Mental Causation: The Mind-Body Problem

The issue is about properties: events cause other events in virtue of their microphysical properties but, apparently, never in virtue of their mental properties. Using our terminology, physical properties of events are causally relevant to physical properties of their effects, but mental properties, apparently, aren’t ever causally relevant to anything. This is because properties are related by causal relevance when laws of nature link them, and, apparently, mental properties are never linked that way.

The instances of some properties—action properties, for instance—must have mental causes. This won’t help us, however, since the connection between the properties isn’t causal if it is required by logic.

So far I’ve said that mental properties are not identical with physical properties, but I haven’t said anything else about how the two kinds of properties are related. In this chapter I make two positive proposals about their relation: (1) mental properties supervene on physical properties, and (2) mental properties are higher-level properties relative to physical properties.

Jaegwon Kim has made the point forcefully in recent years (1993a, 1998, 2005) that supervenience is the most fundamental obstacle to the causal relevance of the mental. He is right that there is something intuitively (and intellectually) problematic about supervenient causation. This concern survives even when we argue (as I shall below) that Kim’s detailed arguments against supervenient causation don’t quite work. Despite this intuitively problematic feel, the intellectual cost of giving up supervenient causation is extremely high. And in fact the supervenience of mental properties on physical properties provides the key to explaining how (in chapter 9) mental properties can meet the gold standard for causal relevance that I articulated in chapter 6 above.

Kim is right that supervenience is not a solution to the mind/body problem, and it is not a solution to the mental causation problem. Two things are needed in addition to supervenience. First, some more substantive relation has to hold between mental properties and physical properties than bare supervenience. Supervenience is a very thin relation. Since it is so easy to supervene (for instance, the property of being either a spoon or a snowflake supervenes on the set of properties {being a spoon, being a snowflake}), lots and lots of clearly causally irrelevant properties supervene on microphysical properties. I’ll argue in this chapter that the more substantive relation we need is that mental properties are higher-level properties relative to physical properties.

Second, just because a property supervenes on microphysical properties in this more substantive way doesn’t mean it is causally relevant to anything. In chapter 9, I will work out a way to connect higher-level properties to laws of nature: I will work out a leveled conception of property causal relevance. This will permit me to say that mental properties can be causally relevant to other properties—and hence to solve the mental causation problem.

Donald Davidson was the first to suggest that the right way to understand the dependence of the mind on the body is supervenience: “Although the position I describe denies there are psychophysical laws, it is consistent with the view that mental characteristics are in some sense dependent, or supervenient, on physical characteristics. Such supervenience might be taken to mean that there cannot be two events alike in all physical respects but different in some mental respect, or that an object cannot alter in some mental respect without altering in some physical respect” (1970b:214). The basic idea is that mental facts depend on physical facts, and the physical facts completely determine (or fix, or settle) the mental facts. Given the whole truth about the physical realm, the whole truth about the mental realm is settled: no mental variation is possible. On the other hand, fixing the mental facts permits a great deal of physical variation. A person could have exactly the same mental life even if she were physically slightly different. (For instance, if the core idea of functionalism is right [chap. 4, sec. 4.3], then there can be small physical differences that would make no difference to how mental states causally interact with one another, and perception, and action, and hence would make no difference to a person’s mental life.) Much of the world is not mental at all. Lots of nonmental parts of the world could change without any mental changes at all (for instance, tiny bits of rock millions of light-years from any sentient creature).

Davidson argued that supervenience is consistent with irreducibility: we can say that psychology is not reducible to physics, but psychological things are nevertheless still physical things through and through. Davidson’s aim in pointing out the compatibility of supervenience and irreducibility was to indicate what we might call “minimal materialism.” The concept is designed to answer this question: what is the weakest possible connection between psychology and the physical world, consistent with materialism, that we could describe that would still let us say the many and various things that we think must be true about our minds and behavior? (Kim [1998:15] introduces the expression “minimal physicalism” used in this way.) As I noted in chapter 1, physicalism is the view that everything is physical, including all the properties; materialism is the view that all the things are physical, but not necessarily the properties. The mental causation problem depends on the idea that mental properties are not identical to physical properties. Since physicalism says that all properties are physical, it also says that if there are any mental properties, they are also physical properties—hence identical to physical properties. The mental causation problem thus makes sense given materialism, but it does not make sense if physicalism is true. It is important to see that although materialism concedes that something isn’t physical, the concession is very small. Materialism says that all the actual minds are physical things. The fact that a nonphysical thinker is logically possible doesn’t mean that the actual world has any nonphysical bits to it—if it did, then materialism would be false.

