December 04, 2004

Archer on Knowledge

A couple of interesting dialogues about knowledge out of Ross MacDonald. (Mild spoilers for The Doomsters and The Zebra-Striped Hearse ahead.) The first is an example of assertion-without-knowledge that's much better than the last one I got out of a detective story. The first speaker is Slovekin, a reporter; the second is the detective, Lew Archer.

"So if you know something that will let Hallman off the hook, you better spill it. [1] I can have it on the radio in ten minutes."

"[2] He didn't do it."

"[3] Do you know he didn't for certain?"

"[4] Not quite. [5] I'd stake my reputation on it, but I have to do better than that. [6] Hallman's being used as a patsy, and a lot of planning went into it." (Ross MacDonald, Archer in Jeopardy, p. 178; my numbering)

Here Archer explicitly acknowledges [4] that he doesn't know what he originally asserted [2]. Slovekin's asking if Archer knows for certain [3] might be taken to shift the standards for knowledge up, but Archer goes on to assert something [6] that pretty much entails [2]; so he evidently doesn't know [6] by the new standard either. Not only that, Archer makes clear [5] that he'd stake his reputation on the truth of his assertions, which is eerily like my account of assertion on which the speaker stakes her credibility on the truth (or at least justification) of what she says. I could hardly have stacked the deck better myself.

There's an interesting cross-play of purposes in this dialogue. A Hawthornean practical-environment view of knowledge has a natural connection to the knowledge account of assertion, or rather an assertability account of knowledge. Suppose your evidence that p is good enough for knowledge only if it's good enough to make it practically rational to act on p. Suppose also in your practical environment you're faced with this choice: assert p, or don't. Then you know that p only if your evidence is good enough to make p assertable.

But in this case there are two different purposes to which the assertion [2], that Hallman didn't do it, can be put. One question is whether Slovekin is to believe that Hallman didn't do it (which will affect the information he gives Archer, and how he directs his further investigations); another question is whether Slovekin will broadcast a report that Hallman didn't do it. Archer's evidence is good enough for him to try to get Slovekin to believe, but not good enough for him to try to get Slovekin to broadcast the report. Hence it's OK for Archer to assert that Hallman didn't do it, but it's not OK for Archer to claim knowledge. Slovekin can't broadcast anything unless he knows, and Archer can't sign off on that higher epistemic standard.

The second dialogue is about weak standards for knowledge, and doesn't really require scene-setting.

[1] "But why would anybody want to kill Ralph?"

[2] "There's one obvious possibility. He may have known who murdered Dolly."
[3] "Why didn't he say so, then?"
[4] "Perhaps he wasn't sure. I believe he was trying to investigate Dolly's murder...." (Archer in Jeopardy, p. 447, my numbering)

The purpose of the modals in [2] and [4] clearly isn't to establish that in one epistemically accessible world Ralph knows who murdered Dolly and in a different epistemically accessible world he isn't sure. [2] and [4] pick out one and the same possibility. So in [2] we have a knowledge ascription that's compatible with the subject's not being sure of what he knows. (It's a knowledge-who ascription rather than a knowledge-that ascription, but I don't think that makes much difference.) Nor do I think there's much plausibility to the idea that standards for knowledge have shifted between [2] and [4], though I don't have a knockdown argument against it either. Perhaps Ralph's doubts, whatever they may be, become salient.

Anyway, the standards for the knowledge-ascription in [2] seem to be determined by the purpose of the action it is meant to explain, as specified in [1]. And this explanation requires very permissive standards for knowledge--even more permissive than saying that knowledge is true belief. If you're thinking of rubbing out someone who knows too much, you're not going to worry about whether he is Gettiered* or has a true but unjustified belief, or even whether he has full-fledged belief. Even someone who has a little evidence that you murdered Dolly and is thinking about the possibility that you did could be dangerous enough.

Then [3] calls for the explanation of a different action, and that explanation raises different concerns. Ralph wouldn't have told people who killed Dolly unless he was quite sure about it. A half-hearted true belief with some reservations wouldn't have been enough for him to tell people. But it would be enough to get him killed. That's why, under the standards for knowledge set by the inquiry in [1], Ralph can have known who killed Dolly, without being sure of it (as [3] brings out).

