by Ranpal Singh Dosanjh - University of Toronto T …...Abstract A Defense of Reductive Physicalism Ranpal Singh Dosanjh Doctor of Philosophy Graduate Department of Philosophy University

A Defense of Reductive Physicalism

by

Ranpal Singh Dosanjh

A thesis submitted in conformity with the requirements

for the degree of Doctor of Philosophy

Graduate Department of Philosophy

University of Toronto

c© Copyright 2014 by Ranpal Singh Dosanjh

Abstract

A Defense of Reductive Physicalism

Ranpal Singh Dosanjh

Doctor of Philosophy

Graduate Department of Philosophy

University of Toronto

2014

In this dissertation, I defend reductive physicalism, according to which (in particular) mental proper-

ties are type-identical to physical properties. I use a burden of proof strategy to defend the position,

established by appealing to scientific practice, methodological considerations, and analogies between

higher-level properties and paradigm cases of reduction. This strategy allows various non-reductive and

anti-physicalist positions to be explored in the context of arguments presented in their favour.

Non-reductive physicalists argue that some higher-level properties are not type-identical to any phys-

ical properties. Perhaps the principal argument along these lines is the Multiple Realizability Argument,

while a common reductivist reply appeals to the Disjunctive Strategy. I establish a framework for un-

derstanding this dialectic that reveals the logical space available to non-reductive physicalists, allowing

me to make comprehensive rejoinders to the best available responses to the Disjunctive Strategy. For

example, according to a recent, powers-based response, a higher-level property instance has a proper

subset of the causal powers of its realizer on the occasion, making the higher-level property instance

ontologically distinct and causally autonomous from its realizer. Using my framework, I argue that the

higher-level property is not token-distinct from its realizer. Furthermore, even if it were, its supposed

causal autonomy is just a special case of the counterfactual stability of disjunctions.

Anti-physicalists argue that some properties—specifically phenomenal properties—are “over and

above” physical properties, as per some strong form of property dualism. The main anti-physicalist

arguments are the Knowledge and Conceivability Arguments. I argue that, perhaps surprisingly, these

arguments actually exclude popular versions of strong emergentism (those according to which phenom-

enal properties are lawfully, downwardly causal), or at least render them less plausible than physicalist

counterparts.

Strong emergentism therefore relies mostly on its bare empirical possibility for support, which makes

it resistant to a priori philosophical counterarguments. However, philosophical analysis can be used to

show that strong emergentism requires the existence of an emergent causal law, which itself requires a

discontinuity in some effect as a function of complexity. I argue that the existence of such a discontinuity

is empirically very unlikely given what is already known about neural systems.

ii

Acknowledgements

I extend sincere thanks to my supervisor, Jessica Wilson, and my committee members, Bill Seager and

Diana Raffman, for philosophical guidance and practical advice; to my friends and colleagues for personal

and philosophical interactions; and to my wife, Geetu, and the rest of my family for love and support.

iii

Contents

1 Introduction 1

1.1 Preliminary Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.1.1 Broadly Scientific Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.1.2 Physicalism and Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.1.3 Physicality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.1.4 Being Nothing Over and Above . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.1.5 Ontological Innocence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.1.6 Non-reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.2 The Plausibility of Reductive Physicalism . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.2.1 Physicalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.2.2 Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2 Against Non-reductive Physicalism 19

2.1 Multiple Realizability and the Disjunctive Strategy . . . . . . . . . . . . . . . . . . . . . . 20

2.2 Denying Type-Identity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.2.1 Difference in Kindhood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

2.2.2 No Difference in Kindhood . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.2.3 Token-Distinctness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.2.4 Denying the Argument from Token-Distinctness . . . . . . . . . . . . . . . . . . . . 27

2.3 Denying Ontological Innocence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2.3.1 Argument Against Ontological Innocence . . . . . . . . . . . . . . . . . . . . . . . 33

2.3.2 Denying the Argument Against Ontological Innocence . . . . . . . . . . . . . . . . 34

2.4 Denying Status as Physical . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.4.1 Fundamentally Non-physical Realizers . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.4.2 Empirical Science and Physical Realizers . . . . . . . . . . . . . . . . . . . . . . . 47

2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

3 Anti-physicalism and the Physical 52

3.1 Physicality Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.1.1 Physical and Paraphysical Properties . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.1.2 Physically and Paraphysically Acceptable Properties . . . . . . . . . . . . . . . . . 54

3.1.3 Stray Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.1.4 Physicalism and Paraphysicalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.1.5 Physical and Paraphysical Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.1.6 Deep Physical-Phenomenal Unity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

iv

3.2 The Knowledge Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.2.1 Overview of the Knowledge Argument . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.2.2 The Knowledge Argument and Physicality . . . . . . . . . . . . . . . . . . . . . . . 61

3.3 The Conceivability Argument . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.3.1 Overview of the Conceivability Argument . . . . . . . . . . . . . . . . . . . . . . . 64

3.3.2 The Conceivability Argument and Physicality . . . . . . . . . . . . . . . . . . . . . 67

3.4 Lawfully Interacting Strong Dualism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.4.1 Varieties of Dualism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.4.2 Interactionist Dualism and Stray Properties . . . . . . . . . . . . . . . . . . . . . . 74

3.4.3 Interactionist Dualism, p-Paraphysicalism, and DUx-p . . . . . . . . . . . . . . . . 75

3.4.4 Interactionist Dualism and Anti-physicalist Arguments . . . . . . . . . . . . . . . . 75

4 Characterizing Strong Emergentism 79

4.1 Problems with Supervenience-Based Accounts . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.2 Characterizing Strong Emergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.2.1 Biased and Unbiased Modal Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.2.2 Counterlegal Reasoning and Strong Emergence . . . . . . . . . . . . . . . . . . . . 85

4.2.3 Strong Emergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.3 Supervenience-Based Accounts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.3.1 Van Cleve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

4.3.2 O’Connor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4.3.3 McLaughlin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4.3.4 Seager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

4.4 Real Operation Accounts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

4.4.1 O’Connor and Wong . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.4.2 Humphreys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

4.5 Powers-Based Accounts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4.5.1 Wilson . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

4.5.2 Shoemaker . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.6 Emergent Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4.6.1 Pepper and Emergent Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.6.2 Meehl and Sellars: Emergent Qualities . . . . . . . . . . . . . . . . . . . . . . . . . 98

4.6.3 Emergent Laws and Discontinuities . . . . . . . . . . . . . . . . . . . . . . . . . . . 98

5 Evidence Against Strong Emergentism 101

5.1 Empirical Character of Emergent Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

5.1.1 Emergence Engendering and Manifesting Systems . . . . . . . . . . . . . . . . . . 102

5.1.2 Levels of Complexity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

5.2 Empirical Evidence Against Strong Emergentism . . . . . . . . . . . . . . . . . . . . . . . 106

5.2.1 Behavioural Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

5.2.2 Evidence from Physical Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . 110

5.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

Bibliography 112

v

Chapter 1

Introduction

Suppose that scientists discover physical (e.g., neural) properties that correlate with mental properties

on a type-by-type basis, perhaps many physical types to a mental type. So far, the metaphysical

status of such mental properties would be an open question. In this dissertation, I defend reductive

physicalism, according to which mental properties just are physical properties, or some ontologically

innocent combination of physical properties.1 I use a burden of proof strategy to defend that position. In

the absence of good arguments to the contrary, I assume it is reasonable to believe that both physicalism

and reductivism are true.2 I implement this strategy by appeal to (a) the role of physicalism and

reduction in scientific practice and explanations; (b) methodological considerations involving ontological

parsimony and physical plenitude; and (c) analogies between higher-level properties and paradigm cases

of reduction. The advantage of this strategy is that it allows various non-reductive and anti-physicalist

positions to be explored in the context of arguments presented in their favour.

The first (and most serious) alternative position to reductive physicalism is non-reductive physical-

ism.3 Non-reductive physicalists argue that, although higher-level properties like mental properties (and

special science properties in general) are “nothing over and above” the physical, at least some such prop-

erties fail to be type-identical to any ontologically innocent combination of physical properties. Perhaps

the principal argument along these lines is the Multiple Realizability Argument, according to which

higher-level properties are realized by very different physical properties, so that higher-level properties

cannot be type-identical to any (one) physical property. A common reductivist reply appeals to the

Disjunctive Strategy, according to which the higher-level property is type-identical to the disjunction of

its realizers.

In what follows, I will defend the Disjunctive Strategy.4 Non-reductive physicalists can respond to

this strategy by making one of three claims: (1) higher-level properties are not type-identical to the

disjunctions of their realizers; (2) type-identity to a disjunction of physical properties is not adequate for

1Some form of reductive physicalism has been advocated by numerous philosophers, including Place (1956), Feigl (1958),Smart (1959), Lewis (1966, 1972), Armstrong (1968), Kim (1998, 2005), and Perry (2001).

2Perry (2001) uses this strategy, which he calls “antecedent physicalism” (p.26ff), to defend physicalism against variousanti-physicalist arguments. Since I am also concerned with non-reductive physicalist arguments, I require antecedentreductivism as well.

3Some form of non-reductive physicalism has been advocated by even more numerous philosophers, including Putnam(1967), Davidson (1970), Fodor (1974), LePore and Loewer (1989), Shoemaker (2001, 2007), and Pereboom (2002).

4The Disjunctive Strategy is rejected by non-reductive physicalists on various grounds. One popular rejection appealsto the fact that the relevant disjunctions are open-ended, or have an indefinite or infinite number of disjuncts, which makesthem unsuitable for type-identity. See, for example, LePore and Loewer (1989) and Pereboom and Kornblith (1991).Against this view, see Fodor (1997) and Jaworski (2002). Below, I present a framework for understanding and categorizingresponses to the Disjunctive Strategy.

1

Chapter 1. Introduction 2

reduction; and (3) higher-level properties have fundamentally non-physical realizers, so the disjunctions

to which they are type-identical outstrip the physical. I will argue that none of the arguments for

these claims succeeds. Against the first response, I will argue that higher-level properties and the

disjunctions of their realizers are relevantly similar in those respects that are typically taken to indicate

their distinctness. Against the second response, I will argue that type-identity to disjunctions is always

adequate for reduction, and that appearances otherwise are merely a result of our categorizing practices.

Against the third response, I will argue that the properties of interest in empirical sciences do not

outstrip their physical realizers.

On the other hand, according to anti-physicalists, some properties—in particular phenomenal prop-

erties—are “over and above” physical properties, as per some strong form of property dualism. The main

anti-physicalist arguments are the Knowledge Argument and the Conceivability Argument. According to

the Knowledge Argument,5 a scientist who learns all the physical facts while never having experienced

colour would learn something new upon experiencing colour for the first time; so phenomenal colour

properties (and thus phenomenal properties in general) cannot be physical properties. According to

the Conceivability Argument,6 our ability to conceive of a world that is physically identical with but

phenomenally distinct from ours entails the possibility of such a world, which implies that phenomenal

properties cannot be necessitated by physical properties, much less that the former are nothing over and

above the latter.

I argue that, perhaps surprisingly, the Knowledge Argument and the Conceivability Argument ac-

tually rule out some forms of dualism, namely those according to which phenomenal properties are

lawfully, downwardly causal. The traditional division of dualist accounts of mental properties rests on

whether the mental properties in question interact with the world. According to some accounts, at

least some mental properties are directly and independently efficacious, just as physical properties are

typically efficacious. For example, mental properties may exhibit so-called downward causation; they

arise from particular physical configurations, but they (rather than the physical configurations) are the

properties that are efficacious at the physical level. These are interactionist dualist accounts. According

to other dualist accounts, mental properties either fail to be efficacious at all, or fail to be efficacious at

the physical level. Examples include epiphenomenalism and parallelism.7 These are non-interactionist

dualist accounts.8

I argue that there is an important and underappreciated distinction to be made within the inter-

actionist variety of dualism: those dualist accounts according to which mental properties are lawfully

causal, and those according to which they are anomalously causal. Lawfully causal dualist accounts, I

will argue, are ruled out by the Knowledge Argument and the Conceivability Argument. Thus, such

accounts are orphaned, at least in terms of the strongest arguments typically taken to be in their favour.

Nevertheless, there remain some other (albeit less convincing) arguments in favour of particular

lawfully causal dualist accounts.9 I will therefore consider in detail another alternative to reductive

5The Knowledge Argument was originally formulated by Jackson (1982).6See Chalmers (2009).7Modern discussions of epiphenomenalism are typically taken to begin with Huxley (1874), while Leibniz’s pre-

established harmony account of mind and body is perhaps the most famous example of parallelism.8The distinction between interactionist and non-interactionist dualist accounts corresponds to Chalmers’ distinction

between type-D and type-E dualism. See, for example, Chalmers (2002a).9For example, consider the following line of argument. There is a psycho-physical monist view called (constitutive)

panpsychism, according to which mental properties are built up, so to speak, from the mental properties of their physicalcomponents in their physical configurations. This view suffers from the combination problem (James, 1890; Seager, 1995,1999; Seager and Allen-Hermanson, 2012; Chalmers, 1996, 2002a, 2012), since there appears to be no way that phenomenalproperties can combine in this way. Some lawfully causal strong property dualist accounts avoid the combination problem


physicalism, one that encompasses such accounts: strong emergentism.10 According to strong emergen-

tism, mental properties arise from a certain configuration of physical properties in a way that cannot be

explained by the underlying physical components and their interactions according to the laws of physics.

A new, non-physical, brute law of nature activates, instantiating the mental property and resulting in

the altered behaviour of the physical components. The key feature of strong emergentism is the existence

of a non-physical property that is nevertheless (lawfully) interactionist. The position therefore includes

lawfully causal dualist accounts. But it also includes epiphenomenal and parallelist accounts according

to which phenomenal properties merely correlate with those non-physical properties.

Strong emergentism is an empirical position, and while there is no uncontroversial evidence for it,

there is no conclusive empirical evidence against it either, so it remains an empirical possibility that (to

some) appears plausible in the unique case of the mental. This makes it resistant to a priori philosoph-

ical counterarguments. However, philosophical analysis can be used to show that strong emergentism

requires the existence of an emergent causal law, and that a necessary condition for the latter is a finite

discontinuity (as a function of complexity) in the interactions of the physical components of the system

instantiating the emergent property. I will then show that the existence of such a discontinuity is very

unlikely given what is already known about neural systems.

The remaining dualist positions are either non-interactionist dualist positions or anomalously causal

dualist positions. According to all such accounts, mental properties (phenomenal properties in partic-

ular) are fundamentally non-physical properties of physical things, and the correlations between the

mental properties and the physical properties are a consequence of brute psychophysical laws of nature.

According to non-interactionist property dualism, the properties are not physically efficacious. Accord-

ing to anomalously causal property dualism, the properties are physically efficacious in a non-lawful

way. These accounts are the only dualist accounts that are not ruled out by the Knowledge Argument

and the Conceivability Argument, but they are not without problems. For example, they imply either

that phenomenal properties are not efficacious, or that the laws of nature are not complete (i.e., that

the laws fail to govern every event). The possibility of mental properties being non-efficacious has a

long and contentious history in the Philosophy of Mind.11 And we have reasons to believe that it is

metaphysically impossible for the laws of nature to be incomplete (Lange, 2009). However, it will not

be possible to argue against such accounts within the scope of this document. I will content myself with

having defended reductive physicalism against its most commonly held rivals: non-reductive physicalism

and lawfully causal versions of dualism (along with strong emergentism).

In the remainder of this chapter, I will make some preliminary remarks about physicalism and

reduction, before making the case that, in the absence of arguments otherwise, it is reasonable to believe

reductive physicalism. In chapter 2, I will defend the Disjunctive Strategy against arguments for non-

reductive physicalism. In chapter 3, I will argue that the Knowledge Argument and the Conceivability

Argument exclude lawfully causal dualist positions. In chapter 4, I will argue that strong emergentism

requires a discontinuity in the behaviour of the physical parts of the system instantiating the emergent

property. In chapter 5, I will argue that, given what is already known about neural systems, such

discontinuities are not to be found, making strong emergentism empirically very unlikely.

by having the mental properties be nomologically necessitated by the combinations. Such accounts can hijack the argumentsin favour of constitutive panpsychism.

10See, for example, McLaughlin (1992), Van Cleve (1990), Wilson (2002), Shoemaker (2002), O’Connor and Wong (2005),and Seager (2012).

11An interesting Early Modern example of that is Elisabeth of Bohemia, who argued that Descartes’ supposedly inter-actionist substance dualism was not interactionist at all, since the mental, in lacking extension, could not cause physicalevents. See Elisabeth of Bohemia and Descartes (2007).


1.1 Preliminary Remarks

In this section, I shall attempt to clarify some of the terms I’ll be using and the assumptions I’ll be

making. First, I will briefly discuss the domain of discourse: broadly scientific properties. Then, I

will characterize physicalism, and explain how reductive physicalism, non-reductive physicalism, and

anti-physicalism can be characterized in light of it. The remainder of this section is devoted to making

more precise the contents of the characterization. I will describe what it means for a property to be

physical, for a property to be nothing over and above another, and for a combination of properties

to be ontologically innocent. Finally, I will discuss the merits of my characterization of non-reductive

physicalism.

1.1.1 Broadly Scientific Properties

The various accounts presented above are concerned with broadly scientific properties. They include

(but are not limited to) those properties that scientists discover and investigate in the natural and

social sciences. Among these are physical properties, chemical properties, biological properties, and

(importantly) mental properties, both representational and phenomenal. Also among them are, plausibly,

economic properties, political properties, and other social properties. Broadly scientific properties do not

include, for example, purely logical properties, purely mathematical properties, and (plausibly) ethical

properties.

1.1.2 Physicalism and Reduction

Schematically, physicalism is the position according to which all broadly scientific properties are nothing

over and above the physical.12,13 There are a host of concomitant commitments that a physicalist holds.

The two most important are as follows. The first is the causal closure of physical events.14 Physical

events (or the instantiation of physical properties) are causally closed if and only if all physical events

that have a cause have a physical cause. The second is the causal inheritance principle.15 According

to this principle, all broadly scientific properties (except fundamental physical properties) inherit their

causal powers from those properties that realize them on the occasion of their instantiation.16

Reductive physicalism is the position according to which all broadly scientific properties are not

just nothing over and above physical properties, but are (furthermore) type-identical to them. In what

follows, I will take for granted that ontologically innocent combinations of physical properties, like

mereological sums of particles or averages of determinate values of a physical variable, are also physical

in this narrow sense.17 Non-reductive physicalism, then, is the position according to which all broadly

scientific properties are nothing over and above the physical, but some are not type-identical to any

12The relation denoted by ‘nothing over and above’ will be described in more detail in section 1.1.4. For now, it isenough to understand the phrase in the ordinary, non-technical sense.

13It follows from this schematic presentation that physicalism is truth-evaluable, that is, it might be true or it might befalse. An argument that physicalism is not a truth-evaluable position, but instead is a commitment have one’s ontologyconform to physics, can be found in Ney (2008b).

14See, for example, Davidson (1970), Kim (1993, 1998, 2005), and Montero (2003).15See, for example, Kim (1992).16Realization is the relation (compatible with physicalism) in virtue of which a configuration of lower-level properties

instantiates a higher-level property. For example, the physical properties (like charge) of an arrangement of electrons andnucleons instantiate the chemical property of valence, so charge (etc.) in that configuration realizes valence. I will describethe realization relation in more detail in chapter 2.

17See section 1.1.5.


physical properties (not even to ontologically innocent combinations). Finally, strong emergentism and

the remaining dualist positions are anti-physicalist. According to them, some broadly scientific properties

(specifically, mental properties) are something over and above physical properties. Whether advocates

of such positions deny the physicalist’s concomitant commitments depends on the particular position

advocated.18

1.1.3 Physicality

Despite being in wide use, the term ‘physical’ can be regarded as problematic.19 I will use the term in

two ways.

One way, the more general way, refers to the properties posited by any possible physics, or the

final physics of any possible world (and any ontologically innocent combination of such properties).

Call this ‘p-physical’. The only constraint on p-physical properties is that they conform to our most

general conceptions of physical properties: they must be broadly scientific (Wilson, 2005, 2006), they

must be empirically accessible (Dowell, 2006b), and they must exclude fundamental mentality (Montero,

1999; Papineau, 2002; Montero and Papineau, 2005; Wilson, 2006). For example, the properties of

being ectoplasm (a generally anti-physical substance) and being a philosophical ghost (disembodied

fundamental mentality) are not p-physical. On the other hand, the property of being an orbital-model

atom (a stable atom in which electrons orbit the nucleus according to Coulomb’s law) is p-physical.

The other way, the less general way, refers to the properties posited by the final physics of the actual

world (again, plus ontologically innocent combinations of them).20 Call this ‘a-physical’. A-physical

properties thus have all the constraints of p-physical properties, and have the further constraint that

they be posited by the final physics of our world. So all a-physical properties are p-physical, but not

all p-physical properties are a-physical. To continue with the examples above, the properties of being

ectoplasm and being a philosophical ghost are a fortiori not a-physical properties. But the property of

being an orbital-model atom also fails to be a-physical, since such atoms would be forbidden by our final

physics.21

My treatment of these terms can be generalized as follows. Let w be a variable representing a

metaphysically possible world. A property is w-physical if it is a property posited by the final physics

of world w, or if it is an ontologically innocent combination of w-physical properties.22 A property is

p-physical if and only if there is some world w such that the property is w-physical.23 An a-physical

property is w-physical for the actual world.

The distinction between p-physical and a-physical properties (and the subsequent generalization into

w-physical properties) should help us sort the anti-reductive and anti-physicalist positions. Call a broadly

scientific property w-physically acceptable if it is nothing over and above the w-physical properties. World

w is physicalist if all broadly scientific properties in w are w-physically acceptable. For the actual world,

this means that all broadly scientific properties are a-physically acceptable.

The sense of physical that is relevant to reduction is also a-physical. Call a w-physically acceptable

18For example, plausibly, strong emergentists deny both the causal closure of physical events and the causal inheritanceprinciple. On the other hand, advocates of non-interactionist dualism deny neither.

19See, for example, Wilson (2006) and Ney (2008a).20Compare to futurist formulations of physicalism in, for example, Loewer (2001) and Dowell (2006a). See Ney (2008a).21Classically, the centripetally accelerating electron would emit synchrotron radiation at a rate that would degrade its

orbit almost instantaneously. Quantum mechanically, the electron cannot have such a well defined orbit.22Compare Dowell (2006b).23For some world w, a p-physical property is w-physical.


property merely w-physically acceptable if it is w-physically acceptable without being type-identical to

any w-physical property. Reductive physicalism is true in w when all broadly scientific properties in w

are w-physical properties; i.e., there are no merely w-physically acceptable properties. For the actual

world, reductive physicalism is true when all broadly scientific properties are a-physical, so that none

are merely a-physically acceptable. This makes sense, as type-identity to the physical properties of some

(other) possible world ought not count as reduction in the actual world. Choosing a-physical properties

as the ones to which other properties are to be reduced also conforms to historical usage; for example,

Nagel’s (1961) account of theory-to-theory reduction requires a particular physical theory to which to

reduce (rather than any physical theory), and in discussions about the truth of reductive physicalism,

the theory in question is the final physics of the actual world. On the other hand, if a higher-level

property could have (perhaps p-physical) realizers that were not a-physical (in a-physically impossible

worlds), then it wouldn’t be type-identical to an a-physical property. The higher-level property would

have a causal profile too general to be exhausted by the laws of physics. It would be merely a-physically

acceptable, which is sufficient for a non-reductive conclusion.

The two classes of anti-physicalist positions can be sorted according to the p-physicality of the

properties in question. Suppose an actual instance of a higher-level property was p-physically acceptable

but not a-physically acceptable. Not being a-physically acceptable, the property would be genuinely

novel (it would not be posited by the final physics of the actual world, nor necessitated by any such

properties). But being p-physically acceptable, it would also be on par with a-physical properties in

terms of ontology and causal efficacy. This fits the description of an emergent property: a novel property

(inexplicable with respect to the interactions of the constituents) that causes its constituents to behave

in ways for which the (a-)physical theory cannot account. An actual orbital-model atom, for example,

would be inexplicable with current (and therefore likely with final) physics, and would thus be genuinely

novel. Strong emergentism entails the existence of properties such as these.24

On the other hand, suppose the property is nothing like p-physically acceptable. This leaves only

properties that fail to interact with the physical world in ways typical of a-physically acceptable prop-

erties. Such properties are compatible with anomalously causal dualism and non-interactionist dualism.

Hempel’s Dilemma

By relying on final physics to capture the sense of a-physical and physicalism, I leave myself open to

Hempel’s dilemma (Hempel, 1969).25 Any notion of physicalism will need to be based on some notion of

the physical, which in turn is usually based on some theory of physics. According to Hempel, our notion

of the physical ought to be such that (1) physicalism is not obviously false, and (2) it must have content

that can be, to some substantive extent, determined by us now. The dilemma is that we must base our

notion of the physical on either current physics (which violates the first desideratum) or on a future, final

physics (which violates the second desideratum). Suppose we decide to base our notion of the physical

on current physics. Plausibly, current physics is not the final physics of our world. Therefore, strictly

speaking, current physics is not true of our world. We are thus gored by the first horn of Hempel’s

dilemma: if current physics is false, then physicalism ultimately grounded in current physics becomes

obviously false. The true, final physics of the actual world is the only theory upon which to base our

notion of the physical that will not result in physicalism being obviously false. So suppose we base our

24See section 3.4.25Treatments of the impact of Hempel’s dilemma on characterizing the physical can be found in Wilson (2006) and Ney

(2008a,b).


notion of the physical on final physics. Then physicalism is not obviously false. But we do not know

(completely) the final physics. We are thus gored by the second horn of Hempel’s dilemma: if the final

physics is not known to us, then our notion of the physical is not well determined, and so neither is the

content of physicalism.

My account obviously falls afoul of the second horn. As stated above, I’m taking physicalism to be

true (of the actual world) when all broadly scientific properties are nothing over and above the a-physical,

where the a-physical properties are those properties posited by the final physics of the actual world (and

ontologically innocent combinations of them). I don’t think this is fatal, however. We may not be able

to determine (in the strongest sense of determine) the content of physicalism if it is defined against final

physics. But, based on its extraordinary empirical success, current physics gives us a very good idea

about what the a-physical properties will be, and therefore what a physicalist world will look like. This

assumption is about as reliable as current physics is in describing the physical world; in other words,

the difference between applying the conclusions of current physics and the conclusions of final physics

to the problems of metaphysics is terribly small in areas other than where our knowledge of physics is

incomplete (as in quantum gravity).26

It’s instructive to try to generalize Hempel’s dilemma. As described above, physicalism is true at

world w when all broadly scientific properties in w are nothing over and above the w-physical properties,

where w-physicality is defined against the final physics of w. Hempel’s dilemma would apply to any

creature in w whose epistemic community had not yet developed their final physics. But for us, in

contexts relevant to discussions of physicalism in non-actual worlds, we would typically pick out w by

specifying all the facts of w, including its final physics. In that case, we avoid the second horn of Hempel’s

dilemma. But this only works if we have some idea of the constraints on what could be the final physics

of some world. This is why we must have a prior notion of what sorts of properties can be physical in

the broadest sense.27 It also provides us with a tight circle of concepts. The set of p-physical properties

is the union of the sets of the w-physical properties of all worlds w. The w-physical properties are those

posited by the final physics of w (and ontologically innocent combinations of them). And the universal

constraint on the content of the final physics of any world is that the properties be broadly physical

(i.e., p-physical).

This circle need not be worrisome. The last leg of the circle, involving a universal constraint, is an

epistemic issue. When specifying all the facts about a world w, we know what to include and what

to exclude from the final physics of w with some knowledge of the physical in broadest sense. It also

illuminates the epistemic nature of Hempel’s dilemma. From a position of uncertainty about the details

of the final physics of some world, knowing that all the properties are broadly physically acceptable, i.e.,

p-physically acceptable, is not enough to conclude that physicalism is true of that world. The properties

of world w have to be w-physically acceptable.28 This is why an epistemic appeal to current physics

makes Hempel’s dilemma less problematic.

In light of this, I will occasionally use ‘physical’ to refer to the properties posited by current physics.

But this should be accepted as a best guess of the properties posited by the final physics. Also, for

convenience, I’ll likely use ‘physical’ in place of either ‘a-physical’ or ‘p-physical’, with the context

disambiguating between them.

26This is compatible with a major revision of physics having serious consequences for metaphysics, and with the finalphysics being an incomplete description of broadly scientific properties. But the properties and laws posited by currentphysics are likely very similar to the properties and laws posited by the final physics, at least within well confirmed areasof physics.

27This is the take-away lesson from Wilson (2006). See also Dowell (2006b).28Recall that an emergent property in w can be p-physically acceptable without being w-physically acceptable.


1.1.4 Being Nothing Over and Above

Physicalist worlds are those in which all broadly scientific properties are nothing over and above the

physical. Having discussed in some depth what constitutes physicality, it is worth spending some time

looking at being nothing over and above. For much of this dissertation, it is enough to understand

‘nothing over and above’ in the ordinary, non-technical sense. Below I shall elaborate on the notion for

those occasions where more is required.

Consider the role that ‘nothing over and above’ plays in the characterization of physicalism. Being

nothing over and above the physical is what allows properties that may or may not be type-identical to

physical properties to be compatible with a physicalist world. There are, therefore, two constraints on

being nothing over and above. First, being type-identical has to be one way of being nothing over and

above, but not the only way. A property is always nothing over and above itself, but (at least prima facie)

some properties are nothing over and above properties to which they are not type-identical. Second, the

existence of properties that are nothing over and above the physical ought not entail anti-physicalism,

but the existence of properties that fail to be nothing over and above the physical (that are something

over and above the physical) ought to entail anti-physicalism.

Intuitively, and loosely, for one property to be nothing over and above another is for the former to

be necessitated in some way by the latter. Type-identity is but one way for a property to be necessi-

tated by another, so necessity-like relations satisfy the first constraint. Furthermore, it is prima facie

plausible that in physicalist worlds all properties are necessitated by physical properties, and that in

anti-physicalist worlds there are properties not necessitated by physical properties.

The most common way to cast this necessity-like relation is in terms of supervenience.29 One set

of properties supervenes on another set when a change in the former entails a change in the latter.

Type-identity is a species of supervenience, but it is merely a limiting case, so the first constraint is

satisfied. Furthermore, properties that fail to supervene on physical properties (say, with nomological

necessity) seem like those properties that make a world anti-physicalist. It would appear that filling in

some details (like the modal force of the relation) is all that is required to provide an account of being

nothing over and above.

Though common in the literature, supervenience-based approaches to being nothing over and above

have serious problems (Wilson, 2005). In particular, such approaches fail to distinguish physicalist worlds

from strongly emergent worlds. All such accounts take for granted the contingency of physical laws, and

that separating physical laws from anti-physical laws is metaphysically sound. However, suppose we

live in a strongly emergent world. If properties are individuated by their relationship to the laws of

nature (property-law necessitarianism) and if physical laws cannot be separated from whatever laws

govern the emergent properties (holism about laws), then strongly emergent properties supervene on

physical properties with metaphysical necessity. This would make them nothing over and above physical

properties on all realistic supervenience-based accounts.

Nevertheless, the supervenience-based approach does have some merits. For technical discussions, it is

easier to deal with supervenience (however inadequate) than an unanalyzed relation of being nothing over

and above. Also, for most purposes internal to the reduction/non-reduction debate, where physicalism

is taken for granted, a supervenience-based approach can be used. If physicalism is true, then all

properties supervene (with the appropriate modal force) on physical properties, which puts a constraint

on which relations between physical properties and special science properties are physically acceptable.

29For a survey of attempts at this, see Wilson (2005).


Finally, though it would be better to be neutral to the issues of property-law necessitarianism and holism

about laws, so long as it is meaningful to talk about properties apart from laws, and laws apart from one

another, some sense can be made of being nothing over and above. Therefore, I will propose a provisional

supervenience-based account of being nothing over and above. I will then revisit the problems with the

account.30

My provisional proposal is as follows.

Strong, physical supervenience as nothing over and above: P -properties are nothing over and above Q-

properties iff P strongly supervenes on Q with physical necessity.

The modal force of the supervenience relation in this definition is physical necessity. Logical/metaphysi-

cal necessity is arguably too similar to type-identity to provide a suitable modal force for a supervenience-

based account of being nothing over and above.31 On the other hand, supposing that we can meaningfully

contrast nomological necessity and physical necessity, nomological necessity is too weak to support

physicalism. This is because, relative to any world with strongly emergent properties, the (brute, non-

physical) laws that govern those properties hold in all nomologically possible worlds. Strongly emergent

properties therefore nomologically supervene on physical properties.

The proposal also incorporates a strength to the supervenience relation: the one set of properties

strongly (rather than weakly) supervenes on the other. We can understand weak supervenience as a

relation that holds among property-instances within a possible world.32 Weak supervenience holds when

all things that instantiate a P -property instantiate some Q-property, and, within a world, all things that

instantiate the same Q-property instantiate the same P -property. In contrast, strong supervenience is

a relation that holds among property-instances across possible worlds.33 Strong supervenience holds

when all things that instantiate a P -property instantiate some Q-property, and, for any (physically)

possible worlds, all things that instantiate the same Q-property instantiate the same P -property. Strong

supervenience is clearly more suited than weak supervenience to characterize physicalism. Different P -Q

relations obtaining in different physically possible worlds (owing, perhaps, to a brute, non-physical law

relating the two) are compatible with a world not being physicalist.

For now, I will assume that it is meaningful to talk about physical laws as distinct from any emergent

laws there might be. I will maintain this assumption throughout my discussion of reductive and non-

reductive physicalism. However, I will revisit the assumption when talking about strong emergentism in

chapter 4. In particular, I will take seriously the problem of property-law necessitarianism and holism

about laws in the context of characterizing strong emergentism.

1.1.5 Ontological Innocence

According to reductive physicalism, and in contrast to non-reductive physicalism, all broadly scientific

properties are type-identical to physical properties, where ontologically innocent combinations of physical

properties are themselves physical properties. It is worth taking a closer look at the notion of ontological

innocence, and how it fits into the overall strategy of my argument.

30See chapter 4.31See, for example, Witmer (2001, p.436).32P -properties weakly supervene on Q-properties iff necessarily, if anything has a P -property then there is a Q-property

that it has, and everything that has that Q-property has that P -property.33P -properties strongly supervene on Q-properties iff necessarily, if anything has a P -property then there is a Q-property

that it has, and necessarily everything that has that Q-property has that P -property.


Recall that I intend to establish that the burden of proof is on those who advocate non-physicalist

and non-reductive positions. Furthermore, my approach is to expand the class of properties to which

we (as reductive physicalists) expect broadly scientific properties to reduce.34 The combination of these

ideas motivates what I call a presumption of ontological innocence. Accordingly, for my purposes here,

‘ontologically innocent’ should be interpreted expansively, but defeasibly so.

First, consider the expansive interpretation. If there are any ontologically innocent combinations of

properties, boolean logical combinations of properties are among them.35 For an important example:

when we talk about a disjunction of properties, we are committed to little beyond than the properties

that serve as the disjuncts. However, for a variety of reasons to be explained below, I will take any

logical combination (including, if necessary, quantification), mathematical combinations (like averages of

determinate values of a physical property), and mereological sums as (defeasibly) ontologically innocent.

This expansion is motivated less by metaphysics than by the dialectic. There are interesting meta-

physical questions about whether, for example, talk of wholes commits us to the existence of something

more than their parts, or whether talking about many things (taken together) commits us to the existence

of more than the things themselves.36 But these concerns are outside those central to the reduction/non-

reduction debate. The issues of nomological autonomy, causal proportionality, abstraction, idealization,

or any of the host of issues informed by the existence of the special sciences, have prima facie force

independent of the outcome of the debate in (say) mereology and the logic of plurals.

Furthermore, it would trivialize the debate if reduction to, say, bare spatiotemporal distributions of

properties, or to certain other combinations of properties (like superpositions) specified by the physical

laws governing the fundamental properties, did not count as sufficiently reductive.37 One might believe

such combinations are not ontologically innocent. After all, a distribution is one, while the properties

distributed (and their instances) are many. Similarly, a system of particles can have an average of

determinate values of a physical variable distinct from the determinate values themselves. Let us there-

fore suppose that all higher-level broadly scientific properties are type-identical to bare distributions

of physical properties, but that bare distributions of physical properties are not ontologically innocent

combinations of physical properties. I cannot imagine that any reductive physicalist would despair at

such a result; at worst, it would be a Pyrrhic victory for the non-reductive physicalist. So it is best to

allow such combinations to count as ontologically innocent, and move the debate to the more interesting

areas.

Second, consider that any given sort of combination that I consider to be ontologically innocent is so

only defeasibly. This will become important when considering disjunctive properties. Though disjunc-

tions of properties are, in general, ontologically innocent, I am open to the possibility that disjunctions

of certain properties, or properties coextensive with such disjunctions, are not in fact ontologically inno-

cent. Such a conclusion requires an argument, though, and I will explore such arguments in the context

of undermining arguments for non-reductive physicalism.

That a case must be made that combining properties a given way is not ontologically innocent is the

core of the presumption of ontological innocence. If the reader finds that I do too much violence to the

term ‘ontologically innocent’ with the defeasible expanded interpretation, we can introduce a new term,

34This expansion can be interpreted as an attempt at convergence with non-reductive physicalism. See Ney (2010).35For those who object to talking about logical combinations of properties, we can translate this talk into talk about

those properties denoted by predicates that are logical combinations of predicates that refer to properties.36Boolos (1984) and Lewis (1991) are among those who think not. The contrary position is held by, for example, Yi

(1999).37But see Humphreys (1997a), and my treatment of his account in section 4.4.2.


‘ontologically not-guilty’, that captures my meaning, and replace instances of the former with the latter.

My main target is the purported ontological guilt of special science properties.38

Finally, let me explain why I chose to define ‘physical’ to include recursively ontologically innocent

combinations of physical properties. This choice was made for two reasons: the transitivity of reduction,

and convenience of nomenclature.

Reduction is a transitive relation. If type-identity to ontologically innocent combinations of properties

is to count as reduction to those properties, there has to be a sense in which ontological innocence is

also transitive. If properties of kind P are ontologically innocent combinations of properties of kind Q,

which are themselves ontologically innocent combinations of properties of kind R, then properties of

kind P should be ontologically innocent combinations of properties of kind R. This might seem a trivial

constraint, but it does add non-trivial complications to how fundamental properties can be combined.

For example, a simple average of averages is not necessarily the average of the components of the latter.39

Since P -properties being an average of Q-properties and Q-properties being an average of R-properties

does not result in P -properties being an average of R-properties, there is no guarantee that P -properties

are an ontologically innocent combination of R-properties.

There are two ways to deal with this. One is to insist that the only meaningful ontologically innocent

combinations are transitive, and we thus have to be careful to choose among related operations to find

those to which the fundamental elements are transparent. For example, the temperature of a gas is

a mean of more fundamental properties, viz., the kinetic energies of the molecules of the gas. The

temperature of a system of two gases is not the mean of the temperatures of the gases; rather, it “sees

through” the temperatures (which are themselves means) and is the mean of the kinetic energies of the

molecules of the gases taken together.40 The other way to deal with it is to make ontological innocence

transitive in the standard way (through transitive closure). We just stipulate that if P -properties are

an ontologically innocent combination of Q-properties, and Q-properties are an ontologically innocent

combination of R-properties, then P -properties just are an ontologically innocent combination of R-

properties, in virtue of standing in a chain of ontologically innocent combinations.

It does not matter for my purposes which of these options is chosen, since none of my arguments relies

on a particular option. It is enough that ontological innocence is transitive, so that it can do the work of

reduction. I chose to make ontological innocence transitive, and implemented this choice by recursively

defining ontologically innocent combinations of physical properties as ‘physical’. This implementation

simplifies talk of the class of properties to which reductive physicalists can reasonably expect broadly

scientific properties to reduce. I can call them ‘physical’, without any riders to the effect that chains of

ontologically innocent combinations are to be included as well.

1.1.6 Non-reduction

One final clarification should be made before moving on to establishing the burden of proof. Above, I

characterized non-reductive physicalism as the position according to which all broadly scientific prop-

38Leave such guilt to the existentialists, I say.39For example, 1 and 1 average to 1, and 4, 6, and 8 average to 6. The average of the averages is 3.5, while the average

of all the components is 4.40While a simple mean of some quantity is not transitive, the simple mean of some quantity of the molecular components

is transitive. If P -properties are the simple mean of the energies of the molecular components of Q-properties, and Q-properties are the simple mean of the energies of the molecular components of R-properties, then P -properties are the simplemean of the energies of the molecular components of R-properties. We can restrict ontologically innocent combinations tothese kinds of transitive relations.


erties are physically acceptable, but some are merely physically acceptable, i.e., some properties are

physically acceptable (nothing over and above the physical) without being type-identical to any physical

property. It might be argued that a more precise characterization would be as follows: all broadly scien-

tific properties are physically acceptable, but some higher-level properties fail to be type-identical to any

lower-level properties (including ontologically innocent combinations). The apparent difference between

my characterization and the alternative one can be seen as follows. According to typical non-reductive

physicalist accounts, there is widespread non-reduction, which is to say, special science properties regu-

larly fail to be type-identical not just to physical properties, but to any lower-level property.41 In other

words, a failure to reduce might happen at a level higher than the physical.

In this section, I argue that, for my purposes, the two characterizations are equivalent. This can

be seen under three plausible assumptions: (1) that the set of broadly scientific properties can be

partially ordered based on their level, (2) that there are physical properties, and (3) that all and only

physical properties are minimal elements of the set.42 The two characterizations share the same account

of physicalism,43 so the only part of the characterizations that concerns us is whether a physically

acceptable property is non-identical to a physical property if and only if it is non-identical to any lower-

level property.

First, consider whether my characterization is sufficient for the alternative one. According to my

characterization, if non-reductive physicalism is true, then there are merely physically acceptable proper-

ties, i.e., there are properties that are nothing over and above the physical, but are not type-identical to

any physical property. Since all minimal elements are physical properties, the merely physically accept-

able properties are higher-level properties. Since the properties are partially ordered, any such property

is non-identical to any (strictly) lower-level property. Therefore, there are higher-level properties that

are not type-identical to any lower-level properties.

Second, consider whether my characterization is necessary for the alternative one. According to the

alternative characterization, if non-reductive physicalism is true, then there are higher-level properties

that are not type-identical to any lower-level properties. Since all physical properties are minimal

elements, these higher-level properties cannot be physical. Therefore, there are properties that are not

type-identical to any physical properties.

That these two characterizations appear not to be equivalent can be seen as an implicit denial of one

of the assumptions. An advocate of the alternative characterization cannot reasonably deny the first

assumption, viz., that properties are partially ordered based on their level. That some properties are

higher-level and some are lower-level is essential to the alternative characterization. Interestingly, that

there are higher-level and lower-level properties is not essential to my characterization, though I do refer

to properties as higher- or lower-level for ease of explication.

It is also possible to deny that there are any physical properties, though this has more profound

41For example, if the Multiple Realizability Argument succeeds, then mental properties are type-distinct from physicalproperties because more than one physical type can realize a given mental type. But more than one biological type canrealize a given mental type as well. So by the Multiple Realizability Argument, mental properties are type-distinct frombiological properties. This generalizes.

42In other words, the hierarchy of properties bottoms out at the physical.43Given the above three assumptions, the physicalism part of the alternative characterization can be put in an apparently

more general fashion. Instead of saying that all properties are physically acceptable, we can amend the characterizationso that it says that all higher-level properties are nothing over and above any lower-level property.

The two versions of physicalism are equivalent under the further plausible assumption that for properties P , Q, and Rsuch that P is a higher level than Q and Q is a higher level than R, then if P is nothing over and above R, and Q isnothing over and above R, then P is nothing over and above Q. This means that being nothing over and above is leftEuclidean on the condition that the properties are ordered by level. See also Wilson (2005, p.429).


implications for physicalism and anti-physicalism than an apparent disagreement between two charac-

terizations of non-reductive physicalism.

The most likely reason that these characterizations appear to be not equivalent is an implicit denial

of the third assumption, viz., that all and only physical properties are the minimal elements of the

partially ordered broadly scientific properties. It is not reasonable to deny that only physical properties

are minimal elements; if the hierarchy of broadly scientific properties bottomed out in something other

than the physical, then physicalism would be false. But it may be reasonable to deny that all physical

properties are minimal elements. This can be done in two ways.

