Nozick On Scepticism

Citation preview

Nozick On Scepticism Graeme Forbes The Philosophical Quarterly, Vol. 34, No. 134. (Jan., 1984), pp. 43-52. Stable URL: http://links.jstor.org/sici?sici=0031-8094%28198401%2934%3A134%3C43%3ANOS%3E2.0.CO%3B2-V The Philosophical Quarterly is currently published by The Philosophical Quarterly.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/philquar.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact [email protected].

http://www.jstor.org Mon Jun 4 08:33:17 2007

ne PhilosophiralQuanerlyVol. 34 No. 134 ISSN 0031-8094 $2.00

NOZICK ON SCEPTICISM

In standard sceptical arguments for our lack of knowledge about the external world, there is an assumption roughly to the effect that if a subject S knows that p, and knows that p entails q, then if S knowingly infers q from p and thereby comes to believe q, then he knows q. The sceptic chooses for q, say, that S is not a brain in a vat being stimulated to have experiences as of an external world, a proposition which, he supposes, S does not know to be true. But since all S's ordinary beliefs of the form 'I am now seeing (hearing, touching, etc.) such and such a thing', 'There are such and such things here', and so on, entail the truth of q, and since nothing prevents S from performing the relevant deductions, it follows by the assumption that these ordinary beliefs are not known by S to be true. Let us call this assumption the Transmission Principle: it says that the status of knowledge is always transmitted to the believed conclusion of a knowing inference from known premisses via known entailments. Robert Nozick has recently tried to refute the sceptical argument by providing an analysis of 'S knows that p' on which the Transmission Principle is false. The analysis employs counterfactuals, and has the four clauses (1) pis true; (2) S believes p; (3) if p were false then S would not believe p; (4) if p were true then S would believe p and would not believe

not-^.^ If (1)-(4) give the correct analysis of 'S knows that p' then the Transmission Principle is false. Let p, be the proposition that I am living a normal life in New Orleans and let q, be the proposition that I am not a brain in a vat being stimulated to have experiences as of living a normal life in New Orleans. Then I know p,, and know that p, entails q,, and can knowingly infer q, from p,. I

' Robert Nozick, Philosophical Explanations, Oxford 1981. All Nozick references are to this work. In Lewis-Stalnaker terminology, Nozick holds that if p is true at w, then 'if it were that p it would be that q' is true at w iff q is true at every p-world closer to w than is any not-p world (Nozick, p. 680).

44

GRAEME FORBES

now do this. Yet I have not added q, to my store of knowledge, according to Nozick's analysis, since the appropriate instance of (3) is false; for even if q, were false, I would still believe it. Thus Nozick establishes at least the epistemic possibility of knowledge, that for all we know, we do know some propositions about the external world, since the sceptic's argument to the contrary relies on a false principle. However, this refutation of the sceptic comes at a high price, namely the rejection of the Transmission Principle. Surely deduction from known truths, at least in the hands of people who know what they are doing, is a method guaranteedto add to the stock of things we know?When an apparently plausible principle is rejected, it is insufficient justification to show that the rejection is demanded by a theory with pleasing consequences; an explanation, perhaps of a debunking sort, of why the principle seemed so plausible, is also appropriate, yet Nozick does not provide one. My view is that it is not possible to explain away the plausibility of Transmission, since it is in fact ~ o r r e c tAs . ~ this is incompatible with (1)-(4)'s being the last word on 'S knows that p', it behooves us to find a flaw in Nozick's approach. We consider in turn the sufficiency of the conditions, the necessity of (4) and the necessity of (3), before addressing the main task of this paper, that of giving a positive argument for the Transmission Principle.

