The problem that my essay addresses is mental causation and the extent. In the first part I will discuss Davidson’s solution to this problem, his anomalous monism, that shares the assumptions of his general account of events and causation: extensionality and “super-fragility” (a token view of events). My thesis is that Davidson’s account of events does not permit the distinction between predicates and properties (distinction which is captured by Kim’s distinction between causal relevance and causal efficacy) and that this has the consequence that the mental cannot be causally effective any more than any arbitrary classificatory schema of events. I will also present a criticism of Davidson’s account of events. The second part of my essay will discuss Kim’s solution to the problem of mental causation, and his distinction between weak and strong supervenience. I will suggest that Kim’s strong supervenience is a form of de dicto supervenience and will develop the idea of quasi-necessity, that is, in my opinion, an alternative to both Davidson’s and Kim’s solutions.
1. Causal Relevance and Causal Efficacy
The starting point of my essay is Kim's distinction between causal relevance and causal efficacy, as pressented in the third section of "Can Supervenience Save Anomaluous Monism?". The context in which Kim makes this distinction is the analysis of the accusation that Davidson's anomalous monism is a form of epiphenomenalism, making the mental causally inert. Davidson's reply, as expressed in Thinking Causes, was that, given supervenience, mental properties are not causally inert: it is impossible to change the mental properties of an event without changing its physical properties. Indeed, according to Davidson, it is events that are causally effective: causation is a two-place relation between two causal relata (the events) and this relation is extensional, i.e. it holds or does not hold between events no matter how they are described. There is thus no place, according to Davidson, for an event being a cause of another event qua something, as instantiating a certain definite description and not other: events are not causally effective as mental events or as physical events. (Thus, anomaluous monism is not properly-speaking a form of materialism, but rather of monism). There are no two types of events, mental and physical: the mental and the physical are two types of properties, or predicates; two schemata of classification of the events. The basic difference between these two schemata is that mental classification of events (unlike the physical one) obeys norms of rationality, while the physical classification (unlike the mental one) allows strict laws. Events thus are the relata of "rational" correlations insofar as they are mental and are the relata of nomological correlations insofar as they are physical. As for causal relations themselves, they do not necessarily obey strict laws any more than they obey rational norms: but since all events are also physical events, for any causal relation there is a pair of definite descriptions of the causal relata such that it instantiates a law. Davidson's defense (as expressed in Thinking Causes) of his view of causation as a two-place relation between events is that if we believed diferently, we would be committed to the view that causation is intensional; that events cause other events qua something; that causation is dependent on language and that events (and their causal relations) "care about how we describe them".
In this context, Kim (in Can Supervenience…?) replies that Davidson does not distinguish between the linguistic status of causation ("events cause other events only under certain descriptions") and what I call, for lack of a better term, the implicitly general status of causation ("events cause other events insofar as they belong to certain kinds"). While the former can unproblematically be rejected as wrong (our renaming events isn’t enough to change causal relations), the latter seems much more acceptable:
He [Davidson] is anxious to defend causation as an extensional binary relation between whose relata are concrete events ("no matter how described"). But none of this has much to do with the main issue at hand, and getting rid of these admittedly sinelegant locutions ["c qua P causes e qua M; c under description D causes e under description D"] will not make it go away. The issue has always been the causal efficacy of properties of events – no matter how they, the events or the properties, are described…To talk about the role of properties in causation we don’t need to introduce the qua locution or any other multi-termed causal relation…all that is necessary is the recognition that it makes sense to ask questions of the form 'What is it about events c and e that makes it the case that c is a cause of e?" (Kim, Can Supervenience…? , p. 22-23).
In the fifth section of the same article, Kim introduces the distinction between causal relevance and causal efficacy. He writes that
"An epiphenomenalist may argue, mimicking Davidson, that on his view mental properties are indeed causally relevant, since, according to his doctrine, what mental properties an event has makes a diference to what physical properties it has, and physical properties are causally efficacious. But that doesn’t mean that he contradicts himself in refusing to allow causal efficacy to mental properties. If this is right, supervenience can at best show that mental properties are causally relevant, not that they are causally efficacious". (Can Supervenience…?, p. 23)
As I understand Kim's distinction between causal relevance and causal efficacy, causal relevance is that property of a device, belonging to a certain classificatory schema, that enables us to pick out, an event in such a way that a different device of the same classificatory schema would pick out a different event – at least as long as the classificatory schema itself is not questioned. Causal efficacy is, on the contrary, a property of an event such that if the event had not had it, no matter how we described it, the event would have occupied a different place in the causal space. (By "place in the causal space" I mean "the causes and effects of an event"). If the bearer, or intended bearer, of causal relevance and causal efficacy are properties of events, then causal relevance and causal efficacy are meta-properties: properties of properties.
2. Properties and predicates: predicates are the least essential properties
If what being causally relevant means is nothing more than selecting an event in a satisfactory way, such that a different characterization would have selected a different event, then the most obvious candidates for the position of being causally relevant are predicates and definite descriptions – I mean, predicates and definite descriptions correctly used. But what holds about predicates when we deal with causation, does not necessarily coincide with what holds about properties. The reason is very simple: that predicates are mere linguistic devices. A predicate, like a name, means nothing more than that it happens to pick out the object (or event) to which it is attached. Nothing prevents me from inventing an artificial language in which I will denote all objects (or events) with names chosen by me; and I may perhaps choose them in such a way as to obey supervenience with respect to the predicates (or names) of a pre-existing, in this case subvenient, level of language. This seems, however, to be too shallow a reason to consider that the predicates of my artificial language express properties that are causally effective. The new, artificial language will obviously be causally inert: what it should be said about it is not that it that it expresses properties with no causal role, but rather that it expresses no properties at all, except the property that the objects and/or events are denoted by the predicates of the new artificial language. The only property that a linguistic entity, as such, expresses is the tautological, implicit, property that the non-linguistic entity to which it attaches happens to be designated by it. A linguistic entity attached to an object or event does not add any new information, any new piece of knowledge to what we already know about that object or event, except the tautological one that I happen to call that object or event by means of it. If this linguistic entity is used with a considerable degree of success, then it may become indeed useful as a device to pick out non-linguistic entities (objects or events); but its being a mere device to pick out entities means that it is dispensable, that it does not necessarily characterize the non-linguistic entity that it characterizes; that it is not essential to it.
Causal relevance characterizes predicates used in a certain way. Causal relevance is thus a meta-linguistic property. Causal relevance is about an event causing another event under a certain description. Causal effectiveness is about an event causing an event insofar as it has certain properties, insofar as it belongs to a certain class. Causal relevance is a metalinguistic property; causal effectiveness is an ontological (meta-ontological, if you want) property. Causal relevance is about how we should pick out events in such a way as to appropriately reflect their causal behaviour. Causal effectiveness is about what an event needs to have/to be in order to occupy its place in the causal space. Davidson's supervenience makes room for causal relevance of the mental insofar as it makes room for the usefuleness of using a certain sort of predicates (in this case, the mental ones) to pick out the events that occupy a certain place in the causal space. But the disadvantage of this account is that it may (too easily) be generalized to any arbitrary classificatory scheme of events. To put in simpler words: causal relevance is a property of arbitrary predicates, while causal efficacy is a property of properties. Kim rightly accuses Davidson of maintaining causal relevance (i.e. predicate-relevance) while neglecting causal efficacy (i.e property-efficacy). Davidson's answer is that there is no difference between causal relevance (of predicates) and causal efficacy (of properties); Davidson's answer is that once we have causal relevance (of predicates) we get causal efficacy (of properties). But this answer obliterates the distinction between properties and predicates; it too simply assumes that any predicate belonging to a certain classificatory schema is causally relevant. In my opinion, this conclusion is wrong: a predicate is not causally effective, or it is – at any rate – less effective causally than a property. The reason is that a predicate qua predicate may be arbitrarily attached to an event (or an object), while properties characterize less arbitrarily the events (or objects) to which they are attached.