First, perhaps there are no minimal elements because there is no fundamental level. In such cases,

the alternative characterization makes sense of non-reduction better than my characterization. A phys-

ical fundamental level is treated as a privileged set of properties against which my characterization (but

not the alternative one) defines reduction. Both characterizations, however, suffer to make sense of

physicalism; that the fundamental level is physical is necessary on both characterizations to separate

physicalist worlds from non-physicalist worlds. To accommodate worlds without a fundamental level,

both characterizations would therefore need to be elaborated. For example, instead of having the fun-

damental level properties satisfy a set of physicality conditions, related conditions could be satisfied by

the properties of all levels below some threshold. Or perhaps the sequence of properties ordered by level

could converge on satisfying the physicality conditions at the limit of infinitely low levels. However the

characterizations are elaborated, some privileged set of physical properties is obtained.44 But once there

is a set of physical properties, my characterization can make sense of non-reduction. Furthermore, if that

set is quasi-minimal,45 the equivalence between my characterization and the alternative characterization

is recovered.

Second, perhaps some physical properties are not minimal elements, that is, perhaps some physical

properties are at a higher level than other physical properties. There is some merit to this concern.

Specifically, perhaps ontologically innocent combinations of properties ought to be considered a higher

level than any of the properties so combined. Ontologically innocent combinations of physical properties

would therefore fail to be minimal elements.46 So perhaps a higher-level property is type-identical to

a same-level ontologically innocent combination of lower-level physical properties, but to no property

at a lower level. Such a situation would satisfy the alternative characterization and fail to satisfy my

characterization. This breaks their equivalence.

However, if this is so, then the alternative characterization very likely mischaracterizes non-reductive

physicalism. The mere fact that a higher-level property is type-identical to no lower-level property

is no longer a guarantee of non-reduction. If a higher-level property is genuinely type-identical to an

ontological innocent combination of physical properties, then by definition its existence does not falsify

reductive physicalism. What seems to be behind intuitions otherwise is the belief that those supposedly

ontologically innocent, higher-level properties are not ontologically innocent at all. But then those

44In case all properties below some level satisfy physicality conditions, all such properties can be designated ‘physical’.In case properties converge to physicality, then for any set of conditions arbitrarily close to physicality, there is level atwhich all lower-level properties satisfy those conditions; for any given purpose, there is a set of properties that is ‘physicalenough’.

45In other words, we choose a set of elements that is both physical and can serve as a unique lower bound for everyinterval without a lower bound.

46Assigning ontologically innocent combinations of properties to a higher level than the properties combined would bea good way to generate a hierarchy of levels in nature given reductive physicalism. If all physical properties were minimalelements of the set of properties ordered by level, and all properties were type-identical to physical properties, then therewould be no levels in nature.


properties would not count as physical, and so their status as minimal elements is no longer an issue.

My characterization of non-reductive physicalism is at least as good as the alternative characteriza-

tion; for most purposes they are equivalent, and for others my characterization is likely better. But the

reason I am choosing my characterization is that it is an effective framework within which to understand

the reduction/non-reduction debate. It will allow me to focus not on how levels in nature are constituted,

but rather on whether certain combinations of properties are ontologically innocent, and if so, of what

they are innocent combinations.

1.2 The Plausibility of Reductive Physicalism

In this section, I will try to establish the prima facie plausibility of reductive physicalism. The goal of

this section is to show that the burden of proof is on those who advocate anti-physicalist or anti-reductive

positions. I’ll attempt this in two stages. First, I’ll show that it is reasonable to believe that physicalism

is true of the actual world. Second, I’ll show that, furthermore, reductive physicalism is antecedently

more plausible than non-reductive physicalism.

1.2.1 Physicalism

Explanations compatible with the truth of physicalism have a long and successful record in the empirical

sciences. There is ample empirical evidence that physicalism is true.47 All known substances are physical

substances, and no plausible candidates for physically unacceptable substances are on the horizon. With

respect to properties, the success of modern physics demonstrates the explanatory power of physics at

very small and at very large length scales. High energy and low temperature physics demonstrates the

same for high and low energy scales. Among those empirical sciences dedicated to intermediate length

and energy scales, chemistry is firmly intertwined with physical explanations, and the physical basis for

low-level biological phenomena is firmly established. Models of higher-level biological properties take

advantage of the physical and chemical ground of biology.

Furthermore, if physicalism were false, we would expect good scientific explanations for at least

some high-level empirical phenomena that are either incompatible with physical theory, or incompatible

with the completeness of physical theory. There are no scientifically accepted explanations of any such

phenomena that are incompatible in this way. Properties posited by biological and social sciences

require no mechanisms incompatible with physics for the causal evolution of their physical components.

All mental phenomena that have accepted explanations are explained in terms of the physiology of the

relevant organisms, whose explanations are in turn compatible with the physics of their components.

We can find more reasons to err on the side of physicalism from the practice of scientists. First,

when modeling new or poorly understood phenomena, part of the criteria for the goodness of the model

is that it be compatible with physics. Another, stronger criterion is that there be a plausible physical

mechanism by which the phenomenon evolves.

Second, one of the more important avenues of research for a given special science is the investigation

into said physical mechanisms. This is particularly true of chemistry and biology, where the atomic

and molecular mechanisms can be investigated using the techniques and results of physics. But it can

47For a thorough discussion of the empirical evidence in favour of physicalism, see Melnyk (2003a, ch.6) and Melnyk(2003b). For a description of how empirical evidence for the physical basis of biological processes led to the rise ofphysicalism among philosophers, see Papineau (2001).


also be seen in even higher-level special sciences, as when the chemistry and physiology of psychological

phenomena is investigated.

Third, special sciences often adopt the basic principles of physics. An example would be paleobi-

ology, especially that of very large animals. Basic physical principles such as those of mechanics and

thermodynamics are used along with observations of extant animals to reconstruct (for example) the

locomotion and internal physiology of dinosaurs.

Finally, the attitude of scientists towards under-investigated pseudo-scientific (including paranormal)

fields is revealing.48 Some pseudo-science is dismissed outright because it is (apparently) incompatible

with physics (and the other sciences). Hauntings and telekinesis are examples, since energy and mo-

mentum appear not be conserved in purported examples of such processes.49 Relatedly, pseudo-science

can be dismissed in a prima facie way when there are no plausible physical mechanisms for the process.

Astrology and telepathy are examples of these.

One further reason for a presumption in favour of physicalism can be seen from the trajectory of

scientific advancement.50 There are two observations we can make in scientific history that are relevant

to this argument. The first has to do with advances in our understanding of high-level phenomena.

Advances in chemistry, biology, and psychology have not overturned the results of physics. They have

instead filled in the gaps in our knowledge and explanations of higher-level phenomena in physically

acceptable terms. Similarly, advances at the lower levels are not isolated to the mysterious phenomena

that led to them, but touch on already well understood phenomena. The advances are constrained by

previous knowledge, and in turn lead to an even better understanding of the well understood phenom-

ena.51 Together, these observations suggest that science is advancing toward a complete and unified

picture of the world in which all broadly scientific properties are nothing over and above the physical.

1.2.2 Reduction

There are three reasons why, given that physicalism is the reasonable position to take, reductive physi-

calism ought to be the default position.

Ontological Parsimony and Physical Plenitude

The first reason is methodological (or perhaps aesthetic). The methodological principle of ontologi-

cal parsimony requires a presumption in favour of reduction.52 According to this principle, we ought

not multiply entities beyond necessity. It might be argued that ontological parsimony is neutral to

reduction/non-reduction, since if higher-level properties are type-distinct from physical properties, then

they are necessarily so. But whether they are (and thus, whether they are necessarily so) is the point in

question. Absent an argument one way or another, we ought to presume in favour of the position that

requires the smallest number of entities. If reductive physicalism is true, then the only property types

are the physical property types, and the only property instances are tokens of these types. On the other

48See, for example, Rothman (1988, p.151ff) and Bunge (2001, p.176ff).49Pseudo-scientific investigators often resort to appeals to hidden sources of energy to explain pseudo-scientific phenom-

ena. Even this practice reveals the presumption that models of phenomena that are incompatible with physics are notgood models.

50This is a weaker form of the optimistic induction for reduction.51A recent dramatic example of this is quantum entanglement in photosynthesis. See Engel et al. (2007) and Sarovar

et al. (2010).52Arguments based on ontological parsimony in favour of reductive physicalism go back at least as far as Smart (1959).

See also Churchland (1984, p.18ff).


hand, if non-reductive physicalism is true, then there are all of the above types and tokens, but there

are also further distinct types associated with higher-level properties, and perhaps even further distinct

tokens of those types. Thus, there ought to be a presumption in favour of reductive physicalism, as the

ontologically more parsimonious position.

More interestingly, reductive physicalism is the only position that can satisfy our often conflicting

intuitions about ontological parsimony and the principle of (physical) plenitude.53 According to the

latter principle, given enough time and space, every possible configuration of physical properties will be

instantiated (Baker, 2011). To put it another way, if the laws of physics do not forbid a state, some

system will eventually instantiate that state. This puts the principle of plenitude in apparent tension

with ontological parsimony.

However, configurations of physical properties are plausibly ontologically innocent. The principle

of plenitude can then be interpreted as saying that, in an infinite universe, we should expect every

ontologically innocent combination of physical properties to be instantiated. That they are ontologically

innocent is fortunate with respect to ontological parsimony; their innocence prevents a tension between

the principles of ontological parsimony and physical plenitude. Under reductive physicalism, since all

properties are physical, the principle of plenitude commits us to no further properties. However, under

non-reductive physicalism, either the configurations of physical properties are not ontologically innocent,

or accompanying each of the ontologically innocent combinations of physical properties is at least one

distinct higher-level property type. This would be an explosion of property types, to say nothing of

property instantiations. If non-reductive physicalism is true, our principles of ontological parsimony and

ontological plenitude are in tension. Absent any arguments one way or another, since we can thereby

eliminate the tension between ontological parsimony and physical plenitude, we ought to presume in

favour of reductive physicalism.

The Practice of Scientists

Another reason to presume in favour of reductive physicalism can be found by observing the practice

of scientists. As mentioned above, a crucial part of the advancement of science is the investigation of

the realizers of higher-level properties. For example, chemists investigate the behaviour of electrons in

materials in order to model higher-level chemical properties. In biology, research into the genetic basis of

the evolution and development of organisms involves investigating DNA, RNA, and protein binding. By

investigating the realizers of higher-level properties, experimental scientists frequently believe themselves

to be investigating the self-same property. As far as geneticists and genomicists are concerned, genes just

are the various DNA and RNA sites, and they are individuated in a language descriptive of the molecular

configuration. Furthermore, this concern with the nature of the realizers betrays a belief on the part

of scientists that the connection between a property and its realizers is closer than many philosophers

admit. Discoveries about the nature of the realizers are often immediately (and unreflectively) applied

to the nature of the realized property by experimental scientists.54 This move would be strictly invalid

if the property under experimental investigation were not type-identical to the realized property.

53Not to be confused with mereological plenitude, according to which every combination of parts is a whole.54An interesting case of this presumption is buried in the etymological history of the word ‘gene’. The term was introduced

to refer to a biological unit of inheritance (that which is passed on from parent to child); after modern evolutionary synthesisit referred to a biological unit of inheritance sensitive to natural selection. Once the protein-coding segments of DNA andRNA were discovered to be genes, the term became associated with that kind of realizer exclusively. Since then, otherunits of inheritance sensitive to natural selection (regulatory regions of the genome) have been discovered, but despite alsorealizing being a gene (in the original sense), they are not called genes.


Analogies to Reduced Properties

A final reason to presume in favour of reductive physicalism is the analogy between high-level properties

and paradigm cases of reduced properties. Uncontroversially physical properties are relevantly similar

to typical candidates for higher-level properties that are merely physically acceptable (nothing over and

above the physical without being physical themselves).

First, consider that a characteristic feature of properties typically taken to be non-reducible is the

asymmetry between the higher-level realized property and its lower-level realizing property. For example,

such higher-level properties are often taken to be functional properties, or determinables of some deter-

minates, or otherwise multiply realized.55 Yet each of these has an analogue among uncontroversially

physical properties. Physical properties are frequently taken to be functionally defined. For example,

the property of being (generating, or being subject to) a conservative force is explicitly functional.56

Indeed, it is not a new observation that physics theory can in principle be Ramsified, which would give

all properties a functional definition. Furthermore, there are many determinables that are uncontrover-

sially physical properties. For example, mass (to choose one among many continuous variables) is the

determinable of particular mass values, and of various massive particles. Finally, physical properties are

often multiply realizable. Consider, as a final example, the temperature of a gas. The temperature of a

gas is a higher-level property of the gas, and it is multiply realized by different molecular configurations

(in the same gas) and by different molecular compositions (in different gases). Yet the temperature of

a gas is a paradigmatic case of successful reduction: the temperature of a gas just is its mean molecular

kinetic energy.

Second, consider that another characteristic feature of properties typically taken to be non-reducible

is some form of causal autonomy. High-level properties typically taken to be non-reducible cause other

properties appropriate to their level. While a certain configuration of particles in a brain causes another

configuration of particles in a doorknob mechanism, it is the desire to open the door that causes the

doorknob to be turned. The latter causal event is (supposedly) counterfactually robust in a way that

the former event is not.57 But again, higher-level properties are analogous to uncontroversially physical

properties in this way. Each of the examples mentioned earlier is causally autonomous in the relevant

sense. A conservative force can cause the movement of a particle in a way that washes out what

configuration of particles generated the force. Mass can cause particles to gravitate in a way that is more

counterfactually robust than how the particular determinate masses cause the particles to gravitate. A

difference in temperature can cause heat flow autonomously from the actual configuration of molecules.

These analogies are not definitive. Advocates of non-reductive physicalism might deny them, and

point out that there are relevant differences between the physical properties and paradigm cases of

non-reducible properties. Or perhaps they would argue that these and similar examples from physics

do not in fact reduce to (even) lower-level properties. Perhaps we can Ramsify all of science, and the

lowest order of functional property is the order of those properties to which all must reduce if reductive

physicalism is to be true; each higher-order property (traditionally taken to be physics or not) that is

implemented by a lower-order property is in fact non-reducible. But neither of these is obvious, and any

such reasoning requires an argument in its favour. Absent such arguments, we have another reason to

favour reduction.

55See chapter 2, especially footnote 1 and the surrounding text.56A conservative force is one for which no net work is done when moving a particle through a closed loop.57Had the configuration of particles been different, the brain-particle-to-doorknob-particle causal event would not have

occurred, whereas the desire-to-turning causal event could well have still occurred. See section 2.3.1.


Collectively, these reasons suggest that the burden of proof is on those who advocate anti-reductive

and anti-physicalist positions. Absent such arguments, we ought to be reductive physicalists.

Chapter 2

Against Non-reductive Physicalism

It was argued above that reductive physicalism is the reasonable position to take in the absence of anti-

reductive or anti-physicalist arguments. Fortunately for the non-reductive physicalist, and (perhaps)

sadly for the reductive physicalist, arguments have been advanced for non-reductive physicalism. The

foremost among these is the Multiple Realizability Argument.

Recall that non-reductive physicalism is the position according to which some higher-level properties

of the special sciences are merely physically acceptable—that is to say, nothing over and above the

physical without being type-identical to any physical property. The precise nature of the relation between

higher-level properties and lower-level ones needs to be worked out, but generally speaking it is accepted

that it is asymmetrical, with lower-level physical properties necessitating the higher-level properties, but

not vice versa. Let us call this relation realization.1

Putnam (1967) and Fodor (1974) observe that the higher-level properties of the special sciences are

specified at a high level of abstraction, so that each one corresponds to more than one physical property.

This is the starting point of the Multiple Realizability Argument: since the higher-level property has

many realizers, it isn’t reducible to any one of them. A reply to this argument, the so-called Disjunctive

Strategy,2 is to point out that nothing stops us from reducing the higher-level property to the disjunction

of its realizers; and a disjunction of physical properties is still a physical property.

It is instructive to look at this argument more formally. According to the reductive physicalist:

DS1 Every multiply realized, higher-level property of the special sciences is type-identical to the dis-

junction of the types of its realizers.

DS2 A disjunction of properties is an ontologically innocent combination of the disjuncts.

DS3 The only realizers of a multiply realized, higher-level property of the special sciences are physical

properties.3

1Wilson (2006) gives three examples of how realization has been characterized in the literature. One option is rolefunctionalism, where higher-level properties are characterized by their causal role, and lower-level properties satisfy thoseroles. See, for example, Putnam (1967), Fodor (1974), or Antony and Levine (1997). Another option is the determi-nate/determinable relation, on which lower-level properties (the determinates) are a particular way of being higher-levelproperties (the determinables). See, for example, Yablo (1992) and Wilson (2009). A third option is mereological, withhigher-level properties being mereological parts or sums of the lower-level properties. See, for example, Shoemaker (2001)and Paul (2002).

2Or the Disjunctive Move. See Jaworski (2002).3This premise should be interpreted with some care.First, if non-reductive physicalism is true, then higher-level properties will typically have merely physically acceptable

realizers, so a fortiori they will have non-physical realizers. For example, biological properties might have chemical realizers

19

Chapter 2. Against Non-reductive Physicalism 20

In other words, if a disjunction of realizers is an ontologically innocent combination of its disjuncts,

and the disjuncts are ultimately physical, then the disjunction of realizers is an ontologically innocent

combination of physical properties, i.e., is a physical property. Therefore, if a higher-level property is

type-identical to the disjunction of its realizers, then it is type-identical to a physical property.

The non-reductive physicalist can respond to this argument only by denying one of the premises.

To deny DS1 is to deny type-identity: the higher-level property is not type-identical to the disjunction

of its realizers. In this response, the non-reductive physicalist argues that higher-level properties have

characteristics lacked by the disjunctions of their realizers. To deny DS2 is to deny ontological innocence:

the disjunction of the realizers of a higher-level property is not an ontologically innocent combination of

its realizers. Disjunctions in general are plausibly ontologically innocent combinations of their disjuncts,

so the most promising denial of this premise is to argue that a disjunction of the realizers of a higher-level

property is a special case. Finally, to deny DS3 is to deny the physicality of the disjunction of realizers:

the disjunction to which a higher-level property reduces has fundamentally non-physical disjuncts, and so

is not itself physical. Such a response could plausibly be motivated by the observation that higher-level

properties have fundamentally non-physical realizers (in physically impossible worlds). An ontologically

innocent combination of such realizers is then not a physical property.

In this chapter, I will explore the Multiple Realizability dialectic in the context of these responses. I

will begin by describing the Multiple Realizability Argument, its prima facie plausibility, and the danger

posed by the Disjunctive Strategy (section 2.1). Thereafter, I will describe responses to the Disjunctive

Strategy and address them. In section 2.2, I will describe and address two responses of the first sort,

denying type-identity: one by Fodor (1974), and one by Wilson (2011). In section 2.2, I will reconstruct

from Wilson (2011) a response of the second sort, denying ontological innocence, before replying to it.

In section 2.4, I will construct (on behalf of the non-reductive physicalist) a response of the third sort,

denying the physicality of the disjunction of realizers, and then address it. I will conclude in section 2.5.

Throughout this chapter, I will assume that physicalism is true. If it is false, then both reductive and

non-reductive physicalists are strictly wrong, and the debate between them hollows out considerably. I

will address anti-physicalist arguments in subsequent chapters.

2.1 Multiple Realizability and the Disjunctive Strategy

Recall that, according to the non-reductive physicalist, the higher-level properties of the special sciences

are realized by (lower-level) physical properties. In the zeroth iteration of the Multiple Realizability

Argument, we note that a typical higher-level property of a special science is specified at a level of

abstraction so as to be multiply realizable, i.e., it has more than one possible realizer. Illustrating for

higher-level property M with physical realizers P1, P2, . . . :

which themselves have physical realizers; if the chemical realizers are merely physically acceptable, then the biologicalproperties have some non-physical realizers. So the non-reductive physicalist might be tempted just to insist that DS3 isfalse. However, in that case, the Disjunctive Strategy can be applied to those purportedly merely physically acceptablerealizers. This process will iterate until there is a property whose realizers are all either physical or fundamentally non-physical.

Second, not all fundamentally non-physical realizers are problematic for reduction. A physical realizer of the higher-levelproperty may itself be realized by a fundamentally non-physical property (in physically impossible worlds). The higher-levelproperty having such a fundamentally non-physical realizer is not obviously incompatible with reduction to the physical;for example, the higher-level property would still be necessarily coextensive with the disjunction of its physical realizers.

Stripping away the iterative structure of the argument, and taking the above into account, DS3 should be interpretedas saying that the higher-level property has no fundamentally non-physical realizers that do not themselves realize someof the physical realizers.


M

P1 P2 P3 . . .�� 6

AAK . . .

This blocks a simple version of type-identity: the higher-level property cannot be type-identical to any

one of its realizers, since the realizers are not type-identical to one another.

But this naıve attempt to argue against type-identity has an obvious flaw. Consider a paradigmatic

case of reduction: the temperature of a gas just is its mean molecular kinetic energy, a physical prop-

erty. Nevertheless, it is multiply realized by various molecular states. The problem for the Multiple

Realizability Argument is that some combinations of physical realizers are ontologically innocent. For

example, taking the average of physical quantities uncontroversially results in another physical quantity,

and so mere multiple realizability is not enough to establish a non-reductive conclusion.

The problem is a general one. A disjunction of properties is also an ontologically innocent combi-

nation.4 For every multiply realizable higher-level property, there is an associated disjunctive property

obtained by disjoining its realizers, to which it is necessarily coextensive. Nothing stops us from simply

type-identifying the higher-level property with the disjunction of its physical realizers, thus reducing it

to a physical property. This is the Disjunctive Strategy.

2.2 Denying Type-Identity

Non-reductive physicalists can respond to the Disjunctive Strategy by denying that the higher-level prop-

erty is type-identical to the disjunction of its realizers. In such responses, the non-reductive physicalist

argues that the higher-level property has some characteristic that sets it apart from the disjunction of

its realizers.5 The reductive physicalist is then compelled to reply in one of three ways: (1) neither the

higher-level property nor the disjunction has that characteristic, (2) both the higher-level property and

the disjunction have the characteristic, or (3) the characteristic has no bearing on type-identity.

In this section, I will outline two responses of this sort. The first response arises from Fodor’s (1974)

claim that there is a difference in kindhood between the higher-level property and the disjunction. I

will then describe two rejoinders to the response: one by Kim (1992) that denies the kindhood of both

properties, and (more important for my purposes) one by Clapp (2001) and Antony (2003) that affirms

the kindhood of both.6 The second response arises from Wilson’s (2011) claim that token instances of a

higher-level property have fewer causal powers than the token instances of the disjunction of its realizers.

I will then describe three rejoinders to this response.7

4Again, if the idea of disjoining properties is troublesome, we can take it as an abbreviated way of talking aboutdisjoining the predicates that refer to the properties. The disjunctive property is the property referred to by the predicateconstructed by the latter disjunction. Issues about whether such predicates refer at all (say, if they have indefinite disjuncts)can be subsumed under issues of whether the disjunctive property is a legitimate property. See also footnote 4 of chapter 1.

5Here and elsewhere, ‘characteristic’ should be interpreted in the loosest possible way. The non-reductive physicalistargues that something is true of the higher-level property that is false of the disjunction of its realizers. The purpose offormulating it in terms of properties having characteristics is to demonstrate that only three possible sorts of responsesare available to the reductive physicalist, and to show the resemblance among such responses.

6Arguing that kindhood has no bearing on type-identity seems unpromising. It’s hard to imagine how a kind can betype-identical to a non-kind. It may be worth exploring the subject, but I will not do so here.

7There is a third response of this sort that I could construct on behalf of the non-reductive physicalist. It is at leastprima facie possible that the higher-level property is more abstract than the disjunction of its realizers. This could happenif, using Clapp’s scheme (section 2.2.2), the disjunction of its realizers overlap on a proper superset of the causal powersthat characterize the higher-level property. There may be a lower limit to how small the intersection of the causal powersof the realizers can be, since all realizers that have the higher-level property’s characteristic causal powers might havefurther causal powers in common. In this case, the exhaustively overlapping disjunction of realizers characterizes a realizer


2.2.1 Difference in Kindhood

In the first iteration (and original formulation) of the Multiple Realizability Argument, Fodor (1974)

anticipates the Disjunctive Strategy. To counter it, he claims that a typical higher-level property of

a special science is specified at a level of abstraction so as to be multiply realizable by very different

physical realizers. This makes the disjunction of its realizers heterogeneous. It is then argued that the

higher-level property cannot be type-identical to the disjunction of its realizers since the former is a

natural kind, while the latter is not.

The higher-level property is a natural kind, according to Fodor, because it is treated by a special

science law. Suppose that the relevant laws are causal laws. We can say, provisionally, that a necessary

condition for it to be true that it is a law that P causes Q is that P and Q are natural kinds.8 So the

properties treated by the special sciences, being subject to true natural laws, are natural kinds.

On the other hand, Fodor argues, a heterogeneous disjunction of physical properties is unlikely to

be treated by any laws of physics, for three reasons. First, the mere fact that it is heterogeneous makes

it unlikely to be treated by a basic law of physics. Physics describes the behaviour of systems at a high

(perhaps the highest) level of detail, and very different properties are unlikely to be subject to the same

behaviour at that level of detail.

Second, lawfulness is not truth-functional. Suppose we have two laws, that P1 causes Q1 and that P2

causes Q2. It follows from this that (P1∨P2) causes (Q1∨Q2). But it does not follow that it is a law that

(P1∨P2) causes (Q1∨Q2).9 As Fodor (1974, p.109) says, “it is a law that the irradiation of green plants

by sunlight causes carbohydrate synthesis, and. . . a law that friction causes heat, but. . . not. . . a law that

(either the irradiation of green plants by sunlight or friction) causes (either carbohydrate synthesis or

heat).” So ‘is either carbohydrate synthesis or heat’ fails to pick out a natural kind.

Finally, Fodor observes that special science laws can have exceptions, while physics cannot.10 The

only way for special science laws to have exceptions while physical laws are exceptionless is for a special

science property to be a kind while its disjunctive realizer is not. Suppose P and Q are special science

kinds, and that it is a law that P causes Q. Suppose further that P1, P2, P3, . . . are the physical realizers

of P , and Q1, Q2, Q3, . . . are the physical realizers of Q. If the law that P causes Q has exceptions, then

so does the purported law that (P1∨P2∨P3∨ . . .) causes (Q1∨Q2∨Q3∨ . . .). But the latter cannot be a

physical law, since (by Fodor’s observation) physical laws have no exceptions. Thus, (P1 ∨P2 ∨P3 ∨ . . .)

of the higher-level property, not the higher-level property itself. Since the higher-level property has fewer causal powersthan the disjunction of its realizers at the type-level, they cannot be type-identical.

I don’t think this line of thinking withstands scrutiny. Arbitrary combinations of causal powers don’t have ontologicalsignificance, so a plausible example of such an abstract property would have to be shown, or an argument made that thiskind of abstractness is typical of special science properties. Furthermore, there is no reason that the set of causal powersthat characterizes the disjunction of realizers has to be the largest set on which the disjunction exhaustively overlaps. Theinitial force of the response comes, I think, from hijacking the intuition that the disjunction of physical realizers overlapson a superset of the causal powers of the higher-level property, and that disjoining fundamentally non-physical realizers isneeded to isolate a small enough set to characterize the higher-level property. But such a response is properly placed insection 2.4, since it denies the status of the disjunction as physical, not the type-identity of the higher-level property andthe disjunction.

8Suppose lawlike sentences have the form of a universal generalization modified by a modal operator (�L) interpretedas “it is a law that”. Let the sentence �L∀x(Px → Qx) be a typical lawlike sentence of a special science. We can say,provisionally, that a necessary condition for the sentence to be true is that the properties to which P and Q refer arenatural kinds.

9See footnote 8. According to Fodor, �L∀x(P1x → Q1x) and �L∀x(P2x → Q2x) together entail ∀x((P1x ∨ P2x) →

(Q1x∨Q2x)), but not �L∀x((P1x∨P2x) → (Q1x∨Q2x)). For this to be true, lawfulness and nomological necessity mustcome apart. According to Fodor, the former two law statements entail the latter only if it is supposed that (P1 ∨ P2) and(Q1 ∨ Q2) are natural kinds, which we cannot assume at the outset.

10More precisely, it should be possible for special science laws to have exceptions while physics does not.


and (Q1 ∨ Q2 ∨ Q3 ∨ . . .) cannot be (physical) kinds.11

2.2.2 No Difference in Kindhood

That there should be a difference in kindhood (or legitimacy generally) between the higher-level property

and the disjunction of its realizers is the most controversial part of Fodor’s argument. There are two

ways to resist it: assert that neither is a natural kind, or assert that both are natural kinds.

Neither is a Kind

Kim (1992) takes the former approach. He agrees with Fodor that a heterogeneous disjunction of

natural kinds is not itself a natural kind. According to Kim, properties are individuated by their

causal powers (Kim, 1992, p.17). Natural kinds have tokens with causal powers in common, so they

are projectible: instances of the property can serve as inductive evidence for generalizations about all

instances of the property (Kim, 1992, p.11ff, p.19). Presumably, only a projectible property can be the

subject of a true natural law, and so being projectible is a necessary condition for being a natural kind.

Elaborating on this argument, a disjunction of natural kinds is heterogeneous when the kinds disjoined

have heterogeneous powers. Having heterogeneous powers, the disjunctive property fails to be projectible:

instances of the disjunctive property cannot serve as inductive evidence for generalizations about all

instances of the property. To illustrate, Kim considers jade, which is the disjunction of two mineral

kinds, jadeite and nephrite. Instances of jadeite cannot be used to make conclusions about nephrite and

vice-versa. Jade is therefore not projectible, and so not a kind.

However, Kim goes on to say, the higher-level property and its associated disjunctive property are

necessarily coextensive. Furthermore, if physicalism is true, then every higher-level property instance

inherits its causal powers from one of its realizers (the realizer on that occasion) (Kim, 1992, p.18).

It follows that the disjunctive property and the higher-level property have the same status in terms of

being or not being kinds. If the higher-level property is realized by a heterogeneous disjunction of kinds,

then the higher-level property (whose instances have only those causal powers of their realizers on the

occasion) is not projectible. So if a heterogeneous disjunction of kinds is not a natural kind, then neither

is the higher-level property realized by it.

Dialectically, this result makes the type-identity of the higher-level property and the disjunctive

property moot, since neither are natural kinds, and reductive physicalists are presumably interested in

type-identifying higher-level kinds to physical kinds. As a consequence, Kim proposes a local reduction,12

in which we ‘functionalize’ the higher-level property only to type-identify it with its realizer on a type-by-

type basis. This is known as functional specification: causal/functional language is used to characterize

a property, but the property is type-identified with its realizer on the occasion.13 Kim’s strategy is semi-

eliminativist, since it does away with heterogeneously realized higher-level properties, while keeping the

11According to Fodor, the exceptions of the special science law are explained by some physical realizers of P failing to belawfully connected to any physical realizers of Q. This explanation is unavailable to anyone who type-identifies the specialscience properties with the disjunctive properties.

12Kim finds local reduction “more satisfying” (Kim, 1992, p.24), though he allows for higher-level properties to betype-identical to a heterogeneous disjunction of physical kinds.

13Functional specification goes back as far as Lewis (1966) and Armstrong (1968) (see Levin (2013)). It contrasts withfunctional state identity: use causal/functional language to define a property as the second-order property of being someproperty that satisfies the appropriate causal/functional role. The latter is implicit in, e.g., Fodor (1974).


lower-level kinds.14 If pain is realized heterogeneously by humans, Martians, and androids, then there

is no natural kind pain; instead, there is human-pain, Martian-pain, and android-pain.

However, this semi-eliminativist consequence makes Kim’s approach problematic. It does not respect

our intuitions about the unity and abstractness of higher-level properties. It may be argued that a given

special science predicate fails to refer to a natural kind. But this is unlikely to be true for all special

science predicates. Furthermore, at least intuitively, instances of a higher-level property governed by

some special science law do have common causal powers: those in virtue of which they satisfy the

causal/functional role used in their specification.

Both are Kinds

The alternative rejoinder is to affirm the kindhood of both the higher-level property and the disjunction

of its realizers. This is the approach employed by Clapp (2001) and Antony (2003). They accept that

the higher-level property is a natural kind—it is featured in special science laws, after all. But for that

very reason, we can expect that its realizing disjunctive property is also a natural kind.

Consider the projectibility of a heterogeneous disjunction of kinds, like jade. Instances of jade, being

instances of either jadeite or nephrite, can in fact serve as inductive evidence for generalizations about

all instances of jade. For example, jade has a certain density range and a certain hardness range, so long

as jadeite and nephrite fall in those ranges. Therefore, jade is projectible. The problem with jade is not

that instances of it fail to to serve as inductive evidence for generalizations about all instances of jade.

The problem is that instances of jade fail to to serve as inductive evidence for any generalizations about

only instances of jade; nothing that jadeite and nephrite have in common is not also had by some other

mineral.15,16 This is why jade is not a natural kind.

So, in general, disjunctive properties are not natural kinds. But the analysis above immediately

suggests a class of disjunctive properties that are natural kinds. Consider any disjunctive property that

has as disjuncts (i) only properties that share a set of causal powers (they overlap) and (ii) all properties

that have that set of powers (they are exhaustive). The causal powers that characterize such a disjunc-

tive property are just the intersection of the causal powers that characterize the disjuncts. Disjunctive

properties of this class (as opposed to disjunctive properties in general) are those we expect to be asso-

ciated with higher-level properties characterized by the same set of causal powers. If pain is functionally

characterized, then human-pain, Martian-pain, and android-pain all satisfy that characterization. If pain

satisfies a lawlike regularity, then so do the various realizers of pain, since they (and only they) have the

causal powers that pain inherits. The disjunctive property is a natural kind for the same reason that

the higher-level property is a natural kind.

This account makes the Disjunctive Strategy available again to the reductive physicalist. A higher-

level property just is the disjunction of its realizers, and a disjunction of properties is an ontologically

innocent combination of those properties. This process bottoms out at the level of the physical. Since

an ontologically innocent combination of physical properties is physical, the non-reductive conclusion is

averted.

14It is not quite eliminativist, since the special science studying the higher-level property is not orthogonal to thelower-level science; the special science just (perhaps erroneously) unifies phenomena that are not unified in nature.

15See, for example, Fodor (1997, p.154).16Save, perhaps, the power to induce the utterance ‘jade’ in the appropriate contexts, and the like. Since any combination

of property instances can in principle be given a name, admitting powers of this sort leads to grue-like problems that areoutside the scope of this argument. Note, however, that this is a problem for the non-reductive physicalist, since theconclusion to this line of reasoning is that any disjunctive property can serve as a physical property to which a higher-levelproperty can be reduced.


An advantage of this approach is that it not only respects our intuitions about the unity and ab-

stractness of special science properties, but it accounts for them. Instances of a higher-level property

are unified inasmuch as they have causal powers in common. They are abstract since they share only

a proper subset of their causal powers. The higher-level property constitutes a natural kind since its

realizers exhaust those properties that have its characteristic causal powers.

2.2.3 Token-Distinctness

Another attempt to deny the type-identity of the higher-level property and the disjunction of its realizers

is made by Wilson (2011). According to this argument, the token instances of a higher-level property

have fewer causal powers than the token instances of the disjunction of its realizers, so they cannot be

instances of the same property. To understand the argument, it is necessary to distinguish token-identity

and token-distinct accounts of property realization.

There are two ways that token instances of a higher-level property can be related to token instances

of the property that realizes it on the occasion of its instantiation. The first is token-identity: a token

instance of a higher-level property just is the token instance of its realizing property on that occasion.

Token-identity can be coupled with physicalism to give token physicalism. On this picture of the world,

every token instance of any property is a token of a physical type, but each token can be a token of

many different types.17 Token-identity is compatible with both reductive physicalism and non-reductive

physicalism, since the very same tokens that are tokens of physical types might also be tokens of merely

physically acceptable types. Because of this, and because it appeared to guarantee the physical accept-

ability of the types in question, token-identity was the assumption upon which the Multiple Realizability

dialectic originally proceeded. Both reductive and non-reductive physicalists accepted that all property

instances were tokens of a physical type, and debated whether there were merely physically acceptable

types of which they were also tokens.

The second way that token instances of a higher-level property can be related to token instances of

property that realizes it on the occasion of its instantiation is token-distinctness: a token of a higher-

level type is distinct from the token of its realizer on that occasion. To square such a relationship with

physicalism, advocates of token-distinctness of property instances typically argue for token-identity at

the level of causal powers.18 On this picture, all token causal powers are token-identical to causal powers

of some instance of a physical property. The distinctness of the sets of (token-identical) causal powers

then grounds the distinctness of property instances. Higher-level properties have different sets of causal

powers than their realizers, and so instances of the former are token-distinct from instances of the latter.

One advantage of such an account for that non-reductive physicalist is that, arguably, a higher-level

property being token-distinct from its realizer on the occasion is a necessary condition for non-reduction.

Specifically, Wilson (2011) argues (in part) that a necessary condition for non-reductive physicalism is

that token instances of a higher-level property have a proper subset of the causal powers of the realizer

on the occasion.19 In other words, the Subset Condition on Causal Powers (SCCP) must be satisfied.

17This view goes back at least as for as Donald Davidson’s anomalous monism (Davidson, 1970), according to whichthe mental (constituted by rationalizing explanations) and the physical (constituted by causal closure) cross-categorize thevery same events.

18In contrast, see Clarke (1999), according to whom higher-level property instances have causal powers beyond those oftheir realizers on the occasion.

19The argument is persuasive, but, it seems to me, not quite sound. In particular, non-reductive physicalists can adopttoken-physicalism, as long as they deny the physicality of the disjunctive property to which the higher-level property istype-identical. See section 2.4 below.


This is a strong motivation for a non-reductive physicalist to deny token-identity in favour of an account

of token-distinctness that preserves physicalism.20

The claim most relevant to the current discussion is that a property instance cannot have more causal

powers than those characteristic of its type. Call this thesis Property Instance Power Limits, or PIPL.

The argument for PIPL, however, is underdeveloped. Shoemaker (2001) argues that a property instance

just is the conferring (on some object) of those causal powers characteristic of its type, so that the

conferring of different sets of powers should count as different property instances. Thus, for example, he

claims that an instance of red is token-distinct from the instance of scarlet that realizes it. Wilson (2011)

argues that it makes no sense for a property instance to have more causal powers than is characteristic

of its type.21 This is so

at least if types are supposed to track similarities among tokens of the type. If a [prop-erty instance] has more powers than a given [property] type, that is, or so it seems to me,compelling reason to think that the [property instance] is not of the type. (Wilson, 2011,p.145)

We may be led to believe otherwise since we naturally think it possible for a token to be a token of more

than one type. But such intuitions are misapplied. An object (token) can be of more than one type

because it can have more than one property. However, causal powers are not properties of a property

instance (token).

At the root of these arguments is something like Clarke’s (1999) argument that all instances of

the same property are exactly the same. Suppose properties are universals. Since every instance of a

universal is the same, if any instance of that universal has (or lacks) a causal power, then they all do.

Alternatively, suppose property instances are tropes. Since types are sets of exactly resembling tropes,22

if one trope has (or lacks) a causal power, then all exactly resembling tropes do. It is understandable why

non-reductive physicalists would be attracted to such an argument, given that higher-level properties

are regarded as ontologically significant.

Token-distinctness can be used to argue against the type-identity of a higher-level property and the

disjunction of its realizers. If a higher-level property instance has fewer causal powers than its (token)

realizer on the occasion, then it is token-distinct from its realizer. On the other hand, an instance of

a disjunction of properties is, intuitively, just an instance of one of the disjuncts. So the instance of

the disjunction of the higher-level property’s realizers is token-identical to the realizer on the occasion.

Therefore, a higher-level property instance is token-distinct from the instance of the disjunction of its

realizers. Their type-distinctness (presumably) follows from their token-distinctness.

This argument proceeds in three stages. First, the token-distinctness of the higher-level property

instance and its realizer on the occasion is established. Then, the token-identity of the instance of

the disjunction of its realizers and its realizer on the occasion establishes the token-distinctness of the

20Indeed, satisfying SCCP may do more for the non-reductive physicalist than preserve physicalism. It also purportedlyexplains, contra Kim (1998, 2005), how a type-distinct higher-level property does not causally compete with its realizer;since the two share causal power tokens, there is only one causal event on the occasion. See Wilson (2011).

21Wilson claims to be following Clarke (1999) when she states that it is possible for a property instance to have fewercausal powers than its type, owing to causal powers being conditional on circumstances (i.e., activation conditions) inwhich the property is instanced. This appears to me to be a misreading of Clarke. According to the latter, a propertyinstance can never have fewer causal powers than those characteristic of its type. What might have fewer causal powersthan those characteristic of a higher-level property type is not the higher-level property instance. Rather, it is the complexthat includes both the higher-level property instance and at least some of the background structure that comes along withits realizer on the occasion, and upon which activation conditions of the higher-level property depend.

22Rather, types are classes of tropes with exactly resembling natures, where having causal powers is part of the nature.


higher-level property instance and the instance of the disjunction. Finally, this sort of token-distinctness

is argued to be sufficient for type-distinctness. More formally:

TD1 The set of causal powers characteristic of a higher-level property is the intersection of the sets of

causal powers of its realizers. (Assumption about the case)

TD2 A property instance has no further causal powers than those characteristic of its type. (From

PIPL)

∴ TD3 A higher-level property instance generally has a proper subset of the causal powers of its realizer

instance on the occasion. (From TD1 & TD2)

TD4 An instance of a disjunction of properties is just an instance of one of the disjuncts. (Intuition)

∴ TD5 A higher-level property instance generally has a proper subset of the causal powers of the instance

of the disjunction of its realizers. (From TD3 & TD4)

TD6 Type-identity is incompatible with this sort of overdetermined token-distinctness. (Intuition)

∴ TD7 Higher-level properties are type-distinct from the disjunction of their realizers.

The argument is valid, so to deny the conclusion requires denying at least one of the premises. TD1

follows from Clapp’s analysis of disjunctive properties associated with higher-level properties. TD3,

TD5, and TD7 are conclusions of the three stages of the argument. That leaves TD2, TD4, and TD6

available to be denied.

2.2.4 Denying the Argument from Token-Distinctness

In this section, I will describe three rejoinders to the argument from token-distinctness. The three

rejoinders correspond to denying TD2, TD4, and TD6, respectively, of the above argument.

Denying TD2

TD2 states that a property instance has no further causal powers than those characteristic of its type.

This is just PIPL, so to deny the argument for token-distinctness in this way, the reductive physicalist

must deny the argument for PIPL.

Recall that, according to Shoemaker, a property instance just is the conferring (on some object)

of the causal powers characteristic of its type. From this he concludes that conferring different sets of

causal powers implies instances of distinct properties. However, this argument can be resisted. We have

good reasons to suppose that what are being conferred when properties are instantiated are not sets of

causal powers, but the causal powers themselves. If so, conferring the members of some set of causal

powers just is conferring the members of each subset of that set; there is only one conferring.23 In that

case, the same property instance would be a token of multiple types, one for each subset (of all the

causal powers conferred on that occasion) that characterizes a type.

23For Valentines Day, I gave my wife a card and flowers. I can take the position that there was only one giving (thegiving of the card and flowers). I can take the position that there were two givings (one for each gift given). But I cannotreasonably take the position that there were three givings, the card giving, the flowers giving, and the giving of both. Ametaphysics that requires a separate giving for each member of the power set of the set of gifts appears to be an error. Soit is with conferring causal powers.


Similar considerations apply to Wilson’s argument. Recall that, according to Wilson, since types track

similarities, having more causal powers than characterizes a type is sufficient for a property instance not

being of that type. Furthermore, while objects can be of more than one type (since they can have more

than one property), property instances cannot, since even if they are typed according to their causal

powers, causal powers are not properties of properties. Again, both these arguments can be reasonably

resisted. Types can track similarities without necessarily tracking exact similarities. Indeed, a causal

powers account of properties explains how property instances can be similar without having to be the

same; namely, they can share some but not all their causal powers. Likewise, it is unimportant whether

property instances have causal powers in the way that objects have properties. In the relevant respect,

the analogy between properties (of objects) and causal powers (of property instances) appears perfectly

serviceable. Both can be used as the values of variables the satisfaction of which determines whether the

token is a member of a type. Whatever combination of properties constitutes a type, if the object has

those properties, it is of that type. In the same way, the reductive physicalist can insist that whatever

set of causal powers characterizes a type, if the property instance has those causal powers, it is of that

type. In this way, the intuition that a token can be of more than one type can be recovered.

Of course, we could merely define the token in question to be that set of token causal powers, and

characterize its type by those (type-level) causal powers. But the ontological significance of such a token

is prima facie unremarkable.24 It’s also not clear that anything is gained by talking about such tokens

compared to talking about their types.

The same problem can be seen in Clarke’s related argument that property instances must be identical

in every (intrinsic) respect. Recall that, according to his argument, all instances of a universal are exactly

the same (and so must have the same set of causal powers), and all tropes of the same type have the

same natures (and so must have the same set of causal powers). Thus, whether a property instance is

an instance of a universal or is a trope, every property instance of the same type must have the same

set of causal powers.