The condition which makes the difference between a subject's truly believing that p and his knowing that p is that of his being sensitive to the facts (Nozick, p. 176), or, as it may also be termed, of his trackingthe facts (p. 178). Someone who offers a counterfactual definition of 'S knows that p' may be said to hold the view that tracking is an "inherently counterfactual" notion; thus one must distinguish between objections to his analysis which suggest merely that he has misformulated the counterfactuals, and objections which suggest that tracking is not an inherently counterfactual notion at all. One must also consider the possibility that whether or not a counterfactual approach is correct depends upon the kind of knowledge which is being analyzed. T o show that Nozick's criticism of the sceptic's argument is ineffectual, it suffices that we establish that the counterfactual approach is incorrect for inferential knowledge and so does not undermine the Transmission Principle, on which the sceptic re lie^.^ A familiar type of case involving inference, upon which attempts to suppleThis view is defended in N following. I am grateful to ChristopherPeacocke for pointing out to me that the arguments in an earlier version of this paper told most strongly against applying the counterfactual analysis to inferred beliefs, and for suggesting that counterfactuals can give at least necessary conditions for noninferential cases.

NOZICK ON SCEPTICISM

45

ment (I) and (2) to obtain conditions sufficient for knowledge often founder, is the Gettier type, where a true belief is inferred from a false one, the inferred belief being true in virtue of the fortuitous obtaining of a state of affairs with which the believer is not properly in ~ o n t a c t Harman .~ has suggested the requirement that "the lemmas be true" to exclude such beliefs from the domain of knowledge, a requirement whose effect Nozick hopes to capture; but this cannot be done with (1)-(4), a fact Nozick himself almost recognizes (p. 190):

A person comes to believe that a vase is in a box by seeing an illuminated hologram, part of a machine which alternates between displaying the hologram and the real vase containingin the box . . . the machine, in alternate time periods, displays a hologram of a vase only when a vase is pressing down on a lever (it somehow detects a vase and not another thing). Hence if there weren't a vase there, he wouldn't believe there was one; and if there were one there, he would come to believe, by looking, that there was. Thus, our account has the consequence that he knows a vase is there, even when he is seeing the hologram but thinks he is seeing the vase. Nozick remarks only that this consequence is "somewhat counterintuitive" (loc. cit.), apparently not recognizing that the example is just a Gemer case with an extra twist to make the appropriate instances of the counterfactual schemata (3) and (4) come out true. The case is certainly not one where theory may be allowed to ovemde intuition: the subject infers the true belief that there is a vase in the box from the false belief he would express by the sentence 'That's a vase in the box there7,a belief he acquires when he sees the hologram. It must be emphasized that seeing the vase-hologram is quite unlike seeing a picture of the vase in the box, since the hologram does not represent the vase in any way anaIogous to the way a picture does; for example, if the hologram looks like the vase, this may be purely accidental. So in this case, the lemma is certainly false, yet Nozick's analysis attributes knowledge. Moreover, no readjustment of clauses (3) and (4) will solve the problem, since we can always arrange causal ties between the deceptive state of affairs and the fortuitous belief-verifjmg state to ensure that counterfactual clauses give the wrong verdict about these inferential examples.

Gettier cases show that (3) and (4) are not what (1) and (2) need to be supplemented with to obtain conditions sufficient for knowledge in general. If the Transmission Principle is correct, they are not necessary either, as is See E. Gettier, 'Is Justified True Belief Knowledge?', Analysis 23 (1963) pp. 121-123.

