Harman's Paradox Tom Sorell Mind, New Series, Vol. 90, No. 360. (Oct., 1981), pp. 557-575. Stable URL: http://links.jstor.org/sici?sici=0026-4423%28198110%292%3A90%3A360%3C557%3AHP%3E2.0.CO%3B2-V Mind is currently published by Oxford University Press.
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/oup.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 08:33:21 2007
Mind (1981) Vol
XC,
557-575
Harman's Paradox* TOM SORELL
Gettier's examples show that someone can believe a truth, have good reasons for believing it, and yet lack knowledge.1 How can this be? One explanation is that a man's good reasons can depend upon or include false beliefs. Another possibility is that a justified true belief can fail to stand in an appropriate causal relation to the fact which makes the belief true. Both explanations fit Gettier's examples, but perhaps neither is sufficiently general : each assumes, as indeed Gettier's exposition does, that someone is adequately justified or not in virtue of evidence he actually has for a truth. Might not a man's good reasons be said to be defeated, undermined, or rendered inadequate, by further evidence he would, but happens not to, take into account? This is the possibility adverted to by certain defeasibilist analyses of knowledge and nonknowledge. Such analyses grant that Gettier's subjects are justified in believing a truth, but they explain in a distinctively general way the fact that those subjects lack knowledge. The lack ensues from inconlplete justification. If someone's actual justification for a true belief admits of supplementation or correction that would disclose evidence against the belief, then one's actual basis for believing is incomplete. Conversely, if correction or supplementation were not to detract from one's reasons, one is actually completely justified, and one's true belief is knowledge-ranking. I n Gettier's cases, conditions of complete justification fail to be met. But complete justification might be absent, too, where one's reasons involved no falsehoods, and where one's true belief was non-deviantly caused. In those circumstances it could still turn out that exposure to further evidence would warrant the retraction of some actually held true belief. Defeasibilism thus seems to alert us to, and to account for, cases of non-knowledge that Gettier did not prepare us for.
*
I
An earlier version of this paper, then entitled 'The Instability of Knowledge', was read to the Jowett Society in Oxford, in Hilary Term, 1979. John Sibson presented a helpful reply. I am indebted to Christopher Peacocke for comments on a later draft. Peacocke's ideas have influenced sections 111 and IV, as have considerations in Thomas Nagel's 'Subjective and Objective', in Afovtal Questions (Cambridge : University Press, 1979)~ pp. 196-214. 'Is Justified True Belief Knowledge?', Analysis 23 (1963), pp. 121-123.
5 9 TOM SORELL: There is reason to think, however, that the resulting distinction between knowledge and non-knowledge is improperly drawn. Harmanl has devised examples which suggest that not only justified true belief, but also knowledge, can co-exist with defeating evidence. Briefly, further evidence can be evidence against what one knows. If that is right, the presence or absence of defeating evidence cannot make the difference between non-knowledge and knowledge. So defeasibilism seems to fail-provided there is such a thing as knowing a truth there is further evidence against. And about that there is an air of paradox. As Harman puts the matter, 'If I know that h is true, I know that any evidence against h is evidence against something that is true ; so I know that such evidence is misleading. But I should disregard evidence that I know is misleading. So, once I know that h is true, I am in a position to disregard any future evidence that seems to tell against h.' This is paradoxical, because I am never in a position simply to disregard any future evidence even though I do know a great many different things.2 If paradox is the price of denying that knowledge is undefeated, justified true belief, then perhaps defeasibilism deserves reinstatement. Either that, or the paradox should be dispelled. But does the argument for the paradox create conviction? A doubt is likely to attach to the conditional which functions as initial premiss. Is it true that 'If I know that h is true, I know that any evidence against h is evidence against something that is true'? Unless to know is to know that one knows, so that evidence against h is evidence against something I can tell is true because known, the conditional seems to hold only when the second occurrence of 'I know that' is dropped. But then there is no paradox. It appears that either the puzzle is contrived, or else, implausibly, the iteration principle comes into its own and to know is to know that one knows. Are we in a position now to dispose of Harman's paradox, and with that, defeasibilism? Not yet: I argue that the puzzle can be made independent of the iteration principle (sections I and 11), and that it trades instead on a long-standing dogma in the theory of knowledge (sections V and VI). T h e dogma can be stated as the thesis of the stability of knowledge : I
2
Thought (Princeton: University Press, 1973))ch. 9, p p 142-154. Familiarity with Harman's discussion will be assumed in what follows. Ibid. p. 148.
One's knowledge of a truth cannot co-exist with a reason for doubting, disbelieving, or suspending judgment concerning that truth. More informally, to know a truth is to be forearmed against rational second thoughts about it. T h e upshot of this paper is that new and old versions of the thesis are false. Once that is seen, defeasibilism becomes unpersuasive and Harman's paradox dissolves.
