Undercutting Defeaters and Levels Confusions

Suppose you’ve read Bill Alston’s illuminating piece on levels confusions in epistemology, and agree that the skeptic can’t win the argument by going up a level. That is, where the argument is about a claim to know that p, the skeptic is entitled to ask about one’s basis for p–that’s fair game. But when the answer is that e is one’s evidence for p, you can’t undermine the claim to knowledge by insisting that the person defend that e is evidence for p. That is, it is not a requirement for knowing that p on the basis of e that one knows that e is good evidence for p, or that one can defend the claim that e is good evidence for p, etc. That’s to be guilty of a levels confusion.

Then turn to Pollock on defeaters. According to John, there are two basic kinds, rebutting and undercutting defeaters. Rebutting defeaters are evidence against p, where the belief that p is the belief to be evaluated. Undercutting defeaters, however, are reasons for thinking that the basis of the belief, e, is not a truth indicator of p.

This characterization takes a metalevel claim to be relevant to the object level issue of which beliefs are justified. So, for those convinced by the Alston piece, there is some tension.

For those of us who find John’s position pretty compelling, we now have to say that the argument against skepticism provided by Alston is no longer decisive. All we could claim, if we still find something useful in Alston’s arguments, is that sometimes a levels jump is not relevant to the question of whether one knows or is justified–not that it never is. So to respond to the skeptical challenge of the reasons we have for thinking that our evidence is really truth-indicating, we can only say that it is not obvious that we need such a defense in order to know.

We are left, then, with the following. If we had a defense of the truth indicator claim, the skeptic’s challenge would be met. If the skeptic had a defense that the evidence wasn’t truth-indicating, the skeptic would win in virtue of being able to present us with an undercutting defeater. In the normal case, however, we are stuck between these two positions. Alston’s arguments make us wary of levels confusions, but we are not willing to deny the force of undercutting defeaters. What would be nice is to be in either one of the following positions: either we have good reason to believe that our evidence is truth indicating, or the skeptic can provide good reason that our evidence is not truth indicating.

Instead, we are normally in the following position. When we have reason to think the truth-indicator claim false, we have an undercutting defeater. But in the normal case, we have no reason either way. So it looks like the appropriate epistemic attitude to take is that of withholding: we shouldn’t believe that our evidence is truth-indicating and we shouldn’t believe that it isn’t.

I hope that strikes you as an uncomfortable position. I’m inclined to think it needs to be avoided; that we need an account of justification so that if you’re justified in believing a claim, you have good reason to think that your evidence is truth-indicating. If that’s right, however, then we need a broader conception of undercutting defeaters: not only claims that give reason to think our evidence is not truth indicating count, but so do claims that render withholding the proper attitude to take toward whether our evidence is truth-indicating.

This inclination is troubling in itself, however, for notice that such a characterization of undercutting defeaters jumps another level: if rebutters are object-level defeaters, and Pollock’s undercutters are second-level defeaters, the new account would make room for third-level defeaters. And once we’ve gone this far, it’s pretty easy to see how to motivate a fourth-level account as well, and so on (you just have to remind yourself of the discomfort of having reasons to withhold at any particular level). Can the concept of defeat really be that complicated?


Undercutting Defeaters and Levels Confusions — 11 Comments

  1. *My view*: For your blf B to be justified, it seems that you can’t think B is currently held in an untrustworthy way (or solely on the basis of misleading evidence) since that gives you an undercutting defeater for it. But B can be justified even if you don’t have the additional blf that “B is formed in a trustworthy way” or “B is not held on the basis of misleading evidence”.
    *Problem*: What do you do when pressed by a skeptic who suggests that B is formed in an unreliable way and who asks you what you think about whether your blf B is formed in an unreliable way? It seems your only choices are: (i) blv that it isn’t formed in an untrustworthy way, (ii) blv that it is formed in an untrustworthy way, (iii) withhold on whether it is formed in an untrustworthy way, or (iv) none of the above. (ii) and (iii) seem to be defeaters and (iv) seems to be sticking your head in the sand which seems philosophically irresponsible. So you seem forced to take option (i), which suggests there is a higher-level requirement on justification. What to do?
    *Solution*: Read my paper “Defeaters and Higher-Level Requirements” just published in The Philosophical Quarterly. Yes, I know: shameless self-promotion.

