Strengthening the exclusion argument

Abstract

As conceived by Kim, the causal exclusion argument targets all forms of nonreductive physicalism equally, but by restricting its focus to functionalist varieties of nonreductivism, I am able to develop a version of the argument that has a number of virtues lacking in the original. First, the revised argument has no need for Kim’s causal exclusion principle, which many find dubious if not simply false. Second, the revised argument can be adapted to either a production-based conception of causation, as Kim himself favors, or to any of a number of dependence-based conceptions, like the ones favored by many of Kim’s critics. And, finally, the revised argument does not have the objectionable consequence that all so-called higher-level properties are epiphenomenal, for it does not generalize in the way that Kim’s original version of the argument arguably does. Nor does it concede much to narrow the scope of the argument in the way proposed. Those who adopt nonreductive theories of mind do so, by and large, on the strength of functionalist arguments for the multiple realizability of mental states. If functionalism entails that mental properties are epiphenomenal, this thus deals a critical, if not quite fatal, blow to nonreductivism.

1 Introduction

The causal exclusion argument can be strengthened considerably by narrowing its scope. As conceived by Kim (1989a, 1993a, 1998, 2005), the exclusion argument targets all forms of nonreductive physicalism¹ (NRP) equally, but if we restrict its focus to functionalist varieties of NRP, it is possible to develop a version of the argument that has a number of virtues lacking in the original. First, the revised argument has no need for Kim’s exclusion principle, which many find dubious if not simply false.² Second, the revised argument can be adapted to either a production-based conception of causation, as Kim himself favors, or to any of a number of dependence-based conceptions, like the ones favored by many of Kim’s critics. And, finally, the revised argument does not have the objectionable consequence that all so-called higher-level properties are epiphenomenal, for it does not generalize in the way that Kim’s original version of the argument arguably does. Nor does it concede much to narrow the scope of the argument in the way proposed. Those who adopt NRP do so, by and large, on the strength of functionalist arguments for the multiple realizability of mental states.³ If functionalism entails that mental properties are epiphenomenal, this therefore deals a critical, if not quite fatal, blow to NRP.

I begin making this case in the following section with a bit of Kim exegesis. My aim here is to show that there are two versions of the exclusion argument, that one of these versions contains an assumption missing from the second, and that many of the most important responses to Kim share this assumption. In section three, I argue that this shared assumption is illicit in the context of a role functionalist account of mental states and propose a much more defensible alternative. In section four, I offer an alternative to Kim’s exclusion principle, one that is logically weaker and far less contentious. In section five, I pull these various threads together to formulate what I’m calling the functional exclusion argument, each premise of which is either equivalent to or more strongly supported than the corresponding premise in Kim’s original. Here I follow Kim in assuming a production-based conception of causation. In section six, however, I show that a parallel argument can be made in terms of dependence-based conceptions of causation. Finally, in the concluding section of the paper, I argue that the narrower scope of the revised argument is actually a strength, for there are reasons to believe that the argument does not generalize to threaten the efficacy of all higher-level properties.

2 Kim’s two exclusion arguments

Kim has given two formulations of his causal exclusion argument over the years; sometimes, as in his (2005), one finds both formulations in the same writing.⁴ But neither Kim nor his commentators appear to believe that these different formulations correspond to significantly different problems. Many commentators fail to acknowledge that these different formulations exist, even when they are responding to texts in which Kim is explicit on the point.⁵ And when remarking on their differences, Kim himself says only that one formulation is “simpler” and the other “more intuitive,” insisting that “the moral … is the same” (2005, pp. 44–45).⁶ Kim and his commentators are mistaken on this point, for these different formulations really do correspond to significantly different problems. More precisely, the first, ‘more intuitive’ version of the problem is a special case of the second, ‘simpler’ version—a special case that applies only when further—and, as we’ll see, unrealistic—conditions are met. For this reason, I’ll refer to the former as the special exclusion problem and to the latter as the general exclusion problem. I’ll refer to the corresponding arguments as the special and general exclusion arguments, respectively.

Both versions of the exclusion argument take the form of a reductio ad absurdum. The following theses, each of which is said to be essential to any defensible version of NRP, are claimed to be collectively inconsistent:

Physicalism: Every property is either physical or supervenient on the physical.⁷

Closure: If a physical event has a cause at t, it has a sufficient physical cause at t.

Distinctness: Mental properties are distinct from, and irreducible to, physical properties.

Exclusion: No event can have more than one sufficient cause occurring at any given time—unless it is a genuine case of causal overdetermination.

Supervenience: Any event that causes a supervenient property to be instantiated must cause some minimal physical supervenience base of that property to be instantiated.

Efficacy: Mental properties are causally efficacious.⁸

Kim often provides an independent argument for Supervenience, but I’ll simply treat it as a coequal thesis in what follows. This is harmless because the more basic principles from which it derives are uncontroversial for defenders of NRP.⁹

Supervenience refers to a property’s minimal physical supervenience base, which is the physical supervenience base that is minimally sufficient for the property’s instantiation on a given occasion.¹⁰ If a property supervenes on some base B, it supervenes on everything that contains B as a proper part. But any of a property’s minimal supervenience bases suffices for its realization, so it’s only one of these minimal bases that needs to be caused. I refer to bases in the plural because, of course, multiply realizable properties will have more than one minimal supervenience base. In such cases, no specific realizer is required for the property’s realization, but some realizer or other is. Supervenience thus requires only that some minimal supervenience base of the supervenient property be caused. As these considerations suggest, we must be careful when we refer to a property’s supervenience base in the singular—as Kim often does, and as I will do below. When I refer in what follows to the supervenience base of some property F or write in some other way that implies uniqueness, I should be taken to mean, in the more cumbersome but more accurate longhand, the minimal supervenience base of F that is actually instantiated at some time t and that grounds or explains F’s instantiation at t.¹¹

The special exclusion argument proceeds as follows.¹² Suppose that Physicalism, Supervenience, and Efficacy are true. It follows that some mental event M causes some physical event P* at some time t.¹³

(1) M causes¹⁴ P* at t.

From Physicalism it follows that

(2) M has a physical supervenience base P.

But, as Kim says, “[t]here are strong reasons for thinking that P is a cause of P*” (2005, p. 41). We’ll return to these reasons below. For now, let us simply grant Kim his premise and see where it leads. Doing so leaves us with two causes of P*.

(3) M causes P* at t and P causes P* at t.

From Distinctness, it follows that

(4) M ≠ P.