I begin by setting aside the iteration principle. Associated in our own day with Hintikka,l it seems to be invalidated by two types of counter-example. Both types make something of the intuitive connection, exploited by Ramsey2 and Unger,3 between knowing and non-random, reliable, or non-accidental arrival at truth. I n the first sort of example, the Ramsey-Unger connection is made to engage with intuitions about the liguistically unsophisticated. Specifically, we can imagine a child who has not learned much of a language, but who responds differentially in language to e.g., the presence of its mother. When and only when its mother is present, the child comes out with a noise that sounds like 'Mama'. Granted that there is not just a conjunction of, but a connection between, the making of the sound and the presence of the child's mother, it seems true of the child, on each occasion of its coming out with the noise, that it knows its mother is there. T h e one-word sentence expresses a truth non-randomly, non-accidentally, or reliably. But the evidence does not appear to warrant an attribution to the child of the knowledge that it knows its mother is there. T h e latter attribution insinuates a conceptual sophistication, specifically, a command of intentional concepts, starkly at odds with the evident resources of the child's vocabulary. The second kind of case concerns anyone who in fact arrives at a truth by reliable means, but who does not remember, or who doubts the authority of, the source of his information. This can be the plight of many of us most of the time. I am not sure how I picked up the fact that Schubert's ninth symphony is called 'The I
2
3
Knowledge and Belief: A n Introduction to the .Logic of the Two Notions ( I t h a c a : Cornell U n i v e r s i t y Press, 1962). ' K n o w l e d g e ' , i n Foundations of Mathematics and Other Logical Essays { L o n d o n : Routledge and K e g a n Paul, 193 1). An Analysis o f Factual K n o w l e d g e ' , J. Phil., 65 (1968), pp. 157-170.
560 TOM SORELL: Great'. But if my source was actually an accurate musical encyclopaedia, and if my once reading it is responsible for my coming out now with the title of the symphony, then even if I believe (falsely) that I dreamed up the name of the symphony, as a matter of fact it is no accident that I am right: my source of information is of a type apt for producing knowledge of the names of symphonies. Here are circumstances in which it is plausible that I know, but flatly false that I know that I know.
Harman's paradox seems to depend upon Hintikka's principle: unless knowing a truth is conditional on recognizing as much, it is hard to see how further evidence against the truth can be, as Harman requires, recognizably misleading. But a source for the paradox remains even if, in deference to counter-examples, the possession of knowledge is not made conditional on its recognition. Take Lemmon's modest proposal, once put up as an alternative to Hintikka's principle,l according to which to know something is to have learned and not forgotten that thing. Lemmon's suggestion seems to entail, quite plausibly, that once acquired, an item of knowledge endures so long as one's memory of it does. This means that only two kinds of event-acquisition and loss-are foreseen by Lemmon in the natural history of an item of knowledge. Conditions of loss of knowledge, moreover, are narrowly circumscribed: to lose one's knowledge is to forget. Accordingly, to be on the point of losing one's knowledge is to be on the point of forgetting. Lemmon's account does not make room for a further kind of instability in the possession of knowledge, a kind of instability that owes nothing to portending lapses of memory. Unless the latter phenomenon is discerned, an account of loss of knowledge, hence of conditions of non-knowledge, is incomplete. But once discerned, the phenomenon seems to give rise to Harman's paradox. The instability I have in mind is a function of four things: a subject's route through an environment over time ; his background information (true or false); his disposition at a time to alter judgements ; and the likelihood of his encountering, on his spatiotemporal route, evidence against some true belief he has already formed. T o fix ideas, suppose that someone has arrived by reliable I
'If I Know, Do I Know that I Know?', in A. Stroll, ed., Epistemology ( N e w Y o r k : Harper and Row, 1967), p p 54-82.
