Saturday, March 03, 2012

Dretske, Information, and Knowledge

There's a philosophy of mind reading group at UBC, reading Dretske's (1981) Knowledge and the Flow of Information this spring. I've never made a proper study of Dretske's work before, so I'm finding it extremely useful and interesting. In yesterday's reading group, I had an idea that I'd like to explore a bit further; consider this blog post a rather preliminary rumination.

First, some background -- both to clue in any readers who are interested in reading but don't know the Dretske, and so that I can make sure I have his framework clear in my own head.

Dretske's central idea is to use information theory to elucidate knowledge. The first few chapters of his book are devoted to articulating the relevant ideas about information. At least for the purpose of argument, I want to take all of that on board for now. Here are some central ideas:

Suppose we've adopted a convention whose purpose is to tell me whether John is happy. You'll send me a text message comprising either 'YES' or 'NO'; it is understood that you will definitely send me YES if John is happy, and NO if he isn't, and you (or anyone else) won't send me any other message than the one. In fact, John is happy, so you send me a message, and the word 'YES' appears on my screen.

In Dretske's terminology, my screen carries the information that John is happy, and it does so by virtue of having the word 'YES' on it. The idea is that only if John were happy would my screen say 'YES'; there's no other way for this to be so. There is, under the appropriate background assumptions, no possible way for my screen to be the way it is, without John being happy, so my screen carries the information that John is happy.

Dretske characterizes (at least a certain kind of) knowledge in terms of information:
K knows that s is F = K's belief that s is F is caused (or causally sustained) by the information that s is F.

He clarifies what it is for information to cause a belief:
When ... a signal carries the information that s is F in virtue of having property F', when it is the signal's being F' that carries the information, then (and only then) will we say that the information that s is F causes whatever the signal's being F' causes.

So for me to know that John is happy, I need my belief to be caused (or sustained) by my screen's saying 'YES', and for my screen's saying 'YES' to carry the information that John is happy.

This is prima facie a very strong condition on knowledge. For my belief to be knowledge, it must be based on a signal that nomologically guarantees the content of my belief. If, for instance, there is any possibility that you will confuse the signals, texting 'YES' to try to tell me that John is not happy, or that you will lie to me, or that you yourself will misapprehend John's emotional state, or that the phone company will send me a 'YES' when you sent a 'NO', then my screen does not carry the information that John is happy, so I can't know that he's happy.

It's easy to see that this kind of picture runs into the risk of serious skeptical consequences. Lots and lots of our putative knowledge looks like it's caused by signals that do not contain the information in question, in Dretske's sense. Take testimony, for instance -- suppose that you believe that Dretske's book was published in 1981 because I said it was, and I said it was because I believed it. There's no perfect relationship between my believing it was published in 1981 and its being published in 1981; there are in some intuitive sense possibilities where I have a wrong belief, and pass it on via my blog post. So my believing that it was 1981 doesn't carry the information that it was 1981. (This is so even if my belief amounts to knowledge, because my belief may in fact have been causes by a signal that carries the information, even though there are other possible ways it could have been caused.) So you can't get knowledge via my testimony.

Dretske's own strategy for heading off these skeptical consequences, I understand, is to develop his 'relevant alternatives' approach to knowledge and information, according to which it is in some sense not a (genuine) possibility that I be wrong, if the circumstances are in fact such that I was right. I haven't read that part of the book yet, so I'll hold off on discussing that material. The idea I wanted to consider in this blog post -- wow, that was a lot of exposition! -- is a different strategy.

My believing that p doesn't carry the information that p. But I have more properties than believing that p; might some of them carry the information that p, and causally produce or sustain your belief? Suppose, for instance, that I know that p. If I know that p, and tell you that p, and you believe me, we might say this: your belief was caused by my knowing that p. And my knowing that p (unlike my merely believing that p) carries the information that p. This is so, by Dretske's lights, even absent any moves about relevant alternatives -- there's no possible case at all, not even an 'irrelevant' one, where I know that p but p is false. Knowledge looks like an excellent signal for the transmission of information.

When I mentioned this idea in the reading group, it was met with a fair amount of resistance, but no one was able to give a very clear statement of what was wrong with it. One potential worry concerned whether it was plausible that my knowing could plausibly be the relevantly causally efficacious state; wouldn't my merely believing have had the same affect on you? Maybe so, but that doesn't mean it wasn't my knowledge that had the effect in actuality. Dretske endorses a kind of a counterfactual approach to causation; I think my knowledge passes the test in this case: if I hadn't known that p, I wouldn't have asserted that p, and so you wouldn't have believed that p. So your belief is caused by my knowledge.