We therefore have distinct events M and P that are each causally sufficient for P*. But, as M and P are not independent, this is not a case of what Kim deems genuine overdetermination.¹⁵ Thus, by Exclusion, only one of these events can cause P*. Choosing M over P launches a regress, for Closure tells us that P* must nevertheless have some sufficient physical cause—P′, say—and Distinctness tells us that it must be distinct from M. But M and P′ cannot be independent causes of P*, for “this has the implausible consequence that it makes every case of mental causation a case of overdetermination” (1998, p. 44).¹⁶Exclusion thus entails that M and P′ cannot both cause P*. We could opt for M again, but then we’d need some other physical cause for P*, leaving us in the same situation as before. The only way to stop this regress is to choose one of these Ps over M. The cause of P* must then be a physical event and only a physical event, from which it follows that

(5) M does not cause P* at t.

And because the choice of M, P*, and t was completely arbitrary, the argument generalizes to show that no mental event can cause any physical event. This together with Physicalism and Supervenience allows us to conclude that Efficacy is false.

Kim’s second version of the argument, which I’ll be calling the general exclusion argument, is presented in his (2005) and (2009). It can be seen as a streamlined version of the first. Suppose, as before, that Physicalism, Supervenience, and Efficacy are true. It follows that some mental event causes some physical event.

(1′) M causes P* at t.

From Closure, it follows that P* has some sufficient physical cause P occurring at t.

(2′) P causes P* at t.

From Distinctness, it follows that

(3′) M ≠ P.

Thus P* has two distinct causes at t, and, once again, this is not a case of genuine overdetermination. But Exclusion tells us that only one of these events can be the cause of P* at t, and, as before, we must choose P over M to avoid a regress. We thereby reach the same conclusion as before.

(4′) M does not cause P* at t.

This conclusion can be generalized and combined with Physicalism and Supervenience to yield a refutation of Efficacy.

I’ve presented both versions of the argument in full so that their difference will be apparent. Clearly the special exclusion argument includes an assumption that the general exclusion argument does not. More precisely, the special exclusion argument assumes that the physical supervenience base of M is also the physical cause of P*. Letting ‘P_M’ designate the former and ‘P_C’ the latter, we can see that (3) doesn’t follow from (2) unless we assume the following premise:

(2.5) P_M = P_C.

This premise is obviously not licensed by any general supervenience thesis, for such theses tell us only that M must supervene on some physical event or other, not that it must supervene on the very physical event that causes P*. Nor does Kim imagine otherwise. When he defends (2.5), he appeals not simply to P’s status as the supervenience base of M but also to M’s status as a cause of P*. In other words, Kim defends (2.5) by appealing to Efficacy. Here he is making this defense in his (1998):

If you take causation as grounded in nomological sufficiency, P qualifies as a cause of P*, for, since P is sufficient for M and M is sufficient for P*, P is sufficient for P* (Kim 1998, p. 43; my italics).

And here he is saying much the same thing in his (2005):

P is (at least) nomologically sufficient for M, and the occurrence of M on this occasion depends on, and is determined by, the presence of P on this occasion. Since, ex hypothesi, M is a cause of P*, P would appear amply to qualify as a cause of P* as well (2005, p. 41; my italics).

As we’ll see below, one must take care when moving between claims of nomological and claims of causal sufficiency. A causally sufficient condition is a nomologically sufficient condition that is also a cause. Causal sufficiency thus implies nomological sufficiency, but the converse is not true. If it were, causal relations would come too cheap. But all of this will be litigated later in the paper. For now, I simply want to point out that these arguments for (2.5) assume that M causes P* and thus that mental properties are causally efficacious.¹⁷

Now Kim is quite within his rights to appeal to Efficacy in this context, for it’s a conjunct in the conjunction of theses he’s attempting to reduce to absurdity. And, by the same token, defenders of NRP are within their rights to appeal to Efficacy in order to show that Kim’s attempted reductio fails. The point I want to emphasize is simply that (2.5) depends on Efficacy for its justification and is thus unjustified whenever Efficacy cannot be assumed. I will argue below that (2.5) is likely false if we assume a role functionalist account of mental states and that its falsity implies the falsity of Efficacy. One cannot in this context appeal to Efficacy in defense of (2.5) without simply begging the question.

It does seem that Efficacy is built into the very meaning of M, which, recall, is defined in the context of the exclusion argument as the mental cause of P*. But if that’s true, and if Kim’s reasoning in the passages quoted above is sound, doesn’t it also follow—by definition, as it were—that P_M = P_C? Indeed it does. But remember that we are dealing with a reductio argument in which nothing that is said cannot be gainsaid. If functionalism gives us reason to believe that the supervenience bases of mental states are not sufficient causes of events like P*, it follows that P_M, in particular, is not a sufficient cause and thus that P_M ≠ P_C. A contradiction! Efficacy tells us that (2.5) is true, but functionalism tells us that it is false. Efficacy and functionalism are inconsistent.¹⁸

It’s the special exclusion argument that begets the diagram that is ubiquitous in the literature and in Kim’s writings, a diagram which is reproduced in Fig. 1 below.

Fig. 1

The ubiquitous exclusion diagram

Here, of course, the solid arrow at the bottom of the figure represents genuine or physical causation, the vertical arrows represent supervenience relations, and the dashed arrow at the top represents the dubious relation of ‘supervenient’ causation. This diagram is so closely associated with the exclusion argument that it’s often used as a kind of visual shorthand for the argument itself, as in Heil (2013, pp. 20–21) and Woodward (2015, p. 306; 2017, p. 251). Indeed, the diagram in Fig. 1 is often taken to be the common ground over which the war for Efficacy is to be fought, with one side arguing that the situation depicted is one in which M can be said to cause P* and the other side arguing that it is not. This is very telling because the diagram clearly embodies the assumption that P_M = P_C, for the event generically named ‘P’ is obviously meant to refer to both the physical cause of P* (by means of the horizontal arrow) and the physical supervenience base of M (by means of the vertical arrow). If (2.5) is false, Kim’s iconic diagram is misleading.

The special exclusion argument and its accompanying diagram give rise to what I’m calling the special exclusion problem, a special case that applies only when P_M = P_C. The general exclusion argument, making no use of such an assumption, gives rise to a more general exclusion problem.

Special Exclusion Problem: How can M cause P* if (i) P_C causes P*, (ii) P_C is distinct from M, (iii) M supervenes on P_M, and (iv) P_M = P_C?

General Exclusion Problem: How can M cause P* if (i) P_C causes P*, (ii) P_C is distinct from M, and (iii) M supervenes on P_M?