means, e.g., on the strength of careful observation, or true, authoritative and sincere testimony, at some true belief or other. At the time and place at which the information is picked up, evidence against its truth is simply not in circulation. The question to ask is this: if the subject subsequently moves to a place where plausible but contrary information is current, or if plausible and contrary information begins to be put about where he remains, what is the status of the subject's true belief? According to me, the true belief was and continues to be knowledge-ranking. But the knowledge is unstable: the subject is explicably susceptible to false information. According to defeasibilists, the true belief comes to be undermined by local counter-evidence and so assumes the status of nonknowledge. I t is a short step from that defeasibilist verdict to Harman's paradox. But the verdict makes it a sufficient condition of non-knowledge that the subject is unable to see through the counter-evidence. According to the view I wish to adopt, it does not count against a man's title to know that he lacks that epistemic X-ray vision. T o distinguish instability from the kindred phenomenon of nonknowledge I need, first, some examples, and later (Sections I11 and IV), principles which explain the intuitions the examples engage. I proceed to construct two cases. T h e first is supposed to illustrate instability, the second non-knowledge. Consider someone who once learnt from an impeccable source that Albany is the capital of New York State. Our man learnt that fact, let's say, in a community in which contrary information was and still is non-existent. Specifically, the community divides up as follows. Some believe that Albany is the capital; the rest profess ignorance. Far away in Outer Mongolia, the textbooks say that in the north-east of America the following convention holds: capital cities of states are always the largest cities in those states. For illustration the textbook cites Boston as the capital of Massachusetts and New York City as the capital of the eponymous state. I t happens that the man with the belief about Albany gets posted to Outer Mongolia. He does not know, when he gets there, that the inhabitants have beliefs different from his own about the political geography of America. But let us add that, if informed of the Mongolian theory about state capitals, he would find the theory plausible and even revise his present belief about Albany. Suppose he is on his way to a party where people are often going to say, and give their reason for saying, that New York City is the capital of
sf)z
TOM SORELL:
New York State. Is our man's justification, hence his knowledge, undermined? My intuitions tell against this conclusion. I am inclined to say that until he actually changes his mind at the party, he knows that Albany is the capital. Our man knows, but at the point at which the party conversation turns geographical, he is on the verge either of suspending judgement, or of believing the falsehood the Mongolians believe. Here is knowledge up to its trick of instability. Now for a variation on the example which arguably sustains a verdict of non-knowledge. We are to imagine a Mongolian schoolboy about to learn about American geography. Before he has looked at his textbook, and in ignorance of what anyone else in Mongolia believes in this connection, our schoolboy happens upon the newlyposted Westerner and asks to be told something about the subject he will soon be taught in school. The schoolboy knows little or nothing about America, and practically anything he is told about its political geography will seem plausible to him. Predictably, one of the things he is told, and believes, is that Albany is the capital of New York State. We are to suppose that in getting over that (correct) information, the Westerner both speaks sincerely and intends faithfully to reproduce what he himself was taught. Granted that the Mongolian schoolboy recognizes the sincerity and the intention, does he acquire knowledge on the strength of the Westerner's testimony? Intuition does not speak clearly. On the one hand, given the truth of the information, and the fact that the schoolboy's trust is well-placed, there is reason to say that the schoolboy knows if his informant does. On the other hand, the schoolboy merely happens upon the Westerner, is unable to tell good from bad information concerning America's political geography, and lives in a place where information contrary to the Westerner's is universally believed and taught. Those facts make it no less than good luck that what he ends up believing about New York State is a truth. On the principles that knowledge excludes lucky arrival at truth, the Mongolian's cognitive condition seems to be one of non-knowledge. Further pairs of cases are easily contrived for perceptual knowledge, specifically, identificatory knowledge. I n an example discussed by Harman, a library guard believes correctly that a certain Tom stole a library book. But what if, to change Harman's details, Tom's twin brother Buck had been stealing a book at about the time and close to the place of Tom's thievery? Both thefts, let's
say, take place on the route the library guard follows as he makes his usual rounds. That stipulation makes it plausible that if the library guard had not noticed Tom stealing a book he would have encountered Buck up to the same thing. But being acquainted only with Tom, not knowing that Tom had a twin, and not being able, much less primed, to tell the two apart, the library guard would still have identified the second thief as Tom. So it appears to be good luck that when the library guard actually makes the identification it is Tom he has latched on to. And again, by the noaccident principle, we seem to have a case of non-knowledge. For an illustration of instability we need only slightly alter the example. As before, Tom steals the book. T h e library guard observes as much and forms his correct identificatory belief. But this time Buck is nowhere near the scene. So it should be uncontroversial that the library guard knows the identity of the thief. Later on, the library guard hears that someone very like the suspected Tom has just been caught stealing books from a local bookshop. This news might start a doubt in the library guard as to whether it was Tom who stole the book in the library: maybe the other, similar man was responsible for both thefts. As in the Albany case, I take the subject's cognitive condition to be one of knowledge until the doubt actually takes effect. I would go further to claim that if the library guard is not given to doubting his judgment, or if he is certain he is right in the case of Tom, so that he disregards the fresh information, then he goes on knowing. His knowledge, moreover, cannot be stigmatized as unstable, since, ex hypothesi, he is immoveable once his mind is made up.
