5 The Causal Efficacy of High-Level Properties

1. Introduction

I have suggested that cognitive properties are emergent properties. According to the conception of emergence developed in Chapter 4, an emergent property is reducible in principle provided that we discover the law of composition responsible for its existence. Moreover, I have suggested a conception of reduction according to which a reducible property preserves an autonomous existence with respect to its reduction base by showing that reduction is not equivalent to the identification of the reduced property with the reducing property. But the most difficult part remains to be done. I still have the task of justifying the reality of mental properties by the causal criterion of reality: as long as I have not shown that such emergent properties can be causally efficacious, my position remains vulnerable to an epiphenomenal interpretation. To the extent that I fail to justify the ability of mental properties to make a causal difference to events in the world, their reality remains doubtful. If mental events had no causal effects of their own, then my mind would be no more than an epiphenomenon, like a shadow accompanying a real causal process taking place at the physical level.

In this chapter, I sketch a conception of causality that allows the emerging macroscopic properties of complex systems to be causally efficacious. Their efficacy complements — in a sense yet to be specified — the efficacy of the physical properties of their parts, which seem to monopolize causal power. Jaegwon Kim (1998) argues that there are only two coherent ways of doing justice to the intuition that the mind exerts an influence on its physical environment through the body’s movements. Either one accepts one of the forms of dualism according to which certain causes are non-physical, for example persons,¹ or one accepts materialism, in its reductionist or eliminativist form, according to which only physical properties possess real and proper (non-derivative) causal powers and in which mental properties are understood to have, at most, derivative causal powers.²

According to reductionist materialism, only properties that can be reduced to physical properties are real and causally efficacious. As far as mental properties are concerned, the cognitive neurosciences of the future will show whether they can be reduced to neurophysiological properties and processes or not.³ This makes two materialist positions conceivable: reductionist materialism (also known as “type physicalism”) is the appropriate position if mental properties turn out to be reducible, and eliminativist materialism (eliminativism) is the appropriate position if they turn out to be irreducible. Now, according to Kim, a materialist (or physicalist⁴) who wishes to avoid dualism cannot maintain that mental properties are both irreducible and causally efficacious. Accordingly, he proposes that anti-reductionist materialism⁵ is an unstable position that cannot be developed coherently without leading to the adoption of dualism, reductionist materialism, or eliminativist materialism.⁶

One of the aims of this book is to show that anti-reductionist materialism is not the only possible position that avoids the radical positions of dualism and eliminativism. We will see that it is not necessary to find an argument for the impossibility — in principle — of reducing mental properties to physical properties in order to avoid reductionism (understood as identification) or eliminativism. It is up to science, not philosophy, to discover which mental properties are locally or globally reducible to neurophysiological properties, possibly through the construction of new psychological or neurophysiological concepts. But questions concerning their reducibility are independent from questions concerning their causal efficacy. The causal efficacy of a property is not threatened by a reduction (as dualists and some reductionists claim), and a reduction is not necessary to ground it (as some physicalists claim). Of course, the reducibility of a property plays an important role in our understanding of its causal efficacy. The reduction of a cognitive capacity, for example through the discovery of a mechanism underlying its exercise,⁷ is the best way of explaining why and how this capacity exerts a causal influence. However, ontologically, its efficacy depends not on its reducibility but on the laws of nature that link it to other properties at the same level: that is, properties of the entire cognitive system and not of its parts.

In what follows, first I will sketch a conceptual framework within which the question of the causal efficacy of mental properties can be posed without prejudging it in the direction of reductionism, eliminativism, or dualism. Next I will show that this framework allows us to defend the possibility of the causal efficacy of the mind against two important objections: first, Kim’s objection that mental properties can have only derivative efficacy; second, Lowe’s objection that such efficacy can be acknowledged only within a dualistic framework.

I will ignore the more specific problem of the intentionality of mental states and properties. At least some mental states possess a content constitutive of their identities. According to the externalist conception of content, the identity of such an intentional mental state is determined by things too remote in space and time from the person in the mental state in question to be able to contribute causally to the efficacy of the mental state. Therefore, we are faced with the problem of understanding how mental states can be efficacious by virtue of their content. However, this problem can be approached independently⁸ of the problem analyzed here: is it conceivable and plausible that global properties of complex objects are causally efficacious so that eliminativism, reduction understood as identification, and dualism can all be avoided? My strategy for answering this question consists of putting the issue of the efficacy of the mind into a more general perspective. Part of the problem of mental causation is just as much a problem for the theory of macroscopic causation in general (Baker 1993, 79; 1998, 261). We can therefore hope to make progress in our understanding of the particular problem by justifying the possibility of macroscopic causation in general.

2. Causality, Causal Responsibility, and Causal Explanation

Let us take a statement expressing mental causation. The thought that street noise disturbs my concentration causes me to decide to close the window. This is a case of mental causation in the strict sense: that is, a situation in which both cause (the thought) and effect (the decision) are mental events. But common sense also naturally conceives of psychophysical causation, in which a mental event causes a physical event, for example if my decision (a mental event) causes my act of closing the window (a behaviour and therefore a physical event). Of course, mental events can also be effects of physical causes, as happens in perception: the (physical) noise in the street causes me to think that this noise disturbs me. Each of these mental events also has physical properties: thoughts and decisions always occur in humans, who are material beings.⁹ Moreover, the mental properties of these events are determined by underlying physical properties: if we intervene on the relevant regions of the brain, then the thought or decision can be altered or disappear. There is no doubt that the causal relationship between these events follows the laws of physics, which apply to them according to their physical properties. We can therefore wonder about the causal contributions of their mental properties: does the fact that it is a thought with a certain content (the noise in the street is disturbing my concentration) contribute causally to my decision to close the window? Does the fact that it is a decision with a certain intentional object (to close the window) contribute causally to my act of closing it? Two intuitively plausible principles appear to challenge the idea that mental properties make a causal difference. First, according to the principle of “causal closure,” every physical event, at every instant preceding it, has a complete and purely physical cause. In particular, my act of closing the window, as a physical event, has a complete physical cause at the moment that I decide to close the window. Second, an event’s mental properties differ from its physical properties. It seems, then, that we are led to the conclusion that the mental properties of the event, in this case the property of being a decision, are causally inert: if my act of closing the window is the result of an uninterrupted and purely physical causal chain, then there seems to be no room left for contributions of mental properties. This line of reasoning, developed with great clarity by Kim (1998), assumes that causal overdetermination is exceptional. There can be rare events caused by two independent causal processes that converge. This can be the case, for example, when someone sentenced to death is shot by two marksmen in such a way that their bullets hit her heart at exactly the same time and in the same place. In this case, her death is overdetermined: each bullet is sufficient for her death, such that the removal of one would not alter the result. The argument for the causal inertia of mental properties depends on the premise that such overdetermination is exceptional. In other words, it is not plausible to assume that, systematically, each effect of every mental cause is overdetermined, in the sense that it has both a physical cause and a mental cause. Kim expresses the thesis that overdetermination is exceptional in his “principle of causal-explanatory exclusion.” There can be only one complete causal explanation of any given phenomenon. This line of reasoning represents a formidable challenge to the conviction of common sense that the mental nature of our thoughts and decisions has a causal impact on the world.

In this chapter, we will see how it is possible to defend this intuition against the thesis that mental properties are epiphenomenal: in other words, that they make no causal difference of their own to the course of events. I will reply to the challenge that all causes are physical and that, more specifically in the case of the physical consequences of our actions, all of their causes are neurophysiological.

Before going into the details of this debate, it is necessary to respond to an important objection to my way of posing the question of whether macroscopic properties have causal efficacy beyond that of the underlying physical microscopic properties. According to some authors, the question of whether certain properties of a cause are efficacious with respect to the properties of the effect is ill posed. In particular, Donald Davidson (1980, 1993) argues that properties (and the predicates that refer to them) belong to the conceptual register of explanation rather than that of causality. According to Davidson (1995), it is true a priori that, for any pair of causally related events, there is a law of which this causal relationship is an instance; therefore, it is always possible in principle to explain why one caused the other. However, Davidson rejects the very question that raises the problem of mental causation. How can a mental event cause something like bodily motion by virtue of its mental properties given that the same event also seems to cause the same motion by virtue of its physical properties?¹⁰ Having adopted an ontological framework that gives no place to properties in the analysis of why one event causes another, Davidson is challenged by critics who argue that he “holds doctrines which commit him to denying that mental events cause physical events in virtue of falling under mental types. On his view, they claim, the mental qua mental is causally inert” (McLaughlin 1993, 28). Davidson’s metaphysics of causation, which admits only particular events and their linguistic descriptions, does not make sense of the question of what makes a given event cause another event by virtue of specific properties. From his point of view, this question betrays confusion between a demand for information about a causal relationship and a demand for a causal explanation. The first is an extensional relationship between particular events. The second is a relationship between statements: one statement explains another, the explanandum, if it is possible to construct a deductive argument whose conclusion is the explanandum, and whose premises, which together constitute the explanans, contain the first statement together with a certain number of nomological statements.

However, the question of whether certain properties are causally efficacious is important. A satisfactory theory of causality must provide a conceptual framework for identifying what it is about the cause that is responsible for its effect having certain properties. If a red billiard ball is the cause of the fact that a white billiard ball starts moving in a given direction with a given speed, then it causes this precise movement by virtue of the energy and momentum that it carries before the impact but not by virtue of its colour. This would be a fact even if there were no science, no language to express it, or any statement of law or explanation that referred to it. Some properties of the causing event (e.g., its momentum) objectively modify the relevant properties of the effect, whereas others (e.g., its colour) do not. Therefore, a complete theory of causation must acknowledge the objective role that the former play in determining causal interactions.