Note that clause (iii) in the formulation of the general exclusion problem says only that M has some physical supervenience base or other, this being required of any form of physicalism. I’ve made this commitment explicit to highlight the fact that it is the addition of clause (iv) that distinguishes the special exclusion problem from the general.

It is the special rather than the general version of the exclusion problem that many of the most important responses to Kim are attempting to solve—which is not to say that this is their stated aim. It’s often unclear whether the authors of these responses even recognize that the exclusion argument can take a more general form. We know that they are proposing a solution to the special problem only because they assume (2.5). One finds Woodward, for example, arguing that the question at the heart of the exclusion problem “has no coherent ‘interventionist’ interpretation” (2008, p. 255) and that it is part of a family of “misguided counterfactual queries about what would happen under antecedents that cannot possibly be realized” (2015, p. 329).

If it is “metaphysically impossible” to change the value of a supervening variable like [M] while holding [P] fixed, then the very fact that this is impossible is itself an indication that counterfactuals with this antecedent do not tell us about the causal effect (or the absence of such an effect) of [M] on other variables (2015, p. 335; substituting ‘P’ for ‘P₁’ and ‘M’ for ‘M₁’).

On Woodward’s interventionist account of causation in order to conclude that it’s P rather than M that causes P*, we are required to consider scenarios in which P is present in the absence of M. But this is said to be impossible. Why? Certainly, it’s impossible for P_M to be present in the absence of M, for the latter is necessitated by the former. But this is irrelevant to the general exclusion problem, which sets up M and P_C rather than M and P_M as competing causes. M and P_M become competing causes only if we assume that P_M = P_C, as Woodward clearly does. We’ll see below that M actually fails the interventionist test for causal relevance if we instead assume a role functionalist account of the relation between P_M and P_C.

Crane (2001) also assumes (2.5) in his attempt to dissolve the exclusion problem with the help of a counterfactual theory of causation. Here is a succinct summary of his position:

A mental cause M counts as a cause of physical effect [P*] because if M had not been there, [P*] would not have been there. That is just to apply the counterfactual criterion. But the simultaneous physical cause P upon which M supervenes (M’s ‘supervenience base’) also counts as a cause of [P*] because if P had not been there, [P*] would not have been there (2001, p. 61; my italics and substituting ‘P*’ for ‘E’).

Note that in the italicized portion of this passage P is introduced as both the supervenience base of M and the physical cause of P*—as both P_M and P_C. Crane is even more explicit just after the quoted passage when he baldly states that “P both causally determines [P*] and metaphysically determines M” (2001, p. 61; ‘P*’ for ‘E’).

List and Menzies (2009) go so far as to include the assumption of (2.5) in their formulation of the exclusion principle:

Exclusion Principle: If a property F is causally sufficient for a property G, then no distinct property F* that supervenes on F causes G (2009, p. 478; italic added and omitted).

Kim, of course, does not restrict the scope of Exclusion in this way.¹⁹ If he did, his second version of the exclusion argument would be a blatant non-sequitur. With their reformulation, List and Menzies collapse the distinction between the two versions of Kim’s argument and the different problems to which they give rise.

It’s not only proponents of dependence-based accounts of causation that fail to acknowledge the more general problem. Here, for example, is Shoemaker’s gloss on exclusion:

The idea is that it is the physical realizers of mental properties that “do the causal work,” and that if these are not identical with the mental property then they preempt whatever causal role the mental property might otherwise seem to have (2007, p. 4).

Shoemaker is clearly describing the special exclusion problem in this passage, for he is assuming that the competing causes are M and P_M, which is true only if P_M = P_C. Moreover, and more importantly, his response to the exclusion problem makes the same assumption. According to the powers subset theory of property realization—defended by Shoemaker in his (2001), (2007), and (2013), but also defended by Wilson (1999, 2011) and Clapp (2001)—the causal powers of instances of realized properties like M are a proper subset of the causal powers of the instances of their realizers like P_M. Every token power of an instance of M is therefore identical to some token power of an instance of P_M. It follows, Shoemaker argues, that the efficacy of P_M cannot compete with the efficacy of M. But notice that this tells us nothing about cases in which the threat to M comes from a P_C that is not also a P_M. Nor for that matter do the closely related attempts by Yablo (1992) and others to interpret the mental/physical relation along the lines of the determinable/determinate relation, for determinables are realized by and thus supervene upon their determinates.

I will not labor the point any further. Suffice it to say that many of the most notable responses to Kim address the special rather than the general version of the exclusion problem. This matters because there’s another special case to consider, the one represented in Fig. 2 below.

Fig. 2

An alternative exclusion diagram

As before, the solid line at the bottom of the figure represents physical causation, the dashed line at the top represents the problematic relation of mental causation, and the doubled lines along the sides represent supervenience relations. The triplet of vertical arrows on the left is meant to indicate that it’s the conjunction of the physical events P₁…P_C… P_n that acts as the supervenience base of M. The solid arrow at the bottom indicates that it’s only P_C that acts as the cause of P*. Thus whereas Fig. 1 represents the physical cause of P* as being identical to the physical supervenience base of M, Fig. 2 represents P_C as being only one among a number of physical events which collectively act as the supervenience base of M.²⁰ That is, Fig. 2 implies that it’s not P_M that causes P* but only some proper part thereof. I’ll argue for this more rigorously below, but it should be intuitively obvious that the situation depicted is not one in which M can be said to cause P* and that this remains true whether we interpret causation along productionist lines or in terms of counterfactual dependence and difference-making. This is all the more obvious if we assume that the cooccurrence of the events P₁…P_C…P_n which constitute P_M is nomologically contingent and that, in particular, P_C can occur (and cause P*) in the absence of the others. This is a point of some consequence because the situation depicted in Fig. 2 is the situation that role functionalism tells us actually obtains.

