Certain Doubts

Tim Williamson holds that knowledge is the norm of assertion, and though I think this is wrong, let’s assume it for present purposes. My interest is not so much in what the norm(s) of assertion are, but rather whether they are defeasible.

Williamson, I recall, holds that they are indefeasible. That means when skeptics tell you of their skepticism and let you in on the philosophical issues that lead them to that conclusion, and then assert, “I don’t know anything,” they violate the norm of assertion. So too do the Churchlands and Stich when they say they have no beliefs (assuming belief is a necessary condition for knowledge). And when a follower of Pascal or James follows their pragmatic advice, and comes to believe God exists in some way or other, if they say “God exists, and I am pragmatically justified in so thinking,” that person too does something wrong.

I pick these cases intentionally, of course, to try to persuade that the norm(s) of assertion are defeasible. I’m interested to see how persuasive others think such cases are.

One more consideration, of a more theoretical sort. Norms that interest philosophers–epistemic, aesthetic, moral, pragmatic, etc.–are defeasible norms, aren’t they? Maybe I’ve been thinking about epistemic norms too much, of the sort Pollock describes, for example, but this seems to me to be true; and it certainly fits with my W.D. Rossian inclinations in ethics. If so, why would we think that the norms for assertion are different?

I don’t think this latter consideration counts for much, even if I’m right about norms generally being defeasible, so the weight of my complaint rests more on the cases above. One additional item, though: what does a theory such as Williamson’s lose by making the norm defeasible rather than indefeasible? One cost is obvious, since you’ll then incur some need for further explanation in order to say when defeat occurs. This cost is balanced, though, by the fact that if you insist on indefeasibility, you’ll have to explain away apparent counterexamples.

As noted earlier, I’ve put a new item on the sidebar. It links to works in progress where contributors can have a link put to drafts on which comments would be welcome (and, I assume, proper blog etiquette requires explicit permission from the author to quote from such work, since draft form and published form might be significantly different). If you have such a paper, you can either email me with the link, or put it in a main entry, and I’ll insert it in the sidebar.

Multi-Premise Closure (MPC) can seem a less secure principle than is Single-Premise Closure (SPC), because MPC is subject to what I’ve for a long time called the problem of the “accumulation of doubt,” which is, or is at least very close to, what John Hawthorne, perhaps a bit more appropriately, would call the aggregation of risk. (I’ve always had to quickly add that the problem needn’t be one of actual doubts accumulating, but may be one where, whether doubt accumulates in the believer or not, it should so accumulate.) Here is Hawthorne on this potential problem:

Deductive inference from multiple premises aggregates risks. The risk accruing to one’s belief in each premise may be small enough to be consistent with the belief having the status of knowledge. But the risks may add up, so that the deduced belief may be in too great a danger of being false to count as knowledge. (Knowledge and Lotteries, p. 47)

Now I should quickly point out that John does not abandon MPC on the basis of these thoughts. On the pages that immediately follow, he rallies to MPC’s defense, closing the first chapter of his book a few pages later with a fairly open mind about the issue, but seemingly leaning toward the pro-MPC position. I, on the other hand, think MPC fails because of the problem of the accumulation of risk. (I agree with John the relevant notion of risk is far from unproblematic. But I think there is some good notion there on which this problem is real.)

But what I want to discuss here is conditional in nature: If you think MPC fails because of this problem, what should you think about SPC? Since, as its name so clearly indicates, there’s only a single premise involved in cases of SPC, there seems to be no room for risk to accumulate. And what immediately follows the quotation above is this:

Granted, deductive inference from a single premise does not seem like a candidate for risky inference. (p. 47)

But wait! Though there’s only one premise involved in cases to which SPC would apply, there are two pieces of knowledge involved. For SPC – in order to be at all plausible – is not formulated as the principle that if S knows that p, and p entails q, then S knows that q. Rather, it’s the principle that if S knows that p, and S knows that p entails q, then S knows that q. (Other complications, which I’m ignoring here, have to added as well.) And this gives rise to the possibility that risk can accumulate even in cases of deduction from a single premise. Given how I think risk (or, perhaps, “dis-warrant”) aggregates, I think the failures of SPC that would arise through the aggregation of risk would have to be cases in which S’s knowledge of p counts as knowledge, but just barely so counts, and S’s knowledge of the entailment is also just barely a case of knowledge, while S’s belief that q falls just barely short of being knowledge. And what that means is that you shouldn’t expect to find clear counter-examples to SPC, even though we have reason to think not-so-clear ones exist. Which is good, because clear counter-examples to SPC of this type seem impossible to find!

