On the Gettier Problem problem

William G. Lycan

It took about ten years for people to get the idea that there was something wrong with the Gettier Problem. By the early 1970s, a number of analyses had been offered to accommodate Gettier’s (1963) counterexamples to the traditional ‘JTB’ view: Michael Clark’s (1963) simple no-false-lemmas proposal, various ‘indefeasibility’ analyses beginning with Lehrer (1965) and Lehrer and Paxson (1969), and Goldman’s (1967) original causal theory, among others. Those analyses had run into further counterexamples; revision after revision had been offered, only to meet further and more elaborate counterexamples. Not only was there no end in sight; there was not even a sense of beginning to converge.

In itself, that was hardly an unusual situation in philosophy. We might expect the optimism of the most enthusiastic practitioners to have been attended by merely the normal degree of professional pessimism. But, no: the Gettier Problem was not doing so well as that; it had begun to get some bad press.

I The Gettier Problem problem

Some epistemologists wrote pointedly larger and more general works, being careful to play down the Gettier Problem and address it only unemphatically, in subordinate clauses (even though they did not want us to miss the solutions they offered).[1] Informally, the Gettier Problem became a leading focus, if not the focus, of disenchantment with the definition-and-counterexample method of analytic philosophy. In some cases the disenchantment spilled over into scorn; there were slighting references to ‘the ‘S knows that p’ crowd’. That attitude combined expansively with the complaint commonly made, that among analytic philosophers the adversarial method had gotten out of hand and that people had begun flinging elaborate counterexamples only to be clever and to score points, with no thought for the larger picture or for positive understanding. (Another popular sneer of the period was, ‘Why don’t you go publish a little note in Analysis?’) Above all, it was suggested that the Gettier project was unfruitful, idle, pointless, almost antiphilosophical.[2]

One feature of the postGettier analyses thought to show their fecklessness was their ungainly, sometimes meandering complexity. For example:

S knows that h iff (i) h is true, (ii) S is justified [by some evidence e] in believing h…, (iii) S believes that h on the basis of his justification and…(ivg)…there is an evidence-restricted alternative Fs* to S’s epistemic framework Fs such that (i) ‘S is justified in believing that h’ is epistemically derivable from the other members of the evidence component of Fs* and (ii) there is some subset of members of the evidence component of Fs* such that (a) the members of this subset are also members of the evidence component of Fs and (b) ‘S is justified in believing that h’ is epistemically derivable from the members of this subset. [Where Fs* is an ‘evidence-restricted alternative’ to Fs iff (i) For every true proposition q such that ‘S is justified in believing not-q’ is a member of the evidence component of Fs, ‘S is justified in believing q’ is a member of the evidence component of Fs*, (ii) for some subset C of members of Fs such that C is maximally consistent epistemically with the members generated in (i), every member of C is a member of Fs*, and (iii) no other propositions are members of Fs* except those that are implied epistemically by the members generated in (i) and (ii).][3]

Faced with such a monster, we may be unable to think of a further counterexample, but that inability is as well or better explained by the very convoluteness of the analysis than by its correctness.

Yet no similar opprobrium attached to other, ostensibly similar chisholming projects -- for example, the Gricean analysis of speaker-meaning, the analysis of social convention inaugurated by David Lewis, the Kripkean causal-historical theory of linguistic referring, the search for criteria of personal identity through time, or the counterfactual theory of causality. Why not? Perhaps the difference was an historical accident of timing or of personality. Or perhaps it was just that those enterprises have never led to the breathtaking complexities that ‘S knows that P’ did.[4] It is well to remind ourselves that no effort of analytic philosophy to provide strictly necessary and sufficient conditions for a philosophically interesting concept has ever succeeded. And there should be a lesson in that. Yet the Gettier project still seems to have outstripped all others in the extent of its failure; why did the other analytic projects never reach the extremes of futile complexity that the Gettier industry did? More deeply, is there something wrong with the Gettier project? Does it rest on some false presupposition?

What I shall call the ‘Gettier Problem problem’ is that of explaining what is distinctively wrong with the Gettier project. There have been a number of attempts over the years. My purpose in this paper is to survey and evaluate those.

II Uninteresting solutions

Originally the Gettier Problem was cast as the search for the ‘fourth condition’ of knowing, a condition to be added to ‘J’, ‘T’, and ‘B’, to block Gettier’s counterexamples to the sufficiency of ‘JTB’. The search took place during the sunset years of ‘conceptual analysis’, the activity of taking a philosophically interesting notion and trying to find a set of conceptually necessary and sufficient conditions for the notion’s being exemplified. Outrageous and fantastical counterexample scenarios were allowed, of course, because a mere conceptual possibility would be enough to refute a claim of conceptual necessity or sufficiency.

That feature alone would have made some philosophers scorn the Gettier project, because of healthy Quinean skepticism about ‘conceptual truth’ and analyticity.[5] If the very idea of a conceptual truth is infirm, then to seek conceptual truths about knowing is obviously misguided. (And Gettier practitioners did typically style themselves as investigating ‘the concept of’ knowing, even into the 1970s.) But even if one is a Quinean skeptic, there are three reasons why that is no satisfactory solution to the Gettier Problem problem. First, the Quinean complaint applies across the board; it does not reveal anything wrong with the Gettier project that is not wrong with any other ‘strictly conceptual’ quest such as that of trying to analyze causality or linguistic referring, or for that matter trying to define ‘doctor’ or ‘bachelor’ or ‘doe’. Second, one cannot complain, as against (e.g.) the personal identity literature, that the hypothetical cases are all fantastical or wildly science-fictional and ordinary concepts just do not bear that close scrutiny. The counterexamples that figure in the Gettier literature, though usually unlikely, are just as usually not fairy-tale or science-fictional or otherwise merely conceptual, but nomologically possible. Some are actual.[6] Third, by the same token, it is entirely possible by any standard to have JTB without knowing; people actually do that. So for all that any Quinean has shown, it is not only reasonable but very interesting to ask what must be added to mere JTB in order to constitute a case of knowing. Analyticity is not required.

I pause in this section to review a few other uninteresting solutions to the Gettier Problem problem, though in slightly ascending order of interestingness.