  2. Mike, thanks for the reference–I’ve been thinking about this ever since reading Rich Feldman’s latest piece, and finally decided to post on it. Looks like good timing!

  3. Pollock’s view of undercutting defeat is (I’m pretty sure) mistaken. Stephen Wykstra has presented a very similar view that has unfortunately caught on. According to Pollock, if I’ve got a (good) reason R to believe it is improbable that (~Q []-> ~P) then R undercuts evidence P for belief Q. For instance, if I have reason to believe it’s improbable that “if x were not red then x would not look red” then “x’s looking red” is defeated as evidence for “x’s being red”. Same argument, other contexts. It’s improbable that: (1) were there goods outweighing evil then there would appear to be goods outweighing evil (:./ the appearence of no such goods is defeated as evidence that there are no such goods) (2) were someone not a saint, he would have an observable flaw (:./ observing no flaw in x is defeated as evidence that x is a saint). All of these arguments are bad. Evidence is not undercut in this way. Quick counterexample.

    Suppose there are very few saints and that everyone who is a saint leads a good life and has no observable flaws. And suppose that all who are non-saints lead less-than-good lives and have observable flaws. But suppose saints (that is, those who actually are saints) are such that they would have no observable flaws were they to lead less-than-good lives. I suppose we can imagine that saints would be good at covering their tracks, ducking responsibility, shifting the blame to others and so on. The principle that Pollock et. al. defend is that P is evidence for H only if it is not improbable that were H not the case then P would not be the case. This is in principle (C).

    C. [Pr(H /P & k) > Pr(H /k)] only if [Pr(~H []-> ~P) > .5]

    Suppose we select a person (Peter) at random and the following are true.

    Let H = Peter led a good life.
    Let P = We observed no flaw in the Peter’s life.
    Let k = Tautology.

    Assume we have made our observation P. So it is true of this selected person Peter that we have observed no flaw in his life. It follows that,

    1. Pr(H / P & k) > Pr(H/ k)

    since P raises the probability of H to 1 (recall, saints are few). So the antecedent of thesis (C) is true But it is also true that,

    2. Pr(~H []–> ~P) < .5 (2) states that it is improbable that had Peter led a less-than-good life then we would have observed a flaw in his life. Since (2) is true, it follows that the consequent of thesis (C) is false. The assumption in (2) that Peter did not lead a good life is counterfactual. Peter, we know, did lead a good life, since P is true and Pr(H/P & k) = 1. But since everyone who actually leads a good life is a saint (recall, all saints lead good lives and all non-saints lead less-than-good lives. From this it follows that anyone who leads a good life is a saint) we know that Peter is a saint. If Peter is a saint, then (2) must be true. Recall that were a saint to lead less-than-good life we would observe no flaw. So the probability that we would observe a flaw were Peter less-than-good is low. And if (1) and (2) are true, then the thesis in (C) is false. Therefore (C) is false. So evidence is not undercut in the way that Pollock and Wykstra suggest. It might well be true that had there been outweiging goods for actual evil, those goods would not be observable. That _does not_ entail that the observation of no outweighing goods is not evidence that there are no such goods (I wish skeptical theists would stop saying that). A. Gibbard suggested a similar informal counterexample, B. van Fraassen offered a very interesting counterexample whose details I unfortunately can't remember. J.H. Sobel, too.