We may pursue this considerations a little bit and gradualize. Call a mere predicate a predicate that is as inessential as it can be to the object to which it is attached; that does not add any new information beyond what was already known about the non-linguistic entity to which it is attached. The event could have remained virtually the same without that predicate: if that predicate had lacked, the event would have been almost the same: the only diference would have been the tautological one that there is one less linguistic device to call the said event. I think that this property (of being called so-and-so) is the farthest possible property from an essential property. At the opposite pole, we have essential properties: properties that are such that the object could not have been without having them. If the object or event would have lacked an essential property, it would have been another object or another event. Between these extremes, we have properties that are almost inessential: that is, properties that are such that their absence would have turned the event into a slightly different event and properties that are quasi essential, that is, their absence (and/or their replacement with a different property) would have turned the event into a very different event.
Complications (and refinements) may arise once we ask whether these remarks hold for tokens or for types. What I hold is that the distinction between essential, almost inessential and quasi-essential properties applies to objects (or events) that are considered as token-as-types. A “token-as-type” is a token that is considered (described) in such a way that its being classified along other tokens under the same type is permitted. A token-as-type is thus different from a token-as-token: a token-as-token is a token that is described in such a way that it cannot be classified under the same type along with other types. If we give names to objects (or events) we consider them as token-as-tokens; if we give them definite descriptions we consider them as token-as-types. (It is easy to realize that in the case of tokens-as-tokens all properties would be essential). A token-as-type has essential, quasi-essential and almost inessential properties. It lacks, however, contingent properties: a contingent property of a token-as-type would be a property such that, if the object lacked it or had another property in its stead, it would remain the same token-as-type. But this is obviously absurd. These remarks also hold for types: by a type I do not mean a class of tokens, but rather an idealization of tokens. In order for a type to have essential, quasi-essential and almost contingent properties, it needs to be considered as a type-as-species: that is, it needs to be described (by a definite description) in such a way that it is apparent under which species it may be classified, along with other types.
A type-as-species’s essential properties are thus those properties whose absence or change would turn the type into a type of a different species. A type-as-species’ quasi-essential properties are such that their absence or change would turn the type-as-species into a type-as-species of the same species, but very different from the original type-as-species. A type-as-species’ almost contingent properties are such that their absence or change would turn the type-as-species into an almost similar type-as-species of the same species. Like in the case of tokens-as-types, types-as-species have no contingent properties: a contingent property would be such that its change or absence would turn the type-as-species into exactly the same type-as-species. (Types-as-species do not have properties that are not essential to their tokens-as-types: the letter “M” , for example, does not have the property that it it written in Times New Roman).
The rationale for considering tokens as tokens-as-types and types as types-as-species (and so on: species-as-genera) is that if all we had were tokens-as-tokens, it would be unintelligible how to classify several tokens under one and the same type; likewise, if we only had types-as-types, it would be unintelligible how to classify several types under the same species, and so on. (This is related to a difficulty with Davidson’s view of events that I will mention in the seventh section of this essay).
In the light of these remarks, the status of the causal efficacy of predicates may be easier to understand.
Causal Efficacy is a gradual matter. It is, inter alia, a measure of how essential a property is to the event that it characterizes if this event enters into causal relations. I am not sure that "being essential to an event" is the only factor that matter in determining how much a property may affect a causal relation. Perhaps there are almost contingent properties of events that still heavily affect the event's place in the causal space: they are contingent in that they do not affect very much the event's identity, but they may affect very much the event's causal behaviour. These properties are almost contingent to an event's identity, but almost essential to an event's causal behaviour. But let's assume there are no such properties: this is a simplification that Davidson would have endorsed since he holds that events are nothing but their place in the causal space. (i.e., their causes and effects). Under this assumption, I take it unproblematic that affecting very much the place of an event in the causal space is tantamount with being essential to an event's identity. Thus, being causally effective is a matter of degree in the same way as being essential to an event is a matter of degree. The more essential a property is to an event, the more effective it is causally.
Mere causal relevance is thus the minimal degree of causal efficacy since properties that are merely causally relevant are arbitrary predicates. (They are not arbitrary in the sense that we could change them as we like while remaining in the same classificatory schema; but they are arbitrary in the sense that the schema is itself arbitrary: we could change the schema as we like, replacing it with another schema or displacing it outright). Predicates are indeed properties: but to the extent that they are properties, their causal role is minimal, since they are almost contingent to the events they characterize. Before passing to the next section, one should pause to make the following remarks:
(1) One should not confuse a predicate's being a property with a predicate's denoting a property. Insofar as they are causally effective, predicates are properties. But they are minimally effective since they are almost contingent properties of the causal relata. Insofar as they denote properties, predicates are causally relevant. But the transition from causal relevance of predicates to the causal effectiveness of the properties denoted (or expressed) by them may be authorized only provided that
(1) Supervenience itself does not guarantee an interesting degree of causal efficacy of the supervenient properties on the subvenient ones. Since the supervenient properties may be mere arbitrary predicates (like in the situation in which I invent an artificial language such that its terms supervene on the terms in a previously established, natural language), supervenience is not a sufficient condition for causal relevance .
In such a situation, causal relevance does not guarantee an interesting degree of causal effectiveness because the predicates that are causally relevant do not denote properties that are to a considerable degree causally effective. Mere causal relevance is a property of arbitrary predicates: and arbitrary predicates are the least essential, the most contingent properties of an event. That is, a mere predicate may be replaced, deleted from the description of an event yielding almost no change in the identity of the event or in its place in the causal space.
3. Davidson's conception of events makes him insensitive to the distinction between predicates and properties, between essential and unessential properties
It is obvious that the distinction between mere predicates and essential properties is understandable only under the assumption that the objects and/or events to which the properties/predicates apply are tokens-as-types. If we lived in a world of mere tokens, then any property, be it as inessential as it may, would be such that the object could not have been the same object without having it. As we saw, even a mere predicate, no matter how arbitrary it is, expresses a minimal property: the property that I happen to call an object by means of it. For tokens-as-tokens, all properties are essential: any change of property would change the token-as-token and - since there is no type above the tokens – there would be no measure of how much this change would affect the token-as-token, i.e. for how much the new token-as-token would be different from the old token-as-token. In a world of tokens-as-tokens, there is nothing more beyond mere occurrences: in such a Heraclitean world, non-linguistic entities cannot survive beyond the mere occurrence of their being designated in language. The same holds for a world of types-as-types: if all we had were types, we would be able to classify tokens-as-types under the types, but unable to classify the types under species and genera. What we need in order to account for essential and contingent properties, and for a gradualization of this distinction, is a world of tokens-as-types, of types-as-species, of species-as-genera and so on.