Denying ‘J’. Early on, it was occasionally suggested that Gettier’s examples were not counterexamples, because they did not in fact satisfy ‘J’; what they showed is rather that no sheer amount of conventional evidence suffices for complete justification. Though the luckless S’s evidence was strong enough ordinarily to count as knowledge-affording, it was not evidence of the right sort or structure, and so was not fully justifying (Pailthorp 1969).

This is uninteresting because it is verbal. What the examples showed is that no amount or strength of evidence short of entailing evidence would do, unless that justification were also not defective in Gettier’s characteristic way; it does have to have the right structure in addition to its strength. There are two distinct factors, strength and structure. Even if we choose to withhold the verdict ‘J’ until the second factor has been established, that does not tell us how to establish it.

The TB analysis. Sartwell (1991; 1992) notoriously argued that knowledge is merely true belief, i.e., that every true belief counts as a piece of knowledge.[7] If Sartwell is right, then Gettier’s cases cannot be counterexamples, because there can be no counterexamples to ‘JTB’. Not even ordinary ‘J’ is required for knowing.

This is uninteresting (as a solution to the Problem problem) because Sartwell’s position is so radical that if one actually accepts it, the rest of one’s theory of knowledge will have little if anything to do with traditional epistemology. Also and in particular, one will lose the distinction aforementioned, between two ways in which justification can be found wanting (insufficient strength, and Gettier defect); for in what respect is a Gettier victim’s justification found wanting, if not in that it fails to constitute knowledge?[8]

Skepticism. The Gettier Problem presupposes that ordinary ‘J’ is fallible, in that a person can have an epistemically justified but false belief. (Recall that according to the usage of the time, ‘epistemic’ justification is justification that would normally be strong enough to afford knowledge, so long as the subject is not gettiered in some way.) This means that (indeed) ordinary ‘J’ is enough for knowledge when the subject is not gettiered or the like. But any skeptic will tell us that that presupposition is false. Epistemic justification requires the truth of the justified belief; otherwise the conceded evil-demon possibilities and other skeptical scenarios would preclude knowledge. If there is no empirical knowledge at all, we should hardly be surprised that Gettier victims lack it.

This is less uninteresting than the TB diagnosis, because skepticism is believed by some epistemologists and taken very seriously (to say the least) by many more. But it is still comparatively uninteresting. Of course the Gettier Problem arises in the first instance only for those of us who are not skeptics.

Also, as in the case of TB, we lose the distinction between failing to know because our justification is not strong enough and failing to know because we have been gettiered. The Problem does arise for a skeptic in the second instance: Any skeptic should admit that knowing is at least a regulative ideal, and that some cognitive conditions come far closer than others to satisfying it. A subject who has what anyone would consider overwhelmingly strong evidence should be counted as just-about-knowing, or as-good-as-knowing, or knowing-for-all-practical-purposes (compare ‘flat,’ for those who hold that nothing is absolutely flat [9]) even if the skeptic is right and no one ever strictly knows. But this holds only so long as the subject is not gettiered. A Gettier victim does not just-about-know or as-good-as know; a Gettier victim simply does not know.[10] That difference remains to be explained, even for a skeptic.

Nomic reliability. Dretske (1971) and Armstrong (1973) argue that one knows only if, in the circumstances, one could not have the reason one has for one’s belief (Dretske) or be holding the belief itself (Armstrong) unless the belief were true. This is not to say that one’s evidence must be entailing. Rather, it is about the natural relation one bears to the relevant chunk of one’s environment: The relation between one’s having that reason or holding the belief is nomic. One cannot be mistaken; by law of nature, the only way in which one could now have that reason or hold the belief is for the belief to be true. If that nomic requirement is satisfied, one cannot be gettiered, because in any Gettier case there is an element of luck or fluke in S’s being right. This explanation is like the skeptic’s in that it preëmpts Gettier by denying the knower a kind of fallibility that Gettier requires, but it does not entail skepticism, and neither Dretske nor Armstrong is a skeptic.

This is by far the least uninteresting of the comparatively uninteresting solutions, because it was independently motivated and also because it (demonstrably) opened up a positive research program. Reliabilism dominated the field for years, both as a theory of knowing and as a theory of justification more generally. But as a solution to the Gettier Problem problem, it is still comparatively uninteresting, for two related reasons. First, though it does not entail skepticism, it militates for great stinginess in knowledge ascription; no one would suppose that the nomic requirement is met in any but the most felicitous of circumstances (Lycan 1984).How often does it happen that even if my eyes and brain are working normally and atmospheric conditions are good, it would be nomically impossible for me to be in my current perceptual-cum-cognitive state and still be fooled? -- much less the other homey sorts of things we take ourselves to know, such as our own names, what we ate an hour or two ago, who chairs our department, etc. If nomic Reliabilism is true, we know hardly anything.

Second, it is now generally conceded by Reliabilists themselves that the Dretske-Armstrong nomic formulation was too strong in more specific ways (e.g., Pappas and Swain 1973). Subsequent versions have variously weakened that formulation (e.g., Goldman 1976). The weakenings burden their respective authors with the need to block Getter cases, and so the Gettier Problem returns for them.

III Interlude: a simple analysis

Before I proceed to the more interesting and viable suggestions that have been offered as solutions to the Gettier Problem problem, I should confess that I have my own favorite solution to the original Problem. (All right, I am announcing that I have one, and with at least a bit of pride.) For a reason that will emerge, I can guarantee that my analysis will convince almost no one; and on inductive grounds I can predict with complete confidence that someone will find a clear counterexample. But here it is.

Start with Clark’s no-false-lemmas proposal, which was immediately suggested by Gettier’s own two cases: S’s belief must not rest upon any false grounds; in particular, S’s reasoning must not pass through any false step. This was of course instantly counterexampled. (Space does not allow detailed description of all the cases; I shall assume some familiarity with them.)

Against the sufficiency of ‘no-false-lemmas’:

Noninferential Nogot (Lehrer 1965; 1970). Mr. Nogot in S’s office has given S evidence that he, Nogot, owns a Ford. By a single probabilistic inference, S moves directly (without passing through ‘Nogot owns a Ford’) to the conclusion that someone in S’s office owns a Ford. (As in any such example, Mr. Nogot does not own a Ford, but S’s belief happens to be true because Mr. Havit owns one.)

