Reduction in Philosophy of Mind: A Pluralistic Account 9783110332131, 9783110322910

Citation preview

Markus I. Eronen Reduction in Philosophy of Mind A Pluralistic Account

EPISTEMISCHE STUDIEN Schriften zur Erkenntnis- und Wissenschaftstheorie Herausgegeben von / Edited by Michael Esfeld • Stephan Hartmann • Albert Newen Band 24 / Volume 24

Markus I. Eronen

Reduction in Philosophy of Mind A Pluralistic Account

Bibliographic information published by Deutsche Nationalbibliothek The Deutsche Nastionalbibliothek lists this publication in the Deutsche Nationalbibliographie; detailed bibliographic data is available in the Internet at http://dnb.ddb.de

North and South America by Transaction Books Rutgers University Piscataway, NJ 08854-8042 [email protected] United Kingdom, Ire, Iceland, Turkey, Malta, Portugal by Gazelle Books Services Limited White Cross Mills Hightown LANCASTER, LA1 4XS [email protected]

Livraison pour la France et la Belgique: Librairie Philosophique J.Vrin 6, place de la Sorbonne ; F-75005 PARIS Tel. +33 (0)1 43 54 03 47 ; Fax +33 (0)1 43 54 48 18 www.vrin.fr

2011 ontos verlag P.O. Box 15 41, D-63133 Heusenstamm www.ontosverlag.com ISBN 978-3-86838-125-2 2011 No part of this book may be reproduced, stored in retrieval systems or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming, recording or otherwise without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use of the purchaser of the work Printed on acid-free paper ISO-Norm 970-6 FSC-certified (Forest Stewardship Council) This hardcover binding meets the International Library standard Printed in Germany by CPI buch bücher.de

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 HanseWissenschaftskolleg 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.

PART I: Reduction in Philosophy of Science

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.

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 Æ 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 Nagelreductions, 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

Ontology: Identity

Replacement 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”) 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).

6

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. 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. 7

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 lowerlevel 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 wellstudied 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 longlasting, involves changes in the gene expression and protein synthesis, which in turn leads to changes in the dendritic spines, resulting in longterm 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 casespecific, 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 samelevel 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 molecularkinetic 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 preanalytic sense of the term. Nevertheless, in my view, seeing downwardlooking 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 causallymechanistically 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, 431Of course, the explanatory generalizations involving higher-level properties might turn out to be false, but that is an altogether different matter.

8

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.

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” (wquestions). 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 mindindependent. 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).

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 lowerorder properties of the same level, not upon lower-level properties. Supervenience thus generates an intralevel hierarchy of lower- and higherorder 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 abovementioned 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 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. 10

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 nonarbitrary 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.

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

11

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.

PART II: Rethinking Reduction in Philosophy of Mind

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).

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 14

This term was coined by Brian McLaughlin (1992). 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. 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.

15

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 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 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, 307308)

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 higherlevel 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.

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

18

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

69

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

70

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 temperatures but are otherwise the same, they count as tokens of the same type, not as different realizers of the psychological property. 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.

19

71

Shapiro spells out this account of realization in terms of “Rproperties:” 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 Rproperties 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 “pain” is manifest in humans and octopi, and then stated that this “pain” cannot be identical to any physical-chemical state.

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. 20

72

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 functions necessary for language, but this does not make them the same sort physiological states. The debate is far from resolved.

73

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 crosstheoretic 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. These approaches assume an intertheoretic (or bridge-law) model of reduction. If we take into account what has been discussed in Part I and

74

take downward-looking mechanistic explanation to be the strongest form of psychoneural reduction, MR is even less problematic. If there are oneto-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 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 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.

