See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/279917743

Reduction in Philosophy of Mind: A Pluralistic Account

Book · September 2011

DOI: 10.1515/9783110332131

CITATIONS READS

23 544

1 author:

Markus Ilkka Eronen

University of Groningen
33 PUBLICATIONS   452 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Nature and History of the Concept of 'Levels of Organization' View project

All content following this page was uploaded by Markus Ilkka Eronen on 09 July 2015.

The user has requested enhancement of the downloaded file.

Reduction in Philosophy of Mind: A Pluralistic Account

PhD thesis

University of Osnabrück

August 2011

Markus I. Eronen
 Abstract

The notion of reduction continues to play a key role in philosophy of mind and
philosophy of cognitive science. Supporters of reductionism claim that psychological
properties or explanations reduce to neural properties or explanations, while antireductionists
claim that such reductions are not possible. In this book, I apply recent developments in
philosophy of science, particularly the mechanistic explanation paradigm and the
interventionist theory of causation, to reassess the traditional approaches to reduction in
philosophy of mind. I then elaborate and defend a pluralistic framework for philosophy of
mind, and show how reductionist ideas can be incorporated into it. This leads to a novel
synthesis of pluralism and reductionism that I call pluralistic physicalism.

Keywords: reduction, reductionism, reductive explanation, functional reduction,

mechanistic explanation, explanatory power, pluralism, explanatory pluralism, levels, levels
of organization, physicalism, pluralistic physicalism, interventionism, causation, causal
exclusion, neuroscience, psychology, cognitive science
CONTENTS

Introduction.................................................................................................. 7  
PART I: Reduction in Philosophy of Science............................................ 11  
Introduction............................................................................................. 13  
1. Reduction: From Derivations of Theories to Ruthless Metascience .. 15  
2. Mechanistic Explanation..................................................................... 25  
3. The Interventionist Account of Causation .......................................... 35  
4. Levels.................................................................................................. 41  
Conclusions: Reductionism vs. Explanatory pluralism .......................... 53  
PART II: Rethinking Reduction in Philosophy of Mind ........................... 55  
Introduction............................................................................................. 57  
5. Traditional Approaches to Reduction in Philosophy of Mind............ 59  
5.1. British Emergentism ..................................................................... 59  
5.2. Logical Behaviorism and Identity Theory.................................... 65  
5.3. Multiple Realizability ................................................................... 68  
5.4. The Disunity of Science as a Working Hypothesis ...................... 74  
5.5. Functionalism ............................................................................... 77  
5.6. The Dream of Nonreductive Physicalism..................................... 82  
6. Functional Reduction.......................................................................... 87  
6.1. The Causal Exclusion Argument and the Functional Model........ 87  
6.2. Kim vs. Nagel ............................................................................... 92  
6.3. Dissecting the Functional Model.................................................. 94  
6.3.1. Functionalization .................................................................... 95  
6.3.2. Realization.............................................................................. 98  
6.3.3. Causation .............................................................................. 102  
6.4. Functional Reduction as Mechanistic Explanation .................... 106  
 7. Phenomenal Consciousness and the Explanatory Gap......................109  
8. New Type Physicalism......................................................................119  
Conclusions: Rethinking Reduction in Philosophy of Mind.................131  
PART III: A New Framework for Philosophy of Mind ...........................133  
Introduction ...........................................................................................135  
9. Explanatory Pluralism for Philosophy of Mind ................................137  
10. From Explanatory Pluralism to Pluralistic Physicalism..................143  
11. Pluralistic Physicalism and Causal Exclusion Worries...................153  
12. Dimensions of Explanatory Power..................................................161  
Conclusions and Directions for Further Research.................................169  
References ................................................................................................173  
Acknowledgements
The process that lead to this book started in 2003, when I began writing my
master's thesis on emergence in philosophy of mind. My supervisor at the
University of Helsinki, Sami Pihlström, introduced me to Achim Stephan's
work on emergence. After I had finished my studies in Helsinki, I visited
Achim in Osnabrück for the summer semester 2005, and he agreed to be
my PhD supervisor. During that summer he also introduced me to Bob
Richardson, who in turn introduced me to a whole new way of doing
philosophy, and helped me see the shortcomings of traditional analytic
philosophy of mind. That was when the basic idea of my PhD thesis began
to take shape: to criticize reduction in philosophy of mind from the point of
view of philosophy of science. In the end, Bob also became my second
PhD supervisor. These three people (Sami, Achim, and Bob) have
influenced my philosophical development probably more than they know,
and I am in deep gratitude for them. Achim was also my main supervisor
during the four years in Osnabrück, and I could not imagine a more
helpful, supporting, and kind "Doktorvater."
In addition, I would like to thank the following people (in no
particular order):

- Everybody at the Center for Research on Networked Learning and

Knowledge Building (University of Helsinki), in particular Kai
Hakkarainen, Liisa Ilomäki, and Sami Paavola, for providing an
unusual but very inspiring start for my scientific career, and for
letting me go to Osnabrück for the summer semester 2005, even
though I was supposed to be performing my non-military service
- Vera Hoffmann-Kolss for helping develop my views on causation
that play a crucial role in this book, for comments on several
presentations and drafts related to this thesis, and for taking me in as
a co-teacher for two inspiring seminars on philosophy of science
- Sven Walter for inviting me to be a co-author of an article on
reduction that substantially influenced parts of this book (Walter &
 Eronen 2011), and for kindly agreeing to be the third reviewer of my
PhD thesis
- Rafael Hüntelmann at Ontos Verlag for helping to prepare this book
for publication
- Brian McLaughlin for extending my philosophical understanding
with the several seminars and talks given at the University of
Osnabrück between 2006 and 2010, and for comments on earlier
drafts of Chapter 8
- Ilaria Serafini for the friendship, for all the crazy ideas, and for
making the years in Osnabrück surprisingly exciting and interesting
- Miriam Kyselo for countless philosophical discussions, an
unforgettable thesis-writing trip to Morocco, and for making me
smile and laugh, even when I tried to work
- The PhD program of the Institute of Cognitive Science, and
especially its coordinators Carla Umbach and Peter Bosch, for all
their help and support
- Other members of the PhD programme, particularly Rudi, Sascha,
Hartmut, Katya, and Mikko, for all the interdisciplinary input
- Dan Brooks for the conversations over beer, for having the same
philosophical vision as I do, and particularly for shedding light on
the details of Woodward's theory
- Jani Raerinne for very detailed and constructively critical comments
on several drafts of this book, since the very beginning
- Ulas Türkmen for the passionate philosophical dialogues (and
monologues) and the technical support
- The "Mäyränkaatajat" group in Helsinki (Antti, Heikki, Ilkka, Jani,
Janne, Jussi, Kari, Reino, Touko): these guys pull me down when I'm
riding too high, and lift me up when I've fallen too deep
- My brother Jussi for all the practical and brotherly support and
guidance
- My parents for always believing in me and supporting me in every
possible way
 - Kai Kuikkaniemi for the walks and talks that have helped me clarify
my ideas through the years
- Petri Savolainen for long discussions on the nature of existence,
science, and God that have also influenced the ideas in this book
- Sami Airaksinen, the second-greatest poet of Nilsiä, for being my
friend in arts and for helping develop my writing skills through the
years
- Antti-Jussi and Nina Pyykkönen, among other things for offering me
a job as an au pair in case I ran out of funding
- Konstantin Todorov for being my best friend during the years in
Osnabrück – keep with the snakes, stupid rocky!
- All the people who have assisted or supported me but who I forgot to
mention here (sorry for that)

This book is a revised and updated version of my PhD thesis, which I

defended in October 2010 at the University of Osnabrück. My PhD
research was financially supported by generous grants from Helsingin
Sanomain 100-vuotissäätiö (2006-2007), DAAD (2007-2009), and the
Finnish Cultural Foundation (2009-2010). In addition, the PhD program of
the University of Osnabrück provided funding for my numerous
conference and workshop trips between 2006 and 2010. I am very grateful
to these organizations for making this project possible.
While finishing this book, I was first a junior fellow at the Hanse-
Wissenschaftskolleg in Delmenhorst and then a postdoctoral researcher at
the Ruhr-Universität Bochum, in the group of Albert Newen. I thank these
institutes for their support and for allowing me to invest time in preparing
this book for publication.
Finally, I would like to thank Laura Bringmann for very useful
comments on different versions of this book, most importantly the
penultimate version, and for keeping me sane during the busy months of
finishing my thesis and this book. In addition, thank you for making me
happy. Nobody does it quite the way you do, nobody does it better.
 7

Introduction
The idea of reduction has surfaced in different forms throughout the
history of science and philosophy. Thales took water to be the fundamental
principle of all things; Leucippus and Democritus argued that everything is
composed of small, indivisible atoms; Galileo and Newton tried to explain
all motion with a few basic laws; 17th century mechanism conceived of
everything in terms of the motions and collisions of particles of matter;
British Empiricism held that all knowledge is derived from experiential
knowledge; current physicists are searching for the TOE, the “Theory Of
Everything,” that would unify the electromagnetic and the weak and strong
nuclear forces with gravity. In a broad sense, all of these projects can be
understood as (attempted) reductions, as they aim at revealing some kind
of unity or simplicity behind the appearance of plurality or complexity. In
philosophy of mind, reduction has figured prominently in the issue of the
relation between the mind and the brain: Does the mind reduce to the
brain? Do mental explanations reduce to neuroscientific explanations?
Does psychology as a science reduce to neuroscience? And so on.
But what exactly is “reduction”? Traditionally, it has been understood
as the derivation of a theory to be reduced from a more fundamental
theory. However, it is now widely accepted in philosophy of science that
this traditional view fails to characterize actual scientific practice, or actual
relations between sciences, at least when it comes to psychology and
neuroscience. In philosophy of mind, reduction is commonly conceived as
“functional reduction,” where reduction consists in defining a property1
functionally and then finding the physical realizers that perform this
function, but this model hardly fits scientific practice any better than the
traditional model, and is plagued with philosophical problems.

1
Often it would be more natural to talk of mental capacities or functions or processes,
but following the venerable tradition in philosophy of mind, I mainly talk about mental
“properties” in this book (without assuming any particular metaphysical theory of
properties). In some contexts I use the term “state” instead of “property,” but this
subtle difference has no relevance for the arguments. I also talk about “mental” and
“psychological” properties interchangeably and make no distinction between them.
8

In this book, I draw from recent developments in philosophy of

science, and explore their consequences for the debates on reduction in
philosophy of mind. I elaborate a pluralistic account of reduction, and
show how and why more strongly reductionistic approaches fail. A
pluralistic account of reduction might sound strange and contradictory.
Aren’t reduction and pluralism mutually exclusive? What I hope to show
in this thesis is that the answer is no. The kind of pluralism defended here
is compatible with certain kinds of reductions or reductive explanations.
And I argue that, in fact, there are no reductions to be expected in any
stronger sense.
This thesis is primarily intended as a contribution to the philosophy of
mind and cognitive science, and what I am focusing on is the purported
reduction of psychology (understood as an empirical science, not “folk
psychology”) to neuroscience.2 The positions and arguments defended in
this thesis do not necessarily apply to relations between other sciences,
although I am happy if they do. The main target of my criticism is
traditional analytic philosophy of mind, which has been largely guided by
conceptual analysis and formal methods instead of actual science. If
philosophy of mind is brought closer to actual science, it can also be more
relevant to scientific endeavors of understanding the mind and
consciousness.3 One of the most prominent proponents of the traditional
analytic philosophy of mind is Jaegwon Kim, who receives the most
attention in this thesis, partly because I am more familiar with his work
2
With “psychology” I mean the empirical study of human behavior and the mind, and
with “neuroscience” the empirical study of the human nervous system. Of course, this
distinction is becoming increasingly blurry, and is to some extent conventional. I make
the distinction mainly for the sake of continuity with the traditions in philosophy of
mind and philosophy of science, and it is in no way essential for the position defended
in this thesis: if pluralism is the right approach, it is right regardless of whether or not
there is a clear distinction between psychology and neuroscience.
3
A distinction is sometimes made between neurophilosophers and philosophers of
neuroscience. Neurophilosophers (e.g., Patricia Churchland, John Bickle) apply
findings from neuroscience to traditional philosophical problems, such as free will or
consciousness. Philosophers of neuroscience (e.g., William Bechtel, Carl Craver)
consider traditional problems of philosophy of science with regard to neuroscience.
My approach differs from both of these and is somewhere in between. I apply results
and insights from philosophy of neuroscience (and philosophy of science in general) to
address traditional problems in philosophy of mind.
 9

than that of other philosophers of the same tradition (for example, Ned
Block, David Chalmers, Frank Jackson, or Joseph Levine). Although I am
criticizing Kim, it is beyond doubt that his contributions to philosophy of
mind have been groundbreaking. I greatly admire him for the clarity and
beauty of his philosophy and few philosophers have influenced my
intellectual development as much as he has.
While I was already halfway through writing this book, I came across
an excellent recent work with aims strikingly similar to mine: Steven
Horst’s (2007) Beyond Reduction. Horst is also arguing against
reductionism, defending pluralism, and emphasizing the importance of
bringing philosophy of mind closer to philosophy of science. Fortunately,
there are also substantial differences in our arguments and conclusions. In
contrast to Horst, mechanistic explanation and the interventionist account
of causation play a key role in my arguments, and the “cognitive
pluralism” of Horst is more far-reaching and radical than the pluralistic
physicalism I am defending. Furthermore, Horst does not discuss the
functional model of reduction, which receives a lot of attention in this
thesis. On the other hand, he goes far deeper into the details of some other
debates in the philosophy of mind, most importantly the debates on
supervenience and the “explanatory gap.” Therefore, although the spirit of
Horst’s book is very close to that of this one, the two are considerably
different and complementary contributions to philosophy of mind.
The structure of this thesis is as follows. In Part I, I will discuss
reduction and reductionism in philosophy of science, focusing on
psychology and neuroscience. I will go through the problems of the classic
intertheoretic models of reduction and the more recent “ruthless” approach
to reductionism, and defend a position consisting of two main elements:
mechanistic explanation and the interventionist account of causation. This
leads to explanatory pluralism regarding psychology and neuroscience. In
the end of the part, I will also consider the issue of levels and its relation to
reduction.
In Part II, I will criticize the way reduction has been understood in
philosophy of mind, based on what has been presented in Part I. I will go
through classical topics like multiple realizability, functionalism, the
explanatory gap, and nonreductive physicalism, and show how our
10

understanding of them is changed once we have a proper picture of

reduction. An extensive and detailed section is devoted to criticizing the
functional model of reduction, which has become something like a
standard model in philosophy of mind.
In Part III, I will present and defend a new framework for philosophy
of mind. Its main elements are explanatory pluralism, mechanistic
explanation, and the interventionist account of causation. I will also
develop an ontological framework for this position, which consists of a
kind of ontological pluralism based on the idea of robustness.
Subsequently, I will show that the causal exclusion argument does not
make this position incoherent, and that the position is compatible with
certain forms of physicalism, to the extent that it could be called pluralistic
physicalism. In the end, I will argue that many reductionist ideas fit
perfectly into this pluralistic framework, including for example the thesis
that all mental properties can be mechanistically explained.
 11

PART I:
Reduction in Philosophy of Science
12
 13

Introduction
In this part, I will discuss reduction4 as it has been understood in the
philosophy of science of the 20th (and 21st) century. Going through the
history (or prehistory) of reductionist ideas would be interesting, but this
thesis is not a historical one, and therefore I will only discuss the models
that are most relevant to contemporary debates. I will begin with the
development of intertheoretic models of reduction that started in the 1950s,
in the afterglow of logical positivism, and then go on to discuss more
recent accounts of reduction, most importantly “New Wave Reductionism”
and “Ruthless Reductionism.” I will argue that these approaches face fatal
problems, at least in the case of psychology and neuroscience, and that
“mechanistic explanation,” especially when supplemented with the
interventionist account of causation, provides a more accurate and
scientifically credible framework for approaching issues of reduction. In
the end, I will consider the question of levels and its relation to reduction,
focusing on the problems in current accounts of levels.

4
Throughout this thesis, I use the term “reduction” to refer to a single case of
accomplished or purported reduction: the reduction of thermodynamics to statistical
mechanics, the reduction of chemistry to physics, and so on. “Reductionism” refers to
a broader thesis, according to which reductions are to be expected (a predictive claim)
and/or desirable (a normative claim). Of course, different models of reduction yield
different reductionisms, and one can be reductionist regarding some domains of
science but not others. Therefore, for instance, “psychoneural Nagel reductionism”
means the thesis that psychology will be or should be reduced to neuroscience
following Nagel’s model of reduction.
14
 15

1. Reduction: From Derivations of Theories to Ruthless

Metascience
By far the most influential philosophical models of reduction have been the
“intertheoretic” models, where reduction is seen as a relation between
formal theories. The development of intertheoretic models started in the
middle of the 20th century, drawing on the spirit of logical positivism. The
ultimate goal was to show how unity of science could be attained through
reductions. John Kemeny and Paul Oppenheim (Kemeny & Oppenheim
1956) formulated reduction as a relation between theories, where the
reducing theory should be able to explain any observational data that the
reduced theory explains, and the reducing theory should be at least as well
systematized5 as the reduced theory. A few years later, Oppenheim and
Putnam published their extremely influential ”Unity of Science as a
Working Hypothesis” (1958), where they presented the hypothesis that all
sciences will be reduced to the fundamental physical science via
”microreductions.” In a microreduction, the higher-level entities to be
reduced must be fully decomposable into the reducing entities of lower
levels. Oppenheim and Putnam also adopted the conditions for reduction
stated by Kemeny and Oppenheim (1956). That is, according to
Oppenheim and Putnam, a theory T2 microreduces to theory T1 if and only
if (1) any observational data explainable by T2 are explainable by T1, (2) T1
is at least as well systematized as T2, and (3) all the entities referred to in
T2 are wholes which are fully decomposable into entities in the universe of
discourse of T1. This is in effect a model of replacement, since the
successful microreduction makes T2 entirely dispensable. This account
suffers from serious defects that I will only briefly mention here (see, e.g.,
Sklar 1967 for more details): it assumes that we can clearly distinguish
between observational and non-observational terms, the notion of
systematization or systematic power is not clearly defined, and it is hard to
find examples from history of science that would satisfy the requirements.

5
A theory is well systematized if it is simple but predicts or explains a broad range of
phenomena. That is, systematization or systemic power is a measure that combines
simplicity and strength. Kemeny and Oppenheim acknowledge the need for a more
precise definition, but do not give one in the paper.
16

Also in Nagel’s (1951; 1961, 336-397) classic account of reduction,

many ideas of logical positivism are clearly visible. Reduction is seen as a
relation between formal theories, such that the theory to be reduced (T2) is
logically derived from a more fundamental theory (T1). Conditions for a
successful reduction are that (1) we can connect the terms of T2 with the
terms T1, and that (2) with the help of these connecting assumptions we
can derive all the laws of T2 from T1. In Nagel’s model, a reduction can be
seen as a kind of deductive-nomological explanation, where T1 explains T2.
Nagel distinguished between two different kinds of reductions:
“homogeneous” and “heterogeneous” reductions. In a homogeneous
reduction the two theories share the same conceptual apparatus. For
example, the reduction of Galileo’s laws to Newtonian mechanics was a
homogeneous reduction. However, most (interesting) cases of reduction
are heterogeneous reductions, where one of the theories has concepts not
found in the other. In these cases, in order to satisfy the two conditions for
reduction, we need some principles or laws that connect the terms of the
two theories. The exact nature of the connecting principles, or “bridge
principles/laws” as they came to be called, was left open by Nagel, and has
been a matter of much debate. Although the conditions of a Nagel-type
reduction can be fulfilled already when these laws express material
conditionals of the form “∀x (FT1x  FT2x)” (e.g., Richardson 1979), it was
widely accepted that biconditionals of the form “∀x (FT1x ≡ FT2x)” are
necessary for the ontological simplifications that were considered to be one
of the main goals of reduction.
Nagel presented the reduction of thermodynamics to statistical
mechanics as a paradigmatic example of a successful scientific reduction.
He focused on the derivation of the Boyle-Charles’ law for ideal gases (pV
= kT, where p is the pressure of the gas, V is the volume of the gas, T is the
absolute temperature of the gas, and k is a constant) from statistical
mechanics, pointing out that the derivation of the whole thermodynamics
would be immensely complicated, and that even for the derivation of the
Boyle-Charles’ law many idealizing assumptions have to be made: one has
to assume, for example, that the gas is composed of a large number of
perfectly elastic spherical molecules with equal masses and volumes but
with dimensions that are negligible compared to the distances between the
 17

molecules, and that the molecules are in constant motion and subject only
to forces of impact between themselves and the perfectly elastic walls of
the container. Nagel argued that the Boyle-Charles’ law is a logical
consequence of the principles of mechanics, when they are supplemented
with certain idealizing assumptions and connecting principles (bridge
principles), and that this is a representative example of the reductive
relation between thermodynamics and statistical mechanics.
Nagel’s model of reduction is neat and precise, but unfortunately fails
to account for many cases that are regarded as reductions. The model is too
demanding: it is very hard to find a pair of theories that would meet these
requirements. Even Nagel’s prime example, the reduction of
thermodynamics to statistical mechanics, is much more complicated than
Nagel thought (see, e.g., Sklar 1999; Richardson 2007). No one ever came
up with the derivation of the whole thermodynamics from statistical
mechanics, and it is likely to be computationally intractable. The
possibility of such a derivation is further diminished by the fact that many
central thermodynamical concepts, like entropy, are associated with a wide
variety of distinct concepts in statistical mechanics which do not exactly
correspond to thermodynamic entropy, neither separately nor taken
together. Historically speaking, scientists were not even aiming at such a
derivation – the goal was to show that thermodynamics is consistent with
statistical mechanics and to situate it in a broader Newtonian framework
(Richardson 2007).
Nagel’s model also has problems accounting for the fact that the
reducing theory often corrects the theory to be reduced, which entails that
the original theory was strictly speaking false. For example, Newtonian
physics showed that some principles of Galilean physics, such as the
assumption that uniformly accelerated gravitational free-fall is the
fundamental law of motion, were false. However, since logical deduction
is truth-preserving, the new reducing theory cannot both be true and
logically entail a false theory.
Problems like these led Paul Feyerabend (1962) to argue that no
formal accounts of scientific reduction are possible or necessary. The
majority of philosophers, however, responded by developing more
sophisticated models (Causey 1977; Schaffner 1967), culminating in what
18

came to be known as, using John Bickle’s (1998) term, “New Wave
Reductionism” (Bickle 1998; 2003; Hooker 1981; see also P. M.
Churchland 1985; P. S. Churchland 1986; Schaffner 1993). Here I will
focus on Bickle’s model of New Wave Reductionism, since it is the most
elaborate and explicit one.
Like its precursors, the New Wave model is a model with universal
scope that takes reduction to be a relation involving logical derivations
between theories. However, the crucial difference is that what is deduced
from the T1 is not the theory to be reduced itself (T2), but an analogue (or
“equipotent image”) of it (T2a). The fate of theory T2 and its ontological
posits is determined by the relation between T2 and the analogue T2a.
Importantly, the analogue can be formulated entirely in the vocabulary of
theory T1 – no bridge principles are needed to connect T1 and T2a. If the
analogy between T2 and T2a is strong and not much correction is needed,
T2 is reduced “smoothly” to T1, and many of its ontological posits can be
retained. If the theories are only weakly analogical and the amount of
correction implied to T2 is considerably large, the reduction is “bumpy,”
and many or all of the ontological posits of T2 will be eliminated. Thus,
depending on the strength of the analogy, each case of intertheoretic
reduction falls at a certain point on a continuum (Figure 1), where at one
extreme we have extremely smooth reductions resembling Nagel-
reductions, and at the other extreme complete replacement of the old
theory and its ontological posits. In this way, the New Wave model can
accommodate the idea that scientific progress sometimes involves
“revolutions” and replacements of old theories and their ontologies
(Feyerabend 1962; Kuhn 1962).
Bickle (1998) has also specified analytical tools for evaluating the
relation between T1 and T2, based on the structuralist/semantic view of
theories. His central example is a familiar one: the reduction of
thermodynamics to statistical mechanics (Bickle 1998, 33-40). Bickle
shows that we can derive an analogue structure of the ideal gas law of
thermodynamics from statistical mechanics and the kinetic theory. This
analogue structure exactly mimics the ideal gas law while containing only
terms of statistical mechanics and microphysics.
 19

Strong analogy Weak analogy

------------------------------------------------------------------------

Theory: Retention Replacement

Ontology: Identity Revision Elimination

Figure 1: The New Wave continuum (based on Bickle 1998, 30). The
strength of the analogy between T2a (deduced from T1) and T2
determines the fate of theory T2 and the ontological consequences.

One of Bickle’s assumptions that he shares with Feyerabend (1962)

and the Churchlands (P. M. Churchland 1981; 1985; P. S. Churchland
1986) is that our intuitions about the nature of the mental can be construed
as relying on a primitive “folk” theory. This includes the intuitions
(zombies, the explanatory gap, qualia inversion, etc.) that seemingly show
that the mental domain must be distinct from the physical domain.
Consequently, one of Bickle’s (1998) central claims is that we can
reformulate the mind-body problem as a problem of theory reduction (the
intertheoretic-reduction reformulation or IR-reformulation). That is, when
we have seen whether psychological theories reduce to theories of
neuroscience, we know how mental properties and neural properties are
ontologically related, and there is no further mind-body problem to solve.
Ontological questions are secondary to and dependent on questions of
intertheoretic reduction. Thus, we can replace the “murky” notion of
ontological reduction with a scientifically grounded and well-studied
notion of intertheoretic reduction.
Bickle’s prediction is that folk psychological mental concepts will be
reduced to neuroscientific theories in a “bumpy” way, such that many or
most of the ontological posits of the psychological theories have to be
20

abandoned or revised. This leads to what he calls “revisionary

physicalism:” there will be no intertheoretic identities between folk
psychological properties and neuroscientific properties, but folk
psychology will not be eliminated in the way theories of phlogiston or
caloric heat were eliminated – there will be neuroscientific concepts that
resemble folk psychological concepts or play roughly the same role.
One problem with Bickle’s account is that it gives us only a
relativistic solution to the mind-body problem (Stephan 2001). There are
many different neuroscientific theories out there. To which of these should
we compare the psychological theories? Our psychological theories might
be in need of revision relative to some new neuroscientific theories. But
should we revise our ontologies and abandon talk of beliefs, desires, and so
on, just because they cannot be smoothly reduced to some neuroscientific
theories? Perhaps they could be smoothly reduced to some other
neuroscientific theories, for example ones that will be developed later. On
the other hand, if the claim is that we have to compare the psychological
theories to some completed neuroscience of the future, the problem is that
we do not know what this future neuroscience will look like, and how
different it will be from our current psychological theories.
Furthermore, two assumptions that the New Wave model shares with
the traditional model lead to fatal problems in the case of psychoneural
reduction (Wimsatt 1976a; 1976b; McCauley 1996; 2007b). The first
problematic assumption is that a single model of reduction can account for
all putative cases of reduction.6 Because of this assumption, the New Wave
model is blind to certain fundamental differences in intertheoretic relations.
Most importantly, it fails to account for the intralevel-interlevel distinction.
Intralevel or successional relations hold between competing theories within
a particular science, operating at a single level of analysis, for example
between Newtonian physics and General Relativity theory. Interlevel
relations are relations between theories that reign at the same time at
different analytical levels, for example between cognitive psychology and
cellular neuroscience. All of the examples of eliminative (or “bumpy”)

6
Nagel already distinguished between heterogeneous and homogeneous reductions,
but probably the first one to emphasize that there are also different types of
heterogeneous reductions was Thomas Nickles (1973).
 21

reduction that New Wave reductionists present are intralevel cases, and
give no reason to expect eliminative reductions in interlevel contexts
(McCauley 1996; 2007b). In particular, they provide no support for New
Wave reductionists’ claims that psychology will be reduced to
neuroscience, since psychology and neuroscience are at different levels.
Replacements and eliminations happen across time between
competing theories at the same analytical level, not in interlevel contexts.
A look at scientific practice shows that scientists are not interested in
eliminating sciences at adjacent levels, but look there for support,
guidance, evidence, and so on. Incompatibilities between sciences do not
lead to eliminations, but to further inquiries, ”co-evolution” of theories
(Hooker 1981; Wimsatt 1976a; 2007) and possibly to the development of
interlevel theories (Darden & Maull 1977).
The second problematic assumption of the New Wave model,
inherited from Nagel’s account, is that the relata of reductions are
exclusively theories that are construable in some formal or semi-formal
way, either as sets of sentences (the “received view” of theories), or as sets
of models meeting certain set-theoretic conditions (the
structuralist/semantic view of theories). However, some generally
accepted cases of scientific reduction — for example, the reduction of
genetics to molecular biology — do not seem to involve such formal
theories (Sarkar 1992). In general, well-structured theories that could be
handled with logical tools are rare and peripheral in the special sciences,
including psychology and neuroscience. There are of course “theories” in a
broad sense of the term in psychology and neuroscience, like the LTP
theory of memory consolidation or the global workspace theory, but these
are not formal theories, and can hardly be the starting points or results of
deductions. Reductions and reductive explanations in psychology and
neuroscience cannot be conceived as logical derivations. Instead, these
disciplines typically look for descriptions of mechanisms that can serve as
explanations for patterns, effects, capacities or phenomena (see next
chapter).
An intertheoretic reductionist could still claim that even though
neuroscientists and psychologists do not present their theories in a formal
way, they could be formalized. For example, Kenneth Schaffner (1993) has
22

presented a qualified defense of formal approaches to reduction in

biology.7 However, the formalization of theories of neuroscience or
psychology would undoubtedly be extremely complicated and likely even
impossible without loss of explanatory power. In any case it would abstract
the analysis far away from the actual scientific practice, and make the issue
of reduction rather irrelevant for science. For these reasons, looking at the
relations between formal theories is a wrong starting point, at least in the
case of psychology and neuroscience.
At least partly for these reasons, John Bickle, who has been the most
ardent advocate of New Wave Reductionism, has taken some distance
from the intertheoretic models of reduction and now emphasizes looking at
the “reduction-in-practice” in current neuroscience (Bickle 2003; 2006a;
2006b). He calls this approach “metascientific reductionism” to distinguish
it from philosophically motivated models of reduction that are typically
applied in philosophy of mind. The idea is that instead of imposing
philosophical intuitions on what reduction has to be, we should examine
scientific case studies to understand reduction. We should look at
experimental practices of an admittedly reductionistic field, characterized
as such by its practitioners and other scientists.
According to Bickle, molecular and cellular cognition – the study of
the molecular and cellular basis of cognitive functions – provides just the
right example. “A ruthlessly reductive methodology prevails in the
molecular and cellular cognition. The approach intervenes into cellular or
intracellular molecular pathways and then tracks these interventions in the
behaving animal, using protocols borrowed from experimental
psychology” (Bickle 2006b, 134).
This reductionist methodology of molecular and cellular cognition has
two parts: (1) intervene causally into cellular or molecular pathways, (2)
track statistically significant differences in the behavior of the animals
(Figure 2). When this strategy is successful and a mind-to-molecules
linkage has been forged, a reduction has been established.