There are a lot of different ways to fill in the basic idea of supervenience (McLaughlin and Bennett 2005). There are supervenience relations among sentences, truths, predicates, facts, properties, particulars, and many other things as well. Since I am concerned with the causal relevance of mental properties, I will focus on supervenience relations among properties. People have mental properties, and people have physical properties. Particular mental events, like the event of choosing a raisin muffin rather than a scone, have mental properties, and those particular mental events have physical properties. So the very same things have both mental properties and physical properties. Hence the kind of supervenience I am looking at is local rather than global (Haugeland 1982; Kim 1984). Supervenience is supposed to express the dependence of the mental on the physical. Dependence has to do with not just how actual things are related but how they would be related if things had been different. So, for instance, if I had chosen the scone rather than the raisin muffin, then I would have had different physical properties. Hence supervenience is strong rather than weak (Kim 1984).

Kim’s statement of the kind of supervenience we need runs like this: “Mental properties supervene on physical properties, in that necessarily, for any mental property M, if anything has M at time t, there exists a physical base (or subvenient) property P such that it has P at t, and necessarily anything that has P at a time has M at that time” (1998:9). The first “necessarily” in this definition demands that every instance of a supervening mental property depends on an instance of a physical property. The second “necessarily” articulates the idea that having the physical property determines having the mental property (Yoshimi 2007).

Contemporary discussions (in the philosophy of mind) of what “necessarily” means center on three kinds of necessity. A proposition is physically (or nomologically) necessary if, given the laws of nature, it has to be true (or: its falsity is inconsistent with the laws of nature). A proposition is logically necessary if, given the relations of the concepts involved in the proposition, it has to be true (or: it is strictly conceptually impossible for it to be false). A proposition is metaphysically necessary if it is impossible for it to be false, period. The standard example of a proposition that is metaphysically necessary but not logically necessary is that water = H₂O. Assuming that it is true, it is not true in virtue of the concepts involved. Yet it is nevertheless impossible for there to be water that is not H₂O (Kripke 1980). The proposition that an aunt has a sibling is logically necessary. The proposition that on Earth I cannot fly unaided is nomologically necessary.

It is logically possible for there to be a nonphysical thinker. (A fiction that begins, “It is a truth universally acknowledged, that a single man in possession of a good fortune, though completely devoid of physical properties, must be in want of a wife,” does not immediately contradict itself.) It follows, then, that mental properties are not identical to physical properties. So the necessity of supervenience is not logical or metaphysical. It can’t be logical, since that would mean that it is impossible for there to be a nonphysical thinker. Supervenience can’t be metaphysically necessary, since if it were, then any thinker would have to be a physical thing—so there couldn’t be a nonphysical thinker.

The remaining alternative is physical necessity. So we read Kim’s definition of supervenience as saying that it is physically impossible for something to have physical property P and not have mental property M; also, it is physically impossible for something to have mental property M and not have physical property P.

Kim’s definition leaves open whether M supervenes on more than one physical property. A token-token identity theory, like functionalism or anomalous monism, will say that in general the instances of a given mental property may have physically different realizations. But given supervenience, if anything has any of the physical realizer properties, then it must have the mental property.

Supervenience is supposed to express a minimal materialism. We should ask: is supervenience too much? Can a reasonable philosophy of mind deny supervenience?

The idea of materialism is that everything in the world is physical. The motor of the world is physical also. This means that everything that happens has a physical explanation. In other words, physics is closed: in principle, physics has the resources to explain whatever happens. Here is Kim’s statement of the physical closure principle: “if a physical event has a cause that occurs at t, it has a physical cause that occurs at t” (2005:43). Denying supervenience means either denying mental causation altogether or denying physical closure. Here’s the proof:

Suppose that the mental does not supervene on the physical. Then the physical facts do not determine (fix, settle) the mental facts. There can be two different persons who are physically completely alike yet differ in some mental respect. Suppose Billy and William are physically completely alike, yet Billy chooses to toss a rock through a window, while William chooses not to.

If what happens is determined exclusively by how things are physically, then both bodies do the same (either throw the rock or not). But then one of the bodies is doing something that doesn’t come from what it intends. So mental causation falls by the wayside, since what the bodies do does not depend on their mental properties.

If what happens is (sometimes, at least in part) determined by how things are mentally, then the bodies do different things. But the physical characteristics of their causes are exactly the same. It follows that physical closure would be false.

If we had a strong independent reason to deny that the mental supervenes on the physical, we would then need to decide whether physical closure is false or whether epiphenomenalism is true (or both). Either way the intellectual cost would be very high. The argument for physical closure is essentially a very high level empirical argument deriving from physical theory (Papineau 2001). And I’ve argued repeatedly that epiphenomenalism is in radical opposition to our conception of ourselves (even if it is logically coherent).

The cost of denying mind-body supervenience is high. As I am about to show, the cost of asserting supervenience is low. Supervenience looks like the way to go.

