Colin Klein --- Philosophy of Mind and the Semantic View of Theories

[New Waves in Philosophy of Mind]

Philosophy of Mind and the Semantic View of Theories

Colin Klein

Many debates in philosophy of mind assume that the predicates used in our best explanations are a guide to the psychological ontology we ought to adopt. I argue first that many of the intuitions used to support this ontologically committal stance can be accounted for by appeal to conversational pragmatics. On this alternate view, I show that intuitions in favor of, e.g., higher-level causation can be explained without appeal to higher-level causes. I then argue that the ontologically committal view relies on an axiomatic view of theories, on which scientific theories are identified with sets of laws formulated using a canonical set of predicates. Though once popular, this view of theories has been largely abandoned by philosophers of science. The most popular current contender is the semantic view, on which scientific theories are best understood as sets of models that indirectly represent target phenomena. The semantic view of theories differs from the older received view in several crucial aspects.  Importantly, I argue, it breaks the link between the language in which we give explanations and the ontological structure of the world: our best explanations need not commit us to properties corresponding to the predicates used within them.

A PDF of the paper is ready to view and download in the attachment below. 

A direct link to the PDF: http://goo.gl/awO0y

[Charles Rathkopf]

Hi Colin,

Great paper. The semantic view certainly has consequences for lots of traditional ideas in philosophy of mind. The semantic view divorces scientific commitments from the language in which they are expressed, and this idea is very important for understanding debates about explanation. Still, I was a little unclear about the argument in the first part. Why does the commitment to proportionality amount to the thesis you call literalism? Why can't you hold that (causal) explanations do need to be proportional, but that the proportionality should be located in the model? This seems like a good option because it allows us to deal with the ever-present problem of causal relevance. We cannot treat all causes as equally explanatory, and the proportionality principle is a good way of singling out those that do the significant work. So I guess I don't see the tension between the semantic view of theories and the commitment to proportionality.
The semantic view says that in general, models are more important than theories. So the semantic view is naturally compatible with the view that models, rather than theories, tend to be the most important part of the explanans. But your alternative to literalism, which you call agnosticism, sounds strongly wedded to language. You talk about explaining "by appealing to the language in which the explanations are couched..."
I agree with you that we should be wary of taking the language too literally in scientific explanations, and that we should avoid any simplistic ontological commitment. But the benefit of the semantic view is precisely that it gives us the freedom to talk about a structure that resembles (or is isomorphic to) the target phenomenon. These kinds of model-based representation relations allow for some flexibility that standard linguistic theory does not. Going beyond this point, things get complicated. But I'm inclined to accept a minimal kind of ontological commitment in which something in the target phenomenon comes close to the structure exhibited by the model. Of course, this kind of realism should be applied selectively, and I'm inclined to be doubtful that many models in cognitive psychology make the cut.

[Liz Irvine]

Hi Colin, 

I also really enjoyed the paper! Just a few questions...

I was confused by the Gna Gk example. Surely, for the purposes of many explanations, Gna/Gk is a 'distinct' causal property - it matters whether it's at or above 1. The precise values of Gna and Gk don't matter. So, whatever 'real properties' or 'real causes' are here, Gna/Gk surely is one, unless we want to go the road of 'crude old fashioned reductionism', which is not particularly appealing, (at least to me, but then I'm happy to go the interventionist route).

You also suggest that if we allow Gna/Gk to be a distinct high level property, then we also have to allow 2Gna/2Gk etc. But there would be reason why anyone would proffer these as terms in an explanation. So maybe the move here is just to echo that of Charles Rathkopf above and say that proportionality, which may also cover some norms of explanation, should reside in the model. 

In general, it seems like once you have a 'good' explanation (or a 'good' model), that some kind of minimal ontological commitment is licensed. The problem seems to lie in identifying 'good' models, not necessarily that ontological commitments from them need to be barred. Clearly there are many models in cognitive psychology where one would not want to make any ontological commitments, but surely this is a problem with the models, not with the step of making ontological commitments? I'm just wondering how endemic this problem is, or if there are other ways of flagging the problem that are discipline specific. If we normally aren't warranted in making ontological commitments, then maybe this illustrates something about scientific methodology in general. But if the problems are more discipline-specific, then there may be particular methods or styles of model-building that just don't warrant ontological commitments (and scientists may well know why), in which case we need to careful about these methods and models in particular. Of course, being sensitive to specific types of models and model-building doesn't make things any easier for philosophers of mind, but I suspect neither is a blanket ban on ontological commitments for philosophers of science.

[cvklein]

Thanks for the great comments, Charles and Liz! I have to run to a thesis defense soon, so this may get spread over a few comments--there's a lot to talk about. 