It is thus understandable why in a world of tokens-as-tokens it makes not sense to distinguish between mere predicates and essential properties. This is the reason why, in a world of tokens-as-tokens, it makes no sense to distinguish between linguistic devices of picking out entities and entities themselves: linguistic devices express the property that they designate the entities: this is the minimal property that they linguistic devices are guaranteed to express even if they are used only once. But since non-linguistic entities are themselves nothing beyond their momentary occurrence, linguistic devices deserve to be considered indispensable to entities with the same title as the other properties: there is no situation in which an entity exists having the essential properties but not the inessential properties because there is only one occasion in which the entity is: entities are token-entities, occurrences of entities and nothing beyond.
In the light of the above considerations, it becomes understandable why Davidson should not distinguish between properties and predicates; between essential and inessential properties of events. The reason is that this distinction makes sense only given a background of type-events. This also explains why he does not distinguish between causal relevance and causal efficacy: he is obliged to equating causal relevance with causal efficacy because his conception of events commits him to obliterating the distinction between essential and nonessential properties of events. Causal relevance is a property of something that picks out causal relata in an appropriate way: causal efficacy is a property of something that is indispensable to a causal relatum if it is to be that causal relatum. But if one thought that causal efficacy is nothing more than causal relevance, one would be committed to thinking that all properties are nothing more than predicates. The conclusion of the second section of my essay was that predicates are indeed properties, but they are minimal properties, the most inessential properties that can be. They are inessential because they are arbitrary . Holding that all properties are equally essential (or inessential) has the inconvenient that it is possible to invent new, arbitrary properties (by inventing new, arbitrary predicates) that will thus have the same causal efficacy-relevance as the properties that we intuitively think are essential properties. This is, I hold, a major inconvenient of Davidson's conception of causation. It stems from his conception of events and it cannot be repaired unless this latter conception is amended. His events are token-as-token events.
Davidson indeed writes about the "mental" as being just one more classificatory scheme added to the physical. This (plus supervenience) makes the mental indeed causally relevant (in Kim's sense) but it does not make it causally efficacious. I have in mind the example of table-events in Thinking Causes, p. 12:
"if causal relations inhere in the particular events and objects, then the way those events and objects are described, and the properties we happen to employ to pick them out or characterize them, cannot affect what they have" (Thinking Causes p. 8)
and also protests against the accusation that he makes the mental causally irrelevant (and inefficacious) in that the removal of all mental properties (and, I take, predicates) from the descriptions of the events would not alter their causes. It is curious since, on the one hand, it is very plausible that removing the predicates (and definite descriptions) of events would not alter their causal relations – this is, after all, what Davidson holds while claiming that causation is an extensional relation and that events do not enter into causal relations as mental events any more than they enter as physical events; on the other hand, replacing a mental definite description with another mental definite description entails replacing some physical descriptions of the event. So predicates, definite descriptions, properties both have a role in causation (since modifying them has consequences about the place the events they characterize occupy in the causal space) and lack such a role (because causation, being an extensional relation, works as it does no matter what predicates, definite descriptions or properties are ascribed to events). In my opinion (and I think this is also Kim's opinion), predicates are causally relevant but they cannot be causally efficacious since they are very un-essential to the causal relata: so to speak, there is no reason why a causal relatum is designated by a certain predicate (although I am not sure this is an entirely correct way of speaking). Properties that are essential to events are causally efficacious since the event could not have been the same event unless it had them. But it is, I take, very intuitive that any linguistic device (therefore a predicate) is such that the event that it designates could have existed even if it had been designated by another linguistic device (or had not been designated at all).
I comment that it is indeed immaterial to causal relations that we describe them in physical temrs. (I hesitate about the "they can be described in physical terms", since presumably "they can be described in physical terms" means nothing more but "they have physical properties"). But – if I accepted that it is immaterial for events's place in the causal space that they have physical properties – then it follows that properties are dispensable for causal relations, they are arbitrary and may be "thrown away"; it follows that properties are as parasitic on causal relations as are predicates.
5. A more moderate view: Mental Fragility
If we abandoned the assumption of super-fragility, we might be enticed to preserve from Davidson an assumption of "mental fragility": if an event would have had a different mental property, it would not have been itself. Let's call this assumption (MF). Davidson holds it only as a consequence of super-fragility, but I will make use of MF since I think that superfragility is obviously false.
The main problem with MF is, however, that it makes unintelligible the classification of mental token events under types. If MF is true, then events need not be fragile in physical respects, but they are fragile in mental respects: i.e., they could have lacked the physical properties they have, but could not have lacked any mental property they have. This is too strong a requirement: it makes counterfactuals such as “If this event had a different mental property, then….” appear unintelligible and it makes the classification of token events under types unintelligible. (These two problems are related: in a counterfactual “If M had been the case, then…” the reference of M is a bit different from the reference of “M” in the actual world; the reference of the “M” in the counterfactual is a different token of the same type as the reference of M in the actual world. But – according to MF – it is not the case that M could have been slightly otherwise than it is; it is not the case that the token-M could have been another token of the same type).
6. The tension in Davidson's philosophy between extensionality and superfragility
I am not going to criticize MF for its incompatibility with the extensional view of causation. Davidson's philosophy witnesses a tension between two extremes: the thesis of super-fragility (all properties of an event are essential to it; any event could not have been that same event if any property of it would have been different) and the thesis of extensionality (it is events that do the causal work, not the properties – be they mental or physical). There is an apparent tension between extensionality and super-fragility because superfragility implies that properties (all properties) matter as most as they can, while extensionality implies that properties do not matter in the least. However, the tension is apparent:
What is wrong, in my opinion, with MF is that it does not satisfy its purpose: namely, that mentality be causally efficacious. (I underline the "efficacious": it is a matter of properties, not of predicates). Since – as I understand causal effectiveness – it requires that a properties be something more than a mere device to pick out an event. According to MF, for any mental property M, if an en event that is M had lacked this property, then it would have been a different event. This precludes counterfactuals with mental events in the sequent, such as "If p had been the case, then event e (which is not actually mental) would have been mental". It also precludes counterfactuals with mental events in the antecendent, such as "If event e, which is actually mental, had not been mental, then p would have been the case". But I think that such counterfactuals should be preserved and formalized; moreover, that they express causal relations. If causal relations did not support counterfactuals, it is difficult to see with what account of causation we would remain. I presume that for Davidson causation is a primate relation, not further analyzable. It remains yet obscure how the relation between counterfactuals and causation could be accommodated on such a view; what Davidson would say is, I presume, that counterfactuals are brought into the story by nomologicity and not by causation: Laws, isn’t it, are such that they support counterfactuals, are intensional have a linguistic status: laws indeed are sentences.
But in the light of the distinction between causal relevance and causal efficacy that Kim pressents and that I tried to explore so far, it appear that there is a distinction between a relation's being linguistic and a relation's being implicitly general, i.e. holding between two events insofar as they instantiate a certain property, insofar as they belong to certain classes. This suggests that we may find a place for causation such that it supports counterfactuals, yet it does not have a linguistic status. Whether the term "intensional" is appropriate for such a relation, I don't know.
