I The variabilist reaction against traditional epistemology
It is fair to say that from the time of the Theaetetus until relatively recently, theorists of knowledge tended to conceive their central task as being to explain in what knowledge consists; more exactly, to explain what further conditions need to be satisfied by a true belief if it is to count as knowledgeable. The widely accepted failure of the post-Gettier debates to execute this task convincingly has motivated a very different tendency in mainstream contemporary epistemology. This, influentially promoted by Timothy Williamson in particular, is epistemic primitivism: to concede that knowledge is, as Williamson puts it, âprimeâ â that it is a fundamental, irreducible cognitive relation. Knowledge, on the primitivist view, is a basic epistemological kind, and to know is to be in a basic, sui generis attitudinal state. There can therefore be no correct analysis of it in terms of other, supposedly constitutive or more fundamental cognitive states (true belief + X). The post-Gettier âX knows that P if and only if âŚâ cottage industry was doomed to disappointment for this reason. To the contrary, it is in terms of knowledge that other epistemic notions â justification, evidence, warranted assertion and rational action â are to be understood.1
This primitivism, however, still shares three traditional assumptions with the reductionism it is set against. They can be wrapped together as the compound idea that knowledge is a unique, objective, purely cognitive type of state â hence something at which the aspiration of reductive analysis could be sensibly (even if mis-) directed. If we unpack that, however, we find the following three distinct thoughts. First, ascriptions of knowledge, that X knows that P, are contentually invariant as far as the semantic contribution of âknowsâ is concerned. More specifically, once the referent of âXâ, the identity of the proposition that P and the time reference associated with âknowsâ are settled, the result is a unique proposition, the same for any competent thinker who considers it. Second â although this would normally be taken to be entailed by the first point â this unique proposition has one and the same truth-value, no matter who asserts or assesses it. Third, this truth-value is determined purely by the cognitive achievements of the subject, irrespective of what else, other than that part of her total information relevant to the judgement that P is true of X. In particular, such aspects as Xâs (or anyone elseâs) interest in whether P, is true, or what is at stake for her in its truth, or the range and specifics of counter-possibilities to P that occur, or are salient, to X â in short: such, as they are often described, ânon-traditionalâ or as I shall say pragmatic factors â have no bearing on the matter.
The striking recent tendency that provides the subject matter of this chapter is the rejection of one or more of these traditional assumptions in favour of one or another form of variabilism: broadly, the notion that whether an ascription of knowledge may correctly be regarded as true may depend on pragmatic factors that pertain to the circumstances of the ascriber, or to those of a third party assessing the ascription, or on pragmatic aspects of the circumstances of the ascribee. Although well short of a consensus, a considerable body of opinion has been developed that agrees that some form of epistemic variabilism is called for if justice is to be done to the actual employment of âknowsâ and its cognates. In what follows, I will review some of the principal considerations that are taken to support that view, critically compare and assess some of the resulting variabilist proposals and recommend a conclusion both about them and about the prospects for primitivism.
II Three types of consideration suggestive of variability
(i) Hume remarked long ago on the contrast between the potency of skeptical doubts, at least in their subtler forms, when developed in the philosophical study and their apparent fatuity when considered in the pub over beer and backgammon.2 We may of course address the tension by proposing that one or the other â study or pub â response has to be misconceived; but then we remain in a state of cognitive dissonance until we have given a convincing account of which. No need, however, for such an account if âOne normally knows that one has two handsâ is false in the study but true in the pub â either because the semantic value of âknowsâ is context-sensitive (and the context shifts in relevant respects as one moves from the study to the pub), or because the truth-value of the single proposition expressed by tokenings of that sentence in both locations is not absolute or because we change in relevant pragmatic respects as we move from the study to the pub.