7
On the other hand, Schaffner (particularly Schaffner 1974) has also argued for the
“peripherality of reductionism” in biology. What he means by this, roughly speaking,
is that reduction is not the explicit goal of research in biology, and that the
methodology of biological research is not generally speaking reductionistic.
 23

A radical feature of this model is that it allows reductions to jump

straight from observable behavior to the molecular level, skipping the
functional, computational, etc., levels in between. The cellular and
molecular mechanisms directly explain the behavioral data and set aside
intervening explanatory levels (2006a, 426). This is in stark contrast to the
classic “layer-cake” model (e.g., Oppenheim & Putnam 1958), where
reduction proceeds step-by-step, always between adjacent levels. Bickle’s
prime example is the case of LTP as the mechanism of memory
consolidation, which according to him is an example of an accomplished
mind-to-molecules reduction (more on this in the next chapter).
A central claim of this “reduction-in-practice” is that when lower-
level explanations are completed, the higher-level explanations become
merely heuristic: ”psychological explanations lose their initial status as
causally-mechanistically explanatory vis-á-vis an accomplished ...
cellular/molecular explanation” (Bickle 2003, 110). Psychology is needed
for describing behavior, formulating hypotheses, designing experimental
setups, and so on, but according to Bickle, these are just heuristic tasks,
and when cellular/molecular explanations are completed, there is nothing
left for higher-level investigations to explain.
Metascientific reductionism does not depend on any intertheoretic
model of reduction, and thus is not threatened by problems discussed
above. However, metascientific reductionism has its own share of
problems, as I will show in the next chapter.
24

Figure 2: The reductive methodology in Bickle’s metascientific

reductionism (based on Bickle 2006a, 426). Researchers make
interventions at the cellular and molecular levels and track the
behavior of the animal. The dashed arrows represent levels of
experimental intervention, the solid arrow represents the level at
which these interventions are measured. The role of psychology is
merely descriptive and heuristic.
 25

2. Mechanistic Explanation
Mostly due to the reasons outlined in the previous chapter, theory
reduction is nowadays not considered to be the norm in the special
sciences. What has become something like the new received view on the
nature of interlevel and intertheoretic relations is rather what is known as
“mechanistic explanation” (Bechtel 2008; Bechtel & Richardson 1993;
Craver 2007; Glennan 1996; Machamer et al. 2000; Wright & Bechtel
2007). The basic insight of this approach is that if one takes into account
actual scientific practice in neuroscience and many of the life sciences, it
turns out that instead of focusing on laws or formalizable theories,
practicing scientists formulate explanations in terms of mechanisms.
The most important historical figure who defended mechanistic
explanation is probably Descartes, who claimed that the bodies of humans
and animals work like machines. Later C. D. Broad (1925) developed an
early theory of (reductive) explanation in mechanistic terms. An important
forerunner of models of mechanistic explanation is also Cummins’ (1983)
model of functional analysis in psychology (see Craver (2007, Ch. 4) for
an account of the differences and similarities between functional analysis
and mechanistic explanation).
According to an often-cited definition, mechanisms are to be
understood as ”entities and activities organized such that they are
productive of regular changes from start or set-up to finish or termination
conditions” (Machamer et al. 2000, 3). Or, as Bechtel (2008, 13) puts it, a
“mechanism is a structure performing a function in virtue of its component
parts, component operations, and their organization.” A mechanistic
explanation then describes how the orchestrated functioning of the
mechanism is responsible for the phenomenon to be explained. According
to Wright and Bechtel (2007), various conceptions of mechanisms share
the idea that mechanisms are composite hierarchical systems that are
composed of component parts and their properties. Each component part
performs some operation and interacts with other parts of the mechanism,
and all of this together results in the overall systemic activity of the
mechanism.
26

A mechanistic explanation describes how the mechanism accounts for

the explanandum phenomenon, the overall systemic activity (or process or
function) to be explained. Mechanistic explanations are constitutive
explanations: they describe how the behavior or phenomenon to be
explained is constituted by underlying causal mechanisms.
To give some rough examples, the propagation of action potentials is
explained by describing the cellular and molecular mechanisms involving
voltage-gated sodium channels, myelin sheaths, etc. The pain withdrawal
effect is explained by describing how nerves transmit the signal to the
spinal chord, which in turn initiates a signal that causes muscle contraction.
The metabolism of lactose in the bacterium E. coli is explained by
describing the genetic regulatory mechanism of the lac operon.
A paradigmatic example of mechanistic explanation that I will present
in more detail here is the case of LTP and memory consolidation (Craver
2002; 2007; Bickle 2003; 2006a). In this example, the overall systemic
activity to be explained is memory consolidation, the transformation of
short-term memories into long-term ones. A mechanistic explanation then
describes how the relevant parts and their activities result in the overall
activity. Central to this explanation is long term potentiation (LTP), a well-
studied cellular and molecular phenomenon that exhibits features that
make it very likely the central part of the memory consolidation
mechanism.
I will briefly summarize the neuroscientific explanation of the LTP
mechanism here. In the hippocampus, which is the brain structure most
closely associated with memory consolidation and where LTP is most
often studied, the synapses where LTP occur use glutamate as a
neurotransmitter. NMDA receptors are a specific type of postsynaptic
glutamate receptors. When glutamate binds into these receptors, they
change shape and expose a pore in the cell membrane. If the postsynaptic
cell is inactive, these pores are blocked by Mg2+ ions. However, if the
postsynaptic cell is depolarized, the Mg2+ ions are detached, and Ca2+ ions
diffuse through the channels into the cell. This triggers a cascade of effects
that leads to the phosphorylation (the addition of a phosphate (PO4) group)
of another type of glutamate receptors, AMPA receptors, which results in
an increase in the efficiency of synaptic transmission. This is the early
 27

phase of LTP (E-LTP). The late phase (L-LTP), which is more long-
lasting, involves changes in the gene expression and protein synthesis,
which in turn leads to changes in the dendritic spines, resulting in long-
term increase in the efficiency of synaptic transmission.
One of the core features of mechanistic explanations is that they are
multilevel: focusing on just one level is not sufficient for full understanding
of the phenomenon. Craver (2002; 2007, 165-170) identifies four levels in
the case of spatial memory and LTP: the behavioral-organismic level,
which involves various types of memory and learning, the conditions for
memory consolidation and retrieval, and other phenomena that are
typically investigated by behavioral and psychological tasks. The
computational-hippocampal level involves the structural features of the
hippocampus and its overall role in the mechanisms of memory, its
connections to other brain regions and the computational processes it is
thought to perform. The electrical-synaptic level includes neurons,
synapses, dendritic spines, axons, action potentials and so on. At the
bottom of this hierarchy is the molecular-kinetic level, where we find
glutamate, NMDA and AMPA receptors, Ca2+ ions, and Mg2+ ions, which
bind to each other, break, phosphorylate, and so on. Craver calls these
“mechanistic levels” or “levels of mechanisms.” They are levels of
composition, where the relata are behaving mechanisms at higher levels
and their components at lower levels. These levels are local and case-
specific, not universal divisions of nature or science. I will discuss the
question of levels in more detail in Chapter 4.
Mechanistic explanations have both a “downward-looking” and an
“upward-looking” aspect (in addition to the obvious “horizontal” or same-
level aspect). In the LTP case, one is looking upward when, in order to
understand the computational properties of the hippocampus, one is
studying its role in the overall cognitive system, or when, in order to
understand the role of the molecular processes of LTP, one is looking at
the larger computational-hippocampal framework. In contrast, one is
looking downward when memory consolidation is explained by appeal to
the computational processes at the hippocampal level, or when the synaptic
LTP mechanism is explained by appeal to activities at the molecular-
kinetic level.
28

Several philosophers have argued that the process of “looking

downward” and invoking parts of the mechanism to understand its
behavior as a whole is close enough to what scientists generally take to be
a reductive explanation to warrant treating the downward-looking aspect of
mechanistic explanation as a kind of reductive explanation (Bechtel 2007;
2008; Sarkar 1992; Wimsatt 1976a; 2000; 2007; Richardson & Stephan
2009). For instance, Wimsatt writes: “A reductive explanation of a
behavior or a property of a system is one showing it to be mechanistically
explicable in terms of the properties of and interactions among the parts of
the system” (Wimsatt 2000, 288). Carl Gillett (2007) has even argued that
successful mechanistic explanations entail ontological reductions. On the
other hand, Craver (2005; 2007) considers the framework of mechanistic
explanation to support anti-reductionism, since it differs so much from
traditional approaches to reduction and allows for the causal and
explanatory relevance of nonfundamental things.
It is true that mechanistic explanation differs fundamentally from
intertheoretic reduction and is less “reductive,” at least in some pre-
analytic sense of the term. Nevertheless, in my view, seeing downward-
looking mechanistic explanation as reductive explanation is unproblematic,
as long as one keeps it clearly distinct from intertheoretic reduction and
stronger forms of reductive explanation (like Bickle’s “ruthless” reduction
or Kim’s functional reduction).
Mechanistic explanation as a model of scientific explanation is of
course a radical departure from the classic (but thoroughly problematic)
deductive-nomological (DN) model of explanation (Hempel and
Oppenheim 1948). In the DN-model, an explanation consists in the
derivation of the explanandum from general laws and statements
describing the situation. That is, the explanandum is shown to be a logical
consequence of the explanans. In mechanistic explanations, it is not
necessary to invoke general laws, and logical derivations play hardly any
role. Furthermore, mechanistic explanation is not intended as a universal,
all-encompassing account of scientific explanation.
Mechanistic explanation essentially involves causal explanation: the
overall activity to be explained is defined in causal terms, as well as the
activities of the components. The account of causation and causal
 29

explanation that the mechanists more or less explicitly endorse (Craver

(2007) does it quite explicitly) is the interventionist account, which I will
present in the next chapter. But before turning to causation, I will first
consider how mechanistic explanation relates to Bickle’s metascientific
reductionism.
First of all, mechanistic explanation and metascientific reductionism
share many background assumptions. Both approaches acknowledge the
insufficiency of intertheoretic models of reduction. Both are based on
actual scientific practice in neuroscience. Both abstain from traditional
metaphysical considerations and philosophically-driven analyses of
reduction. Both appeal to the case of LTP and memory consolidation for
support. However, the conclusions Bickle and Craver draw are completely
different: for Bickle it is ”ruthless reductionism,” for Craver multilevel
mechanistic explanations and antireductionism. Here I will briefly explain
how differently these authors analyze the paradigm case of LTP and
memory consolidation, and argue that Craver has a stronger case, largely
due to causal considerations.
According to Bickle, the case of LTP and memory consolidation is a
paradigm example of an accomplished psychoneural reduction (Bickle
2003, Ch. 2; 2006a). He describes the current cellular and molecular
models of LTP in detail, and argues that they are the mechanisms of
memory consolidation. Furthermore, he argues that these mechanisms
(when they are fully understood) explain memory consolidation directly,
setting aside psychological, cognitive-neuroscientific, etc. levels. This is an
example of the ”intervene cellular/molecularly, track behaviorally”
methodology, and in Bickle’s view a successful reduction.
What makes Bickle’s analysis ”ruthlessly” reductive is the claim that
”psychological explanations lose their initial status as causally-
mechanistically explanatory vis-á-vis an accomplished (and not just
anticipated) cellular/molecular explanation” (Bickle 2003, 110). He argues
that scientists stop evoking and developing psychological causal
explanations once ”real neurobiological explanations are on offer,” and
”accomplished lower-level mechanistic explanations absolve us of the need
in science to talk causally or investigate further at higher levels, at least in
any robust ’autononomous’ sense” (Bickle 2003, 111). He claims that
30

psychological research into the mechanisms of memory consolidation hit

an explanatory wall throughout the middle of the last century and
neurobiology took over, and that as cellular/molecular accounts of memory
consolidation emerge, the empirical search for causal explanations at the
psychological level disappears. Bickle shrugs off the fact that many
psychological explanations seem causally-mechanistically explanatory at
the present time by pointing out that the cellular and molecular accounts
are still incomplete and insufficient.
Bickle concedes that even after cellular and molecular accounts are
complete psychological causal explanations may still play a role in
research. However, this role is merely heuristic, for example in generating
and testing neurobiological hypotheses. Psychological explanations are
useful for heuristic purposes because they are approximations of the best
causal-mechanical explanations. This is all that psychological causal
explanations are needed for, and after they have done their job, they can be
kicked away like Wittgenstein’s ladder (Bickle 2003, 130).
Craver’s analysis is quite different (2007, 233-245). First of all, he
looks at the historical development of the LTP explanation, arguing that it
is not a case of downward-looking search for a memory mechanism in the
hippocampus. He points out that hippocampal synaptic plasticity was not
discovered in a top-down, reductive search for the neural correlate of
memory – rather, it was noticed in an intralevel (and interdisciplinary)
research project in which anatomical and electrophysiological perspectives
were integrated. Furthermore, the discoverers of LTP did not have
reductive aspirations – they saw LTP as a component in a multi-level
mechanism of memory. After the discovery of LTP in 1973, there has been
research both up and down in the hierarchy. Craver claims that the memory
research program has implicitly abandoned reduction as an explanatory
goal in favor of the search for multilevel mechanisms. His conclusion is
that ”the LTP research program is a clear historical counterexample to
those ... who present reduction as a general empirical hypothesis about
trends in science” (Craver 2007, 243).
I have already described above how Craver analyses LTP as a
multilevel mechanism. According to him, successful neuroscientific
explanations span multiple levels, and thus neuroscience itself provides no
 31

support for Bickle’s views. Craver argues for causal and explanatory
relevance of nonfundamental things. That is, he argues that there is no
fundamental level of explanation, and that entities of higher levels can
have causal and explanatory relevance, regardless of how successful and
accomplished lower-level explanations are.
This is of course in sharp contrast to Bickle’s view. Bickle makes a
distinction between ”real” and ”merely heuristic” explanation and claims
that only cellular/molecular explanations are real explanations. However, it
is not clear what this distinction is based on. Why are cellular/molecular
explanations real causal explanations and psychological explanations
merely heuristic? Why do psychological explanations become “merely
heuristic” explanations when cellular/molecular explanations are in place?
Bickle does not give satisfying answers to these questions. He cites
scientific cases that show that we can go straight to the molecular level and
make interventions that cause observable changes in behavior, but these
cases do not show that higher-level explanations are not good explanations.
In accordance to the metascientific attitude, the justification for
Bickle’s claims comes only from scientific practice, not from philosophical
considerations. This is not a problem in itself. The problem is that many
prominent philosophers of neuroscience (including Craver and the other
mechanists) have quite different views of the nature of neuroscientific
practice. Finding some cases that support ruthless reductionism is not
enough if the claim is about the general nature of neuroscientific practice,
and even the case that Bickle has picked out as the central example can be
interpreted in different ways, as we have seen above. Accordingly, Bickle
(personal communication) has recently backed up a little and now accepts
that it is possible that ruthless reductionism characterizes research only in
certain fields of neuroscience (such as cellular and molecular cognition),
while research in other fields may be appropriately characterized as search
for multilevel mechanisms.
A related problem is that Bickle does not provide an account of
causation or causal explanation that would support his conclusions.
Arguing convincingly that psychological explanations lose their status as
causally explanatory and that cellular/molecular explanations are the real
causal explanations would require at least some rough account of what
32

causation or causal explanation amounts to. Scientific practice alone does

not tell us what causal explanation is.
On the other hand, Craver’s defense of the causal and explanatory
relevance of nonfundamental things is explicitly based on the
interventionist account of causation and causal explanation, which I will
discuss in the next chapter. On this account, things that figure in “invariant
generalizations” have causal explanatory relevance. It is clear that in this
sense nonfundamental things can have causal and explanatory relevance
even when the ”fundamental” cellular and molecular explanations are
complete. For example, research in spatial memory has indicated that in a
certain area of the hippocampus (CA2 and CA3) there is a place system
that represents relative location (creating spatial “maps”). In hippocampal
models of spatial memory, representations (e.g., the spatial maps) at what
Craver calls the “computational-hippocampal” level figure in explanatory
(invariant) generalizations. Interventions in the hippocampus are
systematically related to behavioral changes – for example, hippocampal
lesions in rats lead to systematic deficits in their ability to navigate in
mazes. Thus, properties at the computational-hippocampal level have
causal and explanatory relevance, even though the cellular and molecular
mechanisms are also (partly) known. They will continue to have it, even
when the cellular and molecular explanations are complete, since if they
were invariant to begin with, they will continue to be so.8
In order to counter this, a ruthless reductionist would have to show
that the interventionist account is wrong or insufficient, and that there is a
stronger notion of causation and causal explanation that applies to the
cellular/molecular level. I will discuss the problems of such stronger
accounts in more detail in Chapter 6. Furthermore, in this case such a
stronger notion of causation, even if available, would inevitably lead to
further problems. We could always ask the question: why stop at the
cellular/molecular level and not go further down to the
chemical/atomic/quantum level? Bickle is conscious of this, and in fact
seems to admit that it is possible that in the future causal explanations will
be found at the microphysical level (Bickle 2003, 156-157; 2006a, 431-
8
Of course, the explanatory generalizations involving higher-level properties might
turn out to be false, but that is an altogether different matter.
 33

432). This of course means that the cellular/molecular explanations are

only temporarily causal explanations, and that what counts as a causal
explanation is relative to the current state of science. It also implies that at
some point the causal explanations for all human behavior may be
microphysical explanations. While this might not be incoherent, it is a very
strong claim that requires substantial arguments to back it up, and so far
Bickle has not provided such arguments.
On the other hand we have the interventionist account of causation,
which I will now turn to. As this account has already received broad
acceptance among philosophers of science and directly supports multilevel
mechanistic explanation, the prospects of ruthless reductionism do not look
very good.
34
 35

3. The Interventionist Account of Causation

In recent years, several philosophers have presented accounts of causation
in terms of interventions and manipulability (Pearl 2000; Woodward 2003;
2008; Woodward & Hitchcock 2003, also Spirtes, Glymour & Scheines
1993). I will focus here on James Woodward’s (2003) version, which is
exceptional in its scope and clarity. The guiding insight of the account is
that causal relationships are relationships that are potentially exploitable
for purposes of manipulation and control. To put it very roughly, in this
model a necessary and sufficient condition for X to cause Y or to figure in a
causal explanation of Y is that the value of Y would change under some
intervention on X (in some background circumstances).
An intervention can be thought of as an (ideal or hypothetical)
experimental manipulation carried out on some variable X (the independent
variable) for the purpose of ascertaining whether changes in X are causally
related to changes in some other variable Y (the dependent variable). In
more detail, I is an intervention for X with respect Y if and only if (from
Woodward (2003, 98), slightly shortened and adapted):

1. I causes X
2. I acts as a switch for all the other variables that cause X (i.e., it is I
alone that causes X and not any of the other variables)
3. Any causal path from I to Y goes through X
4. I is independent of any variable Z that causes Y and that is on a
causal path that does not go through X

Interventions are not only human activities, there are also ”natural”
interventions, and the definition of an intervention makes no essential
reference to human agency. This sets the interventionist account clearly
apart from previous manipulability theories of causation (e.g., Menzies and
Price 1993).
According to Woodward, causal relationships are relationships that
are invariant under interventions. Suppose that there is a relationship
36

between two variables that is represented by a functional relationship Y =

f(X). If the same functional relationship f holds under a range of
interventions on X, then the relationship is invariant within that range. For
example, the ideal gas law “PV = nRT” continues to hold under various
interventions that change the values of the variables (P, V, and T), and is
thus invariant within this range of interventions. Invariance is a matter of
degree: for example, the van der Waals force law ([P + a/V2][V - b] = RT)
is more invariant than the ideal gas law since it continues to hold under a
wider range of interventions on the variables. One consequence of the
interventionist model is that relata of causation must be represented as
variables, but states or properties can easily be represented as binary
variables, such that, e.g., 1 marks the presence of the property and 0 the
absence of the property. In this framework, token causal claims can be
treated in the following way: variable X’s taking its actual value x causes
another variable Y to take its actual value y. Token causal claims need to be
always backed by type-level causal generalizations (invariant
generalizations).
Invariant generalizations are causally explanatory because they can be
used to answer “what-if-things-had-been-different questions” (w-
questions). For example, the ideal gas law can be used to show what the
pressure of a gas would have been if the temperature had been different. In
this way, the ideal gas law is potentially exploitable for manipulating and
controlling the temperature, pressure and volume of a gas. Generalizations
that are true but not invariant, like ”all the cups on the table of Dan Brooks
on 25.11.2009 are yellow” cannot be used to answer w-questions and are
not exploitable for manipulation and control.
This framework captures the nature of causation as difference-making:
changes in value of variable X make a difference in the value of variable Y
(in a range of circumstances). Another important point is that
interventionist causation is essentially contrastive: It is X’s taking some
value x instead of x’ that causes Y’s taking value y instead of y’. What the
“contrastive focus” or “baseline” for each variable is is not always made
explicit in causal claims, but this is crucial for assessing the truth of these
claims. For example, consider a case where a sick patient is given 200 mg
of penicillin and recovers, but would have recovered even if she had
 37

received only 100 mg of penicillin. The causal claim “giving the patient
200 mg of penicillin caused her recovery” is true if the contrastive focus is
taken to be no penicillin at all. However, a more appropriate interpretation
takes the contrast to be giving the patient less than 200 mg of penicillin,
and in this case the claim is false – the patient would have recovered also
when given, say, 150 mg of penicillin.
The interventionist account also allows for several noncompeting
representations of one and the same system. What variables we choose to
include in the representation depends on the question at hand. For example,
if we are interested in the restoring force that a spring exerts when pulled,
we can use Hooke's law (F = kX, where F is the restoring force exerted by
the spring, X is the displacement of the spring’s end, and k is the spring
constant) and include only variables F and X (and constant k). However, if
we need to know in more detail why the spring exerts the force in
accordance to constant k, we will have to include various variables
representing the physical properties of the spring. Importantly, this does
not make causal judgments subjective, since the counterfactual patterns of
dependence that make the causal claims true or false are mind-
independent. Once the variables and representations are fixed, causal
claims are true or false in a mind-independent way.
The interventionist account is a nonreductive and circular account of
causation, meaning that it does not provide a definition for causation in
non-causal terms. It gives criteria for distinguishing causal relationships
from noncausal relationships (most importantly correlation), but does not
give an answer to the metaphysical question of what causation really is.
However, the history of philosophy of causation is a history of failures to
provide such an account, which suggests that such a reductive and
noncircular account is not forthcoming. Furthermore, if the interventionist
account captures the notion of causation in science and everyday life, and
provides the criteria for distinguishing causal from noncausal relationships,
it seems to be all we need from a notion of causation (with the possible
exception of armchair metaphysics).
In the interventionist model, causation, causal relevance, and causal
explanation are coextensive: X is cause for Y iff X is causally relevant for Y
iff X figures in a causal explanation for Y.
38

The interventionist account abandons the ”nomothetic conception of

explanation,” according to which laws are essential and necessary for
explanations (e.g., Hempel & Oppenheim 1948). The problem with the
nomothetic conception is that most of the traditional criteria for laws
(universality, exceptionlessness, projectibility, etc.) do not seem to capture
the features that make generalizations explanatory, especially in the special
sciences, where the apparent lack of lawlike generalizations has been
traditionally seen as a problem. According to Woodward’s account, what
makes generalizations explanatory is not lawlikeness but invariance. This
solves the problem of the explanatory status of special science
generalizations: they are explanatory insofar as they are invariant, and
there is no lack of invariant generalizations in the special sciences.
In the interventionist framework, the traditional problems that have
plagued nomothetic models, like the asymmetry of explanations or the
problem of determining explanatory relevance, can be easily dealt with.
For illustration, let us consider the classic flagpole example. There is a
flagpole that casts a shadow and we want to explain why the shadow is as
long as it is. In the classic deductive-nomological model, this is done by
deducing the length of the shadow from the length of the pole, the position
of the sun and some general laws of optics. However, unfortunately it is
also possible to deduce the length of the pole from the length of the
shadow and the angle of the sun (and some general laws of optics), and
thus it seems that, according to the deductive-nomological model, the
length of the shadow and the angle of the sun explain the length of the
pole. This is an unwanted conclusion. In the interventionist model this
problem does not arise: interventions on the shadow alone do not change
the length of the pole. We can, for example, remove the upper part of the
shadow by shining a bright light on it, and this will surely leave the pole
intact. On the other hand, we can intervene to change the length of the pole
so that the length of the shadow does change. The interventionist account
shows correctly that the length of the pole is explanatorily relevant for the
length of the shadow, but not the other way around.
The interventionist model is intended to replace older and competing
accounts of causation, including causal process theories (Salmon 1984;
Dowe 1992) and Lewis’ (1973) counterfactual theory of causation.
 39

However, it is not a competitor for the mechanistic explanation model.

While the former is a model of causal explanation and relevance, the latter
is a general account of scientific explanation in neuroscience and many of
the life sciences. Craver (2007, Ch. 3) explicitly endorses Woodward’s
model as the appropriate account of causation for neuroscience. According
to the mechanists, good explanations describe mechanisms, and these
descriptions essentially involve relations that are causal in the
interventionist sense (see also Woodward (2002) for more on the relation
between mechanisms and causes). Both accounts support explanatory
pluralism (see Chapter 9).
40
 41

4. Levels
Talk of levels is ubiquitous. Philosophers talk about levels of nature,
analysis, realization, being, organization, explanation, or existence, to
name just a few. In science, the list is even longer. In the neurosciences
alone, at least the following uses of the term “level” can be found: levels of
abstraction, analysis, behavior, complexity, description, explanation,
function, generality, organization, science and theory (Craver 2007, 163–
164).
Talk of levels has of course also been important in debates about
reduction. Early on (see Chapter 1), when the goal was to reduce all
“higher-level” theories to “lower-level” theories, one important question
was how to sort the various theories into levels. Oppenheim & Putnam
(1958) presented an often-cited preliminary (and not intended as
exhaustive) division of scientific domains into six hierarchical levels —
social groups, (multicellular) living things, cells, molecules, atoms, and
elementary particles — which were supposedly related mereologically in
the sense that the entities at any given level are composed of entities at the
next lower level.
However, instead of reflecting the natural structure of the world, such
divisions give a simplified picture that reminds of elementary school
textbooks. For example, where do we place solar systems in this hierarchy?
What about organs? Brain areas? NMDA-receptors? Ecosystems? It is also
easy to see that there is no neat correspondence between scientific
disciplines and (compositional) levels of nature. There are some disciplines
that can be more or less easily associated with a level, like elementary
particle physics. However, many disciplines span several levels. Molecular
biology deals with entities of at least three different levels: cells, molecules
and atoms. Cognitive neuroscience deals at least with living things, cells
and molecules. On the other hand, at the level of multicellular living things
there are many different disciplines, from evolutionary biology to ethology
and cognitive psychology.
The mereological (compositional) relation that determines the levels
in Oppenheim and Putnam’s account is a feature in nearly all philosophical
accounts of levels of organization. This includes, for example, Jaegwon
42

Kim’s account of levels of properties. The level of a property, Kim (1998,

92) argues, depends upon what it is a property of: properties of objects
with parts are higher-level with regard to the properties of their parts, and
properties of objects with no parts are fundamental properties. In addition
to that, every level of reality has different “orders” of properties, generated
by the supervenience relation: second-order properties are generated by
quantification over the first-order properties that form their supervenience
base (Kim 1998, 20). Each level thus contains lower- and higher-order
properties; higher-order properties are properties supervening upon lower-
order properties of the same level, not upon lower-level properties.
Supervenience thus generates an intralevel hierarchy of lower- and higher-
order properties, while the interlevel micro/macro hierarchy between
properties of wholes and properties of their parts is not generated by
supervenience, but by mereology.
In addition to mereological considerations, size or scale is often
presented as a criterion for organizing things to different levels (e.g.,
Churchland & Sejnowski 1992). Organization by size partly follows from
compositional criteria, as parts are smaller or at least no bigger than
wholes.
However, the traditional criteria of composition and size lead to
anomalies and unwanted conclusions. A pile of snow is composed of
smaller piles of snow, but this does not mean that the larger pile of snow is
at a higher level than the smaller piles, at least not in any interesting sense.
Regarding size, there are bacterium-sized black holes and raindrop-sized
computers, but it does not seem very natural to say that bacteria are at the
same level as black holes, or that raindrops are at the same level as tiny
computers. Size and composition are not sufficient criteria for organizing
the world into levels in a natural or useful way.
McCauley (2007a; 2009) has recently proposed a novel way of
distinguishing analytical levels in science. He rejects the traditional criteria
for distinguishing levels, precisely because they lead to the above-
mentioned anomalies. Instead, McCauley proposes two new general
criteria: scope and age. As we go down the hierarchy of levels, the
sciences’ explanatory scope increases: atoms are everywhere, but cells are
not. Similarly, the lower a science’s analytical level, the longer its principal
 43

objects of study have been around. Conscious beings are newer than less
complex organisms, which are again newer than chemical compounds, and
so on. According to McCauley, these criteria could form a basis for
reviving the standard general framework for analytical levels in science.
However, McCauley’s criteria do not put all the sciences in their
traditional places in the hierarchical order (Eronen 2009). One significant
example of this is thermodynamics, which has had an important role in
discussions of reduction, starting from Nagel (1961). Thermodynamics has
been traditionally conceived as a higher-level science, and intuitively it
should be at a higher analytical level than, e.g., particle physics.
Thermodynamics deals mainly with macroscopic phenomena, while
particle physics deals with the smallest things in nature. However, we
cannot place these sciences at different analytical levels with McCauley’s
criteria. Regarding scope, thermodynamics applies to everything in nature.
Regarding age, we have the same situation: the objects of study of
thermodynamics include both very recent ones and ones that have been
around since the beginning of time. Similarly, particle physics studies
objects that are both ubiquitous and the oldest ones in the universe. Thus, if
we apply the criteria of scope and age, both sciences are located at the
lowest analytical level. This result is counterintuitive and against the
standard view in philosophy of science.9
Perhaps the most comprehensive account of levels of organization has
been developed by William Wimsatt (1976a; 1994; 2007). Wimsatt’s
starting point is that levels of organization are compositional levels that are
non-arbitrary features of the ontological architecture of the world. Wimsatt
is not aiming at a strict definition of levels, but rather at establishing sort of
a “prototype” idea of levels, by characterizing several characteristics levels
typically (but not necessarily) have. For example, levels of organization are
constituted by families of entities usually of comparable size, and the
things at a level mostly interact with other things at the same level, so that
the regularities of the behavior of a thing are most economically expressed

9
Admittedly, McCauley (2009, 619) concedes that his criteria are preliminary, not
intended as a definition, and probably sufficient only for putting the broad families of
sciences (physics, chemistry, biology, psychology, etc.) to their intuitively right places
in the hierarchy. For this purpose and for many cases they may well be sufficient.
44

in terms of variables and properties appropriate for that level. As a kind of

a preliminary definition, Wimsatt (2007, 209) suggests that “levels of
organization can be thought of as local maxima of regularity and
predictability in the phase space of alternative modes of organization of
matter.” Roughly speaking, this means that at the scale of atoms, for
example, there are more regularities than at scales just slightly larger or
smaller, so that at the scale of atoms there is a peak of regularity and
predictability, and thus a level of organization. If we draw a curve of
regularity and predictability against a roughly logarithmic size scale, levels
appears as peaks in the curve (Figure 3). In a world with a nice hierarchical
structure of levels, there are easily distinguishable peaks. In a world with
no levels there are no peaks. Wimsatt assumes that our world is somewhere
in between (Figure 3, alternative c).
Unfortunately, Wimsatt has not spelled out very clearly what he
exactly means by “peaks of regularity and predictability” (these problems
are pointed out and discussed by Craver (2007, 182-184)). If he means that
there are more causal regularities between entities of the same size scale
than between entities of different size scales, the claim is problematic.
Elephants regularly squash insects, nuclear explosions annihilate
everything from humans and ecosystems to macromolecules, humans
interact regularly with dust particles and airplanes, and so on.
On the other hand, Wimsatt’s claim can be interpreted so that there are
more regularities, be they causal or noncausal, between entities of the same
size scale than between entities of different size scales. However, this does
not work very well as a criterion or indicator for a level of organization. If
levels are also levels of composition, as Wimsatt seems to suggest, it
follows that there are countless noncausal regularities that span levels,
starting for example from ”all living beings have organic molecules as
their parts.” In order to flesh out the account we would at least need to
determine what kinds of regularities have to be considered and what not. In
general, it seems that trying to combine all of these criteria (composition,
number of regularities, and size) makes defining levels unnecessarily
complicated.
 45

Figure 3: Levels of organization as peaks of regularity and predictability

when plotted against a roughly logarithmic size scale (from Wimsatt
2007, 224-225, reprinted with permission of the author).
46

In any case, Wimsatt acknowledges that instead of a neat hierarchy of

the Oppenheim & Putnam (1958) kind, his criteria yield a complex and
branching structure of levels (Figure 4). Furthermore, at higher levels, for
example when it comes to psychology and neuroscience, neat
compositional relations break down. According to Wimsatt (2007, 227–
237), levels become less useful here for characterizing the organization of
systems, and it becomes more accurate to talk of “perspectives.”
Perspectives are subjective or at least quasi-subjective views of systems
and their structures that do not give a complete description of all aspects of
the systems in question, and that do not map compositionally onto one
another as levels of organization do. For example, anatomy, physiology,
and genetics can be seen as different perspectives on an organism.
Perspectives sometimes correspond loosely to disciplines, but they need
not to.
When even the boundaries of perspectives begin to break down,
perspectives degenerate into so called “causal thickets” where things are so
intertwined and multiply-connected that it is impossible to determine what
is composed of what and which perspective a problem belongs to (Wimsatt
2007, 237–240). According to Wimsatt, the neurophysiological, the
psychological and social realms are for the most part such causal thickets.
Wimsatt’s account has many virtues, but conceptual clarity is not one
of them. The key notions of perspectives and causal thickets remain rather
vague and unclear. Even the notion of a level of organization is not clearly
defined, although it must be said that Wimsatt intention never was to give a
definition but rather to show that there are “robust” (see Chapter 10) levels
of organization and to describe the properties they typically have.
 47

Figure 4: Levels of composition form a complex and branching structure

instead of a simple hierarchy (from Wimsatt 2007, 232-233, reprinted
with permission of the author).
48

Robert C. Richardson and Achim Stephan (Richardson & Stephan

2007) have proposed that we should extend accounts of levels of
organization by adding another independent dimension: grade of
resolution. They see the dimension of level of organization as ontological,
while the dimension of grade of resolution is related to explanation. We
can change the grade of resolution with which we are examining an entity
without shifting the level of organization. As the resolution gets higher, we
can predict and explain the behavior of the entity more exactly, but lose
generality. For example, we can examine a bacterial culture as a collection
of cells or as a collection of macromolecular systems. These are different
resolutions, but the level of organization does not change. This proposal
can be seen as useful supplement to an account of ontological levels of
organization, but it requires that we already have such an account. The best
candidate for this is Wimsatt’s account, which is impressive and intuitively
appealing, but far too vague to be entirely satisfactory.
Craver (2007, Ch. 5) has recently developed an account of levels that
is quite different from Wimsatt’s and has raised considerable discussion.
He starts by considering levels of composition, which he further divides
into four subcategories. Levels of mereology are levels with formal features
of mereological systems. As Craver points out, the formal apparatus of
mereology is rather inappropriate for describing or analyzing levels in
science. In levels of aggregativity, the relata are properties of wholes and
properties of parts, and the higher-level properties are sums of lower-level
properties. For example, the mass of a pile of sand (property of a whole) is
a sum of the masses of the individual grains of sand (properties of parts).
Levels like this are not very interesting. As Wimsatt (2000; 2007) has
emphasized, it is the failures of aggregativity that are interesting.10 The
third subcategory is levels of mere material/spatial containment, where an
entity is at a lower-level than another entity if the lower-level entity is
within the spatial boundaries of the higher-level entity and is a part of it.
Craver introduces them just to show that mere material/spatial containment

10
According to Wimsatt (2000;2007), testing the conditions for aggregativity reveals
interesting features of the organization of a system, and different kinds of failures of
aggregativity represent different kinds of emergence. Since truly aggregative systems
are rare, emergence is more the rule than the exception.
 49

is not enough for an interesting account of levels; lower-level parts have to

be components of the whole.
This leads to Craver’s own notion of levels, levels of mechanisms:

Levels of mechanisms are levels of composition, but the composition relation is

not, at base, spatial or material. In levels of mechanisms, the relata are behaving
mechanisms at higher levels and their components at lower levels. These relata
are properly conceived neither as entities nor as activities; rather, they should be
understood as acting entities. The interlevel relationship is as follows: X’s [phi]-
ing is at a lower mechanistic level than S’s [psi]-ing if and only if X’s [phi]-ing is
a component in the mechanism for S’s [psi]-ing. Lower-level components are
organized together to form higher-level components. (Craver 2007, 189)

Levels of mechanisms are not universal divisions in the structure of the

world (á la Oppenheim & Putnam 1958 or Wimsatt 1994; 2007). Different
mechanisms can have different structures of levels. For example, the levels
in the spatial memory system are different from those in the circulatory
system. There is no sense in which entities of different mechanisms are at
the same or lower or higher level – such comparisons are not possible.
Levels of mechanisms are always defined only relative to the mechanism
under study.
Bechtel (2008) has defended a similar account of levels. The claim of
these mechanists is that local and case-specific levels are sufficient for
understanding reductive explanations and interlevel relations in many
fields, particularly neuroscience. One limitation of this approach is that
global comparisons become impossible: it does not make sense to ask
whether things that belong to different mechanisms are at the same level or
not. We cannot say that cells are, in general, at a higher level than
molecules. All we can say is that cells in a certain mechanism are at a
higher level than the molecules that are part of the same mechanism. We
cannot even say that a certain molecule in a certain brain is at a lower level
than the hippocampus of that brain, unless the molecule is involved in the
same mechanism as the hippocampus. Even within a certain mechanism it
is not possible to say whether subcomponents of two different components
are at the same level or not, since they do not stand in a part-whole relation
50

to each other. Some of these implications are strongly counterintuitive, so

perhaps there is still need for more general levels of organization. Indeed,
Wimsatt-style levels of organization and levels of mechanisms are not
necessarily incompatible. As we have seen above, Wimsatt’s levels of
organization are said to “break down” in the neurophysiological and the
psychological realms, and these are exactly the realms where levels of
mechanisms are typically applied. In this sense, the two accounts may
simply complement each other (see also Walter & Eronen 2011).
A notion of levels that is more general than levels of mechanisms
would be desirable, since comparing the level of things in contexts broader
than single mechanisms (e.g., “molecules are at a lower level than cells”) is
common, intuitive, and often unproblematic. One possible way of
generalizing from levels of mechanisms to more general levels would be
the following: if a certain level appears in several (independent)
descriptions of (different) mechanisms, then we are justified in considering
it a more general level of organization. For example, the level of synaptic
transmission appears very commonly in neural mechanisms, and it is quite
natural to talk of it as a level of organization in the brain. Of course, this is
just a preliminary suggestion and in need of much refinement.
One further problem with Craver’s account of levels of mechanisms is
that he considers them to be objective, mind-independent levels of nature:
“I propose then that we start by thinking of levels as primarily features of
the world rather than as features of the units or products of science”
(Craver 2007, 177). This is related to the fact that he supports an “ontic”
conception of explanation, according to which explanations are objective
features of the world (Craver 2007, 26-28).
First of all, one problem with this approach is that most of the
scientific accounts of levels of mechanisms are certainly incomplete. There
might be intermediate levels that scientists have not yet discovered, or
some putative levels might turn out not to be levels at all. Relatedly, it is
quite plausible that in many cases there are competing explanatory
accounts of a given mechanism, with different hierarchies of levels. This
suggests that it is implausible that all our current scientific accounts of
levels of mechanisms are ontological levels of nature.
 51

Another way of understanding levels of mechanisms as objective

levels of nature would be that only the completed scientific accounts reveal
the “real” levels of nature. However, this would mean that (at least most
of) our current accounts are not in fact capturing levels of nature, and
describing them as “levels of nature” would be misleading. It is also
plausible that we could never be sure whether we are actually describing
the levels of nature or whether there are further corrections to come.11
We could avoid all these problems by accepting that levels of
mechanisms are levels of description (or analysis or explanation), not of
nature. We would then have local and case-specific levels of mechanism as
levels of description (this comes close to Bechtel’s (2007; 2008) view).
However, if we want to be scientific realists, this is unsatisfactory, since
we would still want to know what the relation is between these descriptions
and reality.
Following Wimsatt, I propose that we should understand the reality of
levels in terms of “robustness”. I will return to robustness in Chapter 10,
but the rough idea is that things are robust if they are accessible
(detectable, measureable, derivable, defineable, produceable, or the like) in
a variety of independent ways. Insofar as levels are robust, they are non-
arbitrary features of the world and represent something mind-independent.
This can also be connected to the considerations briefly mentioned above:
if a level appears in several independent mechanistic descriptions, it is
robust. On the other hand, a level can be considered robust insofar as there
is convergence of various independent (e.g., anatomical, functional,
developmental) considerations. Again, these ideas are preliminary and will
be further elaborated in future work. The main point is that we can make
sense of the reality and the ontological status of levels, but the solution is
not as straightforward as Craver suggests.