True, these accusations against Davidson are intelligible only once we abandon the background of his view of events that are tokens-as-tokens. This commits us to abandon the idea that being essential to an event (to a token-as-type event) is an adequate measure of a property's causal efficaciousness. I explored this idea so far only in order to pressent
7. Criticism of LePore and Loewer
I will criticize in this section LePore and Loewer's distinction between what they call relevance-1and relevance-2. LePore and Loewer's contention is that the critics of anomaluous monism do not distinguish between these two kinds of relevance, claiming that Davidson's anomaluous monism precludes the first kind of causal relevance but not the second:
a) Properties F and G are relevant-1 to making it the case that c causes e
b) c's posesing property F is causally relevant-2 to e's posesing property G
I believe that there is, however, some textual support in Davidson's work for the idea of relevance-2 was suggested to the authors by Davidson's example of the silent versus loud shot that cause a death:
Suppose, Sosa argues, that someone is killed by a loud shot; then the loudness of the shot is irrelevant to its causing the death. 'Had the gun been equipped with a silencer, the shot would have killed the victim just the same'. … The crucial counterfactual is (sorry) ambiguous. Had the gun been equipped with a silencer, a quiet shot, if aimed as the fatal shot was, would no doubt have resulted in a death. But it would not have been the same shot as the fatal shot, nor could the death it caused have been the same death. The ambiguity lies in the definite description 'the shot': if 'the shot' refers to the shot that would have been fired silently, then it is true that that shot might well have killed the victim. But if 'the shot' is supposed to refer to the original loud shot, the argument misfires, for the same shot cannot be both loud and shot. Loudness, like a mental property, is supervenient on basic physical properties and so makes a diference to what an event that has it causes. Of course, both loud and silent (single) shots can cause a death; but not the same death. (Thinking Causes, p. 17).
8. De dicto and de re counterfactuals
We are all familiar with the de dicto/de re distinction in modal contexts. It may happen that a sentence containing a definite description be necessarily true (de dicto) but it should not express a necessary relation between the predicate and the object denoted by the definite description, or vice versa. Thus, the sentence
(there is a unique x such that x is the assassin of Smith and x is the most cruel person in our town)
but obviously entails a conditionals:
(there is a unique x such that x is the assassin of Smith x is the most cruel person in our town
"the assassin of Smith is necessarily the most cruel person in our town" (DE RE)
which says that the person who is actually the assassin of Smith is such that he is, in all possible worlds (that is, even in those in which he does not kill Smith) the most cruel person in our town. This is compatible with the posibility that John is the assassin of Smith in our world, he is the most cruel person in our town in our world, and Mary is the assassin of Smith in other world W and she is not the most cruel person in our town in that world.
Which is perhaps true de dicto (I am inclined to say that it may be true de dicto; that is the chance that we lived in another solar system given that the number of planets is even is greater than the chance that the number of planets became even in our solar system), but is vacuously true if read de re – since the number 9 could not have been even, so the antecedent is necessarily false. Likewise, the sentence
if we had lived in another solar system, then our closest neighbor would not have been the fourth planet from the star
may be read de re, in which case it is trivially true (since Mars – the planet which is our closest neighbor, could not be transplanted to another solar system, or at least the probability that it could be transplanted to another solar system is reasonably low) but it may be true or false if read de dicto – since it is possible that, for any solar system in which we lived, we would live on the third or fifth planet from the main star and our closest neighbour would be the fourth – and not the second or sixth – planet from the sun. (Perhaps this is a still undiscovered law of nature).
And the necessary implication
It is necessary that If yesterday's building happened, then the main building collapsed (de dicto).
The descriptions "yesterday's earthquake" and "the main building" may be read de dicto and de re. Thus we have four ways of interpreting the counterfactual as being de dicto or de re.
To speak only about the first description, "yesterday's earthquake" may be interpreted de re as referring to that precise event which is yesterday's earthquake and de dicto about any event that could have been yesterday's earthquake. To speak in a possible-worlds framework: the de re reading of the description "yesterday's earthquake" entails that when we evaluate the counterfactual, we speak about one and the same event that is hold fixed across possible worlds in order to evaluate the counterfactual; the de dicto reading does not assume this. However, it is wrong to say that a definite description interpreted de re has the same (in the token-sense of 'same') reference in all possible worlds. To hold so is counter-intuitive for at least two reasons: (a) an object (or event) obviously changes its properties as it endures through time. It is thus implausible that it should preserve all its properties across possible worlds. (b) If among the properties of an object (or event) we include relational properties, then an object (or event) obviously does change some of its properties from a possible world to another. Thus, even if I assert the counterfactual “If yesterday’s earthquake had occurred, then …” I do not refer to the same token-counterfactual that is referred to by the description “yesterday’s earthquake” in a non-modal sentence. It is indeed meaningless to say “If yesterday’s earthquake had occurred exactly as it actually did, then…”: when the description “The F” occurs in a counterfactual, it refers to an entity that differes by at least one property from the actual entity that is referred to by that description. slight
Thus, it is incorrect to say that the de re essential class is composed of only one member, that the de re reading of a definite description is such that it refers to the “same” object (or event) in all the worlds – at least if “same” means here token-same. It may be true that more properties of the members of the de re essential class are common to them than the properties of the de dicto essential class: indeed, the members of the de dicto essential class need only have in common the property expressed by the predicate that occurs in the description. (The de dicto essential class of the description “the earthquake” is formed only by earthaquakes”; the de re essential class of the same description is formed by events that may share other properties). The de re essential class may thus be “denser” than the de dicto essential class, insofar as its members may have in common more properties than the members of the de dicto essential class; but even this assumption cannot be taken for granted
8.1. Laws and counterfactuals
The distinction between de dicto and de re counterfactuals may shed a light on the claim that nomological sentences are differ from accidental generalizations in that the first, but not the latter, support counterfactuals. But any accidental generalization may be susceptible of being turned into a counterfactual-supporting sentence if its antecedent is considered de re. Consider the example of “All the cigarettes in my pocket are filter cigarettes”. This is not a nomological generalization; it cannot be counterfactualized since it is not true that “for any x, if x had been a cigarette in my pocket then x would have been a filter cigarette”. It may however be true that for any x, if x had been one of the objects that are actually the cigarettes in my pocket, then x would have been a filter cigarette. This is true if it is true that all filter cigarettes (including the cigarettes that happen to be in my pocket) are essentially filter cigarettes. This is why the counterfactuals that are supported by nomological generalizations are de dicto ones. If the counterfactuals we are dealing with are interpreted de re, then it is possible that even accidental generalizations could be made to support counterfactuals.
Of course, it is not true that any accidental generalization may support counterfactuals if it is taken de re. What I said in the above paragraph is that it may be turned to support counterfactuals. Consider the example of the sentence “All the students in this room wear blue shirts”. This is obviously an accidental generalization, since the counterfactual “for any x, if x had been a student in this room, x would have worn a blue shirt” is false: this is a de dicto counterfactual, since it refers to any object that could have satisfied been described by the predicate “a student in this room”. It is not even true that the de re counterfactual “for any x, if x had been one of those who are actually the students in this room, then x would have worn a blue shirt”, since wearing a blue shirt is not an essential property of any of those who are actually the students in this room. Nevertheless, we may have a super de re reading of the counterfactual: for any x, if x would have been one of the objects who are, in this moment and in this place, the students in this room, then x would have worn a blue shirt” and this super de re counterfactual is, I hold, true. It is true since it refers, as much as possible, to the token-as-token objects that satisfy the property “being a student in this room”. This shows that the de dicto/de re distinction is a gradual matter and that it coincides with the distinction between tokens and types. The more token-oriented the reading of a definite description, the more de re it is; the more type-oriented it is, the more de dicto it is.