Haven't I just provided an explanation for how the standards shift between [2] and [4]? Maybe so. But it doesn't seem as though any of the obvious standard-shifting devices come into play. I don't hear Archer as emphasizing "sure" or any other part of his sentence, and no specific doubt has been made salient. I'm more inclined to say that this indicates that "knowledge" is a somewhat slippery word that can be used to pick out many different kinds of epistemic valuation while seeming to mean the same thing all along. But that's exactly the sort of thing I would say--people who take knowledge seriously will differ.

*Title for a philosophers' thriller: The Man Who Had Too Many Gettiered Beliefs.

Posted by Matt Weiner at December 4, 2004 05:57 PM

I'm not clear on what you mean by saying that knowledge is a slippery concept. If you mean it is ambiguous, then it ought to be possible to spell out what different senses of knowledge are. If you mean it is relative to some (slippery) parameter, then the obvious parameter is still the standard.

You give a good explanation as to why the second passage is invoking standards of a specific kind. It would be silly to deny it just because Archer didn't speak in italics when he said 'sure'. In the subtlety of natural language, standard-shifting must sometimes occur without the obvious standard-shifting flags.

(I will adopt this convention: When I shift standards, I will touch my thumb to my ear and make a beeping noise. *beep*)

Posted by: P.D. at December 4, 2004 07:35 PM

Oh... I meant at the outset to say that these are fabulous passages. I know post a second comment so as to say that.

Posted by: P.D. at December 4, 2004 07:36 PM

My use of "slippery" is itself somewhat slippery. (Isn't that an annoying thing to say?) Here's one thing that might suffice for slipperiness: If a term refers to different concepts in context, and sometimes the context shifts without any apparent markers, so that people can get confused in a way that's not likely with the context-dependence of "I" and such, then that term is probably slippery.

I'm working on a paper that expresses a more official view of what "knowledge" is like. It's going to be based in part on Anil Gupta's "Meaning and Misconceptions," in Language, Logic and Concepts (Table of Contents).

Anil talks about people who think that "up" refers to a particular absolute direction and who also use familiar phenomenological appearances to figure out what's up. So in their use of "up" they take the following inferences to be valid:

(1) If A looks like it's above B, the direction from B to A is up (roughly)
(2) If the direction from B to A is up, then the line from B to A is parallel to the Standard Up line

but (1) and (2) can lead to contradictions. The point is that such a people can use "up" pretty well and that we can evaluate some of their "up"-talk as true or false. We have to use what Gupta calls a "frame," which is distinct from a context (in ways I probably shouldn't require summarizing) to figure out which of the incompatible criteria apply. (And sometimes we won't be able to make sense of it.)

So the thought is something like this: Our use of "know" is governed by rules that are something like this:
(1) "S knows that p" is true iff S bears the appropriate relation to p, determined in the following way [mumble mumble mumble]
(2) If S and T have pretty much the same evidence concerning p, then S knows that p iff T does
(3) Disquotational schemata: modulo shifts in tense and other indexical elements, an utterance of "S knows that p" is true iff S knows that p
(4) Probably some more invariantist-type rules about how you can carry knowledge ascriptions from one context to another

which are meant to lead to incoherence if pushed around a lot. The point being that we don't usually get into trouble here because we don't usually find ourselves trying to carry knowledge ascriptions from one context to another in a way that makes a difference.--For instance, there's no temptation in the second Archer dialogue to say "But if he knew then he was sure!" though that inference might be fine in other settings.--Analogously, the reason Gupta's up-people don't get into trouble is because, when a guy is told that he needs to fix the light above the stove, it's very rare that he will take a gyroscope to the Standard Up and carry it a quarter of the way around the world in order to determine which light lies on a ray through the stove parallel to the Standard Up.

Anyway, that's roughly what I mean when I say "knowledge" is a slippery term; and I would go so far as to suggest that there's a slippery concept at issue too. I talked about this a bit more here. But I'm hoping to work this out in more detail (I'd better, if I'm going to use this paper as a job talk).

Posted by: Matt Weiner at December 5, 2004 01:59 PM