(ii) John Hawthorne and others have emphasised problems generated by our ordinary practices of knowledge-ascription for the principle of the closure of knowledge across known entailment.3 A range of cases exists where one might naturally self-ascribe knowledge of a premise of what one knows is a trivially valid entailment but might then hesitate to self-ascribe knowledge of its conclusion. Some of the most striking are so-called lottery cases. Suppose you buy a ticket in next weekâs UK National Lottery (the first prize has built up to about ÂŁ50 million). You are under no illusions about the odds and sensibly expect (truly, letâs suppose) that you will not win. But, despite this true beliefâs being overwhelmingly strongly justified, there are powerful reasons for denying that it is, strictly, knowledgeable. For one thing, if it is knowledgeable, then your buying the ticket is irrational â but that seems a harsh verdict; indeed if the scale of the prize and the odds suitably combine, the expected utility may actually rationalise the purchase. Moreover having bought a ticket, you will have, if you know that it wonât win, no reason not to tear it up. But actually, once having bought a ticket, tearing it up would seem irrational so long as you have every reason to think that the lottery is fair.
Suppose it agreed that, for such reasons, you donât strictly know that you wonât win the lottery. On the other hand, there are plenty of things that in ordinary contexts you would take yourself to know â for instance, that you wonât be able to afford to buy a new Maserati next week or to retire at the end of the current academic year â that entail that you wonât win the lottery. And in general there are plenty of things we would ordinarily be regarded as in position to know about our future circumstances in all kinds of respects (indeed, had better know if knowledge is the basis of rational practical reasoning to conclusions about what to do) that, in turn, entail that we wonât be the subject of various forms of unlikely happenstance â even in cases, like lotteries, where it is sure that someone will be â which, once contemplated, we will be inclined to acknowledge that we donât strictly know will not occur.
There is the option of regarding such cases as actually challenging the validity of closure, of course. But that is a hard row to hoe.4 Variability offers a different recourse. Perhaps the very act of bringing to mind the conclusion of an entailment of relevant kind âups the anteâ in some way. Maybe the correctness of your self-ascription of knowledge that you will not be able to afford to retire at the end of the current academic year is originally relativised to a range of salient counter-possibilities which do not include lottery wins and which you are a position to rule out â and maybe this range enlarges with the purchase of the ticket.
(iii) Perhaps the dominant motivation towards variabilism, however, springs from a range of putative linguistic âintuitionsâ concerning proprieties of knowledge-ascription provoked, at least among many of the philosophers who think about them, by imaginary cases of a kind first put forward by Stewart Cohen and Keith DeRose.5 We can illustrate by reference to a version of DeRoseâs famous Bank Case. Suppose it is Friday afternoon, and Ashley and Bobbie are considering whether to bank their salary cheques. There are long queues at all the bank counters. Ashley recalls being at the bank on a Saturday morning two weeks ago and says, âLetâs come back tomorrow. I know the bank will be open tomorrow morning.â Suppose that the bank will indeed be open on the Saturday morning.
Case 1 (Low stakes): Suppose that there is no particular reason to ensure that the cheques are banked sooner rather than later â say, by the following Monday. Then
Invited intuition: Ashleyâs recollection of Saturday morning opening two weeks ago suffices for her to speak truly.
Contrast that scenario with
Case 2 (High stakes): The coupleâs mortgage lender will foreclose unless the cheques are in the account by Monday to service their monthly repayment. Ashley and Bobbie know this. Bobbie says, âBut what if the bank has changed its opening hours? Or what if the Saturday morning opening was some kind of one-off promotion?â Ashley says, âYouâre right. I suppose I donât really know that the bank will be open tomorrow (even though I am pretty confident that it will). We had better join the queue.â
Invited intuition: Again, Ashley speaks truly. Too much is at stake to take the risk of e.g. a change in banking hours.
So the suggested conclusion is that âI know the bank will be open tomorrowâ uttered by Ashley is true in Case 1 and false in Case 2 even though all that is different between the two are the costs to Ashley and Bobbie of Ashleyâs being wrong. Only the pragmatic factors have changed. Everything that might be mentioned in a traditional account of knowledge â as we would naturally say, all Ashleyâs relevant evidence or information â remains the same.
