Professional Documents
Culture Documents
Reduction in Philosophy of Mind: A Pluralistic Account: September 2011
Reduction in Philosophy of Mind: A Pluralistic Account: September 2011
net/publication/279917743
CITATIONS READS
23 544
1 author:
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Markus Ilkka Eronen on 09 July 2015.
PhD thesis
University of Osnabrück
August 2011
Markus I. Eronen
Abstract
The notion of reduction continues to play a key role in philosophy of mind and
philosophy of cognitive science. Supporters of reductionism claim that psychological
properties or explanations reduce to neural properties or explanations, while antireductionists
claim that such reductions are not possible. In this book, I apply recent developments in
philosophy of science, particularly the mechanistic explanation paradigm and the
interventionist theory of causation, to reassess the traditional approaches to reduction in
philosophy of mind. I then elaborate and defend a pluralistic framework for philosophy of
mind, and show how reductionist ideas can be incorporated into it. This leads to a novel
synthesis of pluralism and reductionism that I call pluralistic physicalism.
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):
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
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
PART I:
Reduction in Philosophy of Science
12
13
Introduction
In this part, I will discuss reduction4 as it has been understood in the
philosophy of science of the 20th (and 21st) century. Going through the
history (or prehistory) of reductionist ideas would be interesting, but this
thesis is not a historical one, and therefore I will only discuss the models
that are most relevant to contemporary debates. I will begin with the
development of intertheoretic models of reduction that started in the 1950s,
in the afterglow of logical positivism, and then go on to discuss more
recent accounts of reduction, most importantly “New Wave Reductionism”
and “Ruthless Reductionism.” I will argue that these approaches face fatal
problems, at least in the case of psychology and neuroscience, and that
“mechanistic explanation,” especially when supplemented with the
interventionist account of causation, provides a more accurate and
scientifically credible framework for approaching issues of reduction. In
the end, I will consider the question of levels and its relation to reduction,
focusing on the problems in current accounts of levels.
4
Throughout this thesis, I use the term “reduction” to refer to a single case of
accomplished or purported reduction: the reduction of thermodynamics to statistical
mechanics, the reduction of chemistry to physics, and so on. “Reductionism” refers to
a broader thesis, according to which reductions are to be expected (a predictive claim)
and/or desirable (a normative claim). Of course, different models of reduction yield
different reductionisms, and one can be reductionist regarding some domains of
science but not others. Therefore, for instance, “psychoneural Nagel reductionism”
means the thesis that psychology will be or should be reduced to neuroscience
following Nagel’s model of reduction.
14
15
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
molecules, and that the molecules are in constant motion and subject only
to forces of impact between themselves and the perfectly elastic walls of
the container. Nagel argued that the Boyle-Charles’ law is a logical
consequence of the principles of mechanics, when they are supplemented
with certain idealizing assumptions and connecting principles (bridge
principles), and that this is a representative example of the reductive
relation between thermodynamics and statistical mechanics.
Nagel’s model of reduction is neat and precise, but unfortunately fails
to account for many cases that are regarded as reductions. The model is too
demanding: it is very hard to find a pair of theories that would meet these
requirements. Even Nagel’s prime example, the reduction of
thermodynamics to statistical mechanics, is much more complicated than
Nagel thought (see, e.g., Sklar 1999; Richardson 2007). No one ever came
up with the derivation of the whole thermodynamics from statistical
mechanics, and it is likely to be computationally intractable. The
possibility of such a derivation is further diminished by the fact that many
central thermodynamical concepts, like entropy, are associated with a wide
variety of distinct concepts in statistical mechanics which do not exactly
correspond to thermodynamic entropy, neither separately nor taken
together. Historically speaking, scientists were not even aiming at such a
derivation – the goal was to show that thermodynamics is consistent with
statistical mechanics and to situate it in a broader Newtonian framework
(Richardson 2007).
Nagel’s model also has problems accounting for the fact that the
reducing theory often corrects the theory to be reduced, which entails that
the original theory was strictly speaking false. For example, Newtonian
physics showed that some principles of Galilean physics, such as the
assumption that uniformly accelerated gravitational free-fall is the
fundamental law of motion, were false. However, since logical deduction
is truth-preserving, the new reducing theory cannot both be true and
logically entail a false theory.