3 Core and total realizers and a Parthood principle

Thus far we’ve seen that the special exclusion argument assumes that mental events cause physical events in virtue of their supervenience bases. I’ll argue in this and the following section that this assumption is likely false if a functionalist account of the mental states is correct. I’ll propose in its stead a premise that is likely true—namely, that the physical cause of an event like P* is only a proper part of the physical supervenience base of an event like M. My argument for this premise will rest on the distinction between the core and total realizers of a mental state. Very roughly, I’ll argue that mental states cause things in virtue of their core realizers and certain background conditions but supervene on their total realizers and that the conjunction of a mental state’s core realizer and the aforementioned background conditions almost always includes no more than a proper part of its total realizer.²¹

Let us indulge in the familiar philosophical fiction that human pain is realized by the firing of our C-fibers. For role functionalists, this is at most part of the story, for C-fiber firing could only be what Shoemaker (1981) calls the ‘core realizer’ of pain. According to the functionalist²² account, pain and other mental states are second-order states realized by various first-order states when and only when the latter occupy a certain causal role.²³ Thus if C-fiber firing realizes pain in humans, this is only because it is the state that, e.g., is caused by tissue damage, causes grimacing, wincing, and groaning, and, when accompanied by the appropriate beliefs and desires, causes avoidance behavior. For C-fiber firing to occupy such a role, a considerable number of background conditions must also be in place. Obviously, C-fiber firing can’t be caused by tissue damage unless there are tissues to be damaged, and it can’t couple with beliefs and desires to cause avoidance behavior unless there are beliefs and desires with which to be coupled. Less obviously, C-fiber firing can’t have its requisite causes and effects unless the appropriate nervous pathways exist. If, for example, the pathways from our C-fibers to our motor cortex were severed, C-fiber firing couldn’t cause avoidance behavior even in someone who possessed the appropriate beliefs and desires.²⁴ It’s the totality of these background conditions together with its core realizer that Shoemaker calls the ‘total realizer’ of pain.

It’s only the total realizer of a mental state that is properly so-called, for it’s only the total realizer that is sufficient for the state’s realization. This, of course, is a direct consequence of defining mental states in terms of their causal roles. A solitary lump of C-fibers firing away in an isolated petri dish does not occupy the pain role and therefore cannot realize pain. Nor will C-fiber firing realize pain if it occupies the wrong causal role. Consider Lewis’s madman, in whom C-fiber firing is caused not by “cuts, burns, pressure, and the like” but by “moderate exercise on an empty stomach,” causes a person to “cross his legs and snap his fingers” rather than to “groan and writhe,” and acts not as a distraction but rather “turn[s] the mind to mathematics” (1983, p. 122). Mad pain isn’t pain even though the two have the same core realizer because the roles of these realizers differ. Pain and mad pain have the same core but different total realizers. In general, any alteration—whether by addition, deletion, or substitution—to the constitutive causal role of a mental state’s core realizer is an alteration to its total realizer.

Total realizers are usefully thought of as conjunctive states with core realizers as one of their conjuncts. If C-fiber firing acts as the core realizer of pain in our central nervous systems, there must be some complementary physical state that obtains whenever, as Shoemaker puts it, a system is “physically constituted in such a way that [C-fiber firing] plays the causal role definitive of pain” (1981, p. 97). Call this state R. The total realizer of pain would then be the conjunction of C-fiber firing and R—the conjunction, that is, of C-fiber firing and everything else that’s required for C-fiber firing to play the pain role. R itself can be thought of as the difference between pain’s total and its core realizer—as that which remains when the latter is subtracted from the former. It’s a difference in this remainder—a difference in R—that accounts for the difference between the total realizers of pain and mad pain. When C-fiber firing occurs in the context of R, it (core-) realizes pain; when it occurs in the context of R′ (≠ R), it (core-) realizes mad pain.

It should be obvious that what we’re now calling M’s total realizer is just M’s supervenience base—what we took to calling P_M in the previous section—for M’s total realizer is the physical state that is metaphysically sufficient for the realization of M, and this is just the physical state upon which M supervenes. What therefore remains to be shown is that P_C—the physical state that causes P* in the context of Kim’s exclusion argument—is only a proper part of M’s total realizer, for then it will be the diagram in Fig. 2 rather than the diagram in Fig. 1 that correctly represents functionalism’s problem with mental causation. To establish this, it will be helpful to introduce some more exhibits.

Consider the diagrams in Figs. 3 and 4 below. Neuron diagrams of roughly this sort are familiar enough, but I am departing from convention slightly by using them to represent causal networks in addition to the causal processes occurring therein. In fact, it will help to adopt a kind of thoroughgoing literalism and imagine that what’s being represented are networks of actual neurons, stimulating and inhibiting each other by means of their various inter-neuronal connections. Circles will then represent particular neurons in a network, with the shading of a circle indicating that the corresponding neuron is firing. Connections between circles will represent relations of type causation between the states of the corresponding neurons, with connections between shaded neurons indicating that a token of the type has actually occurred. Arrows will indicate excitatory connections, and terminal bulbs will indicate inhibitory connections. And, finally, inhibitory stimulation will be sufficient for inhibition and excitatory stimulation in the absence of inhibition will be sufficient for excitation. The leftmost panel in Fig. 3 thus indicates that the firing of neuron P generally causes the firing of neuron P*, that neuron P is actually firing, and, perforce, that neuron P* is about to fire. (I’m also departing from convention by allowing diagrams to represent momentary ‘snapshots’ of causal processes, which I’ve done to illustrate a subtle but important point.) For ease of presentation, I’ll sometimes speak of P causing P*, or the state P causing the state P*, when what I mean in the more accurate longhand is that the firing of P causes the firing of P*.

Fig. 3

Core and total realizers in a complex network

It’s important that we interpret the causal relations in these diagrams as being nomologically contingent. This is the case in real neural networks and in causal networks more generally. No neuron in our central nervous system will cause another to fire unless the two are connected by means of the appropriate neuronal pathways, and whether or not these pathways exist is a contingent matter. Lewis’s mad pain would be unthinkable unless the various afferent and efferent connections of our C-fibers could be rearranged. Similarly, it is only a contingent matter that the device acting as a carburetor sends a mixture of air and fuel to the engine or that the organ acting as a heart circulates blood throughout the body. Rearrange the inputs and outputs of either apparatus and it will cease to perform its dedicated function.

Consider now the network depicted in Fig. 3. Let M be the second-order state of being in some first-order state that is caused by states A, B, and C, that causes state P*, and prevents the occurrence of states D and E.²⁵ P is the first-order state that occupies this M role and is thus the core realizer of M. This is shown in the center panel of the triptych, which is intended to represent a portion of the network on the left rather than its own standalone network. R, which is shown on the right, is the context in which tokens of P will (core-) realize M and thus consists of P* together with the rest of the network. P* has been left unshaded in the first panel to indicate that, strictly speaking, M is realized at the very instant that P occurs in the context R and is thus realized before the occurrence of P*. Nor does R include the actual tokening of P*. R consists only of whatever makes it true that tokens of P realize M. In neuronal terms, R consists of the presence of the other neurons, their readiness to fire upon being stimulated, the nervous pathways connecting them to P, and the like. M is realized only when its total realizer occurs, which happens only when P occurs in the context R.