Though I myself don’t take things this direction, I should note the possibility of agreeing with me in my conditional thought that if MPC fails for the reason in question, so does SPC, but, because SPC can seem so compelling, go the modus tollens route and take all this as reason to think we shouldn’t abandon even MPC, at least for this reason. For my part, I think both SPC and MPC need to be modified to handle this problem – as of course, they have to be anyway to handle several other problems.

[I should note that, though it was buried in an obscure footnote, I raised this problem to SPC several years ago: See note 14 to my Editor’s Introduction (draft available on-line here) to DeRose & Warfield, ed., Skepticism: A Contemporary Reader (Oxford UP, 1999). However, I think given current interests in lotteries and closure principles, it’s worth re-raising the problem at this time.]

We all know that justified true belief can fail to be knowledge when funny stuff happens (or at least most of us think this). What I want to ask is whether a JTB can fail to be knowledge for a more mundane reason–because the belief is justified, but it isn’t justified enough to count as knowledge.

Another way, perhaps, to put this is to question a line from section 6 of Ralph’s paper “The Aim of Belief”: “[T]here is no way for a rational thinker to pursue the truth except in a way that, if it succeeds, will result in knowledge.” Is this so?

Here’s a case I’d like to survey you on. Charlie Brown, a baseball general manager, is trying to decide who to pick in the amateur draft. He looks at the prospects and comes to believe, based on his high school performance, that Joe Shlabotnik will be a good major league player someday. Indeed, Joe does turn out to be a good major leaguer. So Charlie had a true belief; it also seems as though it may have been justified, because it was based on performance. Yet I would think that it falls short of knowledge, because predicting someone’s eventual major league performance on the basis of his high school performance is too uncertain.

(Apologies to non-baseball fans; the argument probably transfers to any sport, though baseball performance is notoriously difficult to predict.)

Indeed, I’d argue that Charlie is much better off knowing that his pursuit of the truth about Joe’s future performance will not result in knowledge. I’m convinced by Tim Williamson’s argument that one of the advantages of knowledge over JTB is that it is less likely to be abandoned in the face of counterevidence. Yet Charlie should be ready to abandon his belief in Joe’s future in the face of counterevidence. Given the chancy nature of baseball prospects, a general manager has to be prepared to abandon someone who looked promising but who isn’t panning out, or he may damage his team by keeping on an underperforming player. Players who you know to be good will be kept in the lineup after a poor start (I remember Barry Bonds batting under .200 one May when he was in Pittsburgh and going on to win the MVP–er, sorry again to non-baseball fans); players who you think to be good won’t.

Does this case convince you? Do you think Charlie is only justified in believing that Joe will probably be good? Do you think it casts any sort of light on the kind of justification that’s necessary for knowledge?

Assume that sensory experience has no propositional or semantic content. Further assume that Hume has developed two reliable dispositions. He has learned to discriminate colors from each other, and he has also formed the reliable disposition to believe in any color experience that the object in question is not the missing shade of blue. Thus, in every sensory experience involving color, Hume is disposed to form two beliefs, one identifying the particular color in question and the other denying that the object is the missing shade of blue.

After years of experience never having encountered the missing shade of blue, Hume finally finds an object that is the missing shade of blue, and upon seeing it, Hume believes that the object is the missing shade of blue.

My question is how to explain Hume’s justification for this belief. If we appeal only to his reliable dispositions, it looks like there are two competing dispositions. One disposition, a highly reliable one, inclines him to believe that the object is not the missing shade of blue. Another disposition, also highly reliable, inclines him to identify the particular color the object is, which in this case is the missing shade of blue. If we could appeal to some content of the experience itself, it would easy to explain why we favor the latter disposition over the former in accounting for the justification of Hume’s belief. But we are assuming that experiences have no content, so we can’t do that. Notice further that each disposition is perfectly general, and that each can be replaced in a given case with more specific dispositions: upon seeing a magenta flower, Hume is disposed to believe that it is magenta and that it is not the missing shade of blue, etc. (I point this out to try to block the attempt to see this problem as just a version of the generality problem for reliabilism.) So, without appeal to content, what reasons could a theorist have for preferring one of the dispositions in question to the other?