Cautious Nogot (Lehrer 1974; sometimes called ‘Clever Reasoner’). This is like the previous example, except that here S, not caring at all who it might be that owns the Ford [11] and also being cautious in matters doxastic, deliberately refrains from forming the belief that Nogot owns it.

Testimony Nogot (Saunders and Champawat 1964). S’s evidence is all hearsay, but very reliable hearsay. S is told overwhelming evidence that Nogot owns a Ford. S’s grounds are all true: S was (indeed) told all those things, and by a highly reliable informant.

Existential Nogot (Feldman 1974). S does not acquire the evidence itself, but only the existential generalization of it: ‘There is someone in the office of whom it’s true that….’ S has no idea who that person, the protagonist of the evidence, is. But from that existential generalization, S justifiably infers the generalization ‘Someone in the office owns a Ford.’

Stopped Clock (Scheffler 1965, following Russell 1948). S looks at a clock and forms a true belief as to the time of day. S has every reason to believe that the clock is working well, but in fact it has stopped.

Sheep in the Field (Chisholm 1966). Looking into a field, S sees an animal only a few yards off that looks, sounds, smells, etc., exactly like a sheep, and S noninferentially forms the perceptual belief that there is a sheep in the field. Actually the animal is of a different species but has been artfully disguised. Yet there is a real sheep in the field -- ‘way off in a remote corner of the field, completely hidden behind thick hedges.

Sure-Fire Match (Skyrms 1967). S strikes a dry Sure-Fire match and is epistemically justified in thinking it will light. Actually the match has an incredibly rare impurity and could not possibly be lit by friction, but it lights anyway because of a freakish burst of Q-radiation from the sky.

And there is the obvious sort of counterexample to the necessity of ‘no-false-lemmas’ (Saunders and Champawat 1964; Lehrer 1965). Nondefective Chain: If S has at least one epistemically justifying and non-Gettier-defective line of justification, then S knows even if S has other justifying grounds that contain Gettier gaps. For example (Lehrer), suppose S has overwhelming evidence that Nogot owns a Ford and also overwhelming evidence that Havit owns one. S then knows that someone in the office owns a Ford, because S knows that Havit does and performs existential generalization; it does not matter that one of S’s grounds (S’s belief that Nogot owns a Ford) is false.(It does not matter if S has fifty or a thousand other gettiered justifications.)

Further: Togethersmith (Rozeboom 1967). On a Sunday afternoon, S (‘Mrs. Jones’) sees the Togethersmith family car leaving the driveway, and S knows that every Sunday afternoon all the Togethersmiths go for a drive in the country. Because S believes that all the Togethersmiths are in the car today as well, S concludes that Mrs. Togethersmith is not at home, and S is right. But it is false that all the Togethersmiths are in the car today; one of the children is attending a friend’s birthday party.

What the counterexamples to the sufficiency of ‘no-false-lemmas’ have in common, of course, is that in them S does not engage in a process of reasoning that passes through the relevant false step (‘Nogot owns a Ford,’ ‘That clock is working well,’ etc.). Gilbert Harman (1973) argues that the no-false-lemmas strategy should be maintained in the face of such examples, indeed should be aufgehoben into a methodology for investigating the structure of epistemic justification (indeed, the nature of inference itself). He makes a preliminary case for his principle P: ‘Reasoning that essentially involves false conclusions, intermediate or final, cannot give one knowledge’ (47). Then, rather than entertain putative counterexamples to P, he retains P and looks to see what epistemological consequences ensue. A first one is that justification does not proceed by purely probabilistic rules of acceptance, since such rules do not rely on intermediate conclusions at all (120-4). Further appeals to P encourage Harman to posit nonconscious mediating inferences where we might otherwise see none. E.g., he says, a gettiered subject makes tacit inferences concerning causal connections and other explanatory relations, and the falsity of those tacit grounds explains, via P, why the subject fails to know despite being epistemically justified.[12] Harman uses P in this leveraging way to motivate his general idea that all inference is or involves ‘inference to the best explanatory statement’ (ch. 8).

Now, as we saw, our counterexamples to sufficiency are cases in which there does not seem to be reasoning that passes through a false step. Harman’s strategy would be to hypothesize that there is such reasoning nonetheless. I take a different line: What seems more obvious and less potentially controversial is that in each of the counterexample cases, S tacitly believes or assumes something false. This is a weaker notion than that of an unconscious inference that occurrently passes through a false step, for it does not require any occurrent assumption or inference, even an unconscious one. For example, in Noninferential Nogot, we can concede to Lehrer that S does not engage in a reasoning process that passes through ‘Nogot owns a Ford,’ but clearly S does tacitly assume that Nogot owns a Ford -- else why on earth would S form the belief that someone in the office owns one?

Similar remarks hold for the other counterexamples to sufficiency. Perhaps Testimony Nogot and Existential Nogot are less obvious than the rest, but each is fairly obvious: In Testimony Nogot, S falsely assumes that the person S’s informant is talking about owns a Ford. In Existential Nogot, S falsely assumes that the protagonist of the evidence owns one. I propose, then, that (for now) the no-false-lemmas analysis be replaced by the weaker no-false-assumptions theory.

What went wrong? That is, if I am right, why was this simple adjustment to Clark’s proposal not made or even considered? What happened was that theorists tacitly bypassed the notion of tacit assumption and, in effect, tried to analyze it in turn. The nearly instantaneous result was the indefeasibility literature, which began with the useful notion of a defeater, a proposition which if added to S’s epistemically justifying evidence would render the expanded evidence set no longer epistemically justifying. That literature had no success (pace Swain 1974, quoted above); but I say its failures showed, not that the no-false-assumption analysis of knowing was wrong, but that the notion of tacit assumption is itself hard to characterize (in turn) by reference to a defeater. (Indeed, that difficulty was predictable, because (a) it was almost irresistible to start the further analysis with a subjunctive of some kind,[13] and (b) any time any analysis of anything contains a subjunctive, irrelevant counterexamples will ensue. (b) is worth a paper of its own.) In fact, the notion of tacit belief is hard to characterize in any terms at all, never mind subjunctives (Lycan 1986). It was the further ‘defeater’ analyses of assumption that were wrong, not the no-false-assumption analysis of knowing.