Problems like these led Paul Feyerabend (1962) to argue that no
formal accounts of scientific reduction are possible or necessary. The
majority of philosophers, however, responded by developing more
sophisticated models (Causey 1977; Schaffner 1967), culminating in what
18
came to be known as, using John Bickle’s (1998) term, “New Wave
Reductionism” (Bickle 1998; 2003; Hooker 1981; see also P. M.
Churchland 1985; P. S. Churchland 1986; Schaffner 1993). Here I will
focus on Bickle’s model of New Wave Reductionism, since it is the most
elaborate and explicit one.
Like its precursors, the New Wave model is a model with universal
scope that takes reduction to be a relation involving logical derivations
between theories. However, the crucial difference is that what is deduced
from the T1 is not the theory to be reduced itself (T2), but an analogue (or
“equipotent image”) of it (T2a). The fate of theory T2 and its ontological
posits is determined by the relation between T2 and the analogue T2a.
Importantly, the analogue can be formulated entirely in the vocabulary of
theory T1 – no bridge principles are needed to connect T1 and T2a. If the
analogy between T2 and T2a is strong and not much correction is needed,
T2 is reduced “smoothly” to T1, and many of its ontological posits can be
retained. If the theories are only weakly analogical and the amount of
correction implied to T2 is considerably large, the reduction is “bumpy,”
and many or all of the ontological posits of T2 will be eliminated. Thus,
depending on the strength of the analogy, each case of intertheoretic
reduction falls at a certain point on a continuum (Figure 1), where at one
extreme we have extremely smooth reductions resembling Nagel-
reductions, and at the other extreme complete replacement of the old
theory and its ontological posits. In this way, the New Wave model can
accommodate the idea that scientific progress sometimes involves
“revolutions” and replacements of old theories and their ontologies
(Feyerabend 1962; Kuhn 1962).
Bickle (1998) has also specified analytical tools for evaluating the
relation between T1 and T2, based on the structuralist/semantic view of
theories. His central example is a familiar one: the reduction of
thermodynamics to statistical mechanics (Bickle 1998, 33-40). Bickle
shows that we can derive an analogue structure of the ideal gas law of
thermodynamics from statistical mechanics and the kinetic theory. This
analogue structure exactly mimics the ideal gas law while containing only
terms of statistical mechanics and microphysics.
19
------------------------------------------------------------------------
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.
6
Nagel already distinguished between heterogeneous and homogeneous reductions,
but probably the first one to emphasize that there are also different types of
heterogeneous reductions was Thomas Nickles (1973).
21
reduction that New Wave reductionists present are intralevel cases, and
give no reason to expect eliminative reductions in interlevel contexts
(McCauley 1996; 2007b). In particular, they provide no support for New
Wave reductionists’ claims that psychology will be reduced to
neuroscience, since psychology and neuroscience are at different levels.
Replacements and eliminations happen across time between
competing theories at the same analytical level, not in interlevel contexts.
A look at scientific practice shows that scientists are not interested in
eliminating sciences at adjacent levels, but look there for support,
guidance, evidence, and so on. Incompatibilities between sciences do not
lead to eliminations, but to further inquiries, ”co-evolution” of theories
(Hooker 1981; Wimsatt 1976a; 2007) and possibly to the development of
interlevel theories (Darden & Maull 1977).
The second problematic assumption of the New Wave model,
inherited from Nagel’s account, is that the relata of reductions are
exclusively theories that are construable in some formal or semi-formal
way, either as sets of sentences (the “received view” of theories), or as sets
of models meeting certain set-theoretic conditions (the
structuralist/semantic view of theories). However, some generally
accepted cases of scientific reduction — for example, the reduction of
genetics to molecular biology — do not seem to involve such formal
theories (Sarkar 1992). In general, well-structured theories that could be
handled with logical tools are rare and peripheral in the special sciences,
including psychology and neuroscience. There are of course “theories” in a
broad sense of the term in psychology and neuroscience, like the LTP
theory of memory consolidation or the global workspace theory, but these
are not formal theories, and can hardly be the starting points or results of
deductions. Reductions and reductive explanations in psychology and
neuroscience cannot be conceived as logical derivations. Instead, these
disciplines typically look for descriptions of mechanisms that can serve as
explanations for patterns, effects, capacities or phenomena (see next
chapter).