We could simply ignore Davidson’s nominalist scruples, which make Davidson prefer the language of predicates to that of properties and translate his position into realist terms. Instead of simply saying that events satisfy predicates, we would say that they have properties. Davidson (1993) himself uses this realist language; however, he would not accept the next step in the following reasoning. Instead of simply admitting that referring to F₁ (the fact that the red ball hit the white ball with momentum M) causally explains F₂ (the fact that the white ball, after being hit by the red ball, has momentum M), we must acknowledge that, if this explanatory relationship is correct, then its truth has an objective basis, in other words a “truth maker.” We can express this by saying that F₁ is causally responsible for¹¹ F₂. In this way, we can account for the fact that the distinction between F₁ (the fact that the red ball hit the white ball with momentum M) and F₃ (the fact that the red ball was red when it hit the white ball) reflects an objective difference in the causal influence that the constitutive properties of these facts have on the motion of the white ball. This difference does not depend on an explanatory difference; F₁ but not F₃ would be causally responsible for the effect even if there was no one to seek or offer explanations.¹²

However, this argument is still not sufficient to overcome the specific obstacle that prevents, according to Davidson, mental properties from being causally efficacious. My reasoning has no force insofar as mental properties are concerned because they cannot even figure in causal explanations. Therefore, it cannot be said, even within a realist reformulation of Davidson’s position, that these properties can participate in facts causally responsible for anything. The reason is that there are no strict laws involving mental properties, whereas causation presupposes the existence of strict laws. The answer that seems to be plausible to me is that it is not necessary for a property to fall under a strict law in order to be mentioned in a causal explanation and thus to be causally efficacious. If it is true that most laws are not strict,¹³ then Davidson’s version of the nomological theory of causation has the consequence that most physical properties are not efficacious either. This is a reductio of the condition that the relevant laws must be strict. Some authors maintain that the existence of a non-strict law (or ceteris paribus law) that applies to a given situation is sufficient for the existence of a causal relationship between the events to which the law applies.¹⁴ However, if the condition of the existence of a strict law is too strong, then the mere existence of a non-strict law is too weak to account for causal efficacy.¹⁵

The following situation shows why the existence of a law expressing a dependence between certain properties of two events is not sufficient for these events to be linked as cause and effect. Let us assume that A is a radio station that broadcasts a certain program so that the waves carrying the signal travel in all directions at the speed of light. Let us then take two locations, B and C, situated at equal distances but in opposite directions from A. Let us call (B, t) and (C, t) the events of the arrival of the signal at B and C, at time t. There is a locally valid law that links (B, t) and (C, t); in fact, the propagation of electromagnetic waves in a vacuum follows from Maxwell’s equations.¹⁶ Therefore, according to the realistic theories of mental causation proposed by Fodor (1989), McLaughlin (1989, 1993), and Pietroski (1994), (B, t) and (C, t) should be causally related as cause and effect, which they are not. Two events (x, t) and (x’, t’) are “spatially separated,” in the sense of the theory of special relativity, if they cannot be linked by a light signal. (B, t) and (C, t) are spatially separated in this sense and cannot therefore be linked by any causal process. Rather, (B, t) and (C, t) are effects of a common cause (A, t’) taking place at A a little earlier than t.

The existence of a counterfactual dependence, a condition proposed by LePore and Loewer (1987, 1989), is too weak to guarantee the existence of a causal relationship, for the same reason. The events of the arrival of the signal at B and C depend counterfactually on each other. In the imagined situation (i.e., supposing that there are no other sources of signals or screens), if no signal arrived at B at t, then no signal arrived at C at t; if none arrived at C, then none arrived at B. That counterfactual dependence is not sufficient to establish that one of these effects is the cause of the other.¹⁷ Indeed, here it is clear that this is not the case.¹⁸

I propose to draw the following lesson from the case of the radio transmitter. Causality, nomological, and counterfactual dependence are different concepts without being independent. In particular, neither the causal relationship nor the relationship of causal responsibility can be reduced directly to nomological or counterfactual dependence or to increased probability.¹⁹ Neither nomological dependence, nor counterfactual dependence, nor an increase in probability between two types of events (or between properties of events) makes it possible to deduce conclusively that a particular event belonging to one type is the cause of an event belonging to the other type. All of these relationships exist between types, or sets, or properties of events. But we cannot draw any conclusion about the existence of a causal relationship between two particular events from the existence of one of these relationships at the level of types.²⁰ Therefore, what makes c cause e must be a fact about the particular events c and e and the relationship between them as particulars. This fact is added to relationships existing between their properties, such as nomological or counterfactual dependence or probability increase. This is not the place to defend the thesis that the transmission of an amount of a conserved quantity between two events is a necessary and sufficient condition for the existence of a causal relationship.²¹

Let us now return to the question of what makes certain properties of an event efficacious in a given causal relationship while other properties play no such role. The concept of causal responsibility (CR) can serve as a framework for asking this question. It can be analyzed as follows.

(CR) The fact that c is F is causally responsible for the fact that e is G, if and only if c is a cause of e (at the level of particular events, by virtue of transmission of conserved quantities), c exemplifies F, e exemplifies G, and there is a law (which generally is not strict) according to which instances of F tend to produce instances of G.

In (CR),²² the expression “c is the cause of e” designates a causal relationship between particular events based on the transmission of a certain amount of a conserved quantity. An event is a particular whose identity conditions are given by the limits of the spatiotemporal zone that it occupies. The statement “the fact that c is F is causally responsible for the fact that e is G” implies both that c is the cause of e and that the property F of c is causally efficacious in the production of an event that has property G.²³

To guarantee that causal responsibility, which depends on properties, is as local as the causal relationship between events, we need to conceive of the properties F and G that constitute the facts that c is F and e is G as instances of properties or “tropes” and not as universals.²⁴ Only a spatiotemporal entity, such as a trope, can be locally efficacious, whereas universals are not localized in space and time.²⁵ According to an Aristotelian conception of universals, we could say that a universal is localized wherever its instances are localized. But then what is efficacious in a given situation is an instance, not the universal itself, the totality of all instances. This is just another way of saying that the efficient entity is a trope. However, this idea itself is not enough to solve the problem since the crucial question now becomes that of identifying causally efficacious tropes. Under which conditions is one trope identical to another or distinct from another? Our central question thus becomes this: if two tropes of the same event are not identical, then which one was efficacious in relation to a given effect? Robb (1997) proposes the following solution: if I decide to close the window, at the moment of the decision I have a mental trope identical to a physical trope. This decision trope is mental insofar as it belongs to a mental type of decision to close the window, but it is also physical insofar as it belongs to a certain physical and more specifically neurophysiological type. However, it is causally efficacious not by virtue of the fact that it belongs to one type or the other: it owes its efficacy neither to the fact that it belongs to the mental type nor to the fact that it belongs to the physical type.

The identity conditions for Robb’s tropes are less coarse than the identity conditions for Davidsonian events.²⁶ But they are similar with respect to their “coarseness” in a sense that I will specify in a moment. For this reason, Robb’s solution is ultimately very similar to Davidson’s. Robb’s tropes are “coarse” in the sense that they can be both mental and physical, just as Davidsonian events. Mental tropes are causally efficacious because they are identical to physical tropes, assumed to be causally efficacious. In this statement, we need only substitute the word event for all occurrences of the word trope to retrieve the Davidsonian solution to the problem of mental causation. Tropes sufficiently coarse to be both mental and physical prevent us — just as much as Davidsonian events — from asking the question of what it is about these events/tropes that is causally efficacious in bringing about a given effect.²⁷ Robb’s tropes, just like Davidson’s events, lack the internal structure that would make it possible to distinguish between different aspects that could bear different causal responsibility. Davidson’s anomalous monism is confronted with the objection that it renders the mental epiphenomenal (see Sosa 1984; McLaughlin 1993). It prevents us from giving an affirmative answer to the question of whether a given event, for example the decision to open the window (which caused me to open the window), caused the latter act by virtue of its mental properties (the fact that it was a decision) or whether its efficacy is due exclusively instead to its neurophysiological or physical properties. Robb’s theory, which uses the concept of trope, similarly precludes the question of whether the mental and physical trope caused my act of opening the window by virtue of its mental aspect or by virtue of its neurophysiological aspect.

We can hope to answer questions of this kind only on the basis of a criterion of the identity of properties (or tropes).²⁸ The following nomological criterion links the identity of a property (trope) to its nomological relations with other properties (tropes of other types).

(Nomological criterion of property identity) Property P is identical to property Q if and only if, for all properties R, P is in a nomic relation N with respect to R if and only if Q is in the same nomic relation N with respect to R.²⁹

Formally:

$\forall P\forall \underline{Q}\left.P=Q\leftrightarrow \forall R\forall N\left.N\left(P,R\right)\leftrightarrow N\left(Q,R\right)\right]\right\}$

P and Q are identical properties if and only if they share all of their nomic dependency relationships with respect to other properties; moreover, since causal responsibility is determined by nomic dependence, identical properties enter into the same causal responsibility relationships.

Let us consider once again the example of the red billiard ball that causes the movement of a white billiard ball, with a given speed and in a given direction. There are laws linking energy and momentum, and therefore the speeds of these two balls, but there are no laws linking their colours and their speeds. Consequently, the colour trope of the red billiard ball is different from its speed trope, and what is causally responsible for the speed of the white ball after the impact is the speed and not the colour of the red ball.³⁰

According to this criterion, mental and physical properties are different: they are involved in different laws. In this framework, the crucial question becomes whether there are psychological laws between different mental properties. If such laws exist, then mental properties are causally efficacious. This is an empirical question.³¹

I propose to construe a mental property M as a first-order macroscopic property of an individual s. M is determined by a complex physical property P. P is constituted by the logical conjunction of the local properties of the bodily parts of s as well as the relationships among these parts. P determines M in a nomological but non-causal way. The properties of the parts of s and the laws governing the interactions among these properties together give rise to a law of composition. Each individual s that has parts s₁ . . . s_n with properties P₁₁ . . . P_nm possesses, by virtue of the laws of nature, global properties G₁ . . . G_r, some of which are mental. Similarly, the microphysical state of a gas, conceived as the conjunction of the properties of position and momentum of the molecules that compose it, naturally determines, by virtue of a law of composition, the overall temperature of the gas.

Non-causal nomological determination is distinct from two other recent conceptions of the relationship between a person’s mental properties and the underlying physical properties of her body.

First, according to the functionalist view, mental properties are functional properties. Attributing to someone the property of feeling pain is equivalent to attributing to her a property that satisfies a functional condition, in other words a second-order property expressed by means of an existential quantification. In the simplest case, this is a condition expressed in terms of sensory causes and behavioural effects. To say that x feels pain is equivalent to saying that x has one or another of a set of neurophysiological properties that are caused by bodily harm and that in turn cause (among other things) behaviour that leads to an escape from the source of the harm. The cause of this escape behaviour, which at the same time is the effect of the external stimulus, is a first-order neurophysiological property said to “realize” the mental property in that individual. However, according to this view, the mental property itself is causally inefficacious; in other words, it is epiphenomenal. Provided that we accept the causal criterion of reality,³² this has the consequence of denying any reality to the mental property beyond the physical property that realizes it. What is properly mental is merely a second-order predicate constructed by an existential quantification over first-order physical properties; only the latter are real and causally efficacious.³³

Second, according to another proposal (Yablo 1992), the relationship between mental properties and underlying physical properties can be likened to the relationship between a determinable property and one of the properties that determines it. The mental property would be to the underlying physical property what red is to scarlet or what temperature is to 0º Celsius. Indeed, it seems to be plausible that the mental property of feeling pain is an abstract and determinable property, whereas each biological species capable of feeling pain, and perhaps even each individual capable of feeling pain, has its own specific painful sensations.

