Does Anything Break Because it is Fragile?
ABSTRACT: I maintain that dispositions are not causally relevant to their manifestations. The paper begins with a negative argument, which is intended to undermine David Lewis recent attempt to restore causal potency to dispositions by identifying their instantiations with the instantiations of their causal bases. I conclude that Lewis attempt to vindicate the causal credentials of dispositions meets obstacles that are analogous to (though importantly different from) those that beset Donald Davidsons attempt to accord a causal role to the mental. I then consider an argument recently given by Frank Jackson against the causal relevance of dispositions (to their manifestations). Jacksons argument relies on a conception of dispositions that is not likely to be shared by those who defend their causal relevance. I sketch an alternative conception of dispositions that links them more closely to their causal bases, but argue that even on this model dispositions are causally impotent. The paper closes with a defense of the claim that dispositions, in spite of their causal irrelevance to their manifestations, are nevertheless causal-explanatorily relevant to them.
We regard dispositions as being causally responsible for their manifestations. We say that the glass broke because it was fragile, that the rubber band stretched because it is elastic, and that the arsenic killed him because it was lethal. Some philosophers have denied this. According to them, dispositions are causally irrelevant to the effects in terms of which they are defined.
This view was defended by Elizabeth Prior, Robert Pargetter and Frank Jackson, and has been (tentatively) endorsed by David Lewis. According to them, fragility is the second-order property of having some or other first-order property (e.g., a given molecular structure) that tends to cause breaking under certain circumstances. But then, they infer, it is this first-order feature (the `causal basis' of the glass's fragility), and not fragility itself, that is responsible for causing the breaking. Fragility is thus conceptually after the fact as concerns the causation of breaking: the glass counts as being fragile only in consequence of its having some other, first-order property that is causally responsible for its breaking when struck.
Lewis has always seemed uneasy with this view. He has called it a "disagreeable oddity" that must be dispatched if the identification of dispositions with second-order properties is to win our unequivocal support. In a recent paper, he takes himself to have done just that. He begins by saying that, "Sometimes, an event . . . is a having of a certain property by a certain thing"; and sometimes, he continues, "Two different properties are had in the same single event." Consider, for instance, the event that consists in the "having of the [first-order] causal basis" of the glass's disposition to break when struck. According to Lewis, "This same event is a having of the second-order property," (viz., fragility). That is, the glass's possession of the molecular structure which serves as the causal basis of its fragility is all there is to its being fragile. The `havings' of these two properties are one and the same entity. Thus, since the glass's possession of that molecular structure is a cause of its breaking, so too is its fragility.
Unfortunately, this attempt to restore causal potency to dispositions does not succeed. For suppose that some object (c) is F, and that its being F causes e to be G. It does not follow that if c's-being-F is the same thing as c's-being-H (if, in Lewis's terms, these two havings are the same event), then c's-being-H will be causally responsible for e's-being-G. To see why, suppose that I, weighing 165 pounds, tip the scales, and that anyone weighing more than 120 pounds would also tip them. Clearly, my having the property of weighing more than 120 pounds is efficacious with respect to tipping the scales. But I do not tip the scales merely by weighing more than 50 pounds if I had weighed only 55 pounds, the scales would not have tipped. So my having the property of weighing more than 50 pounds is not causally responsible for tipping the scales. And yet my weighing more than 120 pounds and my weighing more than 50 pounds are one and the same event, for my instantiation of the former property (by weighing 165 pounds) is all there is to my instantiation of the latter one.
Stephen Yablo advances a similar criticism against Cynthia and Graham Macdonald's attempt to vindicate the causal relevance of the mental. The Macdonalds couch their discussion in terms of "property instances", but their property instances are the same sorts of entities as Lewis's events. As with Lewis's events, two different properties can be had in the same single property instance. Thus, according to the Macdonalds, even though mental properties cannot be reduced to physical properties, each mental property instance is identical with a physical property instance; and since physical property instances are efficacious with respect to behavioral effects, so too are mental property instances., Against this view, Yablo argues that two properties can share the same property instance without both being causally relevant to that property instance's effects.
Putting his point in terms of the above example, although my weighing more than 120 pounds is efficacious with respect to tipping the scales, my weighing less than 180 pounds is surely not causally responsible for that effect. This is so in spite of the fact that in my case, these two properties are had in the same single property instance. The same sort of objection can be applied to Lewis's account, substituting "event" for "property instance".
To clarify, the criticism is not directed at the claim that the event (or property instance) of my weighing less than 180 pounds is a cause of my tipping the scales. It would be wrong to deny this claim, since its truth follows from the identity of the aforesaid event with the event of my weighing more than 120 pounds (which really did cause the scales to tip), together with the extensionality of the causal relation. Instead, the criticism is intended to show that it is not in virtue of its being an instantiation of the property of weighing less than 180 pounds that the event causes the scales to tip; and that its being a `having' of this property is therefore causally irrelevant to that effect. In short, Lewis's willingness to countenance events that incorporate more than one property leaves him open to the objection that not all of those properties need be relevant to the events' effects.