An intertheoretic reductionist could still claim that even though
neuroscientists and psychologists do not present their theories in a formal
way, they could be formalized. For example, Kenneth Schaffner (1993) has
22
7
On the other hand, Schaffner (particularly Schaffner 1974) has also argued for the
“peripherality of reductionism” in biology. What he means by this, roughly speaking,
is that reduction is not the explicit goal of research in biology, and that the
methodology of biological research is not generally speaking reductionistic.
23
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
phase of LTP (E-LTP). The late phase (L-LTP), which is more long-
lasting, involves changes in the gene expression and protein synthesis,
which in turn leads to changes in the dendritic spines, resulting in long-
term increase in the efficiency of synaptic transmission.
One of the core features of mechanistic explanations is that they are
multilevel: focusing on just one level is not sufficient for full understanding
of the phenomenon. Craver (2002; 2007, 165-170) identifies four levels in
the case of spatial memory and LTP: the behavioral-organismic level,
which involves various types of memory and learning, the conditions for
memory consolidation and retrieval, and other phenomena that are
typically investigated by behavioral and psychological tasks. The
computational-hippocampal level involves the structural features of the
hippocampus and its overall role in the mechanisms of memory, its
connections to other brain regions and the computational processes it is
thought to perform. The electrical-synaptic level includes neurons,
synapses, dendritic spines, axons, action potentials and so on. At the
bottom of this hierarchy is the molecular-kinetic level, where we find
glutamate, NMDA and AMPA receptors, Ca2+ ions, and Mg2+ ions, which
bind to each other, break, phosphorylate, and so on. Craver calls these
“mechanistic levels” or “levels of mechanisms.” They are levels of
composition, where the relata are behaving mechanisms at higher levels
and their components at lower levels. These levels are local and case-
specific, not universal divisions of nature or science. I will discuss the
question of levels in more detail in Chapter 4.
Mechanistic explanations have both a “downward-looking” and an
“upward-looking” aspect (in addition to the obvious “horizontal” or same-
level aspect). In the LTP case, one is looking upward when, in order to
understand the computational properties of the hippocampus, one is
studying its role in the overall cognitive system, or when, in order to
understand the role of the molecular processes of LTP, one is looking at
the larger computational-hippocampal framework. In contrast, one is
looking downward when memory consolidation is explained by appeal to
the computational processes at the hippocampal level, or when the synaptic
LTP mechanism is explained by appeal to activities at the molecular-
kinetic level.
28
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
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
received only 100 mg of penicillin. The causal claim “giving the patient
200 mg of penicillin caused her recovery” is true if the contrastive focus is
taken to be no penicillin at all. However, a more appropriate interpretation
takes the contrast to be giving the patient less than 200 mg of penicillin,
and in this case the claim is false – the patient would have recovered also
when given, say, 150 mg of penicillin.
The interventionist account also allows for several noncompeting
representations of one and the same system. What variables we choose to
include in the representation depends on the question at hand. For example,
if we are interested in the restoring force that a spring exerts when pulled,
we can use Hooke's law (F = kX, where F is the restoring force exerted by
the spring, X is the displacement of the spring’s end, and k is the spring
constant) and include only variables F and X (and constant k). However, if
we need to know in more detail why the spring exerts the force in
accordance to constant k, we will have to include various variables
representing the physical properties of the spring. Importantly, this does
not make causal judgments subjective, since the counterfactual patterns of
dependence that make the causal claims true or false are mind-
independent. Once the variables and representations are fixed, causal
claims are true or false in a mind-independent way.
The interventionist account is a nonreductive and circular account of
causation, meaning that it does not provide a definition for causation in
non-causal terms. It gives criteria for distinguishing causal relationships
from noncausal relationships (most importantly correlation), but does not
give an answer to the metaphysical question of what causation really is.
