The (Mostly Harmless) Inconsistency of Knowledge Attributions

I’ve posted a long paper, “The (Mostly Harmless) Inconsistency of Knowledge Attributions.” It argues for a fourth alternative to contextualism, invariantism, and relativism: that knowledge-talk is governed by inconsistent inference-principles, but that these principles rarely lead us into contradiction. In fact, we can almost always assign an effective content to knowledge attributions even if their absolute content is self-contradictory. (The “effective content”/”absolute content” distinction is borrowed from Anil Gupta’s work on inconsistent discourses, which I draw on heavily.) So there is no reason to abandon knowledge-talk, even if it is inconsistent.

Pretty much the first thing I did after putting the paper up was to surf on over here and see Keith’s post, which seems quite relevant, since my paper compares and contrasts the inconsistency view with a version of contextualism on which an attribution of standards is always implicit (if often unvoiced).

Anyway, comments are of course welcome.


The (Mostly Harmless) Inconsistency of Knowledge Attributions — 11 Comments

  1. First, I’d like to recommend that the paper be shortened, if at all possible. Section 4 (for example) would be welcome in a book chapter, but a paper ought to be more terse. I say this because I think you’re on to something, and do not want you to be robbed of opportunities for feedback from those who might be alienated by the length of the paper.

    Second. So you have identified separate categories of ‘knowledge’, based (largely) upon the three principles; and given an argument that demonstrates that they may be inconsistent in some cases (that an underqualified use of ‘knowledge’ equivocates in a genuine way, depending on our conversation goals); and you have said this doesn’t matter to conversation, because each conversation has its own goals, and conflicting goals usually don’t arise.

    To be blunt, if I thought that was acceptable, then I’d be forced to come to the conclusion that epistemology is mere conversation analysis. But epistemology does not seem to be like that; it asks us for the truth, insofar as we can give it. And because some of these principles are more reliable than others in telling us the truth about things, some are better epistemic candidates than others.

    My reaction is compounded by the fact that the problems raised are easily fixable: we can abandon certain principles as instances of ‘knowledge’.

    Some even deserve to be abandoned, like the practical environment principle. Action-guidance for non-epistemic goals, and what would seem reasonable to do at the time on the basis of evidence, have little to do with knowledge. If a person said to me, “My evidence that p is knowledge because it’s rational to act on the assumption that it’s true”, I would say, “What?”. It is the relationship between our epistemic goals and future experiences which exhausts the nature of knowledge. Practical heuristics have little to do with it (directly).

    The disquotational principle seems true, but unilluminating, since it is always the goal of knowledge-claims. However, it is not always successful, and usually it is the case that we don’t really know if it has been successful. It seems more realistic to keep it as an end-goal, and use the relative successes and failures in our experiences as a way of modifying our standard of what counts as adequate evidence.

    The parity of evidence principle seems the most adequate of the three as a means toward achieving the disquotational end. Moreover, we have easier access to what it requires of us. I can ask myself, “Under what conditions could so-and-so know such-and-such?”, and (if I’m inclined to think about such things) to come up with an answer. It is both relevant and illuminating.

    As a result, I would think that certain conversations with certain goals are ripe for folly, and the socially-minded epistemologist should stay on their toes.

  2. Yentz, thanks for the comment (and the kind words about how I’m onto something). To start with, I regret that the paper is so long; I thought section 4 was necessary to forestall the objection (which has been made to me), “Any philosophical problem could be resolved by declaring that one of our concepts is inconsistent. Why should we think this one is inconsistent?”—by showing that knowledge is akin to a concept that we would acknowledge as inconsistent.

    But the paper is pretty long—anyone who’s put off, you can get the view by reading sections 1, 2, 3, and 5. That’s the teaser version.

    About whether this turns epistemology into conversational analysis; I’m trying to deal with a conception of epistemology on which it involves analyzing our ordinary use of “knowledge” and related words. In the distinction Jon draws here, it’s linguistic as opposed to value-centered epistemology. But part of my project is to show that, though linguistic epistemology might seem to give us an especial reason to concentrate on knowledge (since “know” is used much more in common parlance than other epistemological terms), we don’t get a coherent conception of knowledge from it. A value-centered approach, such as one that declares that epistemology is not concerned with action-guidance for non-epistemic goals, might yield a concept of knowledge, but it would be very different from the ordinary one. But it’s true that I’m not doing value-centered epistemology here.

    And I think you’ll get a lot of dispute from some people about the idea that action guidance should have little to do with knowledge! But I’ll let them make that argument here, if they choose.