However, the relationship between determinable and determinate is inappropriate as a model for the relationship between mental properties and underlying physical properties. A first reason is that the different physical properties on which the existence of a given cognitive property depends in different species and individuals are not ordered in a series, as is the case with the set of determinates of a determinable: the set of temperatures, for example, is ordered in one dimension. It is not clear how this model could be extended beyond the domain of quantitative properties, where more or less fine discriminations give a clear meaning to the distinction between properties more or less abstract and therefore more or less determinable.³⁴

The second and most important reason for considering that the model of the distinction between determinable and determinate is not appropriate to the analysis of the relationship between mental properties and physical properties is that, to be in a relationship of determinable to determinate, two properties must be exemplified by the same object and be of the same logical type. This is the case if both the determinable and the determinate are mental properties, for example the general property of experiencing pain and the human property of experiencing pain. The whole individual possesses both properties. Conversely, the physical property underlying the mental property of experiencing pain, although it too can be technically attributed to the whole individual, is in reality the conjunction of properties of parts of the individual (of nerve cells, neural circuits, or brain regions) as well as of relationships among these parts. In Kim’s terminology, this underlying physical property is a “micro-based property” (see Chapter 4.7) that “belongs to a whole in virtue of facts about its parts” (1988b, 142; 1993b, 124) and by virtue of logic alone: “P is a micro-based property just in case P is the property of having proper parts, a₁, a₂, . . ., a_n, such that P₁(a₁), P₂(a₂), . . ., P_n(a_n), and R(a₁, . . ., a_n)” (1997b, 292). The simplest case of a micro-based property is the property of “being made up of two parts x and y such that x is F and y is G and x is related by R to y” (Kim 1988b, 142; 1993b, 124). Now the mental property of feeling pain is determined by the properties of the parts of the organism and their interactions, by virtue of a law of nature, and not just by virtue of logic. For this reason, it is not a micro-based property in Kim’s sense. Since the mental property is not micro-based, whereas the underlying physical property is, these two properties belong to fundamentally different kinds, making them incapable of being in the relationship of determinate to determinable.

This point can also be expressed as follows: the relationship between a determinable and a determinate is an internal relationship. According to a plausible conception,³⁵ a determinate is a complex property: it is designated by a complex predicate that has the form of a conjunction. When one of the terms of the conjunction is deleted, the resulting predicate designates a determinable property relative to the determinate designated by the original predicate. There is then an internal relationship of subordination between these properties: any object that has the determinate also has the determinable because the elimination of certain terms of the conjunction corresponds to a valid inference. Furthermore, if we know that an individual s possesses the determinate property, and if we know the conjunctive structure of this property, then we can infer a priori, for purely logical reasons, that s also possesses the determinable. This model is unsuitable for analyzing the relationship between mental properties and underlying physical properties insofar as inferring the former from the latter requires knowledge of the laws of nature. The predicates that designate the physical properties of the brain do not have a conjunctive structure such that part of the conjunction corresponds to a mental predicate. We cannot infer a priori, therefore, the possession of a mental property from the possession of a physical property.

3. Mental Causation and Downward Causation

The approach to the determination of macroscopic properties, based on laws of composition, that I have developed in this book solves two important problems. The first is what Kim calls the problem of causal exclusion. The second is to answer the question of whether the emergentist concept of downward causation is compatible with physicalism.

Let us start with the second problem. The view that macroscopic properties are determined by non-causal laws of composition provides the means for settling the debate between those who accept downward causation and those who take its possibility to be refuted by its incompatibility with well-entrenched metaphysical principles. Some emergentists³⁶ defend the thesis that a macroproperty can exert a constraint superposed³⁷ on the constraints exerted by the properties of the microscopic components; in this way, the macroproperty prevents the microproperties from determining the evolution of the system on their own. Their opponents³⁸ seek to show that downward causation is a myth incompatible with the metaphysical principle of “the causal closure of the physical domain” (Kim 2005, 15).

One of the targets of these critics is the concept of downward causation put forward by the neurophysiologist R.W. Sperry. He is an easy prey for defenders of microdeterminism and opponents of the possibility of downward causation because his position combines the fundamental thesis of the existence of macrodetermination with more controversial theses.

First, Sperry seems to take the structural form of a whole to be one of its components, in the same way as its matter. By attributing causal power to this form, he seems to return to a theory “in terms of components” as criticized by Broad (see Chapter 4). “I have repeatedly stressed the important causal role of the non-material space-time, pattern, or form factors and suggested that it is helpful to view any entity as . . . built of space-time components as well as of matter” (Sperry 1986, 266).

Second, the set of emergent properties to which Sperry attributes causal efficacy is vast and varied. It includes cases of macrophysical causality — “drops of water are carried along by a local eddy in a stream. . . . [T]he molecules and atoms of a wheel are carried along when it rolls down hill” (1969, 534) — informational causality — “computer software programs exert downward causal control over their electronic . . . correlates” (1986, 269) — biological causality — “the holistic properties of the organism have causal effects that determine the course and fate of its constituent cells and molecules” (1969, 533) — and psychological causality — “the subjective mental phenomena . . . influence and . . . govern the flow of nerve impulse traffic” (1969, 534). Even social properties can exert downward causal influence. Sperry describes them as “emergent forces of higher and higher levels that in our own biosphere include vital, mental, political, religious, and other social forces of civilisation” (1986, 269).

Critics of downward causation argue that it is incompatible with a number of principles whose truth is presupposed by scientific method. According to an argument put forward by Klee (1984), which I will consider in a moment, downward causation is incompatible with the universal truth of microdeterminism as well as with the principle that all causal relations are exercised through a mechanism (see also Craver and Bechtel 2007). Others, notably Kim, argue that downward causation is incompatible with the principle that there is only one independent complete causal explanation for a given event as well as with the principle of the causal closure of the physical domain. I will come back to this later. We will see that such arguments, even if they seem to be convincing with respect to Sperry’s position, do not refute the theory of macrocausation developed here. In general, these arguments fail because they neglect the crucial distinction between causal and non-causal determination.

3.1. Macrocausation without an Underlying Microscopic Mechanism

According to Klee, downward causal determination is intelligible only insofar as it is exercised through a mechanism. Advocates of downward causation could give us reasons to doubt that “micro-determinism” suffers no exceptions only “if a genuinely plausible mechanism of macro-determination could be provided” (Klee 1984, 61).³⁹ This reasoning suffers from the lack of a clear distinction between causal determination and non-causal simultaneous determination of the properties of an object by the properties of its parts. The conclusion that macrodetermination does not exist can be avoided by arguing that many types of causal determination are due to macroproperties that themselves are determined, in a non-causal way, by microproperties. Before returning to this model at greater length, I would like to offer, as an example of a causally efficacious macroproperty, the ability of hemoglobin molecules to bind oxygen (see Chapter 2.5; Rosenberg 1985, Chapter 4; Feltz 1995). It is the overall structure of the molecule that possesses this capacity. The overall structure is determined by non-causal laws of composition, grounded in the microscopic properties of the components of the hemoglobin molecule and their interactions. However, what is directly responsible causally for oxygen uptake is not the microscopic properties of the molecule but its overall structure. The function of hemoglobin is to carry oxygen from the lungs to the various tissues of the body. However, the property of the molecule that enables it to perform this function is a systemic property of the molecule that is “macroscopic” in a relative sense: it belongs to the whole molecule rather than to its constituents (i.e., its subunits and atoms).⁴⁰

The macroscopic property of the molecule causally responsible for the fact that the molecule tends to bind an oxygen molecule in the lungs, and to release this molecule in the tissues, is its structure (i.e., its “conformation”). This macroscopic conformation can be generated, or determined, by a large number of configurations at the level of the constituent atoms. In a sense, there are many different kinds of hemoglobin in different biological species, each one different from the others at the atomic level. Hemoglobin is a complex molecule — a “tetramer” — made up of four chains of amino acids. Each type of hemoglobin is characterized by its own sequence of amino acids: this sequence is known as its “primary structure.” However, hemoglobins differ from each other in most, but not all, of their 140 constituent amino acids: there are nine amino acids that occupy the same position in the primary structure of all hemoglobin molecules. The interactions between these nine component molecules are sufficient for a sequence to determine the specific conformation common to all hemoglobin molecules, which is then responsible for oxygen uptake. Insofar as there is only one common macroscopic conformation, it is also correct to say that there is only one type of hemoglobin molecule. In this case, “hemoglobin” is identified by its macroscopic property of having the conformation responsible for binding with oxygen.

The natural determination of the conformation of the whole molecule by the primary structure of the chain of amino acids — sketched in Figure 5.1 — involves two intermediate stages: interactions between the amino acids determine where the chain bends and folds, giving rise to the secondary structure. That structure brings together certain amino acids that would have been far apart in the primary structure, giving rise to new interactions that in turn determine the tertiary structure, the shape that the chain takes in space. Finally, hemoglobin is not, strictly speaking, a molecule but an aggregate of four molecules known as its subunits. Given the tertiary structure of the four subunits, they adopt a stable position in relation to each other, constituting the quaternary structure of the molecule: that is, of the aggregate, the oxygen-binding functional unit. This overall structure or conformation of the molecule, a systemic property of the molecule as a whole, is directly responsible causally for its interaction with oxygen: it is the conformation of hemoglobin that, in the relatively low pH environment of the lungs, causes oxygen molecules to bind to the iron molecules contained in the heme groups surrounded by each of the subunits.

The hemoglobin molecule illustrates the possibility that a macroscopic property has a causal responsibility of its own, in the sense that it is not an epiphenomenon of some underlying microscopic causation. It is not the properties of the atoms or amino acids making up hemoglobin that bind oxygen. Nor is this effect explicable on the model of the superposition of independent effects of each atom or each amino acid. None of these microscopic components has the tendency to bind to an oxygen molecule. Microscopic tendencies do not add up to a sufficiently strong attraction. On the contrary, this is a genuine case of macroscopic causality since the tendency to bind an oxygen molecule appears only at the level of the quaternary structure. There are no causal relationships at the level of the components of hemoglobin, the cumulative result of which would be to bind oxygen.

The case of the hemoglobin molecule shows that there is nothing mysterious about a causal relationship at the level of the whole system. On the contrary, on the basis of “what molecular biology has discovered about the hemoglobin molecule,” chemistry provides “a direct and beautiful explanation of why the blood does what it does” (Rosenberg 1985, 74). Contrary to Klee’s claim, the fact that a relationship of causal responsibility must be understood at a systemic — and thus (relatively) macroscopic — level does not entail that the “how” and “why” of its efficacy remain mysterious. The fact that the causally efficacious property is macroscopic does not imply that it is irreducible. On the contrary, the explanation of the relations that determine the causally efficacious property (the quaternary structure) by the primary structure, with its two intermediate stages, is a paradigmatic example of a successful microreduction.

We do not need to choose, as Klee’s argument suggests, between macroscopic but mysterious determination and determination by a microscopic mechanism. According to my analysis, the mystery is unravelled thanks to the reduction of the efficacious property. This reduction involves the discovery and detailed description of non-causal determination relationships among primary, secondary, tertiary, and quaternary structures. However, the existence of the reductive explanation does not prevent the relationship of causal responsibility from involving an emergent property.

