Monday, March 22, 2010

Generality Problems in Biology and Epistemology

Reliabilist definitions of knowledge tend to go something like this:

R: "S knows that P iff S has a true belief in P that was generated by a generally reliable belief-forming mechanism."*

A traditional objection to this is the so-called "generality problem," which goes something like this:

I want to know what time it is, so I pull my cell phone out of my pocket, flip it open and look at the screen. Now, assuming that my phone is working properly, I now have a true belief about what time it is**, etc., we'd want to call my belief about what time it is a case of knowledge, and the reliabilist would point to the process by which I arrived at that belief--checking my phone--as a generally reliable one.

But wait! What level of generality are we talking about? The process of *me* checking *this particular* phone? The process of anyone checking the time on any phone? The process of anyone checking the time on any time-telling device? Depending on what level we're talking about, the statistical (hypothetical limiting) frequency with which the process in question generates true beliefs will vary all over the damned place. Moreover, there seems to be no non-ad hoc way to decide which level is the important one. Ergo, reliabilism is just a hopeless epistemic theory.

Hold that thought for a moment while we switch our attention to evolutionary biology. Whatever one thinks of generation by "generally reliable belief-forming processes" as a condition that makes true belief knowledge, no reasonable person would deny that the claim that "natural selection" is the primary engine of evolution is a respectable, informative, and well-understood bit of biological theory, or indeed that it's extraordinarily well-confirmed by the evidence. Right?


Jerry Fodor has recently attacked natural selection, claiming that, when it comes to pairs of correlated traits, (a) there's no fact of the matter about which trait is "selected for" and which is the free-riding spandrel, or at least that (b) there's no way for us to have epistemic access to which is which, and at any rate that (c) standard evolutionary biology's problems with sorting out which trait is which in a plausible, principled way, whatever exactly the standard for plausibility and principledness is taken to be in this context (which is far from obvious) constitute a huge problem for the theory of evolution by natural selection, indeed a reason to reject theory. (Many of Fodor's formulations--e.g. that the distinction between selected-for traits and spandrels is "invisible to natural selection"--are frustratingly hard to pin down, but fortunately, the distinction between the more radical reading of Fodor in (a) and the more moderate reading (b) isn't central to the point at hand. For our purposes, (c) is what really matters.) This may cause some head-scratching in readers who know enough contemporary biological theory to know something about how biologists investigate and sort out spandrels from selection.

In a review of Fodor's book by Ned Block and Philip Kitcher, they put the issue like this:

"Consider the famous case of industrially induced melanism in the peppered moth. Supposedly, in landscapes where pollution has destroyed the lichens on the trunks of trees, melanic (black) variants of the moth are better camouflaged when they rest on tree trunks than their lighter, speckled relatives. With improved camouflage, birds and other predators are less likely to pick the moths off the tree trunks. In polluted environments, then, melanic moths are more likely to survive, and hence to leave descendants in later generations. So far, so familiar.

"Enter Gould and Lewontin. Maybe moth coloration is a spandrel, and some other property of the moths is both relevant to their proliferation and correlated with their color. For example, evolutionary biologists have observed that moths usually rest by day on the undersides of branches rather than on the trunks of trees. So is the familiar black-as-camouflage story really true? Perhaps a characteristic of the larvae of melanic moths makes them more likely to survive. Or perhaps melanic moths have a tendency to move around less at night, which makes them less vulnerable to being eaten by bats (who care nothing for color). These are interesting alternatives to the familiar story, and the causal hypotheses they introduce can be tested in obvious ways: by examining the rates of larvae survival or by investigating nocturnal motions of moths. And this is what biologists have done. Concerned that an apparent adaptation (a camouflaging color) may be a side effect, they have looked for correlated traits that might figure in some alternative process that would culminate in greater representation of the melanic moths. Despite some controversy in the 1990s, the traditional story seems to be standing up well.

"If Fodor and [his co-author] Piattelli-Palmarini acknowledge the evidence that favors the camouflaging-color hypothesis over the moth-larvae and moth-mobility hypotheses, they will have to say that the biologists have not been imaginative enough—that they have overlooked some other correlated trait for which there could be no fact of the matter about whether it, or the black coloration, caused the reproductive success.