But as with Wilson’s argument, this argument can be resisted by questioning the demand for exact

similarity. Suppose a property type is a universal. The traditional concept of a universal is not that all

instances of it are exactly the same, but rather that the universal is wholly present in any instantiation

of it. This is an important difference. The former demands of a property instance that, in order to be

an instance of a universal, it must have all and only those causal powers that characterize the universal.

According to the latter, it must have merely all those causal powers, with no further constraint on which

causal powers it has.25

Alternatively, suppose a property instance is a trope. It is true that, traditionally, some trope types

have been characterized by exact resemblance,26 an equivalence relation.27 But typing tropes does not

require exact resemblance; resemblance is sufficient,28 and resemblance is a similarity relation.29,30 This

24This is one of the lessons of Ney (2010).25Indeed, this is enough for Clarke to make his point. His concern in making the argument is that a property instance

cannot have fewer causal powers than its type, to support his position that higher-level property instances have morecausal powers than their realizers on the occasion.

26See, for example, Williams (1953), though he does allow for “approximately similar” tropes to compose a “less definiteuniversal”(Williams, 1953, p.9).

27That is, reflexive, symmetric, and transitive.28For one among many examples, Campbell (1981), types tropes on the basis of resemblance, where “[t]he closeness of

resemblance between the tropes in a set can vary.” (Campbell, 1981, p.484).29That is, reflexive and symmetric, but not necessarily transitive.30 Indeed, some trope theorists rely this to distinguish resemblance (a similarity relation) from compresence (an equiva-

lence relation) in order to ground the asymmetry of objects (compresent tropes) and predicates (resembling tropes). See,


is again an important difference. Suppose (for convenience) that tropes resemble (or not) according to

their causal powers. If types are equivalence classes, then, since having all and only those causal powers

in a given set defines an equivalence class, the set of all tropes that share all and only those causal powers

can serve as the type characterized by that set. But having only some of those causal powers does not

(in general) define an equivalence class, and so the set of all tropes that have only some of those causal

powers cannot serve as the type. On the other hand, if types are similarity classes, then, since having

some of the causal powers of a given set defines a similarity class, the set of all tropes that have some of

those causal powers can serve as the type.31

But suppose a trope theorist insists (for whatever reason) that types are equivalence relations. In

that case, as described above, a set of exactly resembling tropes (defining an equivalence class) could

serve as a type, but a set of all merely resembling tropes (defining a similarity class) could not. Still,

there is room for inexact resemblance: two tropes inexactly resemble only if they have a certain set of

causal powers in common. Inexact resemblance also defines an equivalence class, one for each member

of the power set of all possible causal powers. Since having all (but not necessarily only) those causal

powers in a given set defines an equivalence class, the set of all tropes that share all (but not necessarily

only) those causal powers can also serve as the type characterized by that set.32 Any work that exact

resemblance does in trope theory can be done by inexact resemblance.33

These replies to the argument for PIPL resemble one another, which points to a deeper issue. One

might be tempted into a general suspicion over using sets of causal powers to distinguish property

instances in an ontologically revealing way. Without a prior commitment to a particular metaphysical

perspective,34 causal powers, laws of nature, and properties form a circle of concepts. Perhaps properties

can be individuated by sets of causal powers. But it seems just as likely that causal powers can be

individuated by sets of lawlike state transformations (those that represent the activation of the causal

power). And perhaps lawlike state transformations can be individuated by (ordered pairs of) sets of

properties (those instantiated in the initial and final states of the transformation). It seems to me that

relying on sets of causal powers to argue for the ontological significance of a property instance merely

reveals a prior commitment to the ontological significance of the property.

To be fair, this is not a fault of the non-reductive physicalist position. The arguments for PIPL were

not made to convince the reductive physicalist, but to show that the non-reductive physicalist has a

consistent view that can withstand arguments against it. For my purposes here, however, more would

be needed.

Denying TD4

Suppose we accept PIPL, that a property instance can have no further causal powers than those charac-

teristic of its type. We are thereby led to accept that a higher-level property instance is token-distinct

from its realizer on the occasion. But to move from there to the claim that the higher-level property

for example, Mormann (1995).31The relation of having some among a characteristic set of causal powers would be a way for tropes bear a family

resemblance. Against this view, see, for example, Campbell (1981, p.485), who argues that while tropes can resemblewithout exactly resembling, they cannot resemble by family resemblance.

32This is also enough for Clarke to make his point. See footnote 25.33The same point in a very similar context is made by Robb (2013). He argues that inexact resemblance allows for a

trope-theoretic account of multiply realized properties, as part of an overall program of casting (type-identity) physicalismin terms of trope theory. See also Robb (1997). In contrast, see Gibb (2004).

34Such as dispositional essentialism.


instance is token-distinct from the instance of the disjunction of its realizers requires TD4, that an

instance of a disjunction of properties is just an instance of one of the disjuncts.

At first glance, it may appear difficult to deny TD4. While I have seen no argument for it, there

is a very strong intuition in favour of the premise. There is nothing to a disjunction but the disjuncts

themselves. Indeed, suppose we treat properties as something like a set of property instances (objects

exemplifying universals, tropes, etc.), and suppose we treat a disjunction of properties as just a union

of such sets. Then the set corresponding to a disjunction of properties (a union of sets of property

instances) will have as members just those members of the sets disjoined. Therefore, the instances of a

disjunction of properties are just the instances of the disjuncts. This appears to be the picture behind

intuitions that a disjunction of properties has as instances the instances of its disjuncts.

While the picture is a seductive one, it fails to take into account the significance of Clapp’s character-

ization of disjunctive properties. In short, a disjunctive property whose disjuncts exhaustively overlap on

a set of causal powers is characterized by that set. In other words, the set of causal powers that charac-

terizes the disjunctive property is the intersection of the causal powers of the (exhaustively overlapping)

disjuncts. So, by PIPL, the instances of such a disjunctive property have at most the causal powers of

that intersection. Therefore, in general, they will be token-distinct from the property instances of the

disjuncts.

To elaborate this argument, recall Clapp’s condition for a disjunctive predicate to designate a legit-

imate property.35 A disjunctive predicate designates a legitimate property if its disjuncts exhaustively

overlap on a set of causal powers. This condition follows naturally from Clapp’s account of properties as

being characterized by sets of causal powers. Disjoining predicates that designate properties will create a

disjunctive predicate. Consider the property characterized by the common causal powers (the overlap) of

the properties designated by the disjuncts. The disjunctive predicate is satisfied only on those occasions

on which an object has the property. In other words, if the predicate is satisfied, then the property is

exemplified.36 While the contrary is not always true, it is true when the disjunction is exhaustive. In

that case, every exemplification of the property satisfies at least one of the disjuncts, thereby satisfying

the predicate. In other words, for exhaustively overlapping disjunctive predicates, if the property is

exemplified, then the predicate is satisfied.37 So such a disjunctive predicate is satisfied if and only if

the property is exemplified. This biconditional on the disjunctive predicate and the property ensures

that the predicate designates a legitimate property, since the property characterized by the intersection

of their powers answers to such a predicate.

In contrast, if the disjunction does not overlap on any causal powers, then no (legitimate) property

is designated by the predicate. This is because no property (as characterized by a non-empty set of

causal powers) is exemplified on all occasions satisfying the predicate. Similarly, if the disjunction is not

exhaustive, then no (legitimate) property is designated by the predicate. This is because no property

(as characterized by a non-empty set of causal powers) is exemplified only on those occasions satisfying

the predicate. We can still use the truth conditions of such disjunctive predicates to pick out property

exemplifications. But they do not pick out a legitimate property type. It is just a convenient way to

35For the purposes of precision, here and below I shall talk about disjunctions of predicates and the properties to whichthey refer. Clapp’s (2001) argument is also presented in terms of predicates.

36If properties are characterized by a set of causal powers, this is metaphysically necessary.37The modal strength of this statement depends on the domain of properties from which the disjuncts are drawn. If the

domain is restricted to physical properties, and if physicalism is true, then the statement is true with at least nomologicalnecessity. The non-reductive physicalist can exploit a modal strength weaker than metaphysical necessity, by claiming thatthe higher-level property has fundamentally non-physical realizers in physically impossible worlds. See section 2.4 below.See also footnote 36.


talk about multiple properties at once. In such cases it makes sense to say that the instances of the

disjunction of properties are just the instances of the disjoined properties. Instances of jade just are

instances of jadeite or nephrite.

Clapp showed that by disjoining predicates we can build predicates that designate legitimate prop-

erties characterized by fewer causal powers than those designated by any of the disjuncts. This succeeds

only when the disjuncts exhaustively overlap on a set of causal powers, and in that case, the predicate

refers to the property characterized by that set of powers. If so, then a disjunctive property, i.e., the

property designated by an exhaustively overlapping disjunction of predicates, is characterized by a set of

causal powers that is in general smaller than the set of causal powers characterizing any of (the properties

designated by) its disjuncts. So, by PIPL, the disjunctive property instance has fewer causal powers than

those of the instance of the disjunct on the occasion. Just like the case of the higher-level property and

its realizer, they are token-distinct. The argument from token-distinctness is blocked, and nothing stops

the reductive physicalist from type-identifying the higher-level property and the disjunctive realizer.

There is still room for the non-reductive physicalist to object. In particular, the ontological status of

such a disjunctive property can be questioned. Suppose Clapp is correct, and a disjunction can designate

a property with fewer causal powers than any of its disjuncts. Suppose further that PIPL is correct, so

that disjunctive property instances are token-distinct from the disjoined property instances. In that case,

the non-reductive physicalist might wonder if such disjunctions are ontologically innocent anymore. Any

reasons we have to expect a higher-level property to be ontologically guilty (it is distinct, it is causally

autonomous, etc.) apply equally well to disjunctions of this kind. This objection is properly placed

outside arguments against the type-identity of the higher-level property and its disjunctive realizer,

since it grants their type-identity. I will deal with this line of reasoning below, in section 2.3.

Denying TD6

Suppose, however, that we accept that the instance of the higher-level property is token-distinct from

the instance of the disjunctive property. In order to complete the argument against their type-identity,

it needs to be the case that type-identity is incompatible with this sort of token-distinctness. This is

just what premise TD6 says.

On the face of it, TD6 is plausible, and likely not controversial. Still, for the sake of exploring the

logical space available to the reductive physicalist, I shall attempt to show that denying TD6 is a viable

way to respond to the argument against type-identity from token-distinctness. To this end, I will first

discuss why TD6 is plausible, and then explore the consequences of denying it.

Consider that token-distinctness is not per se incompatible with type-identity. Indeed, in general,

types have more than one token. Two different property instances of the same type (say, the red on an

apple and the red on a ribbon) are token-distinct but type-identical. This is why, according to TD6, it is

not general token-distinctness that is incompatible with type-identity. Rather, it is a particular sort of

token-distinctness, one where the distinct tokens overlap on some (but not all) of the same token causal

powers.

The intuition behind forbidding the type-identity of such tokens is akin to a worry about overdeter-

mination. The concern is that more than one token of the same type would have to be instantiated on

the same occasion: in the same spatiotemporal location, with the same causal history and same effects.

This crowding of tokens of the same type offends against our understanding of property instances, more


so that the crowding would be (at least nomologically) necessary.38 Contrast this with type-distinct

property instances being instantiated on the same occasion. This is, in principle, no different than an

object exemplifying two properties (like an apple being both red and round). There may be a lingering

worry that these type-distinct properties share token causal powers, but the sharing is meant to explain

the necessary relation between the two types. This justification is not available when there is only one

type in question.

Nevertheless, it is still an open possibility that the two distinct tokens, the higher-level property

token and the disjunctive realizer token, are type-identical. In particular, PIPL demands that property

instances have no further causal powers than those characterized by their type. It puts an upper limit on

the number of causal powers that a property instance can have. But it does not put a lower limit on them.

Thus, PIPL forbids the disjunctive property from being of a (supposed) higher-level type, but allows

the higher-level token to be of the disjunctive property type. On every occasion that an exhaustively

overlapping disjunctive property type is instantiated, it would have two tokens: the instance of the

disjunct, and a higher-level instance whose causal powers are the intersection of those of the disjuncts.

Indeed, an advocate of local reduction might even say that the higher-level property token is a distinct

token of the same type as its realizer on the occasion.

At the extreme, this paints an unfamiliar metaphysical picture. Robust tokens of physical types are

necessarily co-instantiated with anemic echos of the same type, each with a smaller set of causal powers

than the last, but sharing the same token causal powers. But there is something attractive about this

view. If a sufficient condition for being a distinct token is to have a different set of the same causal

powers, and sets are merely mathematical entities, then there is some plausibility that distinct tokens

should not be a significant burden on our broadly scientific ontology. One way for this to be is for such

tokens to be of no new type.

The only negative consequence of such views appears to be that they are contrary to our expectations

of property instances. Co-instanced property instances of the same type, overlapping on some (but not

all) of the same token causal powers, just seem bizarre.39 But, unless we have a general proscription

against properties having fewer powers than their type,40 nothing appears to forbid it. It is therefore

an available strategy for the reductive physicalist to resist the claim that the token-distinctness of a

higher-level property and the disjunction of its realizers implies their type-distinctness.

2.3 Denying Ontological Innocence

The Disjunctive Strategy against the Multiple Realizability Argument is to type-identify the higher-level

multiply realized property with the disjunction of its realizers. Since a disjunction is an ontologically

innocent combination of its disjuncts, this is meant to save reductive physicalism. In the previous section,

one way to respond to the strategy was addressed: denying the type-identity of the higher-level property

and the disjunction of its realizers. I argued there that these responses do not succeed. In this section,

38There is perhaps less need to worry about causal overdetermination, though. Whatever reason the non-reductivephysicalist has to suppose that sharing token causal powers avoids causal overdetermination in the case of distinct typesis a reason to suppose the same in the case of identical types.

39In fact, two property instances of the same type overlapping on all of the same token causal powers also seems bizarre.40Clarke (1999) makes a strong case that there is such a proscription. See section 2.2.3 and footnote 21. Clarke’s

argument that a property instance cannot have fewer causal powers than those that characterize its type withstands theargument I give in section 2.2.4, except for the argument that trope types can be similarity relations. See footnotes 25and 32, and the surrounding text.


I will address a second way to respond: denying that the disjunctive property (as described by Clapp)

is ontologically innocent.

The non-reductive physicalist denies the type-identity of a higher-level property and the disjunction

of its realizers owing to some difference in status between them; the higher-level property is special

in some way that the disjunctive property is not. In the case of Fodor, the higher-level property is a

kind, while the disjunctive property is not. In the case of Wilson, the higher-level property instance is

token-distinct from its realizer on the occasion, while its disjunctive realizer instance is not. In both

cases, there are two broad strategies to re-establish type-identity.41 One is to deny the specialness of

the higher-level property. Against Fodor, the higher-level property is not a kind. Against Wilson, the

higher-level property instance is token-identical to its realizer. The second way, more relevant to this

section, is to assert that the disjunctive realizer is also special. Against Fodor, the disjunctive realizer is

also a kind. Against Wilson, the disjunctive realizer instance is also token-distinct from the realizer.

The reason why the second broad strategy, asserting the specialness of the disjunctive realizer, is

relevant is that it opens up an opportunity for the non-reductive physicalist. If the higher-level property

just is its disjunctive realizer, then, given that they are both special, whatever reasons the non-reductive

physicalist has for believing that higher-level properties are not ontologically innocent are reasons to

believe that the disjunctive realizers are also not ontologically innocent.

In particular, by drawing a distinction between ordinary disjunctions of properties (like jade) and

disjunctions that are exhaustively overlapping, the reductive physicalist appears to be committed to an

ontological difference between the two. The ordinary disjunctions of properties aren’t characterized by

a set of causal powers, and their instances are just instances of the disjuncts. This is the model case of

ontological innocence. On the other hand, exhaustively overlapping disjunctive properties are character-

ized by a set of causal powers, and their instances are (perhaps) token-distinct from the instances of the

disjuncts.42 Perhaps exhaustively overlapping disjunctive properties are not ontologically innocent after

all, especially if (as I argued is an available option for the reductive physicalist) they are token-distinct

from their disjuncts. In that case, we are not merely recategorizing physical property instances; new

property instances are said to exist.

Below I will outline arguments by non-reductive physicalists for why token- and type-distinctness of

this kind suggests that the properties in question are not ontologically innocent. This will be followed

by a reductive physicalist reply.

2.3.1 Argument Against Ontological Innocence

An argument for the ontological guilt of exhaustively overlapping disjunctive properties can be recon-

structed from Wilson (2011). There, she argues that properties that satisfy SCCP are ontologically

distinct and can be causally autonomous.43 Exhaustively overlapping disjunctive properties are charac-

terized by the intersection of the sets of causal powers of their disjuncts. Suppose that a token instance

of a property cannot have more causal powers than is characteristic of its type. In that case, an exhaus-

tively overlapping disjunctive property instance has a proper subset of the causal powers of the instance

of its disjunct on the occasion. Thus, exhaustively overlapping disjunctive properties satisfy SCCP. It

41We can set aside the third strategy, viz., arguing that what makes the higher-level property special is irrelevant totype-identity.

42Compare with Fodor (1997).43Recall that, by SCCP, higher-order property instances have a proper subset of the causal powers of their realizers on

the occasion.


follows from Wilson’s argument that such disjunctive properties are ontologically distinct and can be

causally autonomous. If ontological distinctness and causal autonomy are sufficient for ontological guilt,

then exhaustively overlapping disjunctive properties are not ontologically innocent.

The first part of Wilson’s argument involves an ontological distinctness claim. Quite simply, the

higher-level property instance is distinct from the lower-level instance, by Leibniz’s Law.44

The second part of Wilson’s argument involves establishing the causal autonomy of the higher-level

property instance. Two arguments are presented, according to which (1) higher-level properties might

be more proportional to an effect, and (2) they might be the subjects of distinct systems of laws.

First, consider a higher-level property instance and its (token-distinct) realizer being differently

proportional with respect to an effect. The higher-level property satisfies SCCP, so its causal powers

are a proper subset of those of the realizer. Suppose there is a causal event, like Monty opening a door.

Monty had a desire to open the door, and this caused him to open the door. Monty’s desire was realized

by a specific physical brain state which also (in virtue of the common causal powers that were activated

on the occasion) caused him to open the door. Monty’s desire is, however, more counterfactually robust

than his brain state. Had things been just a little different, the brain state would have been different,

but (plausibly) the new brain state would have realized his desire, and he would have opened the door.

As Wilson says,

the only powers that matter for the production of [the effect] are the powers associated with[the mental state]; powers differing between [the two brain states] (say, to produce a certainreading on a neuron detector) are irrelevant for [the effect’s] production. (Wilson, 2011,p.129)

So Monty’s desire is more proportional to his opening the door than is his brain state; the former has the

right number of powers, while the latter has excess powers irrelevant to the production of the effect. This

difference in proportionality can ground causal autonomy among causally non-competing properties.

Second, consider that distinct sets of causal powers can be the subject of distinct systems of laws.

On this view, systems of laws “track causal joints in nature” (Wilson, 2011, p.130), and the higher-level

property’s distinct set of powers indicates a distinct causal joint. Such a position relies on our strong

intuitions about the relationship between laws and kinds: laws are about kinds, and kinds are the sorts

of things that figure in laws.45 So, while both Monty’s desire and his brain state cause him to open the

door, they do so as part of distinct systems of laws: the former by the laws of psychology, the latter by

the laws of physics.

2.3.2 Denying the Argument Against Ontological Innocence

In this section, I reply to the argument that properties satisfying SCCP are ontologically guilty. I do it in

three subsections. First, I address the ontological distinctness of higher-level property instances. I won’t

deny it, but I will argue against its significance. In the second and third subsections, I reply to the claim

44Curiously, an appeal to Leibniz’s Law is prima facie in tension with one of Wilson’s arguments for the token-distinctnessclaim. Recall that, to motivate the claim that a higher-level property is token-distinct from its realizer, Wilson arguedthat intuitions about tokens being of more than one type were unreliable. She argued that such intuitions come fromour understanding of objects having properties, but a property instance having causal powers is unlike an object havingproperties. Here, however, causal powers are enough like properties to ground an appeal to Leibniz’s Law, even thoughthe latter typically applies to identical objects having all and only the same properties.

45This seems to be the basis of Fodor’s argument. Before commenting about kinds and laws forming a circle of concepts,he says: “the natural kind predicates of a science are the ones whose terms are the bound variables in its proper laws”(Fodor, 1974, p.102).


that a higher-level property satisfying SCCP is causally autonomous. In the second subsection, I address

the argument that having a distinct set of causal powers can lead to a difference in proportionality with

respect to some effect. In the third subsection, I address the argument that being a part of a distinct

system of laws gives a property claim to causal autonomy with respect to that system.

Ontological Distinctness

Recall Wilson’s argument that property instances satisfying SCCP are ontologically distinct from in-

stances of their realizers. We can grant this as an assumption; denying it was the subject of section 2.2.4,

and here we grant that both the higher-level property instance and its disjunctive realizer instance are

token-distinct from the realizer on the occasion.

Nevertheless, while necessary, mere ontological distinctness is not enough for ontological guilt.46 A

distribution of particles is distinct from the particles distributed, but this is not a worry for the reductive

physicalist. More generally, not every ontological debate in metaphysics is ipso facto germane to whether

reductive or non-reductive physicalism is true. Metaphysicians often argue whether wholes are distinct

from parts, or relations are distinct from the monadic properties of the relata, or whether properties are

distinct from sets of objects; but these arguments are prima facie separate from the issue of whether

all broadly scientific properties are reducible to physical properties. At the very least, the fact that the

properties of interest are broadly scientific should play a role in the debate.

This is the attitude that a reductive physicalist can take to ontologically distinct higher-level property

instances. A higher-level property is just an exhaustively overlapping disjunctive property. Instances of

the latter are ontologically distinct from the instances of the properties disjoined because the disjunctions

instantiate a real relation (or something very much like a relation) among the properties disjoined.

Arbitrary disjunctions do not, and so do not have distinct instances. An imperfect analogy would be to

conjunctions. Sometimes, the conjunction of property instances instantiates a new property, a relation

between the properties. Other times, as when the conjoined properties are causally isolated, there

is no real relation between them. Just as reductive physicalists are unconcerned about the ontological

distinctness of (on the one hand) relations holding among fundamental particles and (on the other hand)

the fundamental particles themselves, they are unconcerned about the ontological distinctness of tokens

of disjunctions and the disjuncts.

At the very least, in addition to ontological distinctness, what is needed to make the disjunctive

property instance ontologically guilty is a distinct role in the causal goings-on. Unsurprisingly, the

more important parts of Wilson’s argument attempt to establish the causal autonomy of the higher-level

property.

Causal Autonomy: Proportionality

Consider again the argument that a property satisfying SCCP can be causally autonomous with respect

to its realizer because the former might be more proportional to the production of an effect. According

to this argument, Monty’s desire to open the door is more proportional to his opening the door than

is his brain state. This is because his desire is more counterfactually stable than his brain state with

respect to the production of the effect; had the world been slightly different, his brain state would have

been different, but he still would have had the desire, and he still would have opened the door. This is

46See also section 1.1.5.


explained by his brain state having causal powers (like producing a certain reading on a neuron detector)

that are not relevant to the production of the effect. Call this the Argument from Proportionality.

In this section, I argue that this counterfactual stability is just a special case of the counterfactual

stability of disjunctions, and so does not ground the causal autonomy of higher-level properties. First,

I elaborate the Argument from Proportionality. I show that, with respect to the production of some

effect, E, the counterfactual stability of the higher-level property, M , relative to its realizer, P , is a

result of M having a proper subset of those causal powers of P that are not only irrelevant to the

production of E, but also activated on the occasion. Second, I set up my reply. I do so by making

some simplifying assumptions, and showing that the only way for M to have a proper subset of the

irrelevant activated causal powers of P (with respect to the production of E) is for P to cause an effect

that is neither caused by M nor a determinable of E. Finally, I will reply to the argument. I will show

that, when the above condition is satisfied in the most plausible way, the counterfactual conditionals

whose truth establishes the counterfactual stability of the higher-level property can be interpreted as

having disjunctive consequents. This makes the counterfactual stability of the higher-level property just

a special case of the counterfactual stability of disjunctions.

Refining the Argument from Proportionality

Before looking carefully at whether the Argument from Proportionality can establish the causal auton-

omy of the property satisfying SCCP, it is worth elaborating it.47 As it stands, the explanation for why

Monty’s desire is more counterfactually stable with respect to his opening the door is not quite right.

While it is true that his brain state has causal powers not relevant to the production of the effect, so

too does Monty’s desire. Under other circumstances, his desire would cause him to look for his keys,

or panic, or a host of other (context-dependent) effects. Not every such causal power is relevant to his

opening the door, on the occasion of his opening the door. So merely having extra causal powers is not

enough to establish a lack of proportionality.

Indeed, there is a greater danger lurking in this line of thinking. Suppose an effect type is defined as

the production of a set of property types.48 Suppose further that merely having causal powers that are

not relevant to the production of some effect is enough to make a property fail to be proportional to the

production of that effect. In that case, the only property that would be proportional to the production

of an effect of a given type is a property characterized by that very small set of causal powers (possibly

as few as one causal power) whose activation produce just that effect. The only property proportional in

this way to Monty’s opening the door is the one characterized by just those causal powers whose effect

is Monty opening the door. At best, this would ground the causal autonomy of radically proscribed

properties that can cause only one type of effect, not the causal autonomy of properties (like mental

properties or other special science properties) that can and do cause a variety of effects in multiple

circumstances.

On the other hand, suppose that we bite the bullet and say that, while such radically proscribed

properties are maximally proportional to some effect (and so are, in some sense, maximally causally au-

47Wilson’s primary concern for this argument is defensive. She wants to argue that non-reductive physicalism is a livingpossibility in spite of causal overdetermination arguments against it. For this reason, it is enough to show how higher-level properties might be causally autonomous in spite of sharing token causal powers with its realizer. For my purposes,however, more will need to be said.

48We can remain neutral to whether the properties in question are all the properties that are instantiated on everyoccasion of the effect’s production, or if some subset of those properties is in some way essential to the effect. See, forexample, Yablo (1992).


tonomous), properties that satisfy SCCP are more proportional to the production of certain effects than

their realizers. Perhaps this can ground enough causal autonomy to guarantee ontological significance.

One way for a higher-level property to be more proportional to the production of an effect than its

realizer is for the former to have fewer irrelevant causal powers than the latter. But even grounding

proportionality in having fewer irrelevant causal powers is not quite right. There is a very real possibility

that the properties of interest to special scientists have an infinite (and possibly an uncountably infinite)

number of causal powers.49 If so, then in general, the cardinality of the set of causal powers of the

higher-level property that are irrelevant to the production of some effect is likely equal to the cardinality

of the set of causal powers of its realizer that are irrelevant to the production of that effect. Thus, merely

having fewer irrelevant causal powers cannot ground proportionality.

What is needed is a tighter connection between the irrelevant causal powers of the higher-level

property and those of its realizer, and for that connection to mirror the relationship between the higher-

level property and the realizer. Fortunately, the satisfaction of SCCP by the higher-level property gives

us that relationship. The irrelevant causal powers of the property satisfying SCCP are a proper subset

of the irrelevant causal powers of the realizer. There is then a principled way to claim that a higher-level

property is more proportional to the production of an effect than its realizer: its irrelevant causal powers

are a proper subset of those of its realizer.

However, some reflection shows that the above elaboration is not quite complete. The reason that

the issue of the realizer having irrelevant causal powers was raised in the first place was to explain

the counterfactual stability of the higher-level property with respect to the production of the effect.

However, the relative lack of irrelevant causal powers is, in general, not pertinent to the counterfactual

stability of the higher-level property with respect to the production of the effect. Plausibly, whatever

the individuating conditions of the effect, those conditions have to be present on the occasion of the

production of the effect. Monty’s desire is counterfactually stable because, had things been a little

different, those of its causal powers that produced his opening the door would still have been activated.

So the counterfactual stability of the higher-level property with respect to the production of the effect,

relative to its realizer, is not because it has a proper subset of its realizer’s irrelevant causal powers.

Rather, it is because the higher-level property has a proper subset of its realizer’s irrelevant causal powers

activated on that occasion.

We might even be tempted to go further: that the higher-level property has no irrelevant causal

powers activated on that occasion. But this would be a mistake. On the occasion of his opening the

door, Monty’s desire also had causal powers that are both activated and not relevant to his opening

the door. For example, his desire, combined with whatever conditions existed that activated the causal

power in virtue of which he opened the door, might have caused his anticipation of the door opening.

In similar circumstances, perhaps those in which he is unsure of whether the door is locked, his desire

might cause the door to open but not the anticipation. As long as the individuation conditions of his

opening the door allow type-identical effects under changes such as these, Monty’s desire had at least

some irrelevant activated causal powers. This generalizes. A property will plausibly have distinct causal

powers with activation conditions that are compatible. If the activation conditions are compatible, then

they can be satisfied simultaneously, in which case both causal powers will be activated on a single

occasion. As long as the individuation conditions of one of the effects are independent of the production

49This depends strongly on the individuation of causal powers. Plausibly, causal powers can be individuated by theiractivation conditions and the effects that are produced. Since there are uncountably infinite different circumstances inwhich a property might cause, the property likely has an uncountably infinite number of causal powers.


of the other, a property will have irrelevant activated causal powers. Thus, the better account of causal

autonomy is relative proportionality (grounded in having a proper subset of irrelevant activated causal

powers) rather than absolute proportionality (grounded in having no irrelevant activated causal powers).

Let us pause here to take stock. The non-reductive physicalist argues that a higher-level property

that satisfies SCCP is causally autonomous from its realizer. One reason for this is that, in satisfying

SCCP, its activated causal powers that are irrelevant to the production of certain effects are a proper

subset of those of the realizer, making it more proportional to the production of those effects. This

grounds the counterfactual stability of the higher-level property with respect to the production of the

effect. The more irrelevant activated causal powers a property has, the more circumstances can be

configured so that the property is no longer present but the effect is still produced.50 The realizer fails

to cause the effect in all the circumstances in which the higher-level property does, and more besides.

This picture, I think, maps quite well onto our intuitions about the case. It is also compatible

with accounts of so-called dependent causation, wherein higher-level properties cause other high-level

properties in a way that is dependent on low-level properties causing other low-level properties.51 It also

appears to be a good translation of the account of causal proportionality given by Yablo (1992) in terms

of causal powers.52

Setting Up the Reply

To reply to this argument, and to show that this kind of proportionality is not enough to ground causal

autonomy, I will make a few simplifying assumptions. While not essential to the reply, the assumptions do

reveal important underlying features of the connection between sets of causal powers and counterfactual

stability. I will show that, with respect to some target effect, a higher-level property will have a proper

subset of the irrelevant activated causal powers of its realizer only if the realizer causes an effect that is

neither caused by the higher-level property nor a determinable of the target effect.

First, I will assume that causal powers are individuated in part by their effects, so that distinct

causal powers are required to produce distinct effects. This immediately raises a need to individuate

effects. I will assume, as above, that effect types are some subset of the property types instantiated on

the occasion of the production of the effect. Token effects will thus be a set of property instances of

those types.

Furthermore, I will assume that the only causal powers that are activated on a given occasion are

those of a single physical property. In reality, of course, there will be a network of properties whose

causal power activations collectively produce all the effects (and therefore all the properties that are

instantiated) on a given occasion of causation. The assumption can be interpreted as focusing on one

element of the total causal picture at a time. It will allow me to investigate the causal contribution

50Causal powers characterize properties, so, in our counterfactual reasoning, changing which causal powers are presentchanges which property is present. The presence of irrelevant causal powers can be changed without failing to producethe effect. Therefore, the more irrelevant causal powers are present, the more counterfactually fragile the property iswith respect to the production of that effect. Each independent irrelevant causal power is a degree of freedom in ourcounterfactual reasoning.

51See, for example, Witmer (2003).52Yablo (1992) described events as coinciding when they share all the same non-modal property instances, but distinct

when they have different essences (subsets of those same non-modal property instances). This invites an obvious translationinto causal powers talk: property type exemplifications (contra events) coincide when they share the same causal powertokens (contra non-modal property instances), but are distinct when they have distinct property instances (contra essences),i.e., when they have different subsets of those same causal power tokens (contra non-modal property instances). WhenYablo talks about events determining other events, we can talk about properties realizing other properties via SCCP. Thetough part of this translation is describing Yablo’s counterfactual conditions for causation (his CARE conditions), whichcorresponds to causal proportionality, in terms that are already modally loaded (causal powers).


of a single physical property and those of the properties it realizes, bracketing any other effects caused

exclusively by other properties as outside the scope of the investigation. Any interference effects from

those other properties (where the other properties enhance, alter, or diminish the effect of the physical

property in question) can be incorporated as part of the activation conditions of the physical property’s

causal powers.

Finally, note that some of the property instances produced on a given causal occasion are related

to other such property instances, mirroring how their characteristic sets of causal powers are related.

When one property instance is realized by another, that is, when one property instance satisfies SCCP

with respect to another, call the former a determinable of the latter, and the latter a determinate of the

former. These relations are also mirrored in sets of property instances, like effects. When every property

instance that defines an effect is either token-identical to, or a determinable of, every property instance

that defines another effect, call the former a determinable of the latter, and the latter a determinate of

the former.53 I will also assume that some effects are maximally determinate, i.e., they themselves have

no determinate effects. Call a maximally determinate effect together with all its determinables a family

of effects.

The end result of these assumptions is the following (somewhat simplified) picture of causal powers.

Let M be the higher-level property, let P be its physical realizer, and let M satisfy SCCP with respect

to P . On the causal occasion in question, P is instantiated in some circumstances that are sufficient for

the activation of some the causal powers of P . Each of these causal powers activates, and each produces

a single effect. Every effect produced on the occasion is caused by P : every maximally determinate

effect and all their determinables. M , on the other hand, has a proper subset of the causal powers of

P . If M causes on that occasion a proper subset of the effects produced (by P ), then M has a proper

subset of P ’s activated causal powers.54

However, in order for M to have a proper subset of P ’s irrelevant activated causal powers (with

respect to some effect E produced by both M and P ), one further (very plausible) condition must be

met. On the occasion of the production of E, at least one effect, F , caused by P but not by M , must

fail to be a determinable of E.55 Since F fails to be a determinable of E, the definition of F includes

a property that neither is itself part of the definition of E, nor has determinates that are part of the

definition of E. The presence of that property makes F irrelevant to the production of E.56 If the

condition is met, the irrelevant effects caused by M will be a proper subset of those caused by P . It

follows that M has a proper subset of the irrelevant activated causal powers of P . Thus, if M is more

counterfactually stable than P with respect to E, it must be the case that P causes at least one effect

not caused by M that is also not a determinable of E.

For example, suppose Monty’s brain state causes only two families of effects on the occasion of his

opening the door: (1) the maximally determinate physical state of his opening the door together with

53In other words, the union of the token causal powers of all the property instances of one set of properties is a propersubset of the union of the token causal powers of all the property instances of the other set. This is a version of SCCPapplied to sets of properties.

54This is not a biconditional since, if two of P ’s causal powers activate to overdetermine a single effect, and if M hasone of the causal powers but not the other (and has all of P ’s other activated causal powers), then M can have a propersubset of the activated causal powers of P without producing a proper subset of the effects produced by P .

55Besides being very plausible in its own right, this condition follows from the requirement that P cause an effect notcaused by M under the plausible supposition that, by metaphysical necessity, any property with a causal power to producean effect has the causal power to produce every determinable of that effect.

56At least in the sense pertinent to the Argument from Proportionality. If F were a determinable of E, then F wouldnecessarily be instantiated whenever E is instantiated. Therefore, F could not be used to ground the counterfactualfragility of P relative to M in the production of E.


its determinables (like opening the door and opening something); and (2) a maximally determinate

displacement of air together with its determinables. Monty’s desire to open the door is, by hypothesis,

more counterfactually stable than his brain state with respect to his opening the door. Therefore,

his brain state must cause an effect that his desire does not, and that effect cannot be one of the

determinables of his opening the door (like the effect of opening something).

Replying to the Argument from Proportionality

Using the above picture of causal powers, we can problematize the claim that higher-level properties

are causally autonomous with respect to their realizers. I will argue that the counterfactual stability

of the higher-level property is just a special case of the counterfactual stability of a disjunction. The

counterfactual stability of the higher-level property relative to its realizer with respect to some effect is

established using two counterfactual conditionals: (1) had the realizer not been instantiated, the higher-

level property would still have been instantiated, and (2) had the realizer not been instantiated, the

effect would still have been instantiated. Regarding the first conditional, it has already been argued

that the higher-level property can be identified with the (exhaustively overlapping) disjunction of its

realizers. Regarding the second conditional, I will establish below that the defining properties of the

effect can also be identified with exhaustively overlapping disjunctions.

Consider again the example of Monty. In order for Monty’s desire to be more counterfactually stable

than his brain state, his brain state must cause an effect that his desire does not, and that effect cannot

be one of the determinables of his opening the door There are two ways for this to be the case. The

effect caused by his brain state but not by his desire either is in a different family of effects than his

opening the door (like the displacement of air), or is one of the determinates of his opening the door

(like the maximally determinate physical state of his opening the door). This generalizes. Let M be a

higher-level property, let P be its physical realizer, let M satisfy SCCP with respect to P , and let E be

an effect caused by both M and P . If M is more counterfactual stable than P with respect to producing

E (i.e., if M has a proper subset of the irrelevant activated causal powers of P with respect to E), there

must be an effect caused by P but not by M that fails to be the determinable of at least one effect

produced by M . That effect either is in a different family from E, or is a determinate of E.

This result yields the two (non-exclusive) ways that M can cause a proper subset of the (irrelevant)

effects caused by P . The first way, the prima facie less plausible way, is for M to be less comprehensive

than P . P is comprehensive because it causes (some member of) every family of effects produced on the

occasion. In contrast, M might fail to cause all the members of some family. Perhaps Monty’s desire

produces the maximal physical state of his opening the door (and all its determinables), but fails to

cause the maximally determinate displacement of air (and any of its determinables). The second way,

the prima facie more plausible way, is for M to be less specific than P . P is specific because it causes

the most determinate effect of every family of effects produced on the occasion. In contrast, M might

fail to cause the most determinate effect of the family of E, but does cause some of its determinables

(including E). Perhaps Monty’s desire fails to cause the maximally determinate physical state of his

opening the door, but does cause his opening the door (along with its determinables).

I will not dwell on M ’s (possible) lack of comprehensiveness. It is unlikely that a typical special

science property, like Monty’s desire, is less comprehensive but not less specific than its physical realizer.

But note that, if M is less comprehensive but not less specific than P , then every realizer of M causes

the same maximally determinate effect (plus all its determinables) in the circumstances in which P


causes it. This makes M more counterfactually stable than P with respect not only to E, but even to

the maximally determinate effect of which E is a determinable. If Monty’s desire is less comprehensive

but not less specific than his brain state, then every realizer of Monty’s desire causes the very same

maximally determinate physical state of his opening the door as his brain state does. So Monty’s desire

is more counterfactually stable than his brain state with respect not only to his opening the door, but

even to the maximally determinate physical state of his opening the door.

More likely, it is in virtue of M ’s lack of specificity that M is more counterfactually stable than P

with respect to E. As with comprehensiveness, this lack of specificity on M ’s part ensures that M ’s

irrelevant activated causal powers are a proper subset of those of P . But this time it does so in a way

that mirrors our intuitions. Both Monty’s desire and his brain state caused his opening the door. His

brain state also caused the precise physical configuration of the Monty-door system. Monty’s desire is

more counterfactually stable than his physical state with respect to his opening the door because, had

things been a little different, his desire would have been instantiated and his opening the door would

have been instantiated, but his physical state and that configuration of the Monty-door system would

not have been.

When M is less specific than P , E is a determinable of at least one maximally determinate effect

for every realizer of M . Recall that a determinable effect is one whose defining properties are either

identical to, or determinables of, a property of some other effect. Since every determinable property

is an exhaustively overlapping disjunction of its determinates, every defining property of E is either a

maximally determinate property, or an exhaustively overlapping disjunction of such properties.57

We are now in a position to re-interpret this counterfactual stability in terms other than causal

autonomy. First, note that, by SCCP, the higher-level property (Monty’s desire) is a determinable of

each of its realizers (including Monty’s brain state), and we have seen that such a higher-level property

can be type-identified with the (exhaustively overlapping) disjunction of all its realizers. So the first

counterfactual conditional, had P not been instantiated then M would still have been instantiated, can

be interpreted as a conditional with a disjunctive consequent. Second, note that, because the higher-level

property lacks specificity, the effect with respect to which the higher-level property is counterfactually

stable (Monty opening the door) is a determinable of some effect of each of the higher-level property’s re-

alizers. The determinable effect can also be identified with a set of exhaustively overlapping disjunctions

of properties (with non-disjunctive properties being limiting cases). So the second counterfactual con-

ditional, had P not been instantiated then E would still have been instantiated, can also be interpreted

as a conditional with a disjunctive consequent.

If so, it no longer seems apparent that M is causally autonomous. Had P not been instantiated,

one of a number of properties would have been instantiated. Moreover, had P not been instantiated,

one of a number of effects (or, more precisely, an effect defined as a set whose members are each one

of a number of properties) would have been instantiated. The source of the counterfactual stability

of an exhaustively overlapping disjunction is transparent: it is a special case of the counterfactual

stability of disjunctions. Any disjunction of contingent, nomologically possible properties will satisfy

more counterfactual conditionals than any disjunct will, since there are more nomologically possible

worlds in which any of a number of properties is instantiated than any particular one of them. What is

unique to exhaustively overlapping disjunctions is that they disjoin disjuncts that could easily take one

57If M is less comprehensive but not less specific than P , the above also holds. When E is itself a maximally determinateeffect, every defining property of E is a maximally determinate property. This is a limiting case of an effect being definedby a set of either maximally determinate properties or an exhaustively overlapping disjunctions of properties.


another’s place in modal reasoning. In the nearest possible worlds where the systems instantiating P are

otherwise, other members of the disjunction of realizers are instantiated. And in those worlds, while P ’s

(maximally determinate) effect is not instantiated, other (maximally determinate) effects whose defining

properties are each members of similar disjunctions are instantiated.

We may be tempted to interpret this as causally significant, since such counterfactual conditionals

play an important role in our understanding of causation. But this, I think, would be a mistake. The

causal powers of the properties being disjoined, and their overlap in the case of exhaustively overlapping

disjunctions, already establish causation, and thereby map out the geography of nomologically possible

worlds. In order to extend our modal reasoning, we then collect together the interchangeable properties

in nearby worlds into a disjunction, and since they are the ones that are nearby, they will share causal

powers. To rely then on the counterfactual stability to establish causal autonomy seems to me to be

double-counting.

Causal Autonomy: Systems of Laws

Another way that properties satisfying SCCP can be argued to be causally autonomous is that they are

part of a distinct system of laws. There is a tight connection between lawhood, kinds, and causation.

If laws do indeed mark legitimate properties, and if causal explanations appeal to laws, then, plausibly,

the properties that figure in laws are causally autonomous.

Against this argument, the reductive physicalist can just deny that such laws are distinct enough

to ground causal autonomy. If physicalism is true, then all nomologically necessary statements about

properties satisfying SCCP will be entailed by the law statements about physical properties. A fortiori,

every law statement about a property that satisfies SCCP is entailed by law statements about physical

properties. The reductive physicalist can insist that law statements about properties satisfying SCCP

are just notational variants of law statements about physical properties. In that case, any explanatory

systems that include the former are just notational variants of systems that include the latter.58

It is unsurprising that the reductive physicalist and the non-reductive physicalist would have dif-

ferent ideas of what the laws are. As mentioned above, laws, causal powers, and kinds form a circle

of concepts. Whatever reasons a non-reductive physicalist has for supposing higher-level properties are

causally autonomous are reasons for supposing the laws about them are distinct enough to ground that

autonomy. Whatever reasons a reductive physicalist has for denying the former are reasons for denying

the latter.59

Moreover, by insisting that there are ontologically significant laws beyond the physical laws, and

using this to ground the causal autonomy of the properties figuring in those laws, the non-reductive

physicalist faces a problem. Consider the set of nomologically necessary statements, some subset of

which are the law statements. Suppose further that a property is causally autonomous if it satisfies a

predicate in a law statement.60 There are two natural approaches to deciding which members of the set

are law statements.61 The first is minimal: the members of the smallest set of nomologically necessary

statements from which all other statements can be deduced, the axioms of the set, are law statements.62

58For an opposing view, see Wilson (2010a).59“If we disagree about what is a natural kind, we will probably also disagree about what is a law, and for the same

reasons” (Fodor, 1974, p.102).60Any subset of law statements, plus their entailments, can be considered a system of laws. Thus, any property satisfying

a predicate in at least one law figures in a system of laws, and so is, by hypothesis, causally autonomous.61A third natural approach, that there are no laws, is outside the scope of this argument.62More precisely, any nomologically necessary statement that is an axiom in some axiomatization of the entire set is a

law statement.


