«1 Introduction Naturalistic philosophers of mind must assume some philosophy of science. For naturalism demands that we look to psychology—but to ...»
Psychological Explanation, Ontological Commitment,
and the Semantic view of Theories
Naturalistic philosophers of mind must assume some philosophy of science. For
naturalism demands that we look to psychology—but to be guided by psychological theories,
one must have some story about what theories are and how they work. In this way,
philosophy of mind was subtly guided by philosophy of science. For the past forty years,
mainstream philosophy of mind has implicitly endorsed the so-called ‘received’ or ‘axiomatic’ view of theories. On such a view, theories are sets of sentences formulated in first-order predicate logic, explanations are deductions from the theories, and the ontology of a theory can be read off from the predicates used in explanations.
The persistence of the received view in in philosophy of mind is surprising, given that few philosophers of science these days would endorse the it. An alternative, the so-called semantic view of theories, has become far more popular. With it comes a new view about explanation, and about ontological commitment more generally. One might therefore worry— as I do—that many problems in philosophy of mind are actually psuedoproblems introduced by an outdated notion of theories.
Philosophy of mind has seen some important moves beyond the axiomatic view and the corresponding view of explanation in recent years (Craver (2007) is a notable example). I think, however, that philosophy of mind — and especially the metaphysics of mind — has not fully appreciated how different the landscape looks when one moves away from the old view of theories. The new wave in philosophy of mind will involve re-importing some of these lessons from philosophy of science, and re-thinking some of the old puzzles that arose in the context of the axiomatic theory. What follows is a first step in that process, focusing on the key issue of explanation and ontological commitment.
2 Two Views about Explanation
2.1 Explanatory Literalism
Consider the following pairs of explanations:
(1) (a) The square peg failed to pass through the hole because its cross-section was longer than the diameter of the hole (b) The peg failed to pass through the hole because [extremely long description of atomic movements ] (2) (a) Klein got a ticket because he was driving over 60mph (b) Klein got a ticket because he was driving exactly 73mph (3) (a) Socrates died because he drank hemlock (b) Socrates died because he guzzled hemlock (4) (a) Esther ran because she was scared of the bee (b) Esther ran because [complicated neural description ] Many have the strong intuition that the first sentence in each pair is a better explanation that the second. This is true, note, even though the truth of the second sentence guarantees the truth of the first. I want to take that intuition for granted and explore two different stories about why that might be the case.
There is a well-loved account, tracing at least back to Hilary Putnam, for the superiority of some explanations. Explanation 1a, Putnam claimed, is clearly better because In this explanation certain relevant structural features of the situation are brought out. The geometrical features are brought out. It is relevant that a square one inch high is bigger than a circle one inch around. And the relationship between the size and shape of the peg and the size and the shape of the holes is relevant.
It is relevant that both the board and the peg are rigid under transportation.
And nothing else is relevant. The same explanation will go in any world (whatever the microstructure) in which these higher-level structural features are present. In that sense this explanation is autonomous (Putnam, 1975, p296) On Putnam’s story, (1a) refers to a higher-level property of the peg, the shape, that is most commensurate with the explanandum. Following Yablo, Bontly has called this the ‘Goldilocks Principle’: the peg’s shape is just enough (and no more) to cause its failure to pass; so too with all truly explanatory properties. (2a) is a better explanation because the property of my speed—being above the limit—was sufficient for a ticket; my exact speed was not required. (3a) is a better explanation than (3b) because it was drinking hemlock that was fatal, guzzled or not. (4a) is a better explanation than (4b) because the extra detail is irrelevant: Esther would have run no matter how her fear was instantiated.
Generality does not always make for better explanation. Consider:
This is both true and more general than (4a); nevertheless, it is an inferior explanation if Esther is scared only of bees but indifferent to flies. Rather, it is proportionality between higher-level cause and effect that picks out the most explanatory of the causally relevant properties.
says that good explanations are superior to rivals because they pick out a property that their rivals don’t, and that this property bears the right sort of relationship to the explanans. Our best explanations are thus ontologically committing. If a term φ appears in the best explanation of some phenomena, then we are, all things being equal, justified in believing that φ refers to some unique property. Hence the term ‘literalism’: one can read off the ontological commitments of a good explanation largely by taking it literally, and supposing that each term φ really is meant to refer to a corresponding property or entity.1
1 Note that one could hold a much stricter version of literalism, on which a predicate is ontologically committing only if it is ineliminable, or just in case it appears in the best overall axiomatization of the phenomena. I do not focus on these formulations for two reasons. First, in practice nobody actually adheres to this standard, because figuring out whether a predicate is ineliminable or part of the best axiomation tout court is
too difficult a task. If we held ourselves to such a high standard, the game would be up from the beginning:
no one should have confidence that their predicates refer, and so literalism would be a straw man. Second, formulations of the requirement in terms of axiomitization or the eliminability of predicates is so obviously derived from the axiomatic view of theories that the considerations presented in section 4 will apply directly.