"What exactly could this trait be? One possibility, suggested by remarks in some of Fodor’s previous writings, would be that there are two different properties: being black, on the one hand, and matching the environment on the other. Is there a fact of the matter as to which of these causes the reproductive success?....

...[Fodor and Piattelli-Palmarini] envisage a vast space of properties and expect proponents of natural selection to discriminate among all the rivals. Not only is there a property of being-a-melanic-moth, there is also a property of being-a-melanic-moth-and-smaller-than-Manhattan."

Now, I'm not particularly interested in the Fodor-exegesis question here. Perhaps Block and Kitcher have captured Fodor's position and perhaps they haven't. However, in terms of Fodor as Block and Kitcher understand him, two points seem worth making here:

(1) This doesn't seem to be a particularly good objection to standard biological theory,
(2) It's strikingly similar to the generality problem for reliabilism.

In fact, it seems reasonable to call Fodor's-objection-interpreted-this-way a sort of generality problem for biology, where instead of worrying about how to cut up reliable processes, we're worrying about how to cut up causally-efficacious traits.

So, if we don't take the difficulty in determining whether being-a-black-moth or being-a-black-moth-smaller-than-Manhattan or being-a-moth-smaller-than-Manhattan-that-blends-into-the-color-of-its-background is relevant level of analysis for natural-selection-talk to be much of a problem for evolutionary biology, should we regard the difficulty in determining whether me-consulting-my-cell-phone, me-consulting-any-cell-phone or anyone-consulting-any-time-telling-device is the right level of analysis for reliable-process-talk to be a serious problem for reliabilism? Alternately, should we regard the fact that, for the sake of argument, the latter is clearly a troubling problem to be a good reason to take the generality worry about evolutionary biology seriously? Or are the two cases just so clearly and importantly disanalogous that our reaction to the generality problem for biology and our reaction to the generality problem in epistemology should properly be utterly independent of one another?

These questions aren't rhetorical. I'm open to the possibility that the generality problem really is a deep problem for reliabilism, although its biological analogue is an easily-dismissed problem for natural selection. The only positive claim I'm making here is a relatively narrow, wussy sort of claim:

There's an interesting, suggestive analogy between the two cases, and if one shares my strong intuition that the generality problem for evolutionary biology is a non-problem, this should, at the very least, give us a good reason to pause and take a harder look at the generality problem for reliabilism, and why exactly we take it to be such a problem.

*A note of caution here: Not everyone whose epistemic views would be characterized as reliabilist would put it quite like that, some would add substantial extra conditions and some epistemologists, while emphasizing reliability considerations enough to deserve the "reliabilist" label, might regard the project of trying to precisely formulate necessary and sufficient conditions for knowledge as a fool's errand. For present purposes, though, R should be good enough to capture the aspect of the position I'm interested in.

**"Etc." here papers over, among other things, Gettier-style problems....what do we say about a case where my phone somehow froze exactly twenty four hours ago and I didn't notice? Is that a reliably-caused true belief? This is an interesting question, but it's one that I suspect many people likely to read this are beyond sick of hearing about, and one which I don't have much of anything new to say about in any case.


Daniel Lindquist said...

I agree that this "generality problem" for biology is a non-problem, but think that the analogy is not particularly strong.

The issue with the generality problem for reliabilism is that, depending on what process we look at as being the relevant one when you look at your phone and learn the time, we might get different answers regarding reliability (and so regarding the status of your true belief as knowledge). Suppose you look at the time to the nearest second, and that only cellphone clocks and atomic clocks are reliable there (since other clocks are set by hand, and most people don't bother lining up the second hand precisely), and that the majority of clocks are neither cellphone nor atomic. Then your belief that it is now 4:23:05 PM would be reliably formed if you formed it by looking at a cellphone clock, but not if you formed it by looking at a clock simpliciter. Which means it's reliably formed or not reliably formed, depending on what we say about the level of generality we're looking at. So you know or don't know the time to the nearest second; we can't tell which. For the reliabilist account of knowledge to be workable, we have to have some sort of principled way of saying whether you were looking at a cellphone clock in particular or just a clock in general when you formed your belief, on pains of it being unclear whether or not you can know what the time is (to the second) by looking at the clock on your cellphone.