However, the history of philosophy of causation is a history of failures to
provide such an account, which suggests that such a reductive and
noncircular account is not forthcoming. Furthermore, if the interventionist
account captures the notion of causation in science and everyday life, and
provides the criteria for distinguishing causal from noncausal relationships,
it seems to be all we need from a notion of causation (with the possible
exception of armchair metaphysics).
In the interventionist model, causation, causal relevance, and causal
explanation are coextensive: X is cause for Y iff X is causally relevant for Y
iff X figures in a causal explanation for Y.
38
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
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
10
According to Wimsatt (2000;2007), testing the conditions for aggregativity reveals
interesting features of the organization of a system, and different kinds of failures of
aggregativity represent different kinds of emergence. Since truly aggregative systems
are rare, emergence is more the rule than the exception.
49
11
These considerations are of course closely related to the general debate on scientific
realism.
52
53
PART II:
Rethinking Reduction in
Philosophy of Mind
56
57
Introduction
In this part, I will critically go through the most important approaches to
reduction in philosophy of mind, in light of what has been discussed in Part
I. I will proceed in a roughly chronological order. However, the main focus
will not be on history, but rather on problems and positions that are
relevant for contemporary philosophy of mind.
I will start with British Emergentism (section 5.1.) of the early 20th
century, focusing on C. D. Broad, who was a forerunner in many debates
that are now central in philosophy of mind. Then I will briefly discuss
translational reduction in logical positivism (section 5.2.) and how its
failure lead to the identity theory (section 5.2.) and to models of
intertheoretic reduction (Part I, Chapter 1).
Since the 1960s, the debates about reduction in philosophy of mind
have been quite distinct from those in philosophy science. In philosophy of
mind, Nagel’s model of intertheoretic reduction was widely accepted.
However, the argument from multiple realizability (section 5.3) apparently
made Nagel-reduction of psychology impossible, and also appeared to
refute ontological reduction of mental properties (the identity theory). This
led to something close to an antireductionist consensus, which lasted at
least until the 1990s. However, most philosophers of mind still wanted to
be good physicalists, which resulted in various attempts to formulate
nonreductive physicalism (section 5.6). The most popular form of
nonreductive physicalism was (and perhaps still is) functionalism (section
5.5.).
In the 1990s, the deficiencies in Nagel’s model were finally
acknowledged and accepted also in philosophy of mind, leading to
alternative models, most importantly functional reduction (Chapter 6).
Applying the functional model, many philosophers have recently argued
that phenomenal consciousness is fundamentally irreducible, or that there
is an explanatory gap between phenomenal consciousness and the physical
domain (Chapter 7). On the other hand, many prominent philosophers have
recently argued that the identity theory can still provide the solution to the
problem of phenomenal consciousness, leading to the new type physicalism
(Chapter 8).
58
59
13
This term was coined by Brian McLaughlin (1992).
14
See McLaughlin (1992) and Stephan (1999) for more on the history of emergentism.
60
The other two kinds of theories deny that peculiar components are
necessary for explaining behavior, and try to explain the differences
wholly in terms of difference of structure. According to one of them, the
characteristic behavior of a whole could, at least in theory, be deduced
from a sufficient knowledge of the behavior of the components. This kind
of theory Broad calls “mechanistic.”15 The most obvious example of a class
of objects to which mechanistic theories apply is mechanical devices. For
example, there is hardly any doubt that the behavior of a clock can be
deduced from sufficient knowledge of its components.
The other of these theories holds that “the characteristic behaviour of
the whole could not, even in theory, be deduced from the most complete
knowledge of the behaviour of its components, taken separately or in other
combinations, and of their proportions and arrangements in the whole”
(Broad 1925, 59). Broad calls this the theory of emergence. Its core is
captured in the following definition:
Put in abstract terms the emergent theory asserts that there are certain wholes,
composed (say) of constituents A, B, and C in a relation R to each other; that all
wholes composed of constituents of the same kind as A, B, and C in relations of
the same kind as R have certain characteristic properties; that A, B, and C are
capable of occurring in other kinds of complex where the relation is not of the
same kind as R; and that the characteristic properties of the whole R(A, B, C)
cannot, even in theory, be deduced from the most complete knowledge of the
properties of A, B, and C in isolation or in other wholes which are not of the
form R(A, B, C). (Broad 1925, 61)
15
This resembles contemporary accounts of mechanistic explanation discussed in Part
I, Chapter 2, but also differs in some crucial respects – for instance, Broad’s account
centrally involves the notion of “deduction,” while the contemporary accounts do not.
