Inductive evidence for other minds

JEROME I. GELLMAN INDUCTIVE EVIDENCE FOR OTHER MINDS (Received 14 March, 1973) Some recent discussion of the problem of other minds concerns the inductive legitimacy of the argument from analogy. In God and Other Minds Alvin Plantinga attacks the analogical position for allegedly violating intuitively acceptable inductive principles. And in Reason and Scepticism Michael Slote amends the argument from analogy so as to avoid Plantinga's objection. In this paper, I assess Plantinga's attack and Slote's defense, and offer an alternative evaluation of the argument from analogy considered by Plantinga. Finally, I sketch an analogical-type argument for the existence of other minds that escapes entirely previous criticism of the analogical position. 1 As Plantinga presents it, the analogical position is that a person can inductively infer the existence of other minds from his total evidence, where S's total evidence is a set of propositions such that p is a member of it if and only if (1) p is either necessarily true or solely about physical objects, or a consequence of such propositions, and (2) S knows p to be true. 2 In particular, among others, such common-sense beliefs as (a) I am not the only being that feels pain. and (b) There are some pains that I do not feel. are justified inductively on this view (p. 252). Employing 'determines by observation' in the sense that one can determine by observation that certain behavior is being displayed, but Philosophical Studies 25 (1974) 323-336. All Rights Reserved Copyright 9 1974 by D. Reidel Publishing Company, Dordrecht-Holland 324 JEROME I. GELLMAN not that someone else is in pain, Plantinga defines a simple inductive argument for S as having this form: Every A such that S has determined by observation whether or not A is B is such that S has determined by observation that A is B. Therefore, probably every A is a B. (p. 251) or more generally: m/n A's such that S has determined by observation whether or not .4 is a B, were such that S determined by observation that A is a B. Therefore, probably m/n A's are B's. A direct inductive argument for S is an ordered pair of arguments the first member of which is a simple argument, and the second a valid deductive argument one premiss of which is the conclusion of the simple argument, where the other premisses are from S's total evidence. (p. 251) I can, then, give the following direct argument for (a) and (b), according to the analogical position: (1) Every case of pain-behavior such that I have determined by observation whether or not it was accompanied by pain in the body displaying it, was accompanied by pain in that body. So, Probably every case of pain-behavior is accompanied by pain in the body displaying it. But then, drawing from my total evidence, I observe that a body over there (other than my own) is displaying pain-behavior, but note that I feel no pain in that body. I conclude that that body contains a pain that I do not feel, from which I infer (a) and (b) I am not the only being that feels pain: There are some pains I don't feel. (pp. 253-254) Plantinga's criticism of the analogical position stems from the fact that I (logically) cannot determine by observation that a given bodily area is free from pain. The most I can observe is that I do not feel pain in that area, from which it hardly follows that no one does. Now let the I N D U C T I V E E V I D E N C E FOR O T H E R M I N D S 325 sample class of an inductive argument be the subset of the reference class of which X is a member if and only if it has been determined by observation (by the appropriate person) whether or not X has the sample property. We can then say of the simple argument (1), therefore (2), that it's (logically) impossible for its sample class to contain a counter-instance to its conclusion, since it's impossible for me to determine by observation that a piece of pain-behavior is not accompanied by pain. The inference from (1) to (2), contends Plantinga, is logically on a par with this argument for idealism: (3) Every physical object of which it has been determined whether or not it has ever been conceived O.e., perceived or thought of) has been conceived. So, (4) Probably every physical object is conceived. (p. 255) which obviously is unacceptable, but precisely because it's impossible for its sample class to contain a counter-instance to its conclusion, i.e., no one can determine by observation that an object is not conceived. Plantinga rules out both inductions with the following principle: (A) A simple inductive argument is acceptable only if it is logically possible that its sample class contain a counterinstance to its conclusion. (19. 