Monday, 3 March 2003

A Puzzle about Justification

This is an MA essay which peters out a bit at the end. It's basically part 1, with part 2 to consist of a positive solution to the puzzle which provides a positive account of the right way to approach matters epistemic, expanding on the last para or two. So that's all right then.

I have since discovered Susan Haack, whose 'foundherentism' seems on the right track in understanding the relationship between evidence and belief. I also keep meaning to revisit Morton White's 'corporatism'...

**If anyone finds the essay makes a useful contribution to academic or published work, please cite it at this location. Comments are welcome too, and will be answered.**

For easier reading, a Word version is available here.

A Puzzle about Justification

The puzzle

A paradox presented by Harman and attributed to Kripke some 30 years ago is still attracting attention in the literature:

If I know that h is true, I know that any evidence that seems to tell 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 to disregard any evidence.[1]

In this essay, I examine the three main lines of response to the puzzle, which I label the externalist, evidentialist and contextualist approaches. I conclude that none of these three approaches provides an adequate response to the puzzle. I then present some considerations which I believe can form the basis of an acceptable response to the puzzle.

In order to clarify our discussion of the puzzle, I shall attempt to explicate and formalise the argument.

[Tf =df f is true;

Mx = df x is misleading;

xEAf = df x is evidence against f;

Dax = df a disregards x; Kx(f) = df x knows that f;

Ox(f) = df f is epistemically obligatory for x (‘should’);

Px(f) = df f is epistemically permissable for x (‘in a position to’)].

1 Ka(Th) I know that h is true

The next step is redundant:

R Ka (x [xEAh f{Tf & xEAf}]) I know that any evidence against h is evidence against something that is true

However, for completeness, we make explicit a further premise:

2 Ka(xf [{Tf & xEAf} Mx]) I know that all evidence against something that is true is misleading

3 Ka (x [xEAh Mx]) I know that all evidence against h is misleading (from 1,2)

To avoid problems which might be raised in respect of knowing that one knows, we remove the knowledge operator from the next premise:

4 x (Mx Oa[Dax]) I should disregard evidence that I know is misleading

Now another suppressed premise:

5 x(Ox[Dax] Px[Dax]) If one should disregard e then one is in a position to disregard e

6 x(xEAh Pa[Dax]) I am in a position to disregard any evidence against h (from 3,5)

We now add the sentence which highlights the paradoxical nature of the argument:

(7 xy(xEAy ~Pa[Dax]) I am never in a position to disregard any evidence)

This formulation of the argument will suffice as a starting point for our investigation. Infelicities, objections and revisions will be dealt with as they arise. One comment is in order at this stage, though. The puzzle is cast in the indicative mood, and I intend to leave it that way. The effects of introducing subjunctive conditionals supporting counterfactuals cannot be investigated here, save to point out that the inference from 1 and 2 to 3 would be blocked since 1 is concerned with actuality, whereas the antecedents in 2 and 3 would be concerned with counterfactual situations. The truth of 1 cannot be assumed constant across actual and counterfactual situations.

The externalist approach

A recent response from Duncan Pritchard claims that the material mode and the vacuous truth of 6 combine to make 6 and 7 consistent. 6 is vacuously true on this account, because its antecedent is, as a matter of fact, never satisified. Therefore it cannot (will not) generate a counterexample to 7.

Pritchard suggest that the puzzle rests on a misconstrual of the post-Gettier notion of warrant. Warrant proper is taken to be ‘whatever is sufficient (presumably minimally so) for converting true belief to knowledge’. I shall follow this usage throughout the essay, using the term ‘justification’ to mean purely internalistic justification (this notion will be explicated below).

Pritchard casts the paradox in terms of ‘warranted’ belief, rather than knowledge, and suggests that the paradox is guilty of having overlooked the consequences of Gettier-style counterexamples for our understanding of warrant.

Since any post-Gettier account of knowledge involves, as we saw above, an extra external component, so it follows that in possessing this knowledge the agent must meet the external condition (however it is to be specified) that this external component demands. This will mean that his belief is not only true and (internalistically) justified, but that it is also, in some sense, appropriately related to the truth.

If that is so, however, then how could it be that there exists misleading evidence that our hero should be taking into account? For insofar as there is such evidence, then the subject could not have met the external condition for his belief in the first place because he would not have been appropriately related to the truth.

…it is irrelevant to oppose this agument on the grounds that new evidence could come to light which would be wrongly dismissed by our protagonist. If the new evidence is indeed relevant to whether or not the agent knows, then it will suffice to destroy his knowledge of the antecedent proposition in the first place, and thus his knowledge of the consequent proposition as well.[2]

Pritchard claims that if one is warranted in believing p (and p is true), then for any evidence, either

(a) it is not evidence against p (and therefore not misleading evidence against p), or

(b) one should not (or need not) take it into account (and one may therefore disregard it.)

This seems little more than counterassertion. Pritchard has done nothing to show that being appropriately related to the facts (the external part of warrant) guarantees that the stated disjunction holds. In fact it is not even plausible that it does[3]. (a) suggests either

(a1) an unpalatably strict view of warrant: I am warranted in believing p only if no evidence (that I will encounter) exists which seems to tell against p; or

(a2) an objective conception of the evidence relation, according to which nothing can be evidence against a truth.