These are, plausibly, statements of physical law, and this result is too conservative for the non-reductive

physicalist. The second natural approach is maximal: every member of the set of nomologically necessary

statements is a law statement. This is far too liberal for the non-reductive physicalist. Plausibly, any

combination of causal powers characterizes a property that figures in some system of laws, including

those (like jade) that have previously been rejected as being illegitimate properties.63 This is far too

liberal to establish any meaningful causal autonomy. So the non-reductive physicalist bears the burden

of supplying an account of laws that includes laws about at least some properties satisfying SCCP, but

not arbitrary combinations of causal powers. It seems to me that any reasons a non-reductive physicalist

could supply for such an intermediate account would be reasons the reductive physicalist would reject

it.64

The reductive physicalist can even press further, if somewhat speculatively. There are three broad

conceptions of natural laws: Humeanism, dispositionalism, and realism. Under none of these conceptions

is it unproblematic that a property satisfying SCCP is causally autonomous because it figures in a distinct

system of laws.

According to the Humean conception, laws are just certain types of regularities. The most popular

Humean account of laws is that laws are those axioms of the scientific theory that best balance strength

and simplicity (Lewis, 1973). In that case, the axioms will be the physical laws, and the properties

figuring in them will be the physical properties. Again, accounts that aim to make a principled distinction

between some entailments of these axioms and others have to appeal to considerations that reductive

physicalists will likely deny.65

The second conception of a natural law is dispositionalism. According to this conception, laws

arise from the metaphysically more fundamental dispositions possessed by properties.66 Under such a

conception, genuine lawhood can be argued on the basis of which sets of causal powers are had by causally

autonomous properties. But the reverse gets us nowhere; the properties are not causally autonomous

because they figure in laws, the laws are genuine because they arise from causally autonomous properties.

The final conception of laws is the realist conception, according to which laws are “out there” in

the world. This conception seems to be the most promising for the non-reductive physicalist: either

there is an entity in nature (the law) that grounds the causal autonomy of a higher-level property, or

there is not.67 However, at least if we suppose that real laws govern the course of events, there are

problems. The non-reductive physicalist characterizes properties according to causal powers in order to

avoid problematic causal overdetermination. If, by SCCP, higher-level properties and their realizers share

token causal powers, then there is no causal competition between them, since there was only one causing.

But if they figure in different systems of real laws, then the (real) laws are competing to govern them.

The problem of causal overdetermination is just moved to the level of laws, introducing an analogous

problem of governing overdetermination. To illustrate, suppose the laws of thermodynamics are as real

as the laws of statistical mechanics; the former governing the temperature (i.e., mean molecular kinetic

63A description of any causal power is a nomologically necessary statement. Since nomologically necessity is closedunder logical entailment, the conjunction of any such statements results in another nomologically necessary statement.Thus, arbitrarily combined (but compatible) causal powers will always satisfy a predicate in some nomologically necessarystatement. If the latter are law statements, then the former are legitimate properties.

64The best approach that I can think of is one that appeals to the proportionality considerations detailed in the previoussection. As stated there, this can be resisted.

65See Cohen and Callender (2009) for an alternative Humean account that allows for special science laws.66See, for example, Bird (2005).67Especially if the laws of nature are primitive. Some examples of primitivist accounts of laws are Lange (2000) and

Maudlin (2007). Whether the non-reductive physicalist can use either to establish the causal autonomy of special scienceproperties is an open question.


energy) of a gas, and the latter governing the configuration of gas molecules. For any given thermal

effect, the configuration of gas molecules and the temperature are equal, non-competing causes. But the

configuration of gas molecules produced the effect in virtue of being governed by the laws of statistical

mechanics.68 If there was genuinely only one causing, then there appears to be nothing left for the (real,

distinct) laws of thermodynamics to govern.69

2.4 Denying Status as Physical

The Multiple Realizability Argument as it stands does not succeed, since the Disjunctive Strategy is

available to reductive physicalists. By type-identifying the higher-level property with the disjunction of

its realizers, the reduction of the former to an ontologically innocent combination of physical properties

appears to be secured. In previous sections, it was argued that attempts to deny the type-identity of

higher-level properties with the disjunctions of their realizers are unsuccessful, as is an attempt to deny

that such disjunctions are ontologically innocent.

There remains, however, one more response available to the non-reductive physicalist. Let us accept

the type-identity of the higher-level property and the disjunction of its realizers. Furthermore, let

us accept that an exhaustively overlapping disjunction is an ontologically innocent combination of its

disjuncts. The non-reductive physicalist can argue that the higher-level property is specified at a high

enough level of abstraction to have physically impossible realizers (in physically impossible worlds). If so,

then the disjunction to which it is identical may be an ontologically innocent combination of properties,

but not (exclusively) of physical properties. In that case, the disjunction is not itself a physical property,

and the type-identity of the higher-level property with a physical property is blocked.

In this section, I will present this response on behalf of the non-reductive physicalist. I will then

address it by denying that the properties of interest to special scientists in an experimental context have

physically impossible realizers.

2.4.1 Fundamentally Non-physical Realizers

The non-reductive physicalist can claim a level of abstraction for a higher-level property that is even

higher than merely having heterogeneous physical realizers.70 The higher-level property might outstrip

its physical realizers entirely by being realizable by a fundamentally non-physical property.71 If so, then

the disjunction of its realizers is not an ontologically innocent combination of physical properties. As

long as this preserves physicalism, the non-reductive physicalist has blocked the Disjunctive Strategy.

This sort of response touches on an important way in which we think that higher-level properties

transcend physical properties. Higher-level types are not merely more abstract than any physical type,

but more abstract than physicality itself. This is the plausible motivation for the machine functionalism

68Perhaps the laws of statistical mechanics govern the configuration of the gas molecules just in case the causal powersof the configuration of gas molecules arise from the laws.

69As with the resolution of the problem of causal overdetermination, the problem of governing overdetermination mightbe addressed by appeal to a special relationship between the laws governing the higher-level properties and those governinglower-level properties, analogous to the determinable/determinate relation. I suspect that attempting this argument wouldsimply push the problem up another level, the level at which whatever it is in virtue of which laws govern is overdetermined.

70As a reminder, the physicality in question here and below is a-physicality.71If reductive physicalism is true, then every physically acceptable property is a physical property, so a fortiori every

property would have only physical realizers. Therefore, supposing physicalism is true, it is sufficient for non-reductivephysicalism that a property have any non-physical realizer. But the fundamentally non-physical realizers in question arenon-physical properties with no physical realizers. See footnote 3.


of the early non-reductive physicalists, who drew the distinction between computational properties and

the hardware that implements them.72

Below, I describe two ways that a higher-level property might outstrip the physical with fundamen-

tally non-physical realizers. I will then describe the conditions that have to be met by such realizers

in order for their existence to be compatible with both non-reduction and physicalism. Finally, I will

provide an argument on behalf of the non-reductive physicalist that special science properties typically

have such realizers.

Outstripping the Physical

There are two ways for a higher-level property to outstrip its physical realizers.

One is for it to be so abstract that no disjunction of physical realizers has an intersection that is small

enough to characterize the property.73 There may be a lower limit to how small the intersection can

be, since all physical properties that have the higher-level property’s characteristic causal powers might

have further causal powers in common. In order to overlap on only the causal powers of the higher-

level property, further, fundamentally non-physical properties must be added to the disjunction.74 If

the higher-level property outstrips the physical in this way, the exhaustively overlapping disjunction of

physical properties will (at best) characterize its highest-level physical realizer.

A second way is for the exhaustive disjunction of physical realizers to overlap successfully on the

higher-level property’s characteristic causal powers, but to fail to include all realizers that overlap them.

It might be a brute fact that there are fundamentally non-physical properties that overlap on the causal

powers that characterize the higher-level property. If the higher-level property outstrips the physical in

this way, then, as in the case of jade, the disjunction of physical realizers is not a legitimate property

because there are property instances that have the characteristic (intersecting) causal powers but are

not in the disjunction.

Either way, the exhaustively overlapping disjunction to which the higher-level property is type-

identical includes fundamentally non-physical disjuncts, and so is not itself a physical property.

Physicalism, Non-reduction, and Exphysical Realizers

The implementation of this response requires some subtlety in order to ensure that both physicalism

and non-reduction are preserved in the account. The fundamentally non-physical realizer of the higher-

level property must therefore meet certain conditions. First, it must be physically impossible. This

is because a physically possible, fundamentally non-physical property would be in direct contradiction

to physicalism. Second, it must not metaphysically necessitate any physical realizer of the higher-level

property. This is because, otherwise, the fundamentally non-physical property would plausibly realize

the higher-level property by realizing its physical realizer.75 Call a realizer that satisfies these conditions

an exphysical realizer.76 Together, these conditions ensure that this response neither undermines the

non-reductive physicalist project nor fails to accommodate the Disjunctive Strategy.

72See, for example, Putnam (1967).73For example, perhaps it has only one causal power, the power to cause something, where ‘something’ is topic neutral.74Compare footnote 7.75It would also point to a property that is more fundamental than physical properties, so that in some worlds instantiating

our physical properties, they are not the fundamental ones. Our world happens not to be one of those.76Let M be a higher-level property, and let P1, P2, . . . be its physical realizers. U is an exphysical realizer of M if and

only if (1) U is a realizer of M , (2) U is physically impossible, and (3) U does not metaphysically necessitate any ofP1, P2, . . ..


A higher-level property having exphysical realizers certainly is compatible with non-reduction, in the

sense that the disjunctive property to which the higher-level property is type-identical is not a physical

property.77 But it is also compatible with physicalism. Physicalism is true in a world if all broadly

scientific properties are nothing over and above the physical. Recall that my (provisional) account of

a property being nothing over and above another is that the former strongly supervenes with physical

necessity on the latter. A disjunction of physical and exphysical realizers does in fact strongly supervene

on physical properties with physical necessity. This is because the property is instantiated without a

physical realizer being instantiated only in physically impossible worlds. So an exhaustively overlapping

disjunction of physical and exphysical realizers is compatible with a physicalist world. Moreover, this

conclusion is not a mere technicality; it is compatible with our intuitions about physicalism. First, the

supervenience relation is not merely vacuously true, since the higher-level property is not itself physically

impossible just because it has physically impossible realizers. Second, all nomologically possible instances

of the higher-level property have physical realizers. Finally, the asymmetric relation between physical

realizers and the higher-level property remains, since (depending on one’s preferred metaphysics of causal

powers) the physical realizers still metaphysically necessitate the higher-level property.

Therefore, a higher-level property having exphysical realizers is compatible with non-reductive phys-

icalism, but incompatible with reductive physicalism.

Arguments for Exphysical Realizers

The major premise of this response is that the higher-level properties of the special sciences are specified

at a level of abstraction that allows for exphysical realizers. In this section, I will argue on behalf of the

non-reductive physicalist that special science properties have exphysical realizers.

Most philosophers find it prima facie plausible that special science properties have exphysical re-

alizers. It matters not to the fitness of a population whether the replicators of the system are made

of organic molecules or ectoplasm. And inflation would be inflation whether the economic agents are

mammals or rational ghosts. Indeed, the very intuition used to accept examples of multiple realizability

can also be used to accept exphysical realizers. One of the first in the literature was Fodor’s example

of Gresham’s Law governing monetary exchanges, and how it is immaterial whether such exchanges are

realized by strings of wampum, dollar bills, or signed cheques. We accept this example not because

we understand how wampum was exchanged, but because we know that various objects (and abstracta

like obligations recorded electronically) have been used as currency, and intuit that many others phys-

ically can. The same intuition extends beyond physical possibility. If there were massless particles not

traveling at the speed of light, they could be used as currency, and presumably Gresham’s Law would

apply. Similarly, when talking about the multiple realizability of mental states, it is not uncommon just

to make up realizers, and assume them to be physically possible, though they are only dubiously so.78

Nothing in our intuitions about these examples hangs on the types in question being physically possible.

To motivate another argument for the possibility of exphysical realizers, consider chemistry, where,

arguably, any sub-molecular property that satisfies certain binding regularities of the elements can be

considered valence. Judging at least from the past intuitions of scientists, it is metaphysically possible

77Since the exphysical realizer does not metaphysically necessitate any of the physical realizers, there are (possible)instances of the exphysical realizer that are not instances of a physical realizer. Thus, the higher-level property is notnecessarily coextensive with the disjunction of its physical realizers.

78Kim (1992) talks about Martian pain as one of the realizers of pain. Putnam (1967) mentions that a system combiningan immaterial soul and a body could be a probabilistic automaton, and could thereby implement certain computationalproperties.


that something like an orbital-model atom (small negative charges orbiting a massive, positively charged

core via the classical Coulomb force, but in discrete angular momentum states) can realize valence. Such

an atom is known to be physically impossible.79 Valence thus has an exphysical realizer.80

This example suggests an argument for the existence of exphysical realizers from the progress of

science. When a higher-level law is proposed by a special science, scientists typically require that a plau-

sible physical mechanism be available to realize the associated higher-level property. That mechanism

will involve an idealization of contemporary physics; call this the original plausible realizer. However,

as in the past, a future advance in physics may reveal the original plausible realizer to be physically

impossible. Yet the advance in physics is unlikely to affect the higher-level law, since we expect the

idealization to wash out the differences between the original plausible realizer and whatever realizer re-

places it in the new physics. The higher-level property will therefore have at least two possible realizers,

the actual realizer and the physically impossible original plausible one. Thus, the higher-level property

has an exphysical realizer.81

2.4.2 Empirical Science and Physical Realizers

The reductive physicalist must deny that the higher-level properties of interest to the special sciences

have exphysical realizers. A reductive physicalist can reasonably maintain that the only causal powers

had by properties are physical causal powers.82 But there is a better way to deny it, one that explains the

strength of intuitions that higher-level properties can transcend physicality. The reductive physicalist can

argue that intuitions that the higher-level properties of interest to the empirical sciences have exphysical

realizers rest on confusing two kinds of properties of interest to the special sciences. The first kind may

have exphysical realizers, but like mathematical properties, are not among the properties physicalists

are concerned with reducing. The second kind, on the other hand, are exactly what physicalists are

concerned with reducing, and won’t have exphysical realizers.

In this section, I present a reply to the argument from exphysical realizers. First, I reframe the

argument in terms of what I call restricted properties (with no exphysical realizers) and unrestricted

properties (with exphysical realizers). Second, I present some reasons from the practice of science that

suggest that the properties of concern to the special sciences are restricted. Third, I draw attention to two

attitudes we take towards properties of interest to the sciences, distinguished by how the properties are

specified: theory-fixed (specified by the role it plays in a theory) and world-fixed (specified by an implicit

ostension). Finally, I argue that, inasmuch as the special sciences are concerned with the experimental

investigation of empirical properties, the properties of interest are world-fixed and therefore restricted.

Restricted and Unrestricted Properties

To facilitate my reply, suppose a special science posits a higher-level property. It is uncontroversial that

it has physical realizers. Let us call the property type-identical to the disjunction of its physical realizers

the restricted higher-level property. By the arguments in the previous section, the higher-level property

also might have exphysical realizers. Let us call the property coextensive with the disjunction of both

79See footnote 21.80The orbital-model atom does not metaphysically necessitate the actual realizer of valence, obviously.81Again, the original plausible realizer does not metaphysically necessitate the actual one.82Causation, on this view, would be akin to a natural kind, so that what we discover about physical causation is a

discovery about causation simpliciter. This is the equivalent of denying that there are any “topic-neutral” causal powers.


the physical and the purported exphysical realizers the unrestricted higher-level property. This can be

illustrated as follows:

M =

restricted︷︸︸︷

P1 ∨ P2 ∨ . . . ∨ U1 ∨ U2 ∨ . . .︸︷︷︸

unrestricted

where M is the higher level property, the P ’s are the physical realizers, and the U ’s are the exphysical

realizers. The denial of the major premise can be restated as follows: the higher-level properties of

empirical interest to the special sciences are identical to the associated restricted properties.83

Arguments from the Practice of Science

With respect to scientific practice, there are two reasons to believe that experimental sciences are con-

cerned with restricted higher-level properties over unrestricted ones.

First, research in a higher-level science is not isolated from research into realizers, and in particular

into physics. There is no doubt that higher-level laws and generalizations involve not only properties

proper to the higher-level science, but also lower-level properties. For example, chemists speak not only

of valence, but also of electrons, the very electrons studied by physics. Clinical psychologists move

seamlessly from folk-psychological explanations to those involving neural mechanisms. Generalizations

like ‘aspirin causes pain to cease’ have inter-level significance. This multi-level concern bottoms out

at physics, and would be hard to explain if the higher-level properties of the special sciences were

unrestricted ones.

Second, a subtle dependency between special sciences and physics is generally accepted not only by

scientists but also by many reductive and non-reductive physicalists alike. Special sciences often posit

laws that are not exceptionless. The exceptions are expected to be explained by qualities possessed by

the physical realizers of the exceptional instances. Indeed, this very point was part of Fodor’s original

Multiple Realizability Argument.84 But this epistemic relationship between exceptional instances of a

higher-level property and the physical realizers of that property suggests that the special sciences are

positing restricted higher-level properties, not unrestricted ones.

Two Attitudes to Properties

One further way to deny that the properties of interest to special scientists have exphysical realizers

requires attending to a distinction in the way we specify properties: theory-fixed, and world-fixed. I

draw this distinction below.

Consider economics. We can construct an axiomatic economic theory, wherein the behaviour of

populations of ideally rational agents with desires that satisfy certain reasonableness criteria can be

given. If evidence and theory diverge, the theory does not incorrectly characterize the properties; rather,

the properties posited by the theory (like ideal rational agency) are not instantiated in the system.

83One may be tempted to say that the restricted property cannot be a natural kind, since the higher-level propertywas designed to include exphysical realizers, or that to exclude them is to gerrymander the referring predicate. But theseconcerns are unwarranted. How the higher-level property is specified is precisely the issue. Furthermore, there is nothingobviously deviant or gerrymandered about a predicate that refers to the restricted higher-level property, as compared tothat of the unrestricted one. Less technically, the restricted predicate is the same as the unrestricted one, except thatit includes the word ‘physically’. More technically, where the unrestricted predicate uses general (topic-neutral) causallanguage to characterize (say) a higher-level property, the restricted predicate makes reference to physical laws to do so.

84See section 2.2.1, and footnote 11 therein.


Contrast this with behavioural economics, which may develop laws about herd behaviour. If evidence

and theory diverge here, the properties are still said to be instantiated in the system, it is just that the

theory incorrectly characterizes them. This illustrates two different attitudes we can take to higher-level

properties: theory-fixed (the property is defined by the theoretical specification) and world-fixed (the

property anchored to the world, likely by an implicit ostension when the theory is introduced).

As suggested above, the difference between these two attitudes is most stark when we consider

whether the property being specified exists. A theory-fixed property, as long as it is defined against a

consistent theory, may or may not be instantiated in a given world, but must be instantiated in some

world. In contrast, a world-fixed property, as long the predicate used to introduce it refers at all, must be

instantiated in the world to which it is anchored. If the property is not instantiated, then the predicate

used to introduce it fails to refer at all, so there are no other worlds in which it is instantiated. Relatedly,

we cannot be wrong about the natures of theory-fixed properties, though we can be wrong about whether

they are instantiated. This is one of the benefits of defining a property against a (well-developed) theory.

On the other hand, we are frequently wrong about the natures of world-fixed properties, though rarely

are we wrong that they exist. That is one of the benefits of pointing to the properties.

I’m going to remain neutral to whether this distinction corresponds to a metaphysical distinction

or a (merely) epistemic distinction. For example, some might argue that at least some properties have

quiddities, and so cannot be specified by any theory. In any case, as I argue in the next section, attention

to this distinction can reveal interesting metaphysical conclusions about the properties of interest to the

empirical sciences.

However, I think this distinction has value independent of the use I put it to, especially since philoso-

phers and scientists alike are frequently not sensitive to this distinction when considering the predicates

used in scientific theories. Often, the same term is used to refer to possibly distinct properties depending

on the attitude we take to its specification. And this, of course, can result in a verbal dispute.85 For

example, an A-theorist about time might be reluctant to admit evidence from Special Relativity about

whether there is universal, unambiguous simultaneity. When confronted with such evidence, it’s possi-

ble for an A-theorist to say that we live in a world without time, though it has something very similar

to time. This is because the A-theorist is taking a theory-fixed attitude towards temporal properties.

A B-theorist or a physicist, on the other hand, would argue that discoveries were being made about

the nature of time. They are taking a world-fixed attitude towards temporal properties. For another

example, a dualist might claim that phenomenal properties are essentially intrinsic, and not structural

or dynamic. Any conceptual progress made on the nature of phenomenal properties by science cannot

result in a structural or dynamic explanation for it; at best, such progress would show that there are

no phenomenal properties. A materialist or a neuroscientist might take the opposite view. Again, the

dualist and the materialist would be taking theory-fixed and world-fixed attitudes, respectively, towards

phenomenal properties.

Exphysical Realizers and Empirical Properties

Here, I argue that the extent to which the laws and properties posited by a science are abstract enough

to outstrip physical realizers, is just the extent to which those laws and properties are not world-fixed but

theory-fixed. In other words, world-fixed properties are restricted, while theory-fixed properties (might

be) unrestricted.

85For a discussion on verbal disputes, see Chalmers (2011).


Suppose two competing theories posit two higher-level properties such that: (i) the properties share

all physically possible realizers and not purportedly exphysical ones; and (ii) the theories are compatible

in all physically possible worlds but incompatible in some physically impossible worlds. If the properties

of the theories are interpreted to be theory-fixed, the properties are defined by their theory specifications,

and the two unrestricted properties can exist side by side. This is illustrated below for theories A and

B, with physical realizers P and exphysical realizers U :

A

BU

1

U2

U3

U4

P3

P2

P1

The same cannot be said if the properties of the theories are interpreted as world-fixed. By the context

in which the two theories were introduced, they refer to the same property in the world. Suppose the

property of concern is unrestricted. We must then consider the properties characterized by the theories

to be unrestricted as well. Since in this case the two properties have exphysical realizers, they would not

be necessarily coextensive, so at most one theory would be correct. Yet there is no experimental way to

distinguish them, since the theories are compatible (and indeed treat corresponding property instances

identically) in all nomologically possible worlds. On the other hand, suppose the property of concern is

restricted. In that case, since the two restricted properties are necessarily coextensive, we can say the

properties are type-identical, and the theories are not incompatible after all. This is illustrated below:

A

BU

1

U2

U3

U4

P3

P2

P1

This suggests that the experimentally investigated higher-level properties are restricted.

We therefore have good reasons to believe that, inasmuch as the special sciences are concerned with

the experimental investigation of empirical properties, those properties will not have exphysical realizers,

despite appearances.

2.5 Conclusion

The higher-level properties of the special sciences are specified at a level of abstraction high enough

to be multiply realized physically, but this is not enough to guarantee a non-reductivist conclusion.

This is because the Disjunctive Strategy is available: the higher-level property is type-identical to the

disjunction of its realizers, a disjunction of properties is ontologically innocent, and all the disjuncts are

ultimately physical. Therefore, the higher-level property is type-identical to an ontologically innocent

combination of physical properties, i.e., it is itself a physical property. There are only three ways for a

non-reductive physicalist to respond to this strategy.


The first way is to deny that the higher-level property is type-identical to the disjunction of its real-

izers. Fodor does this by claiming that the higher-level property is a natural kind, while the disjunction

of its realizers is not. Wilson claims that a higher-level property instance is token-distinct from its

realizer, while the instance of the disjunction of its realizers is token-identical to the realizer. Both these

responses were shown to be inadequate. Exhaustively overlapping disjunctions of properties can indeed

be kinds, and we are free either to deny that the higher-level property is token-distinct from its realizer,

or to deny that the exhaustively overlapping disjunctive property is token-identical to the realizer.

The second way is to deny that an exhaustively overlapping disjunction of properties is an ontologi-

cally innocent combination of its disjuncts. Such a denial is possible if we assume that a disjunction of

realizers is token-distinct from the realizers. If it is an exhaustively overlapping disjunction that is token-

distinct from its disjuncts, then it satisfies Wilson’s subset condition on causal powers. In that case, it

is both ontologically distinct and causally autonomous, the latter in virtue of the higher-level property

being counterfactually stable and figuring in a distinct system of laws. But, as argued above, ontological

distinctness is not enough to establish ontological guilt. Furthermore, the claims to causal autonomy

can be resisted by reinterpreting the counterfactual stability of the exhaustively overlapping disjunction

as a special case of any disjunction satisfying more counterfactual conditionals than its disjuncts, and

by problematizing the idea that the disjunctions figure in distinct systems of laws.

The third way is to deny that the disjunction to which the higher-level property is type-identical

is exhausted by physical realizers. If the higher-level property has exphysical realizers in physically

impossible worlds (as higher-level properties appear to), then even if it is type-identical to the disjunction

of its realizers, that disjunction is not itself a physical property. However, we have good reasons to believe

that, inasmuch as the special sciences are concerned with the experimental investigation of empirical

properties, those properties will not have exphysical realizers. Appearances otherwise rest on confusing

such properties with theory-fixed properties of interest to the various sciences, which may have exphysical

realizers, but which are in no more need of reduction than any mathematical property.

Chapter 3

Anti-physicalism and the Physical

Any modern discussion of anti-physicalist positions in the Philosophy of Mind cannot help but be

anchored by two related arguments against physicalism. These are the Knowledge Argument and the

Conceivability Argument. Both these arguments have the same disjunctive conclusion: either physicalism

is false, or a kind of psycho-physical monism is true whereby phenomenal properties are deeply connected

to physically acceptable properties in a way that goes beyond the role played by the latter in physical

theories.

A complete defense of reductive physicalism would require addressing these arguments. While I

do describe below how they have been resisted in the literature, I do not evaluate the success of the

resistance. Moreover, an original defense against these arguments is, unfortunately, beyond the scope

of this document.1 Instead, I have a more modest goal. Below, I will argue that the Knowledge

Argument and the Conceivability Argument are just as problematic for the main anti-physicalist rivals

to physicalism as they are for physicalism.

A little known consequence of the Knowledge Argument and the Conceivability Argument is that

they also rule out historically important anti-physicalist positions. In particular, they rule out any

strong dualist position according to which phenomenal properties interact with physically acceptable

properties in a lawful manner, unless the position entails an analogue to psycho-physical monism. Since

the latter is no more plausible than psycho-physical monism (otherwise) compatible with physicalism,

lawfully interacting strong dualism is left with no support relative to physicalism.

In this chapter, I argue that the Knowledge Argument and the Conceivability Argument have more

general applicability than is normally appreciated. I will do so by generalizing the conception of physi-

cality in two orthogonal ways. First, I will attend to the distinction between a-physicality (physicality

with respect to the final physics of the actual world) and p-physicality (physicality in the most general

sense, corresponding to those properties characteristic of any possible physical theory). Second, I will

draw parallels between physicality proper (being a property posited by a final physics) and what I call

paraphysicality (being a property satisfying a theoretical role in a final physics). I argue that the two

anti-physicalist arguments are neutral with respect to both which conception of physicality, and whether

physicality or paraphysicality, is used to interpret them. Furthermore, lawfully interacting strong du-

alism, while dualist with respect to a-physical properties, is monist with respect to the paraphysical

1The most promising such defense appears to me to be one that argues that knowledge of phenomenal properties isconstituted, at least in part, by being in particular physical states. This is transparently connected to the KnowledgeArgument. But it also has consequences for the Conceivability Argument, since it changes our judgement about how idealconceivability should be interpreted.

52

Chapter 3. Anti-physicalism and the Physical 53

equivalent of p-physical properties. This makes it as vulnerable to the anti-physicalist arguments as

physicalism is.

In section 3.1, I revisit the conception of the physical, and adapt it to a context that includes the

possibility that physicalism is false. In sections 3.2 and 3.3, I provide an overview of the Knowledge

Argument and the Conceivability Argument, respectively, and show them to be neutral with respect

to the various conceptions of the physical. Finally, in section 3.4, I apply the above results to lawfully

interacting strong dualism.

3.1 Physicality Revisited

Much of this chapter involves interpreting the anti-physicalist arguments in light of different conceptions

of the physical. I shall therefore begin by revisiting the characterization of physical properties and related

concepts that I introduced in the first chapter. These characterizations will need to be elaborated, and

adapted to a context that includes discussions of worlds where physicalism is false. In parallel, they will

be generalized to include the possibility of physically unacceptable properties that satisfy the theoretical

roles of physically acceptable properties.

3.1.1 Physical and Paraphysical Properties

Recall the characterization of physical properties given in section 1.1.3. A property is w-physical if it

is a property posited by the final physics of world w.2 An a-physical property is w-physical for the

actual world. A-physical properties are what matter for reduction and physicalism in the actual world;

w-physical properties are what matter for reduction and physicalism in world w. A p-physical property

is one that is w-physical for at least one world w. P-physical properties are physical in the broadest

sense.

Call any property that satisfies the theoretical role (the nomic role or the causal role)3 of a property

posited by a final physics theory (including ontologically innocent combinations)4 a paraphysical prop-

erty. A physical property is trivially a paraphysical property. Perhaps there is nothing more to physical

properties than the powers by which they satisfy the theoretical roles,5 or perhaps they have quiddities

or something else permitted of properties posited by final physics. But it might be that properties not

typically taken to be physical properties are paraphysical, perhaps because they satisfy those roles in

ways incompatible with a property’s being physical. For example, perhaps it is through the action of

immaterial angels that the roles are satisfied. Nevertheless, in virtue of satisfying theoretical roles of

properties posited by final physics, such properties are paraphysical.

The three varieties of physicality described above have paraphysical analogues. Call any property that

satisfies the theoretical role of the final physics of world w a w-paraphysical property. An a-paraphysical

property satisfies the theoretical role of a property posited by the final physics of the actual world. A

p-paraphysical property satisfies the theoretical role of a property posited by the final physics of some

possible world.

2Ontologically innocent combinations of w-physical properties are also w-physical.3A property’s nomic role differs from its causal role since not all natural laws are causal laws.4Below, satisfying a theoretical role of a property posited by a final physics theory should be understood to include

being an ontologically innocent combination of properties that satisfy such roles.5Or perhaps the physical property is the role itself, in which case we can allow being identical to a theoretical role to

be the limiting case of satisfying that role.


3.1.2 Physically and Paraphysically Acceptable Properties

Recall further that, for any world w, a w-physically acceptable property is one that is nothing over and

above w-physical properties. An a-physically acceptable property is nothing over and above a-physical

properties. A p-physically acceptable property is one that is w-physically acceptable in at least one

world w. In section 1.1.4, I introduced a provisional supervenience-based account of being nothing over

and above, viz., one property is nothing over and above another if and only if the former strongly,

physically supervenes on the latter. As stated in that section, problems arise for this account when we

consider worlds where physicalism is false. For now, I will withdraw my provisional account in favour of

an intuitive notion of being nothing over and above.

If my argument against non-reductive physicalism in the previous chapter is successful, then for any

world w,6 all w-physically acceptable properties are w-physical properties, and thus are w-paraphysical

properties. Otherwise, let paraphysically acceptable properties be those properties that satisfy the theo-

retical roles played by physically acceptable properties. One way to understand this would be to impose

deductive closure on physics theories, so that any statement that logically follows from a physics the-

ory is a statement of the theory.7 Then paraphysically acceptable properties will be those properties

that satisfy a predicate that can be constructed from the predicates designating the posited physical

properties.8 In this way, paraphysically acceptable properties satisfy whatever theoretical role those

constructed predicates play in the expanded theory.9

As above, the varieties of paraphysically acceptable properties match those of physically acceptable

properties. For world w, the w-paraphysically acceptable properties are those that satisfy the theoret-

ical roles of w-physically acceptable properties. The a-paraphysically acceptable properties are those

that satisfy the theoretical roles of a-physically acceptable properties. The p-paraphysically acceptable

properties are those that satisfy the theoretical roles of p-physically acceptable properties, i.e., the roles

of w-physically acceptable properties for some world w.

3.1.3 Stray Properties

Prima facie, it is possible for a property in world w to be p-physically acceptable without being w-

physically acceptable. Call these the stray p-physically acceptable properties of w. Such a property

fails to satisfy the theoretical role of any property posited by the final physics of w (i.e., it is not w-

paraphysically acceptable), even though it does satisfy the theoretical role of a property posited by the

final physics of some world. To generalize, call any property in w that fails to be w-paraphysically

acceptable but is a p-paraphysically acceptable a stray p-paraphysically acceptable property of w.

Recall the example of an orbital-model atom in the actual world. An orbital model atom is a stable

atom in which the electrons orbit the nucleus according to the Coulomb force but in discrete angular

momentum states. Such an atom is forbidden by the current (and hence likely the final) physics of the

actual world. If such an atom were instantiated in the actual world, then the property of being an orbital

model atom would not be a-physically acceptable. On the other hand, there is some world w whose final

6At least those of which physicalism is true.7We can include other, possibly non-deductive extensions of the theories as well, such as limiting cases.8There are problems associated with defining paraphysically acceptable properties this way, since some such predicates,

like those that negate physical predicates (e.g., ‘not-an-electron’), designate properties that don’t play theoretical rolescharacteristic of physical theories. See Melnyk (2003a).

9One way to think about this is for paraphysically acceptable properties to include those that satisfy some functionalcharacterization in terms of the final physics.


physics does describe the behaviour of an orbital model atom. The actual orbital model atom would,

plausibly, be nothing over and above the physical properties of w, which makes it w-physically accept-

able. In that case, being an actual orbital model atom is p-physically acceptable and not a-physically

acceptable, making it a stray p-physically acceptable property. This generalizes. Since being an actual

orbital model atom is forbidden by the actual final physics, it is not an a-paraphysically acceptable prop-

erty. Since the property does satisfy the final physics of w, it is w-paraphysically acceptable. Therefore,

being an actual orbital model atom is p-paraphysically acceptable and not a-paraphysically acceptable,

making it a stray p-paraphysically acceptable property.

That there can be stray p-paraphysically acceptable properties (and so a fortiori stray p-physically ac-

ceptable properties) follows from a (Humean) combinatorial approach to modality, according to which ar-

bitrary (but consistent) combinations of fundamental properties can be instantiated in the same world.10

Nevertheless, there may be reasons to deny that any p-paraphysically acceptable (including p-physically

acceptable) properties in a world are stray. Methodologically, perhaps one of the conditions for being a

final physics theory is that the theory must describe (the role of) all p-physically acceptable properties in

the world. In that case, it is impossible for a p-paraphysically acceptable property to play no theoretical

role in the final physics of that world. Metaphysically, perhaps all p-physically acceptable properties in

world w cannot help but be nothing over and above the w-physical properties of that world. Perhaps the

causal powers of all p-physically acceptable properties must be composed of the primitive causal powers

of the w-physical properties. However, all such considerations forbid certain kinds of anti-physicalist

positions, which works in favour of the physicalist.

Alternatively, there may be reasons to suppose that it is impossible for any property of one world to

satisfy a theoretical role in the final physics of some other (nomologically impossible) world. Perhaps the

only way for a single property to satisfy a single theoretical role is by being part of a system of properties

that satisfies all the theoretical roles together, holistically. For example, perhaps being an actual orbital

model atom cannot satisfy the theoretical role of being an orbital model atom in a world w where such

are posited by the final physics of w, since the elements of the actual orbital model atom (the electrons

and nuclei) do not satisfy the corresponding theoretical roles of the elements of w’s orbital model atom

in the final physics of w, owing to their (presumably) differing behaviour outside that atom. In that case,

any stray properties that are apparently p-paraphysically acceptable are not p-paraphysically acceptable

at all, since they do not satisfy theoretical roles in the final physics of the actual world, and cannot satisfy

theoretical roles in the final physics of any other world.

Again, this is not a serious problem for me. I only require that the role played by the stray property

appropriately resembles that of a w-physically acceptable property for some world w. Being an orbital

model atom in the actual world may not satisfy the theoretical role of being an orbital model atom in

a world where such are posited by that world’s final physics. But the roles certainly do resemble one

another.11 The roles resemble one another enough that, if the former can be satisfied by a property

that conforms to the general constraints on being physically acceptable,12 so will the latter. I shall

therefore redefine p-physically acceptable properties to include not only all the w-physically acceptable

10See Lewis (1986) and Armstrong (1989). However, see Wilson (2010b). I am indebted to Wilson for pointing this out.11How to fill out this resemblance relation is, of course, not a trivial task. Perhaps the two properties are counterparts

to one another, where one plays an appropriate role with respect to the other in counterfactual or counterlegal reasoning.Let Pa be the property of being an orbital model atom in the actual world, and let Pb be the property of being an orbitalmodel atom in world b, wherein Pb is posited by the final physics. Then we can say that, had the actual laws been asthey are in b, actual objects exemplifying Pa would behave as objects exemplifying Pb.

12Recall from section 1.1.3 that the only constraints on a p-physical property is that it is broadly scientific, empiricallyaccessible, and not fundamentally mental, with p-physically acceptable properties being nothing over and above them.


properties, but also those that resemble w-physically acceptable properties. Likewise, I shall redefine

p-paraphysically acceptable properties to include not only all the w-paraphysically acceptable properties,

but also properties that satisfy roles that resemble those of w-paraphysically acceptable properties. Then

the stray properties fail to satisfy theoretical roles in the final physics of the world in which they are

instantiated, but either satisfy theoretical roles in the final physics of some nomologically impossible

world, or satisfy roles that resemble them.

Throughout this chapter, I am therefore going to assume that it is sensible to talk about stray

properties. For convenience, I will assume that such properties can satisfy theoretical roles in the final

physics of worlds nomologically impossible relative to the worlds in which they are instantiated. It

should be understood that they may satisfy resembling roles instead.

3.1.4 Physicalism and Paraphysicalism

Recall that physicalism is the position according to which all broadly scientific properties are physically

acceptable. Since we are interested in the actual world, it is a-physicality that is important for physical-

ism. This generalizes easily to other possible worlds: for any world w, physicalism is true of w if and only

if all broadly scientific properties in w are w-physically acceptable properties. Call this w-physicalism

for clarity.

There is an even more general position related to physicalism, which can be obtained by requiring

of the broadly scientific properties of some world only that they be p-physically acceptable. Call this

position p-physicalism: p-physicalism is true of world w if and only if all broadly scientific properties in

w are p-physically acceptable properties. Since all w-physically acceptable properties are p-physically

acceptable, w-physicalism is a variety of p-physicalism. But since there are worlds w such that not

all p-physically acceptable properties are w-physically acceptable (i.e., worlds with stray p-physically

acceptable properties), p-physicalism does not entail w-physicalism. Thus, strictly speaking, those worlds

w with stray p-physically acceptable properties are anti-physicalist worlds, though they may be p-

physicalist.

An orthogonal series of generalizations can be made with reference to paraphysicality. Let paraphys-

icalism be the position according to which all properties are paraphysically acceptable. Physicalism

entails, but is not entailed by, paraphysicalism. Let a-paraphysicalism be the position according to

which all actual properties are a-paraphysically acceptable, w-paraphysicalism the position according

to which all properties in world w are w-paraphysically acceptable, and p-paraphysicalism the position

according to which all properties in world w are p-paraphysically acceptable. As above, for world w, w-

paraphysicalism entails p-paraphysicalism, but not vice versa, since w might have stray p-paraphysically

acceptable properties.

3.1.5 Physical and Paraphysical Facts

It is plausible that any deductively closed physics theory describes only the spatiotemporal distribution

of physically acceptable properties. In light of this, let a physical fact be a spatiotemporal distribution

of physically acceptable properties, along with the laws described by the appropriate final physics.13

For any world w, the w-physical facts are the spatiotemporal distribution of w-physically acceptable

13I will suppose, for the sake of charity, that every nomologically necessary statement in the deductively closed physicstheory is a law. This way, the laws will include any special science laws exclusively about physically acceptable properties.


properties in w, along with the laws described by the final physics of w. The a-physical facts are the

w-physical facts for the actual world. The p-physical facts of a world w include all the w-physical facts.

But they also include the spatiotemporal distribution of any stray p-physically acceptable properties in

w, along with the laws described by the final physics of any world whose predicates are satisfied by those

stray p-physically acceptable properties, and whatever laws govern their interaction with w-physically

acceptable properties.14

Generalizing, let a paraphysical fact be the spatiotemporal distribution of paraphysically acceptable

properties, along with the laws described by the appropriate final physics. Again, for any world w, the

w-paraphysical facts are the spatiotemporal distribution of w-paraphysically acceptable properties in w,

along with the laws described by the final physics of w. The a-paraphysical facts are the w-paraphysical

facts for the actual world. The p-paraphysical facts of a world w include the w-paraphysical facts,

the spatiotemporal distribution of any stray p-paraphysically acceptable properties in w, and the laws

governing them and their interaction with w-paraphysically acceptable properties.

3.1.6 Deep Physical-Phenomenal Unity

There is a family of positions according to which there is a deep connection between physically acceptable

properties and phenomenal properties that goes beyond the theoretical role of the former. For example,

according to dual-aspect theories, physically acceptable properties and phenomenal properties are two

aspects of the same property. Spinoza (1677) and Nagel (1986) are among those who have advocated

dual-aspect theory. Another example is neutral monism, according to which the physical and the phe-

nomenal are two manifestations of a distinct, non-phenomenal, non-physical (and hence neutral) kind

of property. Russell (1921) argues for neutral monism. For a final example, according to panpsychism

(or panprotopsychism), phenomenal (or protophenomenal) properties are the intrinsic natures, or the

ultimate categorical bases, of the otherwise dispositional physically acceptable properties. Arguments

for panpsychism (or panprotopsychism) can be seen in, e.g., Stoljar (2001) and Strawson (2006).

The differences among these positions is subtle, and to some extent the positions blur into one

another. Whatever differences they have are irrelevant to my purposes. Call all such positions Deep

Physical-Phenomenal Unity :

Deep Physical-Phenomenal Unity (DU): There is at least one physically acceptable property that is

metaphysically necessarily connected to a phenomenal property in a way that goes beyond the

physically acceptable property’s theoretical role.

DU positions, in the context of Knowledge Argument and the Conceivability Argument, are meant to

contrast two other positions that metaphysically connect physical and phenomenal properties. Accord-

ing to the first contrasting position (roughly, analytic functionalism),15 perhaps (contra expectations)

conceptual analysis will yield a functional characterization of phenomenal properties in terms of physical

properties. In that case, in virtue of satisfying the theoretical roles that they do, physical properties

will realize phenomenal properties by metaphysical necessity. According to the second contrasting posi-

14Let P be the stray p-physically acceptable properties of world w1. If arbitrary combinations of fundamental propertiesare possible, then there is another world, w2, in which P is w2-physically acceptable (i.e., is not stray), and in which thew1-physical properties are among the w2-physical properties. In that case, the final physics of w2 will describe the lawsgoverning w1-physically acceptable properties, the laws governing P , and the laws governing their interaction. The lattertwo would be among the p-physical facts of w1.

15See, for example, Lewis (1988).


tion (roughly, a posteriori physicalism),16 perhaps there is no a priori connection between physical and

phenomenal properties, just as there is no a priori connection between water and H2O. Nevertheless,

just as it is an a posteriori metaphysical necessity that water is H2O, so it is an a posteriori metaphys-

ical necessity that phenomenal properties are physical properties. Neither of these positions are DU

positions. In contrast to the first position, according to DU the connection between physical properties

and phenomenal properties goes beyond the theoretical role of the physical property. In contrast to

the second position, it is not a merely a posteriori fact that physical and phenomenal properties are

metaphysically necessarily connected; there is, instead, a deeper connection between the natures of the

physical and the phenomenal.

I do not intend to argue (one way or another) whether DU is compatible with physicalism, nor indeed

whether it correctly describes the actual world. My principal concern here is to disambiguate DU with

respect to the conception of the physical.

In the Philosophy of Mind, philosophers are principally concerned with reconciling physically accept-

able properties as revealed to them by physics with phenomenal properties (as experienced by them).

The physically acceptable properties in question are therefore naturally interpreted as a-physically ac-

ceptable properties:

Deep a-Physical-Phenomenal Unity (DU-a): There is at least one a-physically acceptable property that

is metaphysically necessarily connected to a phenomenal property in a way that goes beyond the

a-physically acceptable property’s theoretical role.

Most of the literature about DU is naturally interpreted as being about DU-a. For example, Spinoza

(1677) and Russell (1921, 1922) are transparently talking about the actual world and (at least in the

case of the latter) contemporary physics. Nagel (1986) develops his dual-aspect view in light of his

interest in our world and our place in the world, which suggests a concern for a-physically acceptable

properties.17 Stoljar (2001) describes physical properties as those that are required to account for

paradigmatic physical objects, which are presumably actual objects. Strawson (2006) plainly states that

physicalism is about the actual world.