255) a Given (A) we must reject the analogical position, for it will be impossible to construct arguments that support (a) or (b) or most of our commonsense beliefs about other minds without violating that principle. But if we reject (A), Plantinga contends (further) we can support (a) and (b) but can also then construct equally strong arguments against (a) and (b). Hence our belief in other minds will not be more probable than not on our total evidence. On either horn of the dillema, we cannot inductively support our belief in other minds. This is the essence of Plantinga's attack on the analogical argument. In his discussion of the analogical argument, Michael Slote accepts Plantinga's (A) as a binding principle. 4 Nevertheless, Slote thinks (A) can be avoided, by advancing an argument that is just not a simple inductive argument as defined by Plantinga. (A) places restrictions only 326 JEROME I. G E L L M A N upon simple inductive arguments, so Slote's analogical arguer is untouched by Plantinga's attack. Slote's inductive proof begins in this way: (5) Every case of (fttll-blown) pain behavior on the part of this (i.e. my) (human) body (that I have a memory impression of having had) has been accompanied by pain or the pretence of pain (namely, on my part). So, (6) (It is reasonable for me to believe that) evey case of (fullblown) pain behavior (on the part of any human body) is accompanied by pain or the pretence of pain. (p. 118) Now Slote is correct that his argument does not contravene (A), since it is not a simple inductive argument. However, the case for (A) can be made equally against Slote's argument. If the fact that I cannot determine by observation that pain-behavior is unaccompanied by pain matters, then it ought to matter here as well. For Slote is arguing from painbehavior that he observed to be accompanied by pain. These same cases of pain-behavior, however, could not have been found to be unaccompanied by pain. In fact, if any of these had not been observed to be accompanied by pain, Slote could have simply ignored those and fallen back on the remainder that he knew to be accompanied by pain. But if one accepts (A), as Slote does, by parity of reasoning we ought to reject any argument that admits of no counter-evidence. Consider the following variation on the above argument for idealism: (7) Every object I can remember having seen was conceived (by me). So, (8) Probably every object is conceived. This is not a simple inductive argument, but is no more successful than its predecessor. I cannot possibly remember seeing an object that was not conceived, and neither can I remember a case of my painbehavior which I knew not to be accompanied by pain. The problem here is the same as before: if an inductive procedure allows for no counterevidence, then it is not acceptable. Hence, Slote does not solve the problem of other minds. I N D U C T I V E E V I D E N C E FOR O T H E R MINDS 327 II Plantinga defends (A) for it intuitive plausibility. However, (A) seems to rule out intuitively acceptable simple inductive arguments. For example, consider this inference: (9) Every substance such that I have determined by observation whether or not it had a solvent, was such that it did have a solvent. So, (10) Probably every substance has a solvent. Now I do think it is logically possible to know that a given substance has no solvent, but it is not logically possible to determine that by observation, where to determine by observation is to determine in the evidencegathering sense appropriate for induction. In this sense I examine single A's or some finite number of A's at a time to determine a sample class of A's that ale B's, and project my findings upon the general population. In the sense of determine by observation appropriate for induction, no matter how many solvents I may have tested, it is always possible for the very next observation to determine a solvent of the substance in question. Hence, even if every single substance I have ever known was determined by me to have a solvent, according to (A) I still cannot justify my belief in (10) by inferring it from (9) via inductive procedures of observation. Even if my sample class contains thousands of members, it turns out that I still haven't the slightest inductive evidence for (10). True, I can construct a direct argument for (10)that does not contravene (A). Where 'N' is the number of solutions I have employed in my tests, I can argue as follows: (11) Every substance such that I have determined by observation whether or not solutions # # 1-N were its solvent, was such that at least one of # # 1-N was its solvent. So, (12) Probably, every substance has at least one of # # 1-N as a solvent. Therefore, (10) Probably every substance has a solvent. The inference from (11) to (12) does not contravene (A), since a 328 JEROME I. G E L L M A N counter-instance to the conclusion, (12), can be found in the sample class, viz., I can determine by observation with respect to a given substance that none of # # 