61
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:
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
16
More recently, David Chalmers (1996) has defended the view that there are such
fundamental and inexplicable laws connecting physical properties with phenomenal
properties.
64
thus several philosophers have argued that they are strongly emergent
properties (Stephan 1999; 2006; Kim 1999) or that they present a
fundamental explanatory gap (Levine 1983; 1993). I will return to this in
Chapter 7.
British Emergentism is sometimes seen as a precursor of nonreductive
physicalism. Like nonreductive physicalists, the emergentists claimed that
everything is composed of physical stuff (and only of physical stuff), but
that the special sciences are nonetheless irreducible. Jaegwon Kim (1999,
5) has gone so far as to claim: “The fading away of reductionism and the
enthronement of nonreductive materialism as the new orthodoxy simply
amount to the resurgence of emergentism … It is no undue exaggeration to
say that we have been under the reign of emergentism since the early
1970s.” However, equating emergentism with nonreductive physicalism in
this way is problematic. While the views of Broad sometimes come close
to nonreductive physicalism, other emergentists had more radical views
regarding the causal powers and the ontological status of emergent
properties, for example claiming that emergent properties have novel
causal powers and can also exert causal influence “downwards” onto the
physical level. Furthermore, in contrast to the British Emergentists,
nonreductive physicalists do not claim that higher-level properties are
emergent in the strong sense (see above) – they typically argue that higher-
level properties are physically realized (see section 5.6 and Chapter 10).
Therefore, it is sensible to distinguish between nonreductive physicalism
and emergentism, the latter being a more radical position.
British Emergentism also bears similarity to explanatory pluralism and
what I call pluralistic physicalism (Chapter 10), since the emergentists
claimed that even though everything is composed of physical stuff, there
are irreducible and indispensable chemical, biological, psychological, etc.,
explanations and properties. However, it is again important to emphasize
that the emergentists also made other more radical claims that go far
beyond what I am defending.
British Emergentism eventually drifted into the periphery of
philosophy before the middle of the 20th century. However, this was not
due to philosophical inconsistencies or problems, as McLaughlin (1992)
and Stephan (1999, 129-155) have pointed out. The first main reason for
65
the downfall was advances in science (McLaughlin 1992): for example, the
explanation of chemical bonding in terms of electron bonds and the
discovery of DNA made the existence of emergent properties or laws in
chemistry and biology seem highly improbable.17 The second main reason
was the rise of logical positivism in the 1930s (as Kim (1999) points out).
This trend in philosophy was anti-metaphysical and hyper-empiricist and
the supposedly vague concept of emergence had no place in its view of the
sciences, although some positivists tried to adapt the concept of emergence
for their own purposes by making it weaker and theory-relative (e.g.,
Hempel & Oppenheim 1948).
However, as I have already pointed out, emergence has made a
comeback in contemporary philosophy of mind. Since the 1990s, it has
also been a central and much-discussed concept in cognitive science and
complexity studies. As Stephan (2006) shows, this concept of emergence is
substantially different and weaker that the one in philosophy of mind.
Philosophers who have presented such weaker accounts of emergence
include, for example, Batterman (2001), Bedau (1997), Clark (2001), and
Wimsatt (2000; 2007). Since my main focus is on philosophy of mind,
these alternative accounts are beyond the scope of this thesis.
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)
18
For Place (1956) the ‘is’ was actually ‘is’ of composition, not of identity, and thus
his view differs from the others in this respect.