(Should it be protested that I should not analyze ‘know’ in terms of a notion whose own analysis is so vexed, one would have to make the same complaint against many theorists, in particular against anyone who analyzes anything in terms of causality.)

But the counterexamples to necessity are effective, even against the weakened no-false-assumptions analysis. The mere presence of a false assumption does not blight knowledge.

Harman (1973: 47) provides the obvious fix: What is required is only that a justification must not essentially rest on a false assumption; any false assumption on which it does rest must be dispensable. As before (the reply is prefigured in the description of Nondefective Chain itself), if S has even one nongettiered epistemic justification, it does not matter if S has other justifications containing false assumptions. So we move from no-false-assumptions to no-essential-false-assumptions.

The same applies, though not quite so directly, to Togethersmith. S does assume that all the Togethersmiths are out in the car, and that assumption is false. But it is not essential to S’s justification. As Rozeboom himself insists, it is irrelevant that the one child is not in the car. S also tacitly assumes that Mrs. Togethersmith is in the car, and has just as good inductive evidence for that assumption as S has for the belief that all the Togethersmiths are in it.

But now (the expert reader will have been shouting for some time) two further putative counterexamples to sufficiency loom, each very well known: Harman’s (1973) unpossessed-defeater sort of case, and the Ginet-Goldman barn case (Goldman 1976).

The assassination: Jill reads a true newspaper account of a political assassination. The reporter is known to be entirely trustworthy, and he was himself an eyewitness. Nor is Jill gettiered. But the victim’s associates, wishing to forestall panic, have issued a television announcement saying (falsely) that the assassination attempt failed and that the intended victim is alive. Nearly everyone has heard the television announcement and believes it. However, by a fluke, Jill misses it and continues (epistemically justified and nongettiered) to believe that the victim is dead.

(Harman’s other two very similar examples are those of Tom Grabit’s mother’s false but widely accepted testimony, and Donald in Italy and his faked letters from California.)

Fake Barn Country: Henry is looking at a (real) barn, and has impeccable visual and other evidence that it is a barn. He is not gettiered; his justification is sound in every way. However, in the neighborhood there are a number of fake, papiere-mâché barns, any of which would have fooled Henry into thinking it was a barn.

It is claimed that Jill and Henry do not know.[14] What distinguishes these cases from the preceding counterexamples to sufficiency is that, in them, there are no identifiable false tacit assumptions. In no reasonable sense is Jill tacitly assuming that no one has issued a television announcement that the assassination attempt failed; nor is Henry assuming that there are no papiere-mâché barn replicas in the neighborhood. (Or if there is such a sense, it is a very loose one and not that same clear one in which our previous Gettier victims were making their specific tacit assumptions.) The no-essential-false-assumptions theory does not rule out these examples.

My reply is that, on quite independent grounds, I reject the received intuitions; I do not share them and I also think they are mistaken. I maintain that Jill and Henry do know, despite the chance elements that peripherally invade their situations. I have argued that at length in Lycan (1977), and not on behalf of the present analysis. Readers who do share the unpossessed-defeater and barn intuitions and who have not read my arguments cannot be expected to agree, but I stand by the no-essential-false-assumptions analysis. A bit more argument against Harman and Ginet-Goldman will be enlisted from Hetherington (1999), in section v below.

IV Family resemblances

One cannot help noticing that the Gettier project is a Socratic search for a set of necessary and sufficient conditions for knowing. In the 1960s that would have been even harder not to notice, because under the influence of Wittgenstein, the Socratic assumption had come under siege: it was nearly anathema to suppose that an interesting concept could be defined by a crisp set of necessary and sufficient conditions. Accordingly, it was thought in some quarters that that is what was wrong with the Gettier project; the ‘S knows that p’ crowd had not read their Wittgenstein, and did not understand that ‘know’ is a family-resemblance term (e.g., Saunders and Champawat 1964).

Given the aforementioned nonsuccess of later chisholming projects, the Wittgensteinians’ negative judgment is hard to fault. Perhaps no philosophically interesting concept admits of explication by strictly necessary and sufficient conditions. However, that does not explain why the Gettier project was held to be worse off than the other Socratic quests of the day. Also, the Wittgensteinians’ positive judgment is a substantive commitment and in need of defense: Is ‘know’ a family-resemblance concept? -- i.e., does it in fact have that structure?

Actually there are two or more different structures that have been called ‘family-resemblance’ structures. The most distinctive one is that in which a concept ‘X’ is defined by a paradigm case: There is a list of features, each of which would be possessed by a paradigm case of an X; if a thing has every feature on the list, the thing is an X by any standard, a real X, an X and a half, an X on wheels. To be an X per se, however, is just to have ‘enough’ of the features on the list. (Perhaps the features are weighted in combinations, but not so thoroughly as to constitute a traditional analysis.) We cannot be much more precise than just to say ‘enough.’ There will be pretty-much-Xs, sort-of-Xs, borderline Xs, things that are Xish but not really Xs. Call this the ‘Paradigm’ structure.

‘Know’ does not have the Paradigm structure.I suppose there is a paradigm for inferential empirical knowledge. (Though according to Plato or Descartes, no case of inferential empirical knowledge would be very close to the paradigm of Knowledge itself.) If S has overwhelming amounts of evidence for believing that p, has not the slightest reason to doubt that p, and is not gettiered or beset by fluke in any way at all, then (barring global skepticism) S surely knows that p. But suppose S meets the first of those two conditions but not the third, i.e., S is a classic Gettier victim. Then (as before) S does not pretty-much-know that p; S is not a good though imperfect example of a knower. S simply and flatly does not know. It is not that S fails to have ‘enough’ of the paradigm features of knowing. It is that S is gettiered and so disqualified, period.[15]

Wittgenstein’s own ‘family resemblance’ metaphor does not support the Paradigm interpretation, even though his central examples, such as ‘game,’ do exhibit the Paradigm structure.[16] ‘[W]e see a complicated network of similarities overlapping and criss-crossing; sometimes overall similarities, sometimes similarities of detail…. I can think of no better expression to characterize these similarities than ‘family resemblances’; for the various resemblances between members of a family: build, features, colour of eyes, gait, temperament, etc. etc. overlap and criss-cross in the same way’ (1953: §§66, 67). On this model, there is no one paradigm, because some of the traits in question may be mutually incompatible and nothing could have them all. There might be sub-paradigms, but there need not be those either. As before, though, to fall under the concept to a degree is to have ‘enough’ of the traits in some acceptable combination. Call this the ‘Criss-Crossing’ structure.