Neither of these is argued for, or is particularly appealing. In particular, (a1) (in the absence of (a2)) would imply that all evidence one will subsequently encounter plays a part in determining whether one has warrant for a belief – so if in ten years time I will see an incorrect newspaper report which contradicts what I now believe, then ipso facto the belief is now ‘not properly related to the facts’.

(a2) is of little dialectical use, since it implies that some of the things that seem to be evidence to me aren’t really evidence at all: if I toss a biassed coin 100 000 times and obtain a 500 032 : 499 968 heads : tails distribution, then not only do I have no non-misleading evidence for the coin’s being true, but I have no evidence for it at all. But in any case, I have no way to distinguish real evidence from this ‘fake’ evidence, so the puzzle may be reasserted referring to apparent evidence rather than evidence simpliciter (Harman’s wording does mention evidence which ‘seems to tell against’ h.)

(b) seems to hold only if we hold an objectivist view of the ‘should’ of epistemic duty. That is, what I should disregard is determined merely by what is true, independently of what I may think about the matter. I think this is the most plausible account of what Pritchard has in mind. By endorsing a (partly) externalist view of warrant, he hopes to rule out the appeal to an internalist view of epistemic requirements. This ploy does not succeed, because Pritchard does not put forward a case for rejecting an internalist (quasi-deontological) view of epistemic requirements. And if the ‘ought’ of epistemic ‘duty’ is to discharge a recommendatory as well as an ex post facto evaluative function as any ‘ought’ ought, it cannot merely enjoin us ‘believe what is true, regardless of the evidence’ – if that were all there were to internalistic rules, justification would be reduced to the truth condition.

Perhaps, if justification were irrelevant to knowledge, so that the only ‘internalist’ element were belief, then one might claim that the paradox is revealed as of little interest to theoreticians of knowledge. This would not be a major problem for those of us interested in the theory of justification, and would not dissolve the paradox, which can be reformulated to apply to the internal perspective, without recourse to purely externalist definitions of knowledge.

Even more extremely, if I have (and need have) no reflectively accessible grounds for my belief, the subjective viewpoint drops out of knowledge-attributions altogether. If there is nothing I am (internalistically) justified or unjustified in believing, then why should there be any question of my being (internalistically) justified in disregarding evidence? Pritchard does not suggest reasons for thinking that there is no such thing as internalistic justification. Indeed he agrees that there is. This approach does not seem worth pursuing.

So the puzzle, we conclude, is concerned with justification, rather than knowledge proper. Since the epistemic counterparts of deontic operators are used in the argument, we may conclude that a purely ‘externalist’ reading of the puzzle is beside the point.

Pritchard is aware that the puzzle can be reformulated in terms of justification. He therefore proposes an alternative answer to the problem, reformulated in terms of justification. This can be achieved by removing all the K-operators from our formulation of the puzzle, and regarding the resulting set of sentences as purporting to lay out an argument which could justifiably be worked through by an epistemic subject. Call the resulting steps J1 – J7.

The externalist’s second horn: opposing justifications do not generate inconsistency

The alternative is most plausibly seen as denying J4: that one should disregard misleading evidence. Pritchard suggests that there is no requirement to disregard misleading evidence, since one may be justified with respect to both of a pair of contradictory propositions:

...the possession of justification for one’s belief in an empirical proposition, P, merely involves the possession of reflectively accessible grounds for that belief. Crucially, however, it need not involve the possession of sufficient reflectively accessible grounds for belief, where the sufficiency resides in the fact that, so long as the belief is true, then one knows the proposition in question. [i.e. the fact that justification is sufficient for warrant] In other words, whatever other properties justification has, it is not a ‘threshhold’ concept. Instead, it admits of degrees and is compatible with ‘opposing’ justifications…

On this basis, Pritchard concludes that if applied merely to the case of justification, the paradox would not arise: Since I can have opposing justifications, then J3 is acceptable mutatis mutandis,

and so are the remaining steps (given that…) but so is taking into account opposing justifications. So J7 is not contradicted, and the puzzle disappears.

Pritchard argues that justification is not a threshold concept, on the grounds that it is not sufficient for warrant. He also argues that since it is not a threshold concept, it is possible to hold conflicting justifications, and that justification is always a matter of degree. This appears to me to be a non sequitur.

I find it less awkward to speak of threshhold properties rather than threshhold concepts – this will however not affect the argument. There are two meanings for threshhold concept and therefore for threshhold property in play here. One meaning (threshhold1) is that of a property which does not admit of degrees. In this sense, justification is a threshhold concept iff necessarily, for any belief, that belief is either fully justified or not justified at all. The second sense, a threshhold2 property is one possession of which by a belief is sufficient for that belief’s being warranted.

His attempted dissolution of the paradox takes as a premise that justification (understood as purely internalist justification) is not sufficient, even given true belief, for knowledge. This is regarded as having been adequately settled by Gettier-style counterexamples. From this, Pritchard concludes warrant must involve an external component, and from there, justification is not a threshhold2 concept. Equivocating on the two meanings of ‘threshhold’, he concludes that it is possible to have conflicting justifications (i.e. that justification is a threshhold1 concept.) from this, he concludes that the puzzle will not arise

We may notice that if justification is a non-threshhold1 property then it must also be a non-threshhold2 property. Since warrant is not something which admits of degrees, a property mere possession (in no specified degree) of which is sufficient for warrant ought not to be either. However, the entailment does not hold in the reverse direction: it is quite possible to be a non-threshhold2 property without being a non-threshhold1 property. Being dead is a reasonably good example of a non-threshhold2 property which is a threshhold1 property. So is being justified-all-things-considered, which for all Pritchard has shown might be necessary for warrant, without being sufficient.