Thanks to Mark Sprevak for pressing me on this point.
comfort to nonreductive physicalists. The fact that (4b) is inferior to (4a) suggests that even were psychology to be reduced to neuroscience, the resulting neural explanations would be inferior to the psychological ones because they would no longer refer to the most commensurate high-level properties. Further, the explanatory superiority of proportionate properties might lead us to suppose that we have a solution to the hoary causal exclusion argument. The causal exclusion argument says, in simplified terms, that mental and physical properties must (if distinct) compete for causal influence, and that a plausible physicalism should force us to assign causal priority to the physical one. Not so, literalism responds: both properties are causally relevant, but only the higher-level one counts as the cause. It does so because it is more proportionate, or commensurate with, or otherwise better fitted to the effect. Not only is the exclusion argument avoided, but the mental is given a certain causal priority over the physical. Nonreductive physicalism is saved. For this reason, various forms of literalism are increasingly popular in philosophy of mind and philosophy of neuroscience.
Finally, literalism is simply assumed as uncontroversial by many philosophers of mind.
The alternatives to literalism seem to be some sort of anti-naturalism or scientific anti-realism, neither of which are particularly attractive. That alone seems to be reason to accept it.
2.2 Explanatory Agnosticism Literalism is not the only account of explanatory goodness available to the naturalist, however. For each pair above, it is possible to account for the superiority of one of the explanations by appealing to facts about the language in which the explanations are couched while remaining provisionally neutral about the ontology one is thereby committed to. Call this agnosticism about explanatory goodness. The agnostic denies that we can move easily from language to ontology. Crudely put, the fact that a certain predicate appears in a superior explanation is no reason to believe that there is a property corresponding to that predicate.
I want to defend agnosticism about higher-level properties. Note that the position I favor is properly agnostic, rather than skeptical. I don’t want to take a stand on whether there are higher-level properties (or determinables, or whatever). Maybe there are. Maybe there aren’t. Rather, my claim is that in ordinary and scientific explanation, apparent reference to higher-level properties carries demands no ontological commitment to the existence of such properties. There may well be higher-level causes; I just don’t think that our best explanations are a good guide to what they are.
Agnosticism also has a certain prima facie plausibility. First, many have noted that shifts in the presumed interests of a listener can make a difference in the explanations that it is
appropriate to give (van Fraassen, 1980). Consider the explanation:
In certain contexts (historical/political ones say), explanation (5) is superior to either (3a) or (3b); in other contexts (physiological/medical ones), the reverse is true. Yet presumably the facts about what properties are involved and their commensurability remain unchanged.
A defense of agnosticism is strengthened by reflection on conversational pragmatics and their role in shaping our intuitions about explanations. We find the more general explanations of the pair more acceptable, says the agnostic, because of pragmatic constraints on the descriptive form of explanations (and not because they refer to more commensurate properties). These pragmatic constraints—and in particular, the Gricean maxims that underly cooperative conversation—may favor a more general description of the same circumstance, but that description is not superior because it picks out a more general property.
A few quick examples for how this might look. (2a) is superior to (2b) because the Gricean maxim of Relevance tells me to give only such information as is relevant to my hearer’s interests (Grice, 1989). My wife wants to know why I got another ticket; the fact that I broke the speed limit is sufficient to satisfy her interests, and the specific speed is (we assume) irrelevant to her interests in the conversation. Similarly, the Gricean maxims of Quantity and Quality should forbid me from giving (4b) as an explanation when the equally effective and much shorter description (4a) is available. Indeed, to give (4b) would (on the assumption that I’m being cooperative) produce several false implicatures: that the extra detail is relevant, in the sense that counterfactuals involving small changes to Esther’s neural state would result in her calm, or that I have good evidence in a particular case for the complicated neural process at which I have hinted. Neither of these is likely to be true. So to give (4b) would be misleading, in the sense that I would implicate something false to my listener.
Nevertheless, it is true that the complicated neural process was the cause of her flight, not some additional distinct higher-level property.
Changing conversational demands can produce shifts in explanatory goodness without shifts in interest. The patrolman is testifying. The judge, like my wife, wants to know why I got a ticket. The patrolman would have ticketed me for any speed above 60; if my speed instantiated a higher-level property of having a speed above 60mph that was most proportionate to my ticketing before, it continues to do so now. Yet it would now be more appropriate for the patrolman to utter (2b) than (2a). Why? The patrolman’s testimony must justify his ticket-giving. Uttering (2b) implies that he has precise information about my speed—which is to say that he determined my speed by some suitably precise measurement.
To utter (2a) would give the false implication that he doesn’t have such information (since by the maxim of Relevance he should be as specific as necessary for the demands of the conversation). This implication is cancelable (“He was going over 60mph—in fact, I clocked him at exactly 73mph”), but in ordinary circumstances the patrolman can achieve his ends through the parsimonious (2b).
each pair above are superior, but explain that superiority by appeal to language and conversational context, not the world. Thomas Bontly has argued (convincingly in my opinion) that the implicatures of many causal claims are nondetachable and cancelable, the standard marks of conversational implicatures.2 Antecedents of the strategy might be found in Kim’s insistence that there are only higher-level predicates, not higher-level properties (1998), and in Lewis’ remarks on the pragmatics of causal explanation (1986).
to them: he can say that the explanations cite facts that are causally relevant, and that shifts in context alter which causally relevant factors are appropriate to cite. But we should be suspicious of this: the evidence for literalism above was supposed to be our judgments about the appropriateness or inappropriateness of single explanations. That confidence should be undermined if we find serious context-sensitive effects on our judgments of appropriateness.
2 See especially (Bontly, 2005, p.343). I am indebted to Bontly’s article for prompting many of the reflections in this section.