  3. Hi Matt,

    To be sure, as the way things turn out, pragmatics and epistemology overlap significantly, since when we speak of knowledge we’re often interested in both the nature of knowledge-claims (which is very pragmatic) and knowledge-thoughts (not so pragmatic). The functions that Kvanvig discusses (esp. closure) do not as such force us to examine the pragmatic elements of epistemology; rather, after the linguistic turn, language ought to be understood as a part of the epistemic project as a matter of course, because language ought to be understood to be an active part of almost everything. And in epistemology, this attitude may be taken for good reason: these musings over classification can yeild philosophical dividends in unexpected ways, because they may provide the keystone to a solution for the “sheepdog” Gettier case.

    My complaint about ‘epistemology as conversation analysis’ is motivated by a disgruntled attitude toward certain of those principles you raise, because of a worry that some of them (i.e., the practical-environmental principle) are at best so sub-optimal that they’re not appropriate in treating knowledge according to the only goals which I can recognize as worthy of the name; and at worst, totally irrelevant. In Kvanvig’s terms, I’m an ‘egalitarian’ in terms of my view of the functions of closure and reliability, but ‘value-driven’ in the sense that ultimately I only care about what’s best for knowing the world, ourselves, and the events therein. And ‘value-driven’ does seem like an appropriate name, because I am forced to make a prescriptive or normative statement, because the ultimate goals of epistemology (i.e., striving for truth) are put in jeopardy. The ordinary is often relative, variant, and in many cases, dysfunctional; I see no virtue in a purely linguistic method. (And, indeed, the pure descriptiveness sought in linguistics may be impossible, anyway; for every use of a word I make carries with it the implication that it is the proper use, sort of like Sperber-Wilson’s relevance principle.)

    But even more than that, at an ethical level, the consequences would be bad. I honestly can’t say that the world would be worse off if knowledge were to fail to meet the P-E standard. Sure, people would have to get used to the funny, embaressing moments where they must confess knowing things only tentatively; or be forced to use a neologism, like “I underknow whether or not today is Saturday”. Still, this is far more honest – more epistemologically competent — far more cooperative, as a speaker (which is the point of pragmatics), and far less foolhardy, than present-day trends. I wonder if a serious popular attitude towards epistemology might even make the world a better place in some cases.

    Still, I recognize one intuition behind a P-E principle. For instance — as John Bell notes in his fantastic article on Infinitessimals at Stanford online — while in principle the notion of an infinitessimal is difficult (though not impossible) to define and defend, in practice engineers consider a length to be infinitessimal if its square can be neglected without any practical consequences. But this is an issue of definitions, which, so long as they have the force of stipulation behind them, are not themselves truths; only the corresponding universal statement in predicate logic of the definition is the expression of a truth, because it is ostensibly meant to talk about things in the world (or possible worlds). We never know the definition of a sign; it is always underknown, along with our raw abilities, faculties, and hallucinations. And the corresponding statements in predicate logic are true only by way of other principles, on the basis of other goals.

  4. I do not find the idea of an inconsistent semantics for knowledge ascriptions a sort of desperate move. In the good old times of Frege and Russell, it was costumary to say that natural languages are full of ambiguity, vagueness and inconsistency. Just how have we come to think that it is a plausible assumption we can give a consistent formal semanics for any part of our language?
    Some more serious remark. A possible further source of inconsistency in the concept of knowledge is dictated by the different kinds of epistemic virtues we attach to knowledge. As you observe, we sometimes use “S knows that P” to mean “S believes that P and P”. Decades of debate in epistemology, and particularly between externalist and internalist camps, should have made clear that we also use “S knows that P” to mean “S reliably believes that P, and P” (or something along these lines), e.g. in cases like animals and infants, and we also use “S knows that P” to mean “S rationally believes that P, and P” (or something along these lines).
    Why isn’t this enough to show that sentences of the form “S knows that P” are ambiguous, more or less like sentences of the form “S rightly believes that P”? and wht is the relation, if any, between such a view and yours?

  5. Daniele,

    Since Paul Grice’s “Logic and Conversation”, the idea that natural language is itself lacking has been called into doubt.

    But the underlying discussion is about whether or not we ought to respect ordinary use of “knowledge”, or refine it. I can, on the one hand, admit that language is used felicitously, and admit of some truth-conditions; and at the same time, say it ought not be used in that way, because its purposes are unbecoming as a means of achieving the goal at stake in the concept.

  6. Matt: As you can imagine, there are several points of your paper at which I see things quite differently. As you no doubt suspect, I don’t think the available data really support all three of the mutually inconsistent principles. Maybe we can fight over those issues some time later.