In fact Pritchard has stated that warrant implies justification (see quote above). We only need to add that the justification required is overall justification to some specifiable standard (the threshold) to expose the missing ground between threshhold1 and threshhold2 properties. If on the relevant sense of ‘justified’ being justified with respect to a proposition implies being justified, all things considered, in believing that proposition, then Pritchard’s attempt to reconcile opoosed justifications fails.Since the defeater is supposed to undermine my justification, thus bringing it below the threshhold standard, it is incompatible with that justification, and something has got to give.

The puzzle remains. Although Pritchard may claim that I may be overall justified in believing h, but only prima facie justified in believing e (the supposed defeater of justification), and that therefore I am not epistemically obliged to disregard e, he does not show why the prima facie justification for e fails to combine with one’s existing prima facie justifications to defeat one’s previous justification for h. We therefore conclude that Pritchard’s argument fails to achieve its stated aim, and turn to an approach which is concerned with internalist justification.

The evidentialist approach

This approach accepts that J3, (and indeed we may suppose, J6), can be inferred, but to deny that J3 or J6) can subsequently justify actually disregarding any specific evidence. This approach is common to Harman, Lewis (though his approach adds a significant twist to the position), Ginet, Sorenson and Conee[4] (whose term ‘evidentialist’ is).

As Harman puts his response (using the example of a witness to a theft whose justification for believing Tom the thief is subsequently undermined by evidence that Tom has a kleptomaniacal twin):

The puzzle overlooks the way actually having evidence can change things. 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 Tom stole the book; If I no longer know that, I no longer know that the new evidence is misleading.[5]

This fallibilist position, though, does not seem to have explanatory weight. Harman seems merely to assert that J7 defeats J6, without showing how J6 gained its apparenly acceptable status. He redescribes the problem, without solving it. Cargile( ), whose approach we will examine later, comments on this style of response:

..our h-knower knows as well as we do about his ‘knowledge’ that all cases that will appear to him as counting against h, that he must drop this ‘knowledge’ the minute he encounters a candidate for an instance of it! Some inference ticket this generalisation (J3) is! We know it but as soon as we try to use it, the knowledge disappears![6]

So the evidentialist owes us an account of how it is that this knowledge can fail so signally to be of any use whatsoever.

Ginet seeks to assimilate instances of the generalisation J3 (rendered ‘If there is evidence against p, then it is misleading evidence’) to the category of conditionals (remember we are here talking of ordinary conditionals in the indicative mood) which are defective in the sense that they fail the acid test of a conditional: supporting modus ponens.

An example given is: ‘If there was such a meeting , then I’ll be forever ignorant of it.’ This example is held by Ginet[7] to be a form of Moorean pragmatic contradiction, in that assertion of the conditional is incompatible with assertion of its antecedent. But as Sorenson[8] points out, this is only an analogue of instances of J3 (‘This is evidence against p, therefore it is misleading’if we suppose that one is actually misled by the evidence. The antecedent of the conditionals Ginet discusses contradicts what is held to be known. This is not the case with misleading evidence. Such evidence supports a conclusion which is ex hypothesi known to be false, but doesn’t contradict it. ‘P but I don’t believe p’ may be a pragmatic contradiction, but ‘p and I believe q which suggests not-p’ is not. More is required to show why this conditional fails to provide an inference-ticket.

Sorenson points out that Ginet’s position rests on a strong reading of ‘misleading’ and therefore of ‘disregarding’. To assimilate ‘if e is evidence against h, then e is misleading’ to the paradigm exemplified in ‘if there was such a meeting, I am ignorant of the fact’ depends on construing ‘e is evidence against h’ as meaning something like ‘e is conclusive evidence against h’ (here we see the truth in Pritchard’s account). This in turn allows Ginet to read ‘e is misleading’ as ‘e misleads me’, thus generating a Moore-style pragmatic paradox..

This observation is of some importance in connection with another proposed dissolution of the paradox; one which rests on considerations of subjective probability. Sorenson, alluding to work by Frank Jackson, proposes an analysis of the supposedly defective generalisation in terms of assertibility-conditions, and the notion of robustness:

The conditional ‘If Doug reports my car is not in the lot, then his report is misleading’ is not robust with respect to its antecedent. According to the background story, Doug is a reliable source. So it is unlikely that Doug would have reported my car is not in the lot, given that it is. The appropriate response to the surprise is to lower my confidence that the car is in the lot. Since the conditional is not robust, Jackson’s theory predicts that it lacks assertibility. For although the probability of the conditional is high, the conditional probability [probability of the consequent given the antecedent] is low. Since assertibility goes by conditional probability, we have an example of a highly probable yet unassertible proposition.[9]

The problem with this account is that the explanation of the modus ponens-resistence of instances of the conditional in terms of non-robustness relies on specific assignments of (subjective) probability. This is a problem, as it smacks of petitio. The consequent of the conditional is said to have low probability when conditionalised on the antecedent. But this assertion is too close to the explanandum phenomenon to have any explanatory clout. The paradox arises because it is never, according to 7, permissable to disregard evidence. And this is precisely what Sorenson asserts. In claiming that the consequent of the conditional has low probability given the antecedent, Sorenson is stating that the a correct assignment of personal probabilities will confer lower probability on the consequent if the antecedent is accepted. That is to say, the antecedent is (probabilistic) evidence against the consequent, and it is to be accepted despite this. This may describe his view of the phenomena, but it doesn’t explain the reasons behind them.