1-N is its solvent. So, even if I am barred from giving a simple argument for (10) I can give a direct inductive argument for the same conclusion. However, the inference from (9) to (10) does not seem necessarily faulty at all. And surely it's counter-intuitive to maintain that the direct argument for (10) is evidentially superior to the simple argument for (10), providing inductive justification for its conclusion whereas the latter fails to provide justification of equal value for the same conclusion. The source of the difficulty is the comparison between the argument for idealism and the argument for (2). The former violates (A)because I cannot examine a physical object for the sample property - being conceived- while it lacks that property. That an object is a member of the sample class entails that it is observed or examined, and for that reason my sample class cannot contain an unconceived object. On the other hand, in the argument for (2), it is possible to examine pain-behavior for the presence of pain and for it to lack the sample property. The argument violates (A) because I cannot determine that pain-behavior is unaccompanied by pain. But it can be unaccompanied by pain all the while that I examine it. Let us define: an undecided case for S o f an A being a B as an A such that (1) it's logically impossible for S to determine by observation that an A is not a B; (2) it's logically possible for S to determine that an A is a B; (3) S's examining an A for B-ness and its not being a B is logically possible; and (4) after thorough examination under favorable conditions S fails to determine that this A is a B. We can then say that there cannot be undecided cases for us of physical objects being conceived, but there can be undecided cases of pain behavior accompanied by pain. This difference, furthermore, distinguishes the two arguments from one another with respect to their evidential value. For undecided cases ought not to count epistemically as anything worse than negative evidence. Suppose I have examined n A's for B-ness, out of which m ( < n ) w e r e found to be B's, and the remainder (n-m) were undecided cases (in accordante with the above definition). The probability of the next A being a B is surely no less than where n - m of the A's had been found not to be B's. Likewise, the probability is that m[n A's are B's, as certainly as in INDUCTIVE EVIDENCE FOR OTHER MINDS 329 the case where n-m A's were determined to be non-B's. For, again, as long as undecided cases are possible, that is, that my examination does not entail the possession of the sample property, we can make the evidential strength of my findings no weaker than if we were to suppose my undecided cases to have been cases of A's that are non-B's. I conclude, therefore, that it is not a necessary condition of the acceptability of a simple inductive argument that it be possible for its sample class to contain counter-evidence to its conclusion. Instead, I propose this principle: (B) A simple inductive argument for the conclusion that m/n A's are B's is acceptable for S only flit is possible for S t o examine an A and it not be a B. Furthermore, when it is logically impossible for S to determine by observation that an A is not a B, then a simple inductive argument for the conclusion that m/n A's are B's is essentially incomplete. A necessary condition of its acceptability is that the A's actually determined to be B's form m/n of S's examined class. We will say that X is a member of S's examined class if and only if S observes X to be an A that is a B, or S examines X, which he observes to be an A, to determine (under favorable conditions) if it is a B. (That X is a member of the examined class does not entail that X is in the sample class, though the entailment holds in the other direction.) Suppose m/n of S's sample class of A's are B's, but S has 1 ( > n ) members in his examined class, such that 1-n members are undecided cases. We have already noted that his undecided cases are no less worthy than non-B's, and now we may add that if S is to be empowered to project probabilities from his examined class he may not treat undecided cases as anything better than non-B's. From which it follows, using the above schema, that S is not justified in concluding that m/n of all A's are B's, but only that m/1 A's are B's. We therefore propose this supplementary principle: (C) A simple inductive argument for the conclusion that m/n A's are B's, where it is logically impossible for S to determine by observation that an A is not a B, is acceptable for S only if m/n of S's examined class was determined by observation to be B's. 330 JEROME I. G E L L M A N It might be more accurate to state (C) in application to arguments for the conclusion that at least m/n A's are B's for S may not know the exact proportion of B's amongst the A's, since he