There are varieties of supervenience (although not the one quoted above) that are compatible with mind/body dualism. The core of supervenience is dependence plus determination: no mental difference without a physical difference, plus how things are physically determines how things are mentally. The core of supervenience is consistent with nonmaterialist conceptions of the world.

Huxley’s epiphenomenalism (chap. 3, sec. 3.3), for instance, is a dualist position that involves supervenience. Every change involving a mental property is caused by a physical change in the body. Hence corresponding to each instance of a mental property there must be an instance of a physical property, and the physical change is sufficient for the mental change. According to Huxley’s epiphenomenalism, then, mental changes supervene on physical changes.

There are varieties of supervenience that are inconsistent with physical closure (see Horgan 1993:559–560; and McLaughlin 1992 for details). Since the bare idea of supervenience is compatible with such wildly divergent conceptions of the mind-body relation, saying that the mind supervenes on the body isn’t saying much.

Horgan (1993) argues that we can’t just say, “the mental supervenes on the physical,” and be done with it: we have to explain why and how the mental supervenes on the physical. The explanation for supervenience is thus more basic than the fact of supervenience itself. As Kim puts it, “supervenience is not a mind-body theory” (1998:9): “mind-body supervenience states the mind-body problem—it is not a solution to it” (14). Kim is right: supervenience does no positive work in explaining the mind-body relation (see Heil 1998 for more on this theme). Nor does it do any positive work in solving the mental causation problem.

But for my purposes it is not necessary to explain how and why supervenience holds, since I am not trying to develop any particular theory of mind in this book. Rather my project is to vindicate mental causation in the context of materialism and physical closure. These two, plus the claim that mental properties are not physical properties, entail that mental properties supervene on physical properties.

The package of these three claims—physical closure, mental properties are not identical to physical properties, and supervenience—provides a particularly crisp way of stating the mental causation problem: Kim’s supervenience argument.

Jaegwon Kim argues that if mental properties are not identical to physical properties then there is no mental causation (1998:38–47; 2003:153–159; 2005:39–45). Since he believes that there is mental causation, he presents his argument as a good reason to believe that (some) mental properties are reductively identical to physical properties. My job is to show that the argument fails, that we can believe in mental causation and believe that mental properties are not identical with physical properties.

The argument runs as follows: Assume mental properties supervene on physical properties. Assume that a particular mental event c causes another particular event e and that these two events instantiate mental properties M and M* respectively. (Lowercase letters denote particular events; uppercase letters denote properties.) Assume that c’s having M causes e’s having M* or, as Kim writes, that M causes M*. By supervenience, c and e have physical properties P and P*. I present all these assumptions in figure 8.1 (for clarity, I leave out the names for the particular events c and e and follow Kim’s convention of referring to them by the properties they have: M and M*, etc.).

Now, also by supervenience, e’s having P* is sufficient for, and hence fully accounts for, e’s having M*. There are, then, apparently two answers to the question, “How does M* get instantiated on this occasion?” One is: because it was caused by M. The other is: because P* is instantiated on this occasion. Kim argues that the latter answer excludes the former. The (seeming) resolution of the seeming tension is to say that M causes M* by causing P* (figure 8.2).

But does M cause P*? There are again two candidates for what causes P*: either M or P. As I indicated in chapter 5, section 5.5, mental properties are not identical to physical properties, so M ≠ P. Now assuming Kim’s exclusion principle—“no single event can have more than one sufficient cause occurring at any given time—unless it is a genuine case of causal overdetermination” (2005:42)—either M or P, but not both, causes P*. But P clearly does cause P*. So M does not. The argument is fully general, so mental properties are causally impotent. The M–M* relation (according to Kim) is something like a pseudoprocess, like a causal sequence portrayed in a film (Salmon 1984), determined by the genuine causal process relating P and P* but not itself causal. Figure 8.3 presents the final picture.

Kim’s argument depends on connecting causation and supervenience in a certain way. In a moment, I’ll look at the two principles involved in making that connection, the exclusion principle and Edwards’s dictum. Both are very plausible principles. They have a firm basis in our understanding of causation and the physical world. I’m not going to argue that Kim’s way of treating the relation between causation and supervenience is contradictory or indefensible. There is a consistent overall picture of how the mind and the physical world are related, according to which supervenient properties are never causally relevant to anything. Given that picture, there are only two alternatives: either mental properties are identical to physical properties, so we really can make things happen, or mental properties are not identical to physical properties, so we never make things happen.

There is, however, another consistent overall picture of how the mind, the physical world, and causation are related. Given this picture, we can say that mental properties are not identical to physical properties, that they are irreducible to them and different from them, and also that mental causation happens. Below, in chapter 9, I describe this picture. For the moment I want to dwell on Kim’s reasons for adopting his picture.