This may necome clearer if we consider more closely Sorenson’s argument (ignoring the use of ‘contrary’ which ought to read ‘countervailing’):

Asserting [A]‘Any contrary evidence is misleading’ is equivalent to asserting [I]‘If there is contrary evidence, then it is misleading’. This conditional is not robust. But asserting it signals that it is robust, and is therefore a conversational misdeed.[10]

Given that Sorenson is concerned with conversational implicatures, the use of truth-functional equivalence seems illegitimate. Sorenson moves (backwards) from non-assertibility of [I] to non-assertibility of [A] by way of what appears to be appeal to standard truth-functional equivalence. This move needs to be justified. It is not written on the face of [A] that it is equivalent in assertibility to [I]. Let’s do some logic.[ CEx = x is counter-evidence; MEx = x is misleading]

A: x (CEx MEx)

I: x(CEx) MEx

Notice that these aren’t even equivalent in truth-functional terms: x occurs free in the consequent of I.[11]

However, there is an equivalent to A which involves existential quantification:

I¢: ~ x(CEx & ~MEx)

But I¢ can't be claimed to be non-robust, because it is not clear what common rule of inference is not supported. Soprenson gives examples of disjunction which fail to be robust with respect to the denial of one of the disjuncts (e.g… ), but a conjunction has a unique truth-condition, therefore is not open to claims of non-robustness.

However, let us consider substitution instances of A. Even though we can permissably derive these, Sorenson may claim that they do suffer from robustness problems, if we accept Sorenson’s description of the problem. It is claimed that the conditional


does not support modus ponens. But by contraposition, we have:

~MEa ~CEa

which we may be more inclined to regard as robust. Given the antecedent, that the evidence (Doug’s report) is not misleading, can we infer that his report is not countervailing evidence? It seems we can. For we know that the proposition it would be evidence against is true, and that Doug is trustworthy, so we can conclude that his report is unlikely to be countervailing evidence. Here we see the effects of the probabilistic method – depending on which bits of information we assume known first, we arrive at different consequences. But a solution to the puzzle needs to explain why Sorenson’s order of knowing should be preferred to our reversed order.

Conee’s response is simple. He explains the priority of the new evidence over our old belief by appeal to an ‘evidentialist’ conception of justification and knowledge:

S is epistemically justified in believing at t exactly what is adequately supported by the totality of S’s evidence at t, and S knows at t only propositions that S is epistemically justified in believing at t.[12]

Conee’s task is to explain the failure of a known (justifiedly believed) proposition to survive the advent of new but inconclusive evidence. His account is foundationalist, in that it denies that a belief gained on the basis of evidence (data) can have justification independently of its basis in the evidence. The direction of justification is one way. Such a position appears at first glance to threaten much of our knowledge. Since (what we commonsensically know to be) sensory evidence is fleeting, very little will be justified for long. It seems we need to appeal to data in the form of memory if justification is to be more than transient. This memory-data should not, I contend, be confused with memory-belief. The former (memory of) would, I suspect have to be akin to a phenomenological trace of past sensation. Otherwise it is assimilated to the latter (memory that), which has no such distinct ‘givenness’. It is difficult to see how ‘memory that’ can be distinguished, on the evidentialist view, from any other belief. The classification of such a belief as memory is premised on its aetiology, not on currently discernable character. A past justification appears, on the evidentialist view canvassed by Conee, to be no justification at all. This view we may describe as ‘contemporarian’: A belief remains ‘parasitic’ on its evidential justification: at any time, beliefs are to be derived from the currently available evidence.

Harman[13] puts forward a related argument against a strong foundationalist view, claiming that such a view requires that one ‘keep track of’ one’s past justifications. He claims that it is undesirable to clutter one’s mind with such specific justifications, and moreover that in many cases we are not able to do so. The foundationalist on this account is committed to an implausible denail of justified status to a wide range of ordinary beliefs.

In his review of Harman’s 1986, Feldman disagrees:

[Harman] omits an obvious alternative[:] that one often acquires new justification for retained beliefs despite losing the original justification. I don’t now know my original Jn for my belief that Mario Cuomo is the governor of New York. But I do know that my beliefs about the identity of prominent politicians are generally right, that I have not recently been corrected after making remarks about the identity of the governor, etc. This sort of information can provide a current justification for many memory beliefs. [14]

This appeal to ‘meta-justification’ faces two problems.