3.2. Kim’s Argument against Mental Causation: Preliminaries

Jaegwon Kim has presented an influential argument against the thesis that mental properties themselves can be directly causally efficacious. His argument has the form of a reductio of the hypothesis of mental causation. He begins by considering the hypothesis that one mental event causes another. Take the previous example, in which my thought that the noise in the street is disturbing my concentration caused my decision to close the window. Schematically, the event in which I think that the noise disturbs me is represented by my possession of mental property M and the decision to close the window by my possession of mental property M*. Kim’s argument consists of showing first that M can have only a causal influence on M* through an influence on the physical properties P* underlying M*. In other words, Kim seeks to show that mental causation, if it were possible, would presuppose downward causation. Then he argues that downward causation is incompatible with two principles widely accepted in metaphysics and epistemology, namely the principle of causal-explanatory exclusion and the principle of causal closure of the physical domain. According to the first principle, which is epistemological, “there can be no more than a single complete and independent explanation of any one event” (Kim 1988a, 233).⁴¹ According to the second principle, which is metaphysical, “if you pick any physical event and trace its causal ancestry or posterity, that will never take you outside the physical domain. That is, no causal chain will cross the boundary between the physical and the nonphysical” (Kim 1997b, 282). This last principle directly denies the possibility that non-physical causes, particularly mental ones, can exert causal influences on physical events. However, we will see that the argument does not simply beg the question, insofar as Kim tries to justify this principle.

The debate about whether mental properties can be efficacious in influencing other mental or physical properties takes place against the background of a consensus on a physicalist conception of mental properties, which are instantiated locally in persons or animals endowed with cognition; they are intrinsic systemic properties of these individuals.⁴² These systemic properties are determined exclusively by the physical properties of the individuals’ parts and their interactions. This presupposes the strong supervenience of mental properties on the properties of the parts of the organism. Strong supervenience can be defined as follows.

Necessarily, if a macroscopic object with parts p₁, . . ., p_n has a global property G, then there exist properties P₁(p₁), . . ., P_n(p_n) of the parts and relations between the parts R₁(p₁ . . . p_n) . . . such that, necessarily, any macroscopic object that has parts p₁ . . . p_n, with properties P₁(p₁), . . ., P_n(a_n) and relations R₁(p₁ . . . p_n), . . . has the property G.

In Chapter 4, I explicitly required that emergent properties satisfy this physicalist condition of mereological determination. This conception of emergence is consistent with Kim’s thesis that “a particularly important and promising approach . . . is to explicate mind-body supervenience as an instance of mereological supervenience. That is, we try to view mental properties as macroproperties of persons, or whole organisms, which are determined by, and depend on, the character and organization of the appropriate parts, or subsystems, of organisms” (1993b, 168; see also Kim 1978, 1988b, 1990).

Kim judges correctly that the problem of mental causation is a special case of the more general problem of understanding when and how different properties of an event cooperate in their causal influence on other events and when, on the contrary, the causal efficacy of one property excludes the causal efficacy of another. The dispute concerns the question of whether a mental property M can have its own causal efficacy in bringing about another mental property M*, or whether the impression of such causal efficacy is illusory, since the only relationship of causal efficacy relates the underlying physical properties P and P*. We can sketch the situation as in Figure 5.2.

Before considering in detail Kim’s argument to the effect that only physical properties can be causally efficacious, it is useful to mention three terminological points. The first concerns the ontology of the terms of causal relation: according to Kim, events linked by causal relations are structured entities composed of an object, a property that the object exemplifies, and an instant at which the object exemplifies the property. The effect considered in our example consists for Kim of a triplet <s, M, t>, where s represents the bearer of the property (i.e., the person who decides to close the window), M the property of deciding to close the window, and t the instant at which this decision is made. However, in the context of his analysis of mental causation, Kim expresses himself more simply by talking about the property that causes or determines another property. This does not create any misunderstanding insofar as the context fixes the object and the instant univocally.

In this simplified terminology, Kim’s thesis is this: the fact that M causes M* is only an alternative way of conceiving of the only real causal relationship, namely that between P and P*. Instead of saying that the event <s, P, t> causes the event <s, P*, t*>, we can simply say, with Kim, that P causes P*. There should be no misunderstanding about the relevant conception of properties: when we ask whether a “mental property M is causally efficacious with respect to physical property P*” (Kim 1993c, 207), it is clear that local instances of these properties, not universal properties, are at stake. From now on, it will be understood that this means “that a given instance of M causes a given instance of P*” (Kim 1993c, 207). Similarly, given that we are dealing with causation and that the concept of causation requires the temporal anteriority of the cause with respect to the effect, properties of effects — designated by starred predicates (here P*) — are systematically exemplified later — namely at t* — than properties without a star (here M) — namely at t. A statement that one property M causes another M* can always be translated into the language of causation between events: in the conception of events as concrete particulars, the mental property M and the physical property P on which it supervenes and by which it is determined belong to the same concrete event.⁴³ The question of the respective roles played by M and P is posed in this context in terms of causal efficacy or causal responsibility. Which fact is causally responsible for the fact that the effect-event has the property P*: the fact that the cause-event has M or the fact that it has P? Or, equivalently, which of the properties M and P of the cause is efficacious in ensuring that the effect has the property P*?

Another terminological detail concerns the object to which the properties are attributed. If M belongs to a subject, then we can consider the question of its efficacy in relation to the underlying physical properties, insofar as they belong to the subject’s parts, particularly neuronal parts, or insofar as they belong to the same subject as a whole. In order to express himself in the second way, Kim uses the concept of a micro-based property (see Chapter 4.8): “P is a micro-based property just in case P is the property of being completely decomposable into nonoverlapping proper parts a₁, a₂, . . ., a_n, such that P₁(a₁), P₂(a₂), . . ., P_n(a_n), and R(a₁, . . ., a_n)” (Kim 1998, 84). This is a technical means of attributing the physical properties of the parts of an object to that object as a whole, without prejudging whether the property thus logically constructed and attributed to the object is a real property or only a nominal one: a real property has causal powers of its own, whereas many merely nominal properties do not.⁴⁴ The table that I am sitting at, for example, has the disjunctive property of being rectangular or yellow, but this property is not “natural” or “real”⁴⁵ insofar as it does not give the table any causal power of its own, different from the powers due to the properties of being rectangular and of being yellow.

A third terminological point should be clarified before I examine Kim’s argument. We must not confuse the physicalist thesis that I expressed above as the thesis of mereological supervenience with the “Physical Realisation Thesis” (Kim 1993b, 344; 1993c, 198). According to the latter thesis, “all mental properties are physically realized; that is, whenever an organism or system instantiates a mental property M, it has some physical property P such that P realizes M in organisms of its kind” (1993b, 344; 1993c, 198). The first thesis concerns the micro-macro relationship among the parts of an organism and its global properties; the second concerns the relationship between second-order functional properties and the first-order properties that perform the function. For example, to conceive of the property of experiencing pain as a functional property is to conceive of it as a second-order property: the property of having a first-order property from within a set B. The properties in B satisfy the functional constraints specific to pain: being caused by damage to the body, causing behaviour aimed at avoiding the cause of that damage, causing the desire for the pain to cease, and so on. According to the physical realization thesis, mental properties are actually physical properties conceived of by second-order concepts (Kim 1998, 104). To have a mental property is not, according to this conception, to have a first-order property (i.e., a causally efficacious property). Rather, it means having a second-order property: the property of having a physical property that obeys certain causal or functional constraints. I am not denying that it is possible or appropriate to conceive of mental properties in this functional way. I am only contesting the legitimacy of inferring from this possibility that mental properties are not first-order properties. The fact that a property can be described with a second-order predicate does not prove that it is impossible to describe the same property with a first-order predicate. Thus, it remains possible, as emergentism envisages, that mental properties are macroproperties that “can, and in general do, have their own causal powers, powers that go beyond the causal powers of their microconstituents” (Kim 1998, 85). In other words, properties that fulfill the functional roles corresponding to second-order functional concepts can be macroscopic mental properties that can also be designated by first-order predicates (see Chapter 3).

Here is what is at stake in the debate: I defend the thesis that an emergent mental property is a first-order property, determined by the physical properties of the brain’s components according to laws of composition, which possesses causal powers not possessed by the physical properties that determine its existence. Against this, Kim argues that the only causally efficacious properties are physical or physiological properties (i.e., properties that belong to the parts of the person’s body), although the concept of a micro-based property makes it possible to attribute them formally to the person herself.

3.3. The First Part of Kim’s Argument: No Mental Causation without Downward Causation

As I said earlier, Kim’s argument proceeds in two stages. First, Kim explains that the only way M could cause M* is by downward causation; M can cause M* only by causing P*, the physical property underlying M* by virtue of mereological supervenience. Second, he tries to show that such a downward causal influence is incompatible with the principles of causal-explanatory exclusion and of the causal closure of the physical domain.

Here is the first part of the argument. Kim says that, for a given mental property M* (as sketched in Figure 5.2), there are two possible answers to the question of what is responsible for M*. According to the first, M* is due to M; according to the second, M* is due to P*. Kim then argues that there are only three ways of reconciling these two answers. Of the three, he tries to show that the first two are unacceptable. The third is therefore provisionally retained, although the second part of the argument shows that it too must be abandoned, depriving M of all causal power both to cause P* and to cause M*. The possible answers are

1. M and P* are jointly responsible for M*;
2. M and P* are each responsible for M* so that M* is overdetermined; and
3. M causes M* by causing P*.

According to hypothesis (1), M and P* taken together are responsible for M*, although neither alone is sufficient for M*. Kim rejects this hypothesis on the ground that the supervenience thesis implies that P* alone is sufficient for M*. He argues against assumption (2) — that M and P* overdetermine M*, in the sense that each alone is sufficient for M* — by proposing that the concept of overdetermination would be appropriate in the present case only if M* had “two distinct and independent origins in M and P*” (1993c, 205). However, as the situation is conceived of, M and P* are not independent. Compare the situation with the paradigmatic case of independent overdetermination described above: when two soldiers in a firing squad both fire a bullet at the victim, such that each is sufficient in itself to cause her death, each causes the victim’s death. Since we are not dealing here with such an exceptional situation in which two independent causal chains converge on the same effect, Kim applies the principle of explanatory exclusion: there is only one complete and independent explanation for M*. From the fact that P* alone is sufficient for M*, he concludes that M is not a complete cause of M*, in the sense of constituting a complete causal explanation of M*.

This argument invites the objection that it fails to make the crucial distinction between causal and non-causal determination; this neglect goes hand in hand with the neglect of the difference between the relationships expressed by “determines,” “is a sufficient condition for,” and “is causally responsible for.”⁴⁶ If we distinguish these concepts of determination, then it appears that hypotheses (1) to (3) do not exhaust all possibilities. There is a fourth way of conceiving of the relationships of determination among M, P*, and M* that does justice to the premises of Kim’s argument while avoiding its conclusion.

The argument against hypothesis (2) shows that what Kim means by “overdetermination” is independent causal overdetermination such as it exists in the case of the firing squad, in which several independent causal processes lead to the victim’s death. However, M and M* are indeed in a relationship of causal responsibility, but the determination between P* and M* is not causal: it is the relationship between a neuronal property and the mental property that emerges from it as a function of a law of composition.⁴⁷ First, an emergent property and the properties from which it emerges belong to the same object at the same time, whereas a causal relationship requires its terms not to overlap in space-time.⁴⁸ Second, emergence is based on nomic necessity, whereas causation is contingent. Since P* and M* are not in a causal relationship, M and P* cannot overdetermine M* in the sense of causal overdetermination.