(And of course you were doing both things, and the looking-at-a-cellphone-clock was identical with the looking-at-a-clock-simpliciter (since there's just the one clock you looked at), so trying to say that it's one of these and not the other that lead to the belief being formed looks mighty hard to work out.)

But there's nothing analogous to this consequence in biology. The reason the "generality problem" is a non-issue for biology is pretty nicely laid out in Block & Kitcher's review: "How then could there be a sense in which one of the properties—being-a-melanic-moth—rather than the other—being-a-melanic-moth-and-smaller-than-Manhattan—caused the increased reproductive success?

We suggest that the question deserves a shrug. Serious evolutionary biology is concerned with comparative causal claims among interestingly different alternatives. Is it the black coloration rather than the larval resilience or the nighttime lethargy? Good question. Is it the coloration rather than coloration-and-being-smaller-than-Manhattan? Silly question."

That is, the question posed by the "generality problem" in biology is "silly" because it doesn't matter what we say about it. Being-melanic and being-melanic-and-smaller-than-Manhattan are both good ways for the moth to hide on tree trunks, which is what biology cares about. Biologists can just shrug their shoulders at some questions (as Block and Kitcher suggest) and still have a lot of interesting things to say about their subject. Reliabilists, on the other hand, cannot shrug off the generality problem, as it affects pretty much any case of supposed knowledge you might want to look at. If the reliabilist shrugs off the generality problem, there's nothing left to reliabilism.

Ben said...

This might be right, but I'm not as confident as you are.

It's certainly true that, while (if one glosses reliability in statistical frequency terms) different answers about which process caused the belief will result in different degrees of reliability, whereas different answers about which trait caused increased reproductive success don't lead to importantly different results. As such, there seems to be a sense in which you can 'pick whichever you want' in the latter case but not in the former.

I'm not sure, however, about just how important a disanalogy this. After all, in the biological case, the fact that all of an infinite number of traits, cut up at different levels of generality, produce the right results is the problem. The objection is that Fodor is raising (on this way of understanding Fodot) is that because they would all work the biologist has no good, principled way of deciding which level to pick. This lack of a principled way of picking levels is (for very different reasons) supposed to be the reliabilist's problem as well.

Again, I'm not at all sure how much to make of the analogy. At least some of the question surely revolves around precisely *why* one takes the biological version to be a non-starter. I certainly agree that it is--when I read that passage, I had a hard time believing that someone like Fodor would press such a silly objection in the first place--but I'm less satisfied by Block & Kitcher's diagnosis of exactly what the problem is. After all, the (alleged) problem is that the biologist can't do better than just shrugging their shoulders and picking arbitrarily, so saying that they *can* do this isn't obviously a satisfying response.

Daniel Lindquist said...

First, a personal note: I'll be at the prospective students thing at the University of Miami this weekend. I note this because it is weird to me when people I've talked to online suddenly show up "IRL" so I am giving you advanced warning. Now back to Fodor.

"After all, in the biological case, the fact that all of an infinite number of traits, cut up at different levels of generality, produce the right results is the problem. The objection is that Fodor is raising (on this way of understanding Fodot) is that because they would all work the biologist has no good, principled way of deciding which level to pick."

I think this is right, as a summary of Fodor. The problem is just the fact that Fodor thinks this raises a problem for "Darwinists", when really it just shows that Fodor doesn't know how evolutionary biology proceeds as the science it is. I think this comes out in Block & Kitcher's response, which I like quite a bit; hopefully a little elaboration will make things clearer.

Block & Kitcher's point about Fodor's "it's just natural history" complaint is that Fodor is coming in with an a priori notion of what sciences must be, and then complaining when a science doesn't match his ideal. The "problem" Fodor raises isn't a real problem because that's not how biology works. Biology isn't trying to find The One True Causally Relevant Trait for any given case of fitness, it's trying to answer the "good questions" like "whether it's the black coloration or the nighttime lethargy that makes the moths fitter". Biology can answer those questions. It can't answer the "silly questions", but biologists aren't interested in asking those in the first place.