Someone who holds, as I do, that cases of unstable knowledge resist assimilation to cases of non-knowledge, will be unpersuaded by the defeasibilist multiplication of examples of non-knowledge, and will be unconvinced, in particular, by Harman's handling of the cases reviewed in Chapter 9 of Thought. All three of those cases, in each of their versions, will strike one as illustrations of the possession of knowledge. Now Harman anticipates such disagreement, and takes steps to mitigate it. He first claims, very plausibly, that we are more inclined to say that there is knowledge in the examples where there is no undermining evidence a person
s64
TOM SORELL:
does not possess than in the examples where there is such evidence.1 He goes on reasonably to demand that this fact be explained by those who dissent from his treatment. But surely the unclarity of intuitions over undermining evidence also requires explanation. Why is it that, in the face of considerations of defeated justification, verdicts about knowledge become uncertain? One possibility is that the verdicts are sensitive to what examples lay down as the actual influences on a certain belief: how it is initiated and how it is sustained. Defeasibilist examples of nonknowledge trade on the would-be effects of further evidence on a belief someone holds apart from the influence of that evidence. T o get those cases to engage with our-native conception of knowledge, we have to imagine the would-be effects actualized. We have to imagine someone's changing his mind or suspending judgment as a result of acquiring counter-evidence. What we have then conjured up are circumstances in which a subject abandons a true judgment, and this compels a verdict of non-knowledge. Now the reason we are more inclined, as Harman justly observes, to turn in that verdict where misleading evidence is present than where it is not, is that, in the latter sort of case, there is no likely projection of an actual course of events that culminates in the subject's rejection of a formerly accepted truth. This partly answers Harman's call for explanation. But we do not yet have a complete account. For surely our inclination to withhold an attribution of knowledge in the disputed cases has something to do with the thought that the subject's would-be change of mind is, objectively speaking, gratuitous. There is no basis in reality for the subject's reconsideration of what he believes : at best there is a subjective basis, e.g., the incompleteness or partial falsity of the subject's background beliefs. These considerations crystallize as a strategic problem for analysts of factual knowledge: will analysis enshrine the objective stance and bear out the impression of gratuitousness, or will analysis be geared to the viewpoint of the subject, a viewpoint from which misleading evidence cannot always be recognized for what it is? An objectivist settlement of the question seems to overdraw requirements of knowledge. T o repeat, it is as if the subject is taken in by evidence he ought to have been able to see through. That
seems unconvincing : unless the subject's background information happens to suffice for his explaining away the new evidence, talk of his being taken in, or of the gratuitousness of a change of mind, loses its purchase. T h e subject's background information need not suffice : that is the plight of creatures with less than comprehensive, and possibly idiosyncratic experience, of even a familiar portion of their environment. Partly for that reason, I am drawn to the alternative of reckoning states of knowledge as information states among others, sensitive, like the others, to alterations in the subject's environment and gaps in the subject's experience. Like the beliefs at large, the non-accidentally true beliefs, i.e., the knowledgeranking ones, can explicably give way to false beliefs. Not so according to the objectivist account, which demands of states of knowledge a puzzling imperviousness to misleading evidence. T o make sense of the imperviousness, the objectivist must construct ad hoc a philosophical psychology of knowledge distinct from that appropriate to mere true belief. There is ancient precedent for positing a special rational faculty (the understanding) that is an evidence-independent source of true thoughts. Alternatively, it is open to the objectivist to say that all thoughts are sensitive to evidence, including misleading evidence, and, by that token, none is knowledge-ranking. But this is to demote to non-knowledge any true belief against which there so much as could be misleading evidence. What compels the adoption of one or another of these unappealing strategies, is the assumption that the falsehood of a certain kind of counterfactual is logically sufficient for the falsehood of a knowledge sentence. In the Albany case, it is not true that were the newcomer to be apprised of what the Mongolians falsely believe, he would stick to his true belief: it is supposed to follow logically that the true belief about Albany was not knowledge-ranking in the first place. Symmetrically, if it is true that the man knows that Albany is the capital, it is supposed to follow logically that were he to hear any story to the contrary, it would not change his mind. T o make out these consequences as logical ones, the objectivist has got to make it constitutive of knowing something, that the knowing be evidence-insensitive. Anything short of this would permit the truth-values of the knowledge sentences to be logically independent of the truth-values of the counterfactuals. And that would suffice for the instability of knowledge. The trouble is that, unless instability is permitted, an objectivist account of knowledge, and so a
566
TOM SORELL:
defeasibilist one, ends up at variance with the actualism of the intuitions about knowledge the account was supposed to agree with. Having counted the Albany case as one of non-knowledge in virtue of the presence of defeating evidence, the objectivist has got to reckon with the question of how a would-be change of mind, even a warranted one, can actually defeat what the example lays down as an impeccably generated thought that Albany is the capital. The reason the question crops up is that untutored assent to knowledge sentences simply does not force assent to the relevant counterfactuals, no more than untutored dissent from the truth of a knowledge sentence is forced by dissent from the relevant counterfactuals.