The critical point is that almost all of R (and, hence, much of P_M) could go missing without affecting whether the firing of P will cause the firing of P*. To make this point more vivid, it is helpful to employ a device used by Antony (1994) and Bartlett (2014) in their arguments against functionalist theories of consciousness.²⁶ Imagine that the following process occurs in the network depicted in Fig. 3: neuron B fires, causing the firing of P, which causes the firing of P*. But imagine that just as the nervous impulse from B arrives at P we eliminate all of the connections in the network save the one linking P to P*. We could even imagine eliminating the superfluous neurons as well. None of this will affect P’s capacity to cause the firing of P*. It follows that P would cause P* even if it did not realize M. P_C is only a proper part of P_M

P is the core realizer of M and also, it seems, the physical cause of P*. If so, and if this identity holds generally, it follows straightaway and as a general rule that P_C is only a proper part of P_M. Consider that M’s core realizer will always be a proper part of its total realizer, for otherwise R will be empty and there will be no (nontrivial) sense in which M is even a functional property.²⁷ Thus if P_C is identical to M’s core realizer and if, as we’ve already established, P_M is identical M’s total realizer, it follows that P_C is a proper part of P_M. But, alas, the truth of the matter is slightly more complicated. Note that P causes P* in Fig. 3 only because certain background conditions also obtain. If P weren’t connected to P* by means of an excitatory connection, for example, the firing of the former would not cause the firing of the latter. Background conditions of this sort satisfy both production-based and dependence-based criteria of causal relevance and ought plausibly to be included in P_C.²⁸ Let us grant that they are. What needs to be established, then, is that M’s core realizer and the relevant background conditions are only a proper part of P_M. This will almost certainly be true in cases that are relevant to a functionalist theory of mind. To see why, it is useful to consider the rare case in which P_C and P_M coincide.

Consider the network depicted in Fig. 4. Here we have a kind of limiting case in which core and total realizers nearly coincide. M is now the second-order state of being in some first-order state that causes P*. P is the first-order state that occupies the M role and is thus the core realizer of M. But here we see that the background conditions necessary for P to cause P* are the very same background conditions that allow tokens of P to (core-) realize M. It is only the firing of P in the context R that causes the firing of P*, but the firing of P in the context R is simply the total realizer of M. Here, then, we have a situation in which P_C and P_M are one and the same.

Fig. 4

Core and total realizers in a (maximally) simple network

There’s a lesson to be learned from this counterexample. It fails to be true as a general rule that P_C is a proper part of P_M only because there are cases in which the background conditions necessary for M’s core realizer to cause P* include the background conditions necessary for M’s core realizer to realize M. But such cases are extremely unusual and almost certainly irrelevant to a functionalist account of mental states.²⁹ In Fig. 4, there is a coincidence of these conditions only because P’s functional role is limited to causing the very effect it is currently causing. This will fail to be the case in networks with only slightly more complexity. Let M be the second-order state of being caused by C and causing P*. R is now only minimally more complex than before, involving not just an effect but also a cause, but already the background conditions necessary for P to cause P* do not include the background conditions necessary for P to (core-) realize M, for neither the presence of C nor its capacity to cause the firing of P is necessary for P to cause the firing of P*. P’s firing would cause the firing of P* even if C were completely eliminated.

But a network of this type is still far too simple to represent the functional role of a mental state on any remotely plausible analysis. A given mental state will have many possible causes—how many ways are there to come to believe that p?—and more than a few of these are likely to be part of its constitutive causal role. Moreover, mental states will have different effects depending on the mental contexts in which they are tokened. For a marathon runner who believes that it is necessary to achieve her goal, pain will not cause avoidance behavior; for a recreational runner with no such belief, it likely will. Add to this the commonplace that beliefs and desires have behavioral effects only in concert with other beliefs and desires. Then there are inhibitory effects. If C-fiber firing did not inhibit concentration, could it realize pain? And, finally, there’s the subtle point about the causal relations that must not obtain. A state playing the pain role cannot regularly cause pleasure, a belief that one is not in pain, or a desire for more of the same. I offer all of this in support of the claim that the functional roles of mental states will typically be far too complex for the happy coincidence of conditions depicted in Fig. 4 to obtain. It is almost certainly true that P_C will be a proper part of P_M in the cases that matter.

By way of summarizing and codifying what has been said, let me introduce the following principle, which is to replace the assumption that P_M = P_C.

Parthood: If some functionally realized mental event M (together, perhaps, with certain background conditions B) is causally sufficient for some physical event P* at t, if P* is among M’s characteristic effects, and if P* has a sufficient physical cause at t, then a proper part of M’s total physical realizer P_M is (together with B) causally sufficient for P* at t.

By characteristic effects, I do not mean only those effects that are constitutive of M’s causal role. I mean those effects that are typically associated with M qua M. These would include M’s various behavioral effects, its effects on other mental states, and the like. I suppose there are physical events that are caused only by the whole of P_M—through its gravitational effects on neighboring bodies, for example—but these are not among M’s characteristic effects, and Parthood is not intended to cover them. Nor is Parthood intended to apply to mental states that are not functionally realized. Davidson’s anomalous monism is (at least officially) in the clear. I’ve also made explicit that parthood applies only if P* has a physical sufficient cause, for otherwise neither M’s core nor its total realizer would be candidates. The reference to background conditions B is meant to cover the likely objection that mental events are only parts of causally sufficient conditions. It doesn’t matter, though, how extensive B is imagined to be. If M causes P* only together with B, then Parthood entails that a proper part of M’s supervenience base causes P* together with B—whatever else B may or may not include. Note as well that Parthood follows Kim in conceiving of causation in terms of causal sufficiency. Much more on this shortly. And note, finally, that Parthood does not rule out M or P_M as causes of P*. For that we require another principle.³⁰

It may perhaps be objected that Parthood is insufficiently justified. After all, I’ve offered nothing like a knockdown argument in support of it—only various considerations that make it likely to be true. But my modest goal in this paper is, as the title suggests, to strengthen the exclusion argument. If (2.5) is likely false and Parthood is likely true, replacing the former with the latter yields a stronger argument. And, at the risk of engaging in vulgar burden-shifting, I would argue that the onus is on functionalists to defend (2.5) rather than on their opponents to refute it conclusively. It’s difficult to see how (2.5) could be true or Parthood false when we consider the complex functional roles mental states are supposed to occupy.

4 A nonredundancy principle

Substituting Parthood for (2.5) in the exclusion argument replaces an unjustified assumption with a premise possessing strong prima facie support. It also has the benefit of allowing us to replace the controversial principle of Exclusion with something far more modest.