11
These considerations are of course closely related to the general debate on scientific
realism.
52
 53

Conclusions: Reductionism vs. Explanatory Pluralism

I have gone through the most prominent accounts of reduction in
philosophy of science and argued that, at least when applied to
neuroscience and psychology, they are all fundamentally problematic.
What we are left with is mechanistic explanation, which is only weakly
reductionistic in the sense of providing “downward-looking” mechanistic
explanations. Regarding the question of levels, I have argued that no fully
satisfactory account is currently available, but levels of mechanisms are a
good starting point, and levels can be said to be real in the sense of being
robust.
Mechanistic explanation, the interventionist account of causation, and
the approach of levels I have defended all support causal and explanatory
relevance of nonfundamental things.12 That is, they support the view that
there is no fundamental level of explanation, and that entities of higher
levels can have causal and explanatory relevance, even when lower-level
explanations are complete. This in turn supports explanatory pluralism (see
Chapter 9 for more). Its key idea is that for a full understanding of human
behavior and the mind, it is not enough to focus on one level of
explanation or one type of explanations. We need explanations at different
levels (molecular, cellular, behavioral, etc.) and of different kinds (causal,
mechanistic, psychological, etc.). Often we need also reductive
explanations in the sense of looking “downwards” into the composition of
systems in order to understand their behavior as a whole, but this is just
one type of explanation, and reductions in any stronger sense are not
forthcoming in the psychoneural case.
This is of course in sharp contrast to Bickle’s metascientific
reductionism and other strongly reductionist or eliminative projects. To
provide evidence for their claims, pluralists and mechanists have analyzed
cases from current scientific practice, even the ones that reductionists like
Bickle offer as prime examples of reductions. The pluralists claim that
these cases in fact show that higher-level explanations are indispensable,
even when lower-level explanations are complete, and that higher-level
research plays a crucial role in the advancement of science. For example,
12
This expression is taken from Craver (2007, Ch. 1).
54

Craver (2007, 233-245) discusses the case of LTP and memory

consolidation at length and concludes that reduction is peripheral to the
recent history of LTP, and that the goal of reduction has been replaced by
the goal of building multilevel mechanisms. Cory Wright (2007) takes up
the case of reward and mesocorticolimbic dopamine systems and argues
that it shows that ”top-down” strategies and psychological studies provide
refinement even after successful reductive explanations. Maurice Schouten
and Huib Looren de Jong (Schouten & Looren de Jong 2007) go through
the case of mind reading and mirror neurons, arguing that the molecular
mechanisms of social cognition are not explanatorily sufficient.
In the next part, I will critically examine some central problems in
philosophy of mind in light of the position I have defended in this part. In
Part III, I will consider the ontological implications of explanatory
pluralism, and explore the ways in which we can incorporate reductionist
ideas into a pluralistic framework.
 55

PART II:
Rethinking Reduction in
Philosophy of Mind
56
 57

Introduction
In this part, I will critically go through the most important approaches to
reduction in philosophy of mind, in light of what has been discussed in Part
I. I will proceed in a roughly chronological order. However, the main focus
will not be on history, but rather on problems and positions that are
relevant for contemporary philosophy of mind.
I will start with British Emergentism (section 5.1.) of the early 20th
century, focusing on C. D. Broad, who was a forerunner in many debates
that are now central in philosophy of mind. Then I will briefly discuss
translational reduction in logical positivism (section 5.2.) and how its
failure lead to the identity theory (section 5.2.) and to models of
intertheoretic reduction (Part I, Chapter 1).
Since the 1960s, the debates about reduction in philosophy of mind
have been quite distinct from those in philosophy science. In philosophy of
mind, Nagel’s model of intertheoretic reduction was widely accepted.
However, the argument from multiple realizability (section 5.3) apparently
made Nagel-reduction of psychology impossible, and also appeared to
refute ontological reduction of mental properties (the identity theory). This
led to something close to an antireductionist consensus, which lasted at
least until the 1990s. However, most philosophers of mind still wanted to
be good physicalists, which resulted in various attempts to formulate
nonreductive physicalism (section 5.6). The most popular form of
nonreductive physicalism was (and perhaps still is) functionalism (section
5.5.).
In the 1990s, the deficiencies in Nagel’s model were finally
acknowledged and accepted also in philosophy of mind, leading to
alternative models, most importantly functional reduction (Chapter 6).
Applying the functional model, many philosophers have recently argued
that phenomenal consciousness is fundamentally irreducible, or that there
is an explanatory gap between phenomenal consciousness and the physical
domain (Chapter 7). On the other hand, many prominent philosophers have
recently argued that the identity theory can still provide the solution to the
problem of phenomenal consciousness, leading to the new type physicalism
(Chapter 8).
58
 59

5. Traditional Approaches to Reduction in Philosophy of

Mind

5.1. British Emergentism

The emergentists of the early 20th century can be seen as forerunners of

both nonreductive physicalism (section 5.6) and explanatory pluralism
(Chapter 9), and thus deserve a closer look here. I will focus on C. D.
Broad, whose theory of emergence is the most elaborate one and the one
most relevant for the contemporary debates.
The heyday of British Emergentism13 was in the 1920s, when Samuel
Alexander (1920), C. Lloyd Morgan (1923) and C. D. Broad (1925)
published their main works. The context in which their theories were
formed was the controversy between mechanism and vitalism. According
to vitalists, organic phenomena could not be explained without appealing
to non-physical factors (like “élan vital” or “entelechies”) that make living
beings what they are. According to mechanists, everything could be
explained in mechanic terms, even organic phenomena. The emergentists
rejected both mechanism and vitalism and offered a third alternative: all
beings and structures, whether living or non-living, are composed of the
same basic elements, but there are irreducible chemical, biological, mental,
etc., properties, and different kinds of explanations must be applied to
things at different levels.14
C. D. Broad (1925, Ch. 2) distinguishes three possible types of theory
that account for characteristic differences of behavior. First, there are
theories that hold that “the characteristic behaviour of a certain object or
class of objects is in part dependent on the presence of a peculiar
component which does not occur in anything that does not behave in this
way” (Broad 1925, 55). An example of this kind of theory is substantial
vitalism, which states that a necessary factor in explaining the behavior of
living objects is the presence of an “entelechy,” a peculiar component that
does not occur in inorganic things.

13
This term was coined by Brian McLaughlin (1992).
14
See McLaughlin (1992) and Stephan (1999) for more on the history of emergentism.
60

The other two kinds of theories deny that peculiar components are
necessary for explaining behavior, and try to explain the differences
wholly in terms of difference of structure. According to one of them, the
characteristic behavior of a whole could, at least in theory, be deduced
from a sufficient knowledge of the behavior of the components. This kind
of theory Broad calls “mechanistic.”15 The most obvious example of a class
of objects to which mechanistic theories apply is mechanical devices. For
example, there is hardly any doubt that the behavior of a clock can be
deduced from sufficient knowledge of its components.
The other of these theories holds that “the characteristic behaviour of
the whole could not, even in theory, be deduced from the most complete
knowledge of the behaviour of its components, taken separately or in other
combinations, and of their proportions and arrangements in the whole”
(Broad 1925, 59). Broad calls this the theory of emergence. Its core is
captured in the following definition:

Put in abstract terms the emergent theory asserts that there are certain wholes,
composed (say) of constituents A, B, and C in a relation R to each other; that all
wholes composed of constituents of the same kind as A, B, and C in relations of
the same kind as R have certain characteristic properties; that A, B, and C are
capable of occurring in other kinds of complex where the relation is not of the
same kind as R; and that the characteristic properties of the whole R(A, B, C)
cannot, even in theory, be deduced from the most complete knowledge of the
properties of A, B, and C in isolation or in other wholes which are not of the
form R(A, B, C). (Broad 1925, 61)

In current terms, underlying this definition is a model of reduction and

reductive explanation, according to which a property of a whole is
reducible and reductively explainable if and only if it can be deduced from
the most complete knowledge of the properties of its components in
isolation or in other wholes. If such a deduction is not possible, even in
principle, the property is emergent.

15
This resembles contemporary accounts of mechanistic explanation discussed in Part
I, Chapter 2, but also differs in some crucial respects – for instance, Broad’s account
centrally involves the notion of “deduction,” while the contemporary accounts do not.
 61

One problem of this definition is that it essentially involves the notion

of deduction. First of all, it is hard to see how properties could be involved
in a deduction. If we understand deduction as logical derivation, the relata
of a deduction have to be laws or statements. Therefore, it is misleading to
talk about deduction of properties – one can only deduce laws or
statements involving properties. Taking this into account, Stephan has
proposed the following construal of Broad’s definition:

Irreducibility. Let there be a system S whose systemic property E is

nomologically dependent on the microstructure <c1, … , cn | o> of S (i.e. the
components c1, … , cn and their organization o). E is irreducible, if the law
according to which all systems with the microstructure <c1, … , cn | o> have the
property E can not be deduced, even in principle, from the laws that describe
the behaviour and properties of the components c1, … , cn in isolation or in
systems simpler than S. (Stephan 1999, 36, my translation)

However, this definition still appeals to deduction, and it is not clear how
we should understand deduction here. If we take deduction to be logical
derivation, the statements and laws have to be formalized, preferably in
first-order predicate logic. This again leads to the kinds of problems
discussed in Part I (Chapter 1), at least when it comes to psychology,
neuroscience, and the life sciences. If, on the other hand, deduction does
not mean logical deduction but something else, this would have to be
clearly spelled out.
Ansgar Beckermann (1997) has proposed the following reading of
“deduction” in Broad’s definition:

To deduce F from the properties of S’s components by means of fundamental

laws is to show on the basis of the fundamental properties of S’s components
and the laws of nature generally applying to objects with these properties that S
possesses all features which are characteristic of property F, or that [C1, …, Cn;
R] [that is, system S consisting of the parts C1, …, Cn in the arrangement R]
possesses all features characteristic of the property F. (Beckermann 1997, 307-
308)
62

The problem with this definition is that “to deduce” is just replaced by
another problematic term, “to show.” When explaining how we should
understand this, Beckermann turns to Hooker (1981), and effectively
invokes the New Wave model of reduction (see Part I, Chapter 1). This
again leads to the problem I have already discussed in detail in Part I,
namely that of applying formal models of theory reduction to psychology
and neuroscience.
However, there is another way of making sense of emergence that can
be drawn from Broad’s work, particularly from his much-discussed
archangel example (Stephan 1999; 2006). This way of defining emergence
is less problematic and connects to contemporary accounts of emergence in
an interesting way.
Broad asks us to imagine “a mathematical archangel, gifted with the
further power of perceiving the microscopic structure of atoms as easily as
we can perceive hay-stacks” (Broad 1925, 71). He argues that even if the
mechanistic theory of chemistry was true (instead of the emergent theory),
there would be a theoretical limit to the deduction of the properties of
chemical elements and compounds:

Take any ordinary statement, such as we find in chemistry books; e.g., "Nitrogen
and Hydrogen combine when an electric discharge is passed through a mixture of
the two. The resulting compound contains three atoms of Hydrogen to one of
Nitrogen; it is a gas readily soluble in water, and possessed of a pungent and
characteristic smell." If the mechanistic theory be true the archangel could deduce
from his knowledge of the microscopic structure of atoms all these facts but the last.
He would know exactly what the microscopic structure of ammonia must be; but he
would be totally unable to predict that a substance with this structure must smell as
ammonia does when it gets into the human nose. The utmost that he could predict
on this subject would be that certain changes would take place in the mucous
membrane, the olfactory nerves and so on. But he could not possibly know that
these changes would be accompanied by the appearance of a smell in general or of
the peculiar smell of ammonia in particular, unless someone told him so or he had
smelled it for himself. If the existence of the so-called "secondary qualities," or the
fact of their appearance, depends on the microscopic movements and arrangements
of material particles which do not have these qualities themselves, then the laws of
this dependence are certainly of the emergent type. (Broad 1925, 71-72)
 63

What Broad is suggesting is that it is not a priori impossible that chemistry

and biology were mechanical, but even if they are, they cannot be the
whole truth of the material world, because smells, tastes, colors and other
secondary qualities (in contemporary terms, phenomenal properties) cannot
be mechanically explained. Broad’s point is that even if all else turns out to
be mechanistic, at least secondary qualities have to be emergent.
According to Broad, the laws connecting microscopic particles or
events with secondary qualities must be emergent laws, “[a]nd no complete
account of the external world can ignore these laws” (Broad 1925, 72).
Broad (1925, 65) tells us that a law that connects an emergent property of a
structure with the properties of the components of the structure is a unique,
ultimate and irreducible law. This means that it is not a special case of a
more general law and that it does not arise from a combination of more
general laws. It is a law that could have been discovered only by studying
this particular case.16
Why is Broad so convinced that the archangel, who has unlimited
capabilities of calculation and can directly perceive all microscopical
structures, nevertheless cannot know what is the smell of ammonia? Broad
does not state this explicitly, but a very plausible interpretation is that
Broad thinks that the archangel cannot know this because the smell of
ammonia is not causally/functionally analyzable, and therefore cannot be
deduced from the behavior of the related structures. From this we can draw
the following definition for irreducibility (Stephan 1999, 41):

Irreducibility. Systemic properties that are not causally/functionally analyzable

are (necessarily) irreducible.

This kind of irreducibility is central in contemporary philosophy of mind,

particularly in the arguments for the explanatory gap and qualia
emergentism. The clearest candidates for properties that cannot be
causally/functionally analyzed are phenomenal properties (qualia), and

16
More recently, David Chalmers (1996) has defended the view that there are such
fundamental and inexplicable laws connecting physical properties with phenomenal
properties.
64

thus several philosophers have argued that they are strongly emergent
properties (Stephan 1999; 2006; Kim 1999) or that they present a
fundamental explanatory gap (Levine 1983; 1993). I will return to this in
Chapter 7.
British Emergentism is sometimes seen as a precursor of nonreductive
physicalism. Like nonreductive physicalists, the emergentists claimed that
everything is composed of physical stuff (and only of physical stuff), but
that the special sciences are nonetheless irreducible. Jaegwon Kim (1999,
5) has gone so far as to claim: “The fading away of reductionism and the
enthronement of nonreductive materialism as the new orthodoxy simply
amount to the resurgence of emergentism … It is no undue exaggeration to
say that we have been under the reign of emergentism since the early
1970s.” However, equating emergentism with nonreductive physicalism in
this way is problematic. While the views of Broad sometimes come close
to nonreductive physicalism, other emergentists had more radical views
regarding the causal powers and the ontological status of emergent
properties, for example claiming that emergent properties have novel
causal powers and can also exert causal influence “downwards” onto the
physical level. Furthermore, in contrast to the British Emergentists,
nonreductive physicalists do not claim that higher-level properties are
emergent in the strong sense (see above) – they typically argue that higher-
level properties are physically realized (see section 5.6 and Chapter 10).
Therefore, it is sensible to distinguish between nonreductive physicalism
and emergentism, the latter being a more radical position.
British Emergentism also bears similarity to explanatory pluralism and
what I call pluralistic physicalism (Chapter 10), since the emergentists
claimed that even though everything is composed of physical stuff, there
are irreducible and indispensable chemical, biological, psychological, etc.,
explanations and properties. However, it is again important to emphasize
that the emergentists also made other more radical claims that go far
beyond what I am defending.
British Emergentism eventually drifted into the periphery of
philosophy before the middle of the 20th century. However, this was not
due to philosophical inconsistencies or problems, as McLaughlin (1992)
and Stephan (1999, 129-155) have pointed out. The first main reason for
 65

the downfall was advances in science (McLaughlin 1992): for example, the
explanation of chemical bonding in terms of electron bonds and the
discovery of DNA made the existence of emergent properties or laws in
chemistry and biology seem highly improbable.17 The second main reason
was the rise of logical positivism in the 1930s (as Kim (1999) points out).
This trend in philosophy was anti-metaphysical and hyper-empiricist and
the supposedly vague concept of emergence had no place in its view of the
sciences, although some positivists tried to adapt the concept of emergence
for their own purposes by making it weaker and theory-relative (e.g.,
Hempel & Oppenheim 1948).
However, as I have already pointed out, emergence has made a
comeback in contemporary philosophy of mind. Since the 1990s, it has
also been a central and much-discussed concept in cognitive science and
complexity studies. As Stephan (2006) shows, this concept of emergence is
substantially different and weaker that the one in philosophy of mind.
Philosophers who have presented such weaker accounts of emergence
include, for example, Batterman (2001), Bedau (1997), Clark (2001), and
Wimsatt (2000; 2007). Since my main focus is on philosophy of mind,
these alternative accounts are beyond the scope of this thesis.

5.2. Logical Behaviorism and Identity Theory

In the early 20th century, when British Emergentism was still enjoying its
heyday, a very different philosophical movement was already emerging.
Logical positivists in Vienna and Berlin set out to understand the nature of
science and the relations between sciences, armed with the logical tools
developed by Russell and Frege, most importantly predicate calculus, and
using physics as the model science. One of their goals was to “unify
science,” or to find a common language into which all meaningful (i.e.,
verifiable) scientific statements could be translated.
Physics was considered the most successful, fully developed, and
fundamental science, and thus positivists like Carnap (1932a; 1932b) and

17
However, recently Boogerd et al. (2005) have argued that there is strong emergence
in cell biology, and in a sense that is closely related to Broad’s ideas.
66

Hempel (1949) argued that the language of physics should serve as the
universal language of science. They claimed that all meaningful (i.e.,
verifiable) scientific concepts, statements, and laws should be translatable
into the physical language. They emphasized that this applies also to
psychology and mental concepts:

All psychological statements which are meaningful, that is to say, which are in
principle verifiable, are translatable into statements which do not involve
psychological concepts, but only the concepts of physics. The statements of
psychology are consequently physicalistic statements. Psychology is an integral
part of physics. (Hempel 1949, 18)

This extreme form of reductionism came to be known as logical

behaviorism (although the alternative title semantic physicalism is more
descriptive). However, it became soon obvious that this approach could not
lead very far, since translating even simple (meaningful) psychological
sentences into physical language turned out to be extremely difficult, and
the purported translations became indefinitely or infinitely long. Consider
for example the simple sentence “Kosta wants a beer.” This could be
translated to behavioral dispositions like “If there is beer in the fridge,
Kosta will get one.” However, this dispositional statement is true only IF
Kosta does not want whiskey even more, IF he does not attempt to stay
sober, IF he has money, and so on. It is clear that the full list of conditions
would be indefinitely (perhaps even infinitely) long. Furthermore, the
conditions include psychological terms like “want” or “attempt,” so there
is a threat that the translation becomes viciously circular.
For these and other reasons, it was widely accepted that the
requirement that psychological sentences have to be translated without loss
of meaning into the physical language was a far too strict and unrealistic
condition for reduction. Translational reduction was therefore replaced by
more sophisticated models of reduction (see Part I, Chapter 1).
Also the development of the ontological type identity theory can be
seen as a reaction to the failure of translational reduction. The impossibility
of translating psychological statements to physical statements does not rule
out the following possibility: for any meaningful psychological concept or
 67

sentence M, there is a physical concept or sentence P, such that M and P

refer to the same thing. Expressions may differ in meaning, but still have
the same reference. This was the line of argument adopted by the
advocates of the identity theory (Place 1956; Feigl 1958; Smart 1959).
Their claim was that every mental state or property or process is identical
to a neurophysiological state or property or process of the brain: a given
mental state or property or process simply is a neurophysiological state or
property or process.18 As Smart (1959, 147) writes regarding sensations,
“…in so far as a sensation statement is a report of something, that
something is in fact a brain process. Sensations are nothing over and above
brain processes.”
The identity theorists appealed to analogies from other domains. For
example, the expressions “lightning” and “electric discharge” have
different meanings, and it is possible to talk about lightning without
knowing anything about electric discharges. However, in fact lightning is
electric discharge, nothing over and above it. In the same way,
psychological expressions have different meanings than
neurophysiological expressions, but in fact, according to the identity
theory, psychological states are neurophysiological states.
Identity theory is often called reductive physicalism, since it states
that everything can be ontologically reduced to the physical. It played a
very important role in the philosophy of mind of the 20th century, although
for decades it was considered to be refuted, or at least very problematic,
mostly due to the argument from multiple realizability (next section).
Recently it has made a comeback and new explanatory arguments have
been put forward in its defense (Chapter 8), while also the force of the
multiple realizability argument has come more and more under doubt.

18
For Place (1956) the ‘is’ was actually ‘is’ of composition, not of identity, and thus
his view differs from the others in this respect.
68

5.3. Multiple Realizability

The most influential argument against reductionism and the identity theory,
the argument from multiple realizability, was brought to the discussion by
Hilary Putnam (1967), and has remained in the center of philosophical
attention to this day. Putnam took as an example “pain,” or the state of
being in pain, and argued that the identity theorist would have to claim that
there is a physical-chemical state such that any organism – be it human,
octopus, or an alien – is in pain if and only if its brain is in that physical-
chemical state. He further argued that if we can find even one
psychological state that can be found in different species, but whose
physical-chemical “realizer” is different in these species, the identity
theory has collapsed. It seemed extremely likely that there are such
psychological states, since states like “pain” or “hunger” appeared to be
realized in different ways in different species. Therefore, Putnam
concluded that the identity theory was empirically and methodologically
problematic, and proposed that we replace it with functionalism (see
section 5.5).
Multiple realizability (MR) was generally considered extremely
plausible, both empirically and conceptually, and seemed to refute both the
identity theory and reductionism (of the Nagel kind). As an illustration,
consider this quote from the year 1989:

It is practically received wisdom among philosophers of mind that

psychological properties … are not identical to neurophysiological or other
physical properties. The relationship between psychological and
neurophysiological properties is that the latter realize the former. Furthermore,
a single psychological property might (in the sense of conceptual possibility) be
realized by a large number, perhaps infinitely many, of different physical
properties and even by non-physical properties. (LePore and Loewer 1989, 179)

The wide acceptance of the multiple realizability argument led to a kind of

antireductive consensus that lasted for decades.
However, in recent years it has become more and more
commonplace to question both parts of the multiple realizability argument:
 69

it is not clear whether there is multiple realizability in any philosophically

interesting sense, and even if there is, it is not clear whether this is a
problem for reductionism or the identity theory.
First of all, it is interesting and important to note that both defenders
and opponents of the multiple realizability argument generally took the key
notion of realization as self-evident and unproblematic, and rarely tried to
spell it out or explicate it. Recent analyses (e.g., Polger 2004 and Shapiro
2004) have shown that the idea of realization is in fact very problematic,
and that it is unlikely that there could be one notion of realization that
would fit all the standard examples. For example, a computer realizing an
abstract algorithm or computation can hardly involve the same realization
relation as a brain realizing a mental property, since mental properties are
thought to be individuated causally, but abstract algorithms or
computations are not individuated causally (Polger 2007a). This means that
there might be no general realization relation that applies to all the
different cases that are presented as typical cases of realization (see also
Chapters 6 and 10 for more on problems realization). However, I will not
push this argument any further here. Even if realization is one thing in
psychology and something different in computer science, it is still possible
that psychological properties are multiply realized by neural properties,
which is the main issue here.
Kim (1992), Bickle (1998), and Bechtel and Mundale (1999), have
argued that the practice of neuroscience speaks against multiple
realizability of psychological properties. They claim that the fact that there
are differences between brains has not stopped neuroscientists from
identifying the same brain areas and processes in different members of a
species or across species. For instance, scientists study the macaque visual
system in order to understand the human visual system, and this is possible
because there is continuity between species and the realizations are
sufficiently similar. In general, neuroscientific research has quite
substantially contributed to the research of cognitive functions, and this
would not be possible if realizations of cognitive functions were radically
different across or within species.
These considerations point to a fundamental problem of multiple
realizability: when and based on what should we consider two instances of
70

a psychological property as instances of the same property, and when as

two different properties? The same problem of course applies to the
realizer properties: when do two realizers count as different realizers, when
as the same realizer?
Remember that the identity theory, which is the target of the MR
argument, is a theory about types. It states that mental types (states, kinds,
properties) are identical to neural types (states, kinds, properties). Now it is
clear that tokens of a type need not be exactly alike in order to count as
tokens of the same type. Two tokens can differ in many respects but still
count as tokens of the same type. For instance, consider two plastic bottles
that are otherwise the same but one of them has a scratch. They are not two
different realizers of “plastic bottle,” they are two different tokens of the
same type, or two different instantiations of the property of being a plastic
bottle. If we would count cases like this as multiple realizability, then
claims about multiple realizability would become utterly trivial.19
Therefore, it is not the case that any variation among or difference in
realizers of a property automatically or unconditionally implies multiple
realizability. Sometimes what prima facie appears to be a case of two
different realizer properties (types) turns out to be just two different tokens
of one and the same realizer property. In these cases, there is no genuine
multiple realizability.
Polger (2004; 2009) and Shapiro (2000; 2004) have argued that in
order to count as different realizers, the realizers must differ in ways that
contribute to their sameness. For example, in order to count as different
realizers, two corkscrews must differ in ways that are relevant to their
function as corkscrews, that is, in ways that affect how they pull out corks
(differences in, say, color, are not relevant). Otherwise they count as same
types of corkscrew. Regarding the psychoneural case, the neural realizers
have to have neuroscientific differences in ways that are relevant to them
being the same psychological property. For example, if two neural
properties that realize a psychological property have slightly different

19
Aizawa & Gillett (2009) defend an account of realization that apparently leads to
such promiscuous and ubiquitous multiple realizability, and also try to argue that this
does not make multiple realizability trivial.
 71

temperatures but are otherwise the same, they count as tokens of the same
type, not as different realizers of the psychological property.
Shapiro spells out this account of realization in terms of “R-
properties:”

I propose to use the name R-property as a label for those properties of

realizations whose differences suffice to explain why the realizations of which
they are properties count as different in kind. In short, two types of realization of
some functional kind, like corkscrew or watch, count as different kinds of
realization if (by definition) they differ in their R-properties … R-properties are
those that are identified in the course of functionally analyzing some capacity. It
is to functional analysis, I claim, that we should look for the identification of R-
properties because it is functional analysis that uncovers those properties that
causally contribute to the production of the capacity of interest (Shapiro 2004,
52-53).

If we accept this approach to realization, multiple realizability of

psychological properties becomes far less obvious than it prima facie
seems: it is an open question whether the neural realizers of a given
psychological property differ in their R-properties.20
There is also another fundamental problem for claims of multiple
realizability: what is the right “grain” for psychological properties on the
one hand and neural properties on the other hand? Bechtel and Mundale
(1999) present the following as one possible diagnosis for the appeal of
MR claims among philosophers: ”researchers have employed different
grains of analysis in identifying psychological states and brain states, using
a coarse grain to identify psychological states and a fine grain to
differentiate brain states” (Bechtel & Mundale 1999, 202). For example,
Putnam talked about “pain” in a very coarse grain, assuming that the same

20
The point of this section is not to defend any particular approach to realization, but
to show that the question of multiple realizability is complicated and far from settled,
and that it is not as tightly connected to the question of reduction as has been
traditionally assumed. In Chapters 6 and 10, I briefly return to the notion of realization
and argue that no metaphysical realization relation is needed for understanding
reduction and reductive explanation.
72

“pain” is manifest in humans and octopi, and then stated that this “pain”
cannot be identical to any physical-chemical state.
There are certainly some “grains” for psychological and neural states
that make MR claims false, and some that make them true. As Richardson
(2009) points out, the problem is that of finding some principled reasons
for choosing one grain instead of another, and this is no easy task. Another
way of putting this is that multiple realizability is not absolute, but a
concept that comes in degrees and depends on the conceptual framework(s)
that are relevant for the question at hand (Polger 2008b).
The question of multiple realizability of the mental then turns out to
have three parts: are psychological properties multiply realized by neural
properties when we have (1) settled on the right “theoretical grains,” (2)
agreed upon the right way of classifying properties as same or different,
and (3) agreed upon the right notion of realization? This makes multiple
realizability far less obvious and plausible than has generally been thought.
Furthermore, the answers to these questions also depend on empirical
matters, not just philosophical considerations.
One hotly debated case that prima facie supports multiple
realizability claims is that of neural plasticity and recovery of function. It
has been the most central empirical part in the arguments of the supporters
of MR (e.g., Block and Fodor 1972). For example, children who
experience severe head trauma and lose some of their linguistic abilities
often recover these abilities as other brain areas take over the functions of
the damaged area. Richardson (2009) argues that in these cases the same
psychological functions are realized by different structures, and this is
multiple realization. On the other hand, Polger (2009) and Shapiro (2004)
argue that we cannot conclude so easily that these are cases of MR; one
needs to show that the different realizing brain areas also differ
functionally after they have adapted to the new task. If they do not differ
functionally, they are not really different realizers, but rather the same
realizers located in different brain areas. Polger and Shapiro then show that
in at least some cases of recovery of function the brain areas that take over
the functions actually become functionally and organizationally similar to
the old areas. In response, Richardson (personal communication) claims
that the realizers obviously have to be functionally similar to carry out the
 73

functions necessary for language, but this does not make them the same
sort physiological states. The debate is far from resolved.
Due to constraints of space, I cannot evaluate the empirical evidence
for and against multiple realizability here (see Raerinne & Eronen
(forthcoming) for more discussion). However, in spite of the arguments of
Polger and Shapiro, I find it very plausible that there are at least some
multiple realizations of psychological functions by neural mechanisms,
even if we have a strict notion of realization (see Aizawa & Gillett 2009
for some candidate cases). Whatever the right “theoretical grain” of
psychological functions turns out to be, it will probably be rougher than
that of neural mechanisms. Then it is not surprising if one psychological
function can be realized by different neural mechanisms, in the sense that
these mechanisms perform the same roughly defined function, but perform
it significantly differently from each other.
In any case, even if there is MR, this does not necessarily prevent
reduction, even of the traditional kinds. Richardson (1979) was the first to
show that even Nagel-reduction is compatible with MR, as long as we
don’t require that bridge laws have to express biconditionals, but accept
also material conditionals. Robert Batterman (2001), Clifford Hooker
(1981) and Patricia Churchland (1986), among many others, have pointed
out that there is multiple realizability also in physics: for instance,
temperature is one thing in a gas, but something else in a solid, and yet
something very different in a plasma. Yet, this is no obstacle for reduction
or reductive identification of temperature. The reductions and cross-
theoretic identifications just have to be “domain-specific:” temperature is
identified with different things in different domains. Temperature in a gas
is identical to the mean molecular kinetic energy, while temperature in a
plasma is identical to something else, and so on.
In the same vein, Kim (1992) has argued that instead of a one-shot
general Nagel-reduction of a certain mental property, we should restrict the
bridge laws or identities to appropriately individuated biological species
and physical structure types. For instance, instead of having one bridge law
that connects pain with a physical property (or a disjunction of properties),
we will have a bridge law or identity for each structure type S connecting
pain-in-S with a physical property in S.
74

These approaches assume an intertheoretic (or bridge-law) model of

reduction. If we take into account what has been discussed in Part I and
take downward-looking mechanistic explanation to be the strongest form
of psychoneural reduction, MR is even less problematic. If there are one-
to-many mappings from psychological properties or functions to the
underlying mechanisms, this is no obstacle to mechanistic explanation of
those properties or functions. In these cases, different mechanisms can
perform the same roughly defined function, and therefore there are
different mechanistic explanations for this function. There is nothing
problematic about this.
However, one question still remains open: is there multiple
realizability in a sense that threatens the ontological identity theory? As I
mentioned above, it is plausible that there are one-to-many mappings
across levels or domains in a sense that suggests that there is multiple
realizability, even though it may not be as ubiquitous and obvious as
philosophers mind have assumed for a long time. But I have not argued for
this in detail, since it is not crucial for my position. Instead, I will show
later that the main positive argument for the identity theory, the causal
exclusion argument, becomes highly suspect if we adopt the interventionist
notion of causation.21 The ontological alternative(s) to the identity theory
will then be discussed in Part III.

5.4. The Disunity of Science as a Working Hypothesis

A few years after Putnam (1967) introduced the problem of multiple

realizability, Jerry Fodor (1974) launched another fierce attack against
reductionism. In his aptly named article “Special Sciences (Or: The
Disunity of Science as a Working Hypothesis),” Fodor (1974) attacked
“reductionism,” which he understood as the view that all special sciences
reduce to physics. What Fodor meant by reduction, in turn, was essentially
a strict form of Nagel-reduction (although for some reason Fodor did not
refer to Nagel). That is, a science is reduced if and only if all its laws are
21
The problems of one other positive argument for the identity theory, the one that
new type physicalists have proposed, will be discussed in Chapter 8.
 75

derivable from laws of the reducing science via biconditional bridge laws.
This of course means that if reductionism is true, all the laws of all the
special sciences have to be derivable from the laws of physics. Fodor then
argued that if we assume that laws, including bridge laws, have to connect
“natural kinds” (Fodor defined them roughly as the predicates whose terms
are bound variables in laws), it follows that reductionism requires that
every natural kind is, or is coextensive with, a physical natural kind.
Fodor further argued that this is a very unfortunate conclusion, since
(a) interesting generalizations can be made about events whose physical
descriptions have nothing in common, (b) it is often the case that whether
the physical descriptions have anything in common is entirely irrelevant to
any epistemologically important properties of the generalization, and (c)
the special sciences are very much in the business of making
generalizations of this kind. Fodor took up the example of monetary
exchanges: economics is full of generalizations involving monetary
exchanges, but the physical descriptions of monetary exchanges have
nothing interesting in common. There is no physical natural kind
corresponding to money or monetary exchanges. Similar reasoning can be
applied to the case of psychology and neurology: if psychology is
reducible to neurology, then for every psychological natural kind predicate
there must be a coextensive neurological natural kind predicate. However,
this does not seem to be the case, since it is likely that there are systems
that satisfy psychological natural kind predicates but do not satisfy any
single neurological kind predicate (multiple realizability).
Fodor proposed that we abandon the thesis of “unity of science”
(Oppenheim and Putnam 1958), which according to him was equal to the
thesis that all sciences should be reduced to physics. He claimed that the
reason why many philosophers had accepted this view was that they
wanted to endorse “generality of physics:” all events that fall under the
laws of any science are physical events and hence fall under the laws of
physics. However, according to Fodor, generality of physics is a much
weaker thesis than reductionism, and all we need to assume for it is “token
physicalism,” which states that all events that sciences talk about are
physical events.
76

It is clear that Fodor’s arguments were targeting a very strict form of

Nagel-reductionism. He dismissed this problem by one short remark:

The version of reductionism I shall be concerned with is a stronger one than

many philosophers of science hold, a point worth emphasizing since my
argument will be precisely that it is too strong to get away with. Still, I think
that what I shall be attacking is what many people have in mind when they refer
to the unity of science, and I suspect (though I shan’t try to prove it) that many
of the liberalized versions of reductionism suffer from the same basic defect as
what I shall take to be the classical form of the doctrine. (Fodor 1974, footnote
2)

Unfortunately for Fodor, this is not the case. As Richardson (1979) first
pointed out, material conditionals and one-to-many mappings are enough
to fulfill Nagel’s conditions for reduction. This means that there is even a
form of Nagel-reductionism for which Fodor’s argument is a nonstarter.
Furthermore, at the time when Fodor’s article was published, several
philosophers of science had already pointed out that Nagel’s model
suffered from serious defects, and thus had to be replaced or adjusted (e.g.,
Feyerabend 1962; Schaffner 1967; Sklar 1967). It is not clear to what
extent Fodor’s argument applies to more sophisticated models, such as the
New Wave model. If Bickle (1998) is right, no biconditional bridge laws
linking the reduced and the reducing theory are needed in reductions that
comply to the New Wave model, since the reducing theory simply
“displaces” the older theory. If this is the case, Fodor’s argument loses its
force. It is also clear that the argument says nothing against the possibility
of reduction understood as downward-looking mechanistic explanation.
However, Fodor’s points about the epistemological importance of
special science generalizations still hold, and are strongly supported by the
interventionist model of causation and the insights of explanatory
pluralists. If we look at points (a), (b), and (c) in this light, we see that they
all turn out true. If a generalization is explanatory in the interventionist
sense, i.e., if it is an invariant generalization, the question whether the
physical descriptions of the variables have anything in common is indeed
entirely irrelevant to the epistemologically important properties (for
 77

instance, the explanatory power) of the generalization. It is also true that

the special sciences are very much in the business of making
generalizations of this kind. In this sense, Fodor was more or less right
about the above-mentioned points regarding special science
generalizations, although his argument for their irreducibility was
insufficient. However, Fodor (1991) has later also argued for
nomologically necessary special science laws that are hedged with ceteris
paribus clauses and are in a strong sense autonomous. In contrast, I hold
the view (drawn mostly from Woodward (2000; 2003) and Wimsatt
(2007)) that special science “laws” are not nomologically necessary or
strongly autonomous, but invariant generalizations that are integrated to
and “co-evolve” with generalizations of lower (and higher) levels.

