovery 399
A third type of e m r occurs when visualizing triggers a fallacious dispo-
sition to believe. Such a disposition is not necessarily a disposition to
believe something false. Suppose one visualizes a singIe figure of a certain
sort, and so doing causes one to believe a proposition about all figures of
that sort; an error of the third type occurs just when what is visualized
illustrates the proposition only in virtue of some attribute not shared by
all figures of the relevant sort.
Suppose one visualizes a triangle with a horizontal base and each angle
less than a right angle; then adds the perpendicular line falling from the
upper vertex to the base; then embeds this figure in a rectangle on the
same base with the same height. Now suppose that the visualizer comes
to view the two right angled triangles which compose the original triangle
as exact halves of the two rectangles which compose the encompassing
rectangle. Coming to view the figure this way may cause one to believe
that the area of a triangle, any triangle, equals half its altitude times its
base. That would be the operation of a fallacious disposition: though the
conclusion is correct, this particular way of reaching it depends on a
feature of the visualized triangle not shared by all triangles, namely, that
its base angles are no greater than right angles; this is because a triangle
with a base angle greater than a right angle cannot be visualized as part
of a rectangle having the very same base.16
If you visualize the comer triangles of a square folding over onto its
tilted inner square, and as a direct result come to believe that the corner
triangles of any square can be arranged to fit the tilted inner square exactly,
are you making an error of the third type? No: if the comer triangles of
one square can be arranged to fit, etc., then the comer triangles of any
square can be so arranged, as all squares have the same shape, while any
property independent of shape is irrelevant.
Finally there are errors which are due to vagueness and inconstancy of
the images, something which by hypothesis does not affect the case under
investigation.
Four types of error which beset belief-acquisition by visualizing have
been reviewed; no error of these types is involved when we get belief B
by visualizing in the manner suggested. If the visualizing were a way of
drawing on experience as evidence, then (as pointed out before) certain
errors of inference from experience would be involved. But if the visualiz-
ing has the non-evidential role sketched above, I can at present see no
type of error, other than those just reviewed, which might be involved.
So I am inclined to think that this way of acquiring belief B is reliable,
although I do not claim to have esfablished that it is. Ideally one would
want to be able to establish reliability as one can for certain algorithms

l6 Of course, one can make a special provision for these cases: imagine the triangle
rotate about an inner point until the longest side is the base; then proceed as before.
If this step is suitably incorporated, the third type of error is avoided.
400 Mind 6 Language
for computing a given function or certain chains of sentential reasoning.
But this would require a grasp of the visualizing process in detail, some-
thing which is at present beyond us.

9. Concluding Remarks

It is natural to start off with the suspicion that, in the kind of case discussed
in this paper, geometrical discovery by visual imagining is impossible.
Visualizing is most readily compared with seeing; seeing, in its primary
role of providing observational evidence, cannot deliver this kind of geo-
metrical knowledge; and it is felt that visualizing is no better than seeing.
But if the view outlined in Section 6 is correct, the comparison is mislead-
ing. While the experience of visualizing may be in some respects similar
to the experience of seeing, the role of visualizing, in cases like the one
examined here, can be utterly different from the evidence-providing role
of seeing and closer to the role of deductive sentential reasoning when
one draws on prior conceptual beliefs. In view of its non-evidential role
we might say that the visualizing in this case is an a priori way of acquiring
belief.
However, we should note two respects in which the boot is on the other
foot. First, there is another kind of seeing-seeing a configuration of marks
as e.g. a square17-which I think can play a role very close to that of
visualizing described above. Secondly, valid sentential reasoning can be
regarded as an analytic procedure, one of unpacking what is implicit in
prior beliefs, whereas the visualizing, when it is a way of applying oper-
ations in one's concept group to items in one's concept image, is synthetic,
a putting together of conceptual elements. However, this account (based
on the apparatus adapted from Vinner and Hershkowitz) must be regarded
as tentative, as a possible way of fleshing out a more general hypothesis,
namely, that the role of the visualizing in this case can be to take one from
prior beliefs to a new belief by means sanctioned by some of the concepts
involved, without drawing on experience as evidence.
Let me stress that visualizing can play roles in the acquisition of a
geometrical belief other than the one suggested here. The contentions of
this paper are only that the visualizing can play this role and that when
it does, it is probably a means of discovering the belief arrived at. The
suggestion that my account paves the way for an explanation of the
accompanying sense of immediacy, certainty and lucidity goes beyond
what I would insist upon here. But if this is right, that would be a definite
advantage for my account. The combined feelings of certainty, lucidity and
immediacy are, I think, what prompt mathematicians to talk of 'intuition'.
Even though it is not plausible to suppose that there is a faculty of

l7
This kind of seeing is discussed in Giaquinto (1992).
 Visualizing as a Means of Geometrical Discovery 401
intuition, the psychological phenomena associated with that term must be
accommodated in any adequate account of the nature of mathematical
discovery.

Department of Philosophy
University College London
London W C l E 6BT

References
Giaquinto, M. 1992: Diagrams: Socrates and Meno's Slave. The International
Journal of Philosophical Studies.
Goldman, A.I. 1986: Epistemology and Cognition. Cambridge, MA.: Harvard
University Press.
Goldman, A.I. 1988: Strong and Weak Justification. In J. Tomberlin (ed.),
Philosophical Perspectives Volume 2: Epistemology. Atascadero, CA.: Ridge-
view Publishing Company.
Hershkowitz, R. 1987: The Acquisition of Concepts and Misconceptions in
Basic Geometry. In D. Novak (ed.), Proceedings of The 2nd International
Seminar on Misconceptions and Educational Strategies in Science and Math-
ematics. Ithaca: Cornell University Press.
Hinton G. 1979: Some Demonstrations of the Effects of Structural Descriptions
in Mental Imagery. Cognitive Science, 3, 231-50.
Kline, M. 1972: Mathematical Thought from Ancient to Modern Times. Oxford:
Oxford University Press.
Kosslyn, S. 1980: Image and Mind. Cambridge, MA.: Harvard University Press.
Kosslyn, S. 1983: Ghosts in the Mind's Machine. New York: Norton.
Lycan, W.G. 1986: Tacit Belief. In R. Bogdan (ed.), Belief. Oxford: Oxford
University Press.
Plato: Meno. (Trans. R.W. Sharples). Warminster: Aris & Phillips, 1985.
Vinner, S. 1983:Concept Definition, Concept Image and the Notion of Function.
lnternational Journal of Mathematical Education in Science and Technology,
14, 239-305.
Vinner, S. and Hershkowitz, R. 1983: On Concept Formation in Geometry.
Zentralblatt fur Didaktik der Mathematik, 15, 20-25.