68
The most influential argument against reductionism and the identity theory,
the argument from multiple realizability, was brought to the discussion by
Hilary Putnam (1967), and has remained in the center of philosophical
attention to this day. Putnam took as an example “pain,” or the state of
being in pain, and argued that the identity theorist would have to claim that
there is a physical-chemical state such that any organism – be it human,
octopus, or an alien – is in pain if and only if its brain is in that physical-
chemical state. He further argued that if we can find even one
psychological state that can be found in different species, but whose
physical-chemical “realizer” is different in these species, the identity
theory has collapsed. It seemed extremely likely that there are such
psychological states, since states like “pain” or “hunger” appeared to be
realized in different ways in different species. Therefore, Putnam
concluded that the identity theory was empirically and methodologically
problematic, and proposed that we replace it with functionalism (see
section 5.5).
Multiple realizability (MR) was generally considered extremely
plausible, both empirically and conceptually, and seemed to refute both the
identity theory and reductionism (of the Nagel kind). As an illustration,
consider this quote from the year 1989:
19
Aizawa & Gillett (2009) defend an account of realization that apparently leads to
such promiscuous and ubiquitous multiple realizability, and also try to argue that this
does not make multiple realizability trivial.
71
temperatures but are otherwise the same, they count as tokens of the same
type, not as different realizers of the psychological property.
Shapiro spells out this account of realization in terms of “R-
properties:”
20
The point of this section is not to defend any particular approach to realization, but
to show that the question of multiple realizability is complicated and far from settled,
and that it is not as tightly connected to the question of reduction as has been
traditionally assumed. In Chapters 6 and 10, I briefly return to the notion of realization
and argue that no metaphysical realization relation is needed for understanding
reduction and reductive explanation.
72
“pain” is manifest in humans and octopi, and then stated that this “pain”
cannot be identical to any physical-chemical state.
There are certainly some “grains” for psychological and neural states
that make MR claims false, and some that make them true. As Richardson
(2009) points out, the problem is that of finding some principled reasons
for choosing one grain instead of another, and this is no easy task. Another
way of putting this is that multiple realizability is not absolute, but a
concept that comes in degrees and depends on the conceptual framework(s)
that are relevant for the question at hand (Polger 2008b).
The question of multiple realizability of the mental then turns out to
have three parts: are psychological properties multiply realized by neural
properties when we have (1) settled on the right “theoretical grains,” (2)
agreed upon the right way of classifying properties as same or different,
and (3) agreed upon the right notion of realization? This makes multiple
realizability far less obvious and plausible than has generally been thought.
Furthermore, the answers to these questions also depend on empirical
matters, not just philosophical considerations.
One hotly debated case that prima facie supports multiple
realizability claims is that of neural plasticity and recovery of function. It
has been the most central empirical part in the arguments of the supporters
of MR (e.g., Block and Fodor 1972). For example, children who
experience severe head trauma and lose some of their linguistic abilities
often recover these abilities as other brain areas take over the functions of
the damaged area. Richardson (2009) argues that in these cases the same
psychological functions are realized by different structures, and this is
multiple realization. On the other hand, Polger (2009) and Shapiro (2004)
argue that we cannot conclude so easily that these are cases of MR; one
needs to show that the different realizing brain areas also differ
functionally after they have adapted to the new task. If they do not differ
functionally, they are not really different realizers, but rather the same
realizers located in different brain areas. Polger and Shapiro then show that
in at least some cases of recovery of function the brain areas that take over
the functions actually become functionally and organizationally similar to
the old areas. In response, Richardson (personal communication) claims
that the realizers obviously have to be functionally similar to carry out the
73
functions necessary for language, but this does not make them the same
sort physiological states. The debate is far from resolved.
Due to constraints of space, I cannot evaluate the empirical evidence
for and against multiple realizability here (see Raerinne & Eronen
(forthcoming) for more discussion). However, in spite of the arguments of
Polger and Shapiro, I find it very plausible that there are at least some
multiple realizations of psychological functions by neural mechanisms,
even if we have a strict notion of realization (see Aizawa & Gillett 2009
for some candidate cases). Whatever the right “theoretical grain” of
psychological functions turns out to be, it will probably be rougher than
that of neural mechanisms. Then it is not surprising if one psychological
function can be realized by different neural mechanisms, in the sense that
these mechanisms perform the same roughly defined function, but perform
it significantly differently from each other.