5.5. Functionalism

The primary ontological framework for nonreductive physicalism was

functionalism. There are reductive forms of functionalism, but it is more
often considered to be the main alternative to reductive physicalism.
Indeed, its emergence can be seen as a reaction to the failure of logical
behaviorism and to the problems of the identity theory. Its core idea is that
what makes something a mental state or property of a certain type is the
function or the causal role it has – simply put, what it causes and what it is
caused by. In contrast to logical behaviorism, functionalism focuses on the
role of inner mental states instead of just directly observable behavior. In
contrast to the identity theory, it does not require that mental properties
should be identical to neurophysiological properties of the brain, and
allows that there are multiple physical realizations of one mental property.
In the earliest version of functionalism, machine functionalism, the
mind was considered to be a kind of a Turing machine (namely, a
probabilistic automaton), and mental states were identified with the
machine table states of that machine (Putnam 1967). However, this turned
out to be too restrictive, since machine table states do not correspond in
any natural way to psychological states (see Block & Fodor 1972 for more
on the problems of machine functionalism). For instance, machine table
78

states are total states of the whole system, but psychological states are
typically characterized in a much more fine-grained way (we typically
ascribe to people many distinct beliefs, desires, feelings, and so on, at any
given moment).
The later versions of functionalism no longer invoked the Turing
machine. The idea was still that mental states or properties are defined by
their function or causal role, but the functions or causal roles were to be
specified by empirical (cognitive) psychology. Mental states were
identified with functionally defined states that appear in our best
psychological theories. Mental states and psychological explanations were
also considered to be autonomous and irreducible, mainly based on
considerations outlined in previous sections (multiple realizability and
Fodor’s arguments). This, the most influential branch of functionalism, is
often called psychofunctionalism (Block 1978) or empirical functionalism.
It is a common implicit background assumption in philosophy of mind,
even though it is difficult to find explicit formulations of the position or
detailed positive arguments for it.
A very differently motivated branch of functionalism was analytic
functionalism, where the functional roles of mental states were not to be
provided by science, but by a priori analyses of our ordinary, “folk
psychological,” mental concepts (Armstrong 1968; Lewis 1972; Smart
1971). The idea was that the defining functional roles of mental states were
to be extracted from the huge collections of “common knowledge” about
the mind. For instance, a part of the implicit common knowledge about
pain is that it is typically caused by bodily damage and that it causes
avoidance of the source of pain. This approach can be seen as an extension
of logical behaviorism: mental states are defined a priori, but not just in
terms of observable behavior – they are defined in terms of their causal
relations to stimuli, responses, and other mental states. A large part of the
motivation for this view stemmed from the background assumption that
our common sense ideas about mental states must be approximately true,
or else mental concepts turn out to be meaningless. The most obvious
problem with analytic functionalism is that it seems very likely that at least
some of our commons sense ideas about mental states will turn out to be
fundamentally false (consider for example the case of memory, further
 79

discussed in section 6.3.1). Analytic functionalism is nowadays far less

popular than psychofunctionalism (see, however, Braddon-Mitchell &
Jackson 2003 for a spirited contemporary defense of analytic
functionalism).22
An important distinction that crosscuts analytic functionalism and
psychofunctionalism is the distinction between “role-functionalism” (or
Functional State Identity Theory, FSIT) and “filler-functionalism” (or
functional specification theory). According to role-functionalism, mental
states just are the functionally defined states. For example, being in pain is
identical to the functional state that is defined in terms of what pain causes
and what it is caused by. What realizes being in pain in humans might be
different from what realizes being in pain in octopuses: being in pain is
identical to the functionally defined state, not to any of the realizations or
all of them together. This is the more common view that is closely
associated with psychofunctionalism. It is essentially a metaphysical thesis,
since it makes a claim about the ontology of the mental: mental states are
functional states (Polger 2004).
According to filler-functionalism (Armstrong 1968; Lewis 1972;
1980; Smart 1971), mental states are not functional states, but the physical
states that fill the functionally defined causal roles. For instance, being in
pain is specified by its causal role, but in the end being in pain just is the
physical (neural) state that fills that causal role. “If the concept of pain is
the concept of a state that occupies a certain causal role, then whatever
state does occupy that role is pain” (Lewis 1980, 218). This physical state
can be one thing in humans, another in octopuses, and still something else
in Martians. Therefore, filler-functionalism denies multiple realizability of
mental states: pain in humans is a different state than pain in Martians.
However, these different states are all picked out by the functional concept
“pain,” which non-rigidly designates different physical fillers in different
species. Historically speaking, analytical functionalism and filler-
22
One way of drawing the distinction between analytic functionalism and
psychofunctionalism is the following. In analytic functionalism, the functional
analyses based on common knowledge give the meaning of mental state terms. In
psychofunctionalism, the functional analysis give only “reference-fixing” definitions
of mental terms, and what the states that these terms refer to turn out to be is a matter
of empirical research.
80

functionalism are closely associated, but there is no principled reason why

one could not be a role-analytic-functionalist (or a filler-
psychofunctionalist).
All forms of functionalism face serious objections. The most
notorious problems stem from thought experiments like inverted qualia,
zombies, the “blockhead,” or the “China brain” (Block 1978). There are
also problems related to holism, intentionality, rationality, and
introspection (see, e.g., Levin 2009 or Polger 2004 for more). However,
instead of going through all these problems, I will focus here on the
following questions: If (psycho)functionalism is true, is it a reductionist or
nonreductionist position? Is it compatible with the position that I have
defended in Part I (explanatory pluralism, reduction as mechanistic
explanation, and the interventionist account of causation)?
I will first discuss role-psychofunctionalism. As Fodor pointed out, it
is not compatible with the kind of strict Nagel-reductionism that requires
biconditional bridge laws or identities that go all the way down to the level
of fundamental physics. However, Fodor’s argument may not be a problem
for later, revised, models of theory reduction, which allow for dealing with
multiple realizability (see sections 5.3 and 5.4). Therefore, role-
psychofunctionalism may be reconcilable with some forms of theory
reduction. However, if what I have defended in Part I is right, this is quite
irrelevant, since theory reduction is the wrong framework for considering
psychoneural reduction to start with.
In any case, role-psychofunctionalism is not compatible with New
Wave Reductionism, at least not with Bickle’s (1998) version of it. Role-
psychofunctionalism states that mental states are (ontologically speaking)
functional states, defined by empirical psychology. Bickle claims that the
nature of mental states is resolved by the nature of the intertheoretic
relation between psychology and neuroscience: if psychology reduces
“smoothly” to neuroscience, then most mental states turn out to be
identical to neural states, but if the reduction is “bumpy,” then many of the
mental states have to be eliminated from the scientific ontology. Role-
functionalism is also clearly incompatible with Bickle’s “ruthless”
metascientific reductionism (Part I, Chapter 1), since the latter posits that
 81

mental explanations (and a fortiori mental properties) are eliminable when

neuroscientific explanations are complete.
Role-functionalism is also incompatible with “functional”
reductionism, which will be discussed in the next chapter. Even though the
functional model requires that the properties to be reduced are functionally
defined, a successful functional reduction shows that the properties are
nothing “over and above” the reducing physical properties; they are
ontologically reduced. Insofar as functionalism implies that the ontological
nature of mental states is that they are functional higher-level states, it is
incompatible with this conclusion.
Whether role-functionalism is compatible with reduction as
downward-looking mechanistic explanation, or mechanistic explanation in
general, is a more complicated question. If we read functionalism not as a
metaphysical thesis but as a thesis about the nature of psychological
explanations and generalizations (explanatory functionalism), it is certainly
compatible with mechanistic explanation. In fact, functional
characterization of a phenomenon is often a significant initial step towards
a mechanistic explanation (Craver 2007). The actual function of the state
may turn out to be different from what was originally thought, but this is
not a problem for psychofunctionalism, which does not require that the
mental states should be functionalized a priori.
However, role-functionalism is fundamentally a metaphysical thesis,
and not just a thesis about explanations. Mechanistic explanation, on the
other hand, is largely an antimetaphysical position, and its advocates often
refrain from taking a stance on traditional metaphysical issues. Yet, there is
nothing in mechanistic explanation as such that speaks against role-
functionalism. If we adopt the ontological framework that I will defend in
Part III, mental states can be distinct functional states and real in their own
right, even though they are mechanistically explainable, as long as they are
sufficiently “robust.” In this sense, role-functionalism is compatible with
mechanistic explanation, and with there being reductive (mechanistic)
explanations of mental states.
The filler version of functionalism (the functional specification view)
is at first glance clearly a reductionist position, as it states that mental
82

states are in the end just the neurological realizers of the functional role,
and not functional states themselves. However, this kind of functionalism
is closer to eliminativism than reductionism, insofar as it claims that
functionally defined mental states are not real as such. I will not discuss
this further here, since this reading of filler-functionalism comes very close
to the functional model of reduction, which is the topic of the next chapter.

5.6. The Dream of Nonreductive Physicalism

The main motivation for nonreductive physicalism, including nonreductive

versions of functionalism, was the following. First of all, it was assumed
that the success of reductionism and unity of science would undermine the
status of special sciences and their generalizations (particularly
psychology). This is because an essential feature of traditional (Nagelian)
reductionism is that it denies the independence or autonomy of higher-level
inquiries, as it entails that all the higher-levels laws and theories logically
follow from laws and theories of lower levels. Drawn to the extreme, this
would mean that higher-level theories are just derivatives of physical
theories (or the fundamental physical theory), and thus clearly secondary
or subordinate to them. It was also assumed that a successful reduction of
the special sciences (particularly psychology) would reveal that special
science properties (particularly mental properties) are “just” physical
properties and not properties in their own right.
Fortunately, the argument from multiple realizability entered the
scene and appeared to block both theory reduction and the type identity
theory. This was just what opponents of reductionism needed. Multiple
realizability seemed empirically extremely plausible, and provided the
philosophical justification for the autonomy and irreducibility of
psychology and mental states.
On the other hand, the irreducibility of psychology and mental states,
and especially the apparent failure of the identity theory, seemed to imply
that mental states must be also ontologically distinct from physical states.
This seemed to lead towards some kind of ontological dualism, but hardly
anyone wanted to be a real dualist, or to deny physicalism. Hence
 83

nonreductive physicalism: special sciences and special science properties

must be irreducible, even though everything is physical. Whether any of
the versions of nonreductive physicalism was, or is, a coherent position,
remains an open question. The main problem was (and perhaps still is) to
find a way of spelling out physicalism that would be strong enough to
satisfy our physicalistic intuitions, but weak enough to “save” mental
properties and special science explanations.
Fodor and Davidson23 argued that token physicalism was enough to
save the intuitions about generality of physics and physicalism. According
to token physicalism, every particular thing in the world is a physical
particular (in contrast, type physicalism states that every type or kind in the
world is a physical type or kind). However, it is clear that token
physicalism yields only a very weak kind of physicalism. It is compatible
with the mental and the physical domains being completely uncorrelated –
it only requires that every mental event is a physical event, and this leaves
open the possibility that the mental and the physical domains relate to each
other in a completely random way. It also allows for the possibility that the

23
Donald Davidson (1970) argued for “anomalous monism,” a form of physicalism
(monism) that was supposedly compatible with multiple realizability and the
irreducibility of psychology. Like the position of Fodor, it was supposed to present an
alternative to type identity theory and Nagel-reductionism. Davidson’s argument has
approximately the following structure. We start with three principles about mental
events that are taken to be true: (1) Some mental events causally interact with some
physical events. (2) Events related as cause and effect are covered by strict laws. (3)
There are no strict laws on the basis of which mental events can predict, explain, or be
predicted or explained by other events. It is easy to see that, under some plausible
assumptions, these three principles cannot all be true. However, according to
Davidson, this tension can be resolved by the following thesis: (4) Every causally
interacting mental event is token-identical to some physical event. This leads to
anomalous monism: there are no strict laws for mental event-types (mental
anomalism), but every mental event is identical to some physical event (monism). The
identity here is not the type identity of the identity theory, but token identity, which
merely states that every mental event (token) is identical to some physical event
(token). This position is nonreductive in the sense that it states that there can be no
physical (or non-mental) explanations of the mental. Anomalous monism has been
criticized from many angles, and has been a controversial position from the start. I will
not discuss it in more detail here, since it relies on very strong metaphysical
assumptions about the nature of causation and explanation, and has few proponents
these days.
84

mental is completely and fundamentally inexplicable, since the claim that

every mental event is a physical event does not imply anything about
explaining mental phenomena. It also allows for property dualism, since it
allows physical particulars to have nonphysical properties. Of course, one
might argue that despite these weaknesses token physicalism is enough,
since it does entail that there are no nonphysical entities. However, the
more common and very plausible view is that the token identity thesis at
least needs to be supplemented with something in order to yield an
interesting form of physicalism.
Perhaps the most influential and initially very promising attempt at
trying to add flesh around the bones of nonreductive physicalism was
based on the notion of supervenience (e.g., Davidson 1970; Kim 1982;
1984). The basic idea of supervenience is that properties of type A
supervene on properties of type B if and only if there cannot be a
difference in A-properties without there being a difference in B-properties.
The idea was that even though mental properties are distinct from physical
properties and irreducible, the fact that they supervene on physical
properties could be enough to save physicalism.
Unfortunately, also this turned out to be too optimistic. The fact that
the mental supervenes upon the physical leaves open numerous ontological
possibilities, including property dualism, parallelism, and
epiphenomenalism. For instance, mental-to-physical supervenience as such
is compatible with the view that the mental and the physical are completely
different substances that some divine entity has preordained to be perfectly
coordinated, in accordance to the supervenience relation. Therefore,
supervenience alone (or supervenience + token physicalism) does not seem
to be sufficient to satisfy our physicalistic intuitions. However, some
authors (e.g., Lewis 1983; Jackson 1998) have argued that if we strengthen
the supervenience thesis to (strong) global supervenience, it is sufficient
for formulating physicalism. The claim is that physicalism can be defined
along the following lines (supervenience physicalism): Physicalism is true
at a possible world w iff any world that is a minimal physical duplicate of
w is a duplicate of w simpliciter (Jackson 1998, 12). A minimal physical
duplicate of w is a duplicate that is identical in all physical respects to w,
and does not contain anything additional (such as epiphenomenal
 85

ectoplasm). However, even this may not be sufficient, since, for instance, a
property dualist might claim that the psychophysical laws linking mental
and physical properties are fundamental laws that hold in all nomologically
possible worlds, so that property dualism would be compatible with the
above definition of supervenience physicalism. The issue of supervenience
and physicalism is far from settled; see, for example, Stoljar (2009) or
McLaughlin and Bennett (2008) for more.
Perhaps the most influential argument against nonreductive
physicalism is the exclusion argument, whose most ardent proponent has
been Jaegwon Kim (1993; 1998; 2002; 2005). Very roughly, the exclusion
argument states that mental properties cannot have causal powers of their
own, since the physical domain is causally closed, and all the causal work
is done by the physical properties that realize the mental properties. (I will
discuss the argument in detail in Chapter 11.) Therefore, if mental
properties are irreducible in the sense of being distinct from physical
properties, they cannot make any causal contribution to the physical world.
If Kim is right in this, there are basically two options left for the
nonreductive physicalist. The first is to insist that mental properties are
irreducible but causally impotent properties. This leads to emergent
property dualism and epiphenomenalism, and effectively means giving up
on physicalism. The second option is to accept that mental properties are
identical to physical properties (reductive physicalism), or that they do not
exist at all (eliminativism). However, this effectively means giving up on
the nonreductive part of nonreductive physicalism. Thus, if Kim is right, it
seems that nonreductive physicalism is an incoherent position.
From the point of view of the position I have defended in Part I, the
main problem with the whole debate on nonreductive physicalism is that it
has been based on unrealistic and mistaken ideas about reduction,
reductive explanation and causation. In Part III, I will argue that if we
accept explanatory pluralism and take downward-looking mechanistic
explanation to be the strongest form of reduction in the relevant sciences,
we can in fact have both: autonomous and irreducible (though not
completely independent) special sciences on the one hand, and physicalism
(or something near enough) on the other. I will also argue that the
86

exclusion argument does not threaten higher-level causes if we understand

causation in interventionist terms.
This position, further elaborated in Part III (Chapter 10), could be
seen as a form of nonreductive physicalism, but I consider “pluralistic
physicalism” to be a more appropriate description. The position is
compatible with physicalism in the sense of supervenience physicalism.
However, it states that nonphysical (or nonfundamental) properties are also
real, and that there is no fundamental level of explanation. In a sense, the
dream of nonreductive physicalism might yet come true.
 87

6. Functional Reduction24
In recent years, the “functional reduction model” has become something
like the standard model of reduction in philosophy of mind. The model is
by no means new: its main ideas are already visible in the analytic filler-
functionalism of Lewis (1972). Lewis’ idea was that a given mental state
M is defined functionally in terms of its causal role, but M is nothing more
than the physical states that occupy this role (see also section 5.5. and
Walter & Eronen 2011). More recently philosophers like Levine (1983;
1993), Chalmers (1996), Jackson (Chalmers and Jackson 2001), and Kim
(1998; 2005) have presented somewhat varying models of functional
reduction based on this general approach. All of these authors have then
applied the supposedly general model of reduction to the problem of
phenomenal consciousness, arguing that phenomenal properties are
fundamentally irreducible, or that there is an “explanatory gap” between
phenomenal properties and the physical domain.
I will focus here on Kim’s model of functional reduction, since it is
the most explicit and detailed one. I will argue that the functional model
fails to capture the role and nature of reductive explanation in science
(particularly psychology and neuroscience). Furthermore, I will show that
if we try to revise the functional model in order to make it more
scientifically credible, it turns out that the revised model is not
significantly different from mechanistic explanation. In the next chapter, I
will consider the consequences of this for the explanatory gap argument.

6.1. The Causal Exclusion Argument and the Functional Model

Kim’s main motivation for invoking the model of functional reduction is to

show that mental properties (with the exception of phenomenal properties)
can be saved from the causal exclusion argument, which I will briefly
sketch here. Several different versions of the argument exist; the
formulation here reflects Kim’s most recent accounts (Kim 2002; 2005).
24
This chapter is based on an article published in Philosophia Naturalis (Eronen 2010-
2011).
88

The argument is based on certain principles that together create a problem

for mental causation (Kim 2002, 278):

The Problem of Mental Causation: Causal efficacy of mental properties is

inconsistent with the joint acceptance of the following four claims: (1) physical
causal closure, (2) exclusion, (3) mind-body supervenience, and (4)
mental/physical property dualism (i.e., irreducibility of mental properties).

The principle of physical causal closure states that every physical

occurrence has a sufficient physical cause. The principle of exclusion
states that no effect has more than one sufficient cause, except in cases of
genuine overdetermination, such as two bullets hitting the heart of a victim
at exactly the same time, both causing death.
It is easy to see how the four principles taken together lead to
trouble. Let us start by assuming that (the instantiation of) a mental
property M causes (the instantiation of) another mental property M*. Due
to mind-body supervenience, M supervenes on some physical property P,
and M* supervenes on some physical property P*. Since M* supervenes on
P*, M* must be necessarily instantiated whenever P* is instantiated, no
matter what happened before: the instantiation of P* alone necessitates the
occurrence of M*. Thus, according to Kim, the only way that M can cause
M* is by causing P*.
This is where the principle of causal closure kicks in: P* must also
have a sufficient physical cause. This means that P* has a sufficient
physical cause P and a mental cause M, and the exclusion principle states
that one of these must go – if we would accept cases like this as genuine
overdetermination, we would get massive overdetermination of physical
effects by mental causes, which is highly implausible. Obviously M is the
one that has to go, since if M was the only cause of P*, this would violate
the principle of physical causal closure. Therefore, M cannot be the cause
of M* or of any other mental or physical property. This holds for all mental
properties, and we have the striking conclusion that, under mind-body
supervenience, mental properties are causally impotent.
 89

According to Kim, physical causal closure and mind-body

supervenience are among the inescapable commitments of all physicalists.
The exclusion principle is taken to be a general metaphysical constraint
that can hardly be challenged. This leaves only mental/physical property
dualism (i.e., the irreducibility of mental properties) as the principle that
has to go. Therefore, Kim’s conclusion is what he calls “conditional
reductionism”: “If mentality is to have a causal influence in the physical
domain – in fact, if it is to have any causal efficacy at all – it must be
physically reducible” (Kim 2005, 161).
What does reduction then amount to? Kim’s answer is the functional
model:

To reduce a property, say being a gene, on this model, we must first

“functionalize” it; that is, we must define, or redefine, it in terms of the causal
task the property is to perform. Thus, being a gene may be defined as being a
mechanism that encodes and transmits genetic information. That is the first
step. Next, we must find the “realizers” of the functionally defined property –
that is, properties in the reduction base domain that perform the specified causal
task. It turns out that DNA molecules are the mechanisms that perform the task
of coding and transmitting genetic information – at least, in terrestrial
organisms. Third, we must have an explanatory theory that explains just how
the realizers of the property being reduced manage to perform the causal task.
In the case of the gene and the DNA molecules, presumably molecular biology
is in charge of providing the desired explanations. (Kim 2005, 101)

Let us take a closer look at this model (Kim 1998, 97-103; 1999, 10-13).
The reduction of property M consists of three steps:
Step 1: M must be functionalized – that is, M must be construed, or
reconstrued, as a property defined by its causal/nomic relations to other
properties. As Kim puts it:

[W]e must first ‘prime’ M for reduction by construing, or reconstruing, it

relationally or extrinsically. This turns M into a relational/extrinsic property.
For functional reduction we construe M as a second-order property defined by
its causal role – that is, by a causal specification H describing its (typical)
90

causes and effects. So M is now the property of having a property with such-
and-such causal potential[.] (Kim 1998, 98)

Thus, property M is defined as a “second-order” property: it is a property

that some first-order properties have.
Step 2 consists of finding the realizers of M. These are the first-order
properties in the reduction base domain that have the right causal/nomic
relations, i.e., the properties that fit the causal specification H. The
realizers can be different in different systems, allowing for multiple
realizability. Step 2 is a matter of scientific research, or as Kim puts it, “a
scientifically significant part of the reductive procedure” (Kim 1999, 11).
Step 3 is to find a theory that explains how the realizers actually
perform the causal role specified in Step 1. Like Step 2, Step 3 is also a
matter of scientific research, and these steps are intertwined, since figuring
out what the realizers of M are certainly involves theories about the
causal/nomic relations in the reduction base.
One of the central points of Kim’s account is that functionally
reduced properties are nothing ”over and above” the reducing properties:
”Central to the concept of reduction evidently is the idea that what has
been reduced need not be countenanced as an independent existent beyond
the entities in the reduction base – that if X has been reduced to Y, X is not
something ‘over and above’ Y” (Kim 1999, 15). According to Kim, this
means that reduction has to lead either to identities (conservative
reduction) or eliminations (replacement / eliminative reduction). Is
functional reduction then conservative or eliminative?
First of all, Kim argues that when M has been functionally reduced to
P, instances of M can be identified with the instances of P (Kim 1999, 15-
16). He invokes the “causal inheritance principle,” which states that ”[i]f a
functional property [M] is instantiated on a given occasion in virtue of one
of its realizers, [P], being instantiated, then the causal powers of this
instance of [M] are identical with the causal powers of this instance of
[P].” If we accept this principle, it follows that the instances of M and P
have exactly the same causal powers, and it is hard not to identify the
instances, since if they were not identical, the difference could not even be
 91

detected. However, what is at issue in the exclusion argument is not token

causation (one instance or event causing another instance or event), but
type causation. The problem is whether mental properties can have causal
powers – in other words, whether some event can cause a physical event in
virtue of being an instantiation of a mental property. Therefore, for
avoiding the exclusion argument it is not enough that instances of M are
identical to instances of physical properties, also the property M itself has
to be identical to a physical property P.
The situation is made even more complicated if (as is generally
assumed) M can have multiple realizers. Kim sees only two options for
dealing with multiple realizability: we can (1) identify M with the
disjunction of its realizers, or (2) give up M as a real property and only
recognize it as a property designator that picks out many different
properties (the realizers of M).
Identifying M with the disjunction of its realizers is notoriously
problematic. The realizers must have different causal roles, since otherwise
they wouldn’t be different realizers (Kim supports a causal theory of
properties). If M is identical to a set of causally and nomologically
heterogeneous properties, Kim reasons, then M itself must be causally and
nomologically heterogeneous, and is unfit to figure in laws, and is thereby
not a scientific property (see Kim 1992 for more details of this
argument).25
Therefore, Kim is inclined to accept the second option:

One could argue that by forming ‘second-order’ functional expressions by

existentially quantifying over ‘first-order’ properties, we cannot be generating
new properties (possibly with new causal powers), but only new ways of
indifferently picking out, or grouping, first-order properties, in terms of causal
specifications that are of interest to us. (Kim 1999, 17)

25
Esfeld and Sachse (2007) have argued that by introducing functional sub-types we
can have property identities and conservative functional reductions, multiple
realizability notwithstanding.
92

This makes functional reduction eliminative: we have to accept that mental

properties are not genuine properties in their own right. Kim accepts this
only because the other alternatives (disjunctive identities or property
dualism) are wrought with major philosophical problems (Kim 2008, 112).
I will return the problems of this option in section 6.3.2 below.

6.2. Kim vs. Nagel

Kim presents the functional model as a better and scientifically more

credible alternative to the classic but problematic Nagelian model: ”Nagel
reduction of pain requires an all-or-nothing, one-shot reduction of pain
across all organisms, species, and systems. It is clear that functional
reduction gives us a more realistic picture of reduction in the sciences”
(Kim 2005, 102). “I believe most cases of interlevel reduction conform to
the model I have just sketched” (Kim 1998, 99).
It is interesting that the problems that Kim sees in Nagel’s model are
quite different from the problems that have vexed philosophers of science.
As I have pointed out in Part I, in philosophy of science, Nagel’s model
was considered to be in need of revision mainly for the following reasons:
it failed to account for many clear cases of reduction, it could not
accommodate the fact that the reducing theory often corrects the theory to
be reduced, and many scientific theories cannot be formalized in a way that
the model requires.
However, Kim’s main problem with Nagel’s model is that it gives us
reductions that do not explain (Kim 1998, 90-97; 2005, 98-101). This is
because, according to Kim, the reductive work in Nagel’s model is done by
the biconditional bridge laws that connect properties of the reduced theory
with properties of the reducing theory, and these bridge laws are just
“unexplained auxiliary premises” that are themselves in need of
explanation. Furthermore, Kim points out that the existence of
biconditional bridge laws between the mental and the physical is
compatible with various different ontological positions, including property
dualism, epiphenomenalism, and identity theory. Because of this, Nagelian
 93

reduction does not by itself yield ontological simplification, which

(according to Kim) is essential for reduction. Therefore, a model of
reduction that focuses on bridge laws is not explanatory and does not give
us the answer to the question of the ontological status of the mental.
However, what Kim seems to ignore is that in the later and more
sophisticated models of intertheoretic reduction that succeeded Nagel’s
model, particularly the New Wave model (see Part I, Chapter 1), bridge
laws play no essential role.26 In the New Wave model, reduction is still a
relation between theories and involves logical deduction. The crucial
difference to Nagel’s model is that what is deduced from T1 is not the
theory to be reduced itself (T2), but an analogue (or “equipotent image”) of
it (T2a). Importantly, the analogue can be formulated entirely in the
vocabulary of theory T1. The fate of theory T2 and its ontological posits is
then determined by the relation between T2 and the image T2a. If the
analogy is strong and not much correction is needed, T2 is reduced
“smoothly” to T1, and many of its ontological posits can be retained. If the
theories are only weakly analogical and the amount of correction implied
to T2 is considerably large, the reduction is “bumpy,” and many or all of
the ontological posits of T2 will be eliminated. In this model, bridge laws
do not play a central role. If the reduction is very smooth and T2 and T2a
are isomorphic, then perhaps this warrants postulating bridge laws between
terms of T2 and T2a, but these are special cases, and even in these cases
bridge laws are not essential: the essential part of the reduction is the
deduction of the image T2a from T1 (see Marras 2002 for more).
Hence, the New Wave model avoids both problems that Kim points
out in Nagel’s model. First, the bridge laws connecting T2 and T1 are not an
important part of the reductive procedure, and thus even if it is true that
they are “unexplained auxiliary premises,” this is quite irrelevant.
Secondly, reductions on this model do give us the answer to the
ontological status of the mental: if the reduction turns out to be smooth,
many mental properties (or their analogues) are retained in the ontology of

26
According to Marras (2002, 237), bridge laws are not central even in Nagel’s own
articulation of the model. See also van Riel (2011) for further criticism of the
traditional interpretation of Nagel’s theory of reduction.
94

the reducing theory; if the reduction turns out to be bumpy, many or all
mental properties are eliminated from the scientific ontology.
The New Wave model and intertheoretic models are not applicable
to the purported reduction of the mind or psychology – but for reasons that
Kim does not mention or discuss. As I have shown in Part I, they do not
work because they require that both the theory to be reduced and the
reducing theory are construable in some formal or semi-formal way, either
as sets of sentences (the “received view” of theories), or as sets of models
meeting certain set-theoretic conditions (the structuralist/semantic view of
theories). The problem is that neither psychologists nor neuroscientists are
in the business of formulating theories of this kind, and it would require
considerable philosophical violence to reconstrue the explanatory
frameworks of psychology and neuroscience in a formal or semi-formal
way.
In this sense, the model of functional reduction is prima facie
promising. It does not seem to require formal theories, since it is a model
of property reduction, not theory reduction. But is it really a realistic model
of reduction to which most cases of interlevel reduction conform?

6.3. Dissecting the Functional Model

In recent years, the functional model has been criticized from several
angles. Ausonio Marras (2002; 2005) has argued that when we analyze the
model carefully and accept certain plausible background assumptions, it in
fact leads back to Nagel reduction, which it was supposed to replace. In the
same vein but with different arguments, Max Kistler (2005) has argued that
functional reduction requires local bridge laws that are left just as
unexplained as in a Nagel reduction. John Bickle (2008; personal
communication) does not criticize the model itself, but points out that it is
based almost entirely on logical and metaphysical considerations, and that
the examples given to support it reflect an elementary school
understanding of science. In this sense, the functional model is a step
backward from intertheoretic models, which were at least based on science
(though not psychology and neuroscience).
 95

I will develop the last line of argument in more detail, and show that
from the point of view of philosophy of science and scientific practice, the
functional reduction approach is unacceptable. I will focus on three salient
problems of the model: 1) Where do the functional definitions of the
properties to be reduced come from? (2) What is the “realization” relation
between the property to be reduced and the realizing properties? (3) What
notion of causation does the model require? These are by no means the
only problems or points that need clarification, but they suffice to show
why the model fails as a general account of reduction.