Traditional discussions of causation and the laws of nature focus on what we may call “the causal connection”: that tie or link between events that amounts to causation (see Dowe 2000 for this usage). These discussions do not discuss other kinds of dependency relations. The fact that the internal angles of a triangle sum to 180 degrees depends on its being a plane trilateral figure, but this dependency is not a causal one. Supervenience is another dependency relation that need not be causal and does not show up in traditional discussions. Mill, for example, thinks of nature as a great uniformity or regularity. This uniformity is composed of smaller regularities (1973:315, 3.4.1): “A certain fact invariably occurs whenever certain circumstances are present, and does not occur when they are absent; the like is true of another fact; and so on. From these separate threads of connexion between parts of the great whole which we term nature, a general tissue of connexion unavoidably weaves itself, by which the whole is held together.… these being pre-supposed, the others follow of course. [These] are called laws of nature; the remainder not; because they are in truth mere cases of [the laws of nature]; virtually included in them; said, therefore, to result from them: whoever affirms [the laws of nature] has already affirmed all the rest.” There is some minimal set of physical properties such that the causal patterns that constitute the progress of the world are a composition of the regularities that obtain among members of that set. These causal patterns determine what is and what is not causally relevant. From this picture of the uniformity of nature it is very easy to derive the claim that only the physical properties are the causally relevant ones: those separate threads of “connexion” composed of relations among the properties from the minimal set are the real motor of the world.

Mill does not so much as glance at dependency relations among properties that are not nomic or causal dependencies. He doesn’t look at logical dependencies; nor does he look at the kind of dependency that holds between exposure to the sun and sunburn (chapter 7), a dependency defined logically in terms of causation. He presupposes that some kinds of connection are causal and that other kinds of connection are simply not under consideration. In particular, it appears that he has already presupposed that non–causal dependency relations, like supervenience relations, are not under consideration.

Kim’s supervenience argument can be seen as a justification for this stance. Kim develops his conception of the way supervenience and causation are related from a marvelous and astonishing passage from Jonathan Edwards’s Doctrines of Original Sin Defended. Suppose that God directly determines how things are at each moment at every point in the natural world. Then any appearance of causation—any appearance of the causal connection—in the natural world is only an appearance. There is no sense in which earlier things make later things happen, or generate them, or produce them. Again, the reflections in a glass of the passing scene appear to display the causal connection, but the reflection of a match’s being struck does not cause the reflection of the match’s lighting. Kim expresses this thought as follows: “Edwards’s dictum. There is a tension between ‘vertical’ determination and ‘horizontal’ causation. In fact, vertical determination excludes horizontal causation” (2005:36). (When an object has its macroproperties in virtue of its parts having their properties and structural relations, the macroproperties of the object are “vertically” determined by the microstructure.) Mental properties supervene on physical properties. By the definition of supervenience, when something has one of these subvening physical properties, then, necessarily, it has the supervening mental property. Hence the fact that someone has the mental properties she does is “vertically” determined by her microstructure. Edwards’s dictum thus entails that there is no horizontal causation of mental things (provided, that is, that mental properties are not identical to what Kim calls “total micro-based properties,” which they will be if they are reducible to physical properties).

What produces the tension Kim describes is the possibility that more than one thing is completely responsible for some phenomenon. Edwards’s God is completely responsible for how things are at any time and place, so nothing else can be. Similarly, the microstructural properties of a person at a time are completely responsible for the fact that that person has the mental properties she does, so nothing else can be.

Kim invokes Edwards’s dictum only at the first stage of his argument, to rule out supervenient effects. As stated, the dictum would seem to rule out supervenient causes as well: “vertical determination excludes horizontal causation” can be understood to entail not just that something that is “vertically determined” cannot be “horizontally caused” but also that it cannot “horizontally cause” anything. But Kim does not use Edwards’s dictum to rule out supervenient causes. Instead he invokes the exclusion principle: “No single event can have more than one sufficient cause occurring at any given time—unless it is a genuine case of causal overdetermination” (2005:42). If M supervenes on P, then if M were to cause something, that effect would have to have a physical cause (either P or some other physical cause). There would then be two causes.

Sometimes it does happen that one event has two (or more) causes, that is, the event is genuinely causally overdetermined. The standard, gory example is that of a firing squad. Suppose three of the shots each by itself would have been sufficient for the death. Then the death was overdetermined. Mills (1996) argues that mental causation is precisely a kind of overdetermination. But that doesn’t fit well with experience or with materialism. I move my arm; there do not seem to be two (or more) causes, each of which by itself would have been sufficient for the arm to move. In fact, the conviction that mental causation doesn’t involve overdetermination is so firmly held that it is easy to overlook the obligation to demonstrate that a response to Kim’s supervenience argument does not involve overdetermination (K. Bennett 2003).