may have undecided. cases concerning which he cannot say whether they are B's or not Nonetheless, S can know that at least m/n A's are B's, and if m/n > 89 then S can be justified in believing that a majority of A's are B's, or that the next A will probably be a B. And if all the members of his examined class are members of his sample class, S is justified to conclude that probably all A's are B's. Here we note a possible objection. Undecided cases, it may be contended, ought not to count as either positive or negative evidence, but as evidentially neutral with respect to the presence of the reference property. S ought to be able to proceed from his sample class alone, and treat the remainder (if any) of his examined class (undecided cases) as though he had never observed them. In that case, using the previous schema, S is empowered to conclude that m/n of A's are B's, where n is the number of members in his sample class, rather than m/1, where 1 is the number of elements in the examined class. If accepted, however, this objection would convert inductive procedures where it was logically impossible to determine by observation that an A was not a B into (in a sense) fool-proof arguments for the conclusion that all A's are B's. For as long as S observes any A's to be B's he has a sample class, and once he has a sample class, on this view, he can argue from it and ignore the rest. If he doesn't determine any A's to be B's, all of his observation have no negative bearing on the conclusion that all A's are B's. Here I agree with Plantinga's insight, that an argument that is in this sense fool-proof is not acceptable. By introducing (C), however, we remove the invulnerability of the type of inductive arguments under discussion and create the possibility of justified inferences of that type. We now turn to the application of principles (B) and (C) to the arguments for idealism and the solubility of substances, respectively, and then will consider in turn the analogical argument. As already noted, the argument for idealism is effectively blocked by (B). But the inference from (9) to (10) - that every substance has a solvent - does not violate (B), since it is possible for me to examine a substance to find a solvent and for it not to have a solvent. Does this I N D U C T I V E E V I D E N C E FOR O T H E R M I N D S 331 inference violate (C)? Its impossible to say, for as stated the argument is incomplete. To know whether the conclusion is warranted I must, in addition to (9), know the relationship between the sample class and the examined class. If they are the same, then (C) is not violated, for every member of my examined class then has a solvent, and the argument is acceptable. But suppose my sample class is a proper sub-set of my examined class. In that ease, even if (9) is true, that every member of my sample class had the sample property, still I may not conclude that every substance has a solvent. That this is so is clearest in a ease where, say, my sample class is 10~o of my examined class. Surely, even though every member of my sample class has a solvent, I cannot conclude that every substance has a solvent. For out of the substances that I examined for a solvent 9 0 ~ were undecided eases. Since I cannot determine that a substance has no solvent (C) applies. And since (C) applies, in order to conclude that every substance has a solvent, every member of my examined class would have to have been determined to have a solvent. This result seems intuitively perfectly acceptable. 5 Equipped with principles (13) and (C), in place of (A), we now return to the direct inductive argument for (a) and (b). Recall that it begins with this simple inductive argument: (1) Every ease of pain-behavior such that I have determined by observation whether or not it was accompanied by pain in the body displaying it, was accompanied by pain in that body. So, (2) Probably every case of pain-behavior is accompanied by pain in the body displaying it. We have found that we cannot invalidate this inference on grounds that I cannot observe the absence of pain. We do not accept (A). But neither does the argument violate (B), since an examined case of painbehavior can be unaccompanied by pain. Now, whether it conforms to (C) depends on the truth of (13) Every member of my examined class of pain-behavior is a member of my sample class of pain-behaviors. which itself is true only if (14) I have never observed pain-behavior with respect to which I noted that I experienced no pain along with its occurrence. 