Wednesday, August 6, 2008

Explications and Empirical Revision

In what senses are explications open to empirical revision?

Say you have a term t and an explication of that term t`. For instance, t could be “heat” and t` could be “mean kinetic energy”. Or t could be “volume” and t` could be “amplitude”. Or t could be “more probable than” and t` could be “has a greater limiting relative frequency than”.

One obvious sense in which the explication t` could be empirically revised is that we could have some theory T1 in which t` occurs essentially, and then, in light of new empirical evidence, we replace T1 with some theory T2 in which t` does not occur. For instance, we could have a Newtonian mechanical theory, in which “heat” is explicated with a certain definition of “kinetic energy”, and (loosely speaking) the evidence could suggest that we replace it with a relativistic theory, in which the definition changes. Or the evidence could suggest that some classical theory of statistical mechanics be replaced by a quantum mechanical theory, in which (so I understand) the frequentist explication of "probability" is not appropriate.

Another slightly stronger sense in which t` could be empirically revised would be this. Suppose we have some theory T, containing the term t, then we replace every occurrence of t in T with the explication t`, producing the theory T`. Then, since T is presumably more vague or ambiguous than T`, there is some conclusive evidence against T` which is not conclusive evidence against T. If we discover just this evidence, then we should abandon T`, but we should not necessarily abandon T. I imagine that some would want to say, in this situation, that they had learned that they had chosen the wrong explication. This would be especially tempting if there were another explication t`` of t, such that the theory T``, gotten by replacing every occurrence of t in T with t``, were true, or supported by all of the relevant evidence at hand.

I feel like these senses aren’t strong enough to capture what someone might want to express by saying that explications are open to empirical revision. An explication is empirically revised in this stronger sense not only if the theories in which it occurs are found to be false, and not only if some competing explication better preserves the truth of theories in which the explicandum occurs. But is there such a sense in which explications are open to empirical revision?

Perhaps the idea is that there is some source of empirical evidence that can tell for or against the decision to explicate a term a certain way that is not empirical evidence for or against any particular theory in which the explicatum is used. This seems unlikely to me. What form could this sort of evidence possibly take? If we choose an explication, say, to render the empirical consequences of a theory more transparent, then we might have evidence that the explication fails or succeeds in doing so. We might have evidence, for instance, that “…t`…” is a borderline case but “…t…” isn’t. But would this be empirical evidence? Can’t we tell borderline cases from the armchair, given enough background information? Note that we can supply the necessary background information ourselves, using thought experiments. We might be incapable of telling what the empirical consequences of some explicated or unexplicated theory are if we don’t know what the theory says. So we should be familiar with theories in which explicanda occur in order to make our explications work best. But is this really empirical knowledge, in the relevant sense? What is necessary is not knowledge that any given theory is true, or knowledge of the evidence for or against a theory, but knowledge of what the theory says. (Of course, what makes what one theory says more or less important than what another theory says is the relative likelihood of the truth of the theories, but philosophers qua explicators don’t generally take on the task of empirically assessing the relative likelihoods of theories – generally, I suppose, they assess the importance of a scientific theory on the basis of its prominence in modern scientific discussion.)

Perhaps the idea is that, since no one sentence in which a (non-observational) term occurs is immune to empirical revision, then no sentence we pick out as the explication of a term is immune to empirical revision. This also seems implausible to me. As I understand it, part of what it is to explicate t with t` is to treat “t = t`” (perhaps relative to a given theory) as if it were the definition of t (in that theory). Part of what that amounts to is agreeing to the eliminability of t for t` in all extensional contexts (in a given theory). When we cease to agree to this, we change the subject. In general, as I see it, the explicit adoption of an explication of a term in a theory is an attempt to escape the consequences of various sorts of holism about that theory. We explicate, in part, to make the meanings of terms less diffuse. Of course, if a popular theory containing t is rendered trivially but empirically false by replacing all occurrences of t with t`, then t` is useless as an explication of t in that theory. Nobody would agree to explicate t with t` in such a case, and everyone’s refusal to adopt the explication would be on empirical grounds.