46

GRAEME FORBES

shown by the example in I of my being a brain in a vat. But before turning to a detailed defence of the Principle, we will consider how (3) and (4) fare with certain cases of non-inferential belief, cases which provide further illustration of the need for an articulation of kinds of knowledge. In connection with cases of non-inferential belief, Nozick introduces a relativized notion of knowledge via a method M. For S to know p via method M it is severally necessary and jointly sufficient that p is true, S believes p via M, and that the following refinements of (3) and (4) hold (p. 197): (5) if p were not true and S were to use M to decide whether or not that p, S would not thereby come to believe p; and (6) if p were true and S were to use M as above, S would believe p. T h e final account is that S knows that p iff there is a method M such that he knows p via M, his belief via M satisfies the just-mentioned conditions, and all other methods via which he actually believes p which do not satisfy these conditions are outweighed by M (p. 180). In the examples to follow, just one method will be in question, so weighting considerations will not be operative. It is easy to see why (5) is to be preferred to (3), for S might know that p in circumstances in which there is a dormant failsafe mechanism for the production of the belief that p in S, which would be activated were p false. But typically such mechanisms involve S acquiring the belief without using the method M he actually uses, and so their existing in the actual circumstances does not falsify (5) even if it does falsify (3). In other cases, where because of the functioning of such a mechanism, S would acquire the belief that p via M even if p were false, it seems correct to say that S does not actually know that p. Imagine that a man correctly believes that he is talking to a certain friend of his on the telephone, but an actress intending to imitate his friend's voice only just failed to get through to this man before his friend did, so that even if the man had not been talking to his friend, he might still have believed that he was. Nozick says that in this case the man does not really know that he is talking to his friend, which is surely the correct verdict; for there is a relevant alternative, that he is talking to the actress, which he would not be able to distinguish from what is actually the case.6 Clause (5) gets this recognitional example right, but we can construct a more acute test case for it. If S is to know p, M must be a way of acquiring knowledge, but it seemsprimafaciepossible that although M is such a method in the actual circumstances, it would not be if p were false, and S in using M might then be misled into believing p. Suppose we have built an extremely sophisticated computer, a machine from which we acquire knowledge given In the terminology of Alvin Goldman's "Discrimination and Perceptual Knowledge", The Journal ofPhilosophy 73 (1976) pp. 771-791, the situation which almost occurred is a "perceptual equivalent" of the actual situation (op. cit. p. 783).

NOZICK ON SCEPTICISM

47

input for which it is programmed. Suppose that this machine has the capacity to survey its own circuits and report malfunctions, and that the chances of any malfunction in normal circumstances are extremely slight, but that malfunction affecting its self-surveying capacities is amongst the simplest of ways in which it can malfunction. Let p be the proposition that the computer is functioning properly. Then in the actual situation, we might want to say that S comes to know that p by asking the computer to report on its internal states, just as we would certainly say that he comes to know about, say, the dismbution of a certain word in a text, by asking it to perform a textual analysis. But ifp were false and S were to use the same method to ascertain the truth-value ofp, S might still believe p, since the malfunction might cause the computer incorrectly to pronounce its circuits to be in order. Thus S knows p, but (5) is false, or so it would be claimed by someone who does not hold an "inherently counterfactual" view of tracking, but instead the view that, say, a tracking method is a method which is reliable, i.e. likely to produce true beliefs; for ex hypothesi the computer is reliable, even concerning its internal states. Yet one undeniably has a sense of unease in attributing knowledge here, a sense perhaps rooted in the fact that the computer might say that it is properly functioning even if it were not. If this is the right reaction, then there are no counterexamples to (5) with the structure sketched above, and that is strong confirmation that tracking is inherently counterfactual in such non-inferential cases: the problem with the computer is that it does not track the facts (in the counterfactual sense) about its own internal states. However, these considerations bear only on (5); (6) is more problematic, and Nozick's applications of it are mutually inconsistent. The idea of the clause is that if S knows p by method M, and M is tracking the facts, then if small changes in circumstances not affecting the truth of p were to come about, S would still believe p, using M. But this condition cannot be necessary for knowledge so long as we are willing to allow that knowledge can be acquired opportunistically, when circumstances briefly and coincidentally come together to make its acquisition possible; for a small change in circurnstances might prevent this fortunate coalescence, so that even if S used M he would not come to believe that p. Nozick mentions just such a case (p. 193): a bystander happens to be watching a particular bank-robber just as his mask slips off and recognizes Jesse James. He knows the robber is James, yet (6) appears false, for a world in which James's mask does not slip which is otherwise as like the actual world as possible is one in which the bystander does not come to believe, as a result ofwatching him, that the robber is James. Nevertheless, Nozick says that (6) is true (p. 193): "His method is looking and concluding it is Jesse James on the basis of seeing certain things. By that method (applied in this way) he would know in other cases also". These remarks violate a clear intuitive distinction between the method the