But ‘know’ does not have the Criss-Crossing structure either, for the same reason as before. It is not that poor gettiered S fails to have ‘enough’ of the family features; it is that S is disqualified. Also, there is no very visible ‘family’ composed of people who have one or two or three of the traits: believing that p, its being true that p, having evidence that p, not being gettiered. Rather, there is more of an epistemological hierarchy: believing that p, believing truly that p, justifiedly and truly believing that p, epistemically-justifiedly and truly believing that p, epistemically-justifiedly and truly believing that p and not being gettiered (though one wonders where on this scale to put epistemically-justifiedly and falsely believing that p, and it is not obvious whether justifiably and truly but not epistemically-justifiedly believing that p but not being gettiered should be ranked higher or lower than epistemically-justifiedly believing that p and being gettiered).

Even if ‘know’ does not have any family-resemblance structure, there is a more basic complaint that has sometimes been made: that ‘JTB’ is flawed to begin with, before we get to the question of its sufficiency. In particular, it is said, knowledge is not a kind of believing, indeed is not a psychological state at all (Austin 1961; Vendler 1972). Indeed, knowing does not even entail believing (Radford 1966).

But this is no solution to the Problem problem. The traditional claim that is needed to set up the Problem is only the sufficiency thesis: that if S does believe that p truly and with epistemic justification, then S knows. The falsity of that thesis is interesting and important and raises the Gettier question, whether or not knowing entails believing. There are people who have JTB and accordingly know; but, surprisingly, there are people who have JTB and do not know. What distinguishes the former from the latter?

V More recent complaints about the Gettier Problem

Insolubility. Craig (1990) and Zagzebski (1994) suggest an argument for the claim that the Gettier Problem is insoluble: So long as a particular fourth condition added to the original three still leaves a logical possibility that a belief might meet all four conditions and still be false, there will always be room for further Gettierish flukes and hence there will be counterexamples; S could be supergettiered, even if S is not gettiered in the customary way. But if the fourth condition shuts off that possibility, it will rule out lots of ordinary instances of knowing and hence be too strong. Thus, the Gettier Problem is insoluble and for a predictable reason, and that is what is wrong with it.

Neither Craig nor Zagzebski actually accepts this argument; indeed Zagzebski (1999) rejects the second horn of the dilemma, and goes on to offer her own solution to the Problem. Fogelin (1994) and Merricks (1995) too accept the first horn but not the second, drawing the moral that whatever ‘epistemic justification’ is, it must guarantee the truth of the belief: either S has evidence that entails that p, or S could not possibly be believing that p on the basis of that evidence in the circumstances unless p.[17]

Each horn is somewhat plausible. If there is no guarantee of truth, then it does seem that a Gettierish fluke would always be available, though we have not seen an algorithm for generating one. The second horn is supported by the same problem that afflicted its particular instance, the Dretske-Armstrong nomic reliability theory: that perfectly ordinary cases of knowledge do not seem to meet the guarantee-of-truth requirement.

But, whatever the merits of the Craig-Zagzebski argument, it would not be a very interesting solution to the Gettier Problem problem, because if sound it shows that some version of skepticism is true. It says that to be knowledge, a belief must meet the guarantee condition, and that hardly any beliefs meet the guarantee condition.[18] That is an interesting argument for skepticism, all the more so for being instigated by the Gettier Problem itself, but for our present purposes it still proves too much.

Unanalyzability. There is an ambitious antiGettier claim, seemingly unanswerable if true: that ‘know’ is unanalyzable. Even if knowledge has necessary conditions such as truth and belief, of course it does not follow that ‘know’ is analyzable in terms of those. Williamson (2000) argues at length that ‘know’ should be taken as primitive. If he is right, then of course any project which bills itself as ‘analyzing knowledge’ is doomed to failure. And that is what the Gettier project did bill itself as doing.

However, even if ‘know’ is unanalyzable and has no set of conceptually necessary and sufficient conditions, the claim needed to set up the Problem is (again) only the sufficiency thesis: that epistemically justified true belief suffices for knowledge. The falsity of that thesis still needs explaining, because, as before, there are real people who have JTB but still do not know, and that raises the question of what distinguishes the knower from the Gettier victim. The Gettier project can rage on unabated.

Rejecting the Ur-intuition. Hetherington (1999; 2001) maintains that a Gettier victim does know, though in a somewhat inferior or ‘less-then-ideal’ way: s/he knows very ‘failably.’

‘Failability’ is a generalization of fallible knowing, and means roughly that although S knows, there is a single element of luck, in virtue of which S might not have known or even nearly failed to know. More precisely: Either there is a possible world in which S believes that p and is justified by the same good evidence, but it is not true that p, or there is one in which S correctly believes that p but is not justified by the same evidence in doing so, or there is one in which although S would be both correct in believing that p and justified by the evidence, S does not hold that belief (1999: 567). (Thus, one has infailable knowledge iff ‘[i]n each world where one exists and where one has two of the elements of knowing that p, one also has the third element’: 568.) Note that failability is a matter of degree; some knowledge will admit more and/or closer epistemic-failure worlds than does other knowledge.

Having identified fallible knowing with the first of the three foregoing disjuncts, Hetherington argues that it is arbitrary to single out that disjunct in preference to the other two, and so his generalization from fallibility to failability is natural and nontendentious. There is at least a presumption, then, that knowledge can be failable in either of the other two ways as well.