    But the issue of how to handle such inconsistencies in general is very interesting. It might have application to issues of vagueness surrounding the sorities paradox. Here it seems our talk might well support some principles that are mutually inconsistent — something along the lines of (to be very rough) tiny differences don’t matter and big differences do matter, which can, under suitable understandings and together with auxillary plausible principles, lead to contradiction because enough little differences can add up to a big difference. Anyway, I had an interesting discussion with Terry Horgan about this, he goes into a bit in a paper he co-wrote with Robert Barnard called “Truth as Mediated Correspondence,” which is available on-line at:


    Just a bit from that paper:

    First, vagueness is logically incoherent in a certain specific way: vague thought and discourse are governed by semantic requirements that cannot be mutually satisfied. Second, this kind of logical incoherence is benign in that it can be quarantined in language and thought in order to avoid problems like becoming committed to contradictions. But third, the logical incoherence inherent to vagueness renders ontological vagueness — i.e., vague objects and/or vague properties — impossible.

    So anyway, how they (Barnard & Horgan) work this out may be of some interest to you. I haven’t read the Gupta paper your use, and it doesn’t appear in B&H’s list of references.

    It can be pretty subtle line between a treatment of a puzzle that claims there’s an inconsistency burried in our thought & proposes a fix that does the least damage to its essential elements on the one hand, and a treatment that claims that our thought is consistent and puts forward the same “fix” forward not as a fix, but as a feature already present. The latter type of treatment can be guided by some principle like “Don’t take our thought/talk to be inconsistent, except under strong pressure to do so.”

  7. I was emailed the following last paragraph:

    It can be pretty subtle line between a treatment of a puzzle that claims there’s an inconsistency burried in our thought & proposes a fix that does the least damage to its essential elements on the one hand, and a treatment that claims that our thought is consistent and puts forward the same “fix” forward not as a fix, but as a feature already present. The latter type of treatment can be guided by some principle like “Don’t take our thought/talk to be inconsistent, except under strong pressure to do so.”

    I hereby put it up again, for posterity.

    More responses soon (Yentz, sorry it’s been so long without answering your post #3; basically, I agree with you in that I think we should adopt the value-driven approach, but I don’t think that this means we should worry about what it takes to know things).

  8. Ok, promised responses:

    Yentz, I’m pretty much in agreement with you about the primacy of values-based epistemology over the linguistic model. In fact, one of the things I’m ultimately trying to do here is to undermine the linguistic model from within; to say, “Hey, there’s a good reason to suppose analysis of our actual use of a ‘knowledge’ is a good way to do epistemology (since it’s been a successful practice), but if you actually do it you find that the ordinary use of ‘knowledge’ can’t be the basis for a completely rigorous theory even if it works in practice.” But I think that when we do values-based epistemology we won’t be analyzing knowledge as such, but different concepts. So we shouldn’t ask “What do we want from the concept of knowledge?” but “What do we want from our inquiry?”

    And it seems to me that the practical environment stuff gives one (but only one) answer to this question. One thing we want from our inquiry is: Beliefs that are well enough supported to make it rational to act on them. The degree of support necessary will depend on practical environment.

    Daniele, thanks for the remarks about the different epistemic virtues. I do think that the different epistemic virtues can all be subsumed under knowledge ascriptions at different times. I don’t think this makes ‘know’ ambiguous, at least not in the way that ‘bank’ is; it seems to me that we slip naturally from one use to another naturally in a way that we don’t for ambiguous sentences. In fact, this might lead to more inconsistency in our use of ‘knows’, over and above the practical environment—based inconsistency I talk about in the paper.

    Keith, thanks for the paper. I’ll definitely take a look at it. Matti Eklund’s “Inconsistent Languages” takes an inconsistentist (to coin a word, perhaps) approach to vagueness and to the liar paradox, as well. For the little it’s worth, this sort of analysis definitely seems to me to be promising for the liar paradox; we have a rule for the word ‘true’, it’s inconsistent, but we have no problem assigning effective contents to most truth-talk and the paradoxical sentences just won’t fit into any successful practice anyway. I need to think more about whether I think it works for vagueness.

    The latter type of treatment can be guided by some principle like “Don’t take our thought/talk to be inconsistent, except under strong pressure to do so.”

    Definitely well taken; it’s related to the objection I mention in comment 2, that any problem could be solved this way. One thing is that I’m not actually proposing a fix here; I think knowledge-talk is fine for our everyday purposes. At the level of theory I think (for other reasons, given in part in this paper) that we should be dealing with concepts other than knowledge.

  9. Matt,

    Thanks for the clarification. Sometimes I feel like a lot of people are partisans attempting to capture linguistic territory needlessly, when all we really need is a neologism to make the differences in goals transparent.

Leave a Reply

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