  4. Last attempt.

    2. Pr(~H []-> ~P) is less than .5

    In English,

    2. The probability that [were it true that ~H then it would be true that ~P] is less than .5.

  5. Mike, nice example, and you’re right here, I think. The lesson, as I see it, is that one shouldn’t try to operationalize the notion of a truth indicator in terms of a counterfactual, which is why I put the problem in terms of truth indication rather than in terms of a counterfactual. It still amazes me, this long after we should have learned to be wary of counterfactual accounts, how tempting they remain…

  6. Jon,

    At the end of your first post on this, you mention a problem about ever higher levels of defeaters. I’m not sure what the problem is. In principle, there can be information at those levels. And it can affect what it is reasonable for you to believe. But I’m not seeing anything suggesting that justification requires some infinitely complex belief, or any other kind of troublesome regress here.

    More shameless self-promotion: I’ve written a long paper about this sort of thing. Jon’s post pretty well sums up what it took me a lot of pages to say. I hadn’t read (or at least didn’t remember if I did read) Mike Bergmann’s paper, so my paper does not take his stuff into account.

  7. Rich, here’s the higher order worry. Suppose you get talked into the view that you can’t remain neutral on whether your evidence is truth-indicative (at least not once the skeptic raises the question). And you are uncomfortable with the idea that justification can remain in the face of holding that your evidence isn’t truth-indicative. So, you have to commit: you now hold that your evidence is truth-indicative, and this is required in the circumstances for your evidence to justify your belief.

    Then consider this possibility that a skeptic might suggest: there’s information that suggests that withholding on this truth indication point is the proper doxastic attitude to take. That information isn’t a rebutter and it isn’t an undercutter. So we’ve got another kind of defeater to worry about. You’ve got to block, epistemically, this further information if your first-order belief is going to be justified in these circumstances (the one where you’ve admitted that you have to believe that your evidence is truth-indicative). Then the same dilemma arises as arose earlier: once the skeptic asks about this latest information, you’ll need to hold that there’s no all-things-considered case for withholding here, even if there is some information that might suggest it.

    Notice that, once the skeptic gets this going, s/he can always add another level of possible information suggesting withholding regarding the latest doxastic commitment you’ve made. If the commitment was an n-level commitment, the question asked by the skeptic will be about the existence of an n+1-level defeater. There’s no stopping point here, as far as I can see. There’s no vicious regress either, since (I’m assuming) the possibility of the skeptic raising this question doesn’t have the same epistemic force as the skeptic actually raising this question. But it made me unhappy to think about this infinite hierarchy of levels of defeaters, unhappy enough that I was hoping to find a way out. So I should read your paper and Mike’s as well!

  8. I thought the level confusions Alston counseled against have to do with mistaking questions about whether or not one has evidence for H (lower-level questions) with questions about whether or not one knows (or justifiably believes, et cetera) H (higher-order, epistemic questions). If so, then it is not a level confusion in Alston’s sense to make knowing (…) that E is evidence for the truth of H a condition on knowing (…) H, since this does not require that one know (…) that one knows H or the evidence for H. In short: it may be a condition on knowing (…) H that one know (…) that one has good evidence for H even though it is not a condition on knowing (…) H that one know that one knows that H or know that one knows that one’s evidence is good. So it is a good open question whether one must know (…) that one’s evidence for H is good in order to know (…) H, but this cannot be solved simply by saying that a positive answer is due to a level confusion.

  9. Andy, maybe I need to go re-read Alston, but the classic example I remember of the confusion is when the skeptic asks you how you know p, and you reply, “q”. The level jump occurs when the skeptic then asks you how you know that q is a good enough justification for p. As I read your description, there is no object-level/meta-level difference–or am I misunderstanding?

  10. Jon, I think there is an ambiguity in the question “how do you know that q is good enough justification for p?” On one natural reading this jumps levels: the assumption is that justification requires that one show that one has satisfied a necessary condition of justification. On another reading, however, it might simply express the requirement that one be aware (be justified in believing, know) that q is evidence for p. Since one can be aware that q is evidence for p without being aware that one has satisfied a condition on justification or knowledge (indeed, one need not even have these normative, epistemic concepts) there is no level confusion on the second reading.

Leave a Reply

Your email address will not be published. Required fields are marked *