The conclusion of this section is that the claim that “nomological generalizations, unlike accidental generalizations, support counterfactuals” need at least to be qualified. Any generalization supports a counterfactual if it is read in a super-de re fashion, and it may support a counterfactual if it is read in a de re fashion. What is meant by the slogan “nomological generalizations support counterfactuals” is that nomological generalizations support de dicto counterfactuals. The distinction between de re and de dicto counterfactuals is thus useful in elucidating the diference between accidental and nomological generalizations.
9. Essential Classes
Whenever a modal sentence is necessary (de dicto) but the corresponding de re reading is not true, the reason is that the reference of the definite description taken de re has differs, across possible worlds, from the reference of the same description taken de dicto. "The current prime-minister of Israel is necessarily male" is true since the Ariel Sharon, the person who is actually the current prime-minister of. Israel, is male in all possible worlds (in which he exists). The definite description "the current prime-minister of Israel", taken de re, refers only to objects that have the property of being males. "It is necessary that the prime-minister of Israel should be male" (de dicto) is false, since it is possible that the prime-minister should have been a woman. The definite description "the current prime-minister of Israel", taken de dicto, refers across possiblw worlds to other individuals than to those referred to by the description taken de re.
Let's define an essential class as the class of the objects (and possible objects) that satisfy a definite description in all the worlds. There are thus two essential classes of a definite description, a de re and a de dicto essential class. In the case of disagreement in truth-value between the de dicto sentence "It is necessary that the F is G" and the de re corresponding sentence "The F is necessarily G", where it is true that "The F is G", there is a disagreement between the de re essential class and the de dicto essential class of a definite description. These two classes coincide in the actual world, that's why the sentence "The F is G" is true.
In the case of a complete agreement between the de re and the de dicto essential class of a definite description, there can be no difference between the de dicto "It is necessary that the F is G" and the corresponding de re, "The F is necessarily G". The reason is that the reference of the description taken de re and de dicto are the same in all possible worlds. It is impossible to give examples of definite descriptions, since definite descriptions are considered to be non-rigid designators. But examples may be given with proper names: there may be no difference between a de re sentence such as "John is necessarily bald" and the de dicto "It is necessary that John is bald". The reason is that "John" refers to one and the same individual in all possible worlds (in which he exists). Natural kind terms – at least if we accept that they are rigid designators – behave the same: "Water is necessarily composed of hydrogen and oxygen" has the same truth value (in all possible worlds) as "It is necessary that water should be composed of hydrogen and oxygen". An informal way to express the agreement between the de dicto and the de re essential classes of a term a is: "a is necessarily a", or "it is necessary that a is a". (These last formulations are equivalent; they are neither de dicto nor de re, since they express the identity between the de dicto and the de re). Water is necessarily water; it is necessary that water should be water – that's why the sentence "It is necessary that water be F" and "Water is necessarily F" are equivalent. That's why a counterfactual using only rigid designators cannot be interpreted as de dicto or de re with a resulting diference in truth-value: indeed,
If the water in the Pacific Ocean had been green, then iron would have been a gas
is a counterfactual (I don't want to discuss its truth value) whose terms may be interpreted de dicto (any kind that satisfies the natural-kind term "water", any object that satisfies the name "Pacific Ocean" etc) or de re (the natural kind that is actually water, the object that is actually the Pacific Ocean etc). But the references of these terms coincide in all possible worlds, no matter whether they are considered de dicto or de re. The reason is that these terms are all rigid designators.
10. Quasi-necessity (Quasi-de dicto-de-re-agreement)
When there is a disagreement between the de dicto and the de re essential class of a definite descritption, it may still be the case that the de re reference of a definite description coincides with the de dicto reference in most, or in the closest possible worlds. Thus, the de dicto sentence "It is necessary that the number of solar planets be odd" is not true, since the number of solar planets is not necessarily the number of solar planets. But it is possible that the number of planets be odd in almost, or in the closest, possible worlds. In such a case, we should say that The F is almost necessarily the F.
Kim's distinction between weak and strong supervenience11
Kim’s reaction to Davidson’s anomalous monism is, as it is known, pressented in a number of articles. In Concepts of Supervenience he distinguishes between what he calls weak and strong supervenience respectively. Although Davidson does not speak in terms of this distinction, I presume that weak supervenience is the only kind of supervenience is the only kind of supervenience that he may allow: indeed, according to weak supervenience correlations between physical and mental events have an accidental status, while strong supervenience supports counterfactuals and thus has a nomological status. It is important to note that weak supervenience remains so even if it is defined in such a way as to hold for all possible worlds:
An equivalent formulation is:
A strongly supervenes on B just in case, if for any x and y if x and y share all properties in B then x and y share all properties in B then x and y necessarily share all properties in A –
“It is necessary that whenever someone has C-fibers stimulated, s/he feels pain”
If we want an account of psychophysical counterfactuals that should make explicit both the causal efficacy of events and of properties, then Kim's de dicto strong supervenience is not enough. We need an account that should make apparent that (i) when a psychophysical causal relation holds, an event is such that it (the same de re cause-event) would have caused the same (de re) effect in other possible worlds; (ii) events with similar properties would have caused effects with similar properties in other possible worlds. The last condition is captured by de dicto counterfactuals; but for the former we need de re counterfactuals.
The idea of quasi-necessity that I pressented in the tenth section of this essay is such that it expresses partial agreement between the de re and the de dicto essential class of a definite description. My suggestion is that quasi-necessity is what we need if we want to analyze the causal efficacy of events and of properties. For quasi-necessity is a stipulation on the margins of variations of the definite description: it says that the variations permitted by the de dicto reading of the description are not very different from the variations permitted by the de re reading of the same description.
Secondly, quasi-necessity is not an arbitrary (purely linguistic) stipulation: it means that "The F (de dicto) is almost necessarily the F (de re)". If quasi-necessity is correctly ascribed to a description, the reason may only be that there is a nomological constraint: the earthquake-as-it-was could not have been (very) different from the earthquake-as-it-was. I do not want to suggest that nomologicity means nothing but quasi-necessity. I rather want to suggest that quasi-necessity, where it holds, holds in virtue of a nomologic constraint; that quasi-necessity is what expresses both nomologicity and causation. It is wrong to say that "nomological generalizations differ from accidental generalizations in that the former support counterfactuals, while the latter do not". The reason why this is wrong is that any generalization – even accidental generalizations – may be interpreted so as to support counterfactuals if they are read de re. On the other hand, if we have de dicto counterfactuals, then we will only be able to express relations between properties, not between events. Therefore, if nomologicity is understood as the property of generalizations to support de dicto counterfactuals, then it is unable to express causation between events. Moreover, quasi-necessity is a remedy for a shortcoming in Davidson’s anomalous monism: according to Davidson, causal relations do not support counterfactuals and events are tokens-as-tokens that may not be classified under types.
