Anda di halaman 1dari 15

NATHAN STEMMER

A SOLUTION

TO THE LOTTERY

PARADOX

1. INTRODUCTION It may happen that the basic terms that are involved in a philosophical problem are ambiguous. If this occurs, then it will often be difficult to find a satisfactory solution to the problem, unless the ambiguity is taken into account. It seems that one of the reasons, perhaps the main reason, why the lottery paradox has shown to be a difficult one is related to this point. In the paradox, the terms 'to believe' and 'to accept' play a central role. Yet, these terms are highly ambiguous both in ordinary language and in our informal scientific discourse. 1 This suggests that in order to find a satisfactory solution to the paradox it will be necessary to first identify those meanings of the terms that are relevant to the paradox. Only after this has been achieved can one try to solve the paradox. This idea underlies the present paper. First, I analyze the assumptions that give origin to the lottery paradox. The analysis makes it possible to characterize two notions of uncertainty that will be called Humean and evidential uncertainty. With the help of these notions, I then distinguish between three meanings of 'to believe' and 'to accept' which are directly relevant to the paradox. Finally, I examine the validity of the so called conjunction principle with respect to these meanings. Our discussion will suggest that whereas with respect to the first two meanings the principle is valid, this is not the case with respect to the third meaning. These results can be considered as a satisfactory solution to the lottery paradox.

2. THE LOTTERY PARADOX A principle that seems intuitive to many persons is the following conjunction principle: (CP) If it is rational to believe p, and if it is also rational to believe q, then it is rational to believe p and q. Synthese 51 (1982) 33%353. 0039-7857/82/0513--0339$01.50 Copyright (~ 1982 by D. Reidel Publishing Co., Dordrecht, Holland, and Boston, U.S.A.

340

NATHAN

STEMMER

Another principle that has been proposed by Kyburg, and which also seems intuitive to various persons is the following Kyburg

principle:
(KP) It is rational to believe a hypothesis that is highly probable relative to what is already believed. 2

But the lottery paradox, formulated by Kyburg in (1961, p. 197), shows that the two principles together may produce contradictory results. Suppose that with respect to a fair lottery with one million tickets we consider the hypothesis: (1) Ticket number 1 will not win.

According to Kyburg, it is rational to believe (1), because it satisfies (KP). Thus, in (1970a, p. 176) he says:
Surely, if a sheer probability is ever sufficient to warrant the acceptance of a hypothesis, this [a probability of 0.999999] is the case,

and in Kyburg (1970b, p. 56) he states:


S i n c e . . . there is only one chance in a million that this hypothesis ['ticket number 7 will not win'] is f a l s e . . , this is reason enough to accept the hypothesis.

Now, if it is rational to believe (1), it is also rational to believe that ticket number 2 will not win, that ticket number 3 will not win, etc. By applying the conjunction principle (CP), we should therefore believe: (2) No ticket will win.

But this contradicts our premise that the lottery is fair, which means that one ticket will win. Kyburg concludes from this paradox that the conjunction principle should be rejected. But since this principle appears to many as highly intuitive, it is worthwhile to examine whether perhaps some of the other presuppositions, including Kyburg's principle (KP), are the causes of the paradox.
3. A N U N W A R R A N T E D ASSUMPTION

It is easy to see that one of Kyburg's assumptions is clearly unjustified. A probability of 0.999999 is not sufficient for accepting a hypothesis. N,o rational person will advise me to accept a bet of 100.000 dollars against one dollar cent that (1) will indeed occur.

ASOLUTION

TO THE LOTTERY

PARADOX

341

Hence, even if the probability that hypothesis h is true is 0.999999, this is not yet sufficient to believe or to accept h. 3 To make this point clearer, suppose we have two urns, urn A and urn B. At tl we put in urn A one million green balls, and in urn B we put 999999 green balls and one blue ball. At t2 the urns are well mixed and we now consider the following hypotheses: (3) (4) The first ball randomly drawn from urn A is green. The first ball randomly drawn from urn B is green.