I’ve spent a lot of energy dealing with this skeptical argument, that I’ve called “The Argument from Ignorance” (AI):

1. I don’t know that I’m not a BIV
2. If I don’t know that I’m not a BIV, then I don’t know that I have hands
So, C. I don’t know that I have hands.

I’ve always thought it a very promising object of study, because it’s valid, both its premises are intuitively very plausible, yet its conclusion is highly implausible (for some, it borders on the absurd). So you’re bound to find out something at least somewhat surprising, and perhaps even important, about knowledge in solving the puzzle this argument presents us with.

But in recent years, I’ve noticed something of a trend among philosophers – usually in conversation, but sometimes in print – to think this isn’t a very strong skeptical argument, perhaps not even worth working on. (Often they have some other argument skeptical argument they’re advocating as the “real” skeptical threat. Invariably, I think end up finding this “real” threat to be far less threatening.) Mostly, they’re worried about the argument’s first premise.

Now, I believe they have correctly fingered the argument’s weakest think. But I think it’s pretty strong for a weak link. (The reason I find the alternative threats less threatening is that I invariably find them to include some premise far less plausible than 1, above.)

Often, a key point in the discussion over the power of this argument is the matter of how intuitively plausible that first premise is. (One might also wonder whether our not being BIVs falls into some broader pattern of things that we intuitively seem not to know. I think premise 1 scores pretty highly here, but that’s a long story.)

I have a report from “the front” on the intuitive plausibility of 1. For three years now, I’ve started off my introductory Descartes-to-Kant survey course, a large course which mostly draws students with no previous philosophy classes, with the question of whether the students know that they’re not BIVs.

This is the very first thing I do in the whole semester. (I justify (rationalize?) the exercise by relating it to my lecture on Descartes’s First Meditation later in the opening meeting.) So I’m mostly receiving the verdicts of people “unspoiled by philosophy” – or at least by their contact with philosophy classes.

Here’s how I take my poll. The opening item on the lecture outline is: “A Philosophical Question or Two.” (”Or Two” because, depending on how I’m doing for time, I might also ask whether they think it’s possible that they are BIVs, which can later facilitate discussion of the second great argument by skeptical hypothesis, like unto the first, the “Argument from Possibility.”) I write these options on the board: “I know that I’m not a BIV” and “I don’t know that I’m not a BIV.” I prepare by explaining the BIV hypothesis, but refraining from using terms of epistemic appraisal in my description of the hypothesis. (So, for instance, I’m very careful not to say anything even remotely like, “So, if you are such a BIV, you can’t tell that you are,” since I take “can’t tell” to be a question-begging negative assessment, or at least to be question-beggingly close to being a negative assessment, of any belief one might have to the effect that one is not a BIV.) I then tell them that I’m going to take a show of hands on the question of whether each student thinks she does, or does not, know that she’s not a BIV. The first time I did this, I noticed a lot of glancing about by the students as I waited for the first show of hands. This made me worry that many were being influenced by how others voted. So the last two times I did this, I explained to them that I would wait quietly a few seconds to let everyone make a firm decision in their own minds as to what they thought, so they wouldn’t be influenced by how others were voting when I did ask for a show of hands.

The results were extremely strong. The first time I did it, I got these results (as reported in note 15 of my paper “Sosa, Safety, Sensitivity, and Skeptical Hypotheses,” forthcoming in SOSA AND HIS CRITICS, but on-line in draft form here [word format] and here [pdf format]):

I recently took a poll in a class of over 70 introductory philosophy students….My question was whether or not each student, in her or his own opinion, knew that she or he was not a BIV. A clear majority — about 2/3 of the class, it seemed, though I didn’t count — of the hands went up. So most of these students agreed with the skeptic on this issue. When I asked who thought they did know they weren’t BIVs, only three hands went up.

The last two times, the results were even stronger. I suspect it’s really about as strong as you can hope for from undergraduate classes on just about any question – including whether killing babies for the fun of it is morally wrong. Well over 80% voted that they didn’t know, and in each case, less than 5 (in classes of about 100 students, which is what this course has drawn for the last two years, at least for its opening meeting during Yale’s “shopping period” – before I frighten some off with such questions!) voted that they did know. (So there were still some abstainers. It’s possible that there were fewer abstainers the last two times because of my instructions to firmly make up their own minds before I took the show of hands.)