In other words, both M and P* determine M*, though not in the same way: M causes M*, by virtue of a nomological dependence between these properties, whereas P* determines M* in a non-causal way. Thus, there is a fourth way, neglected by Kim, to reconcile the two explanations of the presence of M*: the first is causal, whereas the second is a non-causal explanation (which we can call compositional). M and P* do not causally overdetermine M*, although each alone is sufficient for M*. In other words, the expression “is sufficient for” can have two meanings: it can express the relationship of causal responsibility and the relationship of non-causal compositional determination.

Here are two simple examples showing that explanations belonging to these two categories answer independent questions so that they complement rather than exclude each other. Why is there an equilateral triangle on the paper on my desk? First, the causal answer: because I drew it. Second, the non-causal answer: because there is an equiangular triangle on the paper, and all equiangular triangles are necessarily equilateral. Why is this gas at a temperature of T = 50° C? First, the causal answer: because I have just increased its temperature by heating its container. Second, the non-causal answer: because its pressure and volume are P and V, because it is an approximately ideal gas, and because T is proportional to the product of P and V, by virtue of the ideal gas law.

The existence of two determinations of the same property instance, each “complete” in its own way and independent of the other, is certainly incompatible with the principle of explanatory exclusion, according to which “there can be no more than a single complete and independent explanation of any one event” (Kim 1988a, 233). In this context, “complete” means “in itself sufficient,” and “independent” means “without conceptual link or link of logical or metaphysical necessity.” However, the coexistence of our two determinations is compatible with another principle, which can easily be confused with the principle of explanatory exclusion. According to the principle of causal-explanatory exclusion, it is exceptional for there to be two complete and independent causal explanations of an event. Such a principle simply does not apply in the cases that we have considered, insofar as one of the two determination relationships is non-causal.

We therefore arrive at the following analysis of our schematic situation of mental causation. Let us assume that there is a psychological law linking instances of M to subsequent instances of M*. The instance of M* can then be causally explained by the preceding instance of M, together with the psychological law in question. This does not prevent the possibility that the same instance of M* can be explained alternatively non-causally, in terms of the physical or physiological properties P*, on the basis of the relevant law of composition by virtue of which P* gives rise to M*.

If this analysis is correct, then I have refuted Kim’s argument that the only way to account for the fact that M* is determined both by M and by P* is to assume that M causes M* by causing P* (i.e., by virtue of downward causation).

Furthermore, we have seen that a “principle of explanatory exclusion” is not plausible. There is no reason to deny the possibility of two explanations of the same fact when these explanations belong to different categories — one being causal and the other non-causal. This is true even if each can be considered complete in its own right, in the sense that it allows us to deduce the explanandum.⁴⁹

In defence of Kim’s argument, it could be argued that the principle of explanatory exclusion should be interpreted as a principle of causal-explanatory exclusion. According to this principle, there cannot be more than one complete and independent causal explanation for a given event. After all, Kim defends an “explanatory realism” according to which any true explanation is based on a causal relationship between the terms of the explanation: “To ‘have an explanation’ of event e in terms of event c is to know, or somehow represent, that c caused e; that is, . . . explanations of individual events are represented by singular causal propositions” (Kim 1988a, 230).

It is crucial here to bear in mind the distinction mentioned above between two ways of conceiving of the terms of causal relationships: either they are events in the sense of particulars that occupy a portion of space-time and have many properties, or they are facts — which Kim calls “events” — that is, propositional entities. In the first conception, the cause c, as a particular event, can possess both the physical property P and the mental property M and the effect event e both the physical property P* and the mental property M*. There is no reason, then, as Marras (1998) has observed, to deny the possibility that the fact that c is M is causally responsible for the fact that e is M* and that, so to speak in parallel, the fact that c is P is causally responsible for the fact that e is P*.

To evaluate the principle of causal-explanatory exclusion, we must distinguish between its application to events and its application to facts: it is doubtful whether it makes sense to speak of a complete explanation of a particular event. What we are trying to explain are always facts: insofar as an explanation always takes the form of an argument whose explanandum constitutes the conclusion, only a propositional entity can be the object of an explanation. However, when it is said that there is only one complete causal explanation of an event, this can mean two things. First, it can mean denying the possibility of downward causation. It denies that two different facts about the same event, in our case the fact that c is P and the fact that c is M, cause the same physical fact, that e is P*. We will come back to this in a moment. Second, it can mean that there cannot be two causal relationships of responsibility concerning the same pair of events: on the one hand, the fact that c is P is causally responsible for the fact that e is P*; on the other hand, the fact that c is M is causally responsible for the fact that e is M*. The principle of causal-explanatory exclusion does not seem to apply to the latter case: the two relations of causal responsibility do not seem to exclude each other. This is not a case of causal overdetermination since their effects are different. Moreover, the existence of two parallel explanations at different levels can be explained by the fact that their truth makers are not metaphysically independent. Since, according to physicalism, for every mental property M there is an underlying physical property that determines it, mental properties depend on physical properties, and mental explanations depend on mental properties. If there is a physical explanation, however, the psychological explanation depends on this physical explanation. However, as we will see, the dependence of mental properties on physical properties does not exclude the possibility that there is no physical explanation for a pair of causally related events, in which case the only accessible explanation is psychological.⁵⁰

As we saw in Chapter 1, certain elementary mechanisms of learning have been reduced to neurophysiology, in the sense that microscopic mechanisms have been discovered that give rise to the regularities observed at the psychological level. As far as classical (i.e., Pavlovian) conditioning is concerned, Rescorla and Wagner (1972) discovered that the reinforcement of the associative strength between an unconditional stimulus (US) and a conditional stimulus (CS) X (see Chapter 1.9), in an experimental session in which X is presented just before the US, obeys a law expressed by the following formula:

This formula corresponds to the general case in which the same US has already been used to condition the subject to another conditional stimulus, A.

This law can form the basis of a relationship of causal responsibility between one mental property and another and thus illustrate the situation outlined in Figure 5.2. The earlier mental state M contains the associations between A, X, and the US created during the first n–1 exposures as well as the experience of X preceding the US in the n^th exposure; the law predicts that this produces a new mental state M* that contains associations whose strength is a function of M, as expressed by the law. This law applies to various animal species (Rescorla and Wagner report the results of experiments with rabbits and rats); this makes it unlikely that there is a single underlying microscopic property common to different species for each mental state that emerges during learning.

I have shown that the first step of the argument by which Kim tries to establish the impossibility of mental causation is not conclusive. This does not diminish, however, the importance of the second step of his argument. In that step, Kim intends to show that the hypothesis of the existence of downward causal relations is incompatible with the principle of the causal closure of the physical domain as well as with the principle of causal-explanatory exclusion. Even if such downward causation does not play a role in every causal relationship between mental events, the emergentist position defended in this book in fact presupposes that emergent properties, such as cognitive properties, can have a causal influence on events situated at lower levels of complexity, in particular on physiological events: my decision to close the window is an emergent property of my person, which leads causally to the modification of the state of the motor neurons involved in the execution of the appropriate action.

3.4. The Second Part of Kim’s Argument: No Downward Causation

In Figure 5.2, the downward causal influence is represented by the diagonal arrow: it corresponds to the hypothesis of a causal relationship between the thought that the noise bothers me (M) and the microscopic neuronal events (P*) that are part of the basis that determines the decision to close the window (M*). To demonstrate the incompatibility of this hypothesis with the principles of causal closure of the physical domain and causal-explanatory exclusion, Kim begins by arguing that there are four possible causes of P*.⁵¹

1. M and P together constitute a sufficient cause of P*;
2. M and P are two mutually distinct sufficient causes of P*, so that M and P overdetermine P*;
3. P causes P* through M; and
4. P causes P* directly, without any causal contribution from M.

Only the first three hypotheses presuppose the existence of downward causation. If there are no other possibilities, then Kim only needs to refute these three scenarios in order to refute the possibility of downward causation.

3.4.1. The Refutation of Scenario (1)

Against hypothesis (1), Kim (1993c, 207) argues that the physical realization thesis implies that P by itself is sufficient for P*. Moreover, if P is sufficient for M, and if P and M together are sufficient for P*, then P alone is sufficient for P*. As I pointed out earlier, the choice of conceiving of M as a second-order functional property and P as the first-order property that realizes it would prejudge the question of the causal efficacy of M in the negative. By its very conception, a second-order property has no causal efficacy of its own because causal efficacy is limited to the level of the first-order properties that realize it. However, we can express an argument similar to Kim’s that respects our conceptual framework: if M is a first-order macroscopic property, and if P is the set of microscopic properties that determines it according to a law of composition, then P is sufficient for M. However, in the latter case, this is a sufficient condition for nomological reasons and not for logical or conceptual reasons, whereas Kim’s functionalist conception takes the relationship between P and M to be the conceptual relationship of realization.

We can interpret (1) in two ways, corresponding to two ways in which P and M can “join forces,” in the sense of each contributing to the effect P*. In both interpretations, P alone is sufficient for M. The relationship of being a sufficient condition here covers both the relationship of non-causal determination — between P and M — and the relationship of causal determination. The difference between the two interpretations concerns the latter: according to the first interpretation, M alone is causally responsible for P* by virtue of a causal law. In other words, P* exists only because — and insofar as — it is determined by M, but P does not take part in the direct causal responsibility for the production of P*. We can then say (although this is true only ceteris paribus, as with all applications of the laws of the special sciences) that M is “sufficient for P* in the circumstances.”

According to the second interpretation,⁵² both P and M exert a constraint on the evolution of the system, but neither determines the state of the system, at the time of the event P*/M*, completely on its own. M determines only a certain framework within which the system must evolve but not its detailed evolution. This constraint can be construed by analogy with the constraints exerted on a mechanical system. The rail line on which a train is travelling does not in itself determine the trajectory of the train: the forces acting on it — notably gravitation, the force released by the locomotive engine, and the force of friction — determine the direction and speed with which it moves. However, the rail line determines a framework that restricts the number of degrees of freedom available to its movements to two: direction and speed. In the case of a ball rolling inside a bowl, the internal surface of the bowl determines a frame that limits the possible trajectories of the ball. This surface exerts a constraint on the trajectory of the ball that leaves it with only four degrees of freedom or four dimensions in what is known as the phase space of its trajectory: the two dimensions of the surface and the two dimensions of the velocity. Without any constraint, such as that exerted by the rail line or the surface of the bowl, the displacements can occupy all six dimensions of the phase space: three dimensions for the position in space and three dimensions for the components of the velocity in the three spatial dimensions.