So mental causation does not involve genuine causal overdetermination. So exclusion forces us to select one of the causes, and the best candidate is the physical cause, P. Hence the same tension that underlies Edwards’s dictum underlies Kim’s exclusion principle: since one thing, the physical cause, is completely responsible for the effect, nothing else can be.

Kim’s reasoning concerning causation and supervenience thus depends on the thought that if there were supervenient causation, it would be too much; it would be redundant. Where there is supervenience, the physical base does all the causing there is to do, and there is therefore no room for the supervening things to do any causing.

But there are difficulties for this thought about causation and supervenience. It has been observed more than once (Loewer 2001; Pereboom 2002) that the exclusion principle does not apply in the case where M supervenes on P, since in that case the two are not independent causes. (Kim doesn’t use the word “independent” in this formulation of the exclusion principle, though he does at 1989:239, but the same point can be made about the expression “no single event.” If M supervenes on P, there is a sense, to be described in a moment, in which M and P do not add up to more than one sufficient cause.) There is a parallel difficulty for Edwards’s dictum. Where M* supervenes on P*, Kim says that it is “vertically” determined by P*. But it is not determined in Edwards’s sense. P* is not an existence distinct from M* that brings M* into existence, since P* cannot exist without M* (since, by supervenience, necessarily, anything that instantiates P* at a time instantiates M* at that time). Hence there would be no redundancy of effects if P were to cause both P* and M*, since P* and M* do not add up to more than one distinct effect of P. Moreover, because M and M* are not independent of P and P*, the M–M* relation is not relevantly similar to a pseudoprocess in Salmon’s (1984) sense.

It may be objected that Edwards’s God can make an instance of P* that is not an instance of M*, since Edwards’s God is omnipotent, and an omnipotent God is not limited by physical necessity. But such an act by Edwards’s God in the actual world would be a miracle, in the sense that it would involve breaking the laws of nature. I do not think that we should bring miracles to bear on our theory of causation.

A key premise in Kim’s argument is that M ≠ P. If this is true, how can M and P fail to be distinct causes of P*? How can it be that M and P are not more than one sufficient cause of P*? The answer is that properties overlap: while M and P are not identical, they are not completely distinct properties, either (chap. 6, sec. 6.3, above). I will develop this thought using the abundant ontology of properties I developed in chapter 5. (The idea that mental things are in some respects distinct from and in other respects not distinct from physical things has been made before, in a variety of different ontological settings; see Yablo 1992; Clapp 2001; Watkins 2002; Pereboom 2002).

Kim uses the uppercase letters M, P, and so on, to denote particulars, in this case, property instances. Property instances are individuated with respect to their constitutive objects, properties and times. Kim, I believe, holds that instances of nonidentical properties are always distinct particulars; for example, if the property M is irreducible to physical properties, then every instance of M is a distinct particular from any instance of any physical property P.

I am using the uppercase letters M, P, and so on, to denote properties, not particulars. Two properties may be nonidentical yet not entirely distinct; they can overlap. There are two ways for them to overlap (for ease of exposition I will write as though properties are sets; see above, chap. 5, sec. 5.1):

When two properties overlap, they are not identical, but neither are they entirely distinct. Similar things may be said about property instances. Consider again the properties {Carnap, Quine} and {Quine, Davidson}. Quine is an instance of both properties. The property instance of Quine’s having the first property is not entirely distinct from the property instance of his having the second property. Davidson is an instance of the second property but not of the first, hence the property instance of Davidson’s having the second property is entirely distinct from any instance of the first property.

Here’s an example using mental properties that supervene on physical properties. Suppose the property of being a red afterimage supervenes on the properties of being in various brain states, say, B₁, B₂, and so on. Then, necessarily, if something has B₁, it is a red afterimage. So the property B₁ is a proper subset of the property of being a red afterimage. But something may be a red afterimage but not B₁. So in this case no instance of B₁ is entirely distinct from an instance of being a red afterimage. Some instances of being a red afterimage are entirely distinct from any instance of B₁. And no instance of being a red afterimage is entirely distinct from every instance of every property on which being a red afterimage supervenes.

Kim’s supervenience argument, and the justification I am imputing to him of the traditional practice of ignoring supervenient properties in accounts of laws of nature and causation, depends on supervenient causes and effects being too much causing. But they are too much only if they are completely distinct from their supervenience bases. But they are not. So the argument, and the justification, doesn’t succeed.