Nonredundancy: If some event C is causally sufficient for some event E at time t, no proper part of C (or C’s physical supervenience base) is causally sufficient for E at t.

Note, first, that Nonredundancy is logically weaker than Exclusion, which, recall, states that an event cannot have two or more distinct sufficient causes at a given time. Exclusion entails Nonredundancy because events and their proper parts are distinct. But Nonredundancy does not entail Exclusion because the former allows for two or more sufficient causes of an event if none is a proper part of any other. In particular, Nonredundancy allows for both a supervening and a subvening event to be causally sufficient for an effect—for both M and its physical supervenience base P_M to be causally sufficient for P*. This, in fact, is the strongest point in its favor. We saw above that Kim’s opponents tend to view Exclusion with suspicion and argue, for one reason or another, that M and P_M are not in causal competition. Nonredundancy thus gives these defenders of NRP everything they wanted! And yet, as we’ll see, it’s not enough to save the theory.

Nor is Nonredundancy simply ad hoc. It lies at the foundation of any view that, like Kim’s, takes causation to be ‘grounded’ in relations of nomological sufficiency. In adopting this framework, Kim follows in a long line of philosophers that includes Hume (1777), Mill (1843), and Mackie (1973). For present purposes, I will assume that some version of such an account is defensible—perhaps one developed along the lines of Strevens (2007), Wright (2013) or Paul and Hall (2013).³¹ I won’t need to choose from among the alternatives because while they differ in their details, they are all committed to something like—indeed, something stronger than—Nonredundancy.

To see why Nonredundancy is required, observe that causation cannot simply be identified with nomological sufficiency. Mr. Jones takes birth control pills and fails to get pregnant.³² His ingestion of the pills is a nomologically sufficient condition for his failure to get pregnant but is not its cause. More generally, whenever C is nomologically sufficient for E, C together with some causally irrelevant factor X will continue to be nomologically sufficient for E. Identifying causal with nomological sufficiency thus forces us to count indefinitely many noncauses as causes. It is for this reason that Mackie (1973, p. 62) defines causes in terms of minimal sufficient conditions. INUS conditions, recall, are insufficient but nonredundant parts of unnecessary but sufficient conditions. Wright (2013) similarly defines (partial) causes as necessary elements of sufficient sets, while Paul and Hall (2013) follow Mackie in proposing that causation be defined in terms of minimal nomological sufficiency.

Nonredundancy follows from such minimal sufficiency accounts. If a cause is minimally sufficient for its effect, no proper part of is. But the converse is not true. Nonredundancy does not entail minimal sufficiency. The former says that no part of a sufficient cause is a sufficient cause; the latter says this and makes additional claims about events that are not related as parts and wholes. Here again the weakness of Nonredundancy is a virtue, for the stronger thesis embroils minimal sufficiency accounts in a number of familiar problems involving preemption and joint causation.³³Nonredundancy avoids these problems by remaining agnostic on the more ambitious claim.³⁴Nonredundancy is also compatible with the intuitive account of overdetermination given by minimal sufficiency accounts.³⁵ (I leave the details here to footnotes). Nonredundancy thus fares no worse and often fares better than the more robust principle that causes are minimally sufficient conditions.

It’s important that Nonredundancy include the parenthetical clause about supervenience bases. Parthood tells us that only a proper part of M’s supervenience base is causally sufficient for P*. For this principle to come into conflict with Nonredundancy, the latter must also be couched in terms of M’s supervenience base. We cannot simply assume that M has parts because its physical supervenience base does—perhaps simple states can supervene on complex ones. Nor can we assume that the parts of M’s supervenience base are ipso facto parts of M. Nonredundancy thus makes explicit its application to the parts of M’s supervenience base: if M is causally sufficient for P*, neither its parts nor the parts of its supervenience base are causally sufficient for P*.³⁶

It’s true that Mackie et al. make no mention of supervenience bases, but the extension to such cases is surely implicit. Moreover, it is easily justified. The intuition that both minimality conditions and Nonredundancy are trying to capture is that no part of a sufficient cause is superfluous or redundant. But if some portion of M’s supervenience base is causally redundant, M itself is as well. Suppose that a proper part of M’s supervenience base P_M is causally sufficient for P*. This proper part of P_M is, of course, insufficient for the realization of M, which means that it could have occurred even in the absence of M. But then P* would have occurred in the absence of M.³⁷ It follows that M is not included in the set of minimally sufficient conditions for P*, for a set that includes only a proper part of P_M is more minimal then a set that includes all of P_M and thus M itself. It also follows that M is redundant, for its addition, which results from the addition of the superfluous parts of P_M, is unnecessary for the bringing about of P*. Thus if we are to remain faithful to the intuitions that motivate Nonredundancy in the first place, it cannot be that both M and a proper part of its supervenience base are causally sufficient for P*.

Of course, much of what makes Nonredundancy so compelling is that it says so little. It doesn’t propose to define causation or causal sufficiency. It doesn’t even rule out obvious ways in which nomologically sufficient conditions can fail to be causes. But this is as it should be. There’s no reason to invoke more ambitious principles when such a modest principle will do.

5 The functional exclusion argument

With Nonredundancy in hand, we are now able to introduce the functional exclusion argument. This argument begins, like the originals, with the assumption of Physicalism, Supervenience, and Efficacy, from which it follows that some mental event causes some physical event.

(1′′) M is causally sufficient for P* at t.

Closure tells us that P* must have a sufficient physical cause at t. We’re not assuming Exclusion, so we cannot conclude that there is only one such cause. Perhaps there are many. But we are assuming that M is functionally realized and that P* is among its characteristic effects. Parthood thus allows us to infer that there is a sufficient physical cause of P* that includes only a proper part of P_M.

(2′′) A proper part of P_M is causally sufficient for P* at t.

But Nonredundancy tells us that it cannot be the case that both M and a proper part of P_M are causally sufficient for P*. From which it follows that

(3′′) M is not causally sufficient for P* at t.

And we have arrived at our contradiction.