Nevertheless, this interpretation of physicality in DU can be generalized quite easily to w-physicality

in other possible worlds:

Deep w-Physical-Phenomenal Unity (DU-w): In world w, there is at least one w-physically acceptable

property that is metaphysically necessarily connected to a phenomenal property in a way that goes

beyond the w-physically acceptable property’s theoretical role.

If a more general dual-aspect theory is true, then w-physically acceptable properties in world w and

phenomenal properties in world w are two aspects of the same property. If a more general neutral monism

is true, then the w-physical and the phenomenal are two manifestations of a distinct, non-phenomenal,

non-w-physical properties in world w. If a more general panprotopsychism is true, then protophenomenal

properties are the categorical bases of the otherwise dispositional w-physically acceptable properties in

world w.

Arguments for or against DU in the literature can easily be adapted to DU-w. Although they

may be naturally interpreted as being about DU-a, little in the arguments is specific to the actual

16See, for example, Block and Stalnaker (1999).17He starts his discussion of the mind with a Physical Objectivity section, which begins: “The natural place to begin is

with our own position in the world” (Nagel, 1986, p.13).


world. Indeed, all such arguments are made from a position of ignorance of the actual final physics,

so if the consequences of a specific physics theory was being considered, it would have been the final

physics of some non-actual world. Indeed, Spinoza and Russell wrote before modern physics was firmly

established, and the arguments of Nagel, Stoljar, and Strawson do not depend on the details of modern

physics. While Stoljar and Strawson rely to some extent on ostensive reference-fixing to establish the

denotation of “physical”, their arguments are still easily generalizable. This can be seen in, for example,

Seager (2006), where the intrinsic nature argument is seen to turn on general features of the physically

acceptable, such as being relational or structural.

The more interesting generalization of DU is to p-physicality in some possible world.

Deep p-Physical-Phenomenal Unity (DU-p): In world w, there is at least one p-physically acceptable

property that is metaphysically necessarily connected to a phenomenal property in a way that

goes beyond the p-physically acceptable property’s theoretical role.

For any world w, if DU-w is true of world w, then DU-p is also true of w. But the contrary is not

the case. If phenomenal properties are appropriately connected to some stray p-physically acceptable

properties, and not to any w-physically acceptable properties, then DU-p, and not DU-w, is true of w.

Arguments in favour of DU can be adapted easily in service of DU-p. First, for any world w, DU-w

implies DU-p, so any argument for the truth of DU-w in w is a fortiori an argument for the truth of

DU-p. But beyond this, the arguments in favour of the appropriate connection between phenomenal

properties and physically acceptable properties are largely insensitive to how that property is connected

to the final physics of the world in question. Such arguments typically focus on the dispositional,

relational, or structural nature of physically acceptable properties, or our epistemic relationship to

them. But all p-physically acceptable properties are dispositional, relational, or structural if and only

if (for all worlds w) all w-physically acceptable properties are. Moreover, our epistemic relationship to

p-physically acceptable properties is the same as to w-physically acceptable properties. Therefore, we

can expect that arguments for or against DU-w can be interpreted as arguments for or against DU-p.

Finally, there is no apparent reason to exclude from DU those positions according to which there is a

deep connection between paraphysical properties and phenomenal properties. Indeed, one interpretation

of panpsychism is that the properties that satisfy the theoretical roles of properties posited by final

physics are phenomenal properties, which would make phenomenal properties paraphysically acceptable.

Therefore, let us expand the definition of DU to include positions according to which there is a deep

connection between paraphysically acceptable properties and phenomenal properties that goes beyond

the theoretical role of the former.

Expanded Deep Physical-Phenomenal Unity (DUx): There is at least one paraphysically acceptable prop-

erty that is metaphysically necessarily connected to a phenomenal property in a way that goes

beyond the paraphysically acceptable property’s theoretical role.

Moreover, we can define expanded analogues to the DU-a, DU-w, and DU-p positions above.

Expanded Deep a-Physical-Phenomenal Unity (DUx-a): There is at least one a-paraphysically accept-

able property that is metaphysically necessarily connected to a phenomenal property in a way that

goes beyond the a-paraphysically acceptable property’s theoretical role.

Expanded Deep w-Physical-Phenomenal Unity (DUx-w): In world w, there is at least one w-paraphysi-

cally acceptable property that is metaphysically necessarily connected to a phenomenal property

in a way that goes beyond the w-paraphysically acceptable property’s theoretical role.


Expanded Deep p-Physical-Phenomenal Unity (DUx-p): In world w, there is at least one p-paraphysically

acceptable property that is metaphysically necessarily connected to a phenomenal property in a

way that goes beyond the p-paraphysically acceptable property’s theoretical role.

The considerations above regarding DU-a, DU-w, and DU-p apply (mutatis mutandis) to DUx-a, DUx-w,

and DUx-p.

3.2 The Knowledge Argument

In this section, I will provide an overview the Knowledge Argument, and interpret it in light of the various

conceptions of physicality described above. When I provide the overview in section 3.2.1, I will leave

the conception of physicality uninterpreted. I will discuss the a-physicality, w-physicality, p-physicality,

and paraphysicality interpretations of the argument in section 3.2.2.

3.2.1 Overview of the Knowledge Argument

The Knowledge Argument is based on the following thought experiment (Jackson, 1982).18 Mary is an

exceptionally gifted scientist who has never seen any colour: she has been confined to a black and white

room for her entire life, and (for good measure) has had surgery to prevent colour vision. In the room,

she is taught all physical facts. She is then released from her confinement, and surgically re-equipped

with colour vision, after which she looks upon a red tomato.

We have a strong intuition that Mary learned something new after being released, viz., what it is

like to see a red tomato. The Knowledge Argument moves from this intuition to the claim that Mary

learned a new fact after being released. Since she knew all the physical facts before being released, there

must be facts about broadly scientific properties (specifically phenomenal facts) that are not physical

facts.

KA1. Mary knew all the physical facts before her release.

KA2. Mary learned a new (phenomenal) fact after her release.

∴ KA3. There are (phenomenal) facts that are not physical facts.

The first premise is that Mary knew all the physical facts before being released. Recall that physical

facts are the spatiotemporal distribution of physically acceptable properties (along with the nomologically

necessary statements of the physical theories whose theoretical roles they satisfy). Thus, the first premise

can be taken to be true by hypothesis.

The second premise is that Mary learned a new (broadly scientific, phenomenal) fact after being

released. It is supported by the intuition that phenomenal knowledge (such as the knowledge of what

it is like to see a red tomato) cannot be communicated the way that knowledge of the spatiotemporal

distribution of physically acceptable properties (along with the nomologically necessary statements of

the relevant physical theories) can. Even with such comprehensive knowledge of the physical facts, Mary

cannot know what it will be like to see a red tomato.

Nevertheless, there are a variety of physicalist responses that deny this premise. Some deny that

Mary gains any knowledge at all (Dennett, 1991). Others deny that the knowledge that Mary gains

18The thought experiment below is a modified version of the original given in Jackson (1982).


is propositional knowledge (Nemirow, 1990; Lewis, 1988). Still others deny that new propositional

knowledge implies new facts known (Tye, 2000). Since my concern is the role of the physical in the

Knowledge Argument, I will not evaluate the success of these responses.

The conclusion that there are facts (about broadly scientific properties, specifically phenomenal

properties) that are not physical facts is meant to contradict physicalism in a straightforward way. But,

as stated above, knowing the physical facts is knowing only the spatiotemporal distribution of properties

that satisfy theoretical roles in physical theories, along with the nomologically necessary statements of

those theories. As we have seen, many argue that phenomenal properties are metaphysically necessarily

connected to physically acceptable properties beyond the theoretical roles of the latter. To accommodate

these positions, and to remain neutral as to whether such positions are compatible with physicalism,

two more statements can be added to the Knowledge Argument:

KA4. If there are (phenomenal) facts that are not physical facts, then either DU is true or physicalism

is false.

∴ KA5. Either DU is true or physicalism is false.

The formulation of the Knowledge Argument that includes these final two statements has three ad-

vantages. First, it makes explicit the move from the existence of facts that are not physical facts to

physicalism being false. Second, it accommodates DU positions. Finally, it corresponds better to the

Conceivability Argument, which makes comparing the arguments easier, and allows me to generalize

more easily.

3.2.2 The Knowledge Argument and Physicality

The thought experiment upon which the Knowledge Argument is based has Mary knowing all the physical

facts before being released. This leads to the conclusion that there are facts that are not physical facts,

which implies that either DU is true or physicalism is false. Some conception of the physical is needed

to interpret three parts of the argument: the physical facts known by Mary, the DU position that is

permitted by the argument, and the physicalism that is claimed to be false. The argument does not

equivocate, so there is a single conception of the physical being used in these parts. But which conception

of the physical is being used can be uniformly interpreted as a-physical, w-physical, or p-physical, as

well as any variation of paraphysical.

The Knowledge Argument and a-Physicality

The natural interpretation of the argument is that it is about a-physicality. Philosophical interest in

the a-physical possibility of Mary’s confinement reveals this interpretation.19 Under this interpretation,

Mary learns all the a-physical facts in a black and white environment, and, upon her release, learns a

new (broadly scientific) fact. It follows that there are facts that are not a-physical facts, which means

that either DU-a is true or a-physicalism is false. Presumably, it is under this interpretation of the

argument that we have the strongest intuitions about whether Mary learns something new. It is also

under this interpretation that we have the lowest confidence that Mary’s confinement (or that she learns

all the physical facts) is physically possible.

19See, for example, Graham and Horgan (2000).


The Knowledge Argument and w-Physicality

Nevertheless, the argument can be generalized to be about w-physicality for any world w similar enough

to the actual world that it is sensible to do the thought experiment (Mary’s confinement and release

are w-physically possible, there are phenomenal properties, etc.). Under this interpretation, for world

w, Mary learns all the w-physical facts in a black and white environment, and, upon her release, learns

a new (broadly scientific) fact about w. It follows that there are facts about w that are not w-physical

facts, which means that either DU-w is true or w-physicalism is false of w.

An advantage of the w-physical interpretation is that the w-physical possibility of Mary’s confinement

and abilities are not an issue, since the domain of worlds under consideration is stipulated to include only

such worlds. On the other hand, our intuitions about whether Mary learns something new are perhaps

somewhat weaker than in the actual case, if only because we may have some difficulty working out the

w-physical details of Mary’s confinement. Still, our ordinary intuitions about what Mary learns appear

not to depend at all on the final physics of the actual world (since we are in a position of ignorance

about it), nor on the details of current physics (which plausibly approximates the final physics). For

this reason, we should expect that our intuitions about Mary in the a-physical interpretation can be

extended to the w-physical interpretation, at least for worlds w that are similar to ours, and at least if

those worlds are treated as actual.

And this appears to be the case. Take any world w where Mary’s confinement circumstances are w-

physically possible (and treat that world as actual, if necessary). Suppose Mary learns all the w-physical

facts in a black and white room, and is surgically prevented from seeing colour. Upon her release, having

her colour vision restored, she looks upon a red object. We still have the intuition that Mary learns

something new upon her release, viz., what it is like to see red. The rest of the argument follows as

normal, and the premises can be defended or denied in a similar fashion.

The Knowledge Argument and p-Physicality

The generalization to p-physicality is not much different than the generalization to w-physicality. Con-

sider worlds w in the domain described above. For world w, Mary learns all the p-physical facts in a

black and white environment, and, upon her release, learns a new (broadly scientific) fact about w. It

follows that there are facts about w that are not p-physical facts, which means that either DU-p is true

or p-physicalism is false of w.

For any world w without stray p-physically acceptable properties, the Knowledge Argument under

the p-physicality interpretation is exactly the same as under the w-physicality interpretation. The more

interesting cases are worlds with stray p-physically acceptable properties. Our intuitions about whether

Mary learns something new do not seem to depend on whether all the p-physically acceptable properties

of the world satisfy theoretical roles in the final physics theory of the world. Suppose, for example,

that chemical properties were realized not by atoms as we know them, but by orbital model atoms,

and so all biological and psychological properties depend upon such atoms. Suppose further that the

sub-atomic (and extra-atomic) world proceeds roughly as our current theories predict, so that the final

physics theory is as we expect, despite the orbital model atoms. We still expect Mary (who knows all the

p-physical facts, including those about the stray chemical and biological properties) to learn something

new after she is released.

This line of thinking generalizes. Our intuition that Mary learns something new is not driven by


how the facts Mary already knows connect to the final physics of the world. Rather, we understand that

Mary is taught facts about properties that satisfy theoretical roles in physical theories, and we intuit

that no such facts illuminate phenomenal properties. Whether the physical theories are the final physics

of the world in question or the final physics of some other world bears not at all on the intuition.

The Knowledge Argument and Paraphysicality

The orthogonal generalization to paraphysicality presents no serious difficulties. Under this interpreta-

tion, Mary learns all the paraphysical facts in a black and white environment, and, upon her release,

learns a new (broadly scientific) fact about the world. If follows that there are facts beyond the para-

physical facts, which means either DUx is true or paraphysicalism is false.

The intuition that Mary learns something new upon her release might be called into question in

this interpretation. Perhaps the additional facts associated with the properties being paraphysically

acceptable (rather than just physically acceptable) can help Mary learn what it is like to see red. But

upon reflection, this seems unlikely. First, the theoretical roles satisfied by the paraphysically acceptable

properties are the same roles satisfied by physical properties. So if Mary cannot learn what it is like to

see red from the theoretical facts if she is taught all the physical facts, then she cannot learn what it

is like to see red from the theoretical facts if she is taught all the paraphysical facts. Furthermore, the

additional facts that she is taught when she learns the paraphysical facts cannot help her learn what it

is like to see red. It doesn’t appear to matter whether those other facts are compatible with a property

being physical (for example, the presence of a quiddity) or incompatible (for example, the activity of

an immaterial angel). Even if she is told that the angels associated with the brain activity of seeing a

tomato are of an order of angels that present redness when they act, we still have the intuition that she

learns something new upon her release.

What lies behind the intuition that drives the Knowledge Argument is that, apparently, there is

no way to teach Mary what it is like to see red, short of having her experience it (or something very

much like it). Whether the facts we are trying to teach are physical facts or paraphysical facts, she

will not come to know the phenomenal facts as long as she stays confined. This intuition can of course

be resisted, but this is no different under the paraphysicality interpretation than under the physicality

interpretation. The rest of the argument follows as normal, with the premises as plausible or not as they

are under any interpretation.

The generalization of the argument to paraphysicality is orthogonal to the generalization to w-

physicality and p-physicality. The interpretations of the Knowledge Argument under the a-physicality,

w-physicality, and p-physicality interpretations can be applied, mutatis mutandis, to a-paraphysicality,

w-paraphysicality and p-paraphysicality interpretations. We can expect that the Knowledge Argument

applies equally well (or poorly) to these, ruling out a-paraphysicalism (except DUx-a), w-paraphysicalism

(except DUx-w), and p-paraphysicalism (except DUx-p) in equal measure.

3.3 The Conceivability Argument

In this section, I provide an overview of the Conceivability Argument, and interpret it in light of the

various conceptions of physicality described in section 3.1. I begin in section 3.3.1 with an outline of

the argument, followed by a discussion of the epistemic two-dimensional semantics that best establishes

its conclusion, and a formalization of the argument. Again, I will leave the conception of physicality


uninterpreted until section 3.3.2, where I discuss the argument under the a-physicality, w-physicality,

p-physicality, and paraphysicality interpretations.

3.3.1 Overview of the Conceivability Argument

According to the Conceivability Argument (Chalmers, 2009), there is no a priori entailment from physical

facts to phenomenal facts. Unless there are strong necessities, i.e., brute a posteriori necessities, physical

facts and phenomenal facts come apart in some possible world. Thus, either the physical and the

phenomenal are connected in a way that transcends the theoretical role of the former (i.e., DU is true),

or physical properties and phenomenal properties are not necessarily coextensive. So either DU is true,

or physicalism is false.

There are two crucial moves in this argument. The first is establishing the lack of a priori entailment

from physical facts to phenomenal facts. This is done by considering the conceivability of physically

identical but phenomenally distinct scenarios, a subject to which I will return below. The second

is establishing the non-existence of strong necessities. This is best seen using Chalmers’ version of

epistemic two-dimensional semantics. Since the best formalization of the argument includes this appeal

to this two-dimensional semantics, I will begin with an overview of it. I will then describe the formalized

argument.

Chalmers’ Two-Dimensional Semantics

In Chalmers’ (2002b; 2004a; 2004b; 2006) epistemic two-dimensional semantical framework, linguistic

expressions like sentences and referring terms have (at least) two intensions: a primary intension and

a secondary intension. A primary intension is a function from scenarios to extensions, where scenarios

are complete descriptions of states of affairs from an epistemic point of view. For the purposes of the

Conceivability Argument, scenarios can be considered possible worlds with a designated agent and time

(a so-called centered world). The primary intension of an expression is constructed as follows. For each

centered world, the extension of the expression is evaluated in that centered world with the centered

world considered as actual. A secondary intension is a function from possible worlds to extensions. The

secondary intension of an expression is constructed as follows. For each possible world, the extension of

the expression is evaluated in that possible world with that possible world considered as counterfactual.

The primary and secondary intensions can be generalized into a two-dimensional intension. A two-

dimensional intension is a function from ordered pairs of scenarios (centered worlds) and possible worlds

to extensions. Consider an array of extensions created by listing rows of centered worlds and columns of

possible worlds. The row-wise list of centered worlds should be in some sensible order (with the agent

in the actual world being the first scenario) and the column-wise list of possible worlds should be in a

corresponding order (with the actual world being the first one). The two-dimensional intension of an

expression can be constructed as follows. For every element of the array, the extension of the expression is

evaluated in the (column-wise) possible world with the (row-wise) centered world considered as actual.

The primary and secondary intensions of the expression can be extracted from the two-dimensional

intension. The primary intension is the ‘diagonal’ of the array, since the possible world in which the

extension is evaluated is the same as the centered world that is considered as actual. The secondary

intension is the first row of the array, since wherever the expression is evaluated, the actual world (the

centered world of the first row of the array) is considered actual.


Primary intensions are meant to represent epistemic possibilities. A sentence that is true in all sce-

narios evaluated as actual is said to have a necessary primary intension. Such a sentence is epistemically

necessary, or a priori. Call such sentences 1-necessary. A sentence that is true in at least one scenario

evaluated as actual, and false in at least one scenario evaluated as actual, is said to have a contingent pri-

mary intension. Such a sentence is epistemically possible, or a posteriori. Call such sentences 1-possible.

In contrast, secondary intensions are meant to represent metaphysical possibilities. A sentence that is

true in all possible worlds evaluated counterfactually is said to have a necessary secondary intension.

Such a sentence is metaphysically necessary. Call such sentences 2-necessary. A sentence that is true in

at least one possible world evaluated counterfactually, and false in at least one possible world evaluated

counterfactually, is said to have a contingent secondary intension. Such a sentence is metaphysically

possible. Call such sentences 2-possible.

The Conceivability Argument Formalized

Chalmers uses the above epistemic two-dimensional semantical framework to underpin the controver-

sial move from conceivability to possibility in the Conceivability Argument. By using the ‘centered

worlds’ interpretation of the primary intension, Chalmers links an epistemic claim (conceivability) to a

metaphysical claim (physicalism is false).

The argument can be formalized as follows. Let P be all the physical facts,20 Q be all the phenomenal

facts, I be all the indexical facts, and T be the claim that there are no further facts beyond P, Q, and I

(i.e., a “that’s all” claim).

CA1. (PTI & ¬Q) is conceivable

CA2. If (PTI & ¬Q) is conceivable, then (PTI & ¬Q) is 1-possible

CA3. If (PTI & ¬Q) is 1-possible, then either (PTI & ¬Q) is 2-possible or DU true.

CA4. If (PTI & ¬Q) is 2-possible, then physicalism is false

∴ CA5. Either physicalism is false, or DU is true.

According to the first premise, there is no a priori entailment from physical facts (plus indexical facts

and a “that’s all” statement) to phenomenal facts. The support for this claim is just the conceivability

(by an ideal rational agent) of physically identical but phenomenally distinct worlds. An appeal is made

to our intuition that we can imagine without contradiction a world with the same physical facts as

ours, but with no phenomenal properties at all (zombie worlds), or with a uniform redistribution of the

same phenomenal properties (inverted worlds), or with some other distinct distribution of phenomenal

properties. Call all of these phenomenally pathological worlds.

The second premise states that the ideal conceivability of a phenomenally pathological world implies

the 1-possibility of such a world. This follows from the more general claim that for every statement

describing an ideally conceivable state of affairs, there is a corresponding centered world in which the

primary intension of the statement is true. If a phenomenally pathological world is conceivable, then

20In the canonical formulation of the Conceivability Argument, P is all the micro-physical facts, which we can gloss asthe spatiotemporal distribution of all the lowest-level physical properties, along with the physical laws that govern them.If there are no high-level stray p-physically acceptable properties, then the micro-physical facts and the physical factsas I define them in section 3.1.5 entail one another, and this presentation is identical to the canonical formulation. Mypresentation of the argument in terms of all the physical facts serves my purposes better by including the possibility ofhigh-level stray p-physically acceptable properties. I justify this choice below.


there is a centered world for which an ideal agent can rationally apply her concepts in such a way that

the physical facts (along with the indexical facts and the “that’s all” claim) are just as they are in the

actual world, but the phenomenal properties are distinct. Such a world is said to verify the statement.

The third premise relates the 1-possibility of a phenomenally pathological world to the 2-possibility of

such a world. This is the crucial step from epistemic possibility to metaphysical possibility. The premise

states that, if there is a centered world that verifies the description of the phenomenally pathological

world, then either there is a possible world that satisfies the description, or DU is true. The premise

depends on how statements about phenomenal facts and physical facts pick out properties across possible

worlds.

Statements about phenomenal facts pick out the same properties in every centered world that verifies

those facts. If it appears to be pain, then, plausibly, it is pain.21 Statements about physical facts may

or may not pick out the same properties in every centered world that verifies those facts. Suppose that

they do, so that there is nothing more to a physically acceptable property than its theoretical role.22 In

this case, every world that verifies the description of a phenomenally pathological world will satisfy that

description. Worlds with identical physically acceptable properties but distinct phenomenal properties

would therefore be metaphysically possible. Alternatively, suppose that statements about physical facts

do not pick out the same properties in every centered world that verifies those facts; there is more to a

physically acceptable property than its theoretical role. Even still, those extra-theoretic details may not

connect the physical properties with phenomenal facts. For example, maybe the physically acceptable

property just has a quiddity beyond its theoretical role. In that case, while not every world that verifies

the description of a phenomenally pathological world will satisfy the description, some of those verifying

worlds will satisfy it. So again, worlds with identical physically acceptable properties but distinct

phenomenal properties would be metaphysically possible. However, things are different if the extra-

theoretic details do connect the physically acceptable properties to phenomenal properties. For example,

perhaps phenomenal properties are the intrinsic natures of otherwise dispositional physically acceptable

properties. In that case, none of the centered worlds that verify the description of a phenomenally

pathological world will satisfy the description. But it also means that DU is true.23

According to the fourth premise, the 2-possibility of a phenomenally pathological world implies

that physicalism is false. This is straightforward:24 if physically acceptable properties and phenomenal

properties come apart in some possible world, then they are not necessarily coextensive. Therefore,

phenomenal properties are not type-identical to physically acceptable properties. This implies that

physicalism is false.

21That is to say, terms referring to phenomenal properties are epistemically rigid.22For example, perhaps being a negatively-charged lepton of mass 511 keV is all there is to being an electron.23The argument in this paragraph assumes that statements of physical fact have the same primary intention when those

facts are theoretically identical (i.e., the same theoretical roles are satisfied in the same locations), but don’t necessarilyhave the same secondary intention. The secondary intention of such a statement “sees through” the theoretical facts tothe nature of the properties satisfying them, whereas the primary intention does not. This is why a world verifying thedescription of a phenomenally pathological world might not satisfy it. Contrast this with the treatment of physical factsin the Knowledge Argument, above, where it was stated without comment that the argument established that there arebroadly scientific facts that are not physical facts.

The reader may object that it is difficult to reconcile the two treatments of physical facts. If so, I can change the thirdpremise of the Conceivability Argument to be that the 1-possibility of phenomenally pathological worlds implies their2-possibility. The support for this new premise would be as above, except that statements of physical facts would certainlypick out the same properties in every centered world. The fourth premise would then have to be modified: the 2-possibilityof phenomenally pathological worlds would imply that either physicalism is false or DU is true. The conclusion of theargument would then be the same. Disconnecting the statement of physical facts from the individuation of propertiesacross worlds just forces the connection elsewhere in the argument.

24But see footnote 23.


As with the Knowledge Argument, there are a variety of physicalist responses to the various premises.

Some argue that there is an a priori entailment from physical facts to phenomenal facts (Dennett, 1991;

Dretske, 1995), which contradicts CA1. Others argue that there are strong necessities (Block and

Stalnaker, 1999; Perry, 2001), so that some metaphysical necessities hold despite epistemic possibilities

given a complete description of the state of affairs, which contradicts CA3. Again, since my concern is

the role of the physical in the Conceivability Argument, I will not evaluate the success or failure of these

responses.

3.3.2 The Conceivability Argument and Physicality

According to the Conceivability Argument, the physical facts (P)25 do not entail the phenomenal facts

(Q), so it is epistemically possible for there to be two worlds that are physically identical but phenome-

nally distinct. The epistemic possibility of such a world implies either its metaphysical possibility, or the

truth of DU. Since the metaphysical possibility of physical duplicates that are not phenomenal duplicates

contradicts physicalism, either physicalism is false or DU is true.

As with the Knowledge Argument, some conception of the physical is needed to interpret three parts

of the argument: the physical facts (P) that fail to entail the phenomenal facts, the DU position that

is permitted by the argument, and the physicalism that is claimed to be false. Again, the argument

does not equivocate, so there is a single conception of the physical being used in these parts. But

which conception of the physical being used can be uniformly interpreted as a-physical, w-physical, or

p-physical, as well as any variation of paraphysical.

The Conceivability Argument and a-Physicality

The natural interpretation of the argument is that it is about a-physicality. The central intuition upon

which the argument relies is that it is ideally conceivable for there to be worlds that are physically

identical to, but phenomenally distinct from, our world. Furthermore, the point of the argument is to

argue that physicalism is false, that is, physicalism is false of the actual world. Finally, in the canonical

formulation of the argument, P is described as the micro-physical facts, along with the physical laws.26

This is an expression of a concern about the entailments of the final physics of the world, and the

distribution of properties treated directly by the final physics. Together, these considerations suggest

that it is natural to interpret the argument as being about a-physicality.

Under this interpretation, a world a-physically identical but phenomenally distinct from the actual

world is ideally conceivable. This means that such a world is 1-possible. If statements about a-physical

facts pick out the same a-physically acceptable properties across all possible worlds, or if whatever

individuates a-physically acceptable properties has nothing to do with phenomenal properties, then the

phenomenally pathological world is 2-possible, and a-physicalism is false. Otherwise, DU-a is true.

Most arguments for or against the premises of the Conceivability Argument are made for or against

the premises under this interpretation. Those who find the Conceivability Argument persuasive will find

it persuasive as an argument about a-physicality. Those who do not find it persuasive will not be moved

by it even if it is about a-physicality.

25We can neglect the indexical facts and the “that’s all” claim for convenience.26See footnote 20.


The Conceivability Argument and w-Physicality

Even more easily than the Knowledge Argument, the Conceivability Argument can be generalized to be

about w-physicality rather than a-physicality.

Recall that P, Q, and I were defined to be the physical, phenomenal, and indexical facts, respectively.

The natural interpretation of this is that they are facts about the actual world. But there is no reason

to limit the definition to such. Let us define P, Q, and I to be the w-physical, phenomenal, and indexical

facts, respectively, of some arbitrary centered world w. The Conceivability Argument maintains its

strength despite this substitution.

Under the w-physicality interpretation, a world w-physically identical but phenomenally distinct from

w is ideally conceivable. This means that such a world is 1-possible. If statements about w-physical

facts pick out the same w-physically acceptable properties across all possible worlds, or if whatever

individuates w-physically acceptable properties has nothing to do with phenomenal properties, then the

phenomenally pathological world is 2-possible, and w-physicalism is false. Otherwise, DU-w is true.

The support for the premises of the argument under the w-physicality interpretation remain as strong

as they are under the a-physicality interpretation. According to CA1, for any world w, the w-physical

facts do not entail the phenomenal facts of w. The support for this premise remains unchanged: we

can imagine without contradiction that the w-physical facts hold with distinct phenomenal facts. Our

intuition on this matter does not depend upon the physical facts being about the actual world.

According to CA2, the conceivability of a phenomenally pathological world entails its 1-possibility.

The generality of Chalmers’ epistemic two-dimensional semantics ensures this is the case. The phenom-

enally pathological world is ideally conceivable, so there is a centered world that verifies its description.

Again, it doesn’t matter whether the world is phenomenally pathological with respect to the actual

world or with respect to some other world.

According to CA3, the 1-possibility of the phenomenally pathological world entails either its 2-

possibility, or the truth of DU-w. For symmetry reasons, if statements about the phenomenal facts of

the actual world pick out the same phenomenal properties in every world that verifies those facts, then we

can expect that statements about the phenomenal facts of any other world pick out the same phenomenal

properties in every world that verifies those facts. Regarding w-physical properties, if statements about

w-physical facts pick out the same w-physically acceptable properties in every world that verifies those

facts, then every centered world that verifies the description of a phenomenally pathological world will

also satisfy the description, making the phenomenally pathological world 2-possible. If such statements

do not, but whatever else individuates w-physically acceptable properties is not connected to phenomenal

properties, then some centered worlds that verify the description of a phenomenally pathological world

will also satisfy the description, again making a phenomenally pathological world 2-possible. But if such

statements do not because w-physically acceptable properties are connected to phenomenal properties

in a way that goes beyond the theoretical roles of the former, then DU-w is true.

According to CA4, if the phenomenally pathological world is 2-possible, then w-physicalism is false.

If w-physically acceptable properties and phenomenal properties come apart in some possible world,

then they are not necessarily coextensive, which makes w-physicalism false.

The Conceivability Argument under the w-physicality interpretation is as strong (or as weak) as the

argument under the a-physicality interpretation. The general form of the Conceivability Argument, plus

the flexibility of Chalmers’ epistemic two-dimensional semantics, makes this so.


The Conceivability Argument and p-Physicality

The generalization to p-physicality requires a bit more thought than the generalization to w-physicality.

Let P, Q, and I be the p-physical, phenomenal, and indexical facts, respectively, of some arbitrary cen-

tered world w. Under the p-physicality interpretation, a world p-physically identical but phenomenally

distinct from w is ideally conceivable. This means that such a world is 1-possible. If statements about

p-physical facts pick out the same p-physically acceptable properties across all possible worlds, or if

whatever individuates p-physically acceptable properties has nothing to do with phenomenal properties,

then the phenomenally pathological world is 2-possible, and p-physicalism is false. Otherwise, DU-p is

true.

For all worlds w that have no stray p-physically acceptable properties, the p-physicality interpretation

of the Conceivability Argument is the same as the w-physicality interpretation. For such worlds, the

argument is as persuasive (or as unpersuasive) under one interpretation as under the other. Therefore,

the relative persuasiveness of the Conceivability Argument under the p-physicality interpretation can

only be evaluated by considering worlds with stray p-physically acceptable properties. I will therefore

restrict the domain of worlds under consideration to all worlds that have stray p-physically acceptable

properties.

So consider any such world w. The distinguishing feature of the stray p-physically acceptable prop-

erties of w relative to w-physically acceptable properties is that the latter, and not the former, satisfy

theoretical roles in the final physics of w. Therefore, while the p-physical facts entail the w-physical

facts, the w-physical facts do not entail the p-physical facts. By defining P to be all the p-physical facts

of w, I am introducing more facts about world w from which Q might be entailed. So it is critical to

ensure that the intuition driving the first part of the argument, viz., that the physical facts do not entail

the phenomenal facts because phenomenally pathological worlds are ideally conceivable, still holds.

Recall once more that in the canonical formulation of the argument, Chalmers defines P to be the

micro-physical facts. Plausibly, these facts entail all the w-physical facts of w, but do not entail the

stray p-physical facts. It is tempting to think that incorporating facts about stray p-physical properties

might undermine our intuition that phenomenally pathological worlds are conceivable. After all, the

Conceivability Argument highlights the difference between, on the one hand, the paradigmatic cases of

higher-level w-physically acceptable properties of w, which are entailed by the micro-physical facts, and

on the other hand, phenomenal properties, which are (purportedly) not so entailed. But this line of

thinking is mistaken. It is not the smallness of the micro-physical properties that guide our intuitions

about whether micro-physical facts entail (or fail to entail) facts about other physically acceptable

properties or phenomenal properties. It is their fundamentality that guides our intuitions.

Plausibly, the fundamental physical facts are the smallest set of facts which entail all the physical

facts. They are, so to speak, the axiomatic physical facts. The set of micro-physical facts is fundamental

because no proper subset of them entails the remainder of them. In worlds without stray p-physically

acceptable properties, all the fundamental physical facts are micro-physical facts, since the micro-physical

facts are the smallest set of facts that entail all the other physical facts. But in worlds with stray p-

physically acceptable properties, the micro-physical facts do not exhaust the fundamental physical facts,

since they do not entail the facts about the stray p-physically acceptable properties. Therefore, in such

worlds, at least some facts about stray p-physically acceptable properties must be among the fundamental

physical facts.27

27This is why, in my presentation of the argument, I defined P to be all the physical facts; the fundamental physical


Other than failing to satisfy theoretical roles in the final physics of w, the stray p-physically acceptable

properties of w are relevantly similar to w-physically acceptable properties. Both satisfy theoretical roles

in some final physical theory. It is no more or less plausible for stray p-physically acceptable properties

to be purely dispositional, or to have phenomenal properties as their intrinsic natures, than it is for

w-physically acceptable properties. If w-physically acceptable properties are individuated exclusively by

their theoretical roles, then so are stray p-physically acceptable properties. If the w-physical facts do

not entail phenomenal properties, then neither do the p-physical facts.28 We should therefore expect

that the support for the premises of the Conceivability Argument under the p-physicality interpretation

is as strong as it is under the w-physicality interpretation.

According to CA1, the p-physical facts do not entail the phenomenal facts of w. This premise is

supported as before: we can imagine without contradiction that the p-physical facts hold with distinct

phenomenal facts. Our intuition on this matter does not depend upon the whether the physical facts in

question are about properties that satisfy the final physics of w or the final physics of some other world.


This premise is no different under the p-physicality interpretation than under the w-physicality inter-

pretations. Both follow from the more general principle, based on Chalmers’ epistemic two-dimensional

semantics, that for every ideally conceivable description there is a centered world that verifies the de-

scription.

According to CA3, the 1-possibility of the phenomenally pathological world entails either its 2-

possibility, or the truth of DU-p. Statements about phenomenal facts pick out the same properties in all

possible worlds that verify those facts; whether the worlds in question have stray p-physical properties

or not has no bearing on this. With respect to statements about physical facts, consider the following.

If DU-w is true of w, then DU-p is also true of w, whether or not there are stray p-physical properties.

So suppose DU-w is false. If statements about stray p-physical facts pick out the same p-physically

acceptable properties in every world that verifies those facts, then every centered world that verifies the

description of a phenomenally pathological world will also satisfy the description, making phenomenally

pathological worlds 2-possible. Otherwise, if such statements do not, but whatever else individuates

stray p-physically acceptable properties is not connected to phenomenal properties, then some centered

worlds that verify the description of a phenomenally pathological world will also satisfy the description,

again making a phenomenally pathological world 2-possible. But if such statements do not because stray

p-physically acceptable properties are connected to the phenomenal properties in a way that goes beyond

the theoretical roles of the former, then DU-p is true. Therefore, the support for the third premise is as

strong under the p-physicality interpretation as it is under the w-physicality interpretation.

Finally, according to CA4, if the phenomenally pathological world is 2-possible, then p-physicalism

is false. As before, if p-physically acceptable properties and phenomenal properties come apart in some

possible world, then they are not necessarily coextensive, which makes p-physicalism false.

This confirms the expectation that the Conceivability Argument under the p-physicality interpre-

tation is as persuasive (or as unpersuasive) as it is under the w-physicality interpretation. Nothing in

facts and all the physical facts entail one another. See footnote 20.28And neither do the combination of w-physical facts and stray p-physical facts. As described in footnote 14, if arbitrary

combinations of fundamental properties are possible, there there is a world where the stray p-physically acceptable prop-erties of w are not stray (being nothing over and above the physical properties of that world), and where the w-physicallyacceptable properties are also nothing over and above the physical properties of that world (perhaps because the w-physicalproperties are among the physical properties). If the physical facts of that world do not entail phenomenal facts, thenneither do the physical facts of w.


the form of the argument makes it specific to the w-physicality interpretation, and Chalmers’ epistemic

two-dimensional semantics is general enough to accommodate the p-physicality interpretation.

The Conceivability Argument and Paraphysicality

The orthogonal generalization of the Conceivability Argument to paraphysicality also poses no serious

difficulties. Let P, Q, and I be the paraphysical, phenomenal, and indexical facts, respectively. Under

the paraphysicality interpretation, a paraphysically identical but phenomenally distinct world is ideally

conceivable. This means that such a world is 1-possible. If statements about paraphysical facts pick out

the same paraphysically acceptable properties across all possible worlds, or if whatever individuates pa-

raphysically acceptable properties has nothing to do with phenomenal properties, then the phenomenally

pathological world is 2-possible, and paraphysicalism is false. Otherwise, DUx is true.

The paraphysical facts entail the physical facts, but the physical facts do not entail the paraphysical

facts. Therefore, as with the p-physicality generalization, by defining P to be all the paraphysical facts, I

am introducing more facts from which Q might be entailed. So again, we must ensure that the intuition

driving the first part of the argument, viz., that the paraphysical facts do not entail the phenomenal

facts because phenomenally pathological worlds are ideally conceivable, still holds.

The ideal conceivability of phenomenally pathological worlds stems from our ability to separate the

theoretical roles that properties satisfy from the phenomenal properties associated with those properties.

If phenomenal properties cannot be given a functional characterization, then what a property does won’t

help us determine how that property feels. The other facts about the property, like whether it has a

quiddity or whether it results from the action of an immaterial angel, are irrelevant to this ability.

Even the property being deeply connected to phenomenal properties appears irrelevant; the possibility

of panprotopsychism doesn’t stop Chalmers from claiming that phenomenally pathological worlds are

conceivable.29 Given that the theoretical roles that are satisfied by paraphysically acceptable properties

are the same as those satisfied by physically acceptable properties, the phenomenally pathological worlds

are as conceivable whether we are considering physical facts or paraphysical facts. Zombie angels are as

conceivable as zombie humans.

The support for the premises of the argument under the paraphysicality interpretation are thus as

strong as they are under the physicality interpretation. According to CA1, for any world w, the para-

physical facts do not entail the phenomenal facts. As argued above, we can imagine without contradiction

that the paraphysical facts hold with distinct phenomenal facts.


Again, this premise follows from the more general principle that for every ideally conceivable description

there is a centered world that verifies the description.

According to CA3, the 1-possibility of phenomenally pathological worlds entails either their 2-

possibility, or the truth of DUx. Again, statements about the phenomenal facts pick out the same

phenomenal properties in every world that verifies those facts. With respect to paraphysical properties,

if statements about paraphysical facts pick out the same paraphysically acceptable properties in every

world that verifies those facts,30 then every centered world that verifies the description of a phenome-

29See footnote 23. It might be thought that being deeply connected to phenomenal properties ought to make phe-nomenally pathological worlds inconceivable. However, recall that the primary intention of statements of physical factsmakes reference only to the theoretical physical facts. The same can be said for statements of paraphysical facts. Theextra-theoretic details of those properties become relevant when considering the secondary intention.

30That is, the paraphysical properties are physical properties of a certain kind.


nally pathological world will also satisfy the description, making the phenomenally pathological world

2-possible. If such statements do not, but whatever else individuates paraphysically acceptable proper-

ties is not connected to phenomenal properties, then some centered worlds that verify the description

of a phenomenally pathological world will also satisfy the description, again making a phenomenally

pathological world 2-possible. It doesn’t matter whether what individuates the paraphysical property

is a quiddity or the activity of an immaterial angel. But if the reason that statements of paraphysical

facts do not pick out the same paraphysically acceptable property in every world that verifies those pa-

raphysical facts is that the paraphysically acceptable properties are connected to phenomenal properties

in a way that goes beyond theoretical roles, then DUx is true.

According to CA4, if the phenomenally pathological world is 2-possible, then paraphysicalism is false.

If paraphysically acceptable properties and phenomenal properties come apart in some possible world,

then they are not necessarily coextensive, which makes paraphysicalism false.

The Conceivability Argument under the paraphysicality interpretation is as strong (or as weak) as

the argument under the physicality interpretation. Since the generalization to paraphysicality is orthog-

onal to the generalization to w-physicality and p-physicality, we can expect the same to hold true for

a-paraphysicality, w-paraphysicality, and p-paraphysicality. The relative persuasiveness of the Conceiv-

ability Argument under the a-physicality and p-paraphysicality interpretations (and all interpretations

in between) is the same.

3.4 Lawfully Interacting Strong Dualism

In this section, I identify a popular anti-physicalist position, which I call lawfully interacting strong

dualism. I argue that this position entails either p-paraphysicalism or DUx-p. In the former case, this

kind of dualism is ruled out by the Knowledge Argument and the Conceivability Argument. In the latter

case, it remains less plausible than its physicalist analogue, DU-a.

3.4.1 Varieties of Dualism

The Knowledge Argument and the Conceivability Argument are arguments against physicalism, with

the possible exception of DU. Under the natural interpretation of these arguments, they are against

a-physicalism (except perhaps DU-a). Both arguments can be generalized quite easily to any possible

world w, so that they are against w-physicalism (except perhaps DU-w). All these positions are monist

accounts of phenomenal properties, which is to say, phenomenal properties are not ultimately distinct

from physical properties.

The anti-physicalist arguments also exclude p-physicalism (except perhaps DU-p). These positions

may or may not be monist. If there are no stray p-physically acceptable properties in world w, then

there is no difference between, on the one hand, p-physicalism and DU-p, and on the other hand, w-

physicalism and DU-w. For such worlds, the former are thus also monist accounts. If there are stray

p-physically acceptable properties in world w, then p-physicalism and DU-p are monist in one sense

and dualist in another. They are dualist in the sense that stray p-physically acceptable properties are

not w-physically acceptable. But they are also monist in the sense that phenomenal properties are not

ultimately distinct from p-physically acceptable properties.

The anti-physicalist arguments also exclude a-paraphysicalism (except DUx-a) and w-paraphysicalism

(except DUx-w). These are monist in the sense that all properties satisfy the theoretical role of some


property posited by the final physics of the world, or are nothing over and above such properties. The

arguments also exclude p-paraphysicalism (except DUx-p). As above, p-paraphysicalism may or may

not be monist, depending on whether there are any stray p-paraphysically acceptable properties. If there

are stray p-paraphysically acceptable properties in w, then p-paraphysicalism is dualist in the sense that

not every property satisfies the theoretical role of some property posited by the final physics of w. But

it is monist in the sense that phenomenal properties are not ultimately distinct from p-paraphysically


While they are strictly arguments against monist positions, the Knowledge Argument and the Con-

ceivability Argument are sometimes used to argue that some form of strong dualism is true. Accord-

ing to strongly dualist accounts of phenomenal properties, phenomenal properties are radically dis-

tinct from physical properties. Such accounts can be distinguished by their position on the causal

efficacy of phenomenal properties with respect to physically acceptable properties: interactionist and

non-interactionist. According to non-interactionist accounts, phenomenal properties are not causally

efficacious with respect to physically acceptable properties. Perhaps they are causally isolated from

physically acceptable properties: neither caused by nor causing physical events, as per a parallelist ac-

count of phenomenal properties. Or perhaps they are caused by, but do not cause, physical events, as

per a strong form of epiphenomenalism. On the other hand, according to interactionist accounts, phe-

nomenal properties cause physical events in the same way that physically acceptable properties cause

them. On such accounts, the physical world is not causally closed. If the instantiation of phenomenal

properties is correlated with the instantiation of (higher-level) physically acceptable properties, then the

phenomenal properties are downwardly causal.

I will set aside non-interactionist accounts of phenomenal properties. While they may go against

intuitions that our own phenomenal states have effects on our behaviour, they are plausibly among

those accounts allowed by the Knowledge Argument and the Conceivability Argument.