6.3.1. Functionalization
As we have seen, Step 1 in the functional model consists in defining or
redefining the property to be reduced in terms of its causal role. However,
it is not clear how we get the causal definition of the property to be
reduced. Kim seems sympathetic to the view of Chalmers and Jackson
(2001) and Levine (1993), according to which reductive explanation
requires analytic definitions grounded in (a priori) conceptual analysis (see
Kim 2005, Ch. 4). The first step of functional reduction would thus consist
in finding the analytic definition for the property to be reduced through
conceptual analysis.
However, if the functional definitions of the mental properties are to
be based on conceptual analysis that is (at least relatively) a priori, this
leads to a fundamental problem: our a priori ideas about psychological
states or processes are often simply wrong. Consider for example memory.
An armchair conceptual analysis would indicate that memory is some kind
of a simple storage, where our past experiences are waiting for retrieval –
Plato compared memory to an aviary of birds, from which we take the
correct bird when memory retrieval is successful, and the wrong bird when
it is not. However, scientific research has revealed that memories are not
just retrieved, but actively constructed, and subjectively compelling
memories sometimes turn out to be radically inaccurate (Neisser & Harsch
1992). Furthermore, memory comprises several subsystems (short term
memory, long term memory, episodic memory, visual memory, etc.),
96

which neither individually nor taken together correspond to the simple

storage envisioned by a priori analysis (see Bechtel (2008, Ch. 2) for
discussion). Similar considerations apply to pain (Hardcastle 2001), which
has for decades been a standard example in philosophy of mind.
It is thus clear that mere conceptual analysis is not sufficient for
working the properties “into shape” for reduction. One has to either allow
for scientific revision of common sense definitions of mental properties, or
simply focus on properties as defined by empirical psychology.27
Furthermore, in both cases we have to allow for the revision and
adjustment of the definitions as science proceeds. Such revision and
interplay across levels is commonplace in science. One of the first
philosophers to emphasize the importance of this co-evolution of concepts
and theories was Wimsatt, drawing from scientific practice in biology:

A lower-level model is advanced to explain an upper-level phenomenon which

it doesn’t fit exactly. This leads to a closer look at the phenomenon, and
perhaps results in some change in the way in or detail with which it is
described. This will also lead to changes in the lower level model and may
suggest new phenomena to look for. (Wimsatt 1976a, 231)

Also Bechtel and Richardson (1993) have described in detail the

complexities involved in characterizing the phenomena to be explained,
based on detailed analyses of cases from history of biology, and one of
their points is that scientists often have to constantly redefine the
phenomena they are trying to explain. More broadly speaking, in the
mechanistic explanation paradigm (see Chapter 2), a crucial point is that
there is constant interplay between different levels of explanation, and new
discoveries at lower or higher levels can lead to a novel understanding of
the function of the system under consideration.
There is also a further problem related to functionalization, even if
take empirical psychological properties to be the targets of reduction and
allow for constant revision of their functional definitions. It is quite

27
This problem is obviously related to the issue of common-sense (analytical) vs.
empirical functionalism (psychofunctionalism), discussed in section 5.5.
 97

possible that in the end we are unable to find any neuroscientific properties
playing the causal role of some psychological properties, and thus we
cannot functionally reduce them. The easiest solution in these cases would
be to revise the functional definitions of the psychological properties, but
this is not always justifiable. We might want to retain some psychological
properties more or less as they are, since they are useful in scientific
explanations. For example, Khalidi (2005) takes up the psychological
property of fear, and shows (based on empirical results in cognitive
neuroscience) that distinctions made at the neurophysiological level cross-
cut the distinctions made at the psychological level. That is, from the
vantage of neurophysiology, there is nothing playing the functional role
associated with the psychological state of fear. Importantly, this is not a
case of multiple realizability, which is a one-to-many relationship. In this
case, there is simply just mismatch: a “one-to-none” relationship.
However, we would not want to eliminate or revise the psychological
concept of fear, since it still plays an important role in research and
scientific explanations.
In this case, it seems that there are no neurophysiological states
playing the causal role of fear, and the option of redefining fear does not
seem very fruitful. Hence, Step 2 in functional reduction of fear fails. But
should we conclude from this that fear is fundamentally irreducible and
threatened by the exclusion argument? Or should we eliminate the property
of fear from our ontology? Both options seem implausible. The framework
of functional reduction seems unsuitable for dealing with situations like
this.
Certainly the basic idea that the properties to be reduced have to
specified causally is correct and in accordance with scientific practice.
However, functionalization is not just a matter of conceptual analysis, it is
not even remotely an a priori matter, and functional definitions can change
as research proceeds. Furthermore, in some cases we might not be able to
find neural realizers that play the functional role definitive of a mental
property. This does not mean that Kim’s functional model is fundamentally
wrong, but it surely is too simplified in this respect.
98

6.3.2. Realization
The second step in Kim’s account of functional reduction is finding the
‘realizers’ of the functionally defined property to be reduced. But what
makes some property a realizer of another property? How should we
understand this realization relation? And what sorts of things are the
realizers of mental properties?
The roots of talk of ‘realization’ in philosophy of mind go back to
multiple realizability (see Chapter 5, section 5.3). In the debate that
followed, very little attention was initially paid to the notion of realization
itself. However, as I have already pointed out, several philosophers (e.g.,
Polger 2004; 2007a; Shapiro 2000; 2004) have recently shown that the
realization relation is much more problematic than has been generally
assumed. One problem is that there seems to be no general realization
relation that applies to all the different cases that have been presented as
paradigmatic cases of realization. However, it might be that Kim’s account
does not need any general notion of realization, and that a more “local”
notion would suffice. In this section I will show that even if we limit the
discussion to psychological properties and their realizers, and accept that
there is no general notion of realization, Kim’s notion of realization leads
to problems.
First of all, it is not entirely clear whether Kim’s notion of realization
is “flat” or “dimensioned”.28 According to the flat view, both the realized
and the realizer properties must be instantiated in the same individual, at
the same level of composition. According to the “dimensioned” view, the
realized property may also be at a higher level of composition than the
realizer properties. Sometimes Kim seems to be explicitly supporting the
flat view: “It is evident that a second-order property and its realizers are
at the same level … they are properties of the very same objects” (Kim
1998, 82). Also one of Kim’s answers to the “generalization argument”
that threatens the exclusion argument relies on flat realization (see Walter
2008 for more).
However, the “flat” or intralevel view of realization is hard to
reconcile with Kim’s (2005) later exposition of the functional model,

28
This distinction was introduced by Gillett (2002; 2003).
 99

where he claims that his account solves the problem of “explanatory

ascent,” which is the problem of how (reductive) explanations can move
from one level to the other, so that higher level phenomena are explained
by resources from lower levels. His solution and the whole discussion do
not make sense unless he takes realizers to be at a lower level than the
realized properties: how could functional reduction solve the problem of
explanatory ascent if both the reduced and the reducing property are at the
same level? However, what Kim has to say elsewhere supports the flat
interpretation.29
The issue of flat vs. dimensioned realization has grown into a large
debate, and I will not delve into the details here. Regardless of whether his
notion of realization is flat or not, Kim’s view leads to problems. Let us
consider the case of mental properties and their neural realizers. Mental
properties are to be functionally defined in terms of their causal relations to
other mental properties. What is it then for a neural property to realize a
mental property? According to Kim, the realizers have to perform the
causal task specified in Step 1, that is, they have to “occupy” or “fill” or
“play” the causal role definitive of the mental property.
But what does this mean? If we take the realizers to be properties, it
seems that the only way to make sense of this is that the realizing neural
property is embedded in a causal structure that is isomorphic to the causal
structure in which the mental property is embedded. That is, the causal
“context” of the neural property is isomorphic to the causal “context” of
the mental property. What else could it mean for the neural property to
occupy the causal role definitive of the mental property?
However, this leads to problems, since Kim’s aim is to reduce all
(non-phenomenal) psychological properties, not just one of them. This
implies that, in order to accomplish a psychoneural reduction, we would
have to figure out the causal roles of all the mental properties we want to
reduce. This would yield a vast network of mental properties and their
causal relations to each other. The task would then be to find an
isomorphic causal structure among the neural properties. If we also assume
29
Here I will just ignore the problem of “explanatory ascent” – anyway the whole
problem arises only because Kim believes that explanations have to follow something
resembling the deductive-nomological model of explanation.
100

that laws underlie causal relations, and that theories are sets of laws (both
assumptions are controversial, but commonly accepted in philosophy of
mind), the implication is that Kim’s model comes very close to theory
reduction: in order to reduce a psychological theory, we need to find in (or
derive from) the neuroscientic theory a structure that is isomorphic to the
psychological theory. This is not so different from the New Wave model of
reduction, where a psychological theory is reduced by deriving from
neuroscience an “analogue” or “equipotent image” that is isomorphic to the
psychological theory.
Marras (2002; 2005) makes a similar point with a somewhat
different reasoning: in a closer analysis, Kim’s model turns out to be a
model of intertheoretic reduction. If this is the case, the functional model
only appears to be an advance over the intertheoretic models, and faces
exactly the same problems (see Part I, Chapter 1).
Another fundamental problem with Kim’s notion of realization was
already mentioned at the end of section 6.1: if we accept multiple
realizability, the realized properties have to be either identical to the
disjunction of the realizers, or just concepts (or predicates or designators).
Kim rejects the first option for philosophical reasons and accepts the
second one. However, in the context of realization, the problem with the
second option is that it seems to leave no room for the idea that neural
properties realize mental properties. According to the second option, the
mental concepts simply (non-rigidly) designate different neural properties
in different species, just like in Lewis’ (1972) filler-functionalism. If this is
true, there is no realization relation here. Mental properties cannot be
realized, since there are no mental properties, just mental concepts (or
property designators) that group physical properties in interesting ways.30
Mental concepts cannot be realized, since concepts in general are not the

30
This distinction between real properties and “mere concepts” is problematic in
general. One of the points of the pluralistic approach I am defending in Part III is that
such a distinction does not make sense: a concept that groups together properties in a
scientifically useful way is not “just a concept;” it picks out a real property. Perhaps
one further solution more congenial to Kim’s approach would be to argue that mental
properties are some special kind of “abstract” properties. However, Kim does not
appear to seriously consider such a solution. In any case, it would require developing
or spelling out the metaphysics for such properties, which is no easy task.
 101

sorts of things that are realized. But if this is the case, the whole talk of
realization has been misleading, and the claim that the functional model
can accommodate multiple realizability turns out false.31
Perhaps, however, there are yet other ways of understanding
realization. As Polger and Shapiro (2008) have pointed out, one
problematic assumption that underlies many of these issues is the
assumption that the realizers have to be properties. Particularly in more
recent writings, Kim himself has been less strict and allows the realizers to
be mechanisms: “Find the properties (or mechanisms) in the reduction base
that perform the causal task C” (Kim 2005, 102, my emphasis).
If we (unlike Kim) take this idea of mechanistic realization seriously,
it leads to a more complicated picture of mental realization than the one the
functional model presents. The idea is that a functionally (causally) defined
psychological state, property, or capacity is realized by a neural
mechanism that plays that functional role. A crucial aspect of this kind of
mechanistic realization is the multilevel nature of the mechanisms: on any
reasonable understanding of neural mechanisms, they have to be
hierarchically organized into levels. Therefore, instead of a simple two-
level model with the mental property and its neural realizers, we have a
more complicated picture where the realizer is also organized into levels.
An often-cited example of a psychological property or capacity that is
realized by a (multilevel) neural mechanism is memory consolidation (see
Part I, Chapter 2). As Wilson and Craver (2007) point out, this comes close
to how the term ”realization” is used in the cognitive sciences: when
scientists state that they are looking for, say, the neural realization of
memory consolidation, what they typically mean is that they are looking
for the neural mechanism of memory consolidation.

31
In fact, Kim sometimes seems ready to reject the multiple realizability of mental
properties and argues for “species-specific identities,” such that “multiply realized
properties are sundered into diverse realizers in different species and structures” (Kim
1998, 105). This leads to problems if there is also multiple realizability within species
or structures: it seems to follow that mental properties are spliced into properties
restricted to very specific neural or physical structures, and it is hard to see how such
properties could be relevant in scientifically explaining human behavior. See also
section 5.3 for more on multiple realizability.
102

Understanding realization in this way has important consequences.

This weaker notion of realization differs fundamentally from the notion of
realization in traditional philosophy of mind. On the mechanistic account,
there is no special realization relation in addition to the relation between
the overall function and the mechanism that performs it. The orchestrated
behavior of the mechanism simply results in the overall function, and there
is no further “realization” involved. In this sense, talk of realization can be
understood as merely metaphoric; the notion of realization is not doing any
metaphysical work.32
Furthermore, instead of making a distinction between realized
properties and realizer properties, it is perhaps more appropriate to
consider psychological properties simply as higher-level properties of
neural mechanisms. For example, it is quite natural to consider
psychological properties of memory consolidation as properties at the
highest level of the memory consolidation mechanism. In this sense, they
are neither identical to the realizing mechanism nor “just concepts” – they
are real higher-level properties (real in the sense to be discussed in Part
III). The general idea is that psychology defines and discovers functional
properties that are then integrated into multilevel mechanistic explanations.
The key requirement for “realization” in the functional model is that
it would somehow save mental causation. On Kim’s account, the only way
this could work is that the “realized” properties turn out to be just concepts.
Now if we adopt the mechanistic approach outlined above, and see the
“realizers” as multilevel mechanisms, what happens to mental causation?
Aren’t the multilevel mechanisms problematic regarding causation? In the
next section, I will argue that the answer is no.

6.3.3. Causation
As we have seen, the properties to be reduced are defined by their causal
roles; they are reduced by finding the first-order properties that have that
causal role; the aim of functional reduction is to save mental properties

32
In general, I am skeptical about the usefulness of the notion of realization; it does
not seem to ”cut nature at its joints”. See also Polger and Shapiro (2008) for more on
this point.
 103

from the causal exclusion argument; reduced properties have no causal

powers of their own, and so on. Causal notions seem to play a key role in
Kim’s account. Indeed, the whole motivation for developing the functional
model comes from the causal exclusion argument and from worries
regarding the causal efficacy of mental properties. But what is causation?
What does it mean to say that X causes Y?
The kind of notion causation Kim has in mind is very strong and
robust:

We care about mental causation, it seems to me, chiefly because we care about
human agency, and evidently agency involves a productive/generative notion of
causation. An agent is someone who brings about a state of affairs for reasons.
If there indeed are no productive causal relations in the world, that would
effectively take away agency—and our worries about mental causation along
with it. (Kim 2009, 44)

As the quote indicates, Kim thinks of causation as a relation where the

cause generates, produces, or brings about the effect. According to Kim, a
weaker account of causation in terms of, for example, counterfactual
relations would not be satisfactory, since we would still need the
metaphysical account of what makes the counterfactuals we want for
mental causation true (Kim 1998, 71).
In Part I, I have defended the interventionist account of causation. In
Part III, I argue that if we understand causation in interventionist terms,
there is no reason to think that mental causes are excluded by neural
causes. However, the interventionist account seems to be exactly the kind
of “weak” account that Kim finds unsatisfactory. Here I will argue that it in
fact is satisfactory, and that there is no reason to require a more robust
notion.
The main problem is that the stronger notion of causation would
have to be somehow grounded in physics. In the end, the metaphysical
question that Kim wants to answer is how there could be mental causes in a
fundamentally physical world. If the stronger notion of causation was not
grounded in physics, it is hard to see what reason there would be to prefer
104

it to the interventionist account, assuming that the latter captures the notion
of causation as it is needed in science and everyday life.
The problem with grounding causation in physics is that notions like
cause and effect do not really play a role in our best physical theories (as
famously argued by Bertrand Russell (1912-13), and more recently by
Ladyman and Ross (2007), Loewer (2007), Norton (2007), and many
others). The fundamental laws of physics relate the totality of a physical
state at one time to the totality of the physical state at later instants, but do
not single out causes and effects among these states. If we want to find
causes that “bring about” or “produce” their effects, or causes that are
“sufficient” for their effects, we have to consider something like the entire
state of the universe as the cause for even a small effect.33
Of course, we can put labels onto relata that appear in physical
equations and call some of them causes and others effects, but this is
entirely superfluous to the physics itself. There is no “principle of
causality” that would in any way guide or restrict physical theory
formation. Furthermore, there are cases even in Newtonian physics which
go straight against our ideas of causation – for instance, effects that take
place with no observable causes (Norton 2007) – not to even speak of
phenomena like quantum entanglement.
John Norton (2007) proposes, based on these considerations, that we
should view causation as a “folk” science. In the same way as fundamental
physical theories under right circumstances yield something like
Newtonian physics, which is strictly speaking a false theory, they also
under right circumstances yield something like our “folk” notion of
causation. Perhaps this reasoning could also be applied to the relation
between fundamental physics and interventionist causation, but this goes
beyond the scope of this thesis.
The interventionist account seems to capture the nature of causation
both in special sciences and everyday life very well, and in fundamental

33
Or at least the state of the universe on the surface of a sphere with a radius of about
300 000 000 meters centered on the effect, assuming that the cause precedes the effect
by one second – the speed of causal influence cannot be faster than the speed of light
(see Loewer 2007 for more).
 105

physics, causal notions are unnecessary and superfluous.34 In Part III

(Chapter 11), I will show that the problems of causal exclusion can be
avoided in the interventionist framework. Do we still need a notion of
causation that shows that mental agency is possible in the sense that
mentality produces or generates physical effects in some metaphysically
strong sense? I think not. From a scientific point of view, the search for the
true nature of causation can be seen as just a metaphysical exercise. As
Woodward (2008, 249) puts it: “We are thus left with possibility that the
only people who think that vindicating the claim that mental states are
causes requires showing that they are causes in a richer, more metaphysical
sense are certain philosophers of mind.”
Thus, with a correct understanding of causation, a large part of the
motivation behind functional reduction disappears. Kim wanted to show
that the mental is functionally reducible in order to save mental causation.
However, it seems that mental causation does not need such a rescue
operation: mental causation in the interventionist sense is no more
problematic than any other kinds of causation (Chapter 11), and the search
for metaphysical (productive, generative, sufficient, etc.) mental causes is
pointless.
What is then the motivation for reducing or reductively explaining
the mental? I think the correct answer is that we want to reductively
explain the mental because we want to explain everything there is to
explain, and some kind of reductive explanation seems to be very fruitful
in this context, as the success of neuroscience in recent decades shows.
But what exactly is the nature of this explanatory enterprise?

34
Not all philosophers of physics would agree that there is no causation in
fundamental physics (see, e.g., Frisch, 2009). However, even if it turns out that causal
notions do play some role in fundamental physics, it is still the case that there is
currently no metaphysically robust and physically grounded account of causation that
would be suitable for considering mental causation and a serious alternative to
interventionist causation.
106

6.4. Functional Reduction as Mechanistic Explanation

Perhaps the functional model could be revised, taking into account all that
has been said above, in roughly the following way. We want to reduce
mental property M. First, we have to find out what the functional role of M
is. However, this is not done through conceptual analysis alone, but
through the interplay of conceptual analysis and empirical research. Also,
it is an ongoing process, and the initial definitions may be refined later.
This first step is not necessarily temporarily prior to the next steps, and
anyway the whole process is integrated and all the steps are intertwined. In
the second step, we figure out what the neural mechanism that is the
“realizer” of M is. M is neither identical to its realizer nor “just a concept”
– M can be seen as a higher-level property of the neural mechanism. Third,
we construct the “theory” that explains why the mechanism is the realizer
of M – that is, we show how the functioning of the mechanism results in M
(i.e., how the mechanism performs the functional role of M).
This quickly sketched revised account of functional reduction looks
very much like mechanistic explanation. As we saw in Part I, the basic idea
of mechanistic explanation is that explaining a phenomenon requires
describing the mechanism (understood as a composite hierarchical system)
that accounts for the phenomenon. This is exactly what is going on in the
revised model of functional reduction. This suggests that functionally
reducing property M amounts to providing a mechanistic explanation for
M. The upshot is that if we want to keep the model of functional reduction
close to science, it turns out that there is no functional reduction over and
above mechanistic explanation.
What does replacing the functional model with mechanistic
explanation mean for the questions of reduction and causation? First of all,
it is important to remember that the main reason for being an ontological
reductionist (at least for Kim) is the causal exclusion argument. If the
problem does not arise when we understand causation in interventionist
terms, then also the motivation for being a strong reductionist fades away.
The mechanistic explanation model, conjoined with the interventionist
account of causation, does not involve the kind of strong ontological
reduction in terms of property identities or eliminations that Kim is after,
 107

since it emphasizes the multilevel nature of mechanisms, and the causal

and explanatory relevance of nonfundamental things.
As I have pointed out in Part I, many philosophers (e.g., Bechtel
2008; Sarkar 1992; Wimsatt 1976a; 2007) have argued that the process of
“looking downward” and invoking parts of the mechanism to understand
the behavior of the mechanism as a whole is close enough to what
scientists generally take to be a reductive explanation to warrant treating
the downward-looking aspect of mechanistic explanation as a kind of
reductive explanation. On the other hand, Craver (2007) considers the
framework of mechanistic explanation antireductive. This issue is mainly a
terminological one, but I see no harm done calling downward-looking
mechanistic explanation reductive explanation, as long as it is clearly
distinguished from stronger forms of reduction.
Regardless of whether we want to call mechanistic explanation
reductive explanation, this approach supports a kind of causal and
explanatory pluralism: higher-level properties (including psychological
properties) do have causal and explanatory relevance, and need not be in
any strong sense reducible to (i.e., identified with) lower-level entities and
properties. However, before discussing this kind of pluralism in more
detail (Part III), I will turn to some further problems related to reduction in
philosophy of mind: the explanatory gap and the new type physicalism.
108
 109

7. Phenomenal Consciousness and the Explanatory Gap35

One of the central topics in philosophy of mind is phenomenal
consciousness and its place in nature. Phenomenal consciousness is the
what-it’s-like aspect of consciousness: what it is like to see green, smell
grass, taste whiskey, and so on. Nearly all philosophers of mind agree that
there is something it is like, for example, to see green. States of
phenomenal consciousness, such that it is like something to be in that state,
are phenomenal states. Phenomenal state types are often also called
phenomenal properties or qualia.
The epistemological and ontological status of these states or
properties has been one of the main topics in recent philosophy of mind. In
this chapter, I will critically examine arguments that purport to show that
there is an explanatory gap between phenomenal consciousness and the
physical domain. Some philosophers (e.g., Levine 1983; 1993) argue that
phenomenal properties are physical (neural) properties that cannot be
physically explained (epistemological gap). Others (e.g., Chalmers &
Jackson 2001, perhaps also Kim 1998; 2005) go even further and argue not
just that phenomenal properties cannot be physically explained but also
that they are in some sense beyond the physical domain (epistemological
and ontological gap). Here I focus on problems in the arguments for the
epistemological gap.
The starting point of the explanatory gap argument(s) is Kripke’s
(1980) argument against psychoneural identities. First of all, Kripke
showed that if ‘A’ and ‘B’ are rigid designators (i.e., referring to the same
thing in all possible worlds), then if “A = B” is true, it is necessarily true
(i.e., true in all possible worlds). It is easy to see why: If ‘A’ refers to the
same thing in every possible world and ‘B’ refers to the same thing in
every possible world, then “A = B” must be true in every possible world or
not true at all. A related insight of Kripke was that identities, or necessities
in general, need not be a priori – they can also be a posteriori and
empirically discoverable. Identities like “Water = H2O” or “Heat = average

35
I thank Max Seeger for comments and discussions that helped clarify the ideas
presented in this chapter.
110

molecular kinetic energy” are a posteriori identities: they are true in all
possible worlds, but we get to know them a posteriori.
However, as Kripke acknowledged, this is prima facie implausible,
since in these a posteriori identities there seems to be an element of
contingency that is in contrast to their supposed necessity. It does not seem
that heat is identical to average molecular kinetic energy (from now on
MKE) in every possible world. We can conceive of worlds where heat is
something else than MKE. If there are indeed such possible worlds, then
the identity seems to be contingent (not true in every possible world). But
there are no contingent identities: if an identity is true, it is true in all
possible worlds. Therefore, the conclusion would be that heat is not
identical to molecular kinetic energy at all.
However, according to Kripke, in cases like this we can explain
away the appearance of contingency. When we think we are conceiving of
heat not being identical to MKE, what we are conceiving is that someone is
having a sensation of heat, but that this is caused by something else than
molecular kinetic energy. For example, we can conceive of a planet where
there are creatures whose sensations of heat are not caused by MKE but by
something else. But this does not amount to conceiving that heat itself is
not MKE. Heat itself cannot be anything else than MKE.
The gist of Kripke’s argument against the identity theory of mind is
that this strategy does not work when it comes to identities involving
phenomenal properties. Let us consider the standard example “Pain = C-
fiber firing.”36 Again, there is felt contingency: it seems that there could be
pain without C-fiber firing. For example, it seems possible that there are
creatures that experience pain but have a neurological system different
from ours and no C-fibers. However, in this case we cannot explain the
contingency away by saying that we are in fact thinking of something else
than pain itself. We cannot make the distinction between the sensation of
pain and the way it appears to us, because the way it appears to us just is
the sensation of pain. Thus, pain is not necessarily identical to C-fiber
firing, and because there are no contingent identities, pain is not identical
36
This statement is in fact empirically false, but we can ignore this for sake of
simplicity and take the statement as a placeholder for some true phenomenal-physical
identity statement involving rigid designators.
 111

to C-fiber firing at all, and the identity theory of phenomenal properties

cannot be true.
Kripke’s argument has of course been disputed, and much ink has
been spilled about it in the recent decades. One problem is that it makes
rather strong assumptions about the nature and philosophical significance
of our imaginative or conceptual capabilities (Hill 1997). However,
discussing these problems in detail here would lead too far off track – see,
for example, Kallestrup (2008) for a recent overview of the argument and
its problems.
In two very influential papers that introduced the notion of an
“explanatory gap,” Joseph Levine (1983; 1993) has critically analyzed
Kripke’s reasoning and examined the epistemic status of the different kinds
of identity statements. The two papers have significant differences, but I
will ignore the differences here and mainly focus on the later paper (Levine
1993), which is more explicit. Levine argues that underlying the difference
between identities (1) “Water = H2O” and (2) “Pain = C-fiber firing” is the
fact that (1) is fully explanatory, with nothing crucial left out, while (2)
leaves something crucial unexplained, namely why pain should feel the
way it does. According to Levine, in statements like (2), there is an
“explanatory gap,” which is responsible for their vulnerability to Kripke’s
argument.
Of course, identities as such are not explanatory. The identity “Water
= H2O” does not alone explain why water behaves the way it does, no more
than “Pain = C-fiber firing” explains why pain feels the way it does. What
Levine has in mind is the following: the identity “Water = H2O” is
explanatory in the sense that microphysical properties of H2O and the
relevant chemical and physical theories “make it intelligible” that water
behaves the way it does. Furthermore, once we understand how H2O
molecules perform the causal role we associate with the concept of water,
there is nothing left to explain.37 The situation with “Pain = C-fiber firing”

37
This comes close to Broad’s (1925) views of reductive explanation (Chapter 2).
According to Broad (in contemporary terminology), a property of a system is
reductively explained by deducing it from the properties of the components of the
system, and in the case of phenomenal properties, this is not possible. See Stephan
112

is different, since the identity does not make it intelligible why pain should
feel the way it does. No matter how much we know about
neurophysiology, the argument goes, the qualitative character of pain is not
explained, and there remains an explanatory gap.
According to Levine, reductive explanations like that of water to
H2O proceed roughly in the following way: First, we give a our
pretheoretic concept of water a conceptual analysis in terms of its causal
role, and then we find out that it is H2O that in fact plays this causal role.
This yields a two-stage picture of reductive explanation:

Stage 1 involves the (relatively? quasi?) a priori process of working the concept
of the property to be reduced “into shape” for reduction by identifying the
causal role for which we are seeking the underlying mechanisms. Stage 2
involves the empirical work of discovering just what those underlying
mechanisms are. (Levine 1993, 132)

The reason why this does not work in the case of phenomenal properties is
that causal role is not all there is to concepts of phenomenal properties:

Reduction is explanatory when by reducing an object or property we reveal the

mechanisms by which the causal role constitutive of that object or property is
realized. Moreover, this seems to be the only way that a reduction could be
explanatory. Thus, to the extent that there is an element in our concept of
qualitative character that is not captured by features of its causal role, to that
extent it will escape the explanatory net of a physicalistic reduction. (Levine
1993, 134)

According to Levine, this explanatory gap constitutes a deep inadequacy in

physicalist theories of mind.
I have shown in the previous chapter that this functional model of
reduction that is also invoked in the explanatory gap argument is not an
accurate picture of reductive explanation, and should be replaced with a

(2004) for a more detailed analysis and comparison of different approaches to the
reductive explanation of phenomenal properties.
 113

model of mechanistic explanation (note also that Levine himself refers to

“underlying mechanisms” in the characterization of two-stage reductive
explanation above). But what happens to the argument if we switch from
talk of functional reduction to mechanistic explanation? Interestingly, the
argument does not change in any substantial way. Although the
functionalization of the properties to be reduced is much more complicated
than is assumed in the functional model, and not even remotely an a priori
matter, it is still true that some kind of causal analysis is required for a
mechanistic explanation of a property. Mechanistic explanations are in a
broad sense causal explanations (constitutive causal explanations). A
property or a phenomenon that cannot be given a causal definition cannot
be mechanistically explained.
However, in spite of this, I do not agree that the impossibility of
causal analysis is enough for arguing that there is a fundamental
explanatory gap.38 What the impossibility of causal analysis shows is that
the property or phenomenon in question cannot be mechanistically
explained, but mechanistic explanation is just one kind of explanation
among many. At the background of the explanatory gap reasoning is the
positivist idea inherited from the deductive-nomological model that there is
only one model of scientific explanation and that all explanations should
conform to this model, but as I have argued in Part I (and will continue in
Part III), we should be pluralists also regarding different kinds of
explanations.
There are many kinds of concepts or properties that cannot be
(exhaustively) causally analyzed, and that for this reason cannot be
mechanistically explained. However, they do not present any explanatory
gap, since they can be explained in other ways. For instance, the property
of being a state of an adding machine is not individuated causally (Polger

38
Seeger (unpublished manuscript) argues that Levine (1993) cannot be claiming that
only properties that can be given exhaustive causal analyses can be reductively
explained, since if this was the case, much of the discussion in Levine’s paper would
be superfluous. On this interpretation, what Levine claims is that what is required for
example in the water case is that the microphysical properties of H2O
“epistemologically necessitate” all the superficial properties or macroproperties (be
they causal or not) of water. I briefly consider this approach to reductive explanation
below.
114

2007a). It is individuated in computational or algorithmic terms, and its

nature is not captured by causal analysis. If Ruth Millikan (1984) is right,
the property of being a heart and other biological properties are not
individuated causally either, but etiologically. We can also think of the
property of being a democratic society, or the property of being a U.S.
dollar. None of these properties can be given an exhaustive causal
definition, neither in terms of physical causal powers nor in the
interventionist sense of causation.
Of course, the difference between phenomenal properties and
computational, etiological, social, etc., properties, is that we actually can
fully explain the non-phenomenal properties – the explanations just are not
causal or mechanistic. The point I want to make here is that arguing for the
explanatory gap based merely on the impossibility of causal analysis,
which is a common strategy, is unconvincing.
The general argument for the functional irreducibility of phenomenal
properties can be taken to have the following structure:

(1) Functional reduction requires that the property to be reduced can

be given a functional analysis in terms of its causal role.
(2) Phenomenal properties cannot be analyzed in terms of their
causal roles.
Conclusion: Phenomenal properties cannot be functionally reduced.

This argument is sound and I accept its premises, even if we switch from
functional reduction to mechanistic explanation. However, it does not yet
entail an explanatory gap. That requires the following further argument:

(1) If a (nonfundamental) property cannot be functionally reduced,

even in principle, it presents an explanatory gap.
(2) Phenomenal properties cannot be functionally reduced, even in
principle.
Conclusion: Phenomenal properties present an explanatory gap.
 115

However, as I have argued above, premise (1) in this argument is false. In

order to rule out cases like adding machines, etiological properties,
economic properties, etc., the premise has to be strengthened. The easiest
way to do this would be the following:

(1) If a (nonfundamental) property cannot be functionally reduced or

explained in any other way, it presents an explanatory gap.
(2) Phenomenal properties cannot be functionally reduced or
explained in any other way.
Conclusion: Phenomenal properties present an explanatory gap.

The problem is that this would make the argument question-begging: the
revised premise (1) basically states that a property cannot be explained if it
cannot be explained, premise (2) states that phenomenal properties cannot
be explained, and the conclusion is that phenomenal properties indeed
cannot be explained. Thus, the problem is that of finding the right way of
reformulating the premises. They should not be so strong that the argument
becomes question-begging, but they should also account for other forms of
explanation besides functional reduction.
One way of doing this, and perhaps this is what at least some of the
supporters of the explanatory gap argument implicitly have in mind, could
be the following:

(1) If a (nonfundamental) property cannot be functionalized, it

presents an explanatory gap.
(2) Phenomenal properties cannot be functionalized.
Conclusion: Phenomenal properties present an explanatory gap.

In the above, “functionalized” is to be understood in a very broad sense,

not just as “causally defined.” If it could be shown that etiological,
116

computational, economic, etc., properties can be functionally defined in

some broad sense of “functional”, the argument could be made to work.
However, it is far from obvious that there is such a broad notion of
“functional” that applies to all the desired cases. Polger (2004) has argued
at length that the prospects of finding such a notion are dim.
If the reasoning in this chapter is correct, the lack of exhaustive
causal definitions of phenomenal properties does not yet entail an
explanatory gap. However, there are certainly other ways of arguing for
the explanatory gap or for the irreducibility of phenomenal properties. I
have not shown that phenomenal consciousness is reducible,
unproblematic, explainable, or anything of the like.39
One further way of arguing for the explanatory gap is based on the
idea of (a priori) entailment as a requirement for reductive explanation.
This argument is closely connected to the causal role argument, and is not
always distinguished from it, but I believe the two should be kept distinct.
Also this argument has its roots in Levine’s (1993) paper. Levine writes:

[A] reduction should explain what is reduced, and the way we tell whether this
has been accomplished is to see whether the phenomenon to be reduced is
epistemologically necessitated by the reducing phenomenon, i.e. whether we
can see why, given the facts cited in the reduction, things must be the way they
seem on the surface. (Levine 1993, 129, emphasis mine)

The idea is that, in the case of “Water = H2O,” the microphysical properties
of H2O necessitate or entail the superficial or macroscopic properties of
water. It is inconceivable that water should not behave the way it does,
given the microphysical properties of H2O. For instance, it is inconceivable

39
One appealing way of pumping intuitions regarding the explanatory gap is the
following. Let us assume we have built a robot with humanlike mental capacities. We
understand completely the mechanics of the robot’s “brain.” Still, it seems to be a
perfectly legitimate philosophical question to ask whether the robot has subjective
phenomenal experiences like we do. In this sense, there does seem to be a “gap”
between phenomenal consciousness and the physical domain, although intuitions may
differ. However, I do not want to defend any position regarding the explanatory gap –
the point of this chapter is merely to show certain problems in the arguments for the
gap.
 117

that water should not boil at 100 °C at sea level, given the microphysical
properties of H2O. However, the neurophysiological properties arguably do
not necessitate or entail the phenomenal quality of pain. It remains
conceivable that there is no subjective experience at all, given the
neurophysiological properties. This argument does not require that the
property to be reduced be given an exhaustive causal analysis – all that is
required is that microphysical properties necessitate or entail the
macroscopic or “superficial” properties of the phenomenon to be reduced.
Of course, if we further assume that macroscopic properties in general are
exhaustively defined by their causal role (adopting a causal theory of
properties), this argument for the explanatory gap is indeed very close to
the other argument, but as should be clear from the above discussion, I do
not think that all macroscopic properties are exhaustively causal properties.
Chalmers and Jackson (2001) have elaborated in detail on this way
of arguing for the explanatory gap. They argue that reductive explanation
of a phenomenon requires that it be a priori entailed by the facts of physics,
and that if there is no such a priori entailment in the case of phenomenal
properties, this constitutes an explanatory gap. There is an a priori
entailment from P to Q if and only if “PQ” is a priori, that is, when it is
possible to know that “PQ” with justification independent of experience.
In the case of “Water = H2O,” Chalmers and Jackson claim, there is such a
priori entailment, since the microphysical facts about H2O imply the non-
basic facts about water (such that it is transparent, boils at 100°C, etc.), and
this implication is knowable a priori. In this sense, an ideal cognizer who
knows all the relevant microphysical facts about H2O could deduce all the
non-basic facts about water without further a posteriori knowledge.
According to Chalmers and Jackson, any account of reductive explanation
that is weaker than this would not yield the desired metaphysical necessity
of the truth of “Water = H2O” (if the connection between water and H2O is
weaker, one has to deal with the Kripke argument). In cases where there is
no reductive explanation in the sense of a priori entailment, there is an
explanatory gap.
Needlessly to say, this approach to reductive explanation is very
different from the approach defended in this thesis. As I have shown in
Part I, actual scientific explanation (including reductive explanation) has
118

little to do with logical entailment. In the background of Chalmers’ and

Jackson’s reasoning is the old deductive-nomological model of
explanation, where logical entailment (or deduction) does play a key role.
However, it is nowadays a platitude in philosophy of science that “PQ”
is neither necessary nor sufficient for P to explain Q. Therefore, Chalmers
and Jackson apparently defend a notion of reductive explanation that,
albeit perhaps metaphysically robust, has almost nothing to do with actual
scientific explanation. This casts doubt on the relevance their conclusions
have for scientific projects of trying to understand the world (or the mind,
or phenomenal consciousness). If the main points in this thesis are correct,
the approach of Chalmers and Jackson is a prime example of philosophy
mind gone awry. Unfortunately, due to constraints of space and time, I
cannot present a more detailed deconstruction here. See Polger (2008a) for
a good overview of the debate and a convincing criticism, and Ladyman &
Ross (2007) for a general attack on the “neo-scholastic” metaphysics that
also Chalmers and Jackson represent.
 119

8. New Type Physicalism

The type identity theory of the mind (also known as type materialism, type
physicalism, or reductive physicalism) is the view that mental properties
are identical to physical properties (see Chapter 5, section 5.2). After its
brief heyday in the late 1950s and early 1960s, this view was for a long
time quite unpopular among philosophers of mind, mainly due the
argument from multiple realizability (see section 5.3). However, since the
end of the 1990s, the multiple realizability argument has come increasingly
under doubt, and partly for this reason, type physicalism has re-emerged as
a serious alternative. Furthermore, several philosophers (Block & Stalnaker
1999; Hill & McLaughlin 1999; McLaughlin 2007; 2010; Papineau 1998)
have advanced new positive grounds for accepting type physicalism,
arguing that it provides a solution for the problem of phenomenal
consciousness.
In this chapter, I will criticize the main argument of this new wave of
type physicalism. I will focus on McLaughlin’s (2007; 2010) version of the
argument, since it is laudably clear and explicitly formulated. In a nutshell,
the argument states that we are justified in believing that type physicalism
is true because it provides the best explanation for the correlations between
phenomenal and physical properties. Recently, Kim (2005) has attacked
this argument, and Bates (2009) and McLaughlin (2010) have responded to
the critique. What I will show here is that none of these authors have taken
sufficiently into account the actual role that identities and correlations play
in scientific explanations. After briefly presenting the explanatory
argument and Kim’s objections to it, I will argue that (contra Kim)
identities do play a crucial role in scientific explanation, but (contra
McLaughlin) they are not put forward as explanations for correlations of
the kind that are involved in the case of phenomenal properties. For this
reason, the explanatory argument fails to provide empirical grounds for
accepting type physicalism.
It is commonly agreed that the hardest aspect of the mind-body
problem is the problem of phenomenal consciousness, which was briefly
discussed in the previous chapter. Phenomenal states are notoriously
resistant to explanations in physical or neural terms, as several passionately
debated arguments purport to show (the explanatory gap, Nagel’s bat,
120

Mary’s black-and-white room, and so on). However, it is plausible that

some day scientists will have discovered exactly how phenomenal states
are correlated with neural states. McLaughlin formulates this as the
following correlation thesis: “For any type of state of phenomenal
consciousness C there is a type of physical state P such that it is true and
counterfactual supporting that a being is in C if and only if the being is in
P” (McLaughlin 2010, 237). This is the starting point of the explanatory
argument for type physicalism.40
Everyone agrees that the truth of the correlation thesis would not yet
settle the question of the place of phenomenal consciousness in nature,
since it is compatible with various ontological positions, including property
dualism, neutral monism, and parallelism. However, according to
McLaughlin, the problem with these positions is that they don’t explain the
correlation thesis. For example, (emergent) property dualism simply asserts
that psychoneural correlations are a fundamental unexplainable feature of
the world.
The core of the explanatory argument is the claim that type
physicalism does explain the correlation thesis. Type physicalism is the
view that phenomenal states are type identical with certain neuro-scientific
states, or more precisely (Type Physicalism): “For every type of state of
phenomenal consciousness C, there is a type of physical state P such that C
= P” (McLaughlin 2010, 266). This seems to provide a straightforward
explanation for the correlation thesis: for each state of phenomenal
consciousness, the reason why a being is in C if and only if the being is in
P is that C = P.
McLaughlin argues that since the alternative ontological positions do
not explain the correlation thesis at all or are otherwise problematic, Type
Physicalism is the best explanation for the correlation thesis, “best on
holistic grounds of overall coherence and simplicity with respect to total
theory” (McLaughlin 2007, 436). Furthermore, the fact that the identity
thesis is the best explanation for the correlation thesis provides the
40
Some authors have raised doubts about the correlation thesis (e.g., Noë & Thompson
2004). Discussing this in detail would lead too far from the main issue here, so I will
simply go along with the type physicalists and assume that the correlation thesis will
turn out true.
 121

justification for holding the identity thesis to be true. The idea behind this
is that if we know that A is true and that B is the best explanation for A,
then we are justified in believing B.
The explanatory argument can also be presented more schematically
as follows (adapted from Bates 2009, 315):

P1. Correlation Thesis: For any type of state of phenomenal

consciousness C there is a type of physical state P such that it is true
and counterfactual-supporting that a being is in C if and only if the
being is in P.
P2. One possible explanation of the Correlation Thesis is Type
Physicalism: For every type of state of phenomenal consciousness C,
there is a type of physical state P such that C = P.
P3. No alternative explanation of the Correlation Thesis is as good as
Type Physicalism.
Conclusion: Type Physicalism is true.