As it stands, this satisfies me that explications are open to empirical revision only in our first two senses. Am I missing something?


Eric Schwitzgebel said...

This may be totally outside the box, but what about empirical evidence that a particular explication of a term has positive or negative moral consequences or consequences for society? For example, defining alcoholism as a "disease"....

iolasov said...

Thanks for the comment! My response got real long real fast. Sorry...

Different sorts of empirical revision might be going on here, depending on how we think of moral consequences and depending on what our purposes are in explicating a term like "alcoholism". We can think of the moral consequences of an explication for a society as (among other things):

(1) The morally non-neutral effects that a potential choice of explication would have, if chosen.
(2) When we explicate t with t`, roughly, the difference in the set of all conceptual entailments to moral propositions from sentences containing t to sentences containing t` (in the theory we are explicating).*
(3) When we explicate t with t`, roughly, the difference in the set of all morally non-neutral states of affairs that a person takes responsibility for in uttering sentences containing t and corresponding sentences containing t` (in a given theory).

With respect to (1), which I think is what you had in mind, I think you have a sense in which the decision to explicate itself is open to empirical revision, without regard for its theoretical consequences. If the explicator is in a position where her explication is likely to be widely adopted - she is writing the new DSM entry on alcoholism, say) - and learns empirically, say, that a social stigma worse than any of the psychophysiological effects of alcoholism attaches to people labeled with similar "diseases", then, if she takes the lesson to heart, she will be careful not to treat alcoholism as a disease in her explication (if that is possible). She reasonably decides against certain explications just in virtue of her newfound empirical knowledge. For what it's worth, any potential decision anybody makes is open to empirical revision in this sense. Perhaps that's not important, but if we were looking for distinctive senses in which philosophical claims could be empirically revised, this won't do.

I think (2) can be subsumed under the second sense of "empirical revision" in the post. If, as seems natural, we take moral consequences in this sense to be truth-apt items, then any theory containing a sentence with moral consequences (in the sense of (2)) also contains those consequences. Any empirical evidence for or against those consequences is empirical evidence for or against the theory. We might explicitly stipulate that our explication is not supposed to capture the moral facts. (One of the nice things about explications is that we can stipulate what aspects of the explicandum we want our explicatum to capture, refine, or ignore.) Thus someone who is interested in refining their conception of alcoholism, but unable to find a suitably non-normative term with the same descriptive content as "disease", would choose to define alcoholism as a disease, but would stipulate that their definition should be read without any of its moral consequences. But, without this sort of stipulation, the moral entailments of a theory are just like any of the other entailments - if the evidence says they are false, then the theory must go.

Similar comments go for (3) as for (1). If, in uttering a sentence, we take responsibility for a putative state of affairs having a high moral valence, and the evidence suggests that this state of affairs does not have the valence we took responsibility for it's having, then we should avoid uttering the sentence. If in saying someone has a certain sort of disease (which is the explicatum of "alcoholism"), we take responsibility for it being the case, say, that she is a fit object of condescension, then we ought not explicate "alcoholism" that way as long as the person is not a fit object of condescension. If the empirical evidence suggests that she is not (if, in fact, this is the sort of thing empirical evidence can suggest), then we should revise our list of potential explications in that light. But again, this is just the general sense of "revision" in which we should revise any course of action in light of information that it is morally misguided.

* - Or, more formally, if m(x) is the set of all moral conceptual entailments of all the atomic sentences containing the term x, then the ordered pair {m(t)-m(t`), m(t`)-m(t)}. Or we could restrict this to all the atomic sentences containing t or t` in a given theory of interest (e.g., a biological and psychological theory of alcohol).

Eric Schwitzgebel said...

I'm not sure about the formalism in your original post and in your versions (2) and (3) in the reply above. I'm somewhat inclined to think formalisms often mask more than they reveal, given their poor fit with our natural cognitive skills.

I agree with your take on (1). But then I don't think there's anything really distinctive about philosophy or philosophical claims. As Alison Gopnik and I put it in a 1998 article, "philosophy is just very theoretical anything" -- no distinctive method, no distinctive topic.