Now, various complaints may be made against this critique of Kim’s argument. We might decide to reject supervenient causes on the ground that doing so yields a simpler, cleaner, clearer picture of what causation is. We may feel that the whole avenue of inquiry is obviously closed off from the start. For if supervenient causes (effects) are not distinct from their supervenience bases, then they aren’t anything new. Kim writes, concerning supervenient causation, “This is only a gimmick with no meaning; … inserting a dotted arrow and calling it ‘supervenient causation,’ or anything else (how about ‘pretend’ or ‘faux’ causation), does not alter the situation one bit. It neither adds any new facts nor reveals any hitherto unnoticed relationships. Inserting the extra arrow is not only pointless; it could also be philosophically pernicious if it should mislead us into thinking that we have thereby conferred on M, the mental event, some real causal role” (2005:62). It’s important to avoid a real danger here (Sider 2003). Kim is right that there is a powerful intuition that there’s something fishy about supervenient causation; I’ll take this up in a moment. But it’s easy to mistake the intuition for a real argument that there is something wrong with supervenient causation. It is worth noting, for instance, that this passage expresses a new argument against supervenient causation, not the supervenience argument at all. The supervenience argument was that supervenient causation would be too much. The claim of this passage is rather that supervenient causation isn’t anything at all. But that’s not correct: since the mental properties are not identical with the physical properties, there are new facts and new relationships (see below, chap. 9, especially sec. 9.5).

In any event, Kim offers a coherent conception of causal relevance and supervenience that entails that irreducible supervenient properties are never causally relevant. Kim is right that there is a tension here, and it is important to acknowledge that.

But there is another coherent conception of causal relevance and supervenience. It makes room for supervenient properties to be causally relevant. Hence it makes room for mental causation even if mental properties are not reducible to (and not identical to) physical properties.

Both conceptions are consistent, coherent pictures of how causation works. Kim’s picture respects an intuition for minimalism: the exclusion principle and Edwards’s dictum compact the collection of putative causes down into the ones that are really doing the work.

But Kim’s picture demands that mental properties are identical to physical properties if they are involved in mental causation. This violates an intuition flowing from our conception of ourselves and how we make things happen. When I move my arm, it’s important that the cause should be my choice and that my choice should be something distinctive about the world. It has been easy (for philosophers and nonphilosophers alike) to inflate this sense of distinctiveness into a conviction that minds occupy a fundamentally different kind of reality from physical things. And it has been easy to counter this inflated sense of self-importance by insisting on reductionist physicalism. But there is a middle way. Mental properties are not identical to physical properties, but they also are not completely distinct from them. Our conception of ourselves as not just atoms swerving in the void is vindicated, but at the same time we can say we are not more than atoms swerving in the void. Hence I think the alternative conception is better grounded in our ordinary thinking about ourselves as making things happen.

In the next chapter I articulate the alternative conception: the leveled account of causal relevance. The account says that properties at the fundamental level are causally relevant and that properties at higher levels, perhaps including psychological properties, can also be causally relevant. The picture is in some respects less clean than just banning supervenient causation would produce. But it allows for the causal relevance of supervenient properties. And it does so in a way that honors the fundamental picture of causation and law that we get from Mill: that causal relevance is a matter of being related by laws of nature.

The idea of levels is a familiar one. We say that the liquidity and transparency of water are higher-level features and the charge and geometry of molecules of H₂O are lower level. Macroeconomic trends are high level; individual economic transactions are low level. A computer application (such as emacs or Microsoft Word) is high level; how the chips in the computer work is low level. There are high-level programming languages (Lisp), and there are low-level programming languages (assemblers, perhaps C). The fitness of a particular animal is high level; the detailed physical microstructure of the animal that (for instance) enables it to see a bit more clearly is low level. Thoughts and feelings seem to be high-level features of people, while action potentials in single neurons are lower-level features. The picture of reality dividing into a hierarchy of levels is standard in science and science education. Physics, chemistry, biology, anatomy, psychology, and political science pick out successively higher levels of beings in the natural world: particles, atoms, molecules, cells, organisms, groups of organisms. The picture constitutes a metaphysical view about what the world is like (Oppenheim and Putnam 1958; Wimsatt 1974, 1986).

The idea of levels is closely related to that of supervenience. Roughly speaking, levels imply supervenience, but supervenience doesn’t imply levels. Take the liquidity of water as an example. Liquidity is a higher-level property of water. It supervenes on lower-level properties of water. This means that any sample of water that is liquid has these lower-level properties and that any set of molecules of H₂O that has these lower-level properties will be liquid. So higher-level properties supervene on lower-level properties.

But supervenience doesn’t imply levels. The supervenience relation is far less specific than the relation of levels. For instance, identity is a supervenience relation. Here’s Kim’s definition of mental/physical supervenience again: “Mental properties supervene on physical properties, in that necessarily, for any mental property M, if anything has M at time t, there exists a physical base (or subvenient) property P such that it has P at t, and necessarily anything that has P at a time has M at that time” (1998:9). If we put in the same property for both P and M, this statement will be true. But clearly no property is at a higher (or lower) level than itself. For another example, the property of being either a spoon or a snowflake supervenes on the set of properties {being a spoon, being a snowflake}, yet the former property isn’t at a higher level than either of the latter. And, more troubling for our purposes, we wouldn’t want to say that being either a spoon or a snowflake is a causally relevant feature of things that have that property. Suppose we are comfortable with the idea that spoons make things happen because they are spoons and snowflakes make things happen because they are snowflakes. We are not likely to be comfortable with the idea that having the additional property of being either a spoon or a snowflake makes things happen.