This criticism is similar to those that have been raised against Davidson's model of mental causation. Note however that the objection takes a different form when applied to Davidson, in light of his coarser-grained conception of events. The criticism of Davidson was that although event c both has property F and causes effect X, it does not follow that its having F is causally relevant to the production of X; it may cause X in virtue of one of its other properties, H. Given Lewis's more fine-grained conception of events, we can say that the having of F is what causes X, for the causing event, c, is simply identified with this `having'. But although c is a having of F and causes X, it does not follow that its being a having of F is relevant to that effect; it may cause X in virtue of its being a having of some other property, H.
The focus thus far has been on a negative argument, to the effect that Lewis has not vindicated the causal relevance of dispositions to their own manifestations. This is just what one would expect, since, as will now be argued, dispositions really are causally impotent.
Supporters of this view sometimes invoke the bogey of overdetermination. The worry is that if dispositions were efficacious, they would overdetermine the effects in terms of which they are defined. For example, glasses break when struck both because they have a property that serves as the causal basis of their fragility (e.g., a certain molecular structure) and because they are fragile. But this worry need not detain us, for, as others have noted, rampant overdetermination is only troubling when it requires that each effect has many different causally sufficient conditions by sheer coincidence. But since fragility is by definition possessed by anything that is such as to break when struck, the overdetermination of the glasses' breaking involves no such coincidence.
According to Jackson, this last consideration points the way to an argument against the efficacy of dispositions. After all (reasons Jackson), if fragility is simply defined as `being such as to be disposed to break when struck (or dropped or . . .)', then the liability to break when struck is essential to being fragile. But, as Hume taught us, types of state do not have their causal powers essentially. Hence, no causal power with respect to breaking is conferred upon a thing by virtue of its being fragile; so we should not number fragility among the causally relevant factors that help to bring about this effect.
Jackson's argument is convincing as far as it goes, but it may be questioned whether it goes far enough. He assumes that dispositions are definable as second-order properties that involve existential quantification over first-order properties. Thus Jackson, like Lewis, interprets, `being such as to be disposed to break when struck' to mean `having the second-order property of having some or other first-order property that tends to cause breaking when struck'. However, those who advocate the causal relevance of dispositions are likely to regard dispositions as being more closely wedded to their causally relevant, first-order bases than this model allows. Specifically, they may favour defining each disposition extensionally, as the simple enumeration of its various causal bases. Thus, fragility is identified with the second-order property of having property F1 or F2 or . . . Fn, where F1-Fn are the empirically possible causal bases of fragility. The causal connection to breaking when struck no longer figures essentially in the definition of `fragility', but has instead been downgraded to the status of a handy reference-fixer that is used for establishing reference to the properties, F1-Fn. `Being fragile' now amounts to no more than having one of these first-order properties. Jackson's argument has no force against the causal relevance of fragility thus construed. After all, given that the extensional definition contains no mention of the causal connection to breaking when struck, and given that none of the properties that it lists (F1-Fn) is essentially a typical cause of breaking when struck, it is hard to see why fragility should be thought to be so.
This view of dispositions is modeled on Jaegwon Kim's erstwhile view that each mental property is reducible to the disjunction of its empirically (or `physically') possible physical realizers. Unfortunately, once this parallel is noticed, it becomes evident that even though the strategy of extensional definition effects a tighter connection between dispositions and their causally relevant first-order bases, dispositions thus conceived do not inherit the causal powers of their first-order realizers. This is because such `wildly disjunctive' properties as the ones to which Kim would reduce mental features (and to which we have considered reducing dispositions) are not fit to appear in genuine causal laws. In saying this, one need not dispute the claim that disjunctions of properties, even infinitely long disjunctions, are themselves properties. The point is rather that even if they are properties, the possession of them in no way augments the causal powers of their bearers. For something can have the disjunctive property, F1 or F2 or . . . Fn (to which fragility has supposedly been reduced), only by having one of the properties that is cited in its disjuncts, say, the molecular structure F1. Any such thing will act as it does by virtue of having that property. The fact that F1 is part of some other, disjunctive property adds nothing to the thing's causal powers.