In his latest book, Jaegwon Kim (2005, Ch. 5) has presented (at least) four
distinct objections against the explanatory argument. The first three of
these concern the strategy of inference to the best explanation, both
generally and as it is applied to the problem of phenomenal properties. I
will not discuss them further here, since Bates (2009) and McLaughlin
(2010) have presented quite detailed and convincing replies to these
objections. It might well be that inference to the best explanation is
problematic either generally or in this context, but Kim has not succeeded
in showing this, and it is not my intention to attempt it here.
Instead, my focus is on Kim’s fourth objection, which he apparently
also considers as his main weapon against the explanatory argument. This
objection states that psychophysical identities do not explain
psychophysical correlations, and hence the argument fails.
In fact, this objection consists of many parts that are at least partly
independent from each other. One of Kim’s claims is that the whole point
122

of identities is that they allow us to transcend and get rid of correlations,

not to explain them. He refers to Smart’s (1959) suggestion that to say that
states of consciousness are correlated with brain states is in fact to say that
they are something “over and above” brain states. If we have the identity C
= P, it is not the case that it explains the correlation between C and P – the
identity shows that there is no correlation to be explained.
According to McLaughlin (2010), the problem with this objection is
that Kim has a different notion of correlation than type physicalists.
Following Smart, Kim believes that things that are identical are not
correlated. If two things are correlated, they must be distinct. A thing is not
correlated with itself. However, McLaughlin’s “correlation thesis” only
says: “For any type of state of phenomenal consciousness C there is a type
of physical state P such that it is true and counterfactual supporting that a
being is in C if and only if the being is in P”. It does not even involve the
term correlation, except in the title. There might be some sense of
“correlation” that excludes identity, but this is irrelevant for the argument.
The relevant sense of correlation for the present context is that if A and B
are correlated, A is present when and only when B is present, and this does
not exclude identity.
In another attempt, Kim argues that scientific explanations of
correlations are nothing like McLaughlin and others take them to be:

In science there seem to be two principal ways of explaining correlations: first,

correlations are sometimes explained by invoking a single lower-level process
or structure underlying the correlated phenomena; second, the explanation may
proceed by showing the correlated phenomena to be collateral effects of a
common cause. … In any case, it is quite obvious that scientists will not in
general attempt to explain correlations by identifying the correlated properties.
(Kim 2005, 134)

The type physicalist answer to this argument is straightforward: there

is a third way of explaining correlations in science, and that is by pointing
out that the correlates are identical. According to McLaughlin, “A = B” can
explain why A is present when and only when B is present. To support this
claim, he provides an example: when Maxwell realized that the speed of
 123

electromagnetic waves in a vacuum is the speed of light, he made the “bold

conjecture” that “light waves = electromagnetic waves.” The identity “light
waves = electromagnetic waves” explains why light waves are present
when and only when electromagnetic waves are present. I will return to
this example below.
Finally, Kim claims that, in fact, identities do not explain anything at
all: “Identities seem best taken as mere rewrite rules in inferential contexts;
they generate no explanatory connections between the explanandum and
the phenomena invoked in the explanans; they seem not to have
explanatory efficacy of their own” (Kim 2005, 132). That is, identities
don’t play any role in generating explanations, they just allow us to rewrite
facts.
Type physicalists also have an answer to this – Bates (2009) and
McLaughlin (2010) have pointed out that Kim’s reasoning relies on
imposing certain questionable conditions on explanation, most importantly
(1) that the explanans must be derivable from the explanandum, and (2)
that the explanation must move from one fact to another fact. Furthermore,
Bates (2009) shows that even if we accept Kim’s idiosyncratic
requirements for explanation, identities can explain correlations.
I agree with the type physicalists that Kim has not convincingly
shown that identities do not play a substantial role in scientific
explanations. However, as I will next show, I do think that Kim’s basic
intuition is right: in science, identities are not put forward to explain
correlations of the kind that the correlation thesis involves.
Let us take a closer look at the main scientific example McLaughlin
presents to support his argument: the case of light waves and
electromagnetic waves. McLaughlin writes:

When Maxwell’s calculations showed that electromagnetic waves have the

same speed in a vacuum as the known speed of light, he famously made “the
bold conjecture” that light waves = electromagnetic waves … The hypothesis
that light waves are electro-magnetic waves was invoked to explain why (1)
electromagnetic waves and light waves occur in the same spatial regions at the
same time, why (2) electromagnetic waves have the same speed in a vacuum as
124

light waves, and why (3) the refractive indices in materials are exactly the same
for light waves and electro-magnetic waves. (McLaughlin 2010, 282)

I grant that this example shows that (contra Kim) identities do play a role
in scientific explanation. However, it does not show that identities are a
way of explaining correlations, understood in McLaughlin’s sense (as the
kind of nomological copresence expressed in the correlation thesis). The
crucial part in Maxwell’s argument was showing that electromagnetic
waves and light waves propagate at the same velocity (speed of light).
After considering the velocity of propagation of “magnetic disturbances”,
Maxwell concluded that “[t]this velocity is so nearly that of light, that it
seems that we have strong reason to conclude that light itself (including
radiant heat, and other radiations if any) is an electromagnetic disturbance
in the form of waves propagated through the electromagnetic field
according to electromagnetic laws” (Maxwell 1865, 466).
Importantly, the fact that light waves are present when and only
when electromagnetic waves are present did not play a big role in making
(or justifying) the “bold hypothesis”. The hypothesis was not presented to
explain this correlation (and such a correlation alone would never have
been enough to justify the bold hypothesis). It was presented to explain the
fact that light waves and electromagnetic waves share a crucial property
(the speed of propagation). The hypothesis was then later confirmed by
empirical results, which showed that light waves and electromagnetic
waves also have other properties in common, such as their refractive
indices (as McLaughlin correctly points out).
Let us consider another example from a field somewhat closer to
philosophy of mind, namely vision research. About 40 years ago, what
came to be known as “luminance units” were discovered somewhat
accidentally while recording extracellularly from the cat retina (Barlow
and Levick 1969). These units were extremely rare (less than 1% of the
retinal ganglion cell population) and responded to light stimuli in an
unusual way: the response was sluggish, relatively straightforwardly
related to the light intensity, and increased monotonically with increasing
light intensity (normally the responses of retinal ganglion cells are much
more complex).
 125

In a different line of research, a population of morphologically

distinct retinal ganglion cells containing the photopigment melanopsin has
been recently discovered in the mammalian retina (see Do & Yau 2010 for
a review). Interestingly, these melanopsin-containing cells are intrinsically
photosensitive – they respond to light even when the synaptic transmission
from the normal photoreceptors (rods and cones) is blocked. They
resemble the “luminance units” in several important ways: their response
to light stimuli is sluggish, the relation between response and light intensity
is monotonic, and they are also very rare (comprising about 1-3 % of the
retinal ganglion cell population).
Based on these common properties, it seems extremely likely that the
melanopsin-containing cells are the luminance units. However, this identity
remains hypothetical, since it has not (yet) been shown that all the
properties of luminance units are also exhibited by the melanopsin-
containing cells (or the other way around) (Do & Yau 2010). Thus, this is a
case of a hypothetical identity claim that is pending empirical
confirmation.
I believe that underlying both of these examples is a more general
pattern of the role identities play in scientific explanation: they are put
forward as hypotheses to explain why two things that were believed to be
distinct (or were discovered by different methods) both have the same or
similar properties. The fact that light waves and electromagnetic waves
propagate at the same speed and have the same refractive indices is
explained by the hypothesis that light waves are electromagnetic waves.
The fact that luminance units and melanopsin-containing ganglion cells
react to light in the same way is explained by the hypothesis that
luminance units are melanopsin-containing ganglion cells. In contrast,
these identities are not put forward to explain phenomena that are
correlated in the sense of being always copresent, as is the case with
phenomenal properties and physical properties.
This fits well with a recent philosophical account of the role of
identities in science. The proponents of the “Heuristic Identity Theory”
(McCauley and Bechtel 2001; the theory is largely based on ideas of
Wimsatt 1976a) argue that identities are not conclusions of scientific work
but hypothetical premises. The principal motivation for formulating
126

identity claims is their potential to advance empirical research. The role of

identities is not to explain correlations, but to connect levels or sciences
and to generate new hypotheses and new avenues for research. This is one
crucial aspect in which identities differ from correlations: they suggest
explanatory connections that demand further empirical exploration, while
correlations do not. In this sense, claims about correlations and claims
about identities are “different conceptual animals that thrive in different
theoretical habitats” (McCauley and Bechtel 2001, 754).
A crucial feature that the above examples show and that the heuristic
identity theory also emphasizes is that hypothetical identity claims are to
be tested just like any other hypotheses in science, by seeing how they
stand up to empirical evidence. One way of doing this is by testing the
further hypotheses that the identity claim suggests. Since we know that if
two things are identical, they have to have exactly the same properties
(indiscernibility of identicals), a hypothetical identity claim immediately
provides ways of testing itself: if A is identical with B and A has property
P, then B has to have property P also, and so on. This is also how
Maxwell’s hypothesis was tested and confirmed: if light waves are
electromagnetic waves, then both must have the same refractive indices,
and this indeed turned out to be the case.
In contrast, it is difficult to see how the identity claim of the new
type physicalists could be tested in this way. The claim is that phenomenal
properties are identical to neural properties, but this claim does not imply
any further hypotheses that could be empirically tested. Type physicalists
would in fact agree that there can be no further empirical evidence (in
addition to the psychophysical correlations) or tests that could distinguish
between Type Physicalism and the other ontological positions – their claim
is that Type Physicalism is superior based “holistic grounds of overall
coherence and simplicity with respect to total theory” (McLaughlin 2007,
436). In this respect, the identity claim of type physicalists fundamentally
differs from the hypothetical identity claims in science.
These considerations point to a further problem in the explanatory
argument for type physicalism. In all of the examples discussed, the
hypothetical identity is an identity of things. This includes also the
examples of successful identifications that McLaughlin (2010) and Bates
 127

(2009) present (Bill Sikes = the burglar, water = H2O, Tully = Cicero, light
waves = electromagnetic waves, etc.). However, what is at issue in the case
of psychoneural correlations and type physicalism is the identity of
properties. The claim is that the correlation of properties is explained by
the fact that the properties are identical. In order to support this claim,
Type Physicalists would be expected to provide examples from science
where the correlation of two properties is explained by these properties
being identical, but so far they have failed to do so.
In fact, I believe such examples will be hard to find, since it is not a
common explanatory strategy in science to explain the copresence of
properties by hypothesizing that they are identical. In contrast, it is often
the case that correlated properties are not identical. Consider for example
the property of having a heart and the property of having a circulatory
system. An organism has a heart if and only if it also has a circulatory
system. However, explaining this correlation with the hypothetical identity
“the property of having a heart = the property of having a circulatory
system” would be deeply mistaken. If type physicalists want to stick to the
claim that their hypothesis is empirical and scientific, the burden of proof
is on them to present evidence from science where the identity of
properties explains the correlation of properties.
Type physicalists have one obvious answer to the concerns raised in
this chapter. They might grant all of the above, but still claim that in some
sense identities do explain correlations. For example, it does seem
intuitively quite plausible that the fact that water is present when and only
when H2O is present is explained by the hypothesis that water is H2O.
Perhaps this could also be translated to talk of properties: the fact that the
property of being water is present when and only when the property of
being H2O is present is explained by the hypothesis that “the property of
being water = the property of being H2O”. However, claims like this are
not convincing as long as they only rely on intuitions or philosophical
models of explanation and are not backed up by actual scientific cases. The
history of the discovery of water is a long story, but it is safe to say that the
fact that water is present when and only when H2O is present played
absolutely no role in the process that led to the hypothesis that water is
H2O. Type physicalists might insist that the fact that water is present when
128

and only when H2O is present is in some sense explained by the fact that
water is H2O, but even if this is the case, such explanations are only
shallow, philosophical, or ad hoc explanations, not real scientific
explanations.
To repeat, the role of identities in science is not to explain
correlations. What leads to the identity claim (initially a hypothetical one)
in cases like “light waves = electromagnetic waves” or “water = H2O” or
“luminance units = melanopsin-containing ganglion cells” is that things on
either side of the identity share important properties. The hypothetical
identities immediately suggest ways of empirically testing themselves. In
the case of phenomenal state C and the correlated neuroscientific state, no
relevant properties are shared. The claim is that the correlation between
physical state P and a phenomenal state C is explained by the identity “P =
C”. This hypothetical identity claim does not indicate any further
hypotheses that could be empirically tested. Therefore, the identity claim
of the type physicalists is of a different kind than identity claims in science,
and the former is not supported by the success and significance of the
latter.
Type Physicalism might still be preferable to other metaphysical
positions for purely philosophical reasons, but the claim that it is on a par
with other scientific hypotheses is not supported by the history of science.
Type physicalists have tried to present the best explanation argument as an
empirically supported argument, while it is in fact pure metaphysics.
Furthermore, the above considerations also provide general reasons
to question the best explanation strategy of arguing for Type Physicalism. I
have argued that the identity thesis explains the correlation thesis, if at all,
only in a very shallow or ad hoc way. How important is it then that the
identity thesis is currently the best explanation for the correlation thesis? If
we have theories that do not explain phenomenon P at all and one theory X
that explains it in a shallow and ad hoc manner, does this provide
justification for holding X true? I believe the answer must be no.
Therefore, the second premise of the explanatory argument (“P2: One
possible explanation for the correlation thesis is type physicalism”) is
either false or too weak to support the argument.
 129

Some type physicalists take the explanatory argument very seriously:

”[t]he explanatory argument presents a formidable argument for reductive
physicalism” (Bates 2009, 317). ”The explanatory argument gives us good
empirical grounds to accept physicalism” (ibid. 324). In this chapter, I
have argued for the exact opposite: the explanatory argument does not give
us empirical grounds to accept physicalism and it is not a good argument
for type physicalism.
130
 131

Conclusions: Rethinking Reduction in Philosophy of Mind

In this part, I have critically analyzed central topics in philosophy of mind
that involve reduction. I will briefly re-iterate some of the main points
here. First of all, in contrast to the received view in philosophy of mind,
multiple realizability is an unsettled question, and one that is relatively
irrelevant for the issue of reduction (section 5.3). Regarding functionalism
(section 5.5), I have argued that in its most plausible form (role-
psychofunctionalism) it is compatible with the picture defended in Part I
(mechanistic explanation and interventionist causation).
In the longest section (Chapter 6), I have argued that the functional
model fails as a scientifically and philosophically plausible account of
reduction, and that mechanistic explanation provides a better framework
for analyzing the questions of psychoneural reduction.
I have also showed (Chapter 7) that in the arguments for the
fundamental inexplicability of phenomenal properties it is not enough to
appeal to the fact that phenomenal properties cannot be causally defined,
since there are many kinds of properties that cannot be causally defined but
nevertheless can be explained. In the last chapter I have argued, based on
the role of identities in scientific explanation, that the central argument for
the new type physicalism fails.
The general conclusion from Parts I and II is that the strongest form
of reduction to be found in the case of psychology and neuroscience is
downward-looking mechanistic explanation. Functional reduction,
metascientific reduction, New Wave reduction, and Nagel-reduction all
face serious and fundamental problems. I have not directly argued against
reductive physicalism in the sense of type identity theory, but I have
argued against the main arguments supporting it, and will continue this in
Part III. Now I will turn to the final part of this thesis, and present an
account of explanatory and ontological pluralism, and a new framework
for understanding reduction and interlevel relations in philosophy of mind.
132
 133

PART III:
A New Framework for
Philosophy of Mind
134
 135

Introduction
In Parts I and II, I have argued against strong forms of reductionism in
philosophy of mind (most importantly intertheoretic reductionism,
metascientific reductionism, and functional reductionism) and defended
explanatory pluralism and the interventionist approach to causation. But
what does explanatory pluralism exactly consist in? What are the
ontological implications of explanatory pluralism and the interventionist
account of causation? Is pluralism threatened by the dreaded causal
exclusion argument? In this part, I will give answers to these questions. I
will argue that if we adopt explanatory pluralism and the interventionist
approach to causation, our understanding of physicalism has to change, and
this leads to what I call pluralistic physicalism. I will then show that this
pluralistic physicalism is not endangered by the causal exclusion argument,
and that it is compatible with a kind of (weak) reductionism.
The structure of this part is as follows. In Chapter 9, I will give a
more precise characterization of explanatory pluralism, and discuss four
theses that form its core. In Chapter 10, I will argue that, under some very
plausible assumptions, explanatory pluralism leads to pluralistic
physicalism. I will also consider how this kind of physicalism relates to
more traditional forms of physicalism. In Chapter 11, I will discuss the
causal exclusion argument and argue that it does not pose a fundamental
problem for the kind of pluralism I am defending, if causation is
understood in the interventionist sense. In Chapter 12, I will argue that
pluralistic physicalism is in fact compatible with many reductionist ideas,
and discuss some largely neglected ways of understanding the reductionist
aspects of scientific explanation. Finally, I conclude this part and the whole
book with a summary of the results and an outlook on future work.
136
 137

9. Explanatory Pluralism for Philosophy of Mind

Recently several philosophers of neuroscience, biology, and psychology
have defended explanatory pluralism as an approach to the relations
between sciences and different analytical levels (e.g., Bechtel 2008;
Brigandt 2010; Craver 2007; Horst 2007; Looren de Jong 2002; McCauley
& Bechtel 2001; Mitchell 2003; Richardson 2009; Wimsatt 1976a; 2007;
Wright 2007). This is also the position I have defended in the first part of
this book. I take the core of explanatory pluralism to consist of the
following four theses:

(1) For full understanding of human behavior (or the mind),

explanations of different kinds are necessary
(2) For full understanding of human behavior (or the mind),
explanations at different levels are necessary
(3) Successful explanations remain explanatory even when
corresponding lower-level explanations are complete
(4) Interlevel connections and explanatory integration across
disciplines are essential in explanatory enterprises

The term “explanatory pluralism” has been mainly used by Robert

McCauley and William Bechtel (McCauley 1996, McCauley & Bechtel
2001), and in a somewhat narrower sense than I use it here. Richardson
(2009) talks of “methodological pluralism,” meaning roughly what I have
outlined above. Mitchell (2002; 2003) describes her view as “integrative
pluralism”, Horst (2007) his as “cognitive pluralism.” However, I prefer to
stick to the expression “explanatory pluralism” for its simplicity and
clarity.
Thesis (1) is an acknowledgement of the fact there is no single
pattern or structure to which all scientific explanations conform.
Historically speaking, the most influential model of scientific explanation
has been the deductive-nomological model (Hempel & Oppenheim 1948).
For a long time it was hoped that this model, or at least something very
138

similar, would capture the general pattern of scientific explanations.

Unfortunately, these hopes were dashed, as it turned out that most
scientific explanations do not fit the model. In fact, it is fairly clear that
scientific explanations are too heterogeneous to fit any single model. Also
when explaining the human mind or brain, we shouldn’t expect the
explanations to conform to a single pattern: we need mechanistic, causal,
computational, evolutionary, etc., explanations. This is what Mitchell
(2002; 2003) calls the horizontal dimension of pluralism.
The second thesis reflects the fact that focusing on just one level of
analysis is in most cases insufficient for full understanding of the
phenomenon of interest. Levels are here best understood as the “levels of
mechanisms” in the system or phenomenon under consideration (see Part I,
Chapter 4). For example, in order to understand the memory consolidation
mechanism, we need to consider several compositional levels, and none of
these levels is fundamental or sufficient for full understanding of the
phenomenon. For instance, the molecular level is not sufficient, because
we also need to understand the functional role of the mechanism and where
it is situated in the overall system. The higher levels are not sufficient,
because often the details of the composition are necessary for making the
right predictions or explanations. Mitchell (2003) calls this the vertical
dimension of pluralism.
The third thesis is related to the second one, but is stronger, since it
states that higher-level explanations are necessary not only now, but also in
the foreseeable future. The importance of higher-level explanations is not
due to some temporary incompleteness of lower-level theories. For
example, even when we know the full story of memory consolidation all
the way down to the molecular level, we will still need higher-level
regularities characterizing the functioning of memory, since going down to
the molecular level to seek explanations is in most cases both pointless and
intractable due to the enormous complexity of the system (see, e.g.,
Dennett 1991 or Wimsatt 2007 for more).
An unrelenting reductionist might still claim that there is “in
principle” derivability from lower to higher levels, meaning that given
enough computational power and time, we could use the molecular level
generalizations to explain anything the higher-level generalizations
 139

explain. However, how could we evaluate such “in principle” claims, given
that we do not have the time and the computational power, and we do not
know what the “completed” sciences will look like? It is true that in
mathematical or computational sciences there are proofs of “in principle”
derivability even when the actual derivations or computations are utterly
intractable. However, it is hard to see how such proofs could be
forthcoming at the interfaces of sciences as “messy” as neuroscience and
psychology (Wimsatt 2007, Ch. 4). Instead, we have to rely on thought
experiments and intuition pumps, which do not give definitive or reliable
answers.
The point of the fourth thesis is to emphasize the importance of
explanatory integration and interlevel connections: the explanations of
different fields and levels are not independent or isolated from each other.
This is a crucial point that sets explanatory pluralism apart from
“promiscuous” pluralism and claims of disunity of science (e.g., Dupré
1993), or from the “dappled world” pluralism41 of Nancy Cartwright
(1999), or from even more radical views, such as the “methodological
anarchism” of Paul Feyerabend (1975).
The pioneers of exploring the integration of disciplines were Lindley
Darden and Nancy Maull (Darden and Maull 1977), who argued that the
development of interfield theories that connect two existing fields is often
necessary to solve problems and answer questions that could not be
answered with the tools of the fields in isolation. Another pioneer has been
Wimsatt (1976a; 2007), who has repeatedly emphasized the importance of
coevolution of theories of different levels. “A lower-level model is
advanced to explain an upper-level phenomenon which it doesn’t fit
exactly. This leads to a closer look at the phenomenon, and perhaps results
in some change in the way in or detail with which it is described. This will
also lead to changes in the lower level model and may suggest new
phenomena to look for” (Wimsatt 1976a, 231). These successive

41
A representative example is Cartwright’s following characterization of her
metaphysical position: ”Metaphysical nomological pluralism is the doctrine that nature
is governed in different domains by different systems of laws not necessarily related to
each other in any systematic or uniform way; by a patchwork of laws” (Cartwright
1999, 31).
140

modifications sew the two levels together ”more closely than Siamese
twins” (1976a, 232). Building on the insights of these predecessors,
McCauley and Bechtel (2001) have proposed a “Heuristic Identity
Theory,” where hypothesized identities contribute to the integration of
different fields or disciplines, suggesting new avenues for empirical
research.
Mitchell (2002; 2003) has defended integrative pluralism in biology.
She points out that integration of disciplines and models is essential in
science, but the types of integration are varied and diverse – no single
model will suffice. In the same vein, Brigandt (2010) has argued that the
way disciplines are integrated depends on the specific scientific problem
(epistemic goal) at hand, and that transient, case-specific, integrations are
sufficient for genuine explanatory integration. Craver (2007, Ch. 7) has
proposed that a “mosaic” unity of neuroscience can be achieved as
different fields contribute constraints on multilevel mechanistic
explanations.
What all these approaches have in common is that they replace the
classic goal of the unity of science with some weaker, patchier and messier
picture of integration that is more faithful to actual science. However, all of
these accounts remain somewhat sketchy and don’t come even close to the
clarity and formal precision of the accounts of unity of science in terms of
intertheoretic reduction. The advantages of intertheoretic models are that
they can be handled with logico-semantical, mathematical and set-theoretic
tools, and that they have been scrutinized and developed through many
decades. Unfortunately, the other side of the coin is that, at least when it
comes to neuroscience and psychology, they are fundamentally inadequate,
since they require formal representations of theories, which takes the
analysis to an abstract level far distant from actual scientific practice.
Is explanatory pluralism compatible with reductionism? Of course,
this depends on what is meant by reductionism. If we understand reductive
explanation as downward-looking mechanistic explanation, and
reductionism as the view that all mental phenomena can be reductively
explained, then explanatory pluralism and reductionism are indeed
compatible. The claim that all mental phenomena can be reductively
explained in the mechanistic sense does not contradict any of the four
 141

theses of explanatory pluralism (more on this in Chapter 12). In fact, the

wide acceptance of explanatory pluralism is closely related to the recent
emergence of mechanistic explanation as the paradigm for the philosophy
of the life sciences.
If, on the other hand, reductionism is understood as New Wave
Reductionism, “ruthless” reductionism, or functional reductionism, then
reductionism is not compatible with explanatory pluralism. Reductionists
of these kinds would deny one or all of the first three theses of explanatory
pluralism. Kim’s functional reductionism conflicts with at least thesis (3).
Ruthless reductionism includes the explicit denial of thesis (2) and (3). The
New Wave Reductionist would claim that theses (1), (2) and (3) may very
well turn out false as science proceeds. However, as I have argued earlier
in this book, all of these forms of reductionism face fundamental problems,
so they do not pose a threat to explanatory pluralism.
142
 143

10. From Explanatory Pluralism to Pluralistic Physicalism

What is the relation between explanatory pluralism and physicalism? What
are the ontological implications of the interventionist account of causation?
These questions have been largely neglected in the literature on
explanatory pluralism and interventionism, mainly due to the tendency of
philosophers working on these topics to eschew traditional metaphysical
issues. However, instead of eschewing the metaphysics, one can also try to
find out a scientifically relevant metaphysical position that fits explanatory
pluralism and interventionism. This is my main goal in this section.
Among other reasons, this is important for connecting the new philosophy
of science with the more classic metaphysical debates in philosophy of
mind, which in the end is one of the main aims of this book.42
Traditionally, causal considerations have played a key role in the
arguments for physicalism. For example, Kim (2005) argues along the
following lines: Causal considerations rule out substance dualism, since it
is inconceivable how the nonmaterial mental substance could causally
interact with the physical substance that has only physical properties. Kim
then continues by arguing that causal considerations also rule out property
dualism: the famous causal exclusion argument purportedly shows that
nonphysical properties cannot have causal powers of their own, which
means that property dualism leads to the highly implausible conclusion that
nonphysical properties are epiphenomenal. Trenton Merricks (2001) goes
even further, and argues that causal considerations lead us to the
conclusion that macroscopic objects such as tables and chairs do not exist,
since if they did, this would lead to an unacceptable form of causal
overdetermination.
However, if we adopt the interventionist account, this reasoning
breaks down. In the interventionist framework, causation is a notion that is
important in the special sciences but not in fundamental physics. Causes at
different levels can happily coexist, and higher-level causes are not
excluded by lower-level causes (more on this in the next chapter). This is

42
In order to keep the discussion reasonably compact and clear, I assume here that
scientific realism is in some sense true, and do not discuss alternatives like
instrumentalism or constructivism.
144

in stark contrast with the view that causation is a physical matter or that all
causes reduce to physical causes. It seems that causal considerations now
lead toward some kind of pluralism instead of traditional physicalism.
Furthermore, interventionism leads toward pluralism if we merely
make the plausible assumption that there is a close connection between our
best explanations and what is real, or if we give the interventionist account
of causation some realistic interpretation. As we have seen, this is also
what Kim would like to have (1998, 76): “when we speak of ’causal
explanation’, we should insist … that what is invoked as a cause really be a
cause of whatever it is that is being explained. Realism about explanation
should at least cover causal explanation.” Also Woodward (2008, 228)
argues that providing a causal explanation of an outcome requires making
true claims about its causes.
I propose that the best way of making sense of this kind of pluralism
is basing it on the notion of robustness. The idea of robustness is drawn
from the practice of scientific modeling, and has been most extensively
discussed by William Wimsatt (1981; 2007). He roughly defines it as
follows (2007, 196): “Things are robust if they are accessible (detectable,
measureable, derivable, defineable, producible, or the like) in a variety of
independent ways.” For instance, the moon is a very robust thing, since it
can be measured and detected and accessed in numerous ways that are
independent from each other. Properties like temperature or mass are
robust, since they are also measurable, detectable, etc., in a variety of
independent ways. It is important that the different ways of access are
independent from each other, since then the likelihood that they all are
mistaken is a product of each one’s independent likelihood to go wrong,
and this product will be a very small number if there are many independent
ways.
According to Wimsatt (1981; 2007), robustness is by no means a
new idea, and has in fact been looming at the background throughout the
history of philosophy, particularly in the works of Aristotle, Galileo,
Peirce, and Whewell. In the last century, the idea was discussed by Levins
(1966) in connection to modeling in population biology, and Levins was
apparently the first to use the term “robust” in approximately the present
sense (see also Hacking (1983), who does not use the term but presents
 145

similar ideas in passing). However, in spite of its importance, robustness

has never received broader attention of the philosophical community –
only very recently there has been renewed interest in the idea (Calcott
2010, Weisberg 2006).
Wimsatt extends robustness to cover also theories, laws,
explanations, and so on, but this makes the notion unnecessarily
complicated. For the present purposes, we can define a version of
robustness that concerns properties and phenomena: a property or
phenomenon is robust if it is detectable, measurable or producible in a
variety of independent ways. Based on this, we can formulate the core idea
of robustness-realism as follows:
We are justified in believing that property (or phenomenon) P is real
if and only if property (or phenomenon) P is robust, that is, it is detectable,
measurable or producible in a variety of independent ways.
This formulation may be in need of further refinement, but the basic
idea is clear and plausible. It is also clear that if we take robustness as a
guideline for building our ontology, plenty of higher-level or special
science properties turn out real. For example, the properties of short-term
memory, such as its approximate capacity, can be measured and studied
with various experimental setups that are independent of each other.
Change blindness is a fairly recently discovered robust phenomenon of
visual perception that is detectable and producible in a variety of
independent ways. The same goes for psychological and special science
properties in general, insofar as they are good scientific properties – as
Wimsatt (2007, Ch. 4) points out, scientists generally use robustness
analyses to determine whether a phenomenon is real or just an artifact.
Using robustness as a guideline for what to consider real leads to a kind of
ontological pluralism and a “tropical rainforest ontology” (Wimsatt
2007).43

43
Interestingly, it is not clear whether phenomenal properties as usually understood
count as robust. Since they are fundamentally subjective and can be experienced only
from a first-person perspective, there are no independent ways of accessing, detecting,
or measuring them. This would imply that they are not real in the sense of being
robust. Perhaps this is not so surprising. We are not talking about some capacity or
function of the cognitive system, like spatial memory or stereovision. We are talking
146

Robustness as a criterion for what is real obviously differs from what

Kim calls “Alexander’s Dictum”: To be real is to have causal powers.44
Kim explicitly commits himself to this view, and takes it to be immensely
plausible (e.g., Kim 1993, 202). Nonetheless, it should be clear from what
has been presented so far in this book that Alexander’s dictum is highly
problematic. In Chapter 7, I have argued that there are many properties that
cannot be causally defined, but are nevertheless real, such as the property
of being a state of an adding machine. Furthermore, if causal powers are
supposed to be something even remotely resembling our causal intuitions
and causation in the special sciences, then it turns out that there are no
causal powers in fundamental physics, since we find there nothing
resembling our intuitive idea of causation. Thus, Alexander’s dictum
would imply that the most fundamental physical entities are in fact unreal.
Robustness also differs from another criterion for reality, one that
according to Polger (2004, 2007b) has been broadly (yet implicitly)
assumed in philosophy of mind. This is what Polger calls the autonomy
thesis: “a property x is real if and only if x is essentially involved in (the
explanation of) a regularity G” (Polger 2007b, 67). The “essentially
involved in” basically means that the (explanation of) regularity G is
irreducible. Thus, the autonomy thesis could be formulated more
straightforwardly: “A property x is real if and only if x is involved in an
irreducible explanatory generalization.” Although I partly disagree with
Polger’s conclusions (he is defending explanatory pluralism and the
identity theory), I agree with him that the autonomy thesis has to be
rejected. To be essentially involved in a regularity is not a sufficient
criterion for a property to be real: a property has to be involved in several
regularities that are independent of each other to be considered real.