332 J E R O M E I. G E L L M A N on the assumption that attending to my present mental states constitutes the only relevant examination I can undertake of a present case of painbehavior. Now, (14) is false in my case, since there are countless cases of painbehavior on the part of bodies (other than this one that I call 'mine') which I have noted not to be accompanied by my feeling pain. But neither can I conclude on the basis of my evidence that most painbehavior is accompanied by pain. For a majority of pain-behaviors observed by me are undecided cases with respect to the presence of pain. Hence, I cannot conclude that all pain-behavior is accompanied by pain, nor that even a majority are accompanied by pain. In my case, at least, the undecided cases far outweigh the positive instances. In so saying, I am not at all capitulating to Plantinga's earlier requirement, (A). For I envisage circumstances in which, with appropriate auxiliary premisses, the inference from (1) to (2) is perfectly justified. This would be so, if my sample class and examined class were identical, if, for example, no matter what body I observed displaying pain-behavior I always felt pain. I would then have good inductive evidence for the proposition that all pain-behavior is accompanied by pain (namely, my own). The inference is therefore not essentially defective. Yet, the argument does fail, but only for lack of evidence, i.e., (14) is not true for me. III In a sense, the preceding section is academic with respect to the problem of other minds, since whether one accepts (A) or (B) and (C) admittedly the analogical argument fails. In this section, however, I want to argue that there is a form of the analogical position that succeeds whether one accepts (A) or instead (B) and (C). In order to state this argument we must first characterize the notion of causal associations of an object, O, which is the network of regularly recurring or ongoing causal relations of O with a typical set of objects or kinds of objects (not identical to O or any part thereof). An example is the causal associations of a gasoline automobile engine, which includes the interaction between the parts of the engine and gasoline and air (the 'objects' with which it stands in ordinary causal relations), and the combustion of the gasoline and the subsequent driving of the pistons. All INDUCTIVE EVIDENCE FOR OTHER MINDS 333 of this helps partially characterize the causal associations of an automobile engine, that is, the normal manner of its interacting, and the objects or sorts of objects with which it interacts. Now I am going to speak of the relationship between mind and body, pain-behavior and pain, as part of the causal associations of the body, for this relationship constitutes part of the fixed network of the body's interaction with other objects or substances. (In addition, there is the digestive-segment of the causal associations, the respiratory, etc.) The philosopher who denies that the mind and body are causally related may, if he likes, speak instead of the accompanying associations of the body, and our purpose will still be served. Further, let us say that 01 and 02 are congruent with one another to the extent that they have similar gross configurations, have isomorphism of parts, have parts which share similar configurations, are composed of the same materials, and share similarities in the internal workings of their respective parts (without reference to objects or parts of objects other than O1 or 02, respectively). Obviously, there are degrees of congruence, and congruence with respect to specific aspects of objects. The argument I am now going to sketch justifies my belief that the more congruent objects are to one another the more probable it is that they have the same sorts of causal associations. For the sake of simplicity, though, I shall speak simply of congruent objects, where the intention is to speak of congruence of an extent, x, such that (a) objects showing degree of congruence x have been found to have similar causal associations and (b) human bodies bear to one another generally speaking congruence of degree x. On this basis, it is probable that human bodies have the same sorts of causal associations as one another, including causal association with the mental, such as pains and the like. Let us define (O, c) as any ordered pair whose first member is an object, and whose second member is a kind of causal association inJ eluded in the causal associations o f an object congruent with O. We can then state the argument more formally as follows: (15) In the vast majority of cases of ordered pairs (O, c) such that I determined by observation whether or not c was part of the causal associations of O, I determined that c was part of the causal associations of O. 334 JEROME I. GELLMAN So probably (16) (17) In the vast majority of cases, ordered pairs (O, c) are such that c is part of the causal associations of O. That body over there is congruent to this body ('mine'). Furthermore, (18) This body has as part of its causal associations a causal association with mental states, e.g., when it displays painbehavior a pain occurs. Therefore, probably (19) (20) The ordered