One is that the meta-justification seems to have a coherentist character: the appeal to memory beliefs (memory-that, not memory-of) is underwritten by appeal to other beliefs. Harman’s point seems to be confirmed: that the original justification, based on contemporary support by evidence, has evaporated, and that what is left is a ‘memory-belief’, an explanation of whose persistence and survival gives it collateral support. Furthermore, there is no reason to think that the transition from contemporaneous justification by experience to the new belief-based support for propositional ‘memory’ is smooth and uninterrupted. The contemporarian foundationalist seems committed either to the implausible position that this does commonly occur, so that justification is preserved, or to the view that in such cases justification is indeed lost, at least in the short term.

So the contemporarian foundationalist requires the retention of memory-data to provide constant support for beliefs. But there is another problem which has similar costs in terms of both lack of plausibility and excessive demands on human information-processing resources: the constant dependence of belief on evidence seems to demand constant reappraisal of one’s total evidence. I can see no way for the evidentialist to motivate any mitigation of this consquence. The upshot is, in my submission, both undesirable and unrealistic: the loss of personal integrity. Since one’s personal identity over time is due in large part to stable and sincere beliefs and other mental states or dispositions, a requirement constantly to revise one’s beliefs seems to undermine integrity, not only in virtue of the instability of belief it implies, but also because espousing the evidentialist view restricts us to a pale and timid form of belief, a form which is constantly open to revision, and has a staus that is little more than provisional, little more than the self-consciously pragmatic startegey of Bratman’s[15] ‘acceptance in a context’, or Popperian conjecture[16].While mundane plans or instrumentalist science may be able to subsist on a diet of provisional assumption, but this kind of ‘misology’ applied to all belief seems quite unsupportable.

Conee’s position, then, faces two difficulties. First, the ‘bottom up’ foundationalist approach seems to require that I rebuild the edifice of knowledge on receiving any new piece of evidence. While in practice, thanks to the remarkable consistency of experience, the adjustments we make on receipt of new evidence are often minor, the evidentialist position requires that belief be apportioned to total evidence at a time. This is a stringent requirement: too stringent for the likes of us. Our experience (even that part of it which we consciously notice) is rich, and every part of it will standardly count as evidence for and against countless propositions: consider, for example, the Bayesian solution to the Ravens paradox. To constantly reappraise our beliefs in response to the onslaught of data would place an immense burden on our cognitive faculties.

Second, Conee’s insistence on contemporaneous support by evidence would appear to require ‘keeping track’ of evidence – indeed it would require actually keeping the evidence. But sensory evidence is transient, and believing only according to current sensory evidence is barely intelligible. Conee must admit at least a way of storing evidence: memory. It is not entirely clear that memory can take the place of sensory data in an evidentialist theory. Since evidence is given, it is not merely another belief: otherwise, why should we be forbidden to ‘disregard’ or reject it? But memory in many cases does take the form of belief: ‘memory that’.

So the evidentialist method of allowing one to infer J3 and J6 but denying that they are usable in inferences seems to have failed.

The contextualist approach.

Before considering Lewis’ contextualist variant of the evidentialists’ approach, we should consider a different ‘contextualist’ view; that of Cargile[17]. He denies that J3 is known at all. He does this on the basis of a denial of the principle of closure under known entailment. In fact for our purposes he should appeal to a stronger principle: denial of transfer of justification across entailment. Whereas I understand a principle of closure (I will use the example of knowledge) to assert that from known premises all logical consequences are also known, a principle of transmission of knowledge merely claims that knowledge can be obtained by performing deductive inference from known premises. The puzzle requires the latter, and in the form we are considering, only with respect to justification.

Any denial of closure or transmission, however, owes an account of why such an intuitive principle should fail. Cargile’s reply is to claim that closure (transmission of justification) is valid only within a given ‘issue-context’. He gives an example: I may be justified in believing Bill is present (I can see him). I also know ‘just because I am good at recognising trivial analytic truths’[18] that if he’s present then an identical twin is not standing in for him.

However, it is not correct to infer…that I know Bill is not being covered by an identical twin. This latter raises an issue for which I do not have adequate qualifications. Seeing Bill in the front row…is not an adequate method if the issue is whether he is being replaced by an identical twin…The move from the antecedent to the consequent, though trivially valid, is a change in perspective which introduces a new issue context.

However, in this example the conditional in question is a singular proposition, not a generalisation like 3. Since Cargile has chosen to deny that we know 3, he needs to provide a closer analogy. He anticipates this objection:

…the innocuous generalization (III) (‘I know that any evidence against h is misleading’) does not break the context in this way. And its not being applicable does not mean it is not known. You can know (IV) ‘All evidence which misleads me, misleads me’ in spite of knowing that you will not do well at recognising instances. This is a very reasonable objection…But IV is a tautology…Though seemingly very innocuous, what is claimed in III to be known is no tautology. My answer turns on assimilating it to the ‘context-breaking’ side of the ledger. But it is worth admitting that it is not initially obvious that it is unjustified. It is a misleading generalization about misleadingness.[19]

The idea seems to be that in considering the very idea of there being evidence against h, I change the ‘issue context’ in which knowledge of h is to be assessed. This seems unconvincing: since 2 is equivalent to:

2¢¢ Ka( f ~ x [{Tf & xEAf} & ~Mx]) I know that there is no non-misleading evidence against something true,

we can move from 1 and 2¢¢ directly to

3¢¢ Ka (~ x [xEAh & ~Mx]) I know that there is no non-misleading evidence against h.