More interesting are the interactionist accounts of phenomenal properties. There are two ways that a

phenomenal property might interact with physically acceptable properties. First, the interaction might

be wholly lawful. In this case, phenomenal properties lawfully arise from a distribution of physically

acceptable properties, and go on to cause changes to that distribution in a lawful manner. The former

indicates a brute psycho-physical law, which may be synchronic or diachronic. The latter indicates

a causal law governing the phenomenal properties. Second, the interaction of phenomenal properties

and physically acceptable properties might be partly anomalous. In this case, either the instantiation

of phenomenal properties bears no relation to the distribution of physically acceptable properties, or,

once phenomenal events are instantiated, the laws don’t determine the objective chances of systems of

physically acceptable properties evolving into various states. The former indicates a radically separate

domain of phenomenal properties, intervening in the world of physically acceptable properties in possibly

lawful ways. The latter indicates a kind of radical libertarian free will for systems that instantiate


There are reasons to believe that no anomalously interacting strong dualist account is true of the ac-

tual world. Prima facie, the instantiation of phenomenal properties is closely related to the instantiation

of certain physically acceptable properties, specifically neural properties, and changes to some of those

physically acceptable properties appear to correlate with changes in phenomenal properties in a regular

way. Furthermore, phenomenal properties causing changes to the distribution of physically acceptable

properties in a way governed by no law whatsoever violates the (so-called) completeness of the laws of


nature, which we have reason to believe is metaphysically impossible (Lange, 2009).

Nevertheless, I will not be arguing in detail against anomalously interacting strong dualism. The

distinction between lawfully interacting phenomenal properties and anomalously interacting ones is sur-

prisingly important when considering the conclusions of the Knowledge Argument and the Conceivability

Argument. As I argue below, lawfully interacting strong dualism is either ruled out by the Knowledge

Argument and the Conceivability Argument, or made less plausible than a-physicalism.

3.4.2 Interactionist Dualism and Stray Properties

In this section, I argue that if lawfully interacting strong dualism is true of a world, then the world

instantiates at least one stray p-paraphysically acceptable property.

According to lawfully interacting strong dualism, when physically acceptable properties are in a

certain (usually complex) configuration, phenomenal properties are instantiated according to a brute

psycho-physical law. These properties then go on cause the distribution of physically acceptable proper-

ties to change; call these the resultant properties. The instantiation of the resultant properties and their

further causal contribution are lawful with respect to physically acceptable properties. The resultant

properties thus satisfy nomologically necessary statements characteristic of (deductively closed) physical

theories, from which it follows that they satisfy theoretical roles in some physical theory. By hypothesis,

they cannot satisfy the theoretical roles in the final physics of the world in which they are instantiated.

Therefore, they must be stray p-paraphysically acceptable properties.

The conclusion can be most easily seen if we set aside our intuitions about phenomenal properties.31

When physically acceptable properties are in a certain complex configuration, new emergent properties

are instantiated according to a brute emergent law. These properties then go on to cause the distribution

of physically acceptable properties to evolve lawfully, but not in the way predicted by the final physics

of the world. The emergent properties therefore satisfy lawlike statements characteristic of physical

theories, but not statements of the final physics. The properties are therefore stray p-paraphysically


Consider a simplistic example of a lawfully interacting strong dualist position. Suppose that phenom-

enal properties of the actual world are correlated with actual orbital model atoms. When fundamental

particles combine in certain pre-atomic configurations, phenomenal properties are instantiated, which

cause those particles to behave as orbital model atoms, contrary to physics. Being an orbital model atom

satisfies no theoretical role in final physics, but it does satisfy a theoretical role in the final physics of some

nomologically impossible world. Therefore, being an orbital model atom is a stray p-paraphysically ac-

ceptable property. Thus, the simplistic dualist account requires the existence of a stray p-paraphysically

acceptable property.

For a more realistic example of a lawfully interacting strong dualist position, consider the follow-

ing. Suppose that actual phenomenal properties arise from some distribution of a-physically acceptable

properties in the brain, and change the way that neurons in the brain interact with one another, so

that the neural activation patterns are not what would be expected based purely on final physics. The

unusual neuron activation patterns are instantiated whenever certain a-physical conditions are satisfied,

i.e., the very conditions that instantiate the phenomenal properties. The patterns themselves evolve

contrary to the predictions of final physics, and go on to cause further patterns (and other changes to

31These intuitions will be used when applying the Knowledge Argument and the Conceivability Argument to lawfullyinteracting strong dualism.


the distribution of a-physically acceptable properties) in a lawful way. There are, therefore, nomolog-

ically necessary statements describing the objective chances of certain brain states instantiating some

neural activation patterns, which, when instantiated, have objective chances to cause certain other brain

states. The statements are not statements of final physics, but they are statements of the final physics of

some nomologically impossible world. Therefore, there are stray p-paraphysically acceptable properties.

3.4.3 Interactionist Dualism, p-Paraphysicalism, and DUx-p

Recall that p-paraphysicalism is the position according to which all broadly scientific properties p-

paraphysically acceptable. This is explicitly the position of at least some strong dualists of historical

importance.32 According to this version of strong dualism, mental properties (including phenomenal

properties) just are emergent properties that interact with a-physically acceptable properties in lawful

ways. The phenomenal properties themselves would therefore satisfy high-level theoretical roles, and

thus theoretical roles in the final physics of some nomologically impossible world.

DUx-p is the position that phenomenal properties and p-physically acceptable properties are meta-

physically necessarily connected in a way that goes beyond the theoretical roles of the p-paraphysically ac-

ceptable properties. In the context of lawfully interacting strong dualism, it is the stray p-paraphysically

acceptable properties to which phenomenal properties are metaphysically necessarily connected. The

stray p-paraphysically acceptable and the phenomenal might be two aspects of the same property. The

stray p-paraphysically acceptable and the phenomenal might be two manifestations of the same, distinct,

non-p-paraphysically acceptable and non-phenomenal property. Or phenomenal properties might be the

intrinsic natures, or the ultimate categorical bases, of the otherwise dispositional stray p-paraphysically


Perhaps surprisingly, no lawfully interacting strong dualist position about phenomenal properties

can be other than p-paraphysicalist (including DUx-p). Any strong dualist position that fails to be

p-paraphysicalist fails to be lawfully causal. Any property that lawfully interacts with p-physically

acceptable properties will satisfy theoretical roles in some physical theory. Thus, that property will be p-

paraphysically acceptable. It is possible for a-physically acceptable properties (or indeed a-paraphysically

acceptable properties) to fail to be lawfully causally closed, since there is no assurance that every property

that lawfully interacts with them will satisfy a theoretical role of final physics. But it is impossible for p-

paraphysically acceptable properties to fail to be lawfully causally closed, since any property that lawfully

interacts with them will satisfy a theoretical role in some final physics. If p-paraphysicalism (including

DUx-p) is false, then phenomenal properties must fail to satisfy any lawlike statements characteristic of

physical theories. Thus, they either anomalously interact with p-paraphysically acceptable properties,

or don’t interact with them at all.

3.4.4 Interactionist Dualism and Anti-physicalist Arguments

In the previous section, I argued that every lawfully interacting strong dualist account of phenomenal

properties entails p-paraphysicalism (including DUx-p). This result places such accounts among the

targets of the Knowledge Argument and the Conceivability Argument. Recall from sections 3.2 and 3.3

that the Knowledge Argument and the Conceivability Argument can be interpreted to be about p-

paraphysicality, so that they rule out p-paraphysicalism (except DUx-p). If so, then the Knowledge

32The early British Emergentists can be read as advocating p-paraphysicalism about chemistry, biology, and psychology.See McLaughlin (1992).


Argument and the Conceivability Argument rule out all lawfully interacting dualist accounts that do

not entail DUx-p. The remaining lawfully interacting dualist accounts, those that do entail DUx-p, can

be seen to be less plausible than analogous monist accounts.

To make this explicit, it is worth revisiting the Knowledge Argument and the Conceivability Argument

in light of a lawfully interacting strong dualist account. Consider the following example of such an

account, one that takes the collapse of a superposed state in quantum mechanics to be caused by

phenomenal properties (Chalmers (2002a)):

On the standard formulation of quantum mechanics, the state of the world is described bya wave function, according to which physical entities are often in a superposed state (e.g.,in a superposition of two different positions), even though superpositions are never directlyobserved. On the standard dynamics, the wave function can evolve in two ways: linear evolu-tion by the Schrodinger equation (which tends to produce superposed states), and nonlinearcollapses from superposed states into nonsuperposed states. [. . . ] Schrodinger evolution isconstantly ongoing, but on the standard formulation, collapses occur only occasion[al]ly, onmeasurement. [. . . ] [I]t is natural to suggest that a measurement is precisely a consciousobservation, and that this conscious observation causes a collapse.

For this account, we can suppose that conscious observation is a phenomenal event that is always

associated with a particular a-physically acceptable brain state, and that final physics is silent on what

causes collapse.

Consider again the Knowledge Argument (see section 3.2.1). Suppose that Mary learns in her con-

finement not only all the a-physical facts, but also all the interacting emergent facts associated with

collapse: which brain states correlate with the collapse of superposed states (inexplicable in light of the

final physics), and the laws that govern collapse with respect to brain state (which are not laws of the

final physics). Mary is then released from confinement, and looks upon a red tomato. The addition of

the emergent facts does not alter our intuitions about whether Mary learns something new after being

released. The argument goes through as normal:

KA1∗. Mary knew all the a-physical and interacting emergent facts before her release.

KA2. Mary learned a new (phenomenal) fact after her release.

∴ KA3∗. There are (phenomenal) facts that are neither a-physical facts nor interacting emergent facts.

KA4∗. If there are (phenomenal) facts that are neither a-physical facts nor interacting emergent facts,

then either some form of DU is true or both a-physicalism and lawfully interacting strong dualism

are false.

∴ KA5∗. Either some form of DU is true, or both a-physicalism and lawfully interacting strong dualism are

false.

The responses available to the advocate of lawfully interacting strong dualism are the same as those

available to the physicalist. Any strategy that the dualist can use to deny the conclusion can be used

with equal effectiveness by the physicalist.

The same treatment can be given to the Conceivability Argument (see section 3.3.1). Let P be all

the a-physical facts, E be all the interacting emergent facts (the brain states leading to collapse and the

laws governing them), Q be all the phenomenal facts, I be all the indexical facts, and T be the claim

that there are no further facts beyond P, E, Q, and I.


CA1∗. (PETI & ¬Q) is conceivable

CA2∗. If (PETI & ¬Q) is conceivable, then (PETI & ¬Q) is 1-possible

CA3∗. If (PETI & ¬Q) is 1-possible, then either (PETI & ¬Q) is 2-possible or some form of DU true.

CA4∗. If (PETI & ¬Q) is 2-possible, then both a-physicalism and lawfully interacting strong dualism are

false

∴ CA5∗. Either both a-physicalism and lawfully interacting strong dualism are false, or some form of DU

is true.

Just as the a-physical facts (purportedly) fail to entail a priori the phenomenal facts, the causal facts

about interacting emergent properties (like which brain states cause collapse in which situations) fail

to entail a priori the phenomenal facts. So either some form of DU is true, or both a-physicalism and

interactionist dualism is false.

This conclusion, while perhaps surprising, should not be controversial. Chalmers (2002a) himself

acknowledges, in passing, that his type-D dualism (which corresponds to interactionist dualism, and

of which conscious observation causing collapse is an example) should be interpreted as a kind of DU,

with “structural/dispositional” properties beyond a-physical ones having phenomenal properties as their

intrinsic natures.33 Again, the responses available to the advocate of lawfully interacting strong dualism

are the same as those available to the physicalist.

The Knowledge Argument and the Conceivability Argument therefore rule out all lawfully interacting

strong dualist positions that do not entail DUx-p, just as they rule out a-physicalism save perhaps DU-a.

It is worth considering, then, whether the sort of DUx-p that is compatible with lawfully interacting

strong dualism is made more plausible than DU-a.

Recall that DUx-p is entailed by DU-a; the former position encompasses the latter, so it cannot help

but be more likely. But the sort of DUx-p relevant to lawfully interacting strong dualism is one where

there are stray p-paraphysically acceptable properties, and it is those properties with which phenomenal

properties are metaphysically necessarily connected in a way that transcends their theoretical roles. For

convenience, call this DU-e.

DU-e is no more plausible than DU-a. Whatever reasons we have to to think that the phenomenal

and the physical or paraphysical are two aspects of the same property apply equally well to DU-e as

to DU-a. Similarly, the existence of neutral properties manifesting variously as phenomenal properties

or paraphysically acceptable properties is no more likely if only stray p-paraphysically acceptable prop-

erties, and not a-physically acceptable properties, are manifestations. Finally, metaphysical arguments

that phenomenal properties can serve as the intrinsic natures of otherwise dispositional paraphysically

acceptable properties are not made any more plausible by restricting the phenomenal properties to being

33He says in an unnumbered footnote in section 11:

In this way, we can see that type-D [interactionist dualist] views and type-F [DU-a] views are quite closelyrelated. We can imagine that if a type-D view is true and there are microphysical causal gaps, we couldbe led through physical observation alone to postulate higher-level entities to fill these gaps—“psychons”,say—where these are characterized in wholly structural/dispositional terms. The type-D view adds to thisthe suggestion that psychons have an intrinsic phenomenal nature. The main difference between the type-Dview and the type-F view is that the type-D view involves fundamental causation above the microphysicallevel.


the intrinsic natures of stray p-paraphysically acceptable properties.34

On the other hand, there are reasons to think that DU-a is more plausible than DU-e. As described

in section 1.2.1, there is no evidence for the existence of stray properties. Whatever reasons we have

to presume in favour of physicalism over non-physicalist positions are reasons to presume in favour of

DU-a over DU-e.

Without the support of the Knowledge Argument and the Conceivability Argument, lawfully inter-

acting strong dualism appears to be an orphaned position. Some such accounts are positively ruled out

by the anti-physicalist arguments, while the remaining ones are prima facie less plausible than physicalist

analogues.

34There is one exception to this. If phenomenal properties are the intrinsic natures of all a-physically acceptableproperties, then it is natural to think that higher-level phenomenal natures are constituted by lower-level phenomenalnatures in a way that is analogous to how higher-level physically acceptable properties are constituted by lower-levelphysically acceptable properties. Call this DU-a view constitutive panprotopsychism. Constitutive panprotopsychism isproblematic since there appears to be no way that one phenomenal property can metaphysically necessitate another. Thisis the combination problem. See footnote 9 of chapter 1.

DU-e may not suffer from the combination problem, since it is possible for phenomenal properties to be the intrinsicnatures of only certain higher-level properties. But this DU-e replaces the combination problem with a requirementthat there be a possibly infinite number of brute relations between particular phenomenal properties and the higher-levelproperties of which they are intrinsic natures. I have no intuitions about which of these problems is more serious. ButDU-e doesn’t seem any more likely than a version of DU-a that similarly restricts phenomenal properties to be the intrinsicnatures of only higher-level properties.

Chapter 4

Characterizing Strong Emergentism

In chapter 3, I argued that lawfully interacting strong dualist accounts are ruled out by the Knowledge

and Conceivability arguments, unless DUx-p is true. Lawfully interacting strong dualism is one species

of a more general position I will call strong emergentism. According to the strong emergentist, (a-

)physically1 unacceptable properties are instantiated in systems of physically acceptable properties in

complex configurations, and go on to lawfully downwardly cause changes to the distribution of physically

acceptable properties. Lawfully interacting strong dualism is a strong emergentist position according

to which phenomenal properties are those physically unacceptable properties, with DUx-p being the

position that phenomenal properties are, e.g., the categorical bases of those (otherwise dispositional)

physically unacceptable properties. It is also possible for the strong emergentist to insist that phenomenal

properties are epiphenomena of those physically unacceptable properties, or even that the phenomenal

correlates with them in a parallelist fashion.

If strong emergentism is true, then there is level of complexity at which systems of physically accept-

able properties no longer evolve exclusively according to the physically acceptable properties instantiated

by the parts of the system and the laws that govern their combination. At that level of complexity, the

system instantiates a novel higher-level property, one that is something over and above the physical;

call this an emergent property.2 Associated with the emergent property is a brute higher-level law of

nature; call this an emergent law. Specifically, the emergent property causes changes to the distribu-

tion of emergent properties and physically acceptable properties according to a brute causal law; call

this an emergent causal law. Since the physically acceptable properties instantiated by (the parts of) a

system that instantiates an emergent property evolve in part according to the emergent causal law, the

system demonstrates lawful downward causation. The existence of a physically unacceptable property

is sufficient for strong emergentism to be an anti-physicalist position. The existence of lawful down-

ward causation distinguishes strong emergentism from other anti-physicalist positions, and gives it an

empirical character.

1For the remainder of this document, I will use the unmodified ‘physical’ to mean a-physical, etc.2Higher-level properties are often called emergent when they arise from a complex combination of lower-level properties.

But this is consistent with a number of very distinct relationships between higher-level and lower-level properties. Theterm ‘emergence’ has been used to describe relationships that are variously reductive (e.g. Boettiger and Oster, 2009), non-reductive but physicalist (Van Gulick, 2001), interactionist anti-physicalist (McLaughlin, 1992), and even epiphenomenalanti-physicalist (McLaughlin, 1997; Chalmers, 1996). See also Wilson (forthcoming) for a survey of accounts of emergentismof the physicalist kind (weak emergentism) and of the anti-physicalist kind (strong emergentism). I will reserve the term‘emergent’ for lawfully interacting, physically unacceptable properties, and the term ‘strong emergentism’ for positionsaccording to which such properties are instantiated.

79

Chapter 4. Characterizing Strong Emergentism 80

In this chapter and the next, I will attempt to argue against the plausibility of strong emergentism

being true of the actual world. To see why this might be difficult, consider that all strong emergentist

positions have an empirical component. The lawful downward causation of systems instantiating emer-

gent properties implies that the physical world is not causally closed; there are systems of physically

acceptable properties that have physically unacceptable causes, and the evolution of such systems is

lawful. Strong emergentism is therefore an a posteriori hypothesis: either physical events have physi-

cally unacceptable causes, or they do not; and if they do, the existence of a brute emergent causal law

must be accepted. This makes it resistant to being excluded by purely a priori philosophical analysis.

Physicalists must therefore use empirical arguments to show that strong emergentism is unlikely to be

true of the actual world.

It is almost impossible to rule out, on empirical grounds, the existence of any emergent property or

law whatsoever. There is no plausible empirical refutation that, for example, when the universe as a

whole reaches a certain level of complexity, a single event (or a single series of events) takes place that is

a consequence of an emergent causal law. But this remote possibility not a concern for the physicalist.

Strong emergentism is advanced in the Philosophy of Mind to account for phenomenal properties that (at

least) correlate with properties both lawfully causally efficacious and impossible to explain in physically

acceptable terms. Thus, evidence that would support strong emergentism (and what physicalists should

attempt to rule out) are inexplicable physical operations of the brain that correlate in some way with


Even still, the physicalist has a difficult task. The only way to rule out strong emergentism with

respect to psychological properties would be to give a complete physicalist account of the physical

operation of the brain. There is currently no such account. So the best to which a physicalist can aspire

(short of engaging in detailed neuroscience) is to mount empirical arguments against the plausibility of

strong emergentism given current empirical evidence. The philosophically interesting part of this work

is to examine those characteristics of strong emergentism that make it vulnerable to existing empirical

evidence, and then to exploit them.

This chapter is broadly concerned with characterizing strong emergentism. Accounts of strong emer-

gentism focus on the relation between the emergent property and the physically acceptable properties

from which it emerges. Call this relation strong emergence. The aim of this chapter is twofold. First, I

will show that accounts of strong emergence require a principled distinction between a physical law and

an emergent causal law. Second, I propose a way of making this distinction: emergent causal laws govern

systems on the complex side of a finite discontinuity in their behaviour as a function of complexity.

In section 4.1, I detail a problem with supervenience-based accounts of strong emergence that mirrors

the problem with supervenience-based accounts of being nothing over and above, such as the account

presented in section 1.1.4 (and withdrawn at the beginning of chapter 3). Such accounts are neutral

neither to the conjunction of property-law necessitarianism and holism about laws (Wilson, 2005), nor to

properties having all their causal powers with metaphysical necessity (O’Connor, 1994). In section 4.2,

I show that, given a principled distinction between emergent causal laws and physical laws, modal

reasoning can be constructed that does not violate the above neutrality condition. I use this result

to propose a neutral supervenience-based account of strong emergence, one that transparently depends

upon the distinction between emergent causal laws and physical laws. In sections 4.3 to 4.5, I examine a

number of accounts of strong emergentism. I show that they also depend upon a principled distinction

between emergent causal laws and physical laws, though this dependence manifests itself in different ways.


Finally, in section 4.6, I discuss the difficulties associated with distinguishing emergent causal laws and

physical laws. I propose that such laws are distinguished not by their content, and not (entirely) by the

characteristics of the complex configuration at which they arise, but rather by a finite discontinuity in

the way a system evolves as the system becomes more complex.

4.1 Problems with Supervenience-Based Accounts

Recall my provisional, supervenience-based account of being nothing over and above. Adapted to the

relevant case of being nothing over and above the physical:

Being nothing over and above the physical (naıve): Property P is nothing over and above physical prop-

erties iff P strongly supervenes on physical properties with physical necessity.

All and only those properties that are nothing over and above the physical are physically acceptable prop-

erties. With this account of being nothing over and above the physical, physicalism becomes similarly

supervenience-based:

Physicalism: Physicalism is true iff all broadly scientific properties are nothing over and above physical

properties.

This characterization of physicalism has the virtue of conforming to the intuitive understanding of

physicalism as requiring of properties that they satisfy the predicates of the (deductively closed) final

physics of the actual world.

A characterization of strong emergentism can be given in a parallel fashion. If strong emergentism is

true, then there are properties that are something over and above physical properties. These emergent

properties lawfully arise from complex configurations of physically acceptable properties, and go on to

interact with them lawfully.

Strong emergence from the physical (naıve): Property P strongly emerges from physical properties iff

P strongly supervenes on physical properties with nomological necessity, but not with physical

necessity.3

This account of strong emergence from the physical is also supervenience-based. It can be used in a

characterization of strong emergentism, which will also be supervenience-based:

Strong emergentism: Strong emergentism is true only if some broadly scientific properties strongly

emerge from physical properties.

This characterization of strong emergentism conforms to the intuitive understanding of strong emer-

gentism as implying that some properties fail to satisfy the final physics of the actual world, but still

interact with the physical world in a lawful way. There are a set of laws, the physical laws, according to

which physically acceptable properties normally evolve. These are the properties that supervene with

physical necessity on physical properties. Then, when systems of physically acceptable properties reach

a certain level of complexity, new, emergent properties are instantiated according to brute emergent

laws of nature (which are not themselves physical laws). These are the properties that supervene with

nomological (but not physical) necessity on physical properties.

3Compare Van Cleve (1990), discussed in section 4.3.1.


Further conditions would have to be added to the characterization in order to capture all the features

of strong emergentism. Two such conditions stand out. First, the emergent properties have to down-

wardly cause changes to the distribution of physically acceptable properties in a lawful manner. This

differentiates strong emergentism from non-interactionist and anomalously interactionist positions that

satisfy the above characterization. Second, the emergent properties have to be properties of (complex)

wholes, nomologically (but not physically) supervening on the physically acceptable properties of the

parts of that whole. This captures our intuitive understanding of strong emergentism as having complex

(rather than simple) systems instantiate emergent properties in virtue of their (as opposed to some other

system’s) complexity. Since these details obscure what I take to the be essential contrast between strong

emergentism and physicalism, I will suppress them, and keep the characterization a necessary condition.

An obvious precondition to these characterizations is that it be possible for nomological necessity

and physical necessity to come apart. The naıve account of being nothing over and above the physical

characterizes it in terms of physical necessity, and not in terms of nomological necessity. This is because,

if strong emergentism is true, then the emergent properties supervene with nomological necessity on

physical properties. If being nothing over and above the physical were to be characterized in terms of

nomological necessity, these physically unacceptable properties would count as physically acceptable.

Their presence would not violate the characterization of physicalism, even though physicalism is false.4

So there needs to be a way for nomological necessity and physical necessity to come apart. The naıve

account of strong emergence is even more obviously dependent on this distinction, since without it the

characterization becomes impossible to satisfy.

Given this critical dependence, these accounts, and supervenience-based accounts of physicalism

and strong emergentism in general, are vulnerable to arguments that collapse physical necessity to

nomological necessity. Unfortunately, such arguments have been forwarded.

One is presented by Wilson (2005). The conjunction of property-law necessitarianism (henceforth

necessitarianism) and holism about laws implies that there is no distinction between nomological neces-

sity and physical necessity. According to necessitarianism, properties are individuated in part by their

relationship to the particular laws that govern them. So any world where those (or very similar) laws

fail to hold is a world that fails to instantiate those properties.5 According to holism about laws, the

laws of nature come together as a package: any world in which one (actual) law holds is a world where

all the (actual) laws hold.

Together, necessitarianism and holism about laws work against supervenience-based accounts of both

being nothing over and above the physical and strong emergence from the physical in a straightforward

way. It follows from holism about laws that any world where one actual law fails to hold is a world where

all the actual laws fail to hold. Therefore, by necessitarianism about laws, no nomologically impossible

world instantiates any property of the actual world.6

For example, suppose we live in a Newtonian world with only two forces: the (Newtonian) gravita-

tional force and the Coulomb force. According to necessitarianism, mass is instantiated only in worlds

where Newton’s Law of Gravitation (or something very much like it) holds, and charge is instantiated

only in worlds where Coulomb’s Law (or something very much like it) holds. According to holism about

laws, Newton’s Law of Gravitation holds only in those worlds where Coulomb’s Law holds, and vice

4The characterization would fail what Wilson (2005) calls the criterion of appropriate contrast.5Necessitarianism allows for properties to be instantiated in worlds where the laws are very similar to the laws that

govern them. What counts as similarity here is a small difference in parameters like proportionality constants. These kindsof differences are not relevant to my argument.

6Unless minor variations of all the laws hold in that nomologically impossible world. See footnote 5.


versa. Therefore, any world where either of these laws fails to hold is a world where both these laws fail

to hold. Therefore, there is no mass and no charge in any nomologically impossible world.7

The lack of actual properties in nomologically impossible worlds implies that nomological superve-

nience, physical supervenience, and indeed metaphysical supervenience converge. Consider one set of

properties (the supervening properties) that nomologically supervene on another set of properties (the

subvening properties) with nomological necessity. Any nomologically impossible world is a world instan-

tiating neither the supervening nor the subvening properties. Therefore, the former also supervene on

the latter with both physical necessity and metaphysical necessity.

Another argument against nomological necessity and physical necessity coming apart is presented by

O’Connor (1994). If property types are individuated at least in part by (all) their causal powers, then it

is metaphysically impossible for one instance of a property type to lack even a single one of the causal

powers of another instance of the type. Since those different laws would imply different causal powers,

neither emergent properties nor physical properties are instantiated in any nomologically impossible

world.8 By the same argument as above, any nomologically necessary supervenience relation among

such properties is therefore both physically and metaphysical necessary.

As I argued in chapter 2, we have reasons to deny that a property instance of some type has all

the causal powers of every instance of that type. We may also have reasons to deny necessitarianism,

or (perhaps more fruitfully) holism about laws.9 Nevertheless, good characterizations of physicalism

and strong emergentism ought to be neutral to whether properties have their causal powers necessarily

(O’Connor, 1994, p.97), and to whether the conjunction of necessitarianism and holism about laws is

true (Wilson, 2005, p.447). Call this the neutrality condition.

There is, therefore, a serious problem with supervenience-based accounts of strong emergence that

assume that nomological necessity and physical or metaphysical necessity can come apart. However, as

I argue below, this problem can be resolved if a principled distinction can be made between a physical

law and an emergent causal law. What is more, non-supervenience-based accounts of strong emergence

suffer from related problems that can also be resolved if a principled distinction can be made between a

physical law and an emergent causal law. For this reason, I will present an account of strong emergence

that explicitly accepts such a distinction as a starting point.

4.2 Characterizing Strong Emergence

I propose a supervenience-based account of strong emergence, one that embraces the need for a distinction

between an emergent causal law and a physical law, and uses this distinction to ground a difference

between two kinds of necessity. But the difference only makes sense given two features of emergent

properties under strong emergentism: (1) the emergent property lawfully, downwardly causes changes

to the distribution of physically acceptable properties; and (2) the emergent property is a property

7Again, unless minor variations of Newton’s Law of Gravitation and Coulomb’s Law hold in that nomologically impos-sible Newtonian world. See footnote 5.

8We can suppose, for the purposes of this discussion, that all the laws are causal laws.9Wilson’s (2005) argument for holism about laws is based on the observation that physicists expect the fundamental

forces to converge at very high energies. This expectation is based (in part) on an extrapolation of the strength of couplingconstants as a function of energy, at least for the strong force and the electroweak force. But these coupling constantsdescribe the strength of interactions between elementary particles. The expectation is, therefore, that the physical forcesconverge. If there were fundamental forces at high levels of complexity (well above the level of elementary particles), itmay be that all the forces are unified, but the expectations of physicists based on the coupling constants of elementaryparticles wouldn’t support it.


of a relatively complex whole, upon the properties of whose parts it supervenes. The former entails

that there is an emergent causal law. The latter entails that there is a domain, the domain of relatively

simple systems (more or less causally isolated from more complex systems), to which the emergent causal

law makes no difference. Even though these features are suppressed in the characterization of strong

emergentism, they both play a critical role in allowing the account of strong emergence (and hence the

characterization of strong emergentism) to satisfy the neutrality condition.

4.2.1 Biased and Unbiased Modal Reasoning

Recall the naıve account of strong emergence, according to which emergent properties strongly supervene

on physical properties with nomological, but not physical, necessity. This can be understood as follows.

A world is nomologically possible iff all the actual natural laws hold in that world. A world is physically

possible iff all the actual physical laws hold in that world. Since the physical laws are a subset of

all the laws, the nomologically possible worlds are a subset of the physically possible worlds. The

controversial move in light of the neutrality condition is as follows. If there are emergent laws (or

indeed any fundamental laws that are not physical laws), then nomologically possible worlds are a

proper subset of the physically possible worlds: that subset of worlds where the emergent law also

holds.10 The emergent properties are thus instantiated in all nomologically possible worlds, but not in

all physically possible worlds.11 Therefore, even if emergent properties supervene on physical properties

with nomological necessity, they do not supervene with physical necessity.

This characterization falls apart if the set of nomologically possible worlds is identical to the set of

physically possible worlds, which is the case if the conjunction of necessitarianism and holism about

laws is true, or if properties have all their causal powers necessarily. It therefore violates the neutrality

condition. The naıve account is based on what can be called biased modal reasoning.

Still, a distinction between physical laws and emergent laws (and therefore some sense of a difference

between physical necessity and nomological necessity) does appear to be a promising way to give an

account of strong emergence. The British Emergentists appealed to a similar distinction, in their case

between homopathic laws and heteropathic laws (McLaughlin, 1992, p.59). What is needed is a way of

talking about the distinction without running afoul of the neutrality condition, and using the distinction

to ground a difference between nomological necessity and physical necessity in an unbiased way.

There are good reasons to expect that an unbiased way of talking about the distinction between emer-

gent laws and physical laws can be constructed. Distinguishing between emergent laws and physical laws

is a species of counterlegal reasoning: reasoning with counterfactual conditionals whose antecedents are

nomologically impossible. Counterlegal reasoning is widespread and accepted in both science and phi-

losophy.12 In fact, the central motivation behind strong emergentism can be interpreted as counterlegal

reasoning: had there been no emergent law, the physical system that instantiates the emergent prop-

erty would have evolved in a different manner. This sort of reasoning is no different, in principle, than

reasoning about whether orbits would be stable had masses obeyed an inverse-cubed law of gravitation.

A metaphysics that connects laws to properties and laws to other laws with metaphysical necessity, or

10In order for this to be the case, the existence of an emergent law must violate the completeness of the physical laws,but must not violate the laws themselves.

11Alternatively, the lawful relationship between emergent properties and physical properties holds in all nomologicallypossible worlds, but not in all physically possible worlds.

12It is significant that most counterfactual reasoning about the actual world proceeds from an epistemic context ofignorance of the true laws of nature.


one that requires properties to have all their causal powers metaphysically necessarily, must be able to

handle such reasoning. However counterlegal reasoning is handled in the general case, that solution can

be applied to distinguishing emergent laws and physical laws.

How such reasoning is to be accommodated needs to be worked out.13 For example, it may be

metaphysically impossible for a property to be instantiated in a world with very different laws, or

to have different causal powers. Nevertheless, there may still be metaphysically possible worlds with

properties or laws that correspond to those of the actual world, without being identical to them.14

Unbiased counterlegal reasoning could then be performed by considering such worlds.

4.2.2 Counterlegal Reasoning and Strong Emergence

It is well beyond the scope of this document to work out the details of a general account of counter-

legal reasoning in light of the neutrality condition. For my purposes, fortunately, there is no need to

accommodate such counterlegal reasoning in the general case. An account of strong emergence just

needs to ground a difference between nomological necessity and physical necessity in case there are

emergent causal laws that govern only relatively complex systems. At least in the special case of strong

emergentism, counterlegal modal reasoning that captures this distinction can be constructed.

What follows below is a way of adapting the biased modal reasoning above to accommodate the

neutrality condition while preserving some kind of distinction between nomological necessity and physical

necessity. Broadly speaking, I use the fact that emergent properties are never instantiated at the

lowest levels of complexity to isolate the physical properties, and functionally characterize them. The

union of, on the one hand, worlds instantiating properties that satisfy those (and only those) functional

characteristics, and, on the other hand, nomologically possible worlds, can play the role of physically

possible worlds in unbiased modal reasoning.

Suppose strong emergentism is true. For any given emergent causal law, there is a physical configura-

tion to whose evolution the law contributes not at all. Specifically, the emergent law does not contribute

to the evolution of any system whose configuration is at a lower level of complexity than a system that

instantiates the emergent property, given that the system is (more or less) causally isolated from any

emergent system. In fact, there must be physical systems, like pairwise particle interactions in deep

space, or any system of the universe prior to the instantiation of the first emergent property, that are at

a low enough level of complexity that no emergent property is instantiated, and so to whose evolution

no emergent causal law contributes. Call the set of all such systems the non-emergent domain. A theory

of the causal interactions of the entire non-emergent domain can be constructed, and this theory can be

deductively extended outside the non-emergent domain.15 We can then Ramsify the theory to get an

extended functional characterization of each property.16 In the actual world, functional properties de-

scribed in that extended theory will be satisfied by physically acceptable properties in the non-emergent

domain. But, plausibly, they will also be satisfied by (possibly distinct) properties in metaphysically

possible worlds that are nomologically impossible. Call the worlds whose properties satisfy only those

13For a discussion on counterlegal reasoning under the supposition that properties have causal powers necessarily, seeHandfield (2004).

14For example, perhaps mass, necessarily connected to the laws of the actual world, corresponds to schmass in worldswith an inverse-cubed law of gravitation. See Fine (2002).

15In other words, we can construct a theory that correctly describes the non-emergent domain but incorrectly describessystems outside this domain. Indeed, that theory would be the physics theory of the actual world. To be sure, this theorycannot be obtained without the general methodological considerations that go toward the production of any theory ofphysics. This does not take away from my general point.

16Ideally, in topic-neutral terms.


functional properties described by that extended theory superficially-physically possible worlds. The laws

of those worlds are the superficially-physical laws, and they govern the superficially-physical properties.

Actual properties can be connected to superficially-physical properties based on the following coun-

terlegal conditional:

Correspondence to superficially-physical: Actual property P corresponds to superficially-physical prop-

erty Q iff: (a) for every system s1 instantiating P , had there been no emergent causal laws, s1

would have instantiated Q, and (b) for every system s2 not instantiating P , had there been no

emergent causal laws, s2 would not have instantiated Q.

Emergent properties, by construction, correspond to no superficially-physical properties. If physicalism is

true, then (also by construction) the physical properties correspond to themselves, i.e., the superficially-

physical properties just are physical properties, the superficially-physical laws just are physical laws, and

the superficially-physically possible worlds just are physically possible worlds. Even if strong emergentism

is true, the superficially-physical properties might just be physical properties, the superficially-physical

laws might just be physical laws, and the superficially-physically possible worlds might just be the

physically possible worlds. But under strong emergentism, if either the conjunction of necessitarianism

and holism about laws is true, or properties have their casual powers necessarily, the physical and the

superficially-physical come apart.

Suppose the conjunction of necessitarianism and holism about laws is true. Under strong emergen-

tism, none of the actual laws are laws that hold in the superficially-physically possible worlds, so no

property instantiated in the actual world will be a superficially-physical property. Nevertheless, there

will still be a one-to-one correspondence between the superficially-physical properties of nomologically

impossible worlds and the physical properties of the actual world, and the superficially-physical laws will

resemble those actual laws that govern systems in the non-emergent domain.

Suppose instead that properties have all their causal powers necessarily. Again, under strong emer-

gentism, no property instantiated in the actual world will be a superficially-physical property. But again,

there will be a one-to-one correspondence between the superficially-physical properties of nomologically

impossible worlds and the physical properties of the actual world. Indeed, if causal powers are given in

topic-neutral terms, the superficially-physical properties will be those properties with all the (type-level)

causal powers that the physical properties have in virtue of the physical laws, but not those that they

have in virtue of the emergent causal laws. There will also be an analogous one-to-one correspondence

between the superficially-physical laws and the physical laws.

Call the union of superficially-physically possible worlds and nomologically possible worlds the quasi-

physically possible worlds. If physicalism is true, then quasi-physical necessity just is physical necessity,

which just is nomological necessity. On the other hand, under strong emergentism, the nomologically

possible worlds are a proper subset of the quasi-physically possible ones, so nomological necessity and

quasi-physical necessity come apart.

In the context of the debate between physicalists and strong emergentists, quasi-physical necessity

can play the role of physical necessity, but in a way that does not violate the neutrality condition. It

is made possible by supposing that the alternative to physicalism is strong emergentism, which entails

an emergent causal law governing only relatively complex systems. Without a principled distinction

between physical laws and emergent causal laws, the non-emergent domain, and thus quasi-physical

necessity, would have been undefined.


4.2.3 Strong Emergence

The role of physicality in the naıve account of strong emergence can be played by quasi-physicality in

a less naıve account. We can quantify over the quasi-physically possible worlds, as long as sentences

containing predicates that refer to actual properties are taken to refer to their corresponding properties

in the superficially-physically possible worlds. The naıve account can be modified in light of this,

while retaining the force of the characterization’s distinction between nomological necessity and physical

necessity.

Strong emergence from the physical: Property P strongly emerges from physical properties iff P strongly

supervenes on physical properties with nomological necessity, but not with quasi-physical necessity.

This account does not violate the neutrality condition. If the conjunction of necessitarianism and holism

about laws is true, then actual properties are not instantiated in any nomologically impossible world.

Likewise, if properties have their causal powers necessarily, no properties instantiated in the actual world

are instantiated in any nomologically impossible world. Nevertheless, properties corresponding to actual

physical properties are instantiated in the superficially-physically possible worlds, while no property

corresponding to an actual emergent property is instantiated there. There is still a difference between

nomological necessity and quasi-physical necessity. This allows for the two distinct kinds of necessity

that are needed to characterize strong emergentism.

If either of necessitarianism or holism about laws is not true, or if properties do not have their causal

powers necessarily, then, independent of the existence of emergent properties, superficially-physical prop-

erties just are physical properties, and this account of strong emergence reduces to the naıve one.

This account of strong emergence, and thus the characterization of strong emergentism, depends

upon there being a principled distinction between emergent causal laws and physical laws. By appealing

to the difference between nomological necessity and quasi-physical necessity, the account makes this

dependence transparent. This is an advantage to the account since, as I argue below, other accounts of

strong emergence also depend upon a distinction between emergent causal laws and physical laws, but in

a less transparent way. Below, I discuss these accounts, beginning with other (synchronic) supervenience-

based accounts, and moving on to diachronic accounts and powers-based accounts. At the end of this

chapter, I will discuss whether a principled distinction between emergent causal laws and physical laws

can be made.

4.3 Supervenience-Based Accounts

In this section, I will describe four other supervenience-based accounts of strong emergence. I will show

that, as with the account presented above, they depend upon a distinction between emergent causal

laws and physical laws. For three of them, this dependence manifests as a need to distinguish two

kinds of necessity, which violates the neutrality condition unless there is a principled distinction between

emergent causal laws and physical laws. For the other, the dependence manifests as a need to distinguish

structural and non-structural combinations of physical properties.

4.3.1 Van Cleve

One such attempt is by Van Cleve (1990), where he uses strong nomological supervenience to define an

emergent property:


Strong Emergence (Van Cleve): Property P of object w is strongly emergent iff P nomologically super-

venes, but does not metaphysical supervene,17 on the properties of the parts of w taken separately

or in other combinations.

On Van Cleve’s account, emergent properties arise from configurations of physical properties in virtue

of the laws of nature, and hence they supervene on physical properties with nomological necessity. But

their arising is contrary to how physical properties (in some sense naturally) combine, so they do not

supervene with metaphysical necessity. Van Cleve’s account, therefore, clearly demonstrates a need for

two distinct kinds of necessity, and therefore violates the neutrality condition as the naıve account does.

There is another flaw in the account, and one that highlights the difference between it and the naıve

account. Van Cleve’s account is too liberal. As O’Connor (1994, p.96) points out, some higher-level

physically acceptable properties, like a knife’s ability to cut bread, nomologically supervene on lower-level

physical properties, but don’t metaphysically supervene on them.18 Yet a knife’s ability to cut bread

should not count as an emergent property. More distressingly, McLaughlin (1997) observes that the mass

of a whole would count as emergent vis-a-vis the mass of its parts. This is because it is metaphysically

contingent that masses combine according to scalar addition.19

The naıve account of strong emergence avoids this problem by requiring of emergent properties that

they fail supervene with physical (in addition, presumably, to metaphysical) necessity. Suppose it is

meaningful to talk about nomological necessity as distinct from physical necessity. Presumably, a knife’s

ability to cut bread supervenes with both nomological and physical necessity on lower-level physical

properties. Similarly, supposing we lived in a Newtonian world, masses might not combine by scalar

addition by metaphysical necessity, but they certainly would do so combine by physical necessity.

4.3.2 O’Connor

O’Connor (1994) offers a revision to Van Cleve’s account that is also meant to correct these flaws. It

has three parts:

Strong Emergence (O’Connor): Property P of object w is strongly emergent iff

1. P strongly supervenes on the properties of the parts of w;

2. P is non-structural;

3. P demonstrates novel causal influence.

In the first condition, the supervenience relation is meant to have the force of causal necessity, which he

leaves ambiguous between nomological and metaphysical necessity, depending on the reader’s favoured

account of causation. Since his account doesn’t require two distinct kinds of necessity, it prima facie

satisfies the neutrality condition.

As for the second condition, a property P of w is non-structural iff (a) P can only be instantiated by

complex objects, (b) P is not had by any of w’s parts, and (c) P is distinct from any structural property

17Van Cleve appeals to logical supervenience here instead of metaphysical supervenience. No harm is done to the accountby treating it as appealing to metaphysical supervenience instead.

18That the molecules of the knife do not disintegrate when they approach bread is plausibly a matter of natural lawrather than a metaphysical necessity, as long as we maintain (as we must by Van Cleve’s account) a distinction betweennomological and metaphysical necessity.

19Masses combine by scalar addition in a Newtonian world. Indeed, that masses combine in a more complicated way in arelativistic world illustrates the point. Again, this supposes that the way masses combine is not a metaphysically essentialfeature of being mass.


of w. This pushes the question back a little; all we need now is to determine what a structural property

is. O’Connor provides the following definition:20

A property, S, is structural if and only if proper parts of particulars having S have someproperty or properties not identical with S, and this state of affairs is, in part at least,constitutive of the state of affairs of the particular’s having S. (O’Connor, 1994, p.93)

This pushes the question back again, since we need some idea of what it means for properties of parts

to constitute properties of wholes. Unfortunately, O’Connor only provides a “rudimentary” (O’Connor,

1994, p.93) account of how the constitution relation works. Beyond an intuitive appeal to the property

of the whole being nothing over and above the properties of the parts, our only hint is that the causal

powers of the whole are just the “summation” (O’Connor, 1994, p.93) of the causal powers of the

parts. How causal powers combine or constitute one another is left unsaid; that some combinations of

powers are non-emergent and some are not seems to be a restatement of the problem that led to the

non-structurality condition in the first place.

The non-structurality condition can, however, be analyzed in terms of emergent laws and physical

laws. Let both structural and non-structural properties (of wholes) be such that they can only be

instantiated by complex objects, and they are not instantiated by the parts of that object. Structural

properties are those had by the objects in virtue of the physical laws. Non-structural properties are

those had by the objects in virtue of the emergent laws.