If defeasibilism fits a conception of knowledge that is not narrowly actualist, one can see why defeasibilist verdicts diverge from, and why defeasibilist examples fail systematically to engage, our prephilosophical dispositions to attribute knowledge and nonknowledge. There is a different explanation of one's ambivalence in the face of the two cases I earlier cited as arguable illustrations of nonknowledge. Concerning the schoolboy who picks up the correct information about New York State's capital, one is inclined to say that the schoolboy knows if the informant does. Again, when the library guard latches on to Tom in the act of theft, a plausible judgement is that the guard knows the identity of the thief-notwithstanding the presence nearby on the guard's route of a thief the guard would have seen and misidentified as Tom. Both of these verdicts seem the right ones if the examples are taken to have just the following feature in common : in each the subject merely would have formed a false belief on the strength of misleading evidence. As things actually stand, however, no false belief is formed, and no misleading evidence operates in the formation of what are actually true beliefs. Until misleading evidence is an actual influence, each case illustrates knowledge. I t looks as if it would take a counterfactualist principle about knowledge, a principle once more at variance with actualist intuitions, to overturn those considerations. But reflection shows that this is not so. There is a further parallel between the examples. Once that is seen a verdict in each case of non-knowledge seems
compelling on actualist grounds. T o get at the further parallel one has to press a pair of questions about each case. First, why does each subject believe what he does? Second, in virtue of what is each belief true? If the answer to the first question invokes nothing from the answer to the second, i.e., if the causal explanation of the formation of the belief makes no essential reference to objects or features of objects which make the belief true, then those objects and features do not contribute (causally) to the generation of the belief. In that case, the belief, though true, is accidentally true, and so not a case of knowledge. According to the details of the first of our two examples, a schoolboy ends up with a true belief about the political geography of New York State. He believes what he does on the strength of what is in fact impeccable testimony, testimony conveyed by someone who seems sincere and in the know. The informant's apparent sincerity and authoritativeness will enter into the explanation of the story's being believed by the schoolboy. But the content of the story is not essential to the explanation of its acceptance : it is written into the example that, on account of the schoolboy's lack of political background information, most testimonies are equally plausible to him. So it is not essential to the schoolboy's believing what he is told that it is that story he is told. A fortiori, it is not that story's being true, and having the source in the facts that it has, that accounts for the schoolboy's believing it. In other words, the schoolboy does not believe the story because of what makes it true. That is why his belief is accidentally true. For analogous reasons the library guard's belief is true by accident. What prompts his identificatory belief is his latching on to someone with a certain familiar appearance, someone in the act of theft. The appearance is the appearance of someone who in fact fits the identificatory belief. I n other words, the belief is to the effect that Tom is the thief, and it is of Tom that the library guard has the visual experience that generates the identification. But it is not because the visual experience has just that source that the identificatory belief is produced. Visual experiences with sources sufficiently similar to Tom could generate identifications actually generated by Tom himself. And in the example one such similar source (Tom's twin) is actually positioned in a place the library guard customarily moves through and surveys. So Tom does not make all the difference to the identificationgenerating experience. His being the thief is not essential to the explanation of the identification's being made, only to the truth
568 TOM SORELL: of the identification. Once more there is a rift between what makes for truth and what explains the generation of the truth-value bearing belief. Depending on what is held to be the salient parallel between the cases, our intuitions pull us in opposing directions. At first we are inclined to attribute knowledge to each subject, and then, given the new parallel, to withhold the attribution. Which, if either parallel, is the right one for an analysis of knowledge to exploit? Which parallel is likely to bring out the system in our dispositions to assent to, and dissent from, knowledge sentences in actual and invented cases? I suggest that the second is the more promising. If the first parallel were the telling one, i.e., if .all the examples had in common was the feature that the subjects do not actually succumb to misleading evidence, then only a negative principle could be drawn from the examples. The principle would be that only actual influences on belief weigh in a reckoning of whether a belief, if true, is knowledge ranking. T h e principle helps to exclude counterfactualist excesses, but it leaves us wondering, concerning the actual influences, what conditions they must satisfy to make a knowledge sentence true. When the examples are taken to parallel one another in the second way, i.e. in respect of there being nothing in common between the explanation of the formation of each belief and what makes each belief true, an