pair (that body, causal association with mental states) is such that its second member is part of the causal associations of the first member. (from (16), (17), (18), and the definition of (O, c)) e.g., when that body displays painbehavior it is in pain. That body is now displaying pain-behavior. So, probably (21) There exists a pain in conjunction with that pain-behavior. (from (19) and (20)) But, (22) I do not now feel pain. So, (23) There are pains I do not feel. And (24) I am not the only person who feels pain. The only part of this argument that is inductive is the inference from (15) to (16). The remainder is deductive, drawing upon the conclusion of the inductive inference as well as upon my total evidence. The simple inductive argument clearly conforms to my principle (B), for it is logically possible for me to examine an object, O, congruent to an object O' such that there is a c which though part of the causal associations of O' is not part of the causal associations of O. In addition, the argument does not violate Plantinga's (A), for it is INDUCTIVE EVIDENCE FOR OTHER MINDS 335 possible for me to determine by observation counter-evidence to the conclusion, (16). It need only be possible that sometimes I determine by observation that congruent objects have disparities in their respective causal associations, and this is certainly logically possible, e.g. with respect to two congruent machines I could discover relevant differences in their causal associations as far as I have observed. And since this is logically possible, the argument from (15) to (16) does not contravene (A). Of course, I cannot determine that congruent human bodies are disparate in that part of their causal associations that relates to mind, but the simple inductive argument we are considering does not have human bodies or human behavior as its sample class. Rather, the sample class is much wider, which very fact allows the opportunity of counterevidence. I then deduce facts about mind and body, viz. that they conform to the general facts I have discovered in the world at large. Neither does the inference violate another principle endorsed by Plantinga: d (A 1) Where a, fl is an inductive argument for S, fl is of form All A's have B, and C is any part of fl; a, fl, is acceptable for S only if the propositions S has examined an A and determine 4 by observation that it lacks C and S has examined an A and determined by observation that it has C are both logically possible. (p. 453) There seem to be no 'parts' of the concepts of 'congruence' or 'causal associations' which do not conform to the conditions in (A1). Finally, (C) is not violated, since like (A) it is inapplicable) As far as the acceptability of (15), it seems to me to be certainly true and is employed by me often in perhaps less philosophically pertinent inductive justifications. Since (15) is known to me, and since the inference to (16) conforms with the necessary conditions set down by Plantinga and Slote, as well as by me, I conclude that there is no known reason to reject this argument. Since, therefore, it otherwise appears to conform to the cannons of inductive and deductive reasoning, I conclude that it justifies inductively my belief in other minds. University of the Negev, Beer-Sheva, Israel 336 JEROME I. GELLMAN NOTES 1 I want to thank Alvin Plantinga for helpful discussion of some of the points in this paper. I am also indebted to Henry Schnur and Ira Schnall for useful suggestions. God and Other Minds, Cornell University Press, Ithaca, 1967, p. 247. Hereafter, pages from this work will be cited in the text. See also, Plantinga's 'Induction and Other Minds', Review o f Metaphysics 19, 441-61. 3 In addition, Plantinga provides a principle, (A1), to cover certain peculiar arguments not covered by (A)-(A1), however, needn't concern us until later (see below). In the meanwhile, our remarks concerning (A) apply equally to (A1). 4 Reason and Scepticism, George Allen and Unwin. Ltd., London, 1970, p. 115, References in the text are to this volume. See also, Slote's, 'Induction and Other Minds', Review of Metaphysics 20, 341-60, and Plantinga's reply, 'Induction and Other Minds, II', Review o f Metaphysics 21, 524-33. 5 This case is complicated by what we shall count as sufficient 'examination' of a substance to find its solvent. We shall have to require 'full' examination, involving something like testing with all known solutions and not simply a partial examination. 6 This argument also obviates completely the so-called 'my case alone' objection to the analogical argument, for the inductive base of this argument is much broader than evidence about my body and my mind. It includes evidence pertaining to all manner of things observed to be congruent in the specified ways. Hence, this oft-cited objection does not stand against this version of the argument.