Fodor thinks evolutionary biology must be trying to do something like "discover essences". As Block & Kitcher say: "Some would wince at the idea that Darwin and his successors were in the business of finding some analog of a Platonic theory of beds.... There is no “theory” of natural selection. So what?... nobody needs a “theory” of the type [Fodor and P-P] demand."

To hypothesize a bit: I think Fodor doesn't like that the "theory of natural selection" doesn't look like how he thinks theories of science should look, and that he thinks this because he thinks of all sciences as being physics (at different levels of generality). A very old, very wrong, philosophical muddle. (I suspect it goes back to the early modern rejections of "final causality" in general, when really all they should've done is maintained that their new physics didn't need any such thing. And then it continues on through Kant's denial that there could ever "be a Newton for the study of life" or a scientific explanation of the growth of a blade of grass, on to the neglect of biology in 20th-century philosophy of science that Ernest Mayr complains about in the first pages of his "Toward A New Philosophy of Biology".)

Fodor raises a complaint that could hold water against an imaginary science that's trying to play the game he's thinking of, but not against actually existing evolutionary biology. Thus Leiter's nice quip that Fodor has to turn in his "Quinean union card": Fodor is here failing to recognize what a science is doing before he tries to raise problems for the scientists doing it. It's bad naturalism.

For what it's worth: I can see some analogy between the problem for reliabilism and the "problem" Fodor raises; I think I may have been coming across as too negative in my past comment. I did like this post and it's been fun to think about just what makes Fodor wrong while the anti-reliabilist is right.

Ben said...

Ah, cool. I'm not sure just how much of the weekend I'll be around for, what with community college teaching on both Friday night and Saturday morning, some parties and whatnot I'm going to for the Winter Music Festival, trying to poke at papers that I've been waiting for too long to start sending out, and, of course, the fact that "Hot Tub Time Machine" comes out this weekend. That said, I am signed up for the dinner on Thursday night, so I'll definitely at least see you then, and I'll probably be at the Searle talk as well.

Anyway, I think everything you say here is right (and yes, I loved the "Quinean union card" line). I'm not sure what the upshot is for the reliabilism I said before, at the moment I'm more interested in raising and kind of poking at the question than pretending to have a well-worked out answer to it.

But yes, I think you've identified one extremely plausible source of Fodor's confusion. In terms of how this relates to the generality problem, one way to frame the issue might be in terms of just what the difference is between how biology works and how philosophy works, e.g. is epistemology necessarily in the game of developing hard and sharp "iff" conditions for concepts?

Kornblith's line in some of his articles about this seems relevant here--he endorses a broadly reliabilist sort of way of thinking about knowledge, but denies that this should be thought of as conceptual analysis, and trots out his stock analogy that if you want to know about gold investigating our concept of gold isn't going to be very useful.

Daniel Lindquist said...

There's a Searle talk? Neat. I have no idea what's happening this weekend; no one's shown me a schedule of any kind. I'll probably not see "Hot Tub Time Machine" until I get back home, sadly. Unless that's what Searle's talking about.

I think you're right that if reliablism is not trying to do something like define knowledge, then the generality problem might not be a problem for reliabilism. And if reliabilist epistemology starts to get interesting enough results, then maybe things like the generality problem will start to look like a philosopher's silly questions. (I'm not familiar with Kornblith; the little reliabilist literature I've read has explicitly been in the business of getting around Gettier problems by finding better criteria for knowledge; this is why I haven't read more of the reliabilist literature.)

Anonymous said...
This comment has been removed by a blog administrator.
Ben said...

Right, standard versions of reliabilism are definitely trying to formulate necessary and sufficient conditions, get around Gettier, etc. My impression of Kornblith is that--although I don't think he puts it quite like this--what he's doing amounts to a sort of version of reliabilism, or a very reliabilism-like theory, that isn't in the business of doing this. Whether that's a viable idea, and whether that would get around these problems, is of course a separate issue.

Ben said...

Oh, and checking the schedule, it looks like Searle is giving two talks on Friday, one before lunch and one in the late afternoon. I'll probably only be able to go to the first one, but it looks like you guys are scheduled to go to both.