First, for Charles:  there's a dialectical trickiness in this paper, and I think that you've put your finger on one part where I'm having trouble getting the argument clear. Let me run the proportionality argument by you in a more compact (and maybe slightly different) way and see what you think: There is a set S of intuitions that favor the proportionality argument for higher-level causes (S = our judgments about the examples at the beginning and those like it). I claim that the pragmatics of explanation are such that we would have S *regardless* of whether there are higher-level causes or not. So the fact that we have S can't be part of an argument for higher-level causes. Does that sound more plausible? 

You're right that one of the attractions of the Semantic view is that it allows for flexible model-world relations, and there are some nice papers on this in the context of Philosophy of mind (see, e.g., PGS's paper on folk psychology, or a lot of Giere. I'm obviously indebted to both). But I think the potentially more revolutionary aspect of it is the flexible language-model relation -- that is, the fact that you can potentially specific models in a variety of ways, including formulations that use substantially different predicates or are even nonlinguistic entirely. And that's what I think breaks the link between the language used in explanations and the ontological commitments of a theory. In some sense, the overarching theme of the paper is that many of the arguments that seem natural in philosophy of mind only seem so because of a lingering commitment to an axiomatic view of theories. They may be defensible on a semantic view, but they need defense if so. 

You raise the tricky question of realism at the end. Every time I give variants on this theme, I get accused of being an anti-realist. I'm not, at least not yet (I do feel the pull of Constructive Empiricism, though). My standard answer is to appeal to something like Hacking's entity realism to settle the question of what things exist, where that turns out to be a largely separate question from which theories are good at predicting and explaining. I'm not sure whether this is a viable position, though, or whether I should just drop down to C.E. and be done with it. 

Finally, to plug my own work a bit, I have a cleaner version of this same argument related to MR that's coming out in Phil Studies: 

(online first) http://link.springer.com/article/10.1007%2Fs11098-011-9839-6    

(free final draft) http://tigger.uic.edu/~cvklein/papers/MRSemantic.pdf

-Colin 

[cvklein]

Ok, round 2: Liz! So I think the first move I ought to make is to ask why you have the intuition that "once you have a 'good' explanation (or a 'good' model), that some kind of minimal ontological commitment is licensed." In some sense, the whole point of the paper is a twofold attack on this: (a) I can explain all of the intuitions that you have about proportionality, etc. without having to assume this, so the source of the intuition must be elsewhere, and (b) here's an explanation: that move seems obvious given realism + an axiomatic view of theories, and it seems especially obvious because that's what we all got as young philosophers when we read Fodor and friends.

As for the Gna/Gk example. You say "But there would be reason why anyone would proffer these as terms in an explanation." This is awkward, though, if you want to be a realist. For surely what exists doesn't depend on what people talk about. So the expression 2Gna/2Gk either refers or not. If it does refer we have an explosion, and if it doesn't I don't see any good argument as to why it doesn't refer that doesn't also impugn Gna/Gk. 

(Though to step back a bit: my real worry is that the defenses of ontological commitment from explanations always start with predicates but then quickly move to encompass whole expressions like Gna/Gk (or H2O, for that matter). And that seems to ignore the important role that the internal structure of those expression plays.) 

Must run to a defense. Enjoying this! 

Colin 

[Carrie Figdor]

Hi Colin,

Nice paper as usual, though I'm happy to say that I'm not convinced. The basic metaphysical issue here involves commitment to properties. You don't have to think that every predicate picks out a property to think that psychological predicates do, and that various psychological models can pick out the same psychological properties. (These need not be couched in folk-psychological terms, or think that there are folk-psychological properties -- that's a separate issue.)

The Hodgkin-Huxley model can't do the dialectical work you want it to do because it doesn't address this basic issue. I'm skeptical that anyone needs to think that the equation refers to a property; introducing a predicate doesn't change that (and really, what's the motivation?). And you don't have to think models, levels or sciences have strict boundaries to think there are psychological models of real psychological phenomena and their relationships.

There's a lot more to say about all these issues. Sorry so terse!

FYI typo at 1.2 "casually". A common one that spellcheck never gets.

Carrie

[Dan Weiskopf]

Hi Colin. I have a quick clarificatory question and a worry about whether the problems for literalism don't also recur if we adopt the semantic view.

If I understand correctly, literalism says that our best explanations are ontologically committing in a very particular way: one can read off the explanatorily relevant properties more or less directly from the linguistic form of the explanation itself, e.g., the predicates and other terms it contains. (Shades here of the Quinean notion linking ontology with quantification.)