The acceptance of (3) is normally considered to be a rational one. Since we usually assign to (3) a probability of I, we cannot propose a bet which shows that it should not be accepted. Thus, it is indeed rational to accept a bet of 100.000 or even of a million dollars against one dollar cent that (3) will occur. But (4) is clearly different. Just as in the case of (1), the probability of (4) is normally considered to be only 0.999999. We can therefore propose a bet which shows that it is not rational to accept (4). For example, one should not accept a bet of 100.000 dollars against one dollar cent that (4) will indeed occur. H e n c e , we cannot say that it is rational to accept or to believe (4). We thus see that a probability of 0.999999 is too low for making it rational to believe a hypothesis h. One can propose a bet which indicates that it is not rational to accept h. But then any probability, whose value is a definite rational number less than 1, is too low. One can always propose a bet concerning the truth of h whose acceptance would not be rational. If this analysis is correct, then it sheds doubt on the principle (KP) itself. Perhaps no hypothesis whose probability is less than 1 ought to be believed.

4.

HUMEAN

AND EVIDENTIAL

UNCERTAINTY

What are Kyburg's reasons for proposing the principle (KP)? His discussion in (1961, p. 2f) suggests that it is the fact that most, perhaps even all of our empirical beliefs are affected by H u m e a n uncertainty. L e t me quote:
. . one should take experience as one's guide. But how are we to do this? The fact that something has happened once provides no logical warrant that it will happen again. If we have observed that A has followed B once, or n times, we are still free to suppose,

342

NATHAN

STEMMER

without contradicting ourselves, that A will not follow B the next time B occurs. This thesis was cogently argued by Hume, and is periodically reasserted in the philosophical periodicals. But while we may readily admit that no amount of evidence will in itself logically entail that a given statement about the future will be true, it is nevertheless easy to say that when we have enough evidence this predictive statement becomes overwhelmingly probable... [Hence] to base one's expectations on past experience is to take probability as one's guide in life.

There is no doubt that Kyburg's conclusion is correct. Many of the beliefs that people consider rational are affected by Humean uncertainty. However, is the uncertainty that occurs in the lottery paradox the Humean uncertainty? A closer look at the paradox shows that this is not the case. Thus, consider again hypothesis (3) concerning the ball of urn A. We have assigned to (3) a probability of 1. But clearly our evidence does not imply that (3) is true, because, as Kyburg says, no amount of evidence will in itself logically entail that a given statement about the future will be true. Perhaps the uniformity of the world is such that objects that look green are actually grue (i.e. green and the time is before t2, or blue and the time is after tz), or perhaps some or all of the balls have become ravens (hence, black), or blood (hence, red). On any of these possibilities, (3) may turn out to be false. 4 (3) is thus affected by Humean uncertainty. Consequently, it is only probably true. It must therefore be assigned a probability that is less than 1. But this implies that the probability we assign to (4) must be different from 0.999999. For besides considering the evidence that one of the balls of urn B is b l u e - m o r e exactly, that at tl one of the balls was b l u e - we also have to consider Humean uncertainty. Perhaps the balls are now black, or red, or colorless. This shows that (4) is affected by two types of uncertainties. One is the evidential uncertainty, i.e. the uncertainty which has its origin in the evidence accordingly we put 999999 green balls in urn B. This evidence stands behind the definite numerical value 0.999999 which one normally assigns to the probability of (4). The other is the Humean uncertainty which modifies this probability. The same holds for hypothesis (1). This hypothesis about the future is affected by two types of uncertainties. One is the evidential uncertainty which gives the probability of 0.999999 to the hypothesis that ticket number 1 will not win. The other is the Humean uncertainty which modifies this probability. But the paradoxical situa-

A SOLUTION

TO THE

LOTTERY

PARADOX

343

tion that arises in connection with hypothesis (1) derives from its evidential rather than from its Humean uncertainty. It is because the evidence tells us that one ticket will w i n - the lottery is f a i r - that we do not believe (2). Perhaps one can define a lottery in such a way that, unlike hypothesis (4) which refers to the balls in box B, hypothesis (1) will not be affected by Humean uncertainty, e.g. if we decide that "miraculous" or "gruelike" changes in the numbers of the roulette wheel (which, say, is the instrument used for determining the winner in the lottery) will not be taken into account. That is, we eliminate by convention those cases where Humean uncertainty shows its effects. But since this move does not block the paradox, it confirms that the paradox derives from the evidential uncertainty that affects (1) and not from its Humean uncertainty. Note also that the possibility of proposing a bet showing the non-rationality of accepting (1) and (4) derives from evidential rather than from Humean uncertainty. For it is evidential uncertainty that enables us to assign to these hypotheses the definite value of 0.999999. Humean uncertainty tells us nothing about this value, just as it does not give us a definite probability value with respect to hypothesis (3).