Am I missing an obvious problem with this strategy? It looks like it'd be pretty helpful for someone attracted to Dretske's approach.


  1. jonathan weinberg3/05/2012 01:32:00 AM

    It's an interesting idea, and I think you're right about knowledge counting as informationally excellent, for Dretske, in much the way you suggest. But here's one hypothesis about why the story about testimony might not work, though. The information that is embodied in your knowing that p has to get to me by some intermediate channel -- it has to, as the title suggests, _flow_ to me. And thus each intermediary causal step between your mental state and mine is a potential informational bottleneck. In this instance, probably it's the causal step of your asserting to me that p: unless you would assert that p only if p (with probability of 1), then there's going to be some information loss between the event of your knowledge that p, and the event of your assertion that p. And probably that's still a pretty hard condition for your assertion to meet, for a line of argument that is mutatis mutandis much like the argument you give above about the inadequacy of belief here.

  2. Thanks, Jonathan.

    In response to the worry you cite, I think one could just run the same kind of move again -- it all depends on how we type the event in the causal chain. If we call it an assertion, then you're right: not enough information. But what I gave you wasn't just an assertion; it was a knowledgable assertion. And I wouldn't have given you one of those if its content weren't true.

    Of course there are other ways to describe what's going on too, and not all of them will work, but I don't see anything illegitimate about going this way. All we need is for you to receive something's being in some kind of factive state as a signal; that will happen when that factive state is causally efficacious.

  3. jonathan weinberg3/05/2012 09:19:00 AM

    The problem there is that I would need to be appropriately sensitive to the difference between your knowledgeable and your other-than-knowledgeable testimony. I was assuming that that is not the case -- that I have no special sensitivity to which of your tellings are or are not caused by your knowledge. (Would that I did! It would make our debates much quicker to resolve!)

    So, if I would come to believe P on your say-so in the same way in either case, then it is merely that you testified that P, and not that you testified knowledgeably that P, which will be doing the causal work, and will bring whatever information it has with it.

  4. I don't think we need to say that. We need for you to be sensitive to whether I testify-with-knowledge or not, but that doesn't mean we need you to be sensitive to the difference between knowledgable and non-knowledgable testimony. To make this line work, we need to say that if I hadn't testified-with-knowledge, then I wouldn't have testified at all. And lots of the time, this is correct: if I hadn't known it, I wouldn't have said it. (These are the sorts of situations in which testimonial knowledge is intuitively plausible.)

  5. jonathan weinberg3/06/2012 09:55:00 AM

    I think that might be right, that if your testifying that P _at all_ is something that can only happen, with probability = 1, when you know that P, then your testifying in and of itself will carry that information that P.

    But: "lots of times" isn't going to cut it -- if there are also times when you would testify without knowledge that P, but I would generally react to your testimony in just the same way as when you do testify with knowledge that P, then knowledge-sufficient information that P will not flow through your testimony, on Dretske's account.

  6. So as I understand Dretske's view, S's being F carries the information that p if there's no way for S to be F without p being the case.

    As you've said, if we let S be me and F be the property of testifying that p, then S's being F cannot carry the information that p. I agree. My suggestion, however, is that we consider a different property that I have. In particular, let S be me and F be the property of knowledgably testifying that p. Now that S is F does carry the information that p; there's no way for S to be F without p being the case.

    Furthermore, I think that you can be causally impacted by S's being F in this case. That means that you respond one way if S is F, and another way if S isn't F. That's good enough for transmission of knowledge, on Dretske's view.

    In your earlier argument, you suggested that you're not causally impacted by S's being F, because there's some possible way for S to be (ignorantly asserting that p) that would have had the same affect on you. While this is true, I don't think it shows that you can't be causally impacted by S's being F -- the implicit principle, that X's being G cannot cause Y unless there is no other possible way for X to be that would have resulted in Y, doesn't look true.

  7. jonathan weinberg3/06/2012 01:07:00 PM

    It's not a question of what is or isn't a cause or "causal influence" more generally. It's about the kind (or maybe degree?) of causation that is information-preserving in a potentially knowledge-conferring sort of way. I think of you look, e.g., at his spy/doorknock discussion, you'll see a degree of specificity is required for that sort of causation. But that specificity is missing when my reception of your testimony isn't sensitive to its being knowledgy.