Perhaps others have obtained different results. I’d then be interested in some details about how the opinions were collected. One very important feature of how I do things is that I do not present any skeptical arguments until after I’ve taken the vote. Explaining and then asking the question of whether they know they’re not BIVs is really the first thing done in the course. If students are presented with the question in a format where it’s clear that a negative verdict on whether they know they’re not BIVs forms an important part of an argument designed to show that they don’t even know they have hands, many might decide that they do know that they’re not BIVs in order to escape the threat the argument poses. (Of course, some may worry about such skeptical threats on their own, without having to be presented with the argument.)

All of this can lead one to think that AI really is a remarkable argument. Any argument whose weakest link is that intuitively powerful and yet has such an implausible conclusion is something of a wonder.

My recent experience with refereeing for journals has had an inordinate commonality. I’m seeing a greater percentage of counterfactual analyses playing a role in papers than I can ever recall. I wouldn’t be surprised if my experience is idiosyncratic, and perhaps I notice these things more than others since counterfactual analyses are a pet peeve of mine–it is a sign of the lack of quality control in our discipline, since it has been known for more than two decades how precarious such analyses are.

My proposal is that no article containing a counterfactual account of anything should be allowed without an explicit explanation of how that account avoids the conditional fallacy described by Shope in his 1979 Journal of Philosophy piece (or whatever emendation of Shope’s results are favored). Upon encountering such an analysis, it is usually an easy task to construct counterexamples to the account, but something more is needed than simply rebutting all the counterexamples one can think of or that are brought to one’s attention. Responsible authors ought to be able to say why their particular counterfactual account can avoid the recipe that can be elicited from Shope’s piece (with an important precursor in Chisholm’s underappreciated “The Problem of Empiricism,” though Chisholm doesn’t attempt to generalize to all counterfactual theories).

As noted in my previous post on Sellars’ problem, one option is to jettison the presupposition in my characterization of Sellars’ problem that the notion of justification (or whatever notion is thought to close the gap between true belief and knowledge) is connected with the notions of intelligibility, insight into truth, and the cognitively understood, seen, or grasped. On such an approach, it is better to characterize the cognitive machine more on the model of inputs and outputs, where the outputs are beliefs and the inputs can be whatever the particular theorist wishes to hold generates epistemic value for beliefs. Intelligibility and understanding are jettisoned, and the new language uses notions such as reliability of the mechanism, proper functioning of the system, and production by systems that get us to the truth at least most of the time.

Approaches to the problem that fall into this category include Goldmanians, Plantingites, and Sosians. These groups have a hard time explaining why the relationship between experience and belief is not completely variable as long as the varying is truth-conducive in certain specified ways (the ways that distinguish various versions of reliabilism from each other).

I’m not so interested in the problems and solutions proposed from the reliabilists here, as I am in another group of theories of foundationalist and coherentist sorts. Take coherentism first. Let the kind be that which limits the role of experience to that of belief production. It doesn’t play any role in justification, but it does play a role in producing beliefs with a special status in the system of beliefs. These beliefs, in Lehrer’s memorable phrase, are the products of the “prick of sense.” But the experiences have no content and play no role beyond identifying certain beliefs within the system as spontaneous beliefs or observational beliefs, or whatever else the theorist in question chooses to call them.

After we have the system of beliefs divided in this way into those produced by sensation and those not so produced, we have special rules that tell us when the entire system coheres. In the worst case for coherentism, the rules pay no attention to the distinction between sensory beliefs and the remainder, resulting in a theory that would produce exactly the same justificatory picture even if the role of experience had bizarrely changed so as to produce other beliefs in the same system. A more plausible story is that the special rules for coherence treat the sensory beliefs in a way different from the other beliefs.

This difference won’t help much, though, since it won’t matter which experiences cause which beliefs. All that matters is that the beliefs are sensory ones. There is no guarantee that we couldn’t vary which beliefs are the products of which experiences and get the same results, and even if this possibility could be eliminated, we could replace the experiences in question with entirely different ones or simply permute which experiences cause which beliefs, and as long as the system of beliefs stayed the same, with the same beliefs counting as sensory beliefs, we have problems. (This is precisely Plantinga’s point with the mountain climber/opera case).

So coherentism is in trouble if it doesn’t find a more substantive role for experience, substantive in the sense of finding a way to tie the nature of the experience to an appropriate belief.