3¢¢ makes it more explicit (perhaps) that we are not countenancing the possibility of any evidence against h (psst! except misleading evidence…). I can’t really see how else to put this, but it just seems implausible that a generalisation which mentions no particular possible source of (future) error should alter the issue context within which my knowledge is to be assessed. Cargile’s suggestion is that if I recognise that I might be fooled in specific circumstances in the future, I must acknowledge this fact, but that doesn’t support his denial of closure or transmission of warrant. In fact if anything, it affirms a strong form of closure such that in knowing a generalisation I therefore consciously know and consider specific consequences of it, since in order not to know 3, I would have to be regarded as knowing some such consequences: e.g. ‘I know that any evidence consisting of all my colleagues telling me that not-h would be misleading evidence’.

Lewis[20] is in sympathy with this conclusion, and his contextualist response to the problem avoids this problem by taking the same approach as the evidentialists by accepting that 3 can be known, but denying that it can be used to derive specific instances. Lewis’s approach is a ‘relevant alternatives’ contextualism, where Cargile’s might be characterised as a ‘standard of justification’ contextualism.

This does not mean that Lewis fails to fall within the (partly) internalist camp. His view is partly internalist – that is, he is plausibly read as supposing internal justification to be necessary for knowledge, at least in cases of inferential knowledge.

First, his view is based around the subject’s evidence (the given), and is concerned with what is ruled out by that evidence. Such an evidence–based account seems to be unmotivated unless there is some importance is being placed on what is consciously accessible to the subject. So there is a minimal internalism involved here.[21]

Second, some of Lewis’s rules regarding which alternatives are relevant (and therefore need not be eliminated by the subject’s experience) have an internalist character; in particular, the two we shall be concerned with.

These are the Rule of Belief –

‘A possibility that the subject believes to obtain is not properly ignored…Neither is one that he ought to believe – one that evidence and arguments justify him in believing...[more precisely:] a possibility may not be properly ignored if the subject gives it, or ought to give it, a degree of belief that is sufficiently high, and high not just because the possibility is unspecific.’[22] [emphasis mine]

– and the Rule of Attention – ‘a possibility not ignored at all is ipso facto not properly ignored.’[23]

This second rule concerns whether the possibility in question is ignored by the evaluator of knowledge, but Lewis takes it that ‘The well-informed epistemologists who ask what S knows will have to attend to the possibilities that S thinks up’[24], so it will apply to all possibilities not ignored by the subject.

Both of these rules are dependent on internalist considerations; what the subject believes, what she ought to believe on the basis of evidence and arguments, and which possibilities she is attending to. It is the latter rule that Lewis appeals to in his solution to the puzzle:

At first it was stipulated that S knew, whence it followed that S was properly ignoring all possibilities of error. But as the story continues, it turns out there is evidence on offer which points to some particular possibility of error. Then, by the Rule of Attention, that possibility is no longer properly ignored, either by S himself or by we who are telling the story of S. The advent of that evidence destroys S’s knowledge, and thereby destroys S’s licence to ignore the evidence lest he be misled.[25]

This clearly suggests that a specific possibility of error must be acknowledged if the rule of attention is to operate. This explains why Lewis does not adopt Cargile’s solution, and claim (implausibly, as we have seen) that the generalisation 3 changes the context by bringing to attention the general possibility of error. As in the case of the rule of belief, we might suppose, Lewis does not want to include over-general possibilities. He does not accept that acknowledgement of the possibility that there is ‘evidence against’ p, in bringing to to my attention the (over-general) possibility that I am wrong, should destroy my knowledge of p.

If Lewis does indeed reject this, introducing a specificity rule like that applying to the rule of belief, then a lacuna opens in his treatment of the puzzle. For I may recognise that e is evidence against h, even though e suggests no possibility more specific than the denial of h. In such a case, then, the explicated Rule of Attention would not change the context, so I would not be prevented (by internalist factors) from knowing that the evidence in question was misleading.

For example, e may be purely probabilistic evidence against h, which suggests no alternative possibility more specific than not-h. Equally, a straightforward and unelaborated denial of h (or indeed mention of the mere possibility of not-h) would be insufficient to change the context in which I came to know h. In such a case, no alternative explanation is forthcoming. I merely have evidence against h. The Rule of Belief doesn’t work here, because (ex hypothesi) I don’t yet believe the proposition that not-h – and if the normative part of the Rule is used, (I ought to believe not-h), then we are back with Harman’s solution, which works no better for supporting an ungainly theoretical structure.

There is a further problem, which I believe is at the heart of the puzzle: The Rule of Attention only covers possibilities that are not ruled out by the evidence. Since the possibility is ruled out by knowledge of its falsity, then it would seem open to us to claim that it’s ruled out by my evidence, including propositional memory evidence. I do not wish to press this partuicular point against Lewis: it applies to all of the internalist theories I have considered, as we have seen in the case of Sorenson’s reversible conditional, and Conee’s unidirectional justification. The nature of evidence, misleadingness, and what it is to disregard evidence are matters which have received insufficient attention.