To return to the determination of the physical state P* of a cognitive system, the second interpretation of hypothesis (1) amounts to supposing that the mental state M defines a framework that limits the possibilities of evolution of the system without determining them completely in detail. Let us assume that there is a psychological law M → M* stating that, ceteris paribus, any system in M evolves toward state M*. At the instant of M, the system has a well-defined physical state P that determines M in a non-causal way. But M is compatible with a certain number of other physical states that would determine a mental state of the same type as M. In other words, M does not determine P in detail, only a certain framework that delimits a space of possibilities. This thesis corresponds to the thesis of “multiple realizability” in the sense of micro-macro determination, which we examined in Chapter 2. Similarly, M*, although causally determined by M by virtue of the law M → M*, does not determine the details of its physical realization, P*. Rather, M* constitutes a framework that constrains the evolution of the system, the details of which are determined at the physical level (i.e., by P).

By virtue of the transitivity of the relationship of being a sufficient condition, it is correct to say in both interpretations of hypothesis (1) that the instance of P is sufficient for the instance of P*. Without distinguishing between the two interpretations, Kim rejects (1) because it lacks parsimony. However, his argument is based on a strong presupposition: following the (Davidsonian) principle of the nomological character of causality, Kim assumes that the causal relationship between M/P and M*/P* is determined by physical laws.

According to Kim,

P appears to have at least as strong a claim as M as a direct cause of P* (that is, without M as an intervening link). Is there any reason for invoking M as a cause of P* at all? The question is not whether or not P should be considered a cause of P*; on anyone’s account, it should be. Rather, the question is whether M should be given a distinct causal role in this situation? I believe there are some persuasive reasons for refusing to do so. (1993c, 207)

Since there are physical laws that determine P* on the basis of the earlier state P, the assumption that M also contributes to P* — whether in the first or the second interpretation of (1) — violates the principle of explanatory simplicity, which finds its precise expression in the principle of causal-explanatory exclusion. The hypothesis that the complete physical cause P* is systematically accompanied by a second cause M, itself complete as in the first interpretation of (1) or incomplete as in the second, must be excluded.

The question of the level of the laws that determine P* is empirical and cannot be decided a priori, on mere conceptual grounds, as Kim’s argument presupposes. This applies in particular to the hypothesis that, for any physical event P* taking place at time t, there exists, at each preceding time t, a complete and purely physical cause P of P*. This is an empirical hypothesis that has no more a priori credibility than the hypotheses underlying the two interpretations of (1): according to the first, there is a law by virtue of which M alone determines P*; according to the second, there is a psychological law M → M* by virtue of which M determines a framework that restricts the degrees of freedom of the evolution of the system and physical laws that determine, within the framework fixed by the psychological law, the details of the physical evolution from P to P*.

No a priori argument can establish which of these three possibilities is correct. Therefore, Kim is wrong to assume that the first hypothesis, according to which P alone determines P* by virtue of laws at the physical level, needs no empirical justification. Indeed, it is empirically possible that there is no law at the level of P and P* that would make P* predictable from P: this might be the case of a chaotic system in the physical sense. If P describes the state of a chaotic system in the basin of a strange attractor,⁵³ then for any accuracy or tolerated error there is a time in the future such that the states of the system after that time cannot be predicted with that accuracy. In such a situation, it is possible that (1) is correct, according to one of its two interpretations. I will come back to this later.

3.4.2. The Refutation of Scenario (2)

Against hypothesis (2), Kim takes the hypothesis that M and P causally overdetermine P* to be “absurd” (1993c, 208). This judgment is plausible insofar as causal overdetermination is taken to mean the parallel actions of two independent causal processes that lead to the same effect, so that each would have been sufficient without the other for the effect to occur. The paradigmatic case is the firing squad, in which several bullets reach the heart of the victim by causally independent paths at the same instant. It is certainly not plausible for cerebral and mental causes to act systematically independently of each other to cause P*. Independent but converging causal processes such as those occurring in the scenario of the firing squad are rare and exceptional. Kim expresses this by saying that the assumption that each mental cause is accompanied by an independent physical cause is incompatible with the principle of causal-explanatory exclusion, mentioned above. Two complete causal explanations of a fact create, he says, “an unstable situation requiring us to find an account of how the two purported causes are related to each other” (1998, 65). There are good reasons to abandon the hypothesis that, if a mental cause M brings about a physical effect P*, then it always overdetermines P*, in the sense that P* always also has a parallel complete physical cause P.

It is less easy to show why the other interpretation of overdetermination also leads to an absurd consequence. It seems to be possible that mental causes M and their underlying properties P overdetermine their physical effects P*, in the sense that each is sufficient for P* by virtue of a causal law (purely physical in one case and psychophysical in the other) but without one being independent of the other.⁵⁴ Indeed, several authors have suggested that the efficacy of mental causes is a matter of “dependent overdetermination” (Witmer 2003, 205). In a situation in which the effect is overdetermined in this way, the fact that the mental property is efficacious in causing the effect depends on an underlying physical process. Once dependent overdetermination is clearly distinguished from “autonomous overdetermination,” in which neither of the two causes depends on the other, it seems to be conceivable that the former provides an appropriate model for mental causation. Indeed, the only reason to deny that mental causes systematically overdetermine their effects presupposes that overdetermination is always autonomous. Indeed, a large-scale “systematic coincidence,” without any basis in mutual dependence or in relation to a third factor, is certainly not plausible. Bennett (2003) defends what she calls “compatibilism”: that is, the thesis according to which the causal efficacy of a mental property M is not “excluded” or “pre-empted” but perfectly compatible with the causal efficacy of the underlying physical property P. She writes that, “if a mental cause is efficacious in bringing about some effect, the only physical causes that are also efficacious in bringing about that effect are ones that necessitate the mental cause” (2003, 487).⁵⁵ However, none of these authors considers the possibility that, in certain situations, the mental cause might be an indispensable component of what is causally responsible for some physical fact P* concerning the subject’s body at t*. This would be the case if the evolution of the underlying neural state were chaotic in the sense of systems theory. In this case, there is a time t earlier than t*, such that in principle there is no knowable neuronal state P at t, which could be used to explain P* deductively. The mental cause would be necessary — possibly together with the neuronal state — to explain the bodily movement in question.

To draw a metaphysical conclusion — that the state P* of all the parts of the system at t* is not causally determined by the state P of all the parts of the system at t — from the impossibility of long-term prediction in a chaotic system, two presuppositions have to be made. The first concerns the interpretation of the notion of causal determination. As we saw earlier (Chapter 5.2), causal relationships can be analyzed at two levels.

1. Causal relationships can be conceived of as relationships between particular events, where “particular” is taken to mean a concrete object or event with many properties. We assume that causation is based on the transmission of a quantity of energy (or some other conserved quantity) from one event to the other.

2. However, in most contexts in which we are interested in a causal relationship, in particular in the context of scientific explanation, we conceive of causation in terms of certain well-defined properties of events: for example, we do not just want to establish that there is a causal relationship between two successive episodes in the life of an organism learning an association by classical conditioning. Instead, the aim is to understand a fact F₂ about the organism at time t₃ after a learning episode, as a function of a fact F₁ about the learning episode, so that F₁ is causally responsible for F₂. F₂ might be the fact, for example, that the associative strength V_A between a conditioned stimulus (A) and an unconditioned stimulus (US) increased by ΔV_A during a conditioning episode in which the subject was exposed, at t₁, to a stimulus consisting of A and X, before being exposed, at t₂ immediately following t₁, to the US. According to Rescorla and Wagner (1972), what is causally responsible for F₂ is a fact that relates to the cognitive system at time t₁ and to the learning episode that takes place between t₁ and t₂: F₁ relates to the associative strengths V_A and V_X of stimuli A and X before the conditioning episode and various parameters. F₁ causally determines F₂ by means of a law, according to which the change in associative strength is equal to

$\Delta V_A={\alpha }_A\beta \left(\lambda -V_{AX}\right)$

, where l is the maximal strength of association that can be obtained with the US, α_A and β are “learning rate parameters” (Rescorla and Wagner 1972, 76), α_A being specific for the A and β for the US, and V_AX is the combined associative strength of stimuli A and X at t₁, with V_AX = V_A + V_X. We can interpret the causal determination of F₂ by F₁ in an ontological way: the associative strength after the learning episode is independent of our knowledge about and descriptions of these facts. To interpret this second aspect of the causal relationship, we can start from the deductive-nomological analysis of causal explanation: to explain F₂ causally is to produce a deductive-nomological argument of which F₂ is the conclusion and whose premises contain the initial condition F₁ as well as a certain number of statements of laws of nature. From a realist perspective, we can infer that, for this causal explanation to be true, there must exist facts F₁ and F₂ and laws N expressed by the nomological statements, such that F₁ is causally responsible for F₂, given the laws N.⁵⁶

The second presupposition concerns the interpretation of indeterminacy in a chaotic system:⁵⁷ not only is it epistemic, but also it has an ontological interpretation. When it comes to determining the value of a measurable quantity in a physical system that takes its values in a continuum, we cannot attribute any empirical meaning to the hypothesis that this quantity has a value known with absolute precision. We might suppose that measurable quantities nevertheless objectively possess values of infinite precision. If we assume determinism, then the evolution of a system is objectively determined with infinite precision within any arbitrary time. However, the state of a physical system at time t can only be known, even in principle, with finite precision. If the system is chaotic, then for any finite precision or margin of error there is some time t* in the future such that the state of the system after t cannot be determined with a precision within that margin. Therefore, if we consider states that can be known at least in principle, then such knowable-in-principle states of the system at t do not determine knowable-in-principle states of the system for times after t*.

3.4.3. The Refutation of Scenario (3)

Against scenario (3), Kim offers three arguments. First, “given the simultaneity of the instances of M and P respectively, it is not possible to think of the M-instance as a temporally intermediate link in the causal chain from P to P*” (1993c, 207). However, this argument makes it possible to refute scenario (3) only within the framework of the traditional theory — which we reject — according to which all determination is causal. Kim’s observation that P cannot be a cause of M is correct. However, this does not refute schema (3) as such, only one of its possible interpretations. It refutes

(3a) P causes P* by causing M.

But it leaves open the possibility that

(3b) P causes P* by determining in a non-causal manner property M, which causes P* by virtue of a psychophysical causal law,

correctly represents the situation.⁵⁸ (3b) is equivalent to the first interpretation of hypothesis (1). Kim neglects this possibility because he follows the empiricist tradition according to which there are only two ways that properties M and P can be nomologically correlated: either by a causal law or by identity. After objecting to Searle’s conception of the relationship of P to M as a causal relationship, Kim concludes that the only alternative that avoids the problem of overdetermination is the hypothesis that they are identical (1998, 48).

The second argument against scenario (3) is as follows: by virtue of an “inference to the simplest explanation”⁵⁹ that Kim has already followed in arguing against hypothesis (1), the assumption that scenario (3) describes the situation correctly must give way to the simpler explanation according to which P causes P* without any intervention by M. However, this explanation, whose greater simplicity is indeed undeniable, works only if “there is an appropriate law connecting P-instances with P*-instances” (Kim 1993c, 207). The existence of such a law is not guaranteed a priori. Kim argues that the principle of closure of the physical domain guarantees the existence of a purely physical causal relationship between P and P*. According to this principle, every physical event has a complete physical cause at every instant preceding it. Applied to P*, this principle guarantees the existence of a complete physical cause at the instant of M, namely P.