actualist principle is again suggested, but this time an informative one: one knows only if that in virtue of which one's belief is true makes an essential (non-redundant) contribution to the causal explanation of one's having that belief. I say that the latter principle is more likely than the other to disclose the system in our dispositions to assent to, and dissent from, knowledge sentences. This means that the principle must hold generally, must apply to a range of cases inclusive of the pair I have been at pains to dissect. That the principle does have general application is shown by its power of unifying relatively recherche' cases of non-knowledge, such as the pair minutely examined, with paradigm cases, cases of guessing correctly, having a correct hunch, having a fulfilled premonition, and (differently) conjecturing aright. What unifies these types of thought is that their aetiology either includes no evidence for their truth (guessing, premonition, hunch) or too general evidence, evidence subsumable by many hypotheses. In the most vivid case (guessing), someone's having the thought one has is compatible with any state of the objects that endow the thought with a content (truth conditions). That is why, if the
thought expresses the actual state of things, it is good luck : because the objects are inoperative in the thought's capturing their state. Here is the preferred principle at work in familiar cases. Now defeasibilism, though it purports to unify Gettier's cases of non-knowledge with seemingly genuine and unsuspected further cases of the same thing, does not clearly unify those cases with the humdrum paradigms just mentioned. This is because defeasibilism contrives a new notion of adequate justification (complete justification) where an old notion (pre-Gettier adequate justification) served to explain why guessing and premonition were not the stuff of knowledge. Defeasibilism does not explain why the old justified true belief analysis worked so well before Gettier's paradox was discovered. I n other words, defeasibilism does not explain why a novel notion of complete justification is required for all cases of non-knowledge and not just for Gettier examples. No such complaint attaches to the preferred causal explanatory principle. T h e preferred principle uniformly explains old and new cases, explains our actualist bias, and illuminates thereby the failure of intuition to speak clearly to cases of defeating evidence. Finally, and to revive my main theme, the preferred principle blocks the assimilation of the instability of knowledge to non-knowledge.
I said earlier that unstable knowing was a function of four things: ( I ) a subject's route through an environment over time; (2) his background information; (3) his disposition at a time to alter judgements, and (4) the likelihood of his encountering, on his spatiotemporal route, misleading evidence against the truth of some antecedently formed true belief. Reflection on condition (4) suggests that conditions of instability are irremediably vague. Whether misleading evidence is encountered is a matter of, inter alia, the subject's attentiveness and the proximity of sources of misinformation. Take proximity: misleading information must be present along or near a route that the subject follows. But there is no saying once and for all how close is close enough to make a difference. Misleading evidence in Outer Mongolia against the truth of a belief formed by someone who never leaves his village in Suffolk, is not likely to matter. But there are indefinitely many intermediate cases. Analogous considerations hold for condition (3). This does not mean that the distinction between knowledge, including unstable
TOM SORELL: 570 knowledge, and non-knowledge is never sharp, only that it is not always sharp. Harman's account shows that there is scope for collapsing the distinction where it need not be collapsed, and that is at the source of the impression of paradox. When Harman tries to dispel that impression he invokes considerations which resemble, but actually conflict with, my own. Commenting on his own statement of the paradox, he writes:
The argument for paradox overlooks the way actually having evidence can make a difference. Since I now know that Tom stole the book, I now know that any evidence that appears to indicate something else is misleading. That does not warrant me in simply disregarding any further evidence, since getting that further evidence can change what I know. In particular, after I get such further evidence I may no longer know that it is misleading.1 Here the first sentence is an understatement and the rest false. According to me, the subject cannot explicably know that any counter-evidence is misleading. This obviates the awkward claim that 'after I get such- evidence I may no longer know that it is misleading'. How can evidence of a sort I know in advance to be misleading, produce ignorance of its misleadingness upon being acquired? Once it is conceded that one can know and confront further evidence unprepared for misleadingness, a plausible description of the effects of acquiring the evidence follows easily : the effects are likely to be either my believing a falsehood or my not knowing what to believe. I t is those effects that deprive me of my former knowledge of the truth. If that is right, my getting the evidence makes all the difference: the mere availability of the evidence leaves my epistemic status unaltered. This way of disposing of the paradox, however, seems also to tell against the defeasibilist analysis of knowledge. Harman is drawn to certain verdicts on examples generated by that analysis. But he wants to dismiss the principle which explains the verdicts. My point is that the two stand or fall together.