It may seem odd to deny causal relevance to a disjunctive property in spite of its appearance in generalizations that support counterfactuals and predictions. However, none of these claims in which it appears asserts a law-like, causal connectionbetween the disjunctive property and anything else, for none of these claims is confirmable by its instances. The importance of confirmability as a test of lawhood has been emphasised by David Owens. He notes that while genuine laws are susceptible of confirmation by their instances, generalizations that invoke wildly disjunctive properties in their antecedents and non-disjunctive features in their consequents are not. For example, consider the generalization,
Suppose we observe something that has F1 and that breaks when struck, and that thus provides an instance of (P). This does not count as evidence in favour of (P), since it equally supports the following `rival' of (P) (i.e., a claim that yields predictions that are contrary to those yielded by [P]): (Q) For all x, if x has F1 then it breaks when struck, and if x has F2 then it does not break when struck. In short, while the fact that something which is F1 is observed to break when struck certainly lends credence to the general claim that whatever is F1 breaks when struck, it gives no indication as to the behaviour of things that are F2, and thus equally supports any generalization that conjoins the claim that F1-things break when struck with any claim whatever concerning the behaviour of F2-things. To give equal support to all such generalizations is to confirm none of them.
Note that since any given instance of (P) will count as such an instance only by virtue of instantiating one of the disjuncts in (P)'s antecedent (as well as the property described in [P]'s consequent), we can always in like manner construct a rival of (P) that is equally supported by that instance, thereby showing that the instance at hand does not confirm (P). Since (P) is not confirmed by any of its instances, it is not a causal law.
It is interesting to note that in spite of the incapacity of wildly disjunctive properties to augment the causal powers of their bearers, the generalizations in which these disjunctive properties appear do support counterfactuals. For instance, (P) supports the claim that if the paperweight had been F1 or F2 or . . . Fn, then it would have broken when struck. (P) is also closely bound up with the counterfactual claim that if the glass had not been F1 or F2 or . . . Fn, then it would not have broken when struck. This should give us pause, for some philosophers have proposed the satisfaction of such counterfactuals as a test of causal relevance. For example, according to Ernest LePore and Barry Loewer, it is enough to vindicate the causal relevance of mental features merely to show that they appear in such counterfactuals as, "If he had not wanted a Coke, he would not have opened the fridge." The fact that wildly disjunctive properties can appear in the very same sorts of counterfactuals without thereby having their causal credentials authenticated indicates the weakness of this approach.
It must be concluded that dispositions are not causally relevant to their manifestations. Still, we must acknowledge that they are not entirely devoid of explanatory significance. We do in fact say that the glass broke because it was fragile, and that the rubber band stretched because it was elastic. But this explanatory significance is of a relatively low grade. It does not bring with it causal relevance. One may ask how a feature can be explanatorily relevant, and even causal-explanatorily relevant (i.e., relevant to a causal explanation), without being causally relevant simpliciter. Ned Block has shown how, taking dormitivity as his example of a disposition. According to Block, the claim that I fell asleep because I took a dormitive pill is causal-explanatory, "Because it rules out alternative causal explanations of my falling asleep." For instance, it rules out saying that I fell asleep because I had had no sleep the night before. The appeal to dormitivity rules out alternative explanations (and is thus causal-explanatorily relevant) because it "`brings in'" or "involves an appeal to" the first-order chemical property of the sleeping pill that is genuinely relevant to causing sleep. It is perhaps more accurate to say that our appeal to the pill's dormitivity merely locates the first-order causally relevant property (in the chemical composition of the pill) without specifying it, thereby `flagging' or outlining the causal path that culminated in my falling asleep. Switching metaphors, it helps to trace the contours of this causal chain without mentioning any of the causally relevant features from which it is forged. I dub such accounts "flagging" explanations.
It might be thought that the notion of a flagging explanation extensionalizes explanatory contexts. For it appears that one can mark out (`flag') the causal route that culminated in the effect that is to be explained, thereby ruling out alternative causal explanations, simply by referring to the cause in some way or other. It does not matter how one picks it out; as long as one does so, one will have succeeded in providing a flagging explanation. To see why this is not so, contrast the distinction between flagging explanations and explanations that cite causally relevant features with David-Hillel Ruben's distinction between giving an explanation of something and merely "implying that there is some explanation of that thing, without actually giving it." For example, I `give' an explanation when I say that the hurricane caused the loss of life, while I merely imply that there is some explanation of the loss of life when I say that the event reported in The Times on Tuesday is responsible for the fatalities. In making the latter claim, I refer to the cause of the deaths without explaining them, for I have not referred to the cause in such a way as to provide new information about the causal path that culminated in the loss of life: my claim, taken on its own (in isolation from any further claim about what it was that was reported in the newspaper on Tuesday), does not rule out any of the alternative causal paths that could have resulted in that event. As a result, I have merely indicated that an explanation can be had by differently describing the event to which I have referred. By contrast, I do rule out some candidate causes when I say that Ed fell asleep because he took a sleeping pill: I rule out any causal chain that does not proceed through the pill.
We can thus appeal to dispositions in explanation of their manifestation without being committed to their causal relevance. For our appeals to dispositions occur within intensional, explanatory contexts in which the causally relevant factors are merely located without being specified.