48

GRAEME FORBES

bystander uses (observation, or perhaps, training his attention in a certain direction, at a certain time, from a certain point of view) and the evidence he acquires by use of the method (the content of the sense experiences he undergoes). Nozick does suggest (p. 184) that this distinction should be collapsed, but this unattractive policy is anyway inconsistentwith what he says about other cases. If S cannot distinguish between the identical twins Judy and Trudy, but as the result of a bump on the head acquires the belief thatJudy has a mole at the same time as she develops this differentiatingfeature, S does not know he is meeting Judy if he is then presented with her, although he believes correctly that he is. Nozick says that in this case (6) is false (p. 191); "if Judy were before him, but in the very close situation of not having developed the mole, he wouldn't believe it was she". By the token of his treatment of the Jesse James example, he should rather hold that (6) is true, since ifJudy does not develop the mole, S does not see the same things, and so is not using the same method. It is hard to find an improvement of (6), within the confines of a counterfactual analysis, which gives the correct verdict upon these cases. One might try to rely wholly on (5), holding that its relevant instances in the bystander and twins examples are respectively true and false: (5) is false in the twins example, because circumstances in which Trudy, instead of Judy, develops a mole, are sufficiently close, and in such circumstances with Trudy before him, S still believes it is Judy. But if that is right, why are circumstances in which one of James's henchmen fashions and wears a Jesse James mask not sufficiently close to falsify (5) for the bystander? Another track is to add to (6) the requirement that the evidence S gathers by application of M is to be held constant; then circumstances in which the mask does not slip are irrelevant. But so would be circumstances in which Judy does not develop the mole. It might be said that we can count as sufficientlyclose, circumstances in which S does not suffer a bump on the head and does not acquire the belief that Judy has a mole. This makes (6) false for the one case, but by the same token, also for the other, since circumstances in which the bystander does not bother to pay much attention to the wanted posters for James will also be sufficiently close. A better strategy is to build into (6) the condition that similar types of evidence result from the use of M, so that in relevant worlds our bystander believes that it is James when he looks in the right direction at the right time. We then deal with the Judy-Trudy case by saying that, like the vase case, it is one for which the counterfactual analysis of tracking is inappropriate. For it is quite plausible that S, at least implicitly, uses inference to amve at the belief that it is Judy before him. He knows that it is either Judy or Trudy; he believes that it is Judy, and only she, who has a mole; he knows that the twin before him has a mole; and so he infers that the twin before him is Judy. But he does not

NOZICK ON SCEPTICISM

49

know that it is Judy because he does not know that Judy has a mole, since he acquired the belief that she does as a result of a bump on the head. Once again, it is the Transmission Principle which explains failure to know in a case of inferential belief, a case beyond the powers of a counterfactual approach to handle. Finally, before turning to defence of Transmission, a caveat should be entered. The most that could be concluded in favour of Nozick from the discussion above is that (5) and a modified version of (6) are necessay for non-inferential knowledge. We will not pursue the question of whether they are sufficient here, other than to note that (5) and (6) are satisfied ifa hypnotist induces in me a belief which in him is knowledge, a hypnotist with the intention only to make me believe what he knows; but I find the idea that such hypnotism transmits knowledge somewhat dubious.