about the experiences the system undergoes, from the perspective of the system. If
phenomenal properties are real, they are real in a very different sense than other
objects of scientific inquiry, and we certainly should not expect same models of
reductive explanation to apply to them, or draw any conclusions from the impossibility
of such reductive explanation. However, this would be a topic for a treatise of its own.
44
This dictum is named after Samuel Alexander (1920), who wrote that
epiphenomenalism “supposes something to exist in nature which has nothing to do, no
purpose to serve, a species of noblesse which depends on the work of its inferiors, but
is kept for show and might as well, as undoubtedly would in time, be abolished”
(1920, vol. 2., 8).
 147

Recently Ladyman and Ross (2007) have also argued for ontological
pluralism, but in somewhat different terms. First of all, they show that
prominent philosophers of mind have misunderstood or neglected
contemporary physics. The idea (implicit also in Kim’s work) that
everything comes down to “microbangings” amongst elementary things at
the fundamental level makes absolutely no sense from the point of view of
current physics. There is no causation in fundamental physics, no
“fundamental level,” and not even elementary things in the sense of self-
subsistent individuals. Therefore the thesis that everything that is real is
ultimately composed of fundamental level microphysical things (e.g., Pettit
1993) is simply false. For similar reasons, Ladyman and Ross also reject
general (mereological) reductionism, type identity theory and traditional
forms of physicalism. Instead, they defend a form of ontological structural
realism conjoined with the idea of real patterns (Ladyman & Ross 2007,
Ch. 2 and 4).
The talk of real patterns goes back to Dennett’s (1991) well-known
paper. The idea that Dennett, Ladyman, and Ross defend is that to be real
is to be a real pattern. However, what real patterns exactly are is a rather
complicated matter. Dennett (1991, 34) gives only a rough and weak
definition: there is a real pattern in some data if there is a description of the
data that is more efficient than a verbatim bit map description. Ladyman
and Ross provide a very technical definition based on information theory,
and going through it here would lead too far astray. In any case, the basic
idea is similar to that of Dennett’s: a pattern in a structure of events S is a
real pattern if describing S as a bit map would be information-theoretically
less efficient than describing it in terms of the pattern, and there are aspects
of S that cannot be tracked without the pattern (Ladyman & Ross 2007,
233). Real patterns are not real just relative to human capabilities and
limitations: any computational system that is efficient will make use them,
and real patterns are real even if there is no one to observe them.
Since there are real patterns all over the place, also in the domains of
all the special sciences, pattern-realism leads to Rainforest Realism: “Ours
is thus a realism of lush and leafy spaces rather than deserts, with science
regularly revealing new thickets of canopy” (Ladyman & Ross 2007, 234).
It is interesting that Ladyman and Ross end up using the same rainforest
148

metaphor as Wimsatt, approaching from a very different direction. The

authors apparently reach their conclusions completely independently from
each other (at least Ladyman and Ross never refer to Wimsatt, or vice
versa).
The main difference between these two versions of rainforest
pluralism is that Wimsattian robustness-realism implies that there are real
things and properties, while pattern-realism states that there are strictly
speaking no things in the world, just patterns (hence the title of Ladyman
and Ross’ book, Every Thing Must Go). However, according to Ladyman
and Ross, what are taken to be “things” in everyday life and the special
sciences can generally be seen as real patterns, and thus are real. I also take
it to be fairly plausible that robust phenomena and properties can be
interpreted as real patterns, although the converse need not hold.
Therefore, the ontological conclusions of the two positions might end up
being rather similar.
The main point here is that there are (several) scientifically and
philosophically plausible ways of making sense of ontological pluralism.
However, I prefer robustness-realism to pattern-realism, since robustness is
a far more intuitive and less technical criterion, and more scientifically
relevant. As Wimsatt (2007, Ch. 4) shows, scientists constantly use
robustness analyses to determine whether a phenomenon is real or just an
artifact. It is not clear whether the criterion of pattern-reality has any
practical use – at least the authors do not give many concrete examples.
Therefore, I will adopt robustness-realism as the ontological framework for
the pluralism I am defending.
One should not understand ontological pluralism based on robustness
as some kind of “spooky” pluralism that asserts that there are
fundamentally different substances in the world. It merely expresses the
fact that there are many different kinds of properties in the world, and that
requiring that everything real is reducible to something physical or has
physical causal powers does not make much sense.
Another important caveat is that I am not advocating a form of
constructivism. The pluralism I am defending is rather a form of scientific
realism. Our ideas about what is robust may change as science proceeds,
 149

but this does not mean that reality itself changes. The fact that property P
is robust in our current analyses gives us justification for believing that P is
real, but it does not in any sense “make” P real.
Let us now turn to the question whether robustness pluralism is an
alternative to physicalism or a kind of physicalism. In addition to causal
arguments that were discussed above, another motivation for physicalism
has come from considerations based on the history of science. All
hypotheses concerning non-physical forces that affect physical processes in
a way that conflict with the laws of physics have consistently failed.
Relatedly, as science has progressed, more and more phenomena have been
successfully explained in broadly speaking physical terms – also
phenomena that were previously thought to resist physical explanations.
Perhaps the biggest triumph in this respect was the explanation of the
fundamental processes of life in terms of DNA molecules. However, these
inductive arguments do not directly support physicalism. They support a
weaker thesis, which Ladyman and Ross (2007, 43) have dubbed the
Primacy of Physics Constraint (PPC): “Special science hypotheses that
conflict with fundamental physics, or such consensus as there is in
fundamental physics, should be rejected for that reason alone. Fundamental
physical hypotheses are not symmetrically hostage to the conclusions of
the special sciences.” That is, physics sets constraints for the theories of
special sciences.45
A robustness pluralist can happily accept the Primacy of Physics
Constraint. The claim that there are irreducible higher-level properties in
no way conflicts with the claim that fundamental physics constrains the
theories or hypotheses of special sciences. This takes us to the point that
instead of seeing robustness pluralism as an alternative to physicalism, it is
perhaps more appropriate to see it as a kind of physicalism. Consider the
following definition of physicalism (often called “supervenience
physicalism”, see Chapter 5, section 5.6): Physicalism is true at a possible
45
PPC should not be taken as a metaphysical a priori principle. If we would one day
discover a special science generalization that is empirically confirmed to the highest
degree and conflicts with the fundamental laws of physics, then this would perhaps
lead to revision of the fundamental physical theory. However, so far we have
encountered no such cases. PPC is a result of inductive reasoning based on the history
of science, and could be falsified by empirical evidence.
150

world w if and only if any world which is a (minimal) physical duplicate of

w is a duplicate of w simpliciter (Jackson 1998, 12). Nothing what has
been said above is in conflict with this. A robustness pluralist could also
accept that the fundamental physical level in some sense determines all the
higher-level properties. A robustness pluralist could accept token
physicalism. If criteria of this kind are sufficient for physicalism, then the
position I have defended could be called pluralistic physicalism. It
provides a scientifically credible and philosophically interesting middle
ground between reductive physicalism and more radical forms of
pluralism.46
What is then wrong with traditional forms of physicalism and why is
robustness pluralism preferable to them? I take the main problem with
reductive physicalism (type physicalism) to be the familiar one: multiple
realizability. As I have pointed out in Chapter 5 (section 5.3), recent
analyses have cast serious doubt on claims of multiple realizability, both
conceptually and empirically. I agree with these critics in that philosophers
of mind have overestimated the significance of multiple realizability.
However, I also believe that proponents of multiple realizability are right
in one sense: there are no one-to-one mappings from all higher-level
properties to physical properties. The type physicalist solution to the reality
of higher-level properties would require the following: for every single
higher-level property that we want to retain in our ontology we will find a
physical property that is identical to that higher-level property. I find this
extremely implausible. Furthermore, if we look at scientific practice,
special science properties are not considered real only insofar as they are
identical to some physical properties – they are considered real insofar as
they are robust.

46
Andreas Hüttemann (2004) also seems to be a pluralistic physicalist (though he calls
his position ‘pragmatic pluralism’), but in a somewhat different sense: he accepts
physicalism, but denies microphysicalism, and argues that physical systems at all
scales are ontological equals. The position I am defending also resembles that of
Carrier and Mittelstrass (1991, particularly Ch. 6), who argue that psychological states
are real in a sense that is compatible with a kind of physicalism, and that one criterion
for the reality of psychological states is “construct validity” – a psychometric concept
that is related to robustness.
 151

The relation between pluralistic physicalism and traditional

nonreductive physicalism is more complicated. If nonreductive
physicalism is understood as consisting of a moderate kind of physicalism
(such as supervenience physicalism) and the view that special science
properties are distinct from physical properties, then pluralistic physicalism
is a form of nonreductive physicalism. However, traditional nonreductive
physicalism carries more baggage than this. Most importantly, it also
includes the following thesis about the ontological status of higher-level
properties: higher-level properties are not identical to physical properties,
but are physically realized. The problem with this “realization
physicalism” is that its success hinges on the notion of realization, but it
has turned out to be extremely difficult to spell out a notion of realization
that would yield a plausible form of nonreductive physicalism and make
scientific sense (see also Chapter 6, section 6.3.2). Without such an
account, realization physicalism collapses into either type physicalism or
property dualism. In contrast, pluralistic physicalism abandons the idea of
realization. It states that higher-level properties are real insofar as they are
robust; they need not be “realized” (i.e., made real) by physical properties.
In general, realization physicalists have been very ambitious in
assuming that (1) there is a single notion that relates all mental properties
to physical properties, and that (2) this notion is sufficient for giving a
satisfying answer to the ontological status of mental properties. The
relations between mental and physical properties are complex and have to
be understood in terms of many different notions. Mental properties are
very heterogeneous: even if we restrict our focus to the properties studied
by cognitive psychology, they include properties varying from change
blindness to the capacity of short-term memory to cognitive dissonance.
There is no reason to expect that we could use a single notion, such as
realization, to account for the way in which these properties are related to
lower-level properties. These relations can spelled out in terms of
constitution, mechanisms, determination, satisfaction of function, and so
on, but there is no reason to expect a single notion, such as “realization”, to
apply in every case, and it is questionable whether “realization” even
captures anything important that could not be accounted for with the other
notions (see also Polger & Shapiro 2008).
152

There is one further general point worth emphasizing – a point that is

independent of the question of physicalism and of the form of pluralism
one wants to adopt. From a causal and explanatory point of view, there is
no fundamental difference between psychological properties on the one
hand, and biological or neuroscientific properties on the other hand. Causal
and explanatory relevance in the interventionist framework is a matter of
there being stable, invariant generalizations between variables, and these
kinds of generalizations can just as well hold between psychological
variables as between biological variables. Hence, if we take biological or
neuroscientific concepts that are involved in good explanations and
theories to refer to “real” states or properties, there is no reason to claim
that psychological concepts that are involved in good explanations and
theories do not. If we are realists regarding biological or neuroscientific
states, we should also be realists regarding psychological states.
 153

11. Pluralistic Physicalism and Causal Exclusion Worries47

Perhaps the most formidable challenge to nonreductive ontological
positions, including pluralistic physicalism, is the causal exclusion
argument. I have already presented the argument in Chapter 6, but to
repeat, it has the following basic structure (Kim 2002, 278):

The Problem of Mental Causation: Causal efficacy of mental properties is

inconsistent with the joint acceptance of the following four claims: (1) physical
causal closure, (2) exclusion, (3) mind-body supervenience, and (4)
mental/physical property dualism (i.e., irreducibility of mental properties).

The argument is targeting mental properties, and I will mainly discuss

mental causes in this section, but it should be noted that the argument
works just as well for any nonphysical properties. One reason why mental
properties are seen as particularly problematic is that it is generally
assumed that biological, neural, chemical, etc., properties either count as
broadly speaking “physical” properties, or are ontologically reducible to
physical properties. Therefore, premise (4) does not hold for these
properties, and they are not threatened by the argument. Yet, the pluralism
I have defended above can be taken to imply that these kinds of properties
are in a sense distinct from physical properties, and therefore face the
exclusion argument. For this reason, it is particularly important to show
that there are no serious worries of causal exclusion. This is my goal in this
chapter.
To briefly reiterate, the guiding insight of the interventionist account
of causation is that causal relationships are relationships that are potentially
exploitable for purposes of manipulation and control. To put it very
roughly, the core of the model is that a necessary and sufficient condition
for X to cause Y or to figure in a causal explanation of Y is that the value of
Y would change under some intervention on X. An intervention can be
thought of as an (ideal or hypothetical) experimental manipulation carried
out on variable X (the independent variable) for the purpose of ascertaining
47
This chapter found its shape and content largely based on extensive discussions with
Dan Brooks and Vera Hoffmann-Kolss, for which I am very grateful.
154

whether changes in X are causally related to changes in some other variable

Y (the dependent variable). Causal relationships are relationships that are
invariant under some range of interventions (see Part I, Chapter 3, or
Woodward (2003) for more specific definitions).
Prima facie, it seems that mental causation is unproblematic in the
interventionist framework. There are invariant psychological
generalizations such that we can make interventions to mental states in
order to change other mental states or physical behavior. For example, as
Woodward (2008) points out, when you persuade someone, you
manipulate her beliefs by providing information or material things, in order
to change her other beliefs. Also many psychological and social science
experiments involve intervening on the beliefs of the subjects, usually
through verbal instruction, in order to change some other beliefs and
observable behavior.
In a closer philosophical analysis, it indeed seems that the
interventionist account vindicates mental causation. Recently several
authors (e.g., Menzies 2008; Raatikainen 2010; Woodward 2008) have
argued that if the interventionist account is correct, mental states can be
causes of physical behavior, and they are not excluded by their physical
realizers. On the other hand, Michael Baumgartner (2010) and Vera
Hoffmann-Kolss (unpublished manuscript) have argued that there is an
interventionist version of the exclusion argument, and thus adopting the
interventionist account does not make the problem of exclusion go away.
Instead of going through the details of these arguments, I argue that
there is a deeper underlying problem that kicks in already before the
arguments of either side can take off. The problem is that typical causal
representations of mental causation fail to satisfy the conditions required of
interventionist causal models.48 One of these conditions is that variables
that are not related as cause or effect or as effects of a common cause have
to be uncorrelated. In other words: conditional on its direct causes, each
variable has to be independent of every other variable except its effects
(this is often called the Causal Markov Condition, see Hausman &
Woodward 1999 for other formulations and an extensive discussion of the

48
This was originally pointed out to me by Dan Brooks, for which I am very grateful.
 155

condition49). Although the exact formulation of this condition has been a

matter of some debate, it is widely agreed that the condition (or at least
something very close to it) is integral to causal modeling. If this condition
is not satisfied, the model is not a well-formed causal model, and drawing
causal inferences from it is not possible.
The typical representations of mental causation in philosophy of
mind fail to satisfy this condition (Figure 5). Kim’s formulation of the
exclusion argument is a good example: in this representation, mental
property M causes another mental property M*, physical property N causes
another physical property N*, M supervenes on N, and M* supervenes on
N*. Due to supervenience, the values of M and N (as well as M* and N*)
are correlated, and M depends on N. Whenever M changes, N also changes,
and when the value of N is fixed, the value of M is also fixed. However, M
does not cause N, N does not cause M, and they are not both effects of a
common cause. Mind-body supervenience implies a non-causal correlation
and dependency between the variable describing the mental property and
the variable describing the physical property. Therefore, from an
interventionist point of view, the representation is incorrect and has to be
modified. 50

49
Another condition that is also extensively covered in the same paper, and that could
perhaps be also used as a basis for the arguments in this section, is modularity: a
system consisting of several causal relationships is modular to the extent that these
various causal relationships can be changed or disrupted while leaving the others
intact. Both the Causal Markov Condition and modularity have been under intense
discussion in recent years – see, for example, Cartwright (2002) or Steel (2006).
50
Recently Shapiro and Sober (2007) have also argued that supervenient causes are
problematic in the interventionist framework, but in a slightly different manner. Let us
consider again a situation where we want examine whether M, which supervenes on N,
is a cause of physical behavior B. We have to make an intervention on M such that
other causes of B, including N, remain unchanged. The problem is that this is
impossible, since the value of N determines the value of M (due to supervenience). It is
not acceptable or nomologically possible to wiggle M while holding N fixed. Hence,
this must be a wrong way of conceptualizing the situation.
156

Figure 5: A typical representation of mental causation in philosophy of

mind. The arrows represent causation, the dotted lines represent
supervenience.

The obvious reductive solution to this problem would be to get rid of

the mental variables, either by eliminating them or identifying them with
physical variables. Then we would have only physical variables in the
representation, and no non-causal relationships. However, the problem
with this approach becomes obvious when we consider the fact that we can
apply just the same reasoning to biological, chemical, neural, and
macrophysical properties. They all supervene on lower-level physical
properties. Therefore, we can simply draw the same picture again,
replacing the mental variables by, say, neural variables, and the neural
variables by chemical variables. Then it seems that since we got rid of the
mental variables in the first case, we also have to get rid of the neural
variables in the second case. Causation seems to be draining away towards
some fundamental physical level, which is particularly strange if we
consider the fact that there seems to be nothing resembling our ideas of
causation at the fundamental physical level (see Chapter 6, section 6.3.3).
This is a version of the generalization argument that has often been
raised against Kim’s exclusion argument (e.g., Block 2003; van Gulick
1992). The generalization argument states that if Kim’s reasoning about
mental properties is correct, then we can apply it to all higher-level
 157

properties or macroproperties, which are then excluded. However, this is

an absurd conclusion, so there has to be something wrong with Kim’s
argument. Kim has provided several answers to the generalization
argument, but it is widely agreed that none of them is satisfactory (see,
e.g., Walter 2008).
The reductive approach of replacing or reductively identifying the
higher-level variables also runs counter to scientific practice: when
scientists have to choose between causal representations of a system, it is
not the case that they always choose the maximally precise or lowest-level
representation. The interests of the scientist determine the explanandum,
and once this is fixed, various empirical and theoretical considerations
determine the right level at which the causal explanation is sought
(Woodward 2010). One does not get rid of a good causal model just
because the properties represented in it supervene on some lower-level
properties.
This leads to a more scientifically plausible way of dealing with
supervenience in causal representations. This would allow higher-level
causal representations, but not allow including the supervenient base
variables in the same representation. For example, we would not include
neural variables in the same representation as the supervenient mental
variables. We would have a plurality of causal representations, but no
representations that include both supervenient variables and their base
variables. As Hausman and Woodward (1999, 531) put it in a different
context: “One needs the right variables or the right level of analysis –
variables that are sufficiently informative and that are not conceptually
connected.”
This approach is simple, coherent, and scientifically credible.
However, defending it convincingly also requires showing what exactly
goes wrong in the exclusion argument. The argument seems to be valid, so
at least one of its premises has to turn out false.
I will focus on the most likely candidate, the exclusion principle.
This principle states that no effect has more than one sufficient cause,
except in cases of genuine overdetermination. A straightforward
interventionist rendering of this principle would be something along these
158

lines: If variable M is a difference-making cause for B, there is no other

difference-making cause for B, unless this is a genuine case of
overdetermination. It is easy to see that this principle does not hold: there
can be many difference-making causes to a single variable. However, this
formulation is too general and not very fair – it should at least include the
requirement that the competing causes are acting at the same instance in
time (Menzies 2008). Taking this into account, we could formulate the
principle as follows: If this particular instantiation of M (the variable M
taking, say, value 1 instead of 0) is a difference-making cause for this
particular instantiation of B (the variable B taking value 1 instead of 0),
then there is no other difference-making cause for this particular
instantiation of B (unless this is a case of overdetermination).
In my view, this principle is also problematic, due to the fact that it is
possible to find several difference-making causes at different levels for a
given effect. Consider for example (an instantiation of) mental property M
that causes (an instantiation of) property B. Since mental properties
supervene on neural properties, it is very plausible that in some cases we
can find some neural property N that is also a difference-making cause for
B. The value of variable B changes under interventions on M, but also
under interventions on N. This works also the other way around: if we start
with a neural property N that is a difference-making cause for B, it is likely
that in many cases we can find a supervenient mental property M that is
also a difference-making cause for B.
Therefore, it seems that we can form several noncompeting
representations of the same situation: in one, the particular instantiation of
M is a difference-making cause for the particular instantiation of B, and in
another, the particular instantiation of N is the difference-making cause.
Which level we focus on depends on the question at hand. It is also
important that variables of different levels are not included in the same
representation, since this would lead to the problem discussed above (i.e.,
violation of the Causal Markov condition). Therefore, before considering
causal relations one has to fix the level of analysis, which is determined by
the (research) question that one seeks to answer.
This can be seen as leading to a denial of the exclusion principle, or
alternatively as acceptance of systematic overdetermination. I have argued
 159

that it is possible to build several noncompeting representations of the

same situation, each having (instantiations of) different properties as the
difference-making cause of one and the same (instantiation of a) property.
If one does not count this as genuine overdetermination, then the exclusion
principle is false. If one does count this as overdetermination, then we have
systematic overdetermination. Both options have been traditionally
considered unacceptable, but if we understand causation as a matter of
difference-making and manipulation and control (and not as physical
“bringing about”), this kind of violation of the exclusion principle or
acceptance of overdetermination is unproblematic (see also Bennett (2003;
2008), who casts doubt on the exclusion principle, independently of the
notion of causation applied). There simply can be several difference-
making causes at different levels for a given effect, and which level we
focus on depends on contextual matters.
The most promising strategy for trying to save the causal exclusion
argument would probably be to admit that all of the above is true for causal
relevance and causal explanation, but to claim that what is relevant for
mental causation is causal efficacy or causal powers, which are taken to be
something metaphysically stronger. However, as I have already argued in
Chapter 6 (section 6.3.3), it is difficult to see for what purpose we would
need such stronger notions of causation, barring armchair metaphysics.
To summarize, if we understand causation in difference-making
terms, it is true that causal claims become very problematic when
conjoined with supervenience claims. However, this does not mean that
higher-level causes are excluded by the lower-level causes they supervene
on. Which variables are retained in the representation depends on the
question at hand. The exclusion argument can be tackled either by denying
the exclusion principle or by accepting systematic overdetermination. The
issue is far from settled, but there are good reasons to believe that the
exclusion argument does not rule out interventionist higher-level causes
(see, e.g., Woodward (unpublished manuscript) and Baumgartner (2010)
for more discussion on this topic).
160
 161

12. Dimensions of Explanatory Power

Above I have argued for explanatory pluralism and a kind of ontological

pluralism (or pluralistic physicalism). This might seem like a defense of
antireductionism, and when reductionism is understood along traditional
lines, it indeed is. However, in this chapter I will make the position more
intricate by arguing that certain central ideas associated with reductionism
are in fact to some degree correct, and can be accommodated within a
generally pluralistic framework.51
As I have already pointed out in several places, downward-looking
mechanistic explanations can be seen as reductive explanations. Let us call
the thesis that all mental phenomena can be given such mechanistic
explanations the mechanistic explanation of the mental thesis (MEM).
Already this thesis alone satisfies many reductionistic intuitions: if it is
true, all psychological phenomena can be mechanistically explained in
terms of underlying neural mechanisms. However, this is completely
compatible with pluralism, since mechanistic explanations do not make the
psychological properties any less real, or psychological explanations any
less true. Furthermore, from a pluralistic point of view, mechanistic
explanations are not always exhaustive: they do not rule out psychological,
etiological, etc., explanations of the same phenomenon.
This point becomes clearer if we relate it to the situation in
philosophy of biology, where it is widely accepted that there can be
different non-competing explanations for the same phenomenon. For
example, we can explain a given biological function in terms of its
evolutionary origin (answering the “Why?” question), in terms of its
ontogenesis, or in terms of the underlying causal mechanism (answering
“How?” questions). The same applies for philosophy of psychology. For
instance, the psychological phenomenon of memory consolidation can be
explained mechanistically, but this does not rule out an explanation in
terms of its psychological function, its evolutionary origin, its ontogenesis,

51
The idea that pluralism and reductionism are compatible is not entirely new. One
author that has explicitly argued for it is Daniel Steel (2004), and Wimsatt (2007) is
also a pluralist who emphasizes the importance of reductive explanation.
162

and so on. These are different kinds of explanations, and thus non-
competing. This idea is reflected in the thesis (1) of my definition of
explanatory pluralism, which states that for fully understanding the human
mind we need explanations of different kinds (see Chapter 9). The MEM
thesis is not in conflict with this, since MEM as such does not rule out non-
mechanistic explanations for mental phenomena.52
Another reductionist thesis that can be accommodated to pluralism is
related to the special nature of fundamental physics. The history of science
shows that there appears to be the following fundamental asymmetry:
higher-level explanations are constrained by physical explanations, but the
converse does not hold. This is captured in the Primacy of Physics
Constraint (PPC) already presented in Chapter 10: “Special science
hypotheses that conflict with fundamental physics, or such consensus as
there is in fundamental physics, should be rejected for that reason alone.
Fundamental physical hypotheses are not symmetrically hostage to the
conclusions of the special sciences” (Ladyman & Ross (2007, 43). That is,
physics sets constraints for the theories of special sciences. As Ladyman
and Ross (2007, 210) suggest, this asymmetry can perhaps be extended to
apply more broadly: for instance, computer science asymmetrically
constrains psychology in the sense that what is computationally impossible
is also impossible for human brains, but not vice versa.53
In Chapter 10, I defended robustness as a criterion for considering
something real. Importantly, robustness is a notion that comes in degrees:
some things are more robust than others. For instance, photons are
obviously more robust than cognitive dissonance (i.e., the condition of
conflict or anxiety that arises from holding two conflicting ideas or beliefs

52
I have not argued that the MEM thesis is true, and I do not know whether it is – the
point is to show that it amounts to a form of reductionism that is compatible with the
pluralism I have defended.
53
Some authors (e.g., Steel 2004) have also argued for corrective asymmetry. The idea
is that resources from the lower levels are often necessary to explain exceptions in
generalizations at the higher levels, but not vice versa. For example, exceptions to
Mendelian laws of inheritance are explained by resources from molecular genetics,
and not by any higher-level or same-level sciences. However, it is not clear whether
this asymmetry holds generally – sometimes exceptions in generalizations are
explained also by resources from the same level or higher levels (see Raerinne 2011).
 163

simultaneously). The latter is a psychological phenomenon that is

accessible, measurable, detectable, producible, etc., in a variety of
independent ways – it is a robust phenomenon. However, the different
(independent) ways of accessing it are limited in number. They are
certainly fewer than the numerous independent ways of accessing,
measuring, detecting, etc., photons. This suggests that there is a threshold
of robustness below which things are not considered real, but above this
threshold there are different degrees of robustness. It is plausible that the
things that physics and other physical sciences study are generally more
robust than things that higher-level sciences such as psychology study, but
exploring this in detail would go beyond the scope of this thesis.
Instead, I will focus in the rest of this chapter on another crucial
issue that has been mostly neglected in the debates on reduction and
explanatory pluralism: explanatory power and its different dimensions.
Particularly in philosophy of mind, it is often implicitly assumed that
explanations are either true or false, and that true explanations do not come
in degrees. However, it is clear that not all established or successful or true
explanations are equally good or powerful. Christopher Hitchcock and
James Woodward put this point very vividly:

Some explanations are deep and powerful: Newton’s explanation of the tides,
Maxwell’s explanation of the propagation of light, Einstein’s explanation of the
advance of the perihelion of Mercury. Other explanations, while deserving the
name, are superficial and shallow: Bob lashed out at Tom because he was angry,
the car accelerated because Mary depressed the gas pedal with her foot, the salt
dissolved because it was placed in water. We take this intuition to be very natural
and widely shared. Yet in the vast philosophical literature on explanation, there
have been precious few attempts to give any systematic account of this notion of
explanatory depth. (Hitchcock and Woodward 2003, 181)

Hitchcock and Woodward (2003) then argue that a systematic account of

explanatory depth can be provided with the notion of invariance (see also
Part I, Chapter 3, for more on invariance):
164

One generalization can provide a deeper explanation than another if it provides

the resources for answering a greater range of what-if-things-had-been-different
questions, or equivalently, if it is invariant under a wider range of interventions.
That is, generalizations provide deeper explanations when they are more
general. (Hitchcock & Woodward 2003, 198)

To take a clear example from physics, the van der Waal’s force law ([P +
a/V2][V - b] = RT) is invariant under a wider range of interventions than the
ideal gas law (pV = nRT). This is because it includes the additional
variables a (the attraction between the particles) and b (the average volume
excluded by a particle). In this sense, the van der Waal’s force law can
answer more what-if-things-had-been-different questions, and provides
deeper explanations.
However, it is plausible that there is not just one dimension of
explanatory depth, but several, as has been recently argued by Petri
Ylikoski and Jaakko Kuorikoski (Ylikoski & Kuorikoski 2010). An
explanation A of a phenomenon P can be better than explanation B in some
ways, but worse in some other ways. The main point of the authors is that
there are several distinct dimensions of explanatory power, and that they
do not go hand-in-hand: in fact some of them are systemically in conflict.
Ylikoski and Kuorikoski define the following five dimensions of
explanatory power: non-sensitivity (with respect to background
conditions), precision, factual accuracy, degree of integration (to a larger
body of knowledge), and cognitive salience (i.e., how easily the
explanation can be grasped). The account is very provisional, and it is not
clear why exactly these five dimensions are selected, but this is not crucial
here: the important point is that explanatory power comes in degrees, and
that it has several dimensions.
This has significant consequences for understanding reductionism.
One widely shared intuition that makes reductionism appealing is that
lower-level generalizations or explanations are more fundamental than
higher-level ones. I believe this intuition is not entirely mistaken: it is
plausible that lower-level explanations are better than higher-level
 165

explanations in some but not all dimensions of explanatory power.54 This

provides a more fine-grained framework for comparing higher- and lower-
level explanations and for understanding reductionistic claims. It also adds
one more key element to explanatory pluralism: explanations of different
kinds and levels are not all equal, but can differ with respect to various
dimensions of explanatory power. The reductionistic idea that lower-level
explanations are more fundamental than higher-level ones can be
accommodated in a pluralistic framework, as long as we understand that
explanatory power is a multidimensional thing and lower-level
explanations are not just more powerful simpliciter.
In which dimensions are lower-level explanations then faring better
than higher-level ones? I will briefly mention here some candidates, but a
closer analysis will have to wait for a later treatise. One candidate is
factual accuracy: higher-level explanations (always) abstract away from
(at least some) details that could be taken into account at lower levels.
Nearly all explanatory models in special sciences, including non-
fundamental physics, are to some degree idealized models (see Cartwright
1983; 1999; Wimsatt 2007, Ch. 6). Furthermore, explanations and
predictions based on lower-level generalizations are generally more precise
than those based on higher-level generalizations, and yield more precise
predictions of future events. A further dimension where lower-level
explanations fare better is scope: the generalizations of physics apply
everywhere in the universe, whereas psychological generalizations apply
only in very small areas of our planet. In the case of psychology and
neuroscience, cellular mechanisms are similar across different organisms,
while psychological explanations might be applicable only to our species.55
We can also come back to the idea of degree of invariance that
Hitchcock and Woodward (2003) propose as the measure of explanatory
depth. At least in some cases, generalizations at lower levels have a higher
degree of invariance than generalizations at higher levels, and thus the

54
Related to this, Wimsatt (2007, Ch. 5 and 12) has discussed at length “reductive
biases” and the dangers of being too eager to apply reductive heuristics, especially in
situations where they are not warranted.
55
McCauley (2009) uses scope as one criterion to define higher vs. lower levels. See
Part I, Chapter 4 for more.
166

former provide deeper explanations. Hitchcock and Woodward discuss

several ways in which one generalization can be more invariant than
another, but the most important one for assessing invariance across levels
is the following: an explanatory generalization G’ is more invariant than
explanatory generalization G, if it makes explicit the dependence of the
explanandum on variables treated as background conditions by G. For
example, in Galileo’s law of free fall, the mass of the planet is treated as a
background condition that is not involved in the generalization, but in
Newton’s law of gravitation the mass of the planet is included as a
variable. Hence, Newton’s law of gravitation is invariant under a broader
range of interventions.56 Newton’s law of gravitation and Galileo’s law of
free fall are competing same-level explanations, but perhaps these
considerations can also be extended to interlevel cases.
As a psychological example, consider the “Garcia effect.” It is a
psychological generalization that states that if someone gets nauseous after
consuming a certain food, this leads to an aversion of (the taste of) that
food. This is a fairly robust phenomenon, and the generalization is
invariant. It can be used to explain, for example, why Kosta refuses to eat
cooked bananas (the explanation is that he got nauseous last time after
eating them). However, these kinds of explanations are shallow, and the
generalization has a rather low degree of invariance. It treats the
neuroscientific details as background information. If we would know the
neurobiological mechanisms of the Garcia effect in detail, we could have
generalizations that include the neuroscientific details as variables. These
generalizations would also hold in some cases where the psychological
generalization breaks down, and thus they would have a higher degree of
invariance. If neurobiological generalizations are in general more invariant
than psychological generalizations, this is clearly one sense (or dimension)
in which the former provide deeper explanations than the latter.
I have argued that lower-explanations fare better than higher-level
explanations with regard to some dimensions of explanatory power. But on
the other hand, higher-level explanations generally fare better in some
other dimensions. Perhaps the most obvious one is cognitive salience: e.g.,
56
This kind of invariance comes close to the idea of explanatory scope mentioned
above.
 167

psychological explanations are easier to grasp and to apply than

explanations in terms of the underlying neural mechanisms. Another
dimension is generality or portability: since higher-level explanations often
abstract away from lower-level details, they can be used in a broad range
of circumstances where there are differences in these details (consider
Putnam’s (1975) peg example). Relatedly, higher-level generalizations also
allow more efficient computations, that is, they have a lower computational
cost.
As the above discussion already suggests, there are often systematic
trade-offs among the different dimensions: an increase of explanatory
power in one dimension often leads to a decrease in some other dimension.
This idea of trade-offs of explanatory power has been often mentioned in
different contexts, but has not yet been explored in sufficient detail.57 For
instance, Dennett (1991, 36) talks of a trade-off that is “ubiquitous in
nature”: “Would we prefer an extremely compact pattern description with a
high noise ratio or a less compact pattern description with a lower noise
ratio?” He also asks (ibid.): “Is it permissible in science to adopt a carving
system so simple that it makes sense to tolerate occasional misdivisions
and consequent mispredictions? It happens all the time. The ubiquitous
practice of using idealized models is exactly a matter of trading off
reliability and accuracy of prediction against computational tractability.”
To take another example, Sober (1999) points out that higher-level
explanations often score well on generality but less well on depth, while
lower-level explanations are often deeper but less general. That is, there is
often a trade-off between the generality and the depth of an explanation for
a phenomenon.
Matthewson and Weisberg (2009) discuss trade-offs in modeling in
more detail. They distinguish between three types of trade-offs: there can
be a trade-off between modeling attributes A and B in the sense that (1) it
is impossible to maximize both A and B, or (2) it is impossible to increase

57
Levins (1966) discusses trade-offs in ecological model building, but relatively
briefly. Recently, Matthewson and Weisberg (2009) have returned to the topic and
analyzed trade-offs in modeling, building on the work of Levins. Also Ylikoski and
Kuorikoski (2010) briefly discuss trade-offs among the different dimensions of
explanatory power.
168