Fallible knowing is of course presupposed by the Gettier Problem; hence so is failable knowing. Now Hetherington suggests that a Gettier case is one in which, although S does know, S knows only very failably; ‘the epistemic subject almost fails to have his well-justified true belief’ (573). A classic Gettier example falls under the first disjunct; there are scads of very nearby worlds in which S believes (on the very same very strong evidence grounded in Mr. Nogot) that someone in the office owns a Ford, but in which neither Havit nor anyone else owns one. A Harman unpossessed-defeater case falls under the third disjunct, for there are lots of nearby worlds in which the assassination does occur and Jill has her same evidence for it, but in which (because she did there hear the government denials) she has abandoned her belief.

Obviously, if the Gettier victim knows, her/his knowledge is not just failable but very failable. But why should we forsake all received judgment and concede that s/he does know? Rather than giving a positive reason, Hetherington spends the rest of his article suggesting diagnoses of our failure to fall in with his view. The mainstream epistemologist has made a tacit fallacious inference: from the fact that ‘there must be a difference in the quality of the instances of knowing in, respectively, a normal situation where there is failable knowledge that p, and a Gettier situation where there is failable knowledge that p’ (575); or from ‘how easy it is to imagine changes to the circumstances within a Gettier case, changes which would have led to the case’s epistemic subject not having the well-justified true belief he actually has’ (579); or from the subject’s true belief owing anything at all to luck (581); or from the subject’s epistemically justified belief being not robust, a ‘near thing’ (581-2); or from the fact that ‘the more failable…[a piece of knowledge] is, the less confident we might be that it is knowledge’ (585); or, when we are already sniffing around knowledge’s ‘lower boundary,’ from the fact that knowledge has a lower boundary (586).

I suspect that few of my fellow mainstreamers will recognize themselves in these diagnoses, and even fewer will be persuaded to adopt Hetherington’s maverick verdict on Gettier cases generally. But, although my most diligent introspection reveals none of the fallacious inferences on my own part either, I am more sympathetic to Hetherington’s view than most will be. He very usefully distinguishes between ‘helpful’ Gettier cases and ‘dangerous’ ones: A helpful case is one in which the Gettierish ‘strange occurrence’ or fluke saves JTB itself, as when Havit owns a Ford even though Nogot does not. A dangerous case is one in which the ‘strange occurrence’ prevents knowledge despite existing normal JTB, as in Harman’s unpossessed-defeater examples and the Ginet-Goldman barn case.

As I declared in section iii, I reject the majority view that the victims in unpossessed-defeater cases and the barn case lack knowledge. And now Hetherington has shown that those examples have something distinctive in common, viz., being ‘dangerous’ as opposed to ‘helpful.’[19] Moreover, I think his interpretation of them is pretty much right: that although their protagonists’ knowledge is failable and some luck is involved in a peripheral way, it is knowledge nonetheless. True, Jill and Henry nearly failed to know; it does not follow that they fail to know. With Hetherington, I maintain that they do know.

(But in my view the same cannot be said about the classic, ‘helpful’ cases. I could not possibly jolly myself into agreeing that S knows that someone in the office owns a Ford, when S’s only reason for believing that is that S thinks Nogot owns one. I find it hard to imagine that anyone would credit S with knowing that there is a sheep in Chisholm’s field when the sheep that quite coincidentally makes S’s belief true is off in a distant corner of the field, hidden from view by hedges. (However, this may only show the poverty of my imaginative powers; see below.))

Weatherson (2003) also urges, though on grounds very different from Hetherington’s, that gettiered people do know. His idea is a very general one about philosophers’ ‘intuitions’: that intuitions about cases should be trumped, as they are often considered overruled in ethical theory, by a good (otherwise) coherent and systematic theory that says otherwise. Though an analysis must respect a majority of intuitions, it may disregard one when that one forces us into ‘unnatural’ complications and draws the analysandum away from comparatively natural properties in the world. (No one who has read the indefeasibility analysis quoted in section i above could deny that the Gettier intuition has been known to do such forcing and drawing.)

Weatherson’s paper is complex and rich, and I cannot do it justice here. I accept his ‘main claim … that even once we have accepted that the JTB theory seems to say the wrong thing about Gettier cases, we should still keep an open mind to the question of whether it is true’ (10).[20] I shall merely state four reasons why, though I agree that the seeming does not entail the falsity of ‘JTB,’ I continue to join in the majority view.

First, though Swain’s indefeasibility analysis is hideously ‘unnatural,’ of course I chose it as an extreme case. Not every proffered analysis is so complex or so disjunctive. Just to take a random example, my own no-essential-false-assumptions analysis is not so unnatural. It is rather neat, I think.

Second, I do not believe that JTB is a conspicuously more natural kind than nongettiered JTB. If anything, I would say a Gettier victim has more in common with a not fully justified believer than with a knower.[21] (Also, those of us who think that ‘know’ inherits normativity from its relation to justification would not expect knowledge to be a particularly natural kind.)

Third, I believe intuitions have enough authority that if we want to reject one, we ought to explain it away. I think Weatherson agrees, and of course he is well aware that this happens often in philosophy. Why, then, is there so widespread instant agreement that Gettier victims do not know? As noted above, Hetherington put in some work on this, however plausible or implausible we think his explanations are; but unless I have missed it, Weatherson does not offer anything comparable.

Finally, Weatherson’s argument does not single out the Gettier Problem, even though the Problem is his stalking horse. Similar points could be made about all the other analytic projects mentioned in section I. So Weatherson has not solved the Problem problem, so far as that requires exhibiting some special defect in the Gettier Problem that distinguishes it from analytic projects generally.

Actual diversity of intuitions. Weinberg, Stich and Nichols (2001) present data they have collected, according to which the intuitions of subjects from different ethnic groups vary statistically. In particular, 60% of subjects originally from the Indian subcontinent, presented with a Gettier example, judged that its protagonist does ‘really know’ as opposed to ‘only believe.’ For that matter, nearly 25% of the European-descended American subjects made the same antiGettier judgment. This raises two issues: First, is there cultural relativity in the concept of knowing? Second, even within the class of, say, educated European-descended Americans, is the Gettier intuition reliable? If the answer to either question, especially the second, is ‘no,’ then the Gettier project is parochial at best, and is not an augustly Socratic inquiry into the nature of Knowledge Itself.