What is interesting, I think, is that a version of foundationalism has the same problem. Suppose you’re a foundationalist who denies that experiences have content, but that certain beliefs are justified in part by being produced by such experiences. If experiences have no content, then the beliefs which result are not justified in virtue of some shared content with an experience. So it is not in virtue of any content of the state of it’s appearing to you that it is raining that either your belief that it is raining or your belief that it seems to you that it is raining is justified. Just as with the coherentist picture above, there is the threat that we could vary the experiences and the foundationalist would have no explanation why the sensory beliefs produced are justified in one case but not in the other.

The result of this reasoning is as follows. If we use the presupposition of Sellars’ problem in our metatheory, we carve up theories of knowledge in such a way that some versions of foundationalism and coherentism end up in the same category as reliabilism, proper functionalism, and standard virtue epistemology. Somewhat strange bedfellows, wouldn’t you think?

Here’s a followup issue regarding Keith’s post that could be lost in the comments and is significant enough that it shouldn’t be. It’s from Matt McGrath, and so I’m lifting it to a main entry to promote further discussion of it:

A question for Keith. Suppose a third party speaker is mistaken about what is at stake for a subject S in whether p, but S is not. In fact, S has little at stake in whether p, though the speaker believes S has a great deal at stake. Suppose also that S has good evidence for p. The speaker then thinks of various counterpossibilities of S’s error. It seems to me that for Keith this should normally raise the standards associated with ‘know’ in the speaker’s mouth. As Keith suggests, what affects a speaker’s meaning depends not on the facts but on the speaker’s beliefs, doubts, etc. And if the standards are raised, the speaker will speak truly in saying, “S doesn’t know that p.” Doesn’t this seem wrong? Similarly, if the speaker mistakenly thought nothing much was at stake for S in whether p, she would not speak truly in saying, “S knows that p.”

I have in mind the Cohen 1999 paper on contextualism and the Gettier problem. Cohen argued that a contextualist solution (like Lewis’s) to the Gettier problem is implausible: when the third party speaker wrongly believes a subject isn’t in a Gettier case the speaker can’t speak truly in saying “S knows.” I would think that it would be every bit as implausible – though I don’t recall if Cohen makes this point – to think a speaker who wrongly believed a subject was in a Gettier could speak truly in saying “S doesn’t know.”

Keith, if you think Cohen’s argument works for the Gettier problem, why doesn’t the parallel argument above work to show that contextualist maneuvers are useless for explaining the epistemic relevance of stakes?

I’ve been following the fascinating thread on pragmatic encroachment, and it led me to wonder about a situation William James poses in The Will to Believe. He takes our situation with regard to belief in God to be one in which (1) the risk of being wrong is quite high, (2) the evidence on each side balanced, and (3) the probability of future inquiry providing decisive evidence in favor of one conclusion over the other is quite low. This separates two considerations that are typically kept together in the posts on pragmatic encroachment-the risk of error and the perceived relevance of further inquiry. What are we to say about cases like James’s?

Suppose that James is right about all three of his claims. Can one know either that God exists or that God doesn’t exist? I take it that if the evidence is balanced, very few people are going to be willing to grant that one can know either way. Yet James offers some pretty persuasive reasons to think that it isn’t obviously irrational to believe one way or the other. And it seems crucial to his point that future inquiry will never yield decisive evidence for either conclusion. He says that it is unreasonable to make it epistemically repugnant to form a belief in this case because it puts out of epistemic bounds forever a possible truth whose importance is huge.

So, we might say that one cannot know either way, but it is still epistemically rational to believe either way. Given that the lack of knowledge in this case is not due to a failure to meet some fiddling condition to ward off gettier cases and the like, but rather a straightforward lack of evidence, this puts some strain on the usual assumptions about the connection between knowledge and rationality. It also demands an account of why the unavailability of future evidence renders these propositions rational when other propositions that are equally (badly) justified are not rational to believe.

Or, we could say that we don’t know either way, and it’s not rational to believe either way. Then one has to face James’s point about putting out of epistemic bounds forever a hugely important truth.

What I find interesting about all this is that the perceived risk of error here seems to be motivating a *lowering* of epistemic standards rather than the usual case of making them higher (speaking loosely here, not meaning to distinguish one form of contextualism from another). And there’s no reason to think that belief in God is the only case that has this feature. I may have balanced evidence for the proposition that my recently deceased mother loved me. Assuming the unlikelihood of relevant future evidence, and the importance of the positive belief for my future happiness, I seem to be in the same situation.

Having written all this, it strikes me that perhaps no one else is bothered by the possibility that beliefs about such matters might be epistemically out of bounds forever. If so, I guess I’ll just have to commune with old William’s ghost for commiseration!