Perhaps an appeal to the third condition on emergent properties can help distinguish structural

properties and non-structural properties in a way that doesn’t appeal to a distinction between emergent

laws and physical laws. The third condition requires emergent properties to have novel causal powers, and

these causal powers act on objects in a downward fashion. If a whole instantiates an emergent property,

it has causal powers that are lacked by any of the parts. In contrast to the emergent property’s new

causal powers, the structural properties of the whole act “via the activity of the micro-properties that

constitute it” (O’Connor, 1994, p.94). Thus, perhaps no combination of the powers of the parts can

determine the emergent property’s powers.

Sadly, this does not get us very far. If the emergent properties are downwardly causal, then, plausibly,

the physical properties of the parts of non-structural wholes sometimes act via the activity of the

emergent property that arises from them. There are, therefore, laws that govern the evolution of the

parts of structural wholes in virtue of which those wholes act, and laws that govern the evolution of non-

structural wholes in virtue of which the parts of those wholes act. Without an independent account of

the structural/non-structural distinction, these laws appear to be the physical laws and emergent causal

laws, respectively. Rather than taking the place of a distinction between emergent laws and physical

laws, O’Connor’s third condition reinforces the need for such a distinction, and adds a further need for

the emergent laws to be causal.

4.3.3 McLaughlin

Recall that one problem with Van Cleve’s account is that it is too liberal: mass, for example, does

not metaphysically necessarily combine by scalar addition, so total mass would count as emergent. In

response to this, McLaughlin (1997) offers a revision:

20This is a modified version of a David Armstrong’s definition. See O’Connor (1994, fn.3).


Strong Emergence (McLaughlin): Property P of object w is strongly emergent iff (1) P nomologically

supervenes, but does not metaphysically supervene,21 on the properties of the parts of w taken

separately or in other combinations, and (2) some of the supervenience principles linking properties

of the parts of w with w’s having P are fundamental laws.

The first condition is the same as Van Cleve’s, and so uses ‘supervenient’ in the strong sense. Given

that the condition requires a distinction between nomological necessity and metaphysical necessity,

McLaughlin’s account, like Van Cleve’s account, violates the neutrality condition.

The second condition is meant to distinguish those properties that arise from a natural combination

of the parts (like mass) and those that strongly emerge.22 The idea here is that the principles of

combination for non-emergent properties are derivative of more fundamental laws, while the principles

of combination for emergent properties are brute. For instance, the supervenience principle that connects

the total mass of an object to the masses of its parts is not fundamental, being an instance of “the general

compositional [law] of. . . the additivity of mass” (McLaughlin, 1997, p.16). Those that connect emergent

properties of wholes with the properties of their parts are fundamental.

If totaling masses counts as derivative of a more fundamental general law of the additivity of mass, the

supervenience principles that connect emergent properties to physical properties have to be very specific

indeed. They must connect maximally specific physical properties to distinct emergent properties in an

unpatterned way. This is because patterns between physical properties and emergent properties would

indicate a more general, fundamental law of combination, of which the supervenience principle would be

an instance. But, according to McLaughlin, if properties have any (lawful) causal role at all, they are

functionally analyzable,23 and such general laws of combination can be constructed. Therefore, emergent

properties, which cannot be connected to physical properties by derivative supervenience principles, are

epiphenomenal.

This account of strong emergence, therefore, cannot be used to characterize strong emergentism,

since there is no lawful, downward causation. Moreover, since, by McLaughlin’s account, there can be

no emergent causal law, my argument for unbiased modal reasoning cannot go through. The violation

of the neutrality condition is thus without an obvious remedy.

One way around this problem is to revisit the idea that all principles of combination that follow from

a functional analysis of properties are derivative in the same way. It is true that, if emergent properties

are lawfully causal, they must be functionally analyzable in some sense. But they are not functionally

analyzable the way masses are. Masses satisfy their causal roles in virtue of the physical laws, and they

combine according to laws of combination that are themselves physical laws. Emergent properties also

satisfy causal roles, but not in virtue of the physical laws. They do so in virtue of emergent causal

laws, and are instantiated according to emergent laws. A principled distinction between emergent causal

laws and physical laws therefore allows McLaughlin’s account of strong emergence to characterize strong

emergentism. The same distinction also allows the account to satisfy the neutrality condition.

21Again, McLaughlin appeals to logical supervenience, which, for convenience, I have replaced with metaphysical super-venience. See footnote 17.

22In an earlier, provisional proposal in the same paper, McLaughlin merely tacked on the following rider to Van Cleve’scondition: “together with compositional principles that apply to the parts in other combinations” (McLaughlin, 1997,p.15). It appears that the second condition is meant to replace this rider, as a way of specifying which compositionalprinciples count as emergent and which don’t.

23From McLaughlin (1997, p.11): “To functionally analyze a disposition or capacity is to analyze it as a second-orderstate of being in a state that plays a certain causal role.”


4.3.4 Seager

A supervenience-based account of emergent properties that seems to incorporate similar intuitions can

be found in Seager (2012).

Strong Emergence (Seager): A domain of properties, P , is strongly emergent on the domain of the

physical properties, Φ, iff (1) P weakly physically supervenes on Φ;24 and (2) physics is not a total

theory.

The use of weak physical supervenience in the first condition is a recognition of the need to distinguish

emergent laws and physical laws: as far as the physical laws are concerned, physical properties and

emergent properties can come apart.

The second condition amounts to the condition that the physical world fails to be causally closed.25

This implies downward causation of emergent properties on the distribution of physical properties.

This account has many advantages. The appeal to only physical possibility naturally distinguishes

non-emergent properties of wholes from emergent ones: it is physically necessary that masses combine

the way they do, but it is not physically necessary that the properties of the parts of an emergent whole

do. It also builds in downward causation quite naturally, via the violation of physical closure. Finally,

it explains why emergent properties would be hard to find. There is an illusion that physics is a total

theory, since it is complete and has resolution, and, since every physical event at a given time weakly

(but not strongly) supervenes on events at prior times, there appears to be (though in fact there is not)

closure.

Nevertheless, this account violates the neutrality condition. If emergent properties are lawfully

causal, then they will strongly supervene on physical properties with nomological necessity. If either the

conjunction of necessitarianism and holism about laws is true, or properties have all their causal powers

metaphysically necessarily, then nomological supervenience, physical supervenience, and metaphysical

supervenience converge. This means that emergent properties both weakly and strongly supervene on

physical properties with physical necessity, just like every physically acceptable property. Again, a

remedy to this situation is to accept a principled distinction between emergent causal laws and physical

laws, which enables unbiased modal reasoning to be used.

4.4 Real Operation Accounts

Another way of characterizing the relationship between a strongly emergent property and the physical

properties from which it emerges is to abandon (synchronic) supervenience in favour of a real, physical

process. According to O’Connor (2000b,a) and O’Connor and Wong (2005), physical properties cause

emergent properties. According to Humphreys (1997a), physical properties fuse into emergent prop-

erties. In both cases, I argue that, to whatever extent they are appropriate for characterizing strong

24Recall that weak supervenience holds when, within a world, all things that instantiate the same subvening propertyinstantiate the same supervening property, but between worlds different combinations of subvening-supervening pairs mightobtain.

25A theory is total iff it is complete (all entities are in the theory’s domain), it is closed (there are no outside laws orforces on the domain of the theory), and it has resolution (every process or object is resolvable into entities of the theory’sdomain). If strong emergentism is true, then physics is complete (there are no non-physical substances) and has resolution(the properties of the physical parts can be resolved even in wholes that instantiate emergent properties). So a violation ofthis condition is, specifically, the violation of closure: the physical laws are insufficient to determine the causal evolutionof all physical systems.


emergentism, the accounts still depend upon a principled distinction between emergent causal laws and

physical laws.

4.4.1 O’Connor and Wong

O’Connor (2000b,a) and O’Connor and Wong (2005) present a causal account of strong emergence.

There are two parts to the account. The first part is a non-structurality condition: emergent prop-

erties are ontologically basic, non-structural properties of wholes. The second part is causal: When

physical properties are configured in certain complex ways, they cause emergent properties to come into

being, and these properties are sustained by the underlying physical ones. The physical and emergent

properties together go on to cause further physical and emergent properties to be instantiated.

The non-structurality part of the account is apparently no different than the non-structurality con-

dition of the O’Connor (1994) account, described in section 4.3.2. A structural property is one to which

there is nothing more than “being composed by parts having certain other properties and bearing cer-

tain relations to one another” (O’Connor and Wong, 2005). In contrast, non-structural properties are

ontologically basic properties of wholes. When combinations of parts count as nothing more than the

properties of the parts, and when their combination reveals something more, is left unsaid. As above,

this issue can be resolved with a principled distinction between emergent laws and physical laws.

The causal part of the account may provide another resolution. On this account, an emergent property

of a system does not (synchronically) supervene on the properties of the parts of the system, having been

caused by physical and emergent properties at a prior instant. Systems with physically type-identical

parts can instantiate distinct emergent properties, as long as they have type-distinct causal histories.

Perhaps this can ground a difference between structural and non-structural properties.

Unfortunately, the problem is just relocated elsewhere. The emergent property may not synchroni-

cally supervene on physical properties, but it does diachronically supervene on them. To be sure, the fact

that both physical and emergent properties cause physical and emergent properties guarantees that a

given emergent property type depends on both the physical and the emergent property types at the prior

instant. But since emergent properties are not among the initial conditions of any system,26 the emer-

gent property, just like any physical property, supervenes on the entire physical history of the system.

This is global, diachronic supervenience.

Thus, just in virtue of the causal part of the account, there is no accounting for the difference

between emergent properties and physical properties. The causal history of system could just as well be

encoded in a new, basic physical property of the parts of the system, or in a new basic physical relation

of the parts. There appears to be no way to distinguish the emergent property from a higher-level

physical property that is structural relative to these new physical properties, except by appeal to the

structural/non-structural distinction.

It is worth noting that O’Connor and Wong (2005, p.665ff) argue that the causal part of the ac-

count does indeed imply that emergent properties fail to globally diachronically supervene on physical

properties. The reason they give is that the evolution of systems with physical and emergent properties

might not be deterministic. But this only implies that emergent properties fail to globally diachron-

ically supervene on physical properties in just the same way that physical properties fail to globally

diachronically supervene on physical properties. A simple redefinition, whereby the objective chance of

26In the development of an organism, the brain does not appear spontaneously. For a given system, the physical basesof the emergent properties must evolve from purely physical states.


an emergent property being instantiated globally supervenes on the history of physical properties, would

recover global diachronic supervenience.27

The real reason for a failure of global diachronic supervenience is not indeterminism per se in the

evolution of systems with emergent properties. Rather, it is that such systems may evolve in an anoma-

lous way. Under this interpretation, endorsed by O’Connor (2000b,a), the causal contribution of one

emergent property to the instantiation of another might be anomalous. This allows for free will, which

seems to be the ultimate goal of the account. But in this case, it is not the causal part of the account,

but an additional anomalousness condition, that is responsible. An anomalousness condition of this kind

would make the account incompatible with strong emergentism.

The account therefore remains dependent on a structural/non-structural distinction, which likely

requires a prior distinction between emergent laws and physical laws.

4.4.2 Humphreys

Humphreys (1997a,b) provides an account of strong emergence whereby two (or more) physical property

tokens undergo a real physical operation in which they go out of existence, and in their place another

(fused) property token appears.28 If properties are ordered in a hierarchy of levels, the fused property

type is a higher level than the types of those property tokens that were destroyed. Fused property tokens

can also undergo the same operation. Emergent property tokens are token-identical to fused property

tokens (but might not be type-identical if they are multiply realized).

The account is not a synchronic supervenience account since the physical properties from which

emergent properties emerge are destroyed; there are no subvening physical properties after the fusion. It

is, however, a global diachronic supervenience account, since the physical properties fuse by nomological

necessity.

This account is mostly meant as a proof of principle: there is nothing mysterious or anti-scientific

about emergentism, and since there are perhaps a variety of ways in which one property can emerge from

another, simply demonstrating one to be possible and non-mysterious is beneficial. However, it does

have a few drawbacks. First, the examples for the existence of fused properties (quantum entanglements)

arise as a consequence of the application of paradigmatically physical laws to complex systems. The

quantum mechanical approach is to determine the objective probabilities of the various ways a system

could evolve on the basis of the distribution and characteristics of its components.29 If such a thing is

emergent, it is not one that interestingly contrasts physicalism with strong emergentism. Second, it is

not a likely account for the emergence of mental properties. It is hard to imagine which properties of the

particles of the brain (charge? mass? spin?) are fused (and thus eliminated) in order for the organism

to be in pain, for example.

To adapt this account to one more relevant to strong emergentism, there would have to be a higher-

level operation that fuses physical properties into emergent properties, which operation contrasts the

27This is a lesson from Seager (2012).28If the physical properties go out of existence upon fusion, so that the fused property does not (synchronically) supervene

on any physical property, then physics lacks what Seager (2012) calls resolution. This is an interesting alternative way tosatisfy the second condition of Seager’s account of strong emergence. See section 4.3.4, and especially footnote 25.

29While there is a potential energy term in Hamiltonians, this fact does not undermine the point. First, there is noway for potential energy to be manifest physically except through the distribution and characteristics of the componentsof systems. The potential energy term is the resultant influence of the context of the system on it. Second, the potentialenergy term does not play a role in Humphreys’ argument. The features he alludes to, like quantum entanglements inwhich (under his interpretation) properties can be assigned to the system and only derivatively to its parts, do not dependessentially on a potential energy term.


physical operation that fuses (for example) particle properties into entangled states. This again points

to the need for a principled distinction between emergent laws and physical laws.

4.5 Powers-Based Accounts

Two powers-based accounts of strong emergence are described in this section. Both depend upon a

principled distinction between emergent laws and physical laws.

4.5.1 Wilson

The first powers-based account is proposed by Wilson (2002, forthcoming). Initially, Wilson (2002)

provides a force-relative powers-based account of strong emergence. Adapted to the case of interest:

Strong emergence (Wilson—Force-Relative): Property P of system s strongly emerges from the physical

properties Q of s iff (1) P strongly nomologically supervenes on Q; and (2) P has a causal power

distinct from those causal powers had by Q that are grounded only in the fundamental physical

forces.

Under the plausible assumption that the causal powers in question are associated with the lawful evolu-

tion of systems instantiating both emergent properties and physical properties, this account successfully

captures our intuitive understanding of strong emergentism. The first condition, common to a number

of accounts,30 characterizes the emergent property’s lawful correlation with complex configurations of

physical properties.31 The second condition characterizes the downward causation typical of emergent

properties. What makes it downward causation on this account, as opposed to causation inherited from

its physical parts, is that the emergent property has a causal power that its parts don’t have, at least

not in virtue of the fundamental physical forces.

Given that fundamental forces are associated with fundamental laws, this account transparently

depends on a principled distinction between emergent causal laws and physical laws. Physical properties

have causal powers in virtue of the fundamental physical forces, governed by the fundamental physical

laws. If the emergent property has a causal power not had by the physical properties in virtue of the

fundamental physical forces, then there must be another fundamental force, an emergent force, governed

by another fundamental law, an emergent causal law. Without a prior conception of what makes a

fundamental law an emergent causal law and what makes a fundamental law a physical law, the account

cannot isolate the strongly emergent properties relevant to this discussion.32

Wilson (forthcoming) provides a variation of this account. Again, adapted to the case of interest:

Strong emergence (Wilson): Token property P strongly emerges from token physical property Q iff (1)

P ’s type strongly nomologically supervenes on Q’s type; and (2) P has at least one token power

not identical with any token power of Q.

The first condition is the same as above, and serves the same function.33 The second condition is the New

Powers Condition. As above, it accounts for the downward causation typical of emergent properties,

30Wilson (forthcoming) calls this “minimally nomological supervenience”.31In Wilson (2002), this condition is described as same subject necessitation.32Wilson might appeal to the notion of a physical entity (as in the physics-based, no-fundamental-mentality notion from

Wilson (2006)) to ground such a distinction. But see section 4.6, below.33In Wilson (forthcoming), this condition is described as synchronic dependence.


but departs slightly from Wilson (2002) in a number of ways. The most important departure is that it

is no longer committed to the force-relative character of emergent causation.

It might be thought that, owing to this difference, the account is no longer dependent on a principled

distinction between emergent causal laws and physical laws. After all, a token power of a higher-level

property that is distinct from any token power had by a physical property might ground, rather than

depend upon, a distinction between an emergent causal law and a physical law. Perhaps an emergent

causal law governs just higher-level properties, and not lower-level ones.

But there is reason to think otherwise. If the properties of a whole are bound to the properties of

its parts by nomological necessity, then any token power of a property of a whole nomologically must

correspond to token powers of properties of its parts. If an emergent property has a power, it is at least

nomologically necessary that, if the emergent property is instantiated in certain circumstances, then some

effect is instantiated. In that case, there must be some set of circumstances, perhaps even circumstances

that include the instantiation of the emergent property, for which a given subvening property of each

part is at least nomologically necessary for that effect.34 These powers of the physical properties of the

parts may be token-distinct from the corresponding power of the emergent property of the whole. They

may be “manifested differently, or in different conditions” (Wilson, forthcoming, p.8 of the manuscript).

Indeed, they are, plausibly, inherited by the physical properties from the emergent property, inverting

the causal inheritance typical of causation by physically acceptable properties. Nevertheless, they are

powers had by the physical properties in virtue of an emergent causal law. The need to distinguish

emergent causal laws from physical laws becomes apparent again.

To be sure, the New Powers Condition allows for the physical properties of the parts to lack entirely

any power to cause an effect caused by the emergent property of the whole in virtue of its token-distinct

power.35. But this cannot be the case while simultaneously having the emergent property strongly

nomological supervene on the physical properties. A physically unacceptable property that does not

supervene on physical properties, or that only weakly nomologically supervenes on them, also satisfies

the New Powers Condition. Such a property might have the power to produce an effect that no physical

property has the power to produce. But the interaction of strong nomological supervenience with the

New Powers Condition guarantees that subvening physical properties have powers to produce every effect

of a supervening emergent property, albeit in virtue of an emergent causal law.

Therefore, as a way of characterizing strong emergentism, this account, like the force-relative account,

depends on a principled distinction between an emergent causal law and a physical law. What is more,

if Wilson (forthcoming) is correct that all accounts of strong emergence aim to satisfy this account, all

accounts of strong emergence will depend on this distinction.

4.5.2 Shoemaker

Another powers-based account of strong emergence is provided by Shoemaker (2002).36 On his picture,

there are situations in which physical properties combine in ordinary ways, and other situations in

which they combine in ‘emergence engendering’ ways. When they combine in ordinary ways, he calls the

34See Wilson (2002) for an elaboration of this issue.35See Wilson (forthcoming, p.8 of the manuscript), though she considers it less plausible than the physical properties

having distinct powers to cause the same effect.36Wilson (forthcoming) interprets this account as aiming for an account of non-reductive physicalism (weak emergence),

but compatible with strong emergentism. I found the natural interpretation of the account to be an account of strongemergence.


causal powers they manifest micro-manifest powers. On the other hand, their other causal powers, those

that are not manifest until they are in emergence engendering situations, he calls micro-latent powers.

The higher-level properties in ordinary situations are physical micro-structural properties (the micro-

manifest properties and relations among the components); these are physically acceptable properties.

The higher-level properties in emergence engendering situations are emergent micro-structural properties

(the micro-latent properties, the micro-manifest properties and relations among the components); these

are emergent properties.

By Shoemaker’s account, the emergent micro-structural properties strongly supervene with nomo-

logical necessity on the physical micro-structural properties. The micro-manifest powers determine (in

some way or another) the micro-latent powers, since for each micro-latent power there is a set of micro-

manifest powers such that a system of properties that has those micro-manifest powers nomologically

necessarily must have that micro-latent power. Furthermore, it gives an account of the downward cau-

sation typical of emergent properties. The emergent property has new causal powers. The micro-latent

properties, not (just) the micro-manifest properties, cause a change to the distribution of both emergent

and physical micro-structural properties.

This account is compatible with the observation that downwardly causal emergent properties of

wholes cause the parts of the whole, including the elementary particles, to behave in ways that are

inexplicable by physics. This activity on the part of the elementary particles is associated with causal

powers that the particles have, but those powers are not had in virtue of the physical laws.37 Instead

of an account that describes strong emergence in terms of higher-level properties, Shoemaker’s account

focuses on the causal powers of the lowest-level properties of systems instantiating emergent properties.

In this regard, it inverts the approach by Wilson (2002, forthcoming) above.

The account is, however, incomplete. As it stands, it requires a prior conception of the difference

between an ordinary situation and an emergence engendering one. To see the problem, consider that

different physical properties (grounded in different physical laws) can be given the same treatment as

emergent properties. For example, suppose we decide by fiat that systems in which the weak force is

manifest are emergence engendering systems. Then mass, charge, colour, and so on, would have the

micro-manifest powers associated with laws of gravitation, electromagnetism, and the strong force. But

there would also be micro-latent powers associated with the weak force. The latter would only manifest

in emergence engendering systems. As the example shows, the account is coherent, but does not help us

determine what an emergent property is; we need to know why a context should be considered emergent

and not physical. Again, this is best done by considering the difference between an emergent law and a

physical law.

4.6 Emergent Laws

It has been argued that an appropriate account of strong emergentism depends upon a principled distinc-

tion between emergent causal laws and physical laws. This section shows why drawing such a distinction

is problematic, but proposes a way of doing so.

37See section 4.5.1.


4.6.1 Pepper and Emergent Laws

That there is a problem distinguishing emergent causal laws and physical laws was noticed early in the

history of the physicalism/emergentism debate. Pepper (1926) remarks that any candidate emergent law

faces a dilemma: either it is epiphenomenal (in which case it is not a genuine case of strong emergence)

or it is properly classified as a physical law. His argument can be summarized as follows. The laws of

physics describe the mathematical relations that govern the transformation of lower-level characteris-

tics (“shifts”). If there are emergent laws, they must instead describe transformations of higher-level

characteristics (“cumulative change”) that cannot be deduced from the those at the lower level (i.e.,

unpredictable change). He then presents a preliminary dilemma: either the higher-level characteristics

bear on the lower-level transformations, or the emergent law is epiphenomenal. Suppose that the higher-

level characteristics have no bearing on the transformation of the lower-level characteristics, but obey

lawful transformations amongst themselves. This would mean that the emergent law is an epiphenome-

nal law,38 contra strong emergentism. Therefore, in order to be an emergent causal law, the higher-level

characteristics must have a bearing on the transformation of the lower-level characteristics.

So suppose instead that they do bear on the transformation of the lower-level characteristics. The

lower-level transformation can be described in terms of a function of several variables (say, q, r, s, and

t), and some range of values for some of the variables (say, r and s) constitute a level of integration at

which the emergent causal law obtains. The transformations governed by this law cannot be a described

by a different function of the same set of variables, since only one function can accurately describe

the relationships among the variables that actually obtain. Thus, the higher-level transformation has

to be described by a function of a different set of variables, (say, r, s, a, and b) which includes some

genuinely novel emergent variables (a and b). Pepper then presents his dilemma: either the variables are

functionally related to the (remaining) lower-level variables, or they are not. Suppose that the emergent

variables are not functionally related to the rest of the lower-level variables. In this case, they are

epiphenomenal. Again, the law governing the variables would not be an emergent causal law, contra

strong emergentism. Therefore, the variables must be functionally related to the lower-level variables.

But now suppose that they are functionally related. Then the variables are part of the correct

description of the lower-level transformation. This would mean that they are, properly speaking, physical

variables. As Pepper says, “they have to drop down and take their place among the lower level variables

as elements in a lower level shift”. Even if the new variables look very different from what is typically

taken to be physical,39 such variables are needed to describe the transformations at the fundamental

level, and so are physical.

The crux of his argument is this: higher-level transformations are either logically necessitated by

lower-level transformations, or they are not. If they are not, then they are epiphenomenal. If they are,

then they are not strongly emergent. They may appear physically unacceptable, since their description

requires variables that do not normally obtain that the lower level. But if this is the case, those variables

are required for the proper description of the lower-level transformation as well, and so are properly called

physical.

38Pepper appears to use the term ‘epiphenomenal’ to refer to laws which fail to be causal at the level of the physical.39Pepper uses Lovejoy’s (1924) example of the movement of complexes of particles being functions of the future movement

of other complexes.


4.6.2 Meehl and Sellars: Emergent Qualities

There are several ways to respond to this argument. One of these is provided by Meehl and Sellars (1956).

Configurations in which emergent variables become necessary for the proper description of the lower-

level transformations are, by hypothesis, different from normal configurations where the mathematical

relations of ordinary lower-level variables is sufficient for the description. In other words, there are some

configurations where the lower-level transformations are governed by an emergent causal law, whereas

in most configurations the lower-level laws suffice. Using Pepper’s toy mathematical relations, there is a

region of (q, r, s, t)-space for which one function describes the relations among these variables (and hence

the transformation of the state), and another region where some other function describes it. The first

region is the ordinary one, and the second is the emergent one.

Note the similarities this view bears to Shoemaker’s account in section 4.5.2 above. According to

Shoemaker, there are emergence-engendering situations where latent causal powers become activated.

Making the simple translation to the language of laws, there are some situations in which emergent

causal laws (partially) govern the evolution of systems. But the incompleteness of Shoemaker’s account

also applies to Meehl and Sellars: there needs to be some way of distinguishing either the situation or

the emergent causal law, for otherwise the new situation might be manifesting a new physical law (as

the example of that section, considering the weak force as emergent, suggests).

Meehl and Sellars suggest an answer to this. The emergent region is one in which (what I’ll call)

emergent qualities are instantiated. A quality is emergent if it cannot be completely defined using only

the theoretical resources adequate to describe the non-living states of physical systems.40 So, some

subset of living systems correspond to an emergent region if the qualities instantiated therein cannot

be adequately defined using the theoretical resources of non-living systems. But again, other than the

peculiar (and arbitrary) appeal to life, such an answer does not differentiate between an emergent region

and a region in which a new fundamental physical force is present. And this, indeed, appears to be

Pepper’s point.

4.6.3 Emergent Laws and Discontinuities

What is needed is a way to distinguish an emergent causal law and a physical law. I propose to distinguish

emergent causal laws and physical laws not by the nature of the law, and not (entirely) by the nature

of the regions of the configuration space they (respectively) govern, but rather by a discontinuity at the

boundaries of the regions.

Recall that the evolution of a physical system can be described as a function of physical variables. In

Pepper’s toy universe, variables q, r, s, and t are sufficient to describe a system in any way that matters

to its evolution. The physical laws that together describe a change in the system are some function,

f1, of the four variables. By convention, f1 describes the change in a region of (q, r, s, t)-space that

includes the least complex configurations of the system. Call it the ordinary region. If emergentism is

true, then there is some other region of (q, r, s, t)-space, an emergence engendering region, at which f1

does not adequately describe the evolution. Some other function, f2, does. I propose that the boundary

between the ordinary region and the emergence engendering region is finitely discontinuous. A path in

(q, r, s, t)-space from the ordinary region to the emergence engendering region has a finite discontinuity

40Emergent qualities are not “definable in terms of theoretical primitives adequate to describe completely the actualstates though not necessarily the potentialities of the universe before the appearance of life” (Meehl and Sellars, 1956,p.252).


at some point if and only if the limit of f1 as (q, r, s, t) approach the point along the path exists, the

limit of f2 as (q, r, s, t) approach the point along the path from the other side exists, and the limits are

not equal. My proposal is that every path from any point in the ordinary region to any point in the

emergence engendering region has a finite discontinuity at the boundary between the regions.41 If f1 and

f2 are landscapes in the (q, r, s, t)-space, the boundary between the ordinary region and the emergence

engendering region is a sheer cliff.42

Metaphysically, such a finite discontinuity can be interpreted in several ways. The natural interpre-

tation, and the one that is relevant to this chapter, is that it represents the activation of a new law of

nature. Because of the discontinuity, the function describing the interaction of elementary particles in

the emergence engendering region can be distinguished from the function describing their interaction in

the ordinary region. Two distinct functions are naturally interpreted as two distinct sets of laws. Re-

latedly, the finite discontinuity can be interpreted as marking a new force, which is felt by the particles

only when they are in a configuration on one side of the discontinuity. Alternatively, the finite discon-

tinuity can be interpreted as indicating that some fundamental force that already exists at low levels of

complexity has a finite discontinuity in its intensity as a function of complexity. These interpretations

may not ultimately be distinct; perhaps there is no difference between a force with a finite discontinuity

in intensity and two forces, and perhaps there cannot be distinct forces without distinct laws. In any

case, a finite discontinuity marking an emergence engendering region can be used to ground a principled

distinction between emergent laws and physical laws.

This proposal appears to be sufficient to avoid Pepper’s dilemma. Suppose strong emergentism is

true, and there is a finite discontinuity at the boundary between two regions of (q, r, s, t)-space, an

ordinary region and an emergence engendering region, the changes within which are described by f1

and f2 respectively. While it is true that the emergent variables of f2 are required to describe lower-

level changes in the emergence engendering region, it does not follow that they must be treated as

physical variables. There is no continuous function that describes the changes described by f1 and f2 in

terms of all the variables simultaneously. The only way to describe the changes with a single function

is to compose a piecewise function that disjoins f1 and f2, each to the appropriate region. But this

composition introduces a new constant, one that delineates the emergence engendering region.43

Furthermore, the proposal appears to be necessary to avoid Pepper’s dilemma. Suppose there is no

finite discontinuity at the boundary between the ordinary region and the emergence engendering region.

Then there is a continuous function, f3, that describes the changes described by both f1 and f2 in their

respective regions in terms of all the variables simultaneously. The laws that hold at the lowest levels

of complexity are therefore subtly different from those that are believed to hold, and this difference is

revealed when they give rise (as a natural consequence) to supposedly emergent behaviour at higher

levels of complexity. Nevertheless, these laws hold at the lowest level of complexity, and are therefore

physical laws. Such is Pepper’s conclusion above, and I think it holds absent a finite discontinuity.

41Contrast a finite discontinuity with the infinite divergences discussed in Batterman (2011).42This is not to say that the functions themselves can’t give objective probabilities for discrete final states. But it

does assume that the change that a system undergoes is not completely (or perfectly) chaotic. In other words, thoughthe function may vary wildly, there will be local linearity, so that at every point not on the boundary of the emergenceengendering region, the limit as (q, r, s, t) approach that point will be the same in every direction. If the variables themselvesare discrete, this characterization would have to be modified accordingly.

43A piecewise function, f3, can be defined as the disjunction of f1 and f2:

f3 =

f1 : (q, r, s, t) /∈ Rf2 : (q, r, s, t) ∈ R

The new parameter, R, represents the emergence engendering region, as delineated by the finite discontinuity.


It might be argued that even a piecewise function with a finite discontinuity is an acceptable function

for describing how a system changes in virtue of a physical law. A function is a function after all, and

both continuous and discontinuous functions map the values of the variables in (q, r, s, t)-space to values

describing the change in the system. This would take Pepper’s conclusion the rest of the way, making

an emergent causal law metaphysically impossible. If so, strong emergentism becomes a metaphysically

suspect position. In the interest of charity, I will therefore assume that functions describing the behaviour

of changes to a system should be continuous.

If my proposal is correct, emergent causal laws and physical laws can be, in principle, anything at

all. It is not the nature of the laws that make them emergent or physical. Rather, it is their relation to

one another. To say that strong emergentism is true is to say that there is a finite discontinuity in the

behaviour of the fundamental constituents of matter as they evolve into some configuration. By conven-

tion, the set of laws that include only physical laws are those that govern the simpler configurations, and

the set of laws that include emergent laws are those that govern the more complex ones.44 The proposal

grounds accounts of strong emergence in a way that conforms to our intuitive understanding of it.

44It might be that a finite discontinuity separates two regions both of which are maximally simple, and so both haveequal claim to being the ordinary region. In that case, plausibly, the finite discontinuity represents not a set of laws thatincludes an emergent law, but a new set of physical laws. But I will assume that the finite discontinuity does not extendto the maximally simple regions, and so delineates an emergence engendering region.

Chapter 5

Evidence Against Strong Emergentism

According to strong emergentism, certain complex configurations of physically acceptable properties

instantiate an emergent property. The emergent property is governed by an emergent causal law, and

downwardly causes changes to the distribution of both emergent and physically acceptable properties.

Strong emergentism therefore has an empirical component: either there is such a property and its

accompanying causal law, or there is not. The position cannot be ruled out a priori.

Furthermore, despite the lack of evidence in favor of the position, there are serious barriers to ruling

out strong emergentism on empirical grounds. In the first place, the bare empirical possibility of strong

emergentism in some domain or other cannot be practically refuted. As stated earlier, at some point

in the distant future, the universe as a whole may reach a level of complexity at which a single atom

is governed by an emergent causal law. But even after restricting the domain of interest to mental

properties, ruling out strong emergentism faces difficulties. A detailed account of the mind in terms of

the physical operations of the brain is, at best, a distant hope. Such an account is the only evidence

that could sway a dedicated strong emergentist.

For the less dedicated strong emergentist, novel experiments could be performed to determine whether

strong emergentism is true. For example, Wilson (2002) argues that the discovery of a new fundamental

force that appears at only high levels of complexity would be evidence for strong emergentism. Presum-

ably, empirical tests could be constructed to find such a force, and the results of a series of such tests

would provide evidence for accepting or rejecting strong emergentism. However, designing and executing

tests to obtain new empirical evidence for or against strong emergentism is beyond the scope of this

chapter.

Fortunately, there are ways to demonstrate the implausibility of strong emergentism given current

empirical evidence. In chapter 4, I argued that accounts of strong emergentism depend on a principled

distinction between an emergent causal law and a physical law. Moreover, I argued that the only plausible

candidate for such a distinction requires a finite discontinuity in the evolution of systems as a function

of complexity. Emergent causal laws govern systems only on the complex side of that discontinuity (the

emergence engendering region of the state space) while physical laws might govern systems on both sides

(the emergence engendering region and the ordinary region).

The requirement of a finitely discontinuous effect puts a strong constraint on the empirical plausibility

of strong emergentism. The absence of a physicalist explanation of mental properties in terms of current

physics is not enough to motivate strong emergentism. Indeed, even positive evidence that current

physics is inadequate to account for mental properties is not enough, since it might be evidence of new

101

Chapter 5. Evidence Against Strong Emergentism 102

physics. What is needed (in addition) is an effect whose intensity is finitely discontinuous as a function

of complexity.

5.1 Empirical Character of Emergent Effects

In the Philosophy of Mind, strong emergentism is a concern only insofar as it accounts for the character

and operations of the mind under ordinary circumstances. Thus, if strong emergentism is true, some

mental properties either are emergent properties, or at least correlate with emergent properties. Given

that at least part of the appeal of strong emergentism is the tension between objective physical properties

and the subjective character of the mind, phenomenal properties should be among those that correlate

with emergent properties. In fact, if phenomenal properties were not correlated with emergent properties,

strong emergentism would lose a big part of its motivation.

This provides some guidance on where to look for emergent properties. Broadly speaking, emergent

properties should be instantiated in brains and related cognitive systems when they reach an appropriate

level of complexity. In order to be more specific, some philosophical reflection is required.

Call a spatiotemporal distribution of causally interacting physically acceptable properties a system.

Systems that also instantiate an emergent property are emergent systems. Evidence for strong emergen-

tism must take the form of a finite discontinuity in the intensity of some effect as systems become more

complex. Call any change in the distribution of physically acceptable properties (downwardly) caused

by an emergent property (in virtue of an emergent causal law) an emergent effect. Emergent effects will

be discontinuous with physically acceptable effects.

Finding the most promising candidates to provide evidence of emergent effects requires addressing

two separate issues. The first issue is identifying which kinds of systems are likely to become emergent

systems at some level of complexity. The second issue is determining at what level of complexity those

systems can be expected to manifest an emergent effect. By looking at a sequence of such systems

ordered from least to most complex, straddling both sides of the level of complexity at which emergent

effects are expected, evidence for or against strong emergentism can be found in the presence or absence

(respectively) of a finite discontinuity.

5.1.1 Emergence Engendering and Manifesting Systems

Consider first the issue of identifying which systems are good candidates for manifesting an emergent

effect at some level of complexity. A system manifests an emergent effect when an emergent property

(downwardly) causes a change in the distribution of physically acceptable properties. There are two

complications to identifying the most promising systems for discovering such effects. First, if a system

manifests an emergent effect, a second system that has the first system as a part also manifests that effect.

One reason for this is mereological: if a part manifests an emergent effect, the whole manifests the same

effect. A more substantive reason is that an emergent effect (say, the inexplicable change in the trajectory

of a particle) can propagate by ordinary physical means. Since causation is transitive, spatiotemporally

distant events will have emergent properties as distal causes, even if the chain of proximal causes between

them is entirely physical.1 Taken to the extreme, the entire universe might manifest an emergent effect

if strong emergentism is true. But this is not particularly helpful when looking for empirical evidence.

1Save, of course, the initial cause in the chain.


Instead, to evaluate evidence that an effect is emergent, it is better to look for (something like) the

smallest system that manifests a given emergent effect. Call this the emergence manifesting system.

The second complication is that the emergence manifesting system needs to be distinguished from the

related system whose complexity (synchronically) determines that the emergent effect manifests. Call

the latter the emergence engendering system.2 The emergence manifesting system and its associated

emergence engendering system could, in principle, come apart. It is not contrary to strong emergentism

that the universe as a whole has to reach a certain level of complexity for a single atom to manifest an

emergent effect.

Both the emergence manifesting system and the emergence engendering system are important for

determining plausible sources of empirical evidence for strong emergentism. On the one hand, since

evidence for strong emergentism is ultimately evidence of an emergent effect, emergence manifesting

systems would have to be the ones investigated. On the other hand, since evidence of strong emergentism

requires a finite discontinuity in the intensity of some effect as a function of complexity, emergence

engendering systems would have to be the ones included in the sequence of systems ordered by complexity.

Fortunately, both the emergence manifesting and engendering systems are likely brains and related

cognitive systems.

Emergence Manifesting Systems

If strong emergentism about mental properties is true, then it is likely some large fraction of the brain

that manifests the emergent property. Perhaps that fraction manifests the emergent effect when taken

together as a whole (like a whole-brain, whole-cortex, or whole-lobe effect), or perhaps the effect is

manifested individually but pervasively among various components (each of many neurons or neural

pathways).

There are several arguments for why the emergence manifesting system cannot be much smaller.

One is an argument from symmetry. While brains have specialized components and layers, the relations

among the neurons (and sub-neuronal parts) within large parts of the brain are similar in terms of

connectivity. For example, layers of the neocortex have characteristic cell types and connectivity. Under

the plausible assumption that the complexity relevant to the manifestation of the emergent effect is

related to composition and connectivity, there seems to be no reason why an emergent causal law would

prefer one neuron in such a situation to another.

A related argument has to do with neural plasticity. The brain has (spatial) regions that are char-

acterized by the typical function performed. But regions can be co-opted to perform functions by

neighbouring ones. This can occur when an injury renders a region’s function useless by depriving it of

a sensorimotor channel, or damages a region performing an important and otherwise available function.

This fact poses a dilemma for those who would argue that the emergence manifesting system is much

smaller than a large fraction of the brain. If the co-opting and the co-opted regions are neurally similar,

then, as argued above, there is no reason to expect an emergent effect in some small fraction of one

but not in the other. Otherwise, if the two regions are neurally different, then it is implausible that a

function resulting from an emergent causal law can be performed by the neurally different regions by

(what must be) the physical propagation of an effect of a very small emergence manifesting system.

Either way, both regions should manifest emergent effects pervasively.

2The nomenclature here is an intentional reference to Shoemaker’s (2002) powers-based account of emergence, describedin section 4.5.2. I used related nomenclature in section 4.6.3.


Third, and again relatedly, the brain is a robust organ. There are usually multiple ways that the

various functions of the brain can be performed. It is unlikely that a very small emergence manifesting

system, say, a lonely neuron, could be essential to the performance the same function under so many

different paths of physical propagation.

Finally, the operations of the brain that would likely be associated with emergent properties are

subtle and complex. If strong emergentism is to account for such activity, the emergence manifesting

system likely has a very large number of degrees of freedom. This is implausible unless the emergent

effects were pervasive.

A lower limit on the scale of the emergence manifesting system is therefore a large fraction of the

brain. But there is also a likely upper limit: it is unlikely that the emergence manifesting system is

much larger than the brain. First, it is implausible that the emergence manifesting system is as big as

the whole body, since the mechanical activity of the body is well understood, and the propagation of

neural signals to (and from) the body are reasonably well understood based on physical principles. It is

much more likely that the brain (or some part of it) manifests an emergent effect, and this propagates

through the understood pathways through the body and out into the world. Furthermore, the possibility

that ordinary, non-cognitive biological processes are themselves strongly emergent can be set aside. No

such processes are thought to be incompatible with current physics.3 But more importantly, from a

dialectical standpoint, if mental properties arose in a non-emergent way from strongly emergent biological

properties, strong emergentism about mental properties would lose its primary motivation.

Emergence Engendering Systems

As with emergence manifesting systems, if strong emergentism about mental properties is true, then it

is likely some large fraction of the whole brain, perhaps including the extended nervous system, that is

the emergence engendering system.

The emergence engendering system is unlikely to be much smaller than a significant fraction of a

brain. In the first place, it is implausible that the emergence engendering system is much smaller (and

less complex) than the emergence manifesting system. Furthermore, if the complexity at which emergent

effects are manifested was much smaller than a large fraction of the brain, it likely would have been

discovered by now. The growth and cultivation of samples of brain tissue in a laboratory setting is

compatible with that of samples of other kinds of tissue, and reveals no inexplicable effects. In addition,

reasonable simulations exist of cognitive operations through neural nets that have a small number of

neurons and connections (small even compared to smaller mammal brains). Any plausible emergent

effect would have to be engendered by a (biological) neural network significantly larger than this.

More interestingly, the emergence engendering system is unlikely to be much larger than the brain plus

(possibly) the extended nervous system. The complexity of a system has to do with the composition and

causal connectivity of the system. The brain and nervous system have components that are similarly

composed and tightly connected through a multitude of complex causal channels. This satisfies our

intuitions of what it is to be a complex system. The brain-body interface, while absolutely complex, is

far less complex in terms of connectivity than the brain. And from the brain-body interface to the rest

of the world, the complexity of the interactions drops off.

One reason this conclusion might be doubted is an apparent tension with externalism in the Philos-

ophy of Mind. Some representational mental states, like beliefs and desires, might be individuated not

3See, for example, Papineau (2008, p.139) and Papineau (2002, appendix).


just by the brain state of the subject, but also by the state of the world outside the subject’s brain,

and indeed by the causal history in which the brain is embedded. A classic example of an argument

for externalism is Putnam’s Twin-Earth argument, according to which two physically identical brains

have distinct beliefs (one has beliefs about water, the other about twin-water) owing to differences in

the subsurface composition of the substances (water is H2O and twin-water is XYZ) with which they

historically interacted. If mental states are individuated in part by the physical state of the world outside

the brain (never mind its causal history), it might seem that the complexity of that world plays a part

in the manifestation of an emergent effect.

However, this conclusion is mistaken. To be sure, if strong emergentism is true, the world does

contribute to the manifestation of an emergent effect in the brain. But it does so through normal causal

channels, not synchronically in virtue of its complexity. If anything, the Twin-Earth argument supports

the opposite conclusion: that the emergence engendering system does not include the world outside the

brain. Recall that the emergence manifesting system does not include the world outside the brain. So if

strong emergentism is true, the two physically and functionally (and, intuitively, phenomenally) identical

brains from Putnam’s thought experiment manifest the same (maximally type-identical) emergent effects.

This is implausible if the emergence engendering system were not also maximally type-identical. Since

the only things that the brains have in common is their internal composition, the brains themselves are

the likely emergence engendering systems.

Therefore, both the emergence manifesting system and the emergence engendering system can be

expected to be large fractions of the brain. If strong emergentism is true, emergent effects should be

manifested pervasively throughout the brain, and the complexity of that same brain is what gives rise

to them.

5.1.2 Levels of Complexity

Consider next the issue of the level of complexity at which an emergent effect can be expected to be

manifested. This is only meaningful when considering some sequence of systems ordered by complexity,

with emergence engendering systems, in this case brains, included in the sequence. The sequences can be

organized into two categories. The first category, cognitive sequences, include sequences of biologically

realized neural networks from fewer neurons and neural connections to more. The second category, non-

cognitive sequences, include sequences of systems from physical to chemical to biological to neurological.4

Sequences that fall into the first category include systems ordered developmentally (from zygote cells

to adult brains), evolutionarily (cognitive systems from early ancestral organisms to modern humans),

and morphologically (from simple to complex cognitive systems of extant organisms or parts thereof).