A and B simultaneously, or (3) whenever there is a decrease in A, there is

an increase in B, and vice versa. Matthewson and Weisberg then argue that
there is a trade-off between precision and generality in the sense (1): all
other things being equal, it is impossible to increase both the generality and
precision of the model at the same time.
A more thorough discussion of these trade-offs and their
philosophical importance has to wait for another occasion. The main point
of this chapter is that the different dimensions of explanatory power and
their trade-offs is a topic of crucial significance that has not received
enough philosophical attention, and that a large part of the appeal of
reductionism is due to these differences and trade-offs. Yet, they have none
of the drastic consequences that hardcore reductionists have argued for. In
fact, they fit perfectly into the pluralistic framework I have defended in this
thesis.
 169

Conclusions and Directions for Further Research

I have now come to the end of the third and final part of this thesis. In this
part, I have defended explanatory pluralism and pluralistic physicalism as a
framework for philosophy of mind, and pointed out ways in which
reductionistic ideas and intuitions can be accommodated to this framework.
Now it is time to take stock of the results of this thesis in general, to
summarize the consequences for philosophy of mind, and to consider
topics that need to be addressed by further research.
First of all, from an epistemological point of view, it is clear that the
prospects of strongly reductive or eliminative programs do not look very
good. Both the history of science and considerations based on the
philosophy of causation and scientific explanation show that explanations
of different kinds and at different levels will be needed for fully
understanding human behavior and the mind. This is not just due to the
incompleteness of our current scientific knowledge or our cognitive
limitations: it is due to the overwhelming complexity of the world we live
in. The strongest form of explanatory reduction warranted by scientific
practice is downward-looking mechanistic explanation. Its claim to the title
“reduction” can be disputed, but it does satisfy some pretheoretic intuitions
about reduction, and is close to what many scientists mean when they talk
about reduction.
From an ontological point of view, matters are more complicated. I
am not very sympathetic to the type identity theory, but I have not
explicitly argued against it. Instead, I have argued that the main argument
for the identity theory, the causal exclusion argument, loses its force when
causation is understood in interventionist terms. For the same reason, a
large part of the motivation for being a traditional physicalist dissolves.
Therefore, I have suggested replacing traditional physicalism with what I
call pluralistic physicalism: properties of different kinds and at different
levels are real, but fundamental physics still has a special status among the
sciences, as it asymmetrically constrains the theories and explanations of
other sciences. Regarding the mind-body problem, this position implies
that psychological properties (as discovered by empirical psychology) are
170

real and not in any strong sense reducible to neural properties. But on the
other hand, psychological properties are not strongly emergent or
completely independent from neural properties: they are integrated into
neural mechanisms, and can be seen as their higher-level properties.
This pluralistic framework leaves room for many approaches that
have been traditionally associated with reductionism. The idea that
explanations of lower levels are more powerful than explanations of higher
levels can be accepted, as long as we understand that explanatory power is
a multidimensional thing, and that lower level generalizations fare better
than higher level generalizations in some (but not all) of these dimensions.
Furthermore, the thesis that all mental phenomena can be mechanistically
explained is more than enough for many reductionists, and it is compatible
with pluralistic physicalism.
There are obviously several unresolved issues that need to be
addressed by future research. First of all, I have critically analyzed several
problems in philosophy of mind in light of the results of this thesis, but
there are many more potential targets that I have left untouched. These
include intentionality and representation, the extended mind, self-
consciousness and self-models, emotions, embodiment, enactivism,
perception, free will, and many more.
One thing that this thesis has hopefully made clear is that there is a
huge gap between formal approaches to reduction (e.g., New Wave
reductionism, functional reduction) and non-formal approaches that are
closer to scientific practice (e.g., mechanistic explanation). What is still
unclear is what exactly are the fields of application of non-formal and
formal approaches, and what are the theoretical and practical limitations of
each. For instance, what exactly are the domains where mechanistic
explanations are the norm, and what kinds of phenomena generally resist
mechanistic explanations?
Another cluster of important open questions concerns the idea of
levels, a topic that has received comparatively little philosophical attention.
Does it make sense to talk of general levels of organization or are “levels
of mechanisms” sufficient? What are the criteria for identifying levels and
 171

assigning things to levels? Are there systematic “trade-offs” of explanatory

power across levels?
In this thesis, I have defended pluralism based on the concept of
“robustness.” However, I have defined this key concept only roughly and
preliminarily, and other existing accounts of robustness do not go much
further. What is needed is a detailed analysis of the notion, its different
manifestations, and its relation to scientific practice. It is also not clear
whether or not the weak ontological pluralism I have defended is in some
sense reconcilable with the type identity theory or in fundamental conflict
with it.
The age-old problem of mental causation is a further issue that is still
far from resolved, and especially the implications of the interventionist
model of causation for philosophy of mind need to be further explored. I
have argued that the causal exclusion argument loses its force in the
interventionist framework, but there are many remaining problems, such as
how to deal with supervenience in causal modeling, and it seems that the
debate has only just started.
In general, this thesis opens at least as many new questions as it
answers old ones. But if philosophy is, as I believe, more about asking the
right questions and less about answering them, this can be seen as a good
result.
172
 173

References
Aizawa, Kenneth, and Carl Gillett (2009). The (multiple) realization of
psychological and other properties in the sciences. Mind & Language
24, 181-208.
Alexander, Samuel (1920). Space, Time, and Deity. New York: Dover
Publications.
Armstrong, David M. (1968). A materialist theory of the mind. London:
Routledge.
Barlow, H. B., and W. R. Levick (1969). Changes in the maintained
discharge with adaptation level in the cat retina. Journal of
Physiology 202, 699-718.
Bates, Jared (2009). A defence of the explanatory argument for
physicalism. The Philosophical Quarterly 59, 315-324.
Batterman, Robert (2001). The Devil in the Details. Asymptotic Reasoning
in Explanation, Reduction, and Emergence. Oxford: Oxford
University Press.
Baumgartner, Michael (2010). Interventionism and Epiphenomenalism.
Canadian Journal of Philosophy 40, 359-383.
Broad, Charles D. (1925). The Mind and its Place in Nature. London:
Routledge & Kegan Paul.
Bechtel, William (2007). Reducing psychology while maintaining its
autonomy via mechanistic explanation. In M. Schouten & H. Looren
de Jong (eds.) The Matter of the Mind: Philosophical Essays on
Psychology, Neuroscience and Reduction. Oxford: Blackwell
Publishing, 172-198.
Bechtel, William (2008). Mental Mechanisms. Philosophical Perspectives
on Cognitive Neuroscience. London: Routledge.
Bechtel, William, and Jennifer Mundale (1999). Multiple realizability
revisited. Philosophy of Science 66, 175–207.
174

Bechtel, William, and Robert C. Richardson (1993). Discovering

Complexity: Decomposition and Localization as Strategies in
Scientific Research. Princeton: Princeton University Press.
Beckermann, Ansgar (1997). Property physicalism, reduction and
realization. In M. Carrier & P. K. Machamer (eds.) Mindscapes:
Philosophy, Science, and the Mind. Pittsburgh: University of
Pittsburgh Press, 303-321.
Bedau, Mark A. (1997). Weak emergence. In J. Tomberlin (ed.)
Philosophical Perspectives: Mind, Causation, and World, Vol. 11.
Malden, MA: Blackwell, 375-399.
Bennett, Karen (2003). Why the Exclusion Problem Seems Intractable, and
How, Just Maybe, to Tract It. Noûs 37, 471-497.
Bennett, Karen (2008). Exclusion again. In J. Hohwy & J. Kallestrup (eds.)
Being Reduced. Oxford: Oxford University Press, 280-305.
Bickle, John (1998). Psychoneural Reduction: The New Wave.
Cambridge, MA: MIT Press.
Bickle, John (2003). Philosophy and Neuroscience: A Ruthlessly Reductive
Account. Dordrecht: Kluwer Academic Publishers.
Bickle, John (2006a). Reducing mind to molecular pathways: explicating
the reductionism implicit in current cellular and molecular
neuroscience. Synthese 151, 411-434.
Bickle, John (2006b). Ruthless reductionism in recent neuroscience. IEEE
Transactions on Systems, Man, and Cybernetics – Part C:
Applications and Reviews 36, 134-140.
Bickle, John (2008). Real reduction in real neuroscience: Metascience, not
philosophy of science (and certainly not metaphysics!). In J. Hohwy
& J. Kallestrup (eds.) Being Reduced. Oxford: Oxford University
Press, 34-51.
Block, Ned (1978). Troubles with functionalism. Minnesota Studies in the
Philosophy of Science 9, 261-325. Reprinted in Block (1980), 268-
305.
 175

Block, Ned (ed.) (1980) Readings in the Philosophy of Psychology, Vol. 1.

Cambridge, MA: Harvard University Press.
Block, Ned (2003). Do causal powers drain away? Philosophy and
Phenomenological Research 67, 133-150.
Block, Ned, and Jerry A. Fodor (1972). What psychological states are not.
The Philosophical Review 81, 159-181.
Block, Ned, and Robert Stalnaker (1999). Conceptual analysis, dualism,
and the explanatory gap. The Philosophical Review 108, 1-46.
Boogerd, Fred, F. J. Bruggeman, R. C. Richardson, A. Stephan, and H. V.
Westerhoff (2005). Emergence and its place in nature: a case study
of biochemical networks. Synthese 145, 131-164.
Braddon-Mitchell, David, and Frank Jackson (2007). Philosophy of Mind
and Cognition (Second Edition). Malden, MA: Blackwell Publishing.
Brigandt, Ingo (2010). Beyond reduction and pluralism. Erkenntnis 73,
295–311.
Calcott, Brett (2010). Wimsatt and the Robustness Family: Review of
Wimsatt’s Re-engineering Philosophy for Limited Beings. Biology &
Philosophy 26, 281-293.
Carnap, Rudolf (1932a). Die physikalische Sprache als Universalsprache
der Wissenschaft. Erkenntnis 2, 432–465.
Carnap, Rudolf (1932b). Psychologie in physikalischer Sprache.
Erkenntnis 3, 107–142.
Carrier, Martin, and Jürgen Mittelstrass (1991). Mind, Brain, Behavior.
The Mind-Body Problem and the Philosophy of Psychology. Berlin:
Walter de Gruyter.
Cartwright, Nancy (1983). How the Laws of Physics Lie. Oxford:
Clarendon Press.
Cartwright, Nancy (1999). The Dappled World. A Study of the Boundaries
of Science. Cambridge: Cambridge University Press.
Cartwright, Nancy (2002). Against Modularity, the Causal Markov
Condition, and any link between the two: Comments on Hausman
176

and Woodward. British Journal for the Philosophy of Science 53,

411-453.
Causey, Robert L. (1977). Unity of Science. Dordrecht: Reidel.
Chalmers, David J. (1996). The Conscious Mind: In Search of a
Fundamental Theory. Oxford: Oxford University Press.
Chalmers, David J., and Frank Jackson (2001). Conceptual analysis and
reductive explanation. The Philosophical Review 110, 315-360.
Churchland, Paul M. (1981). Eliminative materialism and the propositional
attitudes. Journal of Philosophy 78, 67-90.
Churchland, Paul M. (1985). Reduction, qualia, and the direct introspection
of brain states. Journal of Philosophy 82, 8–28.
Churchland, Patricia S. (1986). Neurophilosophy. Cambridge: The
MIT Press.
Churchland, Patricia S., and Terence J. Sejnowski (1992). The
Computational Brain. Cambridge, MA: MIT Press.
Clark, Andy (2001). Mindware. An Introduction to the Philosophy of
Cognitive Science. Oxford: Oxford University Press.
Craver, Carl F. (2002). Interlevel experiments and multilevel mechanisms
in the neuroscience of memory. Philosophy of Science 69, S83-S97.
Craver, Carl F. (2005). Beyond reduction: mechanisms, multifield
integration and the unity of neuroscience. Studies in History and
Philosophy of Biological and Biomedical Sciences 36, 373-395.
Craver, Carl F. (2007). Explaining the Brain. Oxford: Oxford University
Press.
Cummins, Robert (1983). The Nature of Psychological Explanation.
Cambridge, MA: MIT Press.
Cummins, Robert (2000). ’How does it work?’ versus ’what are the
laws?’ Two conceptions of psychological explanation. In F. Keil &
R. Wilson (eds.) Explanation and Cognition. Cambridge: MIT
Press, 117-144.
 177

Darden, Lindley, and Nancy Maull (1977). Interfield theories. Philosophy

of Science 44, 43-64.
Davidson, Donald (1970). Mental Events. In L. Foster & J. W. Swanson
(eds.) Experience and Theory. Amherst: University of Massachusetts
Press, 79-101. Reprinted in Block (1980), 107-119.
Dennett, Daniel C. (1991). Real patterns. The Journal of Philosophy 88,
27-51.
Do, Michael Tri Hoang, and King-Wai Yau (2010). Intrinsically
photosensitive retinal ganglion cells. Physiological Reviews 90,
1547-1581.
Dowe, Phil (1992). Wesley Salmon’s process theory of causality and the
conserved quantity theory. Philosophy of Science 59, 195-216.
Dupré, John (1993). The Disorder of Things. Metaphysical Foundations of
the Disunity of Science. Cambridge, MA: Harvard University Press.
Eronen, Markus I. (2009). Reductionist challenges to explanatory
pluralism: comment on McCauley. Philosophical Psychology 22,
637-646.
Eronen, Markus I. (2010-2011). Replacing functional reduction with
mechanistic explanation. Philosophia Naturalis 47-48, 125-153.
Esfeld, Michael, and Christian Sachse (2007). Theory reduction by means
of functional sub-types. International Studies in the Philosophy of
Science 21, 1-17.
Feigl, Herbert (1958). The ‘mental’ and the ‘physical’. Minnesota Studies
in the Philosophy of Science 2, 370–497.
Feyerabend, Paul K. (1962). Explanation, reduction, and empiricism.
Minnesota Studies in the Philosophy of Science 3, 28-97.
Feyerabend, Paul K. (1975). Against Method. London: Verso.
Fodor, Jerry A. (1974). Special sciences: Or, the disunity of science as a
working hypothesis. Synthese 28, 97–115.
Fodor, Jerry A. (1991). You can fool some of the people all of the time,
everything else being equal: Hedged laws and psychological
explanations. Mind 100, 19-34.
178

Frisch, Mathias (2009) ‘The Most Sacred Tenet’? Causal Reasoning in

Physics. British Journal for the Philosophy of Science 60, 459-474.
Gillett, Carl (2002). The dimensions of realization: a critique of the
Standard view. Analysis 62, 316-323.
Gillett, Carl (2003). The metaphysics of realization, multiple realizability,
and the special sciences. The Journal of Philosophy 100, 591-603.
Glennan, Stuart (1996). Mechanisms and the nature of causation.
Erkenntnis 44, 49-71.
Hacking, I. (1983). Representing and intervening. Introductory topics in
the philosophy of natural science. New York: Cambridge University
Press.
Hardcastle, Valerie (2001). The nature of pain. In W. Bechtel, P. Mandik,
J. Mundale, and R. S. Stufflebeam (eds.) Philosophy and the
Neurosciences. A Reader. Malden, MA: Blackwell, 295-311.
Hausman, Daniel M., and James Woodward (1999). Independence,
invariance and the Causal Markov Condition. British Journal for the
Philosophy of Science 50, 521-583.
Hempel, Carl G. (1949). The logical analysis of psychology. In H. Feigl &
W. Sellars (eds.) Readings in Philosophical Analysis. New York,
NY: Appleton-Century-Crofts, 373–384. Reprinted in Block (1980),
14–23.
Hempel, Carl G., and Paul Oppenheim (1948). Studies in the logic of
explanation. Philosophy of Science 15, 135-175.
Hill, Christopher S. (1997). Imaginability, conceivability, possibility, and
the mind-body problem. Philosophical Studies 87, 61-85.
Hill, Christopher S., and Brian P. McLaughlin (1999). There are fewer
things in reality than are dreamt of in Chalmers’s philosophy.
Philosophy and Phenomenological Research 59, 445-454.
Hitchcock, Christopher, and James Woodward (2003). Explanatory
generalizations, Part II: Plumbing explanatory depth. Noûs 37, 181-
199.
 179

Hoffmann-Kolss, Vera (unpublished manuscript). The supervenience

argument is alive and kicking.
Hooker, Clifford A. (1981). Towards a general theory of reduction. Part I:
Historical and scientific setting. Part II: Identity in reduction. Part
III: Cross-categorial reduction. Dialogue 20, 38-59, 201-236, 496-
529.
Horst, Steven (2007). Beyond Reduction. Philosophy of Mind and Post-
Reductionist Philosophy of Science. Oxford: Oxford University
Press.
Hüttemann, Andreas (2004). What’s Wrong With Microphysicalism?
London: Routledge.
Jackson, Frank (1998). From Metaphysics to Ethics: A Defense of
Conceptual Analysis. Oxford: Oxford University Press.
Kallestrup, Jesper (2008). Three strands in Kripke’s argument against the
identity theory. Philosophy Compass 3, 1255-1280.
Kemeny, John G., and Paul Oppenheim (1956). On reduction.
Philosophical Studies 7, 6-19.
Khalidi, Muhammad A. (2005). Against functional reductionism in
cognitive science. International Studies in the Philosophy of Science
19, 319-333
Kim, Jaegwon (1982). Psychophysical supervenience. Philosophical
Studies 41, 51-70.
Kim, Jaegwon (1984). Concepts of supervenience. Philosophy and
Phenomenological Research 45, 153-176.
Kim, Jaegwon (1992). Multiple realization and the metaphysics of
reduction. Philosophy and Phenomenological Research 52, 1–26.
Kim, Jaegwon (1993). The non-reductivist’s troubles with mental
causation. In J. Heil and A. Mele (eds.) Mental Causation. Oxford:
Clarendon Press, 189-210.
Kim, Jaegwon (1998). Mind in a Physical World. Cambridge, MA: MIT
Press.
180

Kim, Jaegwon (1999). Making sense of emergence. Philosophical Studies

95, 3-36.
Kim, Jaegwon (2002). Mental causation and consciousness: the two mind-
body problems for the physicalist. In C. Gillett & B. Loewer (eds.)
Physicalism and Its Discontents. Cambridge: Cambridge University
Press, 271-283.
Kim, Jaegwon (2005). Physicalism, or Something Near Enough. Princeton:
Princeton University Press.
Kim, Jaegwon (2008). Reduction and reductive explanation: is one
possible without the other? In J. Hohwy & J. Kallestrup (eds.) Being
Reduced. New Essays on Reduction, Explanation, and Causation.
Oxford: Oxford University Press, 93-114.
Kim, Jaegwon (2009). Mental causation. In B. McLaughlin, A.
Beckermann & S. Walter (eds.) The Oxford handbook of philosophy
of mind. Oxford: Oxford University Press, 29-52.
Kistler, Max (2005). Is functional reduction logical reduction? Croatian
Journal of Philosophy 14, 219-234.
Kripke, Saul (1980). Naming and Necessity. Cambridge, MA: Harvard
University Press.
Kuhn, Thomas (1962). The Structure of Scientific Revolutions. Chicago:
The University of Chicago Press.
Ladyman, James, and Don Ross (2007). Every Thing Must Go:
Metaphysics Naturalised. Oxford: Oxford University Press.
LePore, Ernest, and Barry Loewer (1989). More on making mind matter.
Philosophical Topics 17, 175-191.
Levin, Janet (2009). Functionalism. In E. N. Zalta (ed.) The Stanford
Encyclopedia of Philosophy (Winter 2009 Edition), URL =
<http://plato.stanford.edu/archives/win2009/entries/functionalism/>.
Levine, Joseph (1983). Materialism and qualia: the explanatory gap.
Pacific Philosophical Quarterly 64, 354-361.
Levine, Joseph (1993). On leaving out what it’s like. In G. Humphreys &
M. Davies (eds.) Consciousness. Oxford: Blackwell, 121–136.
 181

Levins, Richard (1966). The strategy of model building in population

biology. American Scientist 54, 421-431.
Lewis, David (1972). Psychophysical and theoretical identifications.
Australasian Journal of Philosophy 50, 249-258. Reprinted in Block
(1980), 207-215.
Lewis, David (1973). Causation. The Journal of Philosophy 70, 556-567.
Lewis, David (1980). Mad pain and martian pain. In Block (1980), 216–
222.
Lewis, David (1983). New work for a theory of universals. Australasian
Journal of Philosophy 61, 343-377.
Loewer, Barry (2007). Mental causation, or something near enough. In B.
P. McLaughlin & J. Cohen (eds.) Contemporary Debates in
Philosophy of Mind. Malden, MA: Blackwell Publishing, 243-264.
Looren de Jong, Huib (2002). Levels of explanation in biological
psychology. Philosophical Psychology 15, 441-462.
Machamer, Peter K., Lindley Darden, and Carl Craver (2000). Thinking
about mechanisms. Philosophy of Science 67, 1-25.
Marras, Ausonio (2002). Kim on reduction. Erkenntnis 57, 231-257.
Marras, Ausonio (2005). Consciousness and reduction. British Journal for
the Philosophy of Science 56, 335-361.
Matthewson, John, and Michael Weisberg (2009). The structure of
tradeoffs in model building. Synthese 170, 169-190.
Maxwell, James Clerk (1865). A dynamical theory of the electromagnetic
field. Philosophical Transactions of the Royal Society of London
155, 459-512. DOI:10.1098/rstl.1865.0008
McCauley, Robert N. (1996). Explanatory pluralism and the co-evolution
of theories of science. In R. N. McCauley (ed.) The Churchlands and
their Critics. Oxford: Blackwell, 17-47.
McCauley, Robert N. (2007a). Enriching philosophical models of cross-
scientific relations: incorporating diachronic theories. In M.
Schouten & H. Looren de Jong (eds.) The Matter of the Mind:
182

Philosophical Essays on Psychology, Neuroscience and Reduction.

Oxford: Blackwell Publishing, 199-223.
McCauley, Robert N. (2007b). Reduction: models of cross-scientific
relations and their implications for the psychology-neuroscience
interface. In P. Thagard (ed.) Handbook of the Philosophy of
Psychology and Cognitive Science. Amsterdam: Elsevier, 105-158.
McCauley, Robert N. (2009). Time is of the essence: Explanatory
pluralism and accommodating theories about long-term processes.
Philosophical Psychology 22, 611-635.
McCauley, Robert N., and William Bechtel (2001). Explanatory pluralism
and heuristic identity theory. Theory & Psychology 11, 736-760.
McLaughlin, Brian P. (1992). The rise and fall of British Emergentism. In
A. Beckermann, H. Flohr & J. Kim (eds.) Emergence or Reduction?
Essays on the Prospects of Nonreductive Physicalism. Berlin: Walter
de Gruyter, 49-93.
McLaughlin, Brian P. (2007). Type materialism for phenomenal
consciousness. In M. Velmans & S. Schneider (eds.) The Blackwell
Companion to Consciousness. Malden, MA: Blackwell Publishing,
431-444.
McLaughlin, Brian (2010). Consciousness, type physicalism, and inference
to the best explanation. Philosophical Issues 20, 266-304.
McLaughlin, Brian P., and Karen Bennett (2008). Supervenience. In E. N.
Zalta (ed.) The Stanford Encyclopedia of Philosophy (Fall 2008
Edition), URL =
<http://plato.stanford.edu/archives/fall2008/entries/supervenience/>.
Menzies, Peter (2008). The exclusion problem, the determination relation,
and contrastive causation. In J. Hohwy & J. Kallestrup (eds.) Being
Reduced. Oxford: Oxford University Press, 196-217.
Menzies, Peter, and Huw Price (1993). Causation as a secondary quality.
British Journal for the Philosophy of Science 44, 187-203.
Merricks, Trenton (2001). Objects and Persons. Oxford: Clarendon Press.
 183

Millikan, Ruth G. (1984). Language, thought, and other biological

categories. New Foundations for Realism. Cambridge, MA: MIT
Press.
Mitchell, Sandra D. (2002). Integrative pluralism. Biology and Philosophy
17, 55-70.
Mitchell, Sandra D. (2003). Biological Complexity and Integrative
Pluralism. Cambridge: Cambridge University Press.
Nagel, Ernest (1951). Mechanistic explanation and organismic biology.
Philosophy and Phenomenological Research 11, 327-338.
Nagel, Ernest (1961). The Structure of Science. London: Routledge &
Kegan Paul.
Neisser, Ulric, and Nicole Harsch (1992). Phantom flashbulbs: False
recollections of hearing the news about Challenger. In E. Winograd
& U. Neisser (eds.) Affect and accuracy in recall: Studies of
"flashbulb memories". Cambridge: Cambridge University Press, 9-
31.
Nickles, Thomas (1973). Two concepts of intertheoretic reduction. Journal
of Philosophy 70, 181–201.
Noë, Alva, and Evan Thompson (2004). Are there neural correlates of
consciousness? Journal of Consciousness Studies 11, 3-28.
Norton, John D. (2007). Causation as folk science. In H. Price & R. Corry
(eds.) Causation, Physics, and the Constitution of Reality. Russell’s
Republic Revisited. Oxford: Oxford University Press, 11-44.
Oppenheim, Paul, and Hilary Putnam (1958). Unity of science as a
working hypothesis. Minnesota Studies in the Philosophy of Science
2, 3-36.
Papineau, David (1998). Mind the gap. In J. Tomberlin (ed.) Philosophical
Perspectives 12: Language, Mind, and Ontology. Oxford: Blackwell,
373-88.
Pearl, Judea (2000). Causality: Models, Reasoning, and Inference.
Cambridge, UK: Cambridge University Press.
Pettit, Philip (1993). A definition of physicalism. Analysis 53, 213-223.
184

Place, Ullin T. (1956). Is consciousness a brain process? British Journal of

Psychology 47, 44–50.
Polger, Thomas W. (2004). Natural Minds. Cambridge: MIT Press.
Polger, Thomas W. (2007a). Realization and the metaphysics of mind.
Australasian Journal of Philosophy 85, 233-259.
Polger, Thomas W. (2007b). Some metaphysical anxieties of reductionism.
In M. Schouten & H. Looren de Jong (eds.) The Matter of the Mind.
Philosophical Essays on Psychology, Neuroscience and Reduction.
Oxford: Blackwell Publishing, 51-75.
Polger, Thomas W. (2008a). H2O, ‘Water’, and transparent reduction.
Erkenntnis 69, 109-130.
Polger, Thomas W. (2008b). Two confusions concerning multiple
realization. Philosophy of Science 75, 537-547.
Polger, Thomas W. (2009). Evaluating the evidence for multiple
realization. Synthese 167, 457-472.
Polger, Thomas, and Lawrence Shapiro (2008). Understanding the
dimensions of realization. Journal of Philosophy 105, 213-222.
Putnam, Hilary (1967). Psychological predicates. In W.H. Capitan & D.D.
Merrill (eds.) Art, Mind, and Religion. Pittsburg: Pittsburg
University Press, 37–48.
Putnam, Hilary (1975). Philosophy and our mental life. In Mind, Language
and Reality. Philosophical Papers, Volume 2. Cambridge: Camridge
University Press, 291-303.
Raatikainen, Panu (2010). Causation, exclusion, and the special sciences.
Erkenntnis 73, 349-363.
Raerinne, Jani (2011). Generalizations and Models in Ecology:
Lawlikeness, Invariance, Stability, and Robustness (Doctoral
dissertation). Philosophical Studies from the University of Helsinki
34.
Raerinne, Jani, and Markus I. Eronen (forthcoming). Multiple realizability
and biological laws. History and Philosophy of the Life Sciences.
 185

Richardson, Robert C. (1979). Functionalism and reductionism.

Philosophy of Science 46, 533–558.
Richardson, Robert C. (2007). Reduction without the structures. In M.
Schouten & H. Looren de Jong (eds.) The Matter of the Mind.
Philosophical Essays on Psychology, Neuroscience and Reduction.
Oxford: Blackwell Publishing, 123-145.
Richardson, Robert C. (2009). Multiple realization and methodological
pluralism. Synthese 167, 473-492.
Richardson, Robert C., and Achim Stephan (2007). Mechanism and
mechanical explanation in systems biology. In F. C. Boogerd, F. J.
Bruggeman, J. S. Hofmeyr & H. V. Westerhoff (eds.) Systems
Biology: Philosophical Foundations. Amsterdam: Elsevier, 123-144.
Richardson, Robert C., and Achim Stephan (2009). Reductionism (anti-
reductionism, reductive explanation). In M. Binder, N. Hirokawa &
U. Windhorst (eds.) Encyclopedia of Neuroscience. Heidelberg:
Springer, 3395-3398.
Russell, Bertrand (1912-1913). On the notion of cause. Proceedings of the
Aristotelian Society 13, 1-26.
Salmon, Wesley C. (1984). Scientific Explanation and the Causal
Structure of the World. Princeton, NJ: Princeton University Press.
Sarkar, Sahotra (1992). Models of reduction and categories of
reductionism. Synthese 91, 167–194.
Schaffner, Kenneth (1967). Approaches to reduction. Philosophy of
Science 34, 137-147.
Schaffner, Kenneth (1974). The peripherality of reductionism in the
development of molecular biology. Journal of the History of Biology
7, 111-139.
Schaffner, Kenneth (1993). Discovery and Explanation in Biology and
Medicine. Chicago, IL: University of Chicago Press.
Schouten, Maurice, and Huib Looren de Jong (2007). Mind reading and
mirror neurons: exploring reduction. In M. Schouten & H. Looren de
Jong (eds.) The Matter of the Mind. Philosophical Essays on
186

Psychology, Neuroscience and Reduction. Oxford: Blackwell

Publishing, 298-322.
Seeger, Max (unpublished manuscript). The reductive explanation of
boiling water in Levine’s explanatory gap argument.
(http://philpapers.org/archive/SEETRE)
Shapiro, Lawrence A. (2000). Multiple realizations. The Journal of
Philosophy 97, 635-654.
Shapiro, Lawrence A. (2004). The Mind Incarnate. Cambridge, MA: MIT
Press.
Shapiro, Lawrence A., and Elliott Sober (2007). Epiphenomenalism – the
do’s and the don’ts. In G. Wolters & P. Machamer (eds.) Thinking
about Causes: From Greek Philosophy to Modern Physics.
Pittsburgh: University of Pittsburgh Press, 235-264.
Sklar, Lawrence (1967). Types of inter-theoretic reduction. British Journal
for the Philosophy of Science 18, 109-124.
Sklar, Lawrence (1999). The reduction(?) of thermodynamics to statistical
mechanics. Philosophical Studies 95, 187-202.
Smart, John J. C. (1959). Sensations and brain processes. The
Philosophical Review 68, 141-156.
Smart, John J. C. (1971). Reports of immediate experiences. Synthese 22,
346-359.
Sober, Elliott (1999). The multiple realizability argument against
reductionism. Philosophy of Science 66, 542-564.
Spirtes, Peter, Clark Glymour, and Richard Scheines (1993). Causation,
Prediction, and Search. New York: Springer.
Steel, Daniel (2004). Can a reductionist be a pluralist? Biology and
Philosophy 19, 55-73.
Steel, Daniel (2006). Comment on Hausman & Woodward on the Causal
Markov Condition. British Journal for the Philosophy of Science 57,
219-231.
Stephan, Achim (1999). Emergenz: von der Unvorhersagbarkeit zur
Selbstorganisation. Dresden, München: Dresden University Press.
 187

Stephan, Achim (2001). How to lose the mind-body problem. Grazer

Philosophische Studien 61, 277-281.
Stephan, Achim (2004). Phänomenale Eigenschaften, phänomenale
Begriffe und die Grenzen reduktiver Erklärung. In W. Hogrebe & J.
Bromand (eds.) Grenzen und Grenzüberschreitungen. XIX.
Deutscher Kongress für Philosophie. Bonn, 23.-27. September 2002.
Vorträge und Kolloquien. Berlin: Akademie Verlag, 404-416.
Stephan, Achim (2006). The dual role of ‘emergence’ in the philosophy of
mind and in cognitive science. Synthese 151, 485–498.
Stoljar, Daniel (2009). Physicalism. In E. N. Zalta (ed.) The Stanford
Encyclopedia of Philosophy (Fall 2009 Edition), URL =
<http://plato.stanford.edu/archives/fall2009/entries/physicalism/>.
van Gulick, Robert (1992). Three bad arguments for intentional property
epiphenomenalism. Erkenntnis 36, 311-332.
van Riel, Raphael (2011). Nagelian Reduction beyond the Nagel Model.
Philosophy of Science 78, 353-375.
Walter, Sven (2008). The supervenience argument, overdetermination, and
causal drainage: assessing Kim’s master argument. Philosophical
Psychology 21, 671–694.
Walter, Sven, and Markus I. Eronen (2011). Reductionism, multiple
realizability, and levels of reality. In S. French & J. Saatsi (eds.)
Continuum Companion to the Philosophy of Science. London:
Continuum, 138-156.
Weisberg, Michael (2006). Robustness analysis. Philosophy of Science 73,
730-742.
Wilson, Robert A., and Carl F. Craver (2007). Realization: Metaphysical
and scientific perspectives. In P. Thagard (ed.) Handbook of the
Philosophy of Psychology and Cognitive Science. Amsterdam:
Elsevier, 81-104
Wimsatt, William C. (1976a). Reductionism, levels of organization, and
the mind-body problem. In Globus et al. (eds.) Consciousness and
188

the Brain. A Scientific and Philosophical Inquiry. New York:

Plenum Press, 205-267.
Wimsatt, William C. (1976b). Reductive explanation: a functional account.
In R. S. Cohen, C. A. Hooker, A. C. Michalos & J. W. van Evra
(eds.) PSA 1974: Proceedings of the 1974 Biennial Meeting
Philosophy of Science Association. Dordrecht: Reidel, 671-710.
Revised reprint in Wimsatt (2007), 241-273.
Wimsatt, William C. (1981). Robustness, reliability, and
overdetermination. In M. Brewer, & B. Collins (eds.) Scientific
Inquiry and the Social Sciences. San Fransisco: Jossey-Bass, 124-
163. Revised reprint in Wimsatt (2007), 43-74.
Wimsatt, William C. (1994). The ontology of complex systems: levels of
organization, perspectives, and causal thickets. Canadian Journal of
Philosophy S20, 207-274. Revised reprint in Wimsatt (2007), 193-
240.
Wimsatt, William C. (2000). Emergence as non-aggregativity and the
biases of reductionism(s). Foundations of Science 5, 269-297.
Revised reprint in Wimsatt (2007), 274-312.
Wimsatt, William C. (2007). Re-Engineering Philosophy for Limited
Beings. Piecewise Approximations to Reality. Cambridge, MA:
Harvard University Press.
Woodward, James (2000). Explanation and invariance in the special
sciences. British Journal for the Philosophy of Science 51, 197-254.
Woodward, James (2002). What is a mechanism? A counterfactual
account. Philosophy of Science 69, S366-S377.
Woodward, James (2003). Making Things Happen: A Theory of Causal
Explanation. Oxford: Oxford University Press.
Woodward, James (2008). Mental causation and neural mechanisms. In J.
Hohwy & J. Kallestrup (eds.) Being Reduced. Oxford: Oxford
University Press, 218–262.
Woodward, James (unpublished manuscript). Interventionism and causal
exclusion. (http://philsci-archive.pitt.edu/id/eprint/8651)
 189

Woodward, James, and Christopher Hitchcock (2003). Explanatory

generalizations, Part I: A counterfactual account. Noûs 37, 1-24.
Wright, Cory D. (2007). Is psychological explanation becoming extinct? In
M. Schouten & H. Looren de Jong (eds.) The Matter of the Mind.
Philosophical Essays on Psychology, Neuroscience and Reduction.
Oxford: Blackwell Publishing, 249-274.
Wright, Cory D., and William Bechtel (2007). Mechanisms and
psychological explanation. In P. Thagard (ed.) Handbook of
Philosophy of Psychology and Cognitive Science. Amsterdam:
Elsevier, 31-79.
Ylikoski, Petri, and Jaakko Kuorikoski (2010). Dissecting explanatory
power. Philosophical Studies 148, 201-219.

View publication stats