5. H U M E A N

AND

EVIDENTIAL

PROBABILITY

Let us call the probability that derives from Humean uncertainty Humean probability and the one that derives from evidential uncertainty evidential probability. Then our discussion suggests that perhaps the following modified version of Kyburg's principle avoids the paradoxical results: (KP') It is rational to believe a hypothesis which, relative to what is already believed, has an evidential probability of 1 and a very high degree of Humean probability.

Before examining whether (KP') indeed avoids the paradox, we must clarify somewhat more the concepts that occur in it. The principle presupposes a distinction between Humean and evidential uncertainty. Now, I am not sure whether one can exactly define this

344

NATHAN

STEMMER

distinction with respect to all kinds of hypotheses. However, it is not difficult to find paradigm cases that are sufficiently clear for allowing us to rely on this distinction at least with respect to simple hypotheses, i.e. hypotheses of the form 'All P are Q' and their particular instances such as 'This P is Q'. Thus, consider again the two urns mentioned above. Urn A, we recall, contains one million green balls. Here, the available evidence supports the general hypothesis: (5) All balls of urn A are green,

and there is no evidence available that suggests that (5) is false. Therefore, the general hypothesis and all its particular instances such as hypothesis (3) are affected only by Humean uncertainty. Urn B contains 999999 green balls and one blue ball. Here, too, there is evidence (actually a great deal of evidence) that supports the general hypothesis: (6) All balls of urn B are green.

However, in the present case we also have evidence which falsifies this hypothesis, since our evidence tells us that one of the balls is blue (and there is no evidence that suggests that this is not true). Therefore, the general hypothesis (6) and its instances are affected by both Humean and evidential uncertainty. This suggests that with respect to simple hypotheses we can use the following criterion for distinguishing between Humean and evidential uncertainty. If our evidence supports the general hypothesis: (7) All P's are Q,

and there is no evidence available that suggests that there are some P's that are not Q, then the hypothesis and its instances are affected only by Humean uncertainty. But if there is not only supporting but also falsifying evidence, then the hypothesis and its instances are affected by both Humean and evidential uncertainty. Notice that our characterization of Humean uncertainty covers not only the uncertainty that derives from our ignorance of whether the future will be similar to the past. It also covers the uncertainty that affects a hypothesis 'All P ' s are Q' which is supported by positive instances that do not exhaust the generalization. (Cf. Hume (1748, Section IV, part II): "These two propositions are far from being the

A SOLUTION

TO THE

LOTTERY

PARADOX

345

same: 'I have found that such an object has always been attended with such an effect' and 'I foresee that other objects which are in appearance similar will be attended with similar effects'.") The second concept that has to be clarified is that of a hypothesis having a very high (degree of) Humean probability. 5 Now, it is very unlikely that one can precisely state the conditions that have to be satisfied by a hypothesis in order to assign to it a very high Humean probability. Moreover, there is little agreement among different scholars about this, even with respect to simple hypotheses. I shall therefore make only a few, very general, statements on this topic. 6 First a distinction has to be made between the probabilities that reflect the two kinds of Humean uncertainties that were just mentioned. With respect to the one that reflects our uncertainty whether the future will be similar to the past, it seems that the degree of probability of a hypothesis concerning the future is mainly a function of the degree of psychological intuitiveness of the similarity that is being considered. (Cf. Stemmer, 1978, 1979.) For example, the similarity between 'All emeralds were green' and 'All emeralds will be green' is an intuitive similarity (while that between 'All emeralds were grue' and 'All emeralds will be grue' is not intuitive). Therefore, if we attribute a high probability to 'All emeralds were green' we will also attribute a high probability to 'All emeralds will be green'. With respect to the probability that reflects the inductive leap from observed instances to a general hypothesis, we must distinguish between the case where the hypothesis is supported only by the observations of positive instances, and the case where it is a part of, or derives from, a theoretical framework. In the former case, the degree of Humean probability of the general hypothesis seems to depend basically on three factors: the existence of positive instances (not necessarily many), the absence of evidence that suggests that there exist negative instances, and the degree of projectibility of the predicates that occur in the hypothesis.7 In the latter case, it seems to depend mainly on the degree of "goodness" of the theory, 8 on the absence of evidence that suggests that there exist negative instances and, to a certain extent, on the existence of positive instances and the degree of projectibility of the predicates that occur in the hypothesis. If a general hypothesis has a very high degree of Humean probability, then its instances also have a very high degree of this probability. 9