In any case, even if there is MR, this does not necessarily prevent
reduction, even of the traditional kinds. Richardson (1979) was the first to
show that even Nagel-reduction is compatible with MR, as long as we
don’t require that bridge laws have to express biconditionals, but accept
also material conditionals. Robert Batterman (2001), Clifford Hooker
(1981) and Patricia Churchland (1986), among many others, have pointed
out that there is multiple realizability also in physics: for instance,
temperature is one thing in a gas, but something else in a solid, and yet
something very different in a plasma. Yet, this is no obstacle for reduction
or reductive identification of temperature. The reductions and cross-
theoretic identifications just have to be “domain-specific:” temperature is
identified with different things in different domains. Temperature in a gas
is identical to the mean molecular kinetic energy, while temperature in a
plasma is identical to something else, and so on.
In the same vein, Kim (1992) has argued that instead of a one-shot
general Nagel-reduction of a certain mental property, we should restrict the
bridge laws or identities to appropriately individuated biological species
and physical structure types. For instance, instead of having one bridge law
that connects pain with a physical property (or a disjunction of properties),
we will have a bridge law or identity for each structure type S connecting
pain-in-S with a physical property in S.
74
derivable from laws of the reducing science via biconditional bridge laws.
This of course means that if reductionism is true, all the laws of all the
special sciences have to be derivable from the laws of physics. Fodor then
argued that if we assume that laws, including bridge laws, have to connect
“natural kinds” (Fodor defined them roughly as the predicates whose terms
are bound variables in laws), it follows that reductionism requires that
every natural kind is, or is coextensive with, a physical natural kind.
Fodor further argued that this is a very unfortunate conclusion, since
(a) interesting generalizations can be made about events whose physical
descriptions have nothing in common, (b) it is often the case that whether
the physical descriptions have anything in common is entirely irrelevant to
any epistemologically important properties of the generalization, and (c)
the special sciences are very much in the business of making
generalizations of this kind. Fodor took up the example of monetary
exchanges: economics is full of generalizations involving monetary
exchanges, but the physical descriptions of monetary exchanges have
nothing interesting in common. There is no physical natural kind
corresponding to money or monetary exchanges. Similar reasoning can be
applied to the case of psychology and neurology: if psychology is
reducible to neurology, then for every psychological natural kind predicate
there must be a coextensive neurological natural kind predicate. However,
this does not seem to be the case, since it is likely that there are systems
that satisfy psychological natural kind predicates but do not satisfy any
single neurological kind predicate (multiple realizability).
Fodor proposed that we abandon the thesis of “unity of science”
(Oppenheim and Putnam 1958), which according to him was equal to the
thesis that all sciences should be reduced to physics. He claimed that the
reason why many philosophers had accepted this view was that they
wanted to endorse “generality of physics:” all events that fall under the
laws of any science are physical events and hence fall under the laws of
physics. However, according to Fodor, generality of physics is a much
weaker thesis than reductionism, and all we need to assume for it is “token
physicalism,” which states that all events that sciences talk about are
physical events.
76
Unfortunately for Fodor, this is not the case. As Richardson (1979) first
pointed out, material conditionals and one-to-many mappings are enough
to fulfill Nagel’s conditions for reduction. This means that there is even a
form of Nagel-reductionism for which Fodor’s argument is a nonstarter.
Furthermore, at the time when Fodor’s article was published, several
philosophers of science had already pointed out that Nagel’s model
suffered from serious defects, and thus had to be replaced or adjusted (e.g.,
Feyerabend 1962; Schaffner 1967; Sklar 1967). It is not clear to what
extent Fodor’s argument applies to more sophisticated models, such as the
New Wave model. If Bickle (1998) is right, no biconditional bridge laws
linking the reduced and the reducing theory are needed in reductions that
comply to the New Wave model, since the reducing theory simply
“displaces” the older theory. If this is the case, Fodor’s argument loses its
force. It is also clear that the argument says nothing against the possibility
of reduction understood as downward-looking mechanistic explanation.
However, Fodor’s points about the epistemological importance of
special science generalizations still hold, and are strongly supported by the
interventionist model of causation and the insights of explanatory
pluralists. If we look at points (a), (b), and (c) in this light, we see that they
all turn out true. If a generalization is explanatory in the interventionist
sense, i.e., if it is an invariant generalization, the question whether the
physical descriptions of the variables have anything in common is indeed
entirely irrelevant to the epistemologically important properties (for
77
5.5. Functionalism
states are total states of the whole system, but psychological states are
typically characterized in a much more fine-grained way (we typically
ascribe to people many distinct beliefs, desires, feelings, and so on, at any
given moment).