In all the approaches I have surveyed (with the exception of Pritchard), the notion of ‘disregarding’ evidence acquires an undeserved connotation of wilful ignorance . It is worth noting that evidence in the sense of data, the given, may not be ignored or denied. That much, while close to being a conceptual truth, is beside the point. The evidence relation is a rational, intelligible relation which has at least two relata: for convenience, I will somewhat inaccurately call them the probans and the probandum. The probans, taken as a concrete entity is evidence for many probanda, depending on background beliefs. Evidence is misleading with respect to some of its probanda, but not others. Jones’s testimony that it is not raining outside is misleading with respect to the weather, but not to his current location, his ability to speak, his Gricean attitudes, and countless other facts. Evidence conceived as data, then, is disregarded only relative to a specific issue. And this sense of ‘disregard’ does not imply ignoring; rather, it consists in rejection of the evidential import of the data as decisive vis-à-vis a given issue.

This is related to Pritchard’s argument from the so-called compatibility of opposed justifications: it points out that there is no need to do anything as obtuse as deny evidence in order to bring new experience in line with our existing beliefs. This is of course not a new point, but neither need we assume holism or strong coherentism to hold this view: we need only accept that some beliefs are always required to draw conclusions from basic evidence, e.g. sensory data, and that these beliefs must to some extent fight it out amongst themselves for control over our conclusions. Knowledge, I submit, must be regarded as stable. I hope at some time in the future to present a solution to the puzzle, but there is no space here. The approach I adopt is perhaps suggested by Bratman:

The context-relativity of acceptance, and the resulting context-relativity of decision, allow us to be sensitive to various practical pressures on how we set up our decision problems. But we also need to ensure continuity and coordination of our activities at one time, over time, and in the world as we find it. And that is a major role of our context-independent intentions and plans, taken together with our context-independent beliefs.[26]

Adler, Jonathan E. ‘Conservatism and Tacit Confirmation’ Mind 99(396), 1990

Alston, William P. ‘Epistemic Desiderata’ Philosophy and Phenomenological Research 53(3), 1993

‘What’s Wrong with Immediate Knowledge?’ Synthese 55, 1983

Audi, Robert ‘Deductive Closure, Defeasibility and Scepticism: A Reply to Feldman’ Philosophical Quarterly 45(181), 2001

Bonjour, Laurence ‘Haak on Justification and Experience’ Synthese 112, 1997

Bratman, Michael E. ‘Practical Reasoning and Acceptance in a Context’ Mind 101(401), 1992

Anderson, Terence Analysis of Evidence: How to do Things with Facts London: Weidenfeld and Nicolson, 1991

and Twining, William

Cargile, James ‘Justification and Misleading Defeaters’ Analysis 55(3), 1995

Casullo, Albert ‘The Coherence of Empiricism’ Pacific Philosophical Quarterly 81, 2000

Chisholm, Roderick M. ‘Firth and the Ethics of Belief’ Philosophy and Phenomenological Research 51(1), 1991

Christensen, David ‘Conservatism in Epistemology’ Noûs 28(1), 1994

Cocchiarella, Nino B. ‘Reference in Conceptual Realism’ Synthese 114, 1998

Code, Lorraine ‘Towards a `Responsibilist` Epistemology’ Philosophy and Phenomenological Research 45(1), 1984

Conee, Earl ‘Heeding Misleading Defeaters’ Philosophical Studies 103, 2001

Davidson, Donald ‘Actions, Reasons and Causes’ Journal of Philosophy 60(23), 1963

??DeRose, Keith ‘Relevant Alternatives and the Content of Knowledge Attributions’ Philosophy and Phenomenological Research 56(1), 1996

Evnine, Simon J. ‘Learning from one’s Mistakes: Epistemic Modesty and the Nature of Belief’ Pacific Philosophical Quarterly 82, 2001

Feldman, richard ‘Change in View: Principles of Reasoning’ Philosophical Review 98(4), 1989

‘In Defence of Closure’ Philosophical Quarterly 45(181), 1995

Ginet, Carl ‘Knowing Less by Knowing More’ Midwest Studies in Philosophy 5, 1980

Goldman, Alvin I. ‘Varieties of Cognitive Appraisal’ Noûs 13(1), 1979

Haak, Susan ‘Towards Reconstruction in Epistemology’ Synthese 112, 1997

‘Reply to Bonjour’ Synthese 112, 1997

Harman, Gilbert Thought Princeton: Princeton University Press, 1973

‘Reasoning and Evidence One does not Possess’ Midwest Studies in Philosophy 5, 1980

‘Inference to the Best Explanation’ in Pojman(ed.), 1998

Change in View: Principles of Reasoning Cambridge(MA): MIT Press, 1986

Herman, Barbara ‘On the Value of Acting from the Motive of Duty’ Philosophical Review 90(3), 1981

Howard-snyder, Daniel ‘On an `Unintelligible` Idea: Donald Davidson’s Case Against Experiential Foundationalism’ Southern Journal of Philosophy 2002

Katzoff, Charlotte ‘Counter-evidence and the Duty to Critically Reflect’ Analysis 60(1), 2000

Lewis, David ‘Elusive Knowledge’ Australasian Journal of Philosophy 74(4), 1996

‘Radical Interpretation’ Synthese 23, 1974

Monton, Bradley ‘Sleeping Beauty and the Forgetful Bayesian’ Analysis 62(1), 2002