At the beginning I claimed that the impression of paradox is encouraged by a dogma in epistemology, a dogma I formulated as the I
Ibid. pp. 148-149.
thesis of the stability of knowledge. Behind the thesis is some venerable, potent, and misleading philosophy which results in the view that, as a matter of analytic truth, to have knowledge is to be in a state that is proof against changes of mind. Philosophers have sought to frame conditions of knowledge which outlaw both external and internal causes of change of mind. Plato outlaws external causes when he stipulates that knowledge can be knowledge only of what is unchanging (Philebus ggb). Is there anything in that? T o see that there is, one has only to consider that knowledge carries with it an apriori requirement of truth, hence truth-value, in things known. The question of what to count as objects of knowledge is thus related to the question of what to count as bearers of truth-value. As a truth-vehicle the ordinary declarative sentence will not do : 'it is true on one occasion and false on another, because of the tenses of its verbs and the varying references of its pronouns or demonstrative adverbs or other indicator words.'l For fixity of truth-value eternal sentences (in Quine's sense) are wanted, and for objects of knowledge those eternal sentences that are true. Or perhaps as good, since individuatable eternal sentences are relative to unindividuatable languages, the truth-value bearers can be regarded as the eternal propositions, the truths among which will fix the range of the knowable. Knowledge sentences embedding sentences other than eternal truths could not, in view of the a priori requirement, be construed as speaking strictly of knowledge.2 By courtesy, they could be made out as concerned with quasi-knowledge, knowledge a cut below the real thing because it is knowledge of truth varying with a time, a place, a subject of knowledge and the like. But this would still yield the conclusion that full-blooded knowledge is of truths true for good. As it is put in the Philebus, 'reason . . ., and knowledge that gives perfect truth are foreign to (things devoid of permanence)'. Knowledge, if confined to eternal truths, could not clearly become non-knowledge in virtue of a change in nature. So knowledge is, so to speak, safe from external sources of instability. But in Plato knowledge is not just a grasp of the fixedly true, but a fixed or 'permanent grasp' of the fixedly true. Internal (intellectual) I
2
Quine, 'Propositional Objects', in Ontolo.qica1 Relativity and Other Essays (New York: Columbia University Press, 1969), p. 139. I t is a separate problem that the mode of containment of the embedding statements may itself be indexical, so that knowledge sentences embedding even eternal sentences have fluctuating truth-value (cf. Davidson on parataxis).
572
TOM SORELL:
sources of instability are as much in tension with the very idea of knowledge as external sources. T h e reasons come from Plato's moral epistemology, according to which it is part of the very idea of knowledge that it is the kind of thing responsible for virtuous action, i.e. that virtue is knowledge. I t is part of the very idea of knowledge, too, that it be teachable. So if the cause of virtuous actions is virtue, and virtue knowledge, it must be possible to teach virtue. But what would create conditions for the teaching of virtue? Not just the existence of virtuous men: such men can try and fail to instil virtue, and even doubt whether it can be instilled (Meno 96b). Nor can virtue be teachable just in virtue of the existence of men who profess to teach it; for it can be a question whether such men understand the subject, and hence also a question whether such men do right to profess to teach it. But if a virtuous man, one in fact motivated by virtue, were to work out for himself what motivated him, his articulate understanding of what virtue was would make him a model for others (Meno ~ o o a ) and , so would make possible moral training. I t is the self-knowledge of the virtuous man that makes virtue teachable. But is that self-knowledge necessary for virtue? If not, then virtue comes apart from knowledge. And it does at any rate appear that right action is compatible with something less than knowledge, on the agent's part, of why the action is right. The true opinion, i.e., the true belief without reasons, that an action was right, might produce the same action as would be produced by a rationally grounded true belief that an action was right (Meno g6b-d). But then the acts are acts of virtue in the way that an oracle's pronouncements are pronouncements of truths : without the agent's consciousness of their virtue. Just as the truth of the oracle's sayings cannot be accounted to him, but only to a god that speaks through him, so unwittingly virtuous actions cannot be ascribed to the ostensible agent, but only to a divine will for which the ostensible agent is the instrument (Meno ggc). The actions of the man who is not knowingly propelled by virtue are not then his. For the agent to act in propriapersona, he must (Plato suggests by means of recollection) come to an understanding of what motivates him, and the effect of that is to elevate his actions to ones that are done because they are virtuous. Until that happens, it is luck relative to the agent's point of view-a matter of well-aimed conjecture (ggc) -that what he does is regularly the right thing: he acts well, but does not know why. No wonder, then, that he cannot train others