I have several doubts about the experimental procedures described by the authors, and I would not take their results at face value. But they do not claim too much for them. And to make things interesting, let us ignore such doubts, and suppose that the survey results are impeccably produced and robustly replicated: 60% of an Asian ethnic group and 25% of European-descended American undergraduates firmly reject Gettier and insist, clearheadedly and understanding the terms and the issue, that a Gettier ‘victim’ does know.

In that eventuality, I submit, we have a conceptual difference. In the speech of the 60% and the 25%, ‘know’ really does mean justified true belief, period. We would have to regard that speech as a dialect that differs from our own. It would be interesting to go on to ask those subjects whether they see any important difference between the two kinds of ‘knowers,’ ordinary ones and Gettier victims. Perhaps they would stigmatize the Gettier victims in some way for which there is no simple convenient expression. Or, less likely, they would see no important difference, and simply have no stronger conception of successful cognition.

This sort of dialect difference is less rare than one might think. It can lurk unsuspected for decades or whole lifetimes, because it is slight and the sort of hypothetical case that would bring it out is unusual. Here is an example from my own experience. Sartre bemoans the fact that we have no simple expression for the following situation:

A believes that not-p, but for selfish reasons wants B to believe that p. In a persuasive manner, A tells B that p: ‘p, B; trust me, old friend, would I ever lie to you?’ Now in fact, A is mistaken, and it is true that p. A has tried to lie to B, and A’s character is that of a liar. But what A said was true, so it cannot be called a lie.

On many occasions I have mentioned this in my undergraduate classes, and every time, about 40% of the students balk at Sartre’s judgment, and say they have no difficulty in calling A a liar. When I protest that a lie cannot be true, they say, ‘Sure it can’; all that matters to them is the intent to deceive. On the basis of induction, I predict that 40% of my readers will likewise have rejected Sartre’s complaint.

There is no substantive issue here. Neither I nor the 40% are right to the exclusion of the other. It is simply a dialect difference -- one that I did not discover until I was in my 40s.[22]

If another culture has a word that we have been translating as ‘know’ but turns out not to share the Gettier intuition, then their word should not strictly be translated as ‘know’ (though there may be no convenient competing expression of English). And if it is really true that 25% of ordinary English speakers simply do not share the intuition, there is a dialect difference.

Weinberg, Stich and Nichols may urge that such an outcome would diminish the importance of the Gettier project. Gettier practitioners would then be pursuing only the minutiae of a concept possessed by some speakers of English. I reply: So be it. The concept has proved to be an important one among English-speaking philosophers, regardless of how more widespread it may be. If another culture or another dialect group simply does not have that concept, then of course the Gettier Problem does not arise for them.

Now, I take very seriously the cynical suggestion that the Gettier concept is a philosophers’ artifact and does not represent anything possessed by ordinary people. No professional philosopher is qualified to make any pronouncement about the ordinary concept of anything -- period -- though few of us can resist making such pronouncements. I believe that some philosophically important and contentious concepts are such artifacts. My leading example would be Putnam’s (1975) externalist natural-kind concepts. When I teach ‘The Meaning of “Meaning”’ to novices, they invariably resist. At best I can get them to concede that there is a sense in which XYZ on Twin Earth is not water.[23] But there is a sharp contrast here: I have never had the slightest trouble convincing novices by Gettier example that ‘JTB’ is insufficient for knowing. And I do not think that is because of my being the instructor or my natural authority, let alone force of personality or great professional stature.

VI Prognosis?

Fodor et al. (1980) have argued convincingly that no interesting concept can be analyzed in the traditional Socratic way, by a nice set of individually necessary and jointly sufficient conditions. At least on inductive grounds, we should not expect a solution to the Gettier Problem having that form. But it remains to be shown what we should expect, instead.

And as I have argued, none of the going solutions to the Problem problem succeeds. So far as has been shown, there is nothing particularly wrong with the Gettier Problem, and people who work on it do not (for that reason) deserve the sneers that are sometimes sent their way.

I am happy with my simple no-essential-false-assumptions analysis. What about you?[24]

References

Adler, J. (1981).“Skepticism and Universalizability,” Journal of Philosophy 78: 143-56.

Almeder, R. (1974). ‘Truth and Evidence’. Philosophical Quarterly, 24: 365-68.

Armstrong, D.M. (1973). Belief, Truth, and Knowledge.Cambridge: CambridgeUniversity Press.

Austin, J.L. (1961). ‘Other Minds’, in Philosophical Papers. Oxford: OxfordUniversity Press.

Battaly, H. (2001). ‘Thin Concepts to the Rescue’, in A. Fairweather and L. Zagzebski (eds.), Virtue Epistemology. Oxford: Clarendon Press.98-116

Chisholm, R.M. (1966). Theory of Knowledge. Englewood Cliffs: Prentice-Hall.

Clark, M. (1963). ‘Knowledge and Grounds: A Comment on Mr. Gettier's Paper’. Analysis, 24: 46-48.

Craig, E. (1990). Knowledge and the State of Nature. Oxford: Clarendon Press.

Dretske, F. (1971). ‘Conclusive Reasons’. Australasian Journal of Philosophy, 49: 1-22. Reprinted in Pappas and Swain (1978).

Feldman, R. (1974). ‘An Alleged Defect in Gettier Counter-Examples’. Australasian Journal of Philosophy, 52: 68-69.

Fodor, J.A., M. Garrett,E. Walker, and C. Parkes (1980). ‘Against Definitions’. Cognition 8: 263-367.

Fogelin, R. (1994). Pyrrhonian Reflections on Knowledge and Justification. Oxford: Clarendon Press.

Gettier, E. (1963). ‘Is Justified True Belief Knowledge?’. Analysis, 23: 121-3. Reprinted in Roth and Galis (1970).

Goldman, A.I. (1967). ‘A Causal Theory of Knowing’. Journal of Philosophy 64: 357-72. Reprinted in Roth and Galis (1970) and in Pappas and Swain (1978).

----. (1976). ‘Discrimination and Perceptual Knowledge’. Journal of Philosophy 73: 771-91. Reprinted in Pappas and Swain (1978).

