Saturday, October 14, 2006


Pragmatic Encroachment and Scepticism

The world is still waiting for Hawthorne and Stanley's joint paper on the relationship between knowledge and practical reasoning, expanding on the interesting but overly-brief remarks each made in their respective books. In the meantime, I'm finally reading Fantl and McGrath's 'Evidence, Pragmatics, and Justification' from 2002's Philosophical Review. They provide an extended argument for the dependence of knowledge on pragmatic considerations, and in particular they argue for various principles of increasing strength relating knowledge to rational action.

I'm worried that the principle they end up with has the result that if you know that p, it's rational to take pretty much any bet on whether p (so long as the bet isn't such that you're worse off if you win too). Here's Fantl and McGrath's principle:

S is justified in believing that p only if, for all acts A, S is rational to do A, given p, iff S is rational to do A, in fact. (78)

(One quick note about terminology. By 'S is justified is believing that p' they mean 'that S has good enough evidence to know that p' (67fn). So this really is meant to be a necessary condition on knowledge, and I'll continue to talk about knowledge rather than justification.)

Let the bet be as follows. If p, then S wins $1. If ~p, then I win S's salary each month plus the deeds to S's house. I don't have an argument for this, but my intuitions are that S is rational to take the bet, given p, but probably not rational to take the bet in fact. And I'm willing to hold fast to these calls despite not having said anything about what p is. (If that seems implausible, just up the stakes of the bet even further, so that the consequences of losing are really really dire.) So I'm inclined to agree with John Hawthorne when he writes:

'I wouldn't even bet on the law of noncontradiction at any odds, and I think myself rational on that score.' (Knowledge and Lotteries, p29fn72)

I'm worried, therefore, that Fantl and McGrath have just given us the materials for a sceptical argument, and one with unusually enormous scope; as the quote from Hawthorne brings out, it could extend to supposed paradigms of a priori knowledge.

But hang on. Isn't this just the main thesis behind sensitive invariantism of this sort; what one knows varies with how high the stakes are? When the consequences of a subject's being wrong about p are sufficiently dire, then that subject doesn't know that p.

Well, that's true, but usually we're talking about the stakes given a particular context. A subject who faces financial ruin if she doesn't cash a cheque by the end of the weekend doesn't know that the bank will be open on saturday; a subject who's career depends on getting somewhere by a certain time doesn't know whether the train she is on stop at her station just from overhearing a stranger say so, etc. Were the practical facts about those subjects otherwise, which of their true beliefs count as knowledge might be otherwise too.

But look again at Fantl and McGrath's principle. It doesn't mention contexts or scenarios or practical facts. It just has that big ol' fat 'for all acts' quantifier. So a subject fails to know that p so long as there is some act for which the embedded biconditional fails. I'm suggesting betting at ridiculous odds on p is potentially always going to be such an act, for any p. If so, scepticism abounds.

Now, Fantl and McGrath do address something like this worry. They write:

' might seem that we are imposing an unduly severe restriction on justification and therefore knowledge. There will be some cases in which in order to have knowledge, one will need to have absolute certainty, or something close to it. These will be cases in which something of great importance hinges on whether a belief is true. Doesn't this make the requirements of knowledge too demanding? Can't we have knowledge without absolute certainty?
We can. Nor does our view entail otherwise. After all, in Train Case 1 [a low-stakes case], our view is consistent with the claim that you know that the train is going to Foxboro, even though you have only the evidence of casual testimony. There are many cases--in fact, most cases like this--in which we have knowledge without having a strong form of certainty. Requiring certainty in these cases would be otiose. Our account requires certainty for knowledge only in cases in which certainty is important--its importance consisting in the fact that it is required for being rational to prefer and act as it the relevant proposition is true. This should give the skeptic no consolation.' (79)

This sounds like just what they should be saying to meet the worry expressed above. The problem is, it's just not at all clear how to square these remarks with what their principle actually states. The principle stated above requires, for knowledge that p, that there be no act A such that it is rational to do A given p, but not in fact rational to do A. I don't see how whether the subject in question is in a high-stakes or a low-stakes scenario has any bearing on whether there is such an act A. But maybe I'm missing something important.

Update: Kenny raises a good concern about how I'm understanding Fantl and McGrath's principle in the comments.

Further update: And Jeremy Fantl has generously pointed out that I am indeed missing important things.


I guess they have to mean "all acts" means "all actually available acts". If that's right, then in some sense their condition just means that the agent knows p in a situation iff she rationally ought to act the same as if p were a certain/guaranteed truth, in the same situation.

I normally think like you, in terms of "all acts" meaning "all hypothetically conceivable acts", or perhaps "all functions from states to outcomes" or something like that. Certain representation theorems in decision theory require that sort of assumption. But if we explicitly avoid it, then their characterization actually sounds like a pretty nice spelling out of what I think Jason Stanley's position is from having heard people talk about his book a few times.
I had a similar thought. And it seems that's what they're driving at in the long passage I quoted. So perhaps all I can say in criticism is that they've been sadly inexplicit in formulating their principle. I'll give it some more thought.