to virtue (Meno ggb). No wonder either that, albeit unwittingly, he connives in the frustration of his own virtuous desire to make others good (Meno g3c-d). That last is a crucial point. I t shows that mere true opinion about what is right cannot guide at least one virtuous undertaking (moral training), and that, by the same token, mere true opinion is compatible with an avoidable wrong of omission (the omission of moral training). That is why virtue and full-blown knowledge cannot come apart: less than conscious virtue permits avoidable wrongdoing. Put another way, less than conscious virtue, though compatible with right actions, is not compatible with a life of virtue, a life that leaves virtue behind it. I take Plato to have this in mind in the Meno, where he explicitly demands stability in the source of right action, comparing the state of the unwittingly virtuous (the man with mere true opinion) to the statues of the gods that Daedalus cunningly endowed with arms and legs (g7e-g8a) : If you have one of (Daedalus's) works untethered, it is not worth much: it gives you the slip, like a runaway slave. But a tethered specimen is very valuable, for they are magnificent creations. And that, I may say, has a bearing on the matter of true opinions. True opinions are a fine thing and do all sorts of good as long as they stay in their place; but they will not stay long. They run away from a man's mind, so they are not worth much until you tether them by working out the reason. That process . . . is recollection. . . . Once they are tied down, they become knowledge and are stable. That is why knowledge is something more valuable than right opinion. What distinguishes one from the other is the tether. Plato's views are powerful. Are they right? Is it true that a man's unconsciously virtuous actions are not his, and are not, until he knows why he acts, reliably virtuous? Or again, and this time with respect to external causes of instability, is it true that there cannot be knowledge of a sentence with unstable truth-value? Why cannot there be knowledge (in the strict sense), at a time and place, of the truth of a sentence that is (absolutely) true then and there? And why not, in a state of pre-recollective virtue, the possibility of always knowing that, even if not why, one's actions are right? Why would not this suffice for the reliability or control supposed by Plato to be available only post-recollectively? Why would not the same suffice for action in propria persona? In each case it can be
574
SORELL: doubted whether the tension in unstable knowledge is after all genuine. A way of making the tension more compelling is by recasting along Cartesian lines the distinction between states which are intelligibly unstable and those which are at war with instability. What results is the difference between persuasio, an episode of irresistibly believing a discrete truth at the time one carefully considers it, and scientia, the state which ensures the timeless, attentiveness-independent certainty of anything that can momentarily compel my assent and yet become dubitable retrospectively, or become dubitable when considered not by acquaintance but by description. Scientia requires, what persuasio does not, appreciation of the system of truths in which an object of persuasio fits. Descartes's problem is to explain how we can aspire to anything more than the fragmentary and momentary true beliefs constituted by persuasio, given facts about the human condition and (as Descartes thought) facts about the world. Human beings are naturally prone to faulty memory and inattentiveness. Differently, and now with the aid of method, they can induce in themselves a near global doubt. These facts create conditions for both gratuitous and rationally-motivated abandonment of true beliefs. Again, objects in the external world are such that, without a cause to sustain them, they can pass out of existence from moment to moment. So the truth-values of any beliefs we managed to hang on to concerning the external world might, despite the hanging on, be subject to massive instability of truth-value. Persuasio is possible in such a setting, but not a systematic grasp of truth-not scienceunless (i) God exists to sustain other existences; and (ii) we know that He exists. (i) ensures a stable and systematic reality; (ii) ensures a stable and systematic knowledge of reality. I say that this set-up makes the thesis of the stability of knowledge ( = scientia) compelling, because there is a tension between the possibility of the build-up of a body of systematic truths (the object of scientia), and the possibility, at any moment, of either the destruction of all material objects, or the loss of one's grip on discrete truths about them through doubt, inattentiveness or forgetfulness. There is something genuinely incoherent about unstable scientia, but not about unstable knowledge of discrete truths: Descartes's persuasio fits perfectly well a pre-philosophical conception of knowing something. What it does not fit is any conception of science. So I am with Descartes when he says (HR 11, 39): TOM
That an atheist can know clearly that three angles of a triangle are equal to two right angles I do not deny. I merely affirm that, on the other hand, such knowledge on his part cannot constitute true science, because no knowledge that can be rendered doubtful should be called science. Descartes does not say that no knowledge that can be rendered doubtful should be called knowledge. In our own day, a philosophical belief in the stability of knowledge is likely to accrue from considerations which support the Hintikka iteration principle. But the belief need not hinge on this alone. I t can result from another line of thought, concerning consistency, which can be put like this. 'If I know something, then I know a truth ; if I find out another truth, then it should be consistent with what I found out previously, for all I am doing in acquiring knowledge is discovering further features of the world which coexist with one another. How can finding out something new, then, fail to square with, or put into peril, the truths I have already discovered? How could my old knowledge be unhinged by my new knowledge?' Certainly there is something right in all of this: whatever I find out to be true will, barring situations in which what I knew previously was a transient feature of the world, be consistent with my previous knowledge. Rut not everything that is newly found out is a truth, not every new piece of information is correct information, and states of non-accidentally true belief, viz., states of knowledge, are sensitive to new information true or false. T H E OPEN UNIVERSITY