----. (1999). Knowledge in a Social World. Oxford: Clarendon Press.

Harman, G. (1966). ‘Lehrer on Knowledge’. Journal of Philosophy

----. (1973). Thought. Princeton: PrincetonUniversity Press.

Hetherington, S. (1999). ‘Knowing Failably’. Journal of Philosophy 96: 565-87.

----. (2001). Good Knowledge, Bad Knowledge. Oxford: OxfordUniversity Press.

Hintikka, K.J.J. (1962). Knowledge and Belief. Ithaca: CornellUniversity Press.

Lehrer, K. (1965). ‘Knowledge, Truth and Evidence’. Analysis, 25: 168-75. Reprinted in Roth and Galis (1970).

----. (1970). ‘The Fourth Condition of Knowledge: A Defense’. Review of Metaphysics, 24: 122-8.

----. (1974). Knowledge. Oxford: Clarendon Press.

----. and Paxson, T. (1969). ‘Knowledge: Undefeated Justified True Belief’. Journal of Philosophy, 66: 225-37. Reprinted in Pappas and Swain (1978).

Lycan, W.G. (1977). ‘Evidence One Does Not Possess’. Australasian Journal of Philosophy 55: 114-26.

----. (1984). ‘Armstrong’s Theory of Knowing’, in R.J. Bogdan (ed.), Profiles: D.M. Armstrong. Dordrecht: D. Reidel,139-160.

----. (1986). ‘Tacit Belief’, in R.J. Bogdan (ed.), Belief.Oxford: OxfordUniversity Press.61-82

----. (1994a). Modality and Meaning. Dordrecht: Kluwer Academic Publishers, Studies in Linguistics and Philosophy Series.

----. (1994b). ‘Sartwell’s Minimalist Analysis of Knowing’. Philosophical Studies, 73: 1-3.

Merricks, T. (1995). ‘Warrant Entails Truth’. Philosophy and Phenomenological Research, 55: 841-55.Pailthorp, C. (1969). ‘Knowledge as Justified True Belief’. Review of Metaphysics, 23: 25-47.

Pappas, G., and M. Swain (1973). ‘Some Conclusive Reasons against “Conclusive Reasons”’. Australasian Journal of Philosophy, 71: 72-6. Reprinted in Pappas and Swain (1978).

----. (eds.) (1978). Essays on Knowledge and Justification. Ithaca: CornellUniversity Press.

Powers, L. (1978). ‘Knowledge by Deduction’. Philosophical Review, 87: 337-71.

Putnam, H. (1975).“The Meaning of ‘Meaning’,” in K. Gunderson (ed.), Language, Mind and Knowledge. Minnesota Studies in the Philosophy of Science 7. Minneapolis: University of Minnesota Press.215-71.

Quine, W.V. (1936).“Truth by Convention,” in O.H. Lee (ed.), Philosophical Essays for A.N. Whitehead.New York: Longmans.Reprinted in Quine (1966).90-124

----. (1963).“Carnap and Logical Truth,” in P.A. Schilpp (ed.), The Philosophy of Rudolf Carnap.LaSalle: Open Court.Reprinted in Quine (1966).385-406

----. (1966).The Ways of Paradox.New York: Random House.

Radford, C. (1966). ‘Knowledge -- By Examples’. Analysis, 27: 1-11. Reprinted in Roth and Galis (1970).

Roth, M., and Galis, L. (eds.) (1970). Knowing.New York: Random House.

Rozeboom, W.W. (1967). ‘Why I Know So Much More than You Do’. American Philosophical Quarterly, 4: 281-90. Reprinted in Roth and Galis (1970).

Russell, B. (1948). Human Knowledge: Its Scope and Limits. New York: Allen and Unwin.

Sartwell, C. (1991). ‘Knowledge is Merely True Belief’. American Philosophical Quarterly, 28: 157-65.

----. (1992). ‘Why Knowledge is Merely True Belief’. Journal of Philosophy,89: 167-80.

Saunders, J.T., and Champawat, N. (1964). ‘Mr. Clark’s Definition of “Knowledge”’. Analysis, 25: 8-9.

Scheffler, I. (1965). Conditions of Knowledge. Chicago: Scott, Foresman.

Schiffer, S. (1972). Meaning. Oxford: Clarendon Press.

Shope, R. (1978). ‘The Conditional Fallacy in Contemporary Philosophy’. Journal of Philosophy, 75: 397-413.

Skyrms, B. (1967). ‘The Explication of “X knows that p”’. Journal of Philosophy, 64: 373-89. Reprinted in Roth and Galis (1970).

Swain, M. (1972). ‘Knowledge, Causality, and Justification’. Journal of Philosophy, 69: 291-300. Reprinted in Pappas and Swain (1978).

----. (1974). ‘Epistemic Defeasibility’. American Philosophical Quarterly, 11: 15-25. Reprinted in Pappas and Swain (1978).

----. (1978). ‘Some Revisions of “Knowledge, Causality, and Justification”’, in Pappas and Swain (1978).

Unger, P. (1971). ‘A Defense of Skepticism’. Philosophical Review, 80: 198-218.

----. (1984). Philosophical Relativity. Oxford: Basil Blackwell.

Vendler, Z. (1972). Res Cogitans. Ithaca: CornellUniversity Press.

Weatherson, B. (2003). ‘What Good are Counterexamples?’. Philosophical Studies, 115: 1-31.

Weinberg, J.M., S.P. Stich and S. Nichols (2001). ‘Normativity and Epistemic Intuitions’. Philosophical Topics, 29: 429-60.

Williamson, T. (2000). Knowledge and its Limits. Oxford: Clarendon Press.

Wittgenstein, L. (1953). Philosophical Investigations. (Oxford: Basil Blackwell.)

Zagzebski, L. (1994). ‘The Inescapability of Gettier Problems,’ Philosophical Quarterly 44: 65-73.

Zagzebski, L. (1999). ‘What is Knowledge?,’ in J. Greco and E. Sosa (eds.), The Blackwell Guide to Epistemology. (Oxford: Basil Blackwell.)92-116

Footnotes