So if we want to defend the idea that supervening mental properties are causally relevant, we need more than the bare idea of supervenience. Now, we could at this point turn to the project of building up from idea of supervenience to the idea of levels. The project would be to say what it is, in addition to supervenience, we need in order to have something that satisfies our commonsense idea that we have something higher level. Alternatively, we could just assume that some idea or other of levels is perfectly acceptable.

Either strategy is open to a very general philosophical critique: that there is something badly wrong with the very idea of levels. Kim 2002 and Heil 2003a, 2003b, for instance, sharply criticize leveled or layered accounts of reality.

There are several kinds of complaints. Perhaps the most pressing is that a leveled account of properties implies that only the fundamental level has any causal relevance. If things at the fundamental level do the causal work, how could anything at any higher level make any difference? This is my main concern in this book: to explain how higher-level properties could have causal relevance. The full explanation comes in the next chapter; so let me set this complaint aside for the moment.

Another kind of complaint is that adding levels beyond the fundamental level is adding extra unnecessary things to our ontology. Levels are supposed to bloat our ontology. This isn’t a worry for the abundant conception of properties that I described above in chapter 5, since I’ve already said that for every set of actual and possible entities there is a property.

Yet another set of complaints revolves around the ideas of supervenience and multiple realization. Higher-level things supervene on lower-level things and, often, are multiply realized by them. Both supervenience and multiple realization have been argued to be problematic (Heil 1998, 1999; Shapiro 2000). The main problem with supervenience is supposed to be that it doesn’t explain anything; as I argued above, that’s right, but that doesn’t show that mental properties do not supervene on physical properties, and, indeed, there is a good reason to think that they do (sec. 8.1.2, above). Gillett 2003 and Aizawa 2006 argue that the supposed difficulties for multiple realization are only apparent.

But there is another and more serious problem. We don’t have a well-defined conception of what levels are. Consider again the familiar claim that the following sciences are arranged according to increasing level: physics, chemistry, biochemistry, biology, psychology, sociology. What principle governs dividing things into levels: Is it size? How much energy goes into interactions at a level? Some of the principles people have suggested have absurd consequences (Kim 2002). Some properties of things are shared at every level: protons and hearts and people and galaxies all have mass. Further, the hierarchies are local. The heart is an organ and so sits at a certain level relative to organisms. But does it make sense to say that it is at the same level as or at a different level from a part of a mountain or a part of a car? The problem is that it is doubtful that there is “a single hierarchy of connected levels, from higher to lower, in which every object and phenomenon in the natural world finds its ‘appropriate’ place” (Kim 2002:16). There is no monolithic conception of levels, in the sense that for every pair of things in the world, the conception tells us that these two items are on the same level or that one is higher than the other. It seems quite likely that there are several distinct conceptions of levels that divide various portions of reality in different ways for different purposes. For instance, the sense in which the heart is high level relative to its valves is certainly distinct from the sense in which the Lisp programming language is high level relative to the C programming language.

These are serious and interesting problems. But they do not amount to a fundamental problem for the very idea of levels. Rather, they mean that when we talk about levels, we need to be clear about what kind of level we are talking about.

Craver (2007: chap. 5) argues that many or most of the things that people have wanted to say about levels come out true if we take them to be talking about levels of mechanisms. The “levels of mechanisms” conception is not monolithic: many things are not mechanisms or parts of mechanisms or composed of mechanisms at all, and so they get no level in this conception. And even when two things are parts of the same mechanical system, it may not make sense to ask whether they are at the same or different levels. For instance, it might make sense to say that a particular valve is lower level relative to the fuel-mixing mechanism of an automobile engine and that a particular chip is lower level relative to a control computer in the same automobile, but it doesn’t make sense to say that the valve and the chip are higher or lower level relative to one another.

Craver argues that explanation in neuroscience integrates explanations from many levels. Neuroscience aims to understand the brain. Understanding the brain requires many strategies, techniques, methods that explore mechanisms of the brain at different levels. Hence neuroscience presupposes the existence of real levels of mechanisms in the brain and body.