There is an intuition of great plausibility in the Transmission Principle, but we need to spell it out further to see that it does state conditions sufficient for knowledge by deduction. In our rough formulation at the beginning, we used the idea of a "knowing" inference and of "thereby" acquiring a belief to gloss over a number of qualifications. First, a man may not believe the conclusion of an inference because it is so astonishing; hence our principle says only that if the conclusion is believed, then it is knowledge. (Here I note a problem for those like Goldman who would hold some kind of transmission principle for justified belief, to the effect that scientific method knowingly applied to justified belief yields justified belief. An unjustified conclusion may be believed in certain cases: consider the nineteenth century logician who is justified in accepting naive set-theoretic principles, and then simply believes the contradiction Russell inferred from them when he sees the inference.' Secondly, a man might know that p, carefully deduce q from p, survey his deduction for errors, fail to find any, and then come to believe q. But we can imagine that this last stage was not under his intentional control: instead, the belief that q was caused to arise in him by a mechanism triggered in part by the perception of the particular pattern of marks on the sheet which constitute the deduction, a mechanism quite insensitive to the semantic content of those marks. Here the man believes via a deviant causal chain, and it is in some sense just luck that it is q which he comes to believe. In this type of situation, we will say that he does not "thereby" come to believe that q; of course, this is simply a ' Goldman's defence of justification-transmission is in his "What is Justified Belief!", in Justifcation and Knowledge ed. Pappas, Dordrecht 1979 pp. 1-23. See especially principle (6B), p. 14, and clause (lo), p. 20. The example in my text is due to Kit Fine.

50

GRAEME FORBES

stipulation: the problem is to explain the difference between deviant and non-deviant causal chain^.^ Another type of difficulty concerns the competence of the subject at deductive inference. If a subject is incompetent at deduction when arguments reach a certain level of complexity, then even if he knows p and infers q from p while attending to what he is doing, it may be a matter of luck that the inference turns out to be correct. Thus a high degree of competence must be required for knowing inference. There are other minor details. For instance, someone might infer a disjunction from a disjunct he knows, the new disjunct being one whose meaning he does not even grasp. At best he knows as a result of this that a certain sentence is true, so we will not count this as knowing inference either. On the other hand, we will allow that the conclusion of a deduction can be known even if the subject does not know that he knows the premisses. No doubt there are other problems too, but it seems that at most these would demand further constraints on the notion of knowing inference, rather than giving up the Transmission Principle; contrast the counterexample just mentioned to transmission for justified belief. It may seem that so many qualifications are being built into the concept of a knowing inference that the resulting Transmission Principle is not merely true, but wholly innocuous, a principle which no-one would wish to deny. But this is a mistake: Nozick would wish to deny it, if his refutation of scepticism is to stand, for my inference by one step of modusponnzs that I am not a brain in a vat from my knowledge that I am living a normal life in New Orleans and my knowledge that that proposition implies that I am not a brain in a vat, is a knowing inference, even as the content of that idea has been circumscribed in the previous paragraphs (since I already believe I am not a brain in a vat, we should perhaps say just that the belief is thereby reinforced). So Transmission as we now understand it still licenses two conditionals which Nozick says are false, that if I know I am living a normal life here then I know I am not a brain in a vat, and its contrapositive. The case that inference yields knowledge in situations meeting the conditions of the Transmission Principle may be made as follows. Counterexamples to the sufficiency of conditions for knowledge supplementary to (1) and (2) typically involve circumstances in which all the conditions are fulfilled but it is nevertheless in some sense an accident that the subject acquired a belief which is true. In relevant alternative examples, the subject might easily acquire a false belief by contact with a deceptive state of affairs. In Gettier examples, it is the subject's good fortune that some additional state ofaffairs in the offing verifies the belief he infers from a false beliec intuitively, it is fortuitous, for example, that the box is so designed that only a vase pressing on For a theory of this distinction, see Peacocke,Holistic Explanation, Oxford 1979, Ch. 2.