The modified principle would have the consequence that were I in fact to offer you such a bet, you would lose knowledge. But that does now look like standard sensitive invariantism. Even so, I'm still worried that (supposed) logical truths and the like might be in the scope of the proposal (a question which I discussed in a horribly inchoate manner here.
Hi Aidan,

Thanks for the comment. We perhaps should have drawn more attention to the point, but I think we were explicit in meaning only available acts. For instance, on page 72, we explicitly interpeted 'the thing to do' "in terms of what would be the best thing one can do in light of all one's goals," thereby eliminating things one can't do. In footnote 7 (p. 73), we say that "an act (A) is rational for S just in case A is available to S and there is no available competitor B such that it is rational for S to prefer doing B to doing A." Finally, in footnote 8 (p. 75) we explicitly note that the inference that drives our principle "is valid only when restricted to instances in which the premises..., if true, would still be true were S to face the choice of whether to make A true or B true."

We do, however, think that there is an argument in the vicinity for a kind of semi-skepticism. If one grants our principle and still wants to insist on evidentialism, then one can infer a strong pragmatic principle on knowledge --our EPC (roughly, you know that p only if you would be rational to act as if p no matter what your stakes were -- that is, no matter what acts were available). In "Evidence, Pragmatics, and Justification," we think this result is too skeptical, but if you think that certain propositions (Here's a hand, some simple logical propositions, etc.) that would be rational to act on no matter what your stakes, then it might be a skepticism that isn't too horrifying. (This argument is currently in development at the factory.)


Thanks for the favorable interpretation. I should point out that I think we may differ from Stanley (and Hawthorne) in the direction of our conditional, as well as the content of what, for them, is the antecedent and, for us, is the consequent. So, we say:

S knows that p only if S is rational to act as if p.

They say (avoiding some details):

p is appropriately used in practical reasoning only if S knows that p.

The Hawthorne/Stanley line has the disadvantage/advantage (depending on who you ask) of not figuring in a quick argument against evidentialism. So, see Stanley's comments here (, comment 10), in which he agrees with Aidan's characterization (in comment 8) of him as complaining "about statements of his view that characterise it as allowing the possibility of two subjects in the same evidential state to differ in terms of whether they possess knowledge."

Of course, our view explicitly allows for the possibility of such subjects.

Thanks for the response. And thanks for drawing attention to the sections where you were more explicit on this point that I'd realised. The disadvantage of blogging is that it encourages you to discuss ideas which you've probably read over too quickly. (For me, the outweighing advantage is you can be corrected before it really matters).

I'd be interested to know how Jason's position bears on your claim (p84) that EPC is the only way for the evidentialist to embrace PC. If one allows what counts as evidence to shift with what counts as knowledge in a way that preserves the supervenience claim, wouldn't this be a evidentialist position that could be consistent with your pragmatic constraint, but which need not drive the bar for knowledge up so high that one can't meet it even in low-stakes?
See also Joe Salerno's post ( for the suggestion that stakes are relevant to what one's evidence is. And, I take it, there will be similar proposals for all other allegedly epistemically pure dimensions of evaluation.

Something in this vicinity seems like it will be available for evidentialists (and other epistemic purists) who wish to avoid pragmatic encroachment. Of course, one might think of it as a worst case scenario because, in effect, the claim will be that there is no epistemically pure dimension of evaluation. Stakes infect everything. This is not what I might have imagined evidentialists were hoping for.

That said, it's not enough for the evidentialist to say that stakes can be relevant to what evidence one has. If there is even one case in which a knower's evidence stays the same even when the stakes rise sufficiently to make it no longer rational for S to act as if p, evidentialism will be false. So, evidentialists will have to say that one's evidence always varies with the stakes.

But it's neither enough just to say that evidence always varies with stakes. It must always vary in the right way. That is, when the stakes rise, our evidence must always decrease, and decrease sufficiently so that when we are no longer rational to act as if p, the evidentialist will no longer want to say that we know that p.

Is this plausible? I have evidence to suggest that the fair 6-sided die will come up 5 or lower (not to say that I am anywhere near in a position to know that this proposition is true). No bet is currently offered. When a bet on the proposition is offered, does my probability for the proposition go down? To what? Is there a corresponding increase in the probability that the die will land 6? What of the frequencies I will predict for future roles of the die?

To the extent that evidence is tied to probability, I don't think much of the prospects for the view that evidence will always be affected by one's stakes in just the right way. And even if evidence is not tied to probability in the right way, there is a dimension of epistemic evaluation -- namely, probability -- that does not seem to always involve stakes in the right way. Epistemic purism can then be formulated with that dimension in mind and it will be, if our argument works, false.

Again... an argument in development.

Indeed - I remember Kvanvig's comment about the stars alligning in the Certain Doubts thread you mentioned earlier. It's good to see these important issues getting so much attention online recently.
Post a Comment

<< Home

This page is powered by Blogger. Isn't yours?