Beyond being in the same category, these sequences of systems are related, since ontogeny, evolutionary

history, and morphology are themselves closely related. Of the three, it is the morphological sequence

that is most accessible. For example, we have almost no information (and indeed no intuitions) about

the representational and phenomenal states of embryonic cognitive systems except by analogy to similar

but fully-developed systems in other organisms. Similarly, while paleontologists can uncover important

information about the size of hominid brains when certain behaviours like tool-crafting evolved, the

critical period when emergent effects would have first manifested is much earlier, without the associated

4There is, perhaps, another category of sequences. A single neurological system can be ordered mereologically, fromsimpler to more complex subsystems. It’s not clear to me how to obtain evidence for or against strong emergentism froma discontinuity in a mereological sequence of systems.


behavioural evidence. But the cognitive systems of various extant organisms can indeed be compared.

Furthermore, they can be easily ordered by complexity, roughly corresponding to the number of neurons

present.

Organisms with very simple cognitive systems, like the ganglia of lobster digestive tracts, and the

brains of roundworms, sea slugs, and simpler insects, can be expected not to instantiate phenomenality,

and hence not to manifest correlated emergent effects.5 On the other hand, organisms with more

complex cognitive systems, like larger mollusks, vertebrates generally, and especially mammals with

their cerebral cortices, can be expected to instantiate phenomenality (like pain), and hence to manifest

correlated emergent effects. If strong emergentism is true, then somewhere in the sequence of cognitive

systems between the simpler organisms and (at worst) smaller mammals, emergent effects will begin to

manifest, resulting in a finite discontinuity in the operations of the cognitive systems.

In contrast, any sequence of systems from simpler physical systems to chemical, biological, and

finally neurological systems falls into the second category. Complexity in such systems corresponds

roughly to the intuitive notion of levels of nature. The relevant systems are those that have similar

physical components, but are configured differently, so that the sequence represents more and more

complex configurations of the same components. Systems that have similar physical components but are

below the level of complexity of neurological systems should interact with other, non-emergent systems

in ordinary physical ways. Inert chunks of carbon, complex organic chemicals, and (non-cognitive)

biological tissues should all interact with non-emergent physical systems in unsurprising ways. But if

emergent effects are pervasive throughout the brain, there should be a finite discontinuity somewhere

between (non-cognitive) biological tissue and in situ brain tissue in the way that the systems interact

with those non-emergent systems.

5.2 Empirical Evidence Against Strong Emergentism

In this section, I will outline evidence that there is no apparent finite discontinuity in the intensity of

any effect of the brain as a function of complexity.

To establish the context of the evidence, consider the computational power of neural networks. Neural

networks can be divided into two kinds: feed forward (in which signals can move only from the input

to the output) and recurrent (in which signals can be fed back into the network). Recurrent neural

networks are (at least) Turing complete; the solution to any computable function can be found by one.

For any mapping of finite inputs and outputs, a recurrent neural network can realize that mapping.

The brain and other biological cognitive systems are recurrent neural networks.6 Moreover, to good

approximation, the inputs into such systems are finite (roughly, discrete sensory neuron signals over a

short time), and the outputs from them are also finite (roughly, discrete effector neuron signals over a

short time). If a psychological causal law requires of an organism that it behave in a particular way

in particular circumstances (given a particular history of behaviours and circumstances), then, subject

to the constraints above, that law requires of the organism’s cognitive system that it compute a finite

function, i.e., that it map finite inputs to finite outputs. Since recurrent neural networks are Turing

complete, a finite biological neural network operating on purely physical principles, can realize any such

mapping.

5Indeed, if they did, we likely would have found them by now.6See, for example, Lytton (2002, p.102ff).


This does not mean that there are no psychological causal laws that are emergent causal laws. That

there can be a physical neural network that realizes the mapping of any psychological causal law does not

imply that actual cognitive systems do so by purely physical means. But there can be no behavioural

evidence that treats cognitive systems as black boxes that can establish that physics is incorrect or

incomplete, much less that an emergent causal law holds.7

In light of this, evidence for strong emergentism must take the form of potentially emergent effects

with reference to the internal structure of the cognitive systems. Below, I look at two approaches

to finding an emergent effect. One to judge whether the suite of behaviour available to an organism

is compatible with a physically implemented neural network of the level of complexity had by the

organism. The other is to judge whether (non-cognitive) physical interactions with the parts of the

brain are compatible with the same interaction with (non-cognitive) biological tissues.

5.2.1 Behavioural Evidence

The first sort of evidence for emergent effects is a finite discontinuity in the behaviour of organisms

as a function of the complexity of their cognitive systems. If strong emergentism is true, then such a

discontinuity would conform to our naıve intuitions about phenomenal consciousness: that at some point

in the development of a human organism from zygote to adult, it develops phenomenal consciousness;

that at some point in the evolution of human organisms from primitive eukaryotes, organisms evolved

phenomenal consciousness; and that among extant cognitive systems organized by complexity, the more

complex ones belong to organisms with phenomenal consciousness while the less complex ones belong

to those that lack it. The level of complexity at which phenomenal consciousness is first instantiated

should correspond to the threshold at which the emergent causal law begins to hold, and so there should

be a finite discontinuity there.

I argue that there is no empirical evidence for a finite discontinuity in the behaviour of organisms

as a function of complexity. We have a good understanding of the simplest cognitive systems, and the

behaviour of organisms with cognitive systems ordered by complexity does not appear disproportionate

to the complexity of the cognitive system.

The structural complexity of a cognitive system will depend on the kinds of neurons, the number of

neurons, and the number of neural connections. The simplest of these are ganglia, either in the simplest

organisms or in parts of more complex ones. Among the better understood ganglia is the stomatogastric

ganglion of the lobster (and other crustaceans).8 This ganglion is very simple: about thirty large (30–

50µm) neurons of about a dozen types, forming two central pattern generators responsible for rhythmic

motion in the gut of the lobster. The neurons and their connections are well mapped, their interactions

are well understood physically and biologically, and their activity has been simulated.9 We can conclude

that there are no emergent causal laws associated with this system.

Somewhat more complicated, but still remarkably simple, is the neural system of the nematode

(Caenorhabditis elegans).10 The hermaphrodite version of this creature has 302 neurons of 118 types,

and the synaptic connections among these neurons have been completely mapped.11 While there is no

7Unless, of course, the behaviour itself is physically impossible, like accelerating to a speed greater than the speed oflight. But I will set aside such exceptions as plainly incompatible with evidence of human behaviour.

8For an overview of this system, see Selverston (2008).9See, for example, Warshaw and Hartline (1976), Hartline (1979), Marder et al. (1993), and Marder and Bucher (2007).

10See Riddle et al. (1997), especially chapter 22 (Rand and Nonet, 1997).11See White et al. (1986).


complete, working simulation of this neural system,12 many of the functions that it performs have been

successfully modeled on physical and biological principles.13

More complex still is the sea slug (Tritonia diomedea),14 with about 8000 neurons of about 180 types.

Again, owing especially to the size of the neurons, much is known about the neural structure, and some

of its functions (like escape swimming) have been reproduced via simulation.

That so many of these functions can be simulated based on physical principles suggests a distinct lack

of emergent effects. It is exceedingly unlikely that emergent effects in one sort of system, primitive brains

and ganglia in vivo, would be mirrored in a simulation of physical events implemented on completely

different sorts of systems, like integrated in vitro biological and electrical networks or computer circuitry.

These primitive nervous systems are therefore very unlikely to be emergence manifesting systems.

To extend this conclusion to more complex systems, consider that there is a continuum of complexity

among the cognitive systems of organisms. The simplest systems, listed above, range from tens to

thousands of neurons. Insects have a number of neurons ranging from thousands to a million,15 so are

among the organisms occupying the next two to three orders of magnitude. Non-mammalian vertebrates

and small mammals have a number of neurons ranging from tens to hundreds of millions.16 Cephalopods

(like the octopus) and intermediate mammals have hundreds of millions to billions of neurons.17 Large

mammals, including human beings, have tens or hundreds of billions of neurons.18

Crucially, there also appears to be a continuum in the complexity of the behaviour of organisms

(and parts thereof). The suite of behaviours implemented by simple cognitive systems are surprisingly

complex. Stomatogastric ganglia merely produce two distinct rhythmic motions with a few dozen neu-

rons. With an order of magnitude more neurons, the nematode manages to display complex cognitive

behaviour, including habituation, associative learning, and imprinting (Ardiel and Rankin, 2010). More

than an order of magnitude above this, the sea slug is capable of a little more complex behaviour,

including plasticity in its escape response: habituation (a shortened response following frequent stimula-

tion), sensitization (a heightened response following a period of no stimulation), and prepulse inhibition

(suppression of the response given prior tactile stimulation) (Katz, 2007). This meagre increase in be-

havioural complexity is even more remarkable given that the complexity of a cognitive system likely

increases somewhat more than linearly with the number of neurons.19

As cognitive systems become more complex, the rudimentary sensation of the simpler organisms

gives way to more complex sensory systems, and so to a larger quantity and variety of stimuli. At the

level of small insects, there are already dedicated sensory organs whose data can be used represent a

12One may wonder why such a simple neural network has not been simulated. There are two problems. First, theintracellular details of each type of neuron need to be determined to produce an accurate simulation, and this is in generala difficult task. Second, the neurons interact with one another in a specific physical environment, responsive not only todirect electrical signals between one another, but also through more general chemical and biological processes, like therelease of hormones. So an understanding is required of more than just the neurons and their synaptic connections.

13For example, tap withdrawal (Wicks et al., 1996), movement response to chemical stimuli (Ferree et al., 1997; Dunnet al., 2004), head swing motion (Sakata and Shingai, 2004), and forward locomotion (Bryden and Cohen, 2008).

14See Katz (2007).15Small parasitic wasps (genus Megaphragma) can have a number of neurons in the thousands (Polilov, 2012), while

fruit flies (Drosophila melanogaster) have about one hundred thousand neurons (Mery et al., 2009) and honeybees haveabout one million (Menzel and Giurfa, 2001).

16For example, the zebrafish has about ten million neurons (Hinsch and Zupanc, 2007), the mouse about fifty millionneurons (Herculano-Houzel et al., 2006), and the rat about one hundred million neurons (Hochner, 2008).

17An octopus has about five hundred million neurons, a cat about a billion, and a rhesus monkey about two billion. SeeHochner (2008).

18See Hochner (2008).19But see Deamer and Evans (2006) for an argument that the complexity of a system varies as the logarithm of the

number of components.


three-dimensional spatial world.20 They also have the ability to navigate such a world, as a smaller

(and easier to model) range of motions gives way to more extensive motor capacity.21 The range of

functions that can be computed thus naturally increases as the number of possible mappings of inputs

to outputs increases. The learning ability of insects also increases.22 As insects’ cognitive systems get

more complex, they begin to learn through classical conditioning, exhibit instrumental learning, and

can represent, remember, and reliably navigate a three-dimensional world. Higher insects also exhibit

complex social interactions and social learning.23 This complex behaviour is achieved over three orders

of magnitude increase in the number of neurons.

Increasing complexity even more, small mammals exhibit these capabilities reliably and with greater

variety. Again, the quantity and variety of sensory data increases, partly as a result of having larger

bodies with larger surface areas (Chittka and Niven, 2009). While the numbers of muscles are not in

general much greater than those of insects, the number of motor neurons innervating them is much

higher, likely leading to more precise movements (Chittka and Niven, 2009). As with insects, the range

of functions that can be computed naturally increases as the number of possible mappings of inputs to

outputs increases. Learning ability also increases. Mice exhibit a wide range of learning behaviours,

while rats exhibit complex decision-making.24 But these are, again, not discontinuous with the suite of

behaviours available to organisms with somewhat simpler cognitive systems, especially considering the

factor of ten to one hundred higher number of neurons.25 Indeed, the number of distinct behaviours

of higher insects (like the honeybee) is compatible with that of most mammals, and within a factor of

three of that of even large mammals.26

From small mammals to cephalopods, intermediate mammals and large mammals, the increase in

number of neurons ranges from a factor of ten to a factor of one thousand. With all this additional

cognitive capacity, the increase in the complexity of behaviour is not remarkably high. The range of

behaviours is about the same, accounted for by roughly the same sensory and motor capabilities. Some

standout increases in complexity are more sophisticated tool use and manufacture,27 more subtlety in

social behaviour (Dunbar, 1998), and more developed play (Burghardt, 2005). Even these are continuous

with simpler versions of the same behaviour. Among the very most complex cognitive systems, organisms

demonstrate self recognition.28 But even this may not be unique to organisms with the most complex

cognitive systems.29

It may be tempting to suppose that there is a finite discontinuity in behaviour between complex,

non-human mammals and human beings.30 For example, language ability is in some sense discontinuous:

20See, for example, Wang and Spelke (2002).21Insects typically have a number of muscles comparable to vertebrates. See Chittka and Niven (2009).22For a review, see Dukas (2008).23Other learning behaviours include attention, contextual learning, categorization, sequence learning, interval timing,

associative recall, and numerosity (Chittka and Niven, 2009).24See, for example, Whishaw et al. (2001).25Chittka and Niven (2009) provide a review of the literature demonstrating that the more complex cognitive systems

of vertebrates only modestly outperform those of insects.26Ethograms of honeybees list 59 distinct behaviours, while those of bottlenose dolphins list 123 (Chittka and Niven,

2009).27See, for example, Roth and Dicke (2005) and Finn et al. (2009).28Mirror self-recognition has been shown in, among other very complex animals, chimpanzees (Gallup, 1970), dolphins

(Reiss and Marino, 2001), and elephants (Plotnik et al., 2006).29For example, magpies may also show self-recognition (Prior et al., 2008). But given the cognitive ability of corvids

(Emery and Clayton, 2004), perhaps these too should be classified among the organisms with the most complex cognitivesystems.

30See, for example, Penn et al. (2008a), though they take discontinuities to be merely large gaps in a continuum ofabilities that can in principle be filled by some (non-extant) species. See also Penn et al. (2008b) in the same issue.


either the organism has the ability to recursively produce an infinite number of expressions or it does not.

There are many problems with such a position,31 but it is enough to note that it is generally accepted

that larger mammals (at least) possess phenomenal properties analogous to human beings, so emergent

effects would be manifested in them as well.

It is therefore very unlikely that the brain is governed by emergent causal laws associated with

phenomenal properties. There are good physicalist explanations of the behaviour of the simpler cognitive

systems. If the more complex cognitive systems were emergence manifesting systems, governed by an

emergent causal law, evidence of such effects would be expected in a finite discontinuity in the behaviour

of organisms as the structure of their cognitive systems get more complex. There is no evidence of such

a discontinuity.

5.2.2 Evidence from Physical Interactions

The second kind of evidence can be obtained from physical interactions with the brain. If there are

pervasive emergent effects in the brain, these physical interactions will be finitely discontinuous with

similar interactions with non-biological systems or non-cognitive biological tissue.

To see why, consider a phase transition in matter, which can be taken as a physical analogy to a system

going from not manifesting an emergent effect to manifesting one. According to strong emergentism, a

biological cognitive system at a low level of complexity, like that of an embryo, eventually reaches some

critical level of complexity, and becomes an emergent system. The embryonic cognitive system behaves

as would be expected of neurons in that configuration and in that biochemical environment. But after

reaching the critical level of complexity, the cells (or sub-cellular parts) begin to experience (something

like) a new force, and begin to manifest emergent effects. As the cognitive system develops, the effects

become pervasive. In the analogous case of a phase-transition, some matter in liquid form (say, in a

container under constant pressure) begins to cool, reaches some critical temperature (the freezing point)

and finally begins to become a solid. The liquid behaves as physicists would expect from molecules with

the degrees of freedom available to them. But at the freezing point, the molecules begin to experience

a (we can imagine hitherto unknown) binding force that manifests itself as they begin form a crystal

structure. As energy continues to be removed from the matter, the effect becomes pervasive.

Matter in a solid state behaves differently than matter in a liquid state because of the additional

forces experienced by the molecules. Evidence that it behaves differently can be found through a vari-

ety of physical interactions with the matter, from measuring bulk properties (like volume) to particle

interactions (like X-ray scattering). Physical interactions can reveal the effects of the phase of matter

precisely because the phase of matter makes a physical difference causally, and because the effect is

pervasive. By analogy, because the emergent effect is governed by an emergent causal law, and because

the effect is pervasive, physical interactions should reveal a difference between the brain and ordinary

biological matter.

No such evidence is to be found. There is considerable evidence that the brain interacts with non-

emergent physical systems in a way not discontinuous with physical, chemical, and (non-cognitive)

biological systems. One line of evidence comes from particle physics used in the service of medicine.

Energetic particles have diagnostic and therapeutic uses when it comes to the brain.32 The successful

31See footnote 30. Among the responses to that paper (in the same issue) are Burghardt (2008), Emery and Clayton(2008), and Herman et al. (2008).

32A thorough description of the use of physics in medical imaging can be found in Bushberg et al. (2012). A thoroughdescription of the therapeutic use of particles can be found in Khan (2012).


use of particle-tissue interactions in a medical context depends crucially on understanding the nature

of the interaction in order to interpret the data.33 This understanding demonstrates that brain tissue

is continuous with body tissue outside the nervous system, with ordinary variations based on physical

characteristics.34 The success of brain and body imaging using the same techniques is evidence that no

pervasive emergent effect exists in the brain alone.

Another line of evidence relies on the bulk properties of brain tissue. Subjecting parts of the body

to vibrations can reveal the mechanical and structural properties of tissues. These properties vary

between healthy and unhealthy tissue, so the vibrations are an important part of diagnostic procedures

like Magnetic Resonance Elastography.35 Models of the mechanical and structural properties of tissues

can be determined from these procedures. The brain is a special case: it is at least uncomfortable

and possibly dangerous to subject it to vibrations, and the presence of the skull and protective tissues

dampens the vibrations. Complex techniques using physical principles (poromechanics, or the motion of

fluids through porous membranes) have been used to determine the mechanical and structural properties

of brain tissue from vibrations induced by blood flow.36 The success of the technique shows that the

mechanical and structural properties of the tissue are continuous with that of (non-cognitive) biological

tissue.

5.3 Conclusion

If strong emergentism is true, there must be a finite discontinuity in the intensity of some effect as a

function of complexity. In the domain of interest, strong emergentism about mental properties, brains

are both the systems that manifest emergent effects, and the systems whose complexity (synchronically)

determines that the effect is manifested. Sequences of systems ordered by complexity, either cognitive

systems ordered from simplest to most complex or material systems ordered from physical to neurological,

should show a finite discontinuity in the intensity of some effect.

There is no apparent finite discontinuity in the behaviour of extant organisms when their cognitive

systems are organized from least to most complex. There are good physicalist explanations and sim-

ulations of the simplest cognitive systems. As such systems become more complex, the behavioural

repertoire does not increase dramatically, considering eight orders of magnitude increase in the number

of neurons.

There is also no apparent finite discontinuity in the non-cognitive response of brains to physical

interactions when compared to simpler systems with similar components (like ordinary biological tissue).

Particle interactions with brain tissue (for diagnostic and therapeutic purposes) are found not to be

discontinuous with similar interactions with other organs. The brain’s response to vibrations has been

determined using general physical principles, showing that no finite discontinuity in the mechanical and

structural properties of the brain should be expected.

Together, these lines of evidence greatly reduce the plausibility of strong emergentism about mental

properties.

33One interesting approach to this understanding is the use of Monte Carlo simulations of the interactions in (for example)dosimetry (to optimize radiation therapy doses). See Andreo (1991).

34For example, the reaction of brain tissue to low-energy neutrons varies from that of muscle tissue because of the relativehydrogen and nitrogen content. See Burmeister et al. (2000).

35For an overview of Magnetic Resonance Elastography, see Mariappan et al. (2010).36See Weaver et al. (2012).

Bibliography

P. Andreo (1991). Monte carlo techniques in medical radiation physics. Physics in Medicine and Biology,

36(7):861.

Louise M. Antony (2003). Who’s afraid of disjunctive properties? Philosophical Issues, 13:1–21.

Louise M. Antony and Joseph M. Levine (1997). Reduction with autonomy. Philosophical Perspectives,

11:83–105.

Evan L. Ardiel and Catharine H. Rankin (2010). An elegant mind: Learning and memory in Caenorhab-

ditis elegans. Learning and Memory, 17:191–201.

David Armstrong (1968). A Materialist Theory of the Mind. Routledge and Kegan Paul, London.

David Armstrong (1989). A Combinatorial Theory of Possibility. Cambridge University Press, Cam-

bridge.

Alan Baker (2011). Simplicity. In Edward N. Zalta, editor, The Stanford Encyclopedia of Philosophy.

Summer 2011 edition.

Robert Batterman (2011). Emergence, singularities, and symmetry breaking. Foundations of Physics,

41:1031–1050.

Alexander Bird (2005). The dispositionalist conception of laws. Foundations of Science, 10:353–370.

Ned Block and Robert Stalnaker (1999). Conceptual analysis, dualism, and the explanatory gap. Philo-

sophical Review, 108:1–46.

Alistair N. Boettiger and George Oster (2009). Emergent complexity in simple neural systems. Com-

municative and Integrative Biology, 2(6):467–470.

George Boolos (1984). To be is to be a value of a variable (or to be some values of some variables).

Journal of Philosophy, 81(8):430–449.

John Bryden and Netta Cohen (2008). Neural control of Caenorhabditis elegans forward locomotion:

the role of sensory feedback. Biological Cybernetics, 98:339–351.

Mario Bunge (2001). Philosophy in Crisis: The Need for Reconstruction. Prometheus Books, Amherst,

NY.

Gordon M. Burghardt (2005). The Genesis of Play: Testing the Limits. MIT Press, Cambridge, Mass.

112

BIBLIOGRAPHY 113

Gordon M. Burghardt (2008). The sun always rises: Scientists also need semantics. Behavioral and

Brain Sciences, 31(2):122–134.

Jay Burmeister, Chandrasekhar Kota, Richard L. Maughan, John J. Spokas, Jeffrey A. Coderre, Ruimei

Ma, and Lucian Wielopolski (2000). A conducting plastic simulating brain tissue. Medical Physics,

27:2560–2564.

Jerrold T. Bushberg, J. Anthony Seibert, Edwin M. Leidholdt, Jr., and John M. Boone (2012). The

Essential Physics of Medical Imaging. Lippencott Williams and Wilkins, Philadelphia.

Keith Campbell (1981). The metaphysic of abstract particulars. Midwest Studies in Philosophy, 6:477–

488.

David J. Chalmers (1996). The Conscious Mind: In Search of a Fundamental Theory. Oxford University

Press, Oxford.

David J. Chalmers (2002a). Consciousness and its place in nature. In David J. Chalmers, editor,

Philosophy of Mind: Classical and Contemporary Readings. Oxford University Press, Oxford. http:

//consc.net/papers/nature.html.

David J. Chalmers (2002b). On sense and intension. http://consc.net/papers/intension.html.

David J. Chalmers (2004a). Epistemic two-dimensional semantics. Philosophical Studies, 118:153–226.

David J. Chalmers (2004b). The foundations of two-dimensional semantics. In Manuel Garcıa-Carpintero

and Josep Macia, editors, Two-Dimensional Semantics: Foundations and Applications. Oxford Uni-

versity Press, Oxford.

David J. Chalmers (2006). Two-dimensional semantics. In Ernest LePore and Barry C. Smith, editors,

Oxford Handbook of the Philosophy of Language. Oxford University Press, Oxford.

David J. Chalmers (2009). The two-dimensional argument against materialism. In Brian McLaughlin,

editor, Oxford Handbook of the Philosophy of Mind. Oxford University Press, Oxford.

David J. Chalmers (2011). Verbal disputes. The Philosophical Review, 120:515–566.

David J. Chalmers (2012). The combination problem for panpsychism. http://consc.net/papers/

combination.pdf.

Lars Chittka and Jeremy Niven (2009). Are bigger brains better? Current Biology, 19(21):R995–R1008.

Paul Churchland (1984). Matter and Consciousness. MIT Press, Cambridge, Mass.

Lenny Clapp (2001). Disjunctive properties: Multiple realizations. Journal of Philosophy, 98:111–136.

R. Clarke (1999). Nonreductive physicalism and the causal powers of the mental. Erkenntnis, 51:295–322.

Jonathan Cohen and Craig Callender (2009). A better best system account of lawhood. Philosophical

Studies, 145:1–34.

Donald Davidson (1970). Mental events. In Lawrence Foster and J. W. Swanson, editors, Experience

and Theory. Duckworth, London.

BIBLIOGRAPHY 114

David W. Deamer and John Evans (2006). Numerical analysis of biocomplexity. In Joseph Seckbach,

editor, Life As We Know It, pages 201–212. Springer, New York.

Daniel C. Dennett (1991). Consciousness Explained. Little, Brown & Co., Boston.

Janice L. Dowell (2006a). Formulating the thesis of physicalism: An introduction. Philosophical Studies,

131:1–23.

Janice L. Dowell (2006b). The physical is empirical, not metaphysical. Philosophical Studies, 131:25–60.

Fred Dretske (1995). Naturalizing the Mind. MIT Press, Cambridge, Mass.

Reuven Dukas (2008). Evolutionary biology of insect learning. Annual Review of Entomology, 53:145–

160.

Robin I. M. Dunbar (1998). The social brain hypothesis. Evolutionary Anthropology, 6:178–190.

N. A. Dunn, S. R. Lockery, J. T. Pierce-Shimomura, and J. S. Conery (2004). Neural network model of

chemotaxis predicts functions of synaptic connections in the nematode Caenorhabditis elegans. Journal

of Computational Neuroscience, 17:137–147.

Princess Elisabeth of Bohemia and Rene Descartes (2007). The Correspondence between Princess Elis-

abeth of Bohemia and Rene Descartes. University of Chicago Press, Chicago. Edited and translated

by Lisa Shapiro.

Nathan J. Emery and Nicola S. Clayton (2004). The mentality of crows: Convergent evolution of

intelligence in corvids and apes. Science, 306(5703):1903–1907.

Nathan J. Emery and Nicola S. Clayton (2008). Imaginative scrub-jays, causal rooks, and a liberal

application of Occam’s aftershave. Behavioral and Brain Sciences, 31(2):134–135.

Gregory S. Engel, Tessa R. Calhoun, Elizabeth L. Read, Tae-Kyu Ahn, Tomas Mancal, Yuan-Chung

Cheng, Robert E. Blankenship, and Graham R. Fleming (2007). Evidence for wavelike energy transfer

through quantum coherence in photosynthetic systems. Nature, 446:782–786.

Herbert Feigl (1958). Concepts, theories and the mind-body problem. In H. Feigl, M. Scriven, and

G. Maxwell, editors, The “Mental” and the “Physical”. University of Minnesota Press, Minneapolis.

T. C. Ferree, B. A. Marcotte, and S. R. Lockery (1997). Neural network models of chemotaxis in the

nematode Caenorhabditis elegans. Advances in Neural Information Processing Systems, 9:55–61.

Kit Fine (2002). The varieties of necessity. In Tamar Szabo Gendler and John Hawthorne, editors,

Conceivability and Possibility. Clarendon Press, Oxford.

Julian K. Finn, Tom Tregenza, and Mark D. Norman (2009). Defensive tool use in a coconut-carrying

octopus. Current Biology, 19(23):R1069–R1070.

Jerry Fodor (1974). Special sciences (or: The disunity of science as a working hypothesis). Synthese,

28:77–115.

Jerry Fodor (1997). Special sciences: Still autonomous after all these years. Nous, 31:149–163.

BIBLIOGRAPHY 115

Gordon G. Gallup (1970). Chimpanzees: self-recognition. Science, 167(3914):86–87.

S. C. Gibb (2004). The problem of mental causation and the nature of properties. Australasian Journal

of Philosophy, 82:464–476.

George Graham and Terence Horgan (2000). Mary Mary, quite contrary. Philosophical Studies, 99:59–87.

Toby Handfield (2004). Counterlegals and necessary laws. Philosophical Quarterly, 54:402–419.

Daniel K. Hartline (1979). Pattern generation in the lobster (Panulirus) stomatogastric ganglion. II.

Pyloric network simulation. Biological Cybernetics, 33(4):223–236.

Carl Hempel (1969). Reduction: Ontological and linguistic facets. In Patrick Suppes, Sidney Mor-

genbesser, and Morgan White, editors, Philosophy, Science, and Method: Essays In Honor of Ernest

Nagel. St. Martins Press, New York.

S. Herculano-Houzel, B. Mota, and R. Lent (2006). Cellular scaling rules for rodent brains. Proceedings

of the National Academy of Sciences of the United States of America, 103(32):12138–12143.

Louis M. Herman, Robert K. Uyeyama, and Adam A. Pack (2008). Bottlenose dolphins understand

relationships between concepts. Behavioral and Brain Sciences, 31(2):139–140.

K. Hinsch and G. K. H. Zupanc (2007). Generation and long-term persistence of new neurons in the

adult zebrafish brain: A quantitative analysis. Neuroscience, 146(2):679–696.

Binyamin Hochner (2008). Octopuses. Current Biology, 18:R897.

Paul Humphreys (1997a). How properties emerge. Philosophy of Science, 64:1–17.

Paul Humphreys (1997b). Emergence, not supervenience. Philosophy of Science, 64 (Proceedings):S337–

S345.

Thomas Henry Huxley (1874). On the hypothesis that animals are automata, and its history. The

Fortnightly Review, 16 (New Series):555–580.

Frank Jackson (1982). Epiphenomenal qualia. Philosophical Quarterly, 32:127–136.

William James (1890). The Principles of Psychology, volume 1. Henry Holt and Co., New York.

http://psychclassics.yorku.ca/James/Principles/.

William Jaworski (2002). Multiple-realizability, explanation and the disjunctive move. Philosophical

Studies, 108:298–308.

Paul S. Katz (2007). Tritonia. Scholarpedia, 2(6):3504.

Faiz M. Khan (2012). The Physics of Radiation Therapy. Lippencott Williams and Wilkins, Philadelphia.

Jaegwon Kim (1992). Multiple realization and the metaphysics of reduction. Philosophy and Phenomeno-

logical Research, 52:1–26.

Jaegwon Kim (1993). Mind and Supervenience. Cambridge University Press, Cambridge.

Jaegwon Kim (1998). Mind in a Physical World. MIT Press, Cambridge, Mass.

BIBLIOGRAPHY 116

Jaegwon Kim (2005). Physicalism, Or Something Near Enough. Princeton University Press, Princeton.

Marc Lange (2000). Natural Laws in Scientific Practice. Oxford University Press, Oxford.

Marc Lange (2009). Must the fundamental laws of physics be complete? Philosophy and Phenomeno-

logical Research, 78:312–345.

Ernest LePore and Barry Loewer (1989). More on making mind matter. Philosophical Topics, 17:175–

192.

Janet Levin (2013). Functionalism. In Edward N. Zalta, editor, The Stanford Encyclopedia of Philosophy.

Fall 2013 edition.

David Lewis (1966). An argument for the identity theory. Journal of Philosophy, 63:17–25.

David Lewis (1972). Psychophysical and theoretical identifications. Australasian Journal of Philosophy,

50:249–258.

David Lewis (1973). Counterfactuals. Blackwell, Oxford.

David Lewis (1986). On the Plurality of Worlds. Blackwell, Oxford.

David Lewis (1988). What experience teaches. Proceedings of the Russellian Society, 13:29–57.

David Lewis (1991). Parts of Classes. Blackwell, Oxford.

Barry Loewer (2001). From physics to physicalism. In Carl Gillett and Barry Loewer, editors, Physicalism

and Its Discontents, pages 37–56. Cambridge University Press, Cambridge.

Arthur O. Lovejoy (1924). Essays in Metaphysics, Lectures Delivered by John Laird, V. F. Lenzen,

William R. Dennes, J. Loewenberg, David W. Prall, George P. Adams, Arthur O. Lovejoy. University

of California Press.

William W. Lytton (2002). From Computer to Brain: Foundations of Computational Neuroscience.

Springer, Heidelberg.

Eve Marder, Laurence F. Abbott, Frank Buchholtz, Irving R. Epstein, Jorge Golowasch, Scott L. Hooper,

and Thomas B. Kepler (1993). Physiological insights from cellular and network models of the stom-

atogastric nervous system of lobsters and crabs. American Zoologist, 33(1):29–39.

Eve Marder and Dirk Bucher (2007). Understanding circuit dynamics using the stomatogastric nervous

system of lobsters and crabs. Annual Review of Physiology, 69:291–316.

Yogesh K. Mariappan, Kevin J. Glaser, and Richard L. Ehman (2010). Magnetic resonance elastography:

A review. Clinical Anatomy, 23(5):497–511.

Tim Maudlin (2007). The Metaphysics Within Physics. Oxford University Press, Oxford.

Brian P. McLaughlin (1992). The rise and fall of British Emergentism. In Ansgar Beckerman, Hans

Flohr, and Jaegwon Kim, editors, Emergence or Reduction? Essays on the Prospects of Nonreductive

Physicalism, pages 49–93. de Gruyter, Berlin.

Brian P. McLaughlin (1997). Emergence and supervenience. Intellectica, 25:25–43.

BIBLIOGRAPHY 117

P. E. Meehl and Wilfrid Sellars (1956). The concept of emergence. Minnesota Studies in the Philosophy

of Science, 1:239–252.

Andrew Melnyk (2003a). A Physicalist Manifesto. Cambridge University Press, Cambridge.

Andrew Melnyk (2003b). Some evidence for physicalism. In Sven Walter and Heinz-Dieter Heckmann,

editors, Physicalism and Mental Causation. Imprint Academic, Exeter.

R. Menzel and M. Giurfa (2001). Cognitive architecture of a mini-brain: the honeybee. Trends in

Cognitive Sciences, 5:62–71.

F. Mery, E. Danchin, S. Blanchet, D. Parejo, I. Coolen, and R. H. Wagner (2009). Public versus personal

information for mate copying in an invertebrate. Current Biology, 19:730–734.

Barbara Montero (1999). The body problem. Nous, 33:183–200.

Barbara Montero (2003). Varieties of causal closure. In Sven Walter and Heinz-Dieter Heckmann,

editors, Physicalism and Mental Causation. Imprint Academic, Exeter.

Barbara Montero and David Papineau (2005). A defense of the via negativa argument for physicalism.

Analysis, 65:233–237.

Thomas Mormann (1995). Trope sheaves: A topological ontology of tropes. Logic and Logical Philosophy,

3:129–150.

Ernest Nagel (1961). The Structure of Science. Routledge and Kegan Paul, London.

Thomas Nagel (1986). The View from Nowhere. Oxford University Press, Oxford.

Lawrence Nemirow (1990). Physicalism and the cognitive role of acquaintance. In William G. Lycan,

editor, Mind and Cognition. Blackwell, Oxford.

Alyssa Ney (2008a). Defining physicalism. Philosophy Compass, 3:1033–1048.

Alyssa Ney (2008b). Physicalism as an attitude. Philosophical Studies, 138:1–15.

Alyssa Ney (2010). Convergence on the problem of mental causation: Shoemakers strategy for (nonre-

ductive?) physicalists. Philosophical Issues, 20:438–445.

Timothy O’Connor (1994). Emergent properties. American Philosophical Quarterly, 31:91–104.

Timothy O’Connor (2000a). Causality, mind, and free will. Philosophical Perspectives, 14:105–117.

Timothy O’Connor (2000b). Persons and Causes: The Metaphysics of Free Will, chapter 6. Oxford

University Press, Oxford.

Timothy O’Connor and Hong Yu Wong (2005). The metaphysics of emergence. Nous, 39:658–678.

David Papineau (2001). The rise of physicalism. In Carl Gillett and Barry Loewer, editors, Physicalism

and Its Discontents, pages 3–36. Cambridge University Press, Cambridge.

David Papineau (2002). Thinking about Consciousness. Oxford University Press, Oxford.

BIBLIOGRAPHY 118

David Papineau (2008). Must a physicalist be microphysicalist? In Jakob Hohwy and Jesper Kallestrup,

editors, Being Reduced: New Essays on Reduction, Explanation, and Causation, pages 126–148. Oxford

University Press, Oxford.

L. A. Paul (2002). Logical parts. Nous, 36:578–596.

Derek C. Penn, Keith J. Holyoak, and Daniel J. Povinelli (2008a). Darwin’s mistake: Explaining the

discontinuity between human and nonhuman minds. Behavioral and Brain Sciences, 31(2):109–130.

Derek C. Penn, Keith J. Holyoak, and Daniel J. Povinelli (2008b). Darwin’s triumph: Explaining the

uniqueness of the human mind without a deus ex machina. Behavioral and Brain Sciences, 31(2):153–

178.

Stephen C. Pepper (1926). Emergence. Journal of Philosophy, 23:241–245.

Derk Pereboom (2002). Robust nonreductive materialism. Journal of Philosophy, 99:499–531.

Derk Pereboom and Hilary Kornblith (1991). The metaphysics of irreducibility. Philosophical Studies,

63:125–145.

John Perry (2001). Knowledge, Possibility, and Consciousness. MIT Press, Cambridge, Mass.

Ullin T. Place (1956). Is consciousness a brain process? British Journal of Psychology, 47:44–50.

Joshua M. Plotnik, Frans B. M. de Waal, and Diana Reiss (2006). Self-recognition in an Asian elephant.

Proceedings of the National Academy of Sciences of the United States of America, 106(45):17053–

17057.

Alexey A. Polilov (2012). The smallest insects evolve anucleate neurons. Arthropod Structure and

Development, 41(1):29–34.

Helmut Prior, Ariane Schwarz, and Onur Gunturkun (2008). Mirror-induced behavior in the magpie

(Pica pica): Evidence of self-recognition. PLoS Biology, 6(8):e202.

Hilary Putnam (1967). Psychological Predicates, pages 37–48. University of Pittsburgh Press, Pittsburgh.

James B. Rand and Michael L. Nonet (1997). Synaptic transmission. In Donald L. Riddle, Thomas

Blumenthal, Barbara J. Meyer, and James R. Priess, editors, C. elegans II. Cold Spring Harbor Press,

Cold Spring Harbor, NY.

Diana Reiss and Lori Marino (2001). Mirror self-recognition in the bottlenose dolphin: A case of cogni-

tive convergence. Proceedings of the National Academy of Sciences of the United States of America,

98(10):5937–5942.

Donald L. Riddle, Thomas Blumenthal, Barbara J. Meyer, and James R. Priess, editors (1997). C.

elegans II. Cold Spring Harbor Press, Cold Spring Harbor, NY.

David Robb (1997). The properties of mental causation. Philosophical Quarterly, 47:178–194.

David Robb (2013). The identity theory as a solution to the exclusion problem. In S. C. Gibb, E. J. Lowe,

and R. D. Ingthorsson, editors, Mental Causation and Ontology, pages 215–232. Oxford University

Press, Oxford.

BIBLIOGRAPHY 119

Gerhard Roth and Ursula Dicke (2005). Evolution of the brain and intelligence. Trends in Cognitive

Sciences, 9(5):250–257.

Milton A. Rothman (1988). A physicist’s guide to skepticism. Prometheus Books, Buffalo, NY.

Bertrand Russell (1921). The Analysis of Mind. George Allen & Unwin (1978 Reprint), London.

Bertrand Russell (1922). Physics and perception. Mind, 31:478–485.

K. Sakata and R. Shingai (2004). Neural network model to generate head swing in locomotion of

Caenorhabditis elegans. Network: Computation in Neural Systems, 15:199–216.

Mohan Sarovar, Akihito Ishizaki, Graham R. Fleming, and K. Birgitta Whaley (2010). Quantum en-

tanglement in photosynthetic light-harvesting complexes. Nature Physics, 6:462–467.

William Seager (1995). Consciousness, information, and panpsychism. Journal of Consciousness Studies,

2:272–288.

William Seager (1999). Theories of Consciousness. Routledge, London.

William Seager (2006). The intrinsic nature argument for panpsychism. Journal of Consciousness

Studies, 13(10–11):129–145.

William Seager (2012). Natural Fabrications: Science, Emergence and Consciousness. Springer, Heidel-

berg.

William Seager and Sean Allen-Hermanson (2012). Panpsychism. In Edward N. Zalta, editor, The

Stanford Encyclopedia of Philosophy. Winter 2012 edition.

Allen Selverston (2008). Stomatogastric ganglion. Scholarpedia, 3(4):1661.

Sydney Shoemaker (2001). Realization and mental causation. In Carl Gillett and Barry Loewer, editors,

Physicalism and Its Discontents, pages 74–98. Cambridge University Press, Cambridge.

Sydney Shoemaker (2002). Kim on emergence. Philosophical Studies, 108:53–63.

Sydney Shoemaker (2007). Physical Realization. Oxford University Press, Oxford.

J. J. C. Smart (1959). Sensations and brain processes. Philosophical Review, 68:141–156.

Baruch Spinoza (1677). Ethics.

Daniel Stoljar (2001). Two conceptions of the physical. Philosophy and Phenomenological Research,

62:253–270.

Galen Strawson (2006). Realistic monism: Why physicalism entails panpsychism. Journal of Conscious-

ness Studies, 13(3–31):129–145.

Michael Tye (2000). Knowing what it is like: The ability hypothesis and the knowledge argument. In

Consciousness, Color, and Content. MIT Press, Cambridge, Mass.

James Van Cleve (1990). Emergence vs. panpsychism: Magic or mind dust? Philosophical Perspectives,

4:215–226.

BIBLIOGRAPHY 120

Robert Van Gulick (2001). Reduction, emergence and other recent options on the mind/body problem:

A philosophic overview. Journal of Consciousness Studies, 8(9–10):1–34.

Ranxiao Frances Wang and Elizabeth S. Spelke (2002). Human spatial representation: insights from

animals. Trends in Cognitive Sciences, 9(6):376–382.

Howard S. Warshaw and Daniel K. Hartline (1976). Simulation of network activity in stomatogastric

ganglion of the spiny lobster, panulirus. Brain Research, 110(2):259–272.

John B. Weaver, Adam J. Pattison, Matthew D. McGarry, Irina M. Perreard, Jessica G. Swienckowski,

Clifford J. Eskey, S. Scott Lollis, and Keith D. Paulsena (2012). Brain mechanical property measure-

ment using mre with intrinsic activation. Physics in Medicine and Biology, 57(22):7275–7287.

Ian Q. Whishaw, Gerlinde A. S. Metz, Bryan Kolb, and Sergio M. Pellis (2001). Accelerated nervous

system development contributes to behavioral efficiency in the laboratory mouse: A behavioral review

and theoretical proposal. Developmental Psychobiology, 39(3):151–170.

J. G. White, E. Southgate, J. N. Thomson, and S. Brenner (1986). The structure of the nervous system

of the nematode Caenorhabditis elegans. Philosophical Transactions of the Royal Society of London.

Series B, Biological Sciences, 314(1165):1–340.

S. R. Wicks, C. J. Roehrig, and C. H. Rankin (1996). A dynamic network simulation of nematode

tap withdrawal circuit: predictions concerning synaptic function using behavioral criteria. Journal of

Neuroscience, 16:4017–4031.

Donald C. Williams (1953). On the elements of being: I. The Review of Metaphysics, 7(1):3–18.

Jessica Wilson (2002). Causal powers, forces, and superdupervenience. Grazer Philosophische Studien,

63:53–78.

Jessica Wilson (2005). Supervenience-based formulations of physicalism. Nous, 39:426–459.

Jessica Wilson (2006). On characterizing the physical. Philosophical Studies, 131:61–99.

Jessica Wilson (2009). Determination, realization, and mental causation. Philosophical Studies, 145:149–

169.

Jessica Wilson (2010a). Non-reductive physicalism and degrees of freedom. British Journal for the

Philosophy of Science, 61:279–311.

Jessica Wilson (2010b). What is Hume’s dictum, and why believe it? Philosophy and Phenomenological

Research, 80:595–637.

Jessica Wilson (2011). Non-reductive realization and the powers-based subset strategy. The Monist,

94:121–154.

Jessica Wilson (forthcoming). Metaphysical emergence: Weak and strong. In Tomasz Bigaj Christian

Wuthrich, editor, Metaphysics in Contemporary Physics. Poznan Studies in the Philosophy of the

Sciences and the Humanities.

D. Gene Witmer (2001). Sufficiency claims and physicalism. In Carl Gillett and Barry Loewer, editors,

Physicalism and Its Discontents, pages 57–73. Cambridge University Press, Cambridge.

BIBLIOGRAPHY 121

D. Gene Witmer (2003). Functionalism and causal exclusion. Pacific Philosophical Quarterly, 84:198–

214.

Stephen Yablo (1992). Mental causation. Philosophical Review, 101:245–280.

Byeong-uk Yi (1999). Is mereology ontologically innocent? Philosophical Studies, 93:141–160.

Documents

by Ranpal Singh Dosanjh - University of Toronto T …...Abstract A Defense of Reductive Physicalism Ranpal Singh Dosanjh Doctor of Philosophy Graduate Department of Philosophy University