Kim’s third argument refutes scenario (3) by arguing that it is inadequate by virtue of “the problem of causal-explanatory exclusion” (1993c, 207).⁶⁰ If there is a causal determination relationship between P and P*, then the causal explanation that it makes possible “excludes” the adequacy of scenario (3). By virtue of causal-explanatory exclusion, two complete causal explanations cannot coexist in parallel, at least not systematically. Several objections can be raised against this argument. First, the physical parts of the brain that possess P and M have physical properties; however, the hypothesis of the existence of a purely physical law connecting P and P* begs the question against my hypothesis according to which the physical properties P have causal effects only through the intermediary of the global properties that emerge during their interaction, which can be mental properties M. Second, even if a law P → P* exists alongside the law M → P*, the conclusion that M lacks causal efficacy still depends on the controversial thesis that there is no systematic overdetermination of events that have a mental cause. Third, the appeal to simplicity does not in itself settle the matter one way or the other. Even if an explanation of P* in terms of P were possible, it seems to be plausible that it would be far more complex than the explanation that appeals to M and the psychological law M → M*. Rather than justifying Kim’s conclusion, the appeal to the simplicity of explanation provides a strong argument for the realism of the mental properties (see Rey 1997).

It is interesting to compare my critique of Kim’s argument with a suggestion by E.J. Lowe for making the principle of physical closure compatible with the causal influence of the mind on the physical world. Lowe shows that the causal inefficacy of the mind can be concluded only from the assumption that all physical events, at every instant preceding them, have a sufficient physical cause.⁶¹ However, a weaker principle according to which “every physical state has a fully sufficient physical cause” (Lowe 2000b, 30) is compatible with a dualist interactionist conception⁶²: even if P* has a sufficient physical cause at some moment that precedes it, say P, it is possible that the causal chain from P to P* passes through intermediate mental steps. In the scenario considered by Lowe, P (at t₁) causes M (at t₂), which causes P* (at t₃). He explicitly states that this scenario fits in well with the emergentist thesis (in the sense of diachronic emergence, which has it that the evolution of the human mind began with a purely physical state (P) and begins, at some moment, to give rise to mental states M that can then influence physical states (P*). The problem with this scenario — which Lowe does not consider — is that it seems to lead back to the difficulty highlighted by Kim. It seems to be difficult to admit that the physical processes between P and P* are, so to speak, interrupted at the physical level: for someone following the evolution of the brain on a purely physical or neurophysiological level, the influence of M on P* would appear to be a mysterious intrusion.

Conversely, my way of reconciling the causal efficacy of mental properties with the causal closure of physics does not encounter this difficulty. Unlike Lowe’s scenario, my suggestion is that an event with property P causes an event with P* without making the dualist hypothesis that some events in the intermediate causal chain are purely mental.

The contribution of M might be indispensable from the point of view of causal responsibility. The event that has M also has physical properties P. However, there might be no law at the physical level according to which P directly determines P*. Rather, P makes M emerge as a global property of the person, by virtue of the interactions between the parts of the system (where P is a micro-based property determined logically by the properties of the parts of the system and their relations). Causal responsibility for P* is shared between P and M; there is a psychological ceteris paribus law that imposes on the system at the moment of the event P*/M* the constraint of possessing the property M*. It is the state P that determines which one of the possible physical situations that give rise to M* is realized. Similar scenarios are scientifically plausible in many contexts that do not involve mental causation.

As I have pointed out above, the state of a physical system can be known only with finite precision. Even with perfect knowledge of the laws governing the dynamics of a chaotic system, the best available knowledge of the state at t is not sufficient for deducing the state of the system for times later than t* with arbitrary precision. If we ask for the causally responsible state for the state P* of the air molecules above Paris at t*, specified with some margin of error, then there is always some time t earlier than t*, such that there is no description of the state of the air molecules at t precise enough and knowable in principle that would be sufficient to predict P* with a precision that lies within that margin of error. In this sense, if the state P* at t* is given with some finite precision, then there is no knowable-in-principle state at t causally responsible for P*.

From his refutation of the first three scenarios, Kim concludes that (4) describes the situation correctly. This scenario presupposes — as does scenario (2), the argument that Kim offers against scenario (1), and the second and third arguments that he offers against scenario (3) — that there is a law and therefore a direct causal determination relationship between P and P*. Kim justifies this presupposition with a principle that he considers to be part of “the natural picture for the layered physicalist world” (1993c, 208). According to this principle, “all causal relations are implemented at the physical level, and the causal relations we impute to higher-level processes are derivative from and grounded in the fundamental nomic processes at the physical level” (208). According to an equivalent version of this principle, which expresses it in terms of the causal powers of properties, “if M is instantiated on a given occasion by being realized by P, then the causal powers of this instance of M are identical with (perhaps, a subset of) the causal powers of P” (208). However, this “causal inheritance principle” (Kim 1998, 54) applies only to mental properties if we accept the eliminativist and apparently paradoxical assumption (developed in Kim 1998) that there are no mental properties: in fact, according to Kim, there are only physical properties but two kinds of concepts. Physical concepts correspond directly to properties (physical properties, but this specification is redundant since there are only physical properties), whereas mental concepts are second-order concepts that designate physical properties by means of quantification.

I will analyze this conception in a moment. My aim is to defend the plausibility of scenario (3b) — equivalent to the second interpretation of scenario (1) — by showing, contrary to Kim, that the relationship between first-order properties and second-order concepts of these properties is not an adequate model of the relationship between physical properties and mental properties. The causal inheritance principle is plausible only in the context of the assumption that the relationship between the physical and the mental is equivalent to the relationship between first- and second-order predicates. Without the causal inheritance principle, Kim has no argument for his claim that, at each instant preceding the event e that is P*, there necessarily exists a physical fact, that c is P, such that there exists a physical law between P and P* and therefore a relationship of causal responsibility that makes it possible to explain and predict P* at the physical level on the basis of P.

4. Mental Properties or Physical Properties Conceived with Mental Concepts?

According to Kim, there are, strictly speaking, no mental properties. What we mistake for mental properties are mental concepts that apply to physical properties. If this thesis is correct, then the issue of the downward causal influence of mental properties no longer arises: there are no such mental properties with causal powers.

Kim’s suggestion would solve the general problem of downward causation only if all higher-level properties could be construed as functional properties. Kim himself (1997b, 1998, 84–85) denies that mental properties belong in this respect to the same category as other macroscopic properties. On the contrary, he insists on the importance of the difference between two distinctions: on the one hand, the distinction between the macroscopic properties of a complex object and the microscopic properties of its parts; on the other hand, the distinction between the first-order properties — microscopic or macroscopic — of a given object and the functional properties of that object. What appear to be functional properties are really just second-order concepts (expressed by second-order predicates) that quantify over first-order properties.

Like any first-order property, macroproperties of macroscopic objects have causal powers, which can differ from the causal powers of the properties of the objects’ parts. There is no principle equivalent to the causal inheritance principle that would deprive macroproperties in general of any proper causal efficacy, over and above that of the microproperties from which they emerge. The emergentist thesis, according to which mental properties are macroproperties, makes downward causation, as in scenario (3b), conceivable. Each instance of P determines an instance of M, thanks to a non-causal law of composition. If M is causally responsible for P* thanks to a causal law, possibly in conjunction with property P underlying M, then we have a case of downward causation.

Conceiving of mental properties as macroproperties of macroscopic objects, in an analogy with physical, chemical, or biological macroproperties that emerge by virtue of non-linear laws of interaction that apply to the properties of their parts, makes their causal efficacy conceivable. Let us take the example of a laser, which has emergent causal powers. First and foremost, it has the power to emit coherent light (i.e., light whose components are all in phase). A laser causes a beam of extremely monochromatic light; for the first historical laser, the ruby⁶³ laser built in 1960, this wavelength is 6,943 Å (or 694.3 nm). To say that this light is “extremely monochromatic” means that the deviations from the average wavelength of the light emitted are very small compared with this wavelength; in this case, these deviations are of the order of 0.1 Å. This phenomenon is extraordinary in that all other natural bodies emit radiation distributed over a broad spectrum of different wavelengths, which can be modelled by Planck’s law of “black body” radiation. A very specific configuration of atoms in the ruby crystal gives this crystal — as a complex object — the causal power to produce a characteristic beam of coherent, monochromatic light. There is a law of composition, according to which a certain atomic configuration in a ruby crystal determines the structure of the energy levels of the electrons of the Cr³⁺ ions of the ruby crystal that, in specific circumstances,⁶⁴ is causally responsible for the laser emission. This law is deducible, in principle, from quantum mechanics in a way analogous to the deduction of the stability of the H₂⁺ ion that we considered in Chapter 4.

To see the analogy between the structure of causal determination of light emission in a laser and that of causal determination by a mental property, let us call P the set of properties of the atoms making up the ruby crystal alongside their spatial relationships. Let us call C the (chemical) properties that belong to the crystal as a complex object and are responsible for the laser mechanism. C includes in particular the structure of the energy levels of the electrons of the Cr³⁺ ions, as it exists inside the crystal.

The operation of a laser is a case of downward causation from the chemical level to the atomic level: P determines C by virtue of a law of composition, itself determined by the laws governing the interactions between the electrons and protons in the crystal. According to a physico-chemical causal law, C has the causal power to produce, in precise circumstances, a physical effect P*, namely the emission of coherent, monochromatic light.

If this is indeed a case of downward causation, then it can be analyzed according to one of the scenarios described above to elucidate the causal relationships between mental properties M and M* and physical properties P and P*. The only difference is that chemical properties C take the place of mental properties M. The four hypotheses considered by Kim for the case of mental causation correspond to four analogous hypotheses about the respective causal contribution of the physical properties P and chemical properties C of the ruby crystal in the production of laser light P*:

1. C and P together constitute a sufficient cause of P*;
2. C and P are two distinct sufficient causes of P*, which overdetermine P*;
3. P causes P* via C;
   1. P causes P* by causing C;
   2. P causes P* by determining in a non-causal manner property C that causes P* by virtue of a chemical-physical causal law;
4. P causes P* directly, without any causal contribution from C.

Before examining these hypotheses, it is important to note that the production of monochromatic light by the ruby laser is not perfectly analogous to the process of mental causation that I analyzed above. The effect of the laser — the ray of monochromatic light — is a purely physical phenomenon in the sense that the events that constitute it have no chemical properties. This makes a distinction analogous to the one that I made in my interpretation of hypothesis (1) inappropriate; the interpretation of (1), according to which P and C act as two factors neither of which alone is sufficient to produce P*, but which produce P* jointly, does not hold here. In fact, in the case of the laser, there is no chemical law C → C* analogous to the psychological law M → M*, the consequence of which constitutes a constraint that limits the degrees of freedom of the effect P*.

The only possible interpretation of (1) is that P determines C non-causally and that C determines P* causally by virtue of a law C → P*. Therefore, the only sense in which P and C together constitute a sufficient condition for P* is that P is nomologically sufficient under the circumstances for P*, thanks to the intermediate determination of C.