Two further cases may seem to prompt another important conclusion:
Case 3 (Unknowing high stakes): The coupleâs mortgage lender will indeed foreclose unless the cheques are in the account by Monday to service their monthly repayment but Ashley and Bobbie are unaware of this (they habitually leave what looks like circular mail from the mortgage company unopened and have missed the reminder). The dialogue proceeds as first described above, with Ashley asserting, âI know the bank will be open tomorrow morning.â
Invited intuition: This time, Ashley speaks falsely.
Compare that with
Case 4 (Unknowing low stakes): Ashley and Bobbie actually have no good reason to ensure that the cheques are banked before Monday but, misremembering the notice from the mortgage company, they falsely believe that Monday will be too late. The dialogue proceeds as in Case 2.
Invited intuition: This time Ashleyâs disclaimer, âI suppose I donât really know that the bank will be open tomorrowâ is false.
The suggested conclusion from Cases 3 and 4 is this: when changes in pragmatic factors convert a true knowledge-ascription into a false one, or vice versa, it is actual changes that matter, rather than thinkersâ impressions of what changes in such factors may have taken place.
III The varieties of variabilism
We have already, in effect, noted that the space of theoretical options here must include at least three quite different kinds of proposals: one for each of the traditional assumptions distinguished in section I. First, there is the option of maintaining that although knowledge-ascriptions are contentually invariant (in the sense there specified), the proposition thereby expressed may take different truth-values in different circumstances, depending on variation in the pragmatic factors applying to its subject, X. This is the thesis, proposed separately by Stanley and Hawthorne,6 that is most often termed interest-relative invariantism (IRI).7 The details of a proposal of this kind will naturally depend on just what kinds of pragmatic factor are deemed relevant â saliences seemed to be the germane factor for the issue about closure; but variation in stakes is what seems germane in the various scenarios in the Bank Case. IRI allows, apparently, that a pair of subjects may both truly believe that P on the basis of the same evidence or cognitive achievements yet one knows that P, and the other fails to know that P if they suitably differ in pragmatic respects. Iâll come back to this.
Second, there is the option of maintaining that the variability in truth-value of knowledge-ascriptions across the kinds of situation illustrated is actually a product of variation in content. The specific version of this proposal made by DeRose and Cohen is standardly termed ascriber contextualism (henceforward simply âcontextualismâ). In its original and basic form, this view holds that the (level of) cognitive achievement that is required of X by the truth of an utterance of âX knows that Pâ varies as a function of pragmatic aspects â needs, stakes, saliences â of the speaker. Thus, in an example like the Bank Case, variation in pragmatic aspects of a self-ascriber across actual, or hypothetical, cases may result in (actual or hypothetical) tokenings of âI know that Pâ demanding different â more or less exigent â levels of cognitive achievement if they are to count as true. The truth-conditions, hence content, of tokens of such an ascription can vary, even though the only differences in their respective contexts of utterance pertain to the situation of the speaker in purely pragmatic respects.
The third option â that of knowledge-relativism, fashioned on the model of assessment-sensitivity as developed by John MacFarlane8 â shifts the location of the pragmatic factors once again, this time to anyone who evaluates a knowledge-ascription, whether or not they are its original author. So a single token of âX knows that Pâ may properly be assigned different truth-values in differing contexts of assessment, whether or not distinct assessors are involved, depending on the situation in pragmatic respects of the assessor. Thus, Ashley may again quite correctly return different verdicts on a self-ascription of knowledge that the bank will open on the Saturday in the two contexts described. A smooth account of Humeâs observation is likewise in prospect if the knowledge relativist can make a convincing case that travel between the philosophical study and the pub is apt to change the context of assessment in some relevant respect; a relativistic treatment of lottery cases will require a similar story concerning the potential effects of explicit consideration of certain of a statementâs consequences. But I shall not here consider in any detail how such an account might run.
It will not have escaped the attention of the alert reader that the three types of variabilist views distinguished exhibit disagreement in two dimensions. Agreeing that the truth-value of a knowledge-ascription may vary as an effect of variation in non-traditional pragmatic factors, they disagree about the location â subject, ascriber or assessor â of the relevant factors; but they also disagree about the semantic significance of such variation. For both knowledge-relativism and interest-relative invariantism, variation in pragmatic factors is of no semantic significance at all;...