Sunday, January 18, 2009

Concessive knowledge attributions

I've just read Dougherty and Rysiew's piece on concessive knowledge attributions in PPR (here). They offer a pragmatic explanation as to why CKA's sound contradictory. Here's a CKA:

(1) I know that Harry is a zebra, but he might be a painted mule.

Why are CKA's interesting? Because there's really nothing we know that we believe on the basis of evidence that entails that what we believe is true. The worry is that it seems like (1) expresses the basic fallibilist idea, and yet it seems contradictory. If claims akin to (1) are uniformly false, the sceptic wins. In response to a challenge of Stanley's, D&R suggest that the truth-conditions for claims about epistemic possibility ought to be understood in terms of a subject's evidence rather than knowledge. On their view, 'q is epistemically possible for S' is true iff 'not-q is not entailed by S's evidence'.

Here's a worry. On their view, (1) can turn out to be true because what (1) amounts to is the claim that:

(1') I know that Harry is a zebra, but it is consistent with my evidence that Harry is a cleverly disguised mule.

The advantage of their view over, say, Stanley's view is that on Stanley's view 'q is epistemically possible for S' is true iff 'not-q is not obviously inconsistent with something S knows'. Thus, their view is an advance over Stanley's view only if (1') does not entail:

(1'') I know that Harry is a zebra, but it is consistent with what I know that Harry is a cleverly disguised mule.

(1') logically entails (1'') if S's having knowledge that p is sufficient for p's inclusion in S's evidence. They want (1'') to turn out false, but they get that only by rejecting a pretty plausible claim about the relationship between evidence and knowledge. (If knowledge of a proposition's truth were sufficient for that proposition's inclusion in the subject's evidence, then because the subject knows that Harry is a zebra the proposition that Harry is a zebra would be included in Harry's evidence. That proposition quite obviously entails that Harry is not some non-zebra cleverly disguised as a zebra.)

Now, I'm somewhat ambivalent towards K entails E because it implies that deduction is a way of acquiring evidence. I don't quite buy that. But, if we restrict evidence to that which is known directly, we can still generate the problem. Take any proposition we know directly through observation (e.g., that I have hands). It seems that the fallibilist will say that I can know this but will also concede that it might be that I'm mistaken about this. If non-inferential knowledge of this proposition's truth suffices for the proposition's inclusion in my evidence, D&R's account cannot explain how it could be true that:

(2) I know I have a hand but it might be an amazing prosthetic.

Oops. In a footnote, D&R sort of address the worry I raise. They write:
Just as EPk per se is neutral about what knowledge is, EPev itself is neutral as to what counts as evidence—whether it includes, or is restricted to, one’s experiences and (apparent) memories, one’s beliefs, and so on. For the same reason, EPev doesn’t require a specifically internalist handling of evidence. Maintaining the distinction between EPk and EPev does require rejecting Williamson’s (2000) principle that ‘E=K’—i.e., that one’s evidence is one’s knowledge. But that principle is sufficiently controversial that most would join us in assuming that the distinction between EPk and EPev is real.

I don't think this goes far enough. You can reject E = K on the grounds that knowledge of p's truth is not necessary for p's inclusion in S's evidence. You can raise the problem I raise if you think that non-inferential knowledge of p's truth is sufficient for p's inclusion in S's evidence. And while E = K might be controversial, I don't think that the principle I've appealed to is nearly as controversial. Yes, there are people who think that evidence is not propositional. But, that is not something D&R are assuming in their discussion. So far as I can tell, to deny that non-inferential knowledge suffices for a proposition's inclusion in someone's evidence is tantamount to denying that this way of acquiring knowledge is a way of acquiring evidence. And, that calls into question the very idea that the way of acquiring knowledge truly is a source of knowledge. On its face, a necessary condition on some source being a source of knowledge is that it is a source of evidence. So, they don't ignore completely the worry, but I don't think they've addressed it satisfactorily.


Andrew Moon said...

This is an interesting post, and I read the R&D paper just yesterday. I would want to say, if I were R&D, that statements of

i) I know that p, but it's possible that ~p

could be read

i') I know that p, but it's consistent with my evidence (for p) that ~p.

The problem is that I'm not sure how to revise D&R's official definition of epistemic possibility; maybe that can't be done.

Clayton said...


I think that's a good point. I was thinking that one way to go would be to take care to distinguish:

(a) whether it is consistent with S's evidence that ~p.
(b) whether it is consistent with S's evidence _for p_ that ~p.

(I rely on this sort of distinction in the PQ piece I wrote a while back and suggested that this distinction matters to someone like Williamson because of the apparent commitment he has to infallibilism.) One worry I have about your suggestion is this. It seems that your (i') will almost always come out true. That's good. However, it seems that while (i') is a sensible thing to say it never seems sensible to say things like 'I know p, but I might be mistaken', and the question is why this second claim sounds so bad when your claim seems perfectly harmless.

Andrew Moon said...

Hmm, suppose I say,

i') "I know that p, but it's consistent with my evidence (for p) that ~p."

And my interlocutor says, "oh, so you might be wrong about p?"

It seems that I would have to say yes. What is entailed (or conversationally implied, if D&R are right) seems to be the same.

I think that (i') might not sound as bad because we're not as used to using it in ordinary language as "I know that p but I might be wrong". Here's a more ordinary case where we might actually talk that way:

Astronomer talking: "I know that there's an asteroid that passed by Pluto last week, but it's consistent with my evidence for this claim that there was no asteroid that passed by Pluto."

That seems pretty bad to me.

Btw, what did you think of D&R's example "Well it's possible that we're still going to win, but c'mon, we all know that we're going to lose." Did you find that example convincing? (as a concessive knowledge attribution that doesn't sound weird or contradictory?)

Clayton said...

Hey Andrew,
The example you give at the end strikes me as very strange. The only way I can hear that as not being contradictory is if I understand the 'possible' as something other than epistemic possibility.