Assumption (3a) is inappropriate because the determination of C by P is non-causal. In particular, there is no time lag between the instantiation of C by the crystal and that of P by all of its atomic components.

For reasons already explained, it is empirically possible that hypotheses (2) and (4) are also inappropriate. This is the case if there is no causal law P → P* at the level of atomic physics that would determine the effect P* directly, without passing through the chemical level C. In this case, the effect P* is not directly determined causally at the atomic level.

The concept of deterministic chaos allows us to understand the possibility that the evolution of a purely material system, subject only to deterministic laws (this is, I suppose, the case of the human nervous system), might be such that, if its microscopic state P* at t* is given with some finite precision, then there is some time t earlier than t* such that there is no knowable-in-principle microstate P of the system at t causally responsible for P*.

We can make the hypothesis that the brain is chaotic at the level of the properties of its neuronal components. However, the configuration of the activity of neurons and their connections determines — in a non-causal way — a global property (in reality many insofar as there are specialized cognitive modules) that is mental. This property evolves according to psychological laws. Accordingly, the fact that the cognitive system possesses certain mental properties can be causally responsible for cognitive or behavioural facts. In my example, my thought that the noise in the street is disturbing my concentration (M) leads me to make the decision to close the window (M*). More precisely, the fact that I have the property M at t is causally responsible for the fact that I have the mental property M* at t*. A causal law determines the evolution of the system as a function of a global property. However, if the state of the set of all neurons is given with some finite precision, then for some t* later than t the lawful evolution of the neurons does not determine their state at t* with the same precision. The state of the set of neurons at t does determine their state at t*, but this state is determined — at least in part — by the overall psychological property (or properties). In a similar way, it is indeed the set of atoms that makes up the ruby crystal that determines (and indirectly causes), in favourable circumstances, the emission of laser light, but these atoms are not directly causally responsible for this effect. They act in this way only by virtue of their interaction, which produces an overall property of the crystal: the structure of the energy levels of the Cr³⁺ ions. It is this overall property that regularly and causally determines the effect of the laser light emission, according to a causal law. In both cases, there is downward causation because a global property of the entire system causally determines a subsequent state of the system, which is situated at the level of the properties of the components (as in the case of the laser) or which constrains the system at the level of the properties of its components (as in the case of mental causation).

This suggestion bears only a superficial resemblance to a conception of downward causation suggested by Popper. In his words, “the randomness of the movements of the elementary particles — often called ‘molecular chaos’ — provides, as it were, the opening for the higher-level structure to interfere. A random movement is accepted when it fits into the higher-level structure; otherwise it is rejected” (1977, 348). According to this suggestion, chaos would lead to the exploration of different types of evolution, only some of which conform to higher-level laws. The non-conforming courses of events would be eliminated, in an analogy with living beings resulting from harmful genetic mutations. However, this “downward causation” brought about by natural selection differs from the “downward causation” that I have sketched above. In the case of the laser — and, as I assume, in the case of mental causation — the constraint imposed on the lower-level property by the higher-level property is immediate and deterministic. The application of Popper’s model of downward causation to physical causation would imply the existence of contradictory situations in nature: for a certain period of time (i.e., before they are eliminated), there would be molecular movements that contradict higher-level laws.

Another reason for rejecting the analogy with natural selection is the fact that the selective forces at work in natural selection do not operate directly in a downward fashion. An organism poorly adapted to its environment is not breaking any laws. The forces by which the environment eliminates poorly adapted organisms are physical. It is only a question of downward causality in the sense that the presence of certain types of individuals can be explained only by higher-level laws. Campbell (1974) offers the example of the jaws of soldier termites, so large that they make these termites unable to feed. Their presence can be explained only by appealing to the laws of social organization through the division of labour in social species. However, this is a long-term explanation that does not imply any downward physical causation in the short term. The selection model cannot be transferred to downward causation. Microscopic movements have no “elbow room” to try out, even for short periods, movements that violate higher-level laws.

(3b) therefore appears to be the correct analysis. The “problem of causal-explanatory explanation” does not arise because the causal determination of laser light operates only in a downward manner. Similarly, that problem does not arise in the case of mental causation. In the absence of laws that directly determine the neuronal state P* as a function of the previous neuronal state P, the causal determination of P* is partly downward. The state of the set of neurons and the state of their relationships (above all their synaptic relationships) P do not directly determine the next state, P*, of the same set of neurons; they determine it only through the intermediary of the determination of a global mental property M of the whole organism.

It is true that M “inherits” in a sense its causal powers from the underlying properties P that determine its existence. However, the metaphor of inheritance is itself misleading, and Kim is wrong to interpret it literally. For him, the “causal inheritance principle” (1992a; 1993c, 208; 1998, 54) implies that the donor has a property, with its causal powers, at the physical level. This property is transmitted unchanged to the heir (the mental level). At the mental level, we find the same physical property, now referred to with mental concepts, which has conserved its physical causal powers (Kim 2002, 674). According to my model, a better analogy to the influence that the physical level exerts on the mental level is the transmission of ideas during a conference. The speaker’s words do determine the listeners’ ideas. However, aside from cases in which the listener memorizes the speaker’s words identically, the listeners’ ideas are not identical to the speaker’s ideas.

Kim’s assertion that “higher states are to inherit their causal powers from the underlying states that realise them” (1993c, 208) might be true if interpreted in the sense of cultural inheritance: where the heir receives something different from what is transmitted by the donor. Kim’s “causal inheritance principle” seems to be plausible only in the context of a conception of mental properties according to which they are physical properties conceived of by means of second-order concepts, which contain a quantification over these physical properties. Insofar as we construe the relationship between P and M as one of nomological determination of a property M of the organism by different properties of its parts — in an analogy to the determination of C by P in the case of the laser — it is not true that “the causal powers of this instance of M are identical with (perhaps, a subset of) the causal powers of P” (Kim 1993c, 208). Neurophysiological properties P, situated at the level of neurons and their relationships (in particular, synaptic relationships), do not belong to objects at the same level as mental properties. The relationship between the parts and the whole is a matter of (non-causal) nomological determination between different properties, which is not the same as realization, a relationship between different predicates referring to the same properties.

5. Conclusion

We have the intuition that our decisions, through our actions, can change the course of events. Philosophy alone cannot determine whether this is justified or illusory. Concepts alone cannot establish that the mind has causal powers of its own. But philosophy can determine whether this is at least conceivable. Its role is to map out the conceptual terrain and indicate the questions that only science can answer. Chapters 3 and 4 elaborated a conception of emergent systemic properties that can be applied to mental properties. In this framework, representations and decisions — alongside other mental states, processes, and events — can be construed as emergent properties that, though determined by underlying physical properties, have powers of their own. The mind is not just an epiphenomenon of the brain, as seems to follow from the functionalist conception of mental properties. It is true that many cognitive concepts are functional: that is, their logical form is second-order because it contains an existential quantification. It does not follow, however, that the only efficacious properties are physical: the first-order properties over which the second-order functional concepts quantify can be emergent properties, in particular cognitive properties.

However, there are reasons to doubt that cognitive properties can be efficacious, even if we accept the idea that they are first-order properties. Physicalism — which I have accepted as the framework for my inquiry — implies that all real states, processes, or events are ultimately determined by physical states of affairs. True, if mental properties are first-order, then their efficacy is no more doubtful than the efficacy of emergent non-mental properties, such as chemical or biological ones. But the physicalist doctrine according to which all properties ultimately are determined by the physical properties of elementary particles seems to challenge the very intelligibility of the causal efficacy of non-physical properties. However, we have seen that physicalism is compatible with a conception of complex systems in which their non-physical properties contribute to determine their evolution; therefore, such properties have causal powers of their own. Of course, philosophy can only establish that this is rationally conceivable. It is up to the empirical sciences to establish that this or that particular emergent property is real and subject to laws of nature.

The scenario that I have developed is as follows. The determination imposed on a complex system by microscopic laws governing the behaviour of its components is articulated with macroscopic laws. Complex systems have emergent properties subject to laws that constrain the system as a whole. However, these systemic laws do not determine its evolution at the level of its microscopic components. The evolution of the system at the microscopic level is the result of a double determination: systemic laws impose constraints limiting the possibilities of evolution of the components. Within the framework of these constraints, their evolution is determined by microscopic laws. For example, the psychological law discovered by Rescorla and Wagner determines the progress of learning by classical conditioning at the level of the cognitive system. Like other laws in the special sciences, psychological laws are ceteris paribus laws that apply only under very specific conditions. This law does not apply in exceptional circumstances; for example, brain trauma interrupts learning by causing retrograde amnesia. Under normal conditions, however, all animals capable of conditioning evolve in such a way that their behavioural dispositions obey Rescorla and Wagner’s law of conditioning. However, this law does not determine the details of the modification and formation of synapses at the physiological and molecular levels. Physiological, chemical, and physical laws determine the precise sequence of microscopic changes that give rise to learning at the systemic level. Similarly, thermodynamic or hydrodynamic laws determine the evolution of liquid or gaseous bodies at the macroscopic level without determining the microscopic evolution of their components. The detailed evolution of these components is determined by physical and chemical laws within the constraints fixed by macroscopic laws.

This model shows how an emergent property can exert a causal power of its own. It can contribute to determine the evolution of the system without calling into question either the principle of causal closure of the physical domain or the principle of causal-explanatory explanation. Let us consider the latter principle first. If system s has an emergent property G* at t*, then there are generally two ways of fully explaining G*. G* can be the conclusion of two kinds of deductive-nomological arguments. G* can be explained in a non-causal way, on the basis of laws of composition and microstructural properties P* that the system possesses at t*, and G* can be explained causally, by dynamic laws based on the properties that the system possesses at t, some time earlier. In accordance with the principle of causal-explanatory exclusion, there is only one complete causal explanation of G*. However, there is no reason to accept the stronger “principle of explanatory exclusion.” Even if the causal explanation of G* on the basis of the state of the system at t is complete, there can be another complete explanation of G* that is non-causal. This is the explanation of G* by the microscopic state P* of the system at time t*.

The existence of emergent properties, sources of downward causal determination, might also seem to be incompatible with the physicalist principle of the causal closure of the physical domain. According to this principle, the physical state P* of any system at a given instant t* is determined, with respect to any instant t prior to t*, by its physical state P at t. My model is compatible with this principle. Let us assume that P* is causally determined partly by emergent properties G (and macroscopic laws) that the system possesses at t and partly by microscopic properties P that it possesses at t. Even if G makes an essential contribution to the causal determination of P* (and G*), ultimately it is the physical properties P of the system at t that determine P* (and G*). P directly determines, in a non-causal way, the emergent properties G, which then contribute — in the way indicated above — to determining the state of the system at t*. It therefore appears that physicalism, which requires all determination ultimately to be physical, is compatible with the existence of emergent properties having causal powers of their own. The state of some complex systems is only indirectly determined causally by their previous physical state: causal determination is achieved through emergent properties and systemic laws constraining the evolution of these emergent properties.