"Suppose I create a table in which all the entries are definite descriptions of one sort or another of events. I refer to the events by giving the column and the row where the description is to be found: column 179 row 1044 for example is the event of my writing the sentence. Let us call the events listed in the table "table-events". The vocabulary needed to describe (needed to provide a definite description of) each event is just the vocabulary needed to pick out the column and the row. These events have their causes and effects: for example, event 179-1044 caused a certain rearrangement of electric flows in the random access memory of my computer. There are, I imagine, no interesting tablo-physical laws whatever, that is, laws linking events described in the table language and events described in the vocabulary of physics. Yet this fact does not show that table events are not causally efficacious". (p. 12).
It is curious for Davidson to maintain such a view of the mental, since he also holds (in the same article) that
That causation is extensional entails that events do not care about how (or whether) we describe them. Predicates, descriptions are just a way of picking out events. But what I believe (and what I think Kim believes) is that from this extensional, non-linguistic view of causation (which I take to be unproblematic) it does not follow that removing the properties of events from the events themselves would have no consequence on their effects. There is a big step from "events don’t care about how we describe them" to "events don’t care about what properties they have". The first is innocuous, the second is not. Unfortunately, Davidson is, most of the time, prone to confusing the two, thus reaching the conclusion that properties of events may be omitted from any account of causal relations: this is the only way I can understand his comment of the example of table-events:
"It will be retorted that it is simply irrelevant to the causal efficacy of table-events that they are table-events. This is true. But it is also irrelevant to the causal efficacy of physical events that they can be described in the physical vocabulary. It is events that have the power to change things, not our various ways of describing them". (p. 12).
However, this last consequence cannot be accepted wholeheartedly by Davidson, since its consequence is drastic: abolition of supervenience, freedom to ascribe any predicate-property we want to any other event or do away with predicates-properties and name events in a propertiless fashion (for example, by giving them numbers). It is hard to see how any causal relevance – let alone efficacy – of the mental could be accommodated on this view. The way of avoiding this dire consequence is superfragility: any event with property F is essentially F; if it had not been F, it would have not been itself but rather a different event. Call (A) the assumption of superfragility. (I borrow the term "fragility" from David Lewis: a fragile event is such that if it had ocured under a different guise, it would have been a different event; fragility is thus a measure of the malleability of an event, of its ability to occur differently than it did/does in fact occur).
I am convinced that superfragility is not correct. To see this, we can only pause to notice that, according to superfragility any property of an event is essential to it. The thesis of super-fragility is, I take, false, since it is intuitively true that not all arbitrary descriptions that may characterize an event, according to any scheme we may like to invent to classify them, are essential. The lack of causal efficacy of arbitrary classificatory devices is an assumption that I want to preserve and that, I think, so does Kim. As I pointed in the second section of this essay, a predicate – in fact, any linguistic device – expresses a "minimal" property: the (self-referential) property that the non-linguistic entity (the event) to which it applies just happens to be called by means of that predicate (of that linguistic device). These minimal properties are not causally efficacious because they are arbitrary: there is no reason why an event is designated by a predicate rather than by another. Arbitrary properties may be causally relevant: but they cannot be causally efficacious. Making all predicates causally efficacious amounts to turning any arbitrary devices that happen to characterize events in a certain way (for example, obeying supervenience) into causal agents. But this is, I think, too little as a fulfilement of the requirement that the mental be causally efficacious.
There are several problems with MF. First of all, it does not entail, and should not be confused with, the thesis that mental predicates are causally efficacious. Mental predicates are linguistic devices: they happen to be causally relevant (since they designate mental properties and mental properties are – we hope – causally efficacious essential to the events that are causal relata) but they are relevant only provided that mental predicates are correctly used, that is, provided they designate mental properties and that mental properties are causally efficacious. MF is an ontological, not a linguistic thesis. Mental predicates as such are dispensable since they are mere linguistic devices – and it is, I take, unobjectionable that the world as it is correctly described in language would be the same world if the language were absent or if language were wholly different. (Davidson for sure would think this is unobjectionable since it is a consequence of the extensional, i.e. non-linguistic status of causation: as we know, naming the invasion of Panama "operation Just Cause" would not alter the consequences of that event). Secondly, mental predicates are minimally causally efficacious insofar as they are properties and they are minimally essential properties, almost contingent properties. (We saw in the above section that the mere supervenience does not guarantee causal efficacy: in order for something to be causally efficacious, it needs to be essential, or almost essential, to a causal relatum).
Secondly, MF stands in need of justification. In Davidson's philosophy, its only justification is superfragility. But superfragility needs, in my opinion, to be rejected as counterintuitive: it is only suited for a world of tokens-as-tokens. Among the consequences of MF are that causal relations cannot be counterfactualized (indeed, any sentence of the form "if M had had a different mental property, then…" is false, since M could not have had other mental property than those it has) and that causal relations do not work via the causal relata's belonging to certain classes. Davidson is ready to admit these consequences, since they are the consequene of the thesis of superfragility.
What I want to hold is that causal efficacy of mental (properties, not predicates) should be independent from MF. As I understand Kim in "Can Supervenience….?", he rejects even MF implying that for an event with mental property M it is not essential to be M; but if it lacked M, then it would not occupy the same place in the causal space. In my opinion, what is needed is an account of mental properties that are almost essential to events; i.e., that are such that the event they characterize hardly could have lacked them; if it had lacked them, it would have belonged to very distant possible worlds form the actual world. The reason is that if mental properties are not essential (almost contingent) to the token-events, then they are arbitrary and can be replaced; if they are essential, then counterfactuals with mental events are impossible.
The thesis of superfragility implies that the all the properties of an event are efficacious in determining the place the event occupies in the causal space; the thesis of extensionality implies that no property is efficacious. (This tension between two extremes that ultimately converge reminds of the same relation between realism and solipsism in the Tractatus). Both super-fragility and extensionality are opposed to the essential status of causation; both super-fragility and extensionality cannot tolerate counterfactualization of causal relations. (The reasons are, admittedly, opposite: on a super-fragile account of events, no event could have been a different event if it had been slightly otherwise than it is; on an extensional account of events, any event could have been as well any other event – events are somehow like Tractarian objects, two events are different since they difer from one another). Thus, the accusation formulated against Davidson, that on his account "the mental does no causal work since mental properties (any property, for that matter) are not effective" may be formulated in different words: on Davidson's account of events and causation, all properties of an event do equal causal job, and it is only they that do this causal job, therefore the mental is not more effective than other classificatory schemata: if mental properties are effective, they are as effective as any other properties. If what impresses us in Davidson is extensionality, then we should say that properties are causally inefficacious. If what impresses us is superfragility, we should say that no property is particularly efficacious, since all properties of an event (and nothing over and beyond them) are equally efficacious.
If we abandoned superfragility but retained mental fragility, the problems would not be solved. We would no longer be in a world of tokens-as-tokens, yet in a world of mental tokens-as-tokens.
(p. 634)
LePore and Loewer claim that, while Davidson's account of causation is incompatible with relevance-1 of properties but allows for relevance-2, thus allowing for mental properties to play a causal role: true, mental properties cannot be responsible for turning two events into the relata of a causal relation, but when the event that plays the role of the cause in a causal relation has a certain mental property, it may turn be responsible for the effect's having a mental property. Thus, the mental (mental properties) is indeed causally inert in that it cannot cause new events to appear, but it is relevant in the sense that it can cause new properties to characterize events.