There are a few things to note. First, although Distinctness does not make an explicit appearance in this derivation, it is, of course, implicit in the appeal to Parthood. Parthood applies only to functionally realized states, and functional states are distinct from their physical realizers. Second, it doesn’t matter whether we read (1′′) as saying that M is itself causally sufficient for P* or simply part of a causally sufficient condition. Nor does it matter which way we read (2′′) and (3′′)—as long, that is, as we read all three the same way. Even if M is only part of a sufficient cause of P*, Parthood tells us that only part of that part is nonredundant. Excluding the irrelevant bits leaves us with a proper part of the original cause that is also, ex hypothesi, causally sufficient for P*. From this together with Nonredundancy, we can conclude that M is not, in fact, part of a sufficient cause of P*. Finally, and most importantly, the moral of the argument is not that Efficacy is false, just as this is not the moral of Kim’s original exclusion argument. Consistency can be restored by rejecting Efficacy; it can also be restored by rejecting Parthood. But this, I have argued, is tantamount to rejecting role functionalism.³⁸

6 The functional exclusion argument and dependence-based accounts of causation

Some authors have argued that Kim’s exclusion argument operates with a flawed conception of causation—“it mistakes causal sufficiency for causation,” as Menzies (2013, p. 71) puts it—and that the threat to NRP disappears if we instead conceive of causation in terms of counterfactual dependence. In this section, I’ll show that this objection has purchase only against versions of the exclusion argument that presuppose (2.5). When Parthood is instead assumed, dependence-based conceptions of causation actually entail that functionally realized mental properties are causally irrelevant.

Loewer (2002, 2007), appealing to Lewis (1973a, 1986), argues that counterfactual dependence is at least a sufficient condition for causation. He argues, that is, that C is a cause of E whenever (i) C and E both occur, (ii) C and E are distinct events, and (iii) E wouldn’t have occurred if C hadn’t occurred. It follows that both a mental event M and a physical event P cause the same physical event P* whenever both M and P stand in the aforementioned relation to P*, and this is exactly what we would expect whenever M metaphysically supervenes on P. In such cases, P* wouldn’t occur if P hadn’t occurred, for we are assuming that P is the physical cause of P*. But it’s a necessary truth that P doesn’t occur if M doesn’t occur, given that M supervenes on P. It follows (given a plausible assumption³⁹) that P* wouldn’t occur if M hadn’t occurred and hence that M is a cause of P*. Thus, according to Loewer, adopting a Lewisian account of causation dispels the exclusion argument and preserves the causal efficacy of the mental. Similar arguments can be found in Crane (2001) and List and Menzies (2009).

It’s debatable whether this strategy works in cases where M supervenes on P. Kim (2009, p. 44), for example, has argued that counterfactual dependence is too anemic a relation to ground our sense of agency and that only productive causation will do. But it is also clear that this strategy will not work whenever Parthood is assumed. Consider again the neuron diagram in Fig. 3. Here we see that (i) P and P* both occur (or, in the case of P*, will occur), (ii) P and P* are distinct events, and (iii) P* wouldn’t have occurred if P hadn’t occurred. P thus causes P* according to Lewis’s framework. But we cannot conclude that M causes P*, for P* does not depend on the occurrence of M. If the path from A to P had been eliminated just as the nervous impulse reached P, a portion of M’s total realizer would have been missing and M would not have occurred. But P* still would have. And, of course, there is nothing special about this path. The argument can be run for any of the neurons or any of the pathways not involved in the stimulation of P*. More generally, when the total realizer of M includes elements that are redundant with respect to the occurrence of P*—as it surely will whenever Parthood is assumed—P* will not depend on the occurrence of M.

One might object at this point. We assessed the dependence of P* on M by considering cases in which it was the superfluous bits of M’s supervenience base that were excised from history, but why not consider the case in which P itself is excised? If P were eliminated, M would be eliminated, for P is part of M’s total realizer. But if P were eliminated, P* wouldn’t have occurred. So here we have a case in which eliminating M also eliminates P*. Doesn’t that show that P* depends upon M after all?

It does not. Note, first, that the objector’s strategy allows us to count obvious noncauses as causes.⁴⁰ According to the objection, P* depends upon M because (i) P* depends upon P (and background conditions), and (ii) M contains P as a conjunct. But let M* be the conjunction of M and some arbitrary event occurring in a distant region of the universe. The same strategy would show that P* depends upon M*, or that it depends upon M**—the conjunction of P and two such events. And so on. These arbitrary events are irrelevant to the occurrence of P*, but so are the superfluous bits of P_M. The dependence of P* on M is therefore no less spurious than its dependence on M* or M**.

The objector’s strategy not only fails to establish that the right sort of relation exists between P* and M. It reveals, more importantly, what M has in common with M*, M**, and their dubious ilk. M, like M* and M**, contains (i) a core element upon which P* actually depends and (ii) a superfluous or redundant element the presence of which makes no difference to its purported effect. It’s the inclusion of the latter that renders these dubious dependency relations problematic—egregiously so in the case of M* and M**. Dependence-based accounts of causation thus require a principle akin to Nonredundancy in order to rule out these spurious dependencies, but because of their common underlying structure, any principle that rules out M* and M** as causes of P* will rule out M as well. Thus far I’ve been arguing that M fails to meet a sufficient condition for causal relevance—that P* does not depend on M. I’ll now argue that by including redundancies it fails to meet a necessary condition as well.⁴¹

Consider a simple account first. To avoid counting obvious epiphenomena as causes,⁴² Loewer (2002, p. 660) is forced to amend the simple counterfactual account to say that M is a cause of P* only if M is not preempted by some other cause, where M is preempted by P (with respect to P*) whenever (i) –P > –P* and (ii) (P&–M) > P*.⁴³ A brief glance at Fig. 3 will confirm that both counterfactuals are, in fact, true. More generally, both counterfactuals will be true whenever Parthood obtains. This is because Loewer’s condition has the effect of imposing a kind of nonredundancy constraint on possible causes, and Parthood entails that portions of M’s total realizer are redundant to its purported effects. If P* depends on P, and M is the conjunction of P and some set of events that have no influence on P*, M is preempted by P. M is therefore not cause of P* on Loewer’s own account.

Let us now consider Woodward’s more sophisticated account, with its more sophisticated method of excluding redundancies. Woodward’s framework deserves special attention for a variety of reasons. First, few authors have developed a dependence-based account of causation as carefully as Woodward (2003). And, certainly, few authors have developed a dependence-based response to the exclusion argument in as much detail as Woodward (2008, 2015, 2017). Moreover, Woodward’s framework is in many ways more attractive than Lewis’s or Loewer’s, for it does not carry the additional burden of attempting to reduce causation to something more fundamental. It would therefore strike a serious blow against the dependence-based strategy if functional properties were to fail Woodward’s test of causal relevance. We saw above that Woodward believes that the question at the heart of the exclusion argument—Is it M or P that causes P*?—cannot even be stated within an interventionist framework if we assume (2.5). But what does this framework have to say if we assume Parthood instead?