346
6.
THE

NATHAN MODIFIED

STEMMER PRINCIPLE (KP')

Let us now consider whether the principle (KP') avoids the paradox. First, we note that if a hypothesis satisfies its conditions, then it has a very high degree of Humean probability. From our clarification of the latter notion it follows that in this case there is no evidence available suggesting the existence of instances that falsify the hypothesis. Hence, the evidence does not give us data that enable us to determine a definite rational number which expresses its degree of Humean probability. Consequently, we are unable to propose a concrete bet that would show that it is not rational to believe the hypothesis. Moreover, the principle appears to be compatible with the conjunction principle. If we have reasons to attribute to a number of hypotheses a very high degree of Humean probability-e.g., if they satisfy to a high degree the conditions mentioned in the previous sections- then the degree of Humean probability of their conjunction seems to remain sufficiently high to make it rational to believe the conjunction. To illustrate, let me use an example that is given by Peirce in order to make a related point (1934, p. 64):
Here is a stone. N o w I place that stone where there will be no obstacle b e t w e e n it and the floor, and I will predict with confidence that as soon as I let go m y hold u p o n the stone it will fall to the f l o o r . . , and if a n y o n e p r e s e n t has a n y doubt on the subject, I should be h a p p y to try the experiment, and I will bet him a h u n d r e d to one on the result.

Now, it is not only rational to bet a hundred to one but even a million to one that the stone will fall. 1 Moreover, it would be rational to propose such a bet not only with respect to one stone but also with respect to thousands of (well examined, heavy) stones with respect to thousands of iron balls, of gold bars, etc.; hence, with respect to the conjunction of a very large number of hypotheses. Now, hypotheses like 'this stone will fall when dropped', 'that iron ball will fall when dropped' have a very high Humean probability. They satisfy to a high degree the conditions that were discussed in the previous section. This therefore supports the conclusion that if the Humean probability of a hypothesis is very high, then we believe not only the hypothesis itself, but also large conjunctions of similar hypotheses. Nevertheless, as long as we do not attribute to the hypothesis a Humean probability of 1, the hypothesis is still affected by Humean uncertainty. The question now arises: What is the cumulative effect

A SOLUTION

TO THE

LOTTERY

PARADOX

347

of these uncertainties on the conjunction of all the hypotheses to which we assign a very high Humean probability? Maybe the Humean uncertainty of the conjunction is too great to make it rational to believe the conjunction? This is probably the rationale behind the following argument of Kyburg for rejecting the conjunction principle (1970b, p. 77):
Although I claim to have good reasons for believing every statement I believe, I claim also to have good reasons for believing that some of these statements are false. ~

Now, I am not sure whether we have good reasons for believing that some of the hypotheses that have a very high degree of Humean probability are false, such as the hypotheses 'This heavy stone will fall when dropped', 'That chicken will die when its heart is removed', 'The Eiffel Tower will not become a raven in 1980', etc. Still, it seems reasonable to admit that theoretical reasons suggest (but do not prove) the conclusion that if we form the conjunction of all these hypotheses, the cumulative effect of the Humean uncertainties is indeed too high to make it rational to believe the conjunction. But this theoretical conclusion has no practical effects whatsoever since, when making our decisions, we never have to take into account the whole conjunction. We often do consider very large conjunctions, but here we act as if they have a Humean probability of 1. Consider, for example, a person who offers price X for a house. When determining the price he takes into account the evidential probability of several hypotheses. 12 But he does not consider the possibility that the conjunction of the relevant hypotheses that have a very high degree of Humean probability (and are not affected by evidential uncertainty) may be false, such as of the hypotheses 'The first brick will not become a raven', 'The second brick will not become a raven', 'The third brick will not become a raven', etc., The Humean uncertainty that affects the conjunction of these hypotheses does not influence the person's decision to pay price X rather than some other price.