NOZICK ON SCEPTICISM

51

the lever produces a vase hologram. So in each example, there is a sense in which the subject is "close" to failure to acquire a beliefwhich is true (I do not claim that the sense of 'close' is the same for both). The bystander in the Jesse James case may be contrasted, for although it is true that if the mask had not slipped then he would not have acquired his true belief about the identity of the robber, he is not close to failure to acquire a true belief: our modification of (6) to include the specificationthat evidence be held constant excludes a world in which the mask does not slip from the class of worlds where failure to acquire a true belief matters. The main problem here is to find a unified account of inferential and non-inferential cases which explains why we do think that failure to acquire a true belief in some worlds where the design of the box is changed to allow non-vases to produce a vase hologram is relevant to knowledge attributions. We have seen that counterfactuals will not produce such a unification, but I will not pursue any alternatives in this paper. It can now be seen that satisfaction of the Transmission Principle's conditions is sufficient for acquisition of knowledge, because these conditions function to rule out various ways in which it might be an accident that the subject acquires a true belief, ways in which he might be close to failure to acquire a true belief. Assuming the conditions are sufficiently comprehensive (they could always be added to) this means that the only way in which an element of the accidental can enter is through the machinery of inference itself. But it is no accident that employment of this machinery in otherwise favourable circumstances leads from truths to truths, since inference rules are necessarily truth-preserving. Hence knowledge must be acquired from deduction in situations meeting the prin~iple.~ If this defence of the Transmission Principle is correct, the Nozick's extraordinary account of the formal properties of the operator 'K' (p. 227229) is as implausible as it intuitively appears to be, and must be set aside along with the attempted refutation of scepticism it supports. So finally, how should we respond to scepticism?If I can know that I am not a brain in a vat by inference from other things I know, do I not then know that the sceptic's hypothesis is false. Nozick justly remarks that such attempts to show that we do know that sceptical hypotheses are false "strike us as suspicious, strike us even as bad faith" (p. 201). However, it is consistent to hold both that we do know these hypotheses are false, because we know such things as that these are two hands, and that arguments exploiting Transmission to show that we know it are without force if they have such premisses as that we know that these are two hands; and it is the failure of such arguments to have any force This claim is consistent with our verdict about the computer example, since there there is no inference. In his "A Cognitive Cul-de-sac", Mind 91 (1982) pp. 109- 111, Fred Dretske gives an example which he thinks defeats the Transmission Principle. I would simply deny that in his example, the subject does not know the inferred proposition.

52

GRAEME FORBES

which accounts for our feeling that use of them is suspicious, or bad faith. The trouble lies in the structure of the dialectical situation obtaining between the sceptic and his opponent, which requires more for the refutation of sceptical claims than the production of sound arguments that they are false. The sceptic's claim, embedding a hypothesis within an epistemic modality, is, let us say, that for all S knows, S is a brain in a vat being stimulated to have experiences as of leading a normal life. The point of the claim is that if the embedded hypothesis were true then none of S's empirical procedures for arriving at beliefs would yield knowledge, which enables the sceptic to exploit a dialectical principle which says that any knowledge S has acquired using these procedures is subjudice or inadmissible in the dispute between S and the sceptic; so S cannot use an argument beginning 'I know I have two hands'. Someone who doubts that there is a principle of dialectics to this effect should consider the analogous case of the creationist who believes that the world was created circa 4,000 B.C. along with the evidence that it has existed longer. The creationist need only say that for all we know, this belief of his is true; then even though we know he is wrong because we know there are 10,000 year old fossils, we cannot refute him with arguments citing this knowledge in their premisses, since if his belief were true, then our empirical procedures such as carbon dating would not be methods of acquiring knowledge. It is the dialectical principle which makes this type of creationist position so frustrating: in general, to know that an hypothesis is false is not to be in a position to refute it. On this interpretation of scepticism, then, we do know that the sceptic's hypothesis is false, but we do not know how to refute scepticism, since a refutation demands an argument with dialectical force. A demonstration of a fallacy in the sceptic's own argument from 'S does not know not-H' to 'S knows almost nothing' would have such force, but the Transmission Principle relied on to work the argument is wholly unfallacious.

Tulane Universitj