An interventionist account of causation holds, very roughly, that events of type X are causally related to events of type Y if and only if one can manipulate the latter by manipulating the former. Less roughly, X is a direct cause of Y (relative to some set of variables V) if and only if there is “a possible intervention on X that will change Y (or the probability distribution of Y) when all other variables in V besides X and Y are held fixed at some value by interventions” (Woodward 2003, p. 55). (Correlatively, there is a directed path from X to Y just in case X is connected to Y by a series of direct causes). Imagine that we want to determine whether it’s falling barometric pressure or falling barometer readings that cause bad weather. Our set of variables will be V = {B, R, W} for the barometric pressure, barometric reading, and bad weather, respectively. B and R can take any number of values, but we’ll assume that W takes only two: 1 when the weather is bad and 0 when it isn’t. An intervention I on B with respect to W (or on R with respect to W) sets the value of B (/R) in such a way that (i) all other directed paths to B (/R) are broken, (ii) I affects W (if at all) only via the directed path passing through B (/R), and (iii) any directed path to I affects W (if at all) only via the directed path passing through B (/R) (Woodward, p. 100). One may not, for example, set the value of R by changing the barometric pressure, for that would affect W via a path that does not pass through R. It follows from these definitions that falling barometric pressure causes bad weather and falling barometer readings do not. There are some (indeed, many) values of R for which interventions on B change W, but there are no values of B for which interventions on R change W.

Not all causes are direct causes, of course. The relation that matters in discussions of the exclusion argument is that of actual or token causation, which Woodward defines in terms of direct causation: X’s taking the value x is a token cause of Y’s taking the value y if and only if (i) the actual values of X and Y are x and y, respectively, (ii) there is a directed path from X to Y, and (iii) there is a possible intervention on X that will change the value of Y when all other direct causes of Y not on this path are held fixed at their actual values (2003, p. 77).⁴⁴

It should now be obvious why Woodward believes that the original exclusion argument presupposes a metaphysically impossible intervention. Suppose we wish to determine whether it’s M or M’s supervenience base P_M that causes P*. The appropriate set of variables would be V = {M, P_M, P*}, where each variable is binary, taking the value 0 when the corresponding event is absent and 1 when the corresponding event is present. We are assuming that M, P_M, and P* are instantiated. It thus follows that it’s M rather than P_M that causes P* with respect to V if and only if an intervention on M that changes its value from 1 to 0 changes the value of P* from 1 to 0 when P_M is held fixed at its actual value of 1. But M can’t change while P_M remains constant.⁴⁵

It should also be clear that Woodward’s response presupposes (2.5). Variable sets should include candidate causes, and Parthood and Nonredundancy together tell us that it is only a proper part of P_M rather than P_M as a whole that is causally sufficient for P*. The proper choice of variables therefore is not V = {M, P_M, P*} but rather V′ = {M, P_C, P*}, where P_C is the proper part of P_M that causally suffices for P*. And when we run these variables through Woodward’s test, we find that M fails to be causally relevant to P*. Consider again the situation depicted in Fig. 3. Note first that it is not metaphysically or even nomologically impossible to change the value of M while holding P_C fixed at its actual value. P_C includes no more than P, P*, and the connection between them, while M includes the whole network. Thus, if we eliminate A or C, the value of M changes from 1 to 0, but the value of P_C does not. Moreover, because P_C is causally sufficient for P*, the latter will take the value 1 whenever the former does. Here, then, we have an intervention on M that does not change the value of P* when P_C is held constant. There is, in fact, no intervention on M that will change the value of P* when P_C is held constant. It is thus P_C rather than M or even P_M that causes P*. And nothing in this argument depends on the actual configuration of the network in Fig. 3. What matters is only that Parthood obtains. We thus have a general argument for the causal irrelevance of functionally realized states.

7 Conclusion

The functional exclusion argument I’ve offered here is stronger than the original in some rather obvious ways. Parthood is more defensible than the dubious assumption that P_M = P_C, and Nonredundancy is more modest than Exclusion. But I would like to note in closing a subtler strength of the argument. Somewhat paradoxically, this strength concerns its narrower scope.

Many commentators have argued that Kim’s exclusion argument generalizes to all so-called higher-level properties. The properties countenanced by biology and geology—and perhaps even chemistry—appear to be merely supervenient upon those of physics. But it seems absurd to deny that earthquakes and acids are causally efficacious, in which case there must be something wrong with the exclusion argument. Or, if there’s nothing wrong with the argument, what it shows is of little consequence. If our thoughts and desires are merely as potent as natural disasters, it’s not obvious that this is something we should fret about.

The functional exclusion argument does not generalize—not, at least, to the extent of the original. This is because it applies only to cases in which we can draw a distinction between core and total realizers, and some higher-level properties do not admit this distinction. Consider the case of (intrinsic) dispositions. A glass vase is fragile, but so is a ceramic vase, and each case of fragility is realized by a different microphysical structure. Fragility is thus multiply realizable and irreducible. But what is the core realizer of the glass vase’s fragility? What is its total realizer? Don’t say that the core realizer is the vase and the total realizer is the vase in the context of being struck, for the vase maintains its fragile disposition even outside of such contexts. Further consideration suggests that the core/total distinction simply fails to apply. If so, then at least some higher-level properties are unthreatened by the functional exclusion argument, which is thus not so much an objection to a metaphysics of ‘layers’ or ‘levels’ as it is to a functionalist theory of mind.⁴⁶

It follows that there is an important distinction to be made between different sorts of higher-level properties. There’s a tendency among philosophers to treat all such properties as being cut from the same cloth. Indeed, it’s not unusual to find the terms ‘functional’ and ‘dispositional’ being used interchangeably. But the arguments of this paper suggest that this is a mistake. Some higher-level properties can be analyzed in terms of the core/total realizer distinction, some can’t. Some higher-level properties consist of a causally relevant central element and a causally irrelevant surrounding element, some don’t. Some higher-level properties are threatened by the functional exclusion argument, some aren’t. It doesn’t matter how we mark this distinction, but there’s some historical precedent for calling properties of the former type ‘functional’ and those of the latter type ‘dispositional’. The functional exclusion argument applies only to functional properties thusly conceived.⁴⁷

But it’s functional properties thusly conceived that functionalism identifies with mental properties, and it’s functionalism that most of NRP’s defenders want to defend. For that enterprise, the assumption that P_M = P_C is clearly off limits. By adopting this assumption, functionalism’s defenders have unwittingly waged war against a straw man—even if that straw man sometimes bears an uncanny resemblance to Jaegwon Kim.