7.

THREE

MEANINGS

OF 'TO BELIEVE'

This suggests that one must distinguish between a theoretical and a practical notion of belief. The theoretical notion is mostly used in philosophical discourse, while the practical notion is related to con-

348

NATHAN

STEMMER

crete human behavior, such as betting behavior. From now on, I shall use the term 'to believe' or 'to believe to be true' for the theoretical notion, and the term 'to practically accept' for the practical notion. Our discussion now suggests the following conclusions. The revised version of Kyburg's principle (KP') holds only for practical acceptance. We practically accept the hypotheses that have a very high Humean probability and an evidential probability of 1, and we also practically accept their conjunction. But (KP') does not hold for (theoretical) belief. Not only do we not believe that the conjunction of these hypotheses is true, we do not even believe that any of these hypotheses is true. For if we use such a philosophical notion then, of course, we must take into account Humean uncertainty not only with respect to the conjunction of the hypotheses, but also with respect to each of them. This is precisely Hume's point. Whatever our evidence, it does not imply 'This stone will fall when dropped'. Expressing our conclusions more formally, we obtain the following principles for practical acceptance and theoretical belief. (PA) It is rational to practically accept a hypothesis which, relative to what is already practically accepted, has an evidential probability of 1 and a very high Humean probability. It is rational to believe a hypothesis which, relative to what is already believed, has a Humean (and evidential) probability of 1. j3

(PB)

Since the conjunction principle (CP) deals only with rational belief, we now replace it by the following principle: (CP') If it is rational to believe (to practically accept) p and also q, then it is rational to believe (to practically accept) p and q.

It now follows from our discussion that (PA), (PB), and (CP') are compatible. (PB) is a trivial principle But this is unavoidable since, by definition, the philosophical notion of rational belief is sensitive to Humean uncertainty. Nevertheless, there remains a problem. There are cases where the evidential probability of a hypothesis is less than 1 and still, relative to what can happen to a real person, it seems rational to practically

A SOLUTION

TO THE

LOTTERY

PARADOX

349

accept the hypothesis. Thus, let the term 'stonrav' denote the class which contains all the heavy stones on the earth and one living raven. Then, the general hypothesis: (8) All stonravs fall when being dropped,

is false. But the evidential probability of the particular hypothesis: (9) This stonrav will fall when dropped,

is so high that, for all practical matters, we can accept it as true. Notice, moreover, that although (9) is affected by an evidential u n c e r t a i n t y - o u r evidence tells us that there exists at least one negative instance of (8)- we are not able to specify a definite rational number which expresses the evidential probability of (9), since we do not know how many heavy stones there are on the earth. Hence, we cannot propose a concrete bet that would show that it is not rational to practically accept (9). There is therefore almost no doubt that it is rational to practically accept 'This second stonrav will fall when dropped', 'This third stonrav will fall when dropped', etc. Ultimately, the conjunction principle will bring us to contradictory results. Now, there is a clear difference between: (10) This heavy stone will fall when dropped,

and (9). Our evidence tells us that the general hypothesis that stands behind (10), namely: (11) All heavy stones fall when dropped,

if affected only by Humean uncertainty. 14 On the other hand, this evidence tells us that the general hypothesis (8) which stands behind (9) is actually false. This shows that a further distinction has to be made. One must distinguish between the practical acceptance of a hypothesis that is affected ofily by Humean uncertainty and the practical acceptance of a hypothesis that is (also) affected by an evidential uncertainty which is too small to influence the decisions of a (real) person. With respect to the former, the conjunction principle (CP') is valid. With respect to the latter, however, the principle does not hold. And I believe that this agrees with our intuitions. The cumulative effect of the evidential uncertainties that affect a number of hypotheses, even if they are