21

75

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. It is clear that Fodor’s arguments were targeting a very strict form of Nagel-reductionism. He dismissed this problem by one short remark:

76

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 instance, the explanatory power) of the generalization. It is also true that the special sciences are very much in the business of making

77

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 states are total states of the whole system, but psychological states are typically characterized in a much more fine-grained way (we typically

78

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 discussed in section 6.3.1). Analytic functionalism is nowadays far less popular than psychofunctionalism (see, however, Braddon-Mitchell &

79

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 fillerfunctionalism are closely associated, but there is no principled reason why 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. 22

80

one could not be psychofunctionalist).

a

role-analytic-functionalist

(or

a

filler-

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, rolepsychofunctionalism 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. Rolepsychofunctionalism 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. Rolefunctionalism is also clearly incompatible with Bickle’s “ruthless” metascientific reductionism (Part I, Chapter 1), since the latter posits that mental explanations (and a fortiori mental properties) are eliminable when neuroscientific explanations are complete.

81

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 rolefunctionalism. 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 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

82

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 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,

83

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 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 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.

23

84

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 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

85

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 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

86

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 fillerfunctionalism 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 20102011).

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 suchand-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, 1516). 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)

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

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 reduction does not by itself yield ontological simplification, which

93

(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 the reducing theory; if the reduction turns out to be bumpy, many or all mental properties are eliminated from the scientific ontology. 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 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). 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

95

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.), which neither individually nor taken together correspond to the simple storage envisioned by a priori analysis (see Bechtel (2008, Ch. 2) for

96

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 possible that in the end we are unable to find any neuroscientific properties playing the causal role of some psychological properties, and thus we This problem is obviously related to the issue of common-sense (analytical) vs. empirical functionalism (psychofunctionalism), discussed in section 5.5.

27

97

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 crosscut 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 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.

29

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 twolevel 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.

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. 31

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.

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

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

35

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 = Cfiber 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”

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

37

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 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. 38

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 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. 39

117

inconceivable 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 “PÆQ” is a priori, that is, when it is possible to know that “PÆQ” 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 nonbasic 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 “PÆQ” 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 of identities is that they allow us to transcend and get rid of correlations,

122

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

123

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

124

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). In a different line of research, a population of morphologically distinct retinal ganglion cells containing the photopigment melanopsin has

125

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 melanopsincontaining 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 identity claims is their potential to advance empirical research. The role of identities is not to explain correlations, but to connect levels or sciences

126

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 (2009) present (Bill Sikes = the burglar, water = H2O, Tully = Cicero, light waves = electromagnetic waves, etc.). However, what is at issue in the case

127

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 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

128

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. Some type physicalists take the explanatory argument very seriously: ”[t]he explanatory argument presents a formidable argument for reductive

129

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.

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 (rolepsychofunctionalism) 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.

PART III: A New Framework for Philosophy of Mind

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.

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.

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 selfsubsistent 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 differencemaking 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).

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 noncompeting. 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 nonmechanistic 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 simultaneously). The latter is a psychological phenomenon that is 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

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): 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

164

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

165

provides a more fine-grained framework for comparing higher- and lowerlevel 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 nonfundamental 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 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 55

McCauley (2009) uses scope as one criterion to define higher vs. lower levels. See Part I, Chapter 4 for more.

166

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., 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 56

This kind of invariance comes close to the idea of explanatory scope mentioned above.

167

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 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

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

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, selfconsciousness 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.

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: Cambridge, MA: MIT Press.

The New Wave.

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), 268305.

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, 181199.

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, 496529. Horst, Steven (2007). Beyond Reduction. Philosophy of Mind and PostReductionist 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 mindbody 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 = . 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 crossscientific 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 = . 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, 931. 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 Philosophy of Science 46, 533–558.

and

reductionism.

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 (antireductionism, 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 = . 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 the Brain. A Scientific and Philosophical Inquiry. New York: Plenum Press, 205-267.

188

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, 124163. 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), 193240. 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) Woodward, James, and Christopher Hitchcock (2003). Explanatory generalizations, Part I: A counterfactual account. Noûs 37, 1-24.

189

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.