What I want to hold is that the thesis that Davidson's conception of events is compatible with relevance-2 is simply wrong. First of all, relevance-2 is a counterfactual account of causation: but, according to Davidson, it is not causation but nomologicity that supports counterfactuals. By assuming counterfactuals, relevance-2 assumes a view of events that have, so to speak, a nucleus and a detacheable parts: there is a "nucleus" of properties that constitute the essence of the event and a supplement of contingent properties that may vary freely, while the event remains (type) the same. But this view of causation is incompatible with Davidson's extensionality and superfragility: since events are fragile, all their properties are (equally) essential. LePore and Loewer’s relevance-2 could be called a view of property-causation as supervenient on event-causation: that is, a property of an event (the cause) cannot cause a property of another event (the effect) unless the first event causes the second. I presume that their model is committed to the truth of the following rule:
When c causes e and Fe causes Gc, c would still have caused e but even if ~Fc, but then ~Ge. In this situation, there is (would have been) a property H, incompatible with F, such that Hc and Hc would have caused Je.
I believe that this passage witnesses a fatal flaw in Davidson’s conception of events and causation. The ambiguity that Davidson diagnoses is not between a type event (“the shot”, “the death” no matter how they occurred or would have occurred) but between two token events (this-death-here-now versus that-death-there-then and likewise in the case of the shot).
What is more important, it is very intuitive – and I think unquestionably true – that even if the death-that-was-caused-by-a-loud-shot and the-death-that-would-have-been-caused-by-a-silent-shot are (would have been) two different deaths – and all we need for this is a reasonable low standard of fragility, so that the words “same” and “different” be used in a token sense – then still the differences in these deaths could not have been ascribed to the differences in the loudness of the shot. When I think of “two deaths being different”, I can imagine some medical details, like what organs were damaged and how, the duration of agony and so on. It is hard to see how any of these details has anything to do with the loudness of the gunshot that caused the death. I dare to say that Davidson’s sentence, “both loud and silent (single) shots can cause a death, but not the same death” is ambiguous: the ambiguity resides in the verb “to cause”: different shots cause different deaths, but it would be a mistake to conclude from here that different shots cause different deaths to be different. The difference between the two deaths caused by loud and silent shots cannot be caused by the difference in the loudness of the shots: two different shots cause two different deaths only because it is trivially true that two different token deaths are different. If all we can say about the causal role of the mental is this kind of causal relevance, we can again be disappointed.
There is one more reason to be dissatisfied with the account of mental causation that Loewer and LePore ascribe to Davidson. On Davidson’s view of events, all events are tokens-as-tokens. There is no criterion of classifying events as more or less similar. Therefore, to say that “a different shot than this one would have caused a different death” is simply illegitimate on this account. It is, first of all, illegitimate because it can be the case that two different events have the same cause, or can have the same cause (overdetermination or preemption). Leaving aside this objection, we can ask: why should we believe that a different shot would have caused a different death rather than a different event? What authorizes us to classify several token-deaths as deaths? The only answer Davidson could provide to this question is: an arbitrary linguistic decision. But then any classification of token-events under types, any element of generality would not authorize any inference to causal relations. If we forget about preemption and overdetermination, then we should say – with Davidson – that when event c causes event e, then a slight change in event c would result in a new event, d; and this event d¸ had it occurred in exactly the same (token) circumstances in which event c occurred, would have caused an event f, different from e. But from here we cannot infer anything about the degree of resemblance or diference between f and e: it may be the case that f is a very different event from e or very similar to it. After all, similarity and difference depend on laws and laws are linguistic entities: laws are intensional and they hold in virtue of descriptions. Therefore, instead of saying that “a different (silent) shot would have caused a different death”, Davidson should have said either that “a different (silent) shot would have caused a different event” or “if there is a connexion between ‘a loud shot’ and ‘death of kind X’, then there is a connexion between ‘silent shot’ and ‘death of kind Y’”. But this connexion is a nomological connexion: it deals with descriptions, not with events, and it says nothing about causation – since we cannot know to which event the respective descriptions apply.
In Davidson’s world of tokens-as-tokens, little place may be allotted to generality. If we still want a (non-arbitrary) criterion of classifying events, then we should be open to counterfactuals. And, since an event is just its place in the causal space (its causes and consequnecess) it would follow that causation supports counterfactuals. But this contradicts Davidson’s tenet that it is only nomologicity, and not causation, that supports counterfactuals.
If there is a moral that can be detached from the above lines, it is that we need an account of causation that supports counterfactuals. Kim’s distinction between causal relevance and causal efficacy showed how this can be pursued without committing us to a linguistic view of causation, i.e. without committing us to saying that an event c causes an event e only under certain descriptions, but not under other descriptions.
The current prime-minister of Israel is necessarily an Israeli citizen (DE RE)
is false, since the person who is actually the current prime-minister of Israel (Ariel Sharon) might not have been an Israeli citizen. On the other hand, the corresponding de dicto sentence,
It is necessary that the prime-minister of Israel be an Israeli citizen (DE DICTO)
Is true, since it is a matter of (juridical) law that any prime minister of Israel must be, at the time of his being in office, an Israeli citizen. Likewise, the sentence The current prime-minister of Israel is necessarily male is true, at least insofar as we believe that Ariel Sharon could not have been not-male, while It is necessary that the current prime-minister of Israel be male (DE RE) is false, since the law does not prohibit women from being prime-ministers.
The distinction between de dicto and de re reading of definite descriptions does not affect only modal sentences such as the above examples, but also necessary conditionals that contain definite descriptions. It is easy to realize this once we consider examples such as
"it is necessary that the assassin of Smith is the most cruel person in our town" (DE DICTO
This might be formalized as a conjunction:
which can be read de dicto and mean that in all possible worlds in which someone assassinates Smith, that person is the most cruel person in all the worlds in which s/he kills Smith. The de dicto necessary conditional asserts an identity between whoever satisfies, in any world, the first description and whoever satisfies, in that world, the second description. This is compatible with the situation in which John assassinates Smith in this world and he is the most cruel person in our town and Mary assassinates Smith in another world W and she is the most cruel person in our town in that world W, although Mary is not the most cruel person in our town in this world and John is not the most cruel person in our town in W. On the other hand, we may have a de re reading, perhaps better expressed as
The next step of my essay is to generalize the de dicto/de re distincion to counterfactuals. The easiest way to see how this can be done is to consider counterfactuals that have definite descriptions in their antecendent or sequent. An example is:
If the number of planets were even, then we would live in another solar system
The de re/de dicto distinction for counterfactuals can be applied wherever there are definite descriptions. But it can be generalized to necessary implications and counterfactuals in general. The reason is that whenever we have a modal qualification, the de dicto/de re distinction may appear. In order to realize this, we need to regard descriptions of events as definite descriptions. Thus, let events A and B be respectively "yesterday's earthquake" and "the main building's collapse". We have the counterfactual
If yesterday's earthquake had happened, then the main building would have collapsed
Of course, Davdsonian events do manage to preserve all their properties through time (and possible worlds) for the simple reason that they are dated occurrences in space-time; and they are, consequently, fixed within one possible world. But if we want a counterfactual account of event-causation, then we need events such that they may survive, while remaining the same, across possible worlds.