350

NATHAN

STEMMER

minimal, may be too great to make it rational to ignore the effect in the case of a conjunction of these hypotheses. Since such conjunctions may be relevant to the practical decisions of a person, it would no longer be rational to practically accept them. In a recent paper on the lottery paradox, Derksen states the following conclusion (1978, p. 72):
There is thus a probability gap, which may, and often does, increase when the conjunction of more beliefs get involved. In fact, the gap may, and often does, increase to such an extent that the (not specified, and not precisely specifiable) justification-limit (of the probability that p) for the belief that p, will be transgressed.

Our discussion suggests that this conclusion holds for the notion of b e l i e f - i n our terminology, for the notion of practical acceptancewhich is applied to hypotheses that are affected by a very small evidential uncertainty. With respect to the decisions in which only one of these hypotheses has to be considered, the uncertainty is too small to be taken into account. However, when a conjunction of such hypotheses becomes relevant to a decision, then it may no longer be rational to ignore the cumulative effect of the evidential uncertainties.

8. C O N C L U S I O N S

The lottery paradox calls our attention to the need of making a distinction between three different concepts that are expressed in ordinary language by the terms 'to believe' or 'to accept'. One is the philosophical concept which takes into account the Humean uncertainty of a hypothesis even if it is very small. With respect to this concept, the conjunction principle can be accepted without any qualm, since here we deny the rationality of believing (true) even a single hypothesis (if it is affected by the uncertainty). According to the second concept, it is rational to practically accept a hypothesis provided it has a very high Humean probability and is not affected by evidential uncertainty. When using this concept, the conjunction principle can also be accepted. For, with respect to the conjunction of hypotheses that practically influence the decisions of a person, the cumulative effect of the Humean uncertainties is too small to make it irrational to practically accept these conjunctions.

A SOLUTION

TO THE

LOTTERY

PARADOX

351

Finally, the third concept corresponds to the practical acceptance of hypotheses which are affected by evidential uncertainties that are too small to be taken into account in the decisions of a person. When using this concept, however, the conjunction principle cannot be accepted, since the cumulative effect of the evidential uncertainties may be too great for making it rational to ignore this effect in the case of certain conjunctions of these hypotheses. These conclusions can be considered as a satisfactory solution to the lottery paradox. They enable us to distinguish between three meanings of the terms 'to believe' or 'to accept', and they state the validity of the conjunction principle with respect to these meanings. To be sure, the conclusions have been formulated only with respect to simple hypotheses, i.e. hypotheses of the form 'All P are Q' and their particular instances. However, it is likely that the conclusions can be naturally extended to also cover more complex hypotheses. But even if the correct account of the belief or acceptance of more complex hypotheses requires a different treatment, there is no doubt that by solving the lottery paradox for simple hypotheses, an important step has been made towards the formulation of a general theory of rational belief or rational acceptance. In our solution, a fundamental role is played by the distinction between Humean and evidential uncertainty. It is with the help of this distinction that the three meanings of 'to believe' and 'to accept' have been characterized. Now, most people who have discussed the lottery paradox have not considered the distinction between the two uncertainties. The reason is probably that their examples concerned entities denoted by projectible predicates, such as coins, balls, dice, American males, etc. Since such predicates usually conferred a very high degree of Humean probability on the corresponding hypotheses, the need for considering Humean uncertainty did seldom arise. But had they also discussed hypotheses such as 'The next randomly drawn non-raven is grue', their treatment might have been different. I believe that this inattention to the difference between Humean and evidential uncertainty is at least partially the reason why so far no satisfactory solution to the lottery paradox has been given.