As I read the paper there are two argumentative goals, but I’m not sure how they’re related. There’s the bit where you argue against literalism and for agnosticism (roughly, that our best explanations are ontologically neutral in the sense that no ontological commitments can simply be read off of them). And then there’s the bit at the end where you advocate for some version of the semantic view of theories—though really I think you want to advocate for the centrality of models in science, rather than the classical semantic view.

For clarification, I take it that you’re associating literalism with the axiomatic view of theories, so undermining literalism is supposed to support the move to the semantic view. Is this correct? (As a side note, if the only support for literalism was the axiomatic view, that would automatically count against it in my opinion, since effectively nobody believes the axiomatic view anymore. And no one should.)

Suppose we accept some form of model-based explanation in science. This involves taking models to be the central descriptive and explanatory constructs in science, rather than theories. And suppose we read models in something like Giere’s way, as abstracta that bear various similarity relations to real target systems in the world. Does this help avoid the objections to literalism? I think it doesn’t.

To see this, first note that we can couch these models in terms of either high-level or low-level properties, just as we can the descriptions that give rise to them. Issues about proportionality and explanatory aptness, then, can just as easily arise when we are considering models as when we are considering theories. The issues now just turn on which abstract entities make it into the model, rather than which terms comprise the sentences of the theory; but in either case we can raise the exact same question: how are these elements of the model/theory related to the real structures, properties, and causal powers in the target system?

To put the point more directly and forcefully, a model may fit or fail to fit the world in various ways. (See Giere’s discussion of similarity mapping here.) Even a model that is successful in various ways may not correspond well with the structure of a real world system. To know whether a model’s success is due to its tracking real structures, we need to adopt a range of methods to confirm the existence and nature of the entities that the model talks about. We cannot simply read off from the structure of the model any facts about the target system. This is the equivalent of treating models literally, and it is just as much a mistake here as when we were dealing with linguistically formulated theories.

So my question is: how does moving to models help us  avoid the problem of literalism, when we face the very same sort of mapping problem in deciding whether a model is sufficiently similar to the real world target system that it is intended to characterize? The ontological commitments of good models are no more transparent than are those of theories.

[cvklein]

Dan, 

Good points, and you're quite right about the two different argumentative goals. I need to think through your points some more, but I think you've characterized the paper pretty well. 

I will say this as for the last point --- what people find frustrating about the papers I've written on this is that I don't tell a positive story about ontological commitment. So the deeper implication of agnosticism, the one I think I might ultimately be committed to, is that we can do a lot of science without actually worrying about ontological commitment. This is where I start sounding like van Fraassen, basically, which is a bit disconcerting because I set out long ago to sound like Kim and Heil. (That's the other way you can take this, that others have picked up on --- that the only real properties are those of basic physics, and the rest of the predicates are just re-descriptions. I've gotten over any lingering physics envy, though, so that seems unattractive). 

-c

[Mark Sprevak]

Hi Colin!! This was a great paper—really interesting! It relates to some issues on ontological commitment that I’ve been thinking about too recently.

I confess that I had the same worries as Dan that I couldn’t quite see how the two bits of the paper hang together—the choice of literalism vs. agnosticism just seems independent of the semantic model of scientific theories.

I was also a bit worried about the construal of literalism.

I thought that literalism was a bit more exacting than you make out. Literalism says that an ontological commitment appears when you have a term that ineliminably contributes explanatory value. In other words, if you were to eliminate the term from science, or try to paraphrase it away, you end up with a significantly worse science—net explanatory value would go significantly down.

If a term passes this criterion, then this shows that it does ‘explanatory work’, which literalism assumes is the hallmark of hooking onto something real in the world.

I guess that I wasn’t sure how the example you gave threatened this.

On the HH-model, the ratio GNa / Gk doesn’t seem to be ontologically committing for a literalist, since since this ratio talk can easily be paraphrased without incurring explanatory loss (e.g. as the ratio of two conductances).

In contrast, if one were to eliminate or paraphrase talk of high-level geometrical properties (the shape of the square peg), that would have a massive negative effect on the kinds of explanations one could give in science—they’d end up being horribly and infinitely disjunctive in many cases.

So GNa / Gk fails the test of being explanatorily ineliminable, while Putnam’s geometrical properties pass that test...?

[Thomas Polger]

Colin, while confessing that I only skimmed this paper, I want to give three cheers for the general project.  I'm familiar with your Phil Studies paper on MR and the semantic view.  Details of this paper aside--not, of course, that they are not important--I am wholly in favor of pretty much any and every way to get metaphysicians and metaphysicians of mind to stop thinking that Quine is right about ontological commitment in the sciences, Lewis is right about theory structure, and Fodor is right about explanation an reduction.  I share you ambivalence about all the features that are typically associated with model-based science, e.g., tendency towards antirealism. But it's got to be good that we're thinking hard about old questions through a different lens.  Yeah!