Thus, it is possible that the counterfactual "If yesterday's earthquake had occurred, then the main building would have collapsed" may be true even if in the closest possible world in which the description yesterday's earthquake, taken de re, is satisfied, no earthquake occurred yesterday, i.e. no event satisfies earthquake yesterday’s he description Yesterday's earthquake as read de dicto. What this means is that the property of “being yesterday’s earthquake” is not an essential property of yesterday’s earthquake, or is a less essential property than other properties.
Similar examples could be built with events. In the already given example of "If yesterday's earthquake had occurred, then the main building would have collapsed", it is perhaps true that yesterday's earthquake as it actually happened is such that its properties are pressent in the most, or in the closest possible worlds. Perhaps laws of nature are such that the event “yesterday's earthquake” could not have occurred otherwise than it actually did. In such circumstances, it would be safe to affirm that yesterday's earthquake is a denoted by a description whose de re and de dicto essential class are in quasi-complete agreement. This may, again, hold in virtue of a law of nature. If this is true, then it may be said that events whose descriptions are suitable for inclusion in laws of nature are such that the de re and the de dicto essential classes are almost identical.
Quasi-necessity thus involves nomologicity since it holds in virtue of scientific laws if it holds at all. Quasi-necessity is a measure of an event’s being necessarily that event; if quasi-necessity holds, then any event F (de re) is the same event (de dicto) in almost all possible worlds, i.e. in almost all possible situations in which it can occur. If the equation of quasi-necessity with nomologicity is correct, as I hold it is, then scientific laws may be defined as involving descriptions that are quasi-necessary. This characterization is better than the characterization of scientific laws as generalizations that support counterfactuals.
Quasi-necessity is fit for counterfactuals. Where quasi-necessity holds for a description “the F”, then the counterfactuals “If the F had occurred, then…” are identical in truth value no matter whether they are read de dicto or de re. The reason is that quasi-necessity means almost complete agreement between the de dicto and the de re essential class of a description. Complete agreement between the two essential classes is fit for necessary conditionals: thus, the conditionals “If the F occurs, then necessarily …” and “It is necessary that, if F occurs then …” are alike in truth-value if the de dicto and the de re essential classes of “the F” are identical. Quasi-necessity is fit for nomologicity since nomologicity is a matter of counterfactuals and quasi-necessity is fit for counterfactuals.
A weakly supervenes on B if and only if, necessarily, for any x and y if x and y share all properties in B then x and y share all properties in B then x and y share all properties in A – that is, indiscernibility with respect to B entails indiscernibility with respect to A.
As I understand Kim, he introduces the moral operator (the de dicto necessity) in the definition of weak supervenience in order to emphasize that the weak supervenience does not support counterfactuals even if it is defined for all possible worlds. The relation of dependence expressed by weak supervenience is such that it holds only within each possible worlds and it cannot be interpreted so as to support trans-world relations between objects or properties. To consider a famous example, let’s assume that the property of being in pain supervenes on the property of having C-fibers stimulated. (I assume, for the sake of the argument, that there are several kinds of C-fibers stimulation, or even –to make things simpler - several C-fibers that may be stimulated, and only one kind of pain. This is an ultrasimplistic assumption, but it may serve to illustrate the fact that two events may be alike in respect of having the same supervenient property but different in respect of their physical properties. According to weak supervenience, we cannot infer that any event belonging to other possible world that has the property of being a C-fiber stimulation has the property of being an instance of pain. All we can conclude – if weak supervenience is defined for all possible worlds – is that if two events share all their subvenient (physical) properties, then they share all the supervenient (mental) properties. But it is possible that the supervenient property they share is other mental property than being painful – for example, it may be the property of being pleasant. For weak supervenience, even if it is defined for all possible worlds, cannot preclude the possibility that in another possible world than the actual one it is the property of being pleasant that supervenes on the property of being a C-fiber stimulation.
In order to remedy these shortcomings, Kim introduces his version of strong supervenience. Strong supervenience is specifically designed so as to support counterfactual relations between events that have physical, respectively mental properties:
A strongly supervenes on B just in case, necessarily, for each x and each property F in A, if X has F then there is a property G in B such that x has G, and necessarily if any y has G, it has F.
Now, the point that I want to develop is that Kim’s strong supervenience is in a sense too strong and in a sense too weak for the purposes which it is meant to serve.
First, it is too strong because it involves necessary conditionals, while nomologicity only requires counterfactuals. To adopt a possible-worlds standpoint, a counterfactual says less than a necessary conditional: a necessary conditional is true iff the embedded conditional is true in all possible worlds, while a counterfactual is true iff the embedded conditional is true in the closest world in which the antecedent is true.
Second, Kim’s necessary conditionals are too weak to the extent at which they may be interpreted de dicto. If they are interpreted de dicto, they do not guarantee that the same events are preserved across possible worlds. In order to see this, let’s consider the necessary conditional:
if it is read de dicto (as I believe Kim would have us read it), then it asserts something about any event, in any world, that is a C-fibers stimulation, that belongs to the de dicto essential class of the definite description “C-fibers stimulation”, and says that it is followed by the said person’s feeling pain in that same world.
What this necessary conditional tells us is that relations of entailment hold between any events which have a certain physical property and events that have a certain mental property. But it does not tell us anything about which these events are. More specifically, it does not tell us what the C-fiber stimulation is in other worlds, nor about what pain is in another worlds. The only thing that de dicto counterfactuals about the events described is that they have the property that is specified by the (event) description: these events belong to the de dicto essential class of the definite description.
It is obvious, however, that an event such as the C-fibers stimulation has many other properties, besides that of being a C-fiber stimulation. For instance, an instance of C-fibers stimulation may be yesterday’s sudden C-fibers stimulation in my right arm. The de re essential class of this definite description is such that it requires us to refer to the same-type event in all worlds. It is not even necessary that all the members of the de re essential class of a definite description should share the property that is expressed by the predicate which occurs in the description: it is not necessary that the the members of the de re essential class of the C-fiber stimulation should be C-fiber stimulations. On the other hand, it is not necessary that the members of the de dicto essential class of the description "the F" be instances of "what the F could have been", if "the F" is understood de re.
In the light of these remarks, it becomes apparent that, if all we have are de dicto counterfactuals or even necessary conditionals, then we cannot secure stable correlations between events. If we have to choose between de dicto and de re counterfactuals, then we need to choose between the causal efficacy of properties and the causal efficacy of events. Thus, de dicto supervenience indeed secures the causal efficacy of properties, but it does not say anything about which events are causally efficacious. We can also imagine a form of de re (strong) supervenience that would tell us that, if an event A causes another event B in this world, then it causes it in other worlds as well. De re supervenience without de dicto supervenience would tell us which events are causally efficacious, but would not tell us what are the properties that are causally efficacious.
References:
Davidson: Thinking Causes Donald
Mental Events
Thinking Causes
Jaegwon Kim – Can Supervenience Save Anomalous Monism?
- Concepts of Supervenience
Loewer and LePore – Mind Matters (Journal of Philosophy 1987)