Moser, Paul Human Knowledge(2nd ed.) New York: OUP, 1995

and vander Nat, Arnold

Nozick, Robert(ed.) Socratic Puzzles Cambridge(MA): Harvard University Press, 1997

Nozick, Robert The Nature of Rationality Princeton: Princeton University Press, 1993

‘Simplicity as Fallout’, in Nozick(ed.), 1997

‘Newcombe’s Problem and Two Principles of Choice’, in Nozick(ed.), 1997

Pearl, L. ‘Objective and Subjective Duty’ Mind 80(319), 1971

Plantinga, Alvin Warrant: the Current Debate New York: OUP, 1993

Pojman, Louis P.(ed.) Theory of Knowledge(2nd ed.) Belmont(CA): Wadsworth, 1998

Pollock, John ‘The Gettier Problem’ in Pojman(ed.), 1998

Popper, K. Conjectural Knowledge: My Solution to the Problem of Induction, Oxford: OUP, 1973

Pritchard, Duncan ‘A Puzzle about Warrant’ Philosophical Inquiry 23, 2001

Pryor, James ‘Highlights of Recent Epistemology’ British Journal for the Philosophy of Science 52, 2001

Sellars, Wilfred ‘Givenness and Explanatory Coherence’ Journal of Philosophy 70(18), 1973

Shope, Robert K. ‘Cognitive Abilities, Conditionals, and Knowledge: A Response to Nozick’ Journal of Philosophy 81(1), 1984

Sklar, Lawrence ‘Methodological Conservatism’ Philosophical Review 84(3), 1975

Sorenson, Roy A. ‘Dogmatism, Junk Knowledge and Conditionals’ Philosophical Quarterly 38(153), 1988

Sorrell, Tom ‘Harman’s Paradox’ Mind 90, 1981

Tidman, Paul ‘Critical Reflection: an Alleged Epistemic Duty’ Analysis 56(4), 1996

van Fraasen, Bas ‘Belief and the Will’ Journal of Philosophy 81, 1984

Vogel, Jonathan ‘Sklar on Methodological Conservatism’ Philosophy and Phenomenological Research 52(1), 1992

Williams, Michael ‘Contextualism, Externalism and Epistemic Standards’ Philosophical Studies 103, 2001

Zimmerman, Michael J. The Concept of Moral Obligation Cambridge: CUP, 1996

[1] Harman (1973), p.148

[2] Pritchard (2001)

[3] Pollock’s (1998) conception of objective justification, which involves aan externalist element; viz. undefeatedness of resons relative to the set of all truths, still allows that some subset of all truths might defeat justification; so if this subset were all that were subjectively available to me, my (internalist) justification would be defeated.

[4] Harman (1973), Ginet (1980), Sorenson (1988), Lewis (1996), Conee (2001).

[5] Harman (1973), p. 148-9

[6] Cargile (1995), p.216

[7] op. cit.

[8] op. cit.

[9] op.cit., p.444

[10] ibid., p. 445

[11] We might introduce an individual concept or a definite description to

ensure the co-referentiality of the two occurrences of the symbol x:

I: x(CEx & Xx) x(MEx & Xx)

But this loses the generality of A, having the effect of limiting the domain of x to a specified individual. One might as well resort to the individual quantifier, or a singular term (constant) with identity. In effect, we have a complicated way of presenting an instance of A, which is precisely what I claim Sorenson needs to do before the claim of non-robustness can get off the ground. But it is clear that an instance of A is not equivalent to A itself.

[12] Conee (2001), p.101

[13] Harman (1986),pp. 25ff.

[14] Feldman(1989), pp.554-5.

[15] Bratman (1992)

[16] Popper (1973) etc.

[17] Cargile, 1995.

[18] ibid., p219

[19] ibid., p220

[20] Lewis (1996)

[21] Even though Lewis claims:

The given does not consist of basic axioms to serve as premises in subsequent arguments. Rather it consists in a match between possibilities. When perceptual experience E (or memory) eliminates a possibility W, that is not because the propositional content of the experience conflicts with W…Rather, it is the existence of the experience that conflicts with W: W is a possibility in which the subject is not having experience E. Else we would need to tell some fishy story of how the experience has some sort of infallible, ineffable, purely phenomenal propositional content…[sic]Who needs that? Let E have propositional content P. suppose even – something I take to be an open question – that E is, in some sense, fully characterised by P. Then I say that E eliminates W iff W is a possibility in which the subject’s experience or memory has content different from P. I do not say that E eliminates W iff W is a possibility in which P is false. (p.553)

Now this appears to be a thoroughly externalist position. But we may note that Lewis’s account is designed to describe the conditions under which, in conversation, we may attribute knowledge to others. So we clearly have some conception of the character of a person’s evidence. We need not attribute propositional content to the concrete experience-state to note that we take it to signify some range of states of affairs – to be a sign that one of them obtains. That much is given in the necessary implications we regard as attaching to the existence of the ‘evidential’ states in question. And the eliminating relation between experiential states and centred worlds must be regarded as accessible to us, the attributors of knowledge. And if to us, then also to our fellow rational beings, including the holder of knowledge.

[22] ibid., p.555

[23] ibid., p.559

[24] ibid., p.561

[25] ibid., pp.564-5

[26] Bratman (1992), p.13.