There are two philosophical dangers here. One is arguing from the practice of science to metaphysical conclusions. Just because well-established sciences use the idea of levels doesn’t show that the world actually has levels; it might be that the idea is a convenient way to talk about a world devoid of levels. The second danger is arguing from claims about levels and explanation to metaphysical conclusions about how the world actually is. Again, it might be that there are different ways to explain things and that we tend to think of these ways in terms of levels of things, yet nevertheless reality itself doesn’t actually have levels.

These are serious concerns. Again, just because they are possible worries about whether there really are levels of properties, independent of what people (including scientists) say or think about them, doesn’t mean that we should give up on the very idea of levels of properties, that is, the metaphysical idea that in addition to the fundamental properties there are also properties at many levels.

Mechanisms, for instance, are objective features of reality. They have parts, which interact, and when they do, the entire mechanism exhibits particular behaviors. Mechanisms in turn interact with other mechanisms, and when they do, the larger mechanisms of which they are a part exhibit their own distinctive—higher-level—behaviors. The history of the life sciences, and the history of neuroscience, is full of research that explains the behavior of complex entities by revealing the mechanisms that produce that behavior. When a photon hits a cone cell, it can cause a nerve impulse. How does it do this? The first step is that the added energy of the photon causes a molecule of rhodopsin to change from the 11-cis to the 11-trans form. This is part of the mechanism by which cone cells turn light into nerve impulses. There is a higher-level mechanism that takes light as input and produces representations of the world; one of its parts is the mechanism that generates nerve impulses from photons in cone cells.

Hence the levels of mechanisms conception shows how the things that we do (for instance, speaking or acting) can be explained in terms of mechanisms at a lower level. The behavior of these mechanisms in turn is explained in terms of the behaviors and causal powers of the mechanisms that compose them. It may be, for instance, that speaking involves a computational mechanism that looks up words in a lexicon given a semantic representation. This mechanism will be composed of further mechanisms that do simpler things. And so forth.

The idea that there are levels and that things at various levels have causal powers does not depend on any particular account of how the world actually works. It is to be expected that we will change our thinking about the sciences of the mind and the brain as we learn more. But we aren’t going to outgrow the idea that the behavior of complex things can be explained by the composition of the behavior of their simpler parts. And so we aren’t going to outgrow the idea of thinking of reality in terms of levels.

So the very general philosophical complaints against the idea of levels can, I suggest, be answered. I am about to turn to the most serious, the worry that higher-level properties would be causally irrelevant. For present purposes, I need not do further work on the idea of levels. There is every reason to think that there is some reasonable conception of levels, such that mental properties are higher-level properties than microproperties of the brain.

The most basic reason for thinking that mental causation is impossible stems from the claim that when a property supervenes on a causally relevant property, it is not itself causally relevant. The argument for this claim is that if the supervenient property were causally relevant, it would be too much: supervenient causation would be overdetermination.

As I’ve shown, when one property supervenes on another, they overlap and hence are not entirely distinct. Consequently, the supervening property and its supervenience base cannot overdetermine their effect; so supervenient causation would not be too much causation.

I argued that supervenience is necessary for mental causation (given physical closure and the nonidentity of mental properties and physical properties). It is not sufficient. Supervenience is a very weak relation, and many supervening properties are causally irrelevant. We want a stronger relation: mental properties are higher-level properties relative to their supervenience bases.

If mental properties are higher-level properties, and higher-level properties can be causally relevant, then we can solve the mental causation problem. The remaining difficulty, then, is to show how a higher-level supervening mental property may be causally relevant, to show how such a property—like the physical properties on which it supervenes—can stand in the right relation to the laws of nature to count as causally relevant (the gold standard).

Supervenience is almost exclusively a technical term in philosophy—mainly in the philosophy of mind but widely used in many other areas of philosophy as well. Terry Horgan’s “From Supervenience to Superdupervenience” (1993) is an excellent account of the history, character, and usefulness of the supervenience relation. Jaegwon Kim wrote many of the seminal papers on supervenience; they are collected in his Supervenience and Mind (1993a). McLaughlin and Bennett’s “Supervenience” (2005) is a comprehensive account of varieties of supervenience and their relations. John Heil’s “Supervenience Deconstructed” (1998) and his From an Ontological Point of View (2003a) argue that supervenience explains nothing; consequently, we should avoid its use in working on the metaphysics of mind.

Oppenheim and Putnam’s “Unity of Science as a Working Hypothesis” (1958) is an early exposition of a leveled account of reality. William Wimsatt’s papers “Complexity and Organization” (1974) and “Forms of Aggregativity” (1986) characterize levels somewhat differently. Jaegwon Kim’s “The Layered Model: Metaphysical Considerations” (2002) argues that none of the standing conceptions of levels is completely coherent. Carl Craver argues persuasively in his Explaining the Brain: Mechanisms and the Mosaic Unity of Neuroscience (2007) that there are coherent conceptions of levels and that neuroscience makes use of the concept of levels of mechanisms.