Bar-Ilan University

352

NATHAN

STEMMER

NOTES See, e.g. Bar-Hillel (1968). 2 In his (1961), Kyburg states the following condition that has to be satisfied by a logic of rational belief: " . . . we want to be able to show that any statement which is highly probable relative to what is already believed ought also to be believed" (p. 83). 3 In (1961), Kyburg speaks mainly of believing a hypothesis, while in his (1970a) and (1970b) he usually speaks of accepting a hypothesis. But since in the latter publications he also speaks of believing hypotheses, I shall use until Sec. 7 both notions indistinctly. As far as I can discern, Kyburg does not distinguish betwe.en these notions. 4 Notice that the several possibilities that I am considering here do not constitute specific background information that is relevant only to the probability of (3). For these possibilities derive from Hume's conclusion that we do not know whether the future will be similar to the past. This conclusion is universal, since it bears upon any empirical hypothesis about the future. In other words, in every case in which one normally assigns the probability of 1 to a hypothesis concerning the future, the assignment assumes the background "information" that the future will be similar to the past, which implies that situations like those of balls becoming ravens will not occur. (The Goodman paradox shows that the similarity must be qualified. This is discussed in Stemmer, 1978, 1979.) 5 I shall frequently omit the clause ~degree of'. 6 I have studied this problem more closely in Stemmer (1971), (1975), (1978), and (1979). 7 Cf. Goodman (1955), Quine (1969), Quine and Ullian (1970), and the articles mentioned in Note 6. 8 So many publications have appeared on this topic that any selection would be arbitrary. But let me call attention to Thagard's article (1978) which, in my opinion, gives an excellent characterization of theory evaluation. 9 It is likely that a distinction has to be made between the degree of Humean probability of those instances of a general hypothesis that have been actually observed, and of those that have not (yet) been observed. But an examination of this point would lead us too far afield. ~0 Of course, one must be convinced that everything is normal, e.g. that nobody is playing a trick on us. For example, the (heavy) stone has been well examined, the room has been searched, etc. These assumptions are legitimate, since we are testing Humean uncertainty, and not our capacity to avoid being deceived. H See also Kyburg (1974, p. 192): " . . . among the things I am almost completely sure of, is the thesis that at least one of those things is false." n He probably also takes into account the statistical probability of certain hypotheses. But since it is likely that the notion of evidential probability corresponds closely to at least an elementary notion of statistical probability, I shall not deal here separately with statistical probability. ~3 The problem whether there exist (empirical) hypotheses that satisfy (PB) falls outside the scope of the present paper. ~4 See again Note 10.

A S O L U T I O N TO T H E L O T T E R Y P A R A D O X REFERENCES

353

Bar-Hillel, Y.: 1968, 'The Acceptance Syndrome', in I. Lakatos (ed.), The Problem of Inductive Logic, North-Holland Publishing Co, Amsterdam. Derksen, A. A.: 1978, 'The Lottery Paradox Resolved', American Philosophical Quarterly 15, 67-74. Goodman, N.: 1955, Fact, Fiction, and Forecast, Harvard University Press, Cambridge, Mass. Hume, D.: 1748, An Enquiry Concerning Human Understanding. Kyburg, Jr., H. E.: 1961, Probability and the Logic of Rational Belief, Wesleyan University Press, Middletown, Conn. Kyburg, Jr., H. E.: 1970a, Probability and Inductive Logic, MacMillan, New York. Kyburg, Jr., H. E.: 1970b, 'Conjunctivitis', in M. Swain (ed.), Induction, Acceptance, and Rational Belief, D. Reidel, Dordrecht. Kyburg, Jr., H. E.: 1974, The Logical Foundations of, Statistical Inference, D. Reidel, Dordrecht. Peirce, C. S.: 1934, Collected Papers, Vol. 5, Harvard University Press, Cambridge, Mass. Quine, W. V.: 1969, 'Natural Kinds', in N. Rescher (ed.), Essays in Honor of. Carl G. Hempel, D. Reidel, Dordrecht. Quine, W. V. and J. S. Ullian,: 1970, The Web of Belief, Random House, New York. Stemmer, N. 1971, 'Three Problems in Induction', Synthese 23, 287-308. Stemmer, N.: 1975, 'A Relative Notion of Natural Generalization', Philosophy of Science 42, 46--48. Stemmer, N.: 1978, 'A Partial Solution to the Goodman Paradox', Philosophical Studies 34, 177-185. Stemmer, N.: 1979, 'Projectible Predicates', Synthese 41,375-395. Thagard, P. R.: 1978, 'The Best Explanation: Criteria for Theory Choice', Journal of Philosophy 75, 76-92.