The later versions of functionalism no longer invoked the Turing
machine. The idea was still that mental states or properties are defined by
their function or causal role, but the functions or causal roles were to be
specified by empirical (cognitive) psychology. Mental states were
identified with functionally defined states that appear in our best
psychological theories. Mental states and psychological explanations were
also considered to be autonomous and irreducible, mainly based on
considerations outlined in previous sections (multiple realizability and
Fodor’s arguments). This, the most influential branch of functionalism, is
often called psychofunctionalism (Block 1978) or empirical functionalism.
It is a common implicit background assumption in philosophy of mind,
even though it is difficult to find explicit formulations of the position or
detailed positive arguments for it.
A very differently motivated branch of functionalism was analytic
functionalism, where the functional roles of mental states were not to be
provided by science, but by a priori analyses of our ordinary, “folk
psychological,” mental concepts (Armstrong 1968; Lewis 1972; Smart
1971). The idea was that the defining functional roles of mental states were
to be extracted from the huge collections of “common knowledge” about
the mind. For instance, a part of the implicit common knowledge about
pain is that it is typically caused by bodily damage and that it causes
avoidance of the source of pain. This approach can be seen as an extension
of logical behaviorism: mental states are defined a priori, but not just in
terms of observable behavior – they are defined in terms of their causal
relations to stimuli, responses, and other mental states. A large part of the
motivation for this view stemmed from the background assumption that
our common sense ideas about mental states must be approximately true,
or else mental concepts turn out to be meaningless. The most obvious
problem with analytic functionalism is that it seems very likely that at least
some of our commons sense ideas about mental states will turn out to be
fundamentally false (consider for example the case of memory, further
79
states are in the end just the neurological realizers of the functional role,
and not functional states themselves. However, this kind of functionalism
is closer to eliminativism than reductionism, insofar as it claims that
functionally defined mental states are not real as such. I will not discuss
this further here, since this reading of filler-functionalism comes very close
to the functional model of reduction, which is the topic of the next chapter.
23
Donald Davidson (1970) argued for “anomalous monism,” a form of physicalism
(monism) that was supposedly compatible with multiple realizability and the
irreducibility of psychology. Like the position of Fodor, it was supposed to present an
alternative to type identity theory and Nagel-reductionism. Davidson’s argument has
approximately the following structure. We start with three principles about mental
events that are taken to be true: (1) Some mental events causally interact with some
physical events. (2) Events related as cause and effect are covered by strict laws. (3)
There are no strict laws on the basis of which mental events can predict, explain, or be
predicted or explained by other events. It is easy to see that, under some plausible
assumptions, these three principles cannot all be true. However, according to
Davidson, this tension can be resolved by the following thesis: (4) Every causally
interacting mental event is token-identical to some physical event. This leads to
anomalous monism: there are no strict laws for mental event-types (mental
anomalism), but every mental event is identical to some physical event (monism). The
identity here is not the type identity of the identity theory, but token identity, which
merely states that every mental event (token) is identical to some physical event
(token). This position is nonreductive in the sense that it states that there can be no
physical (or non-mental) explanations of the mental. Anomalous monism has been
criticized from many angles, and has been a controversial position from the start. I will
not discuss it in more detail here, since it relies on very strong metaphysical
assumptions about the nature of causation and explanation, and has few proponents
these days.
84
ectoplasm). However, even this may not be sufficient, since, for instance, a
property dualist might claim that the psychophysical laws linking mental
and physical properties are fundamental laws that hold in all nomologically
possible worlds, so that property dualism would be compatible with the
above definition of supervenience physicalism. The issue of supervenience
and physicalism is far from settled; see, for example, Stoljar (2009) or
McLaughlin and Bennett (2008) for more.
Perhaps the most influential argument against nonreductive
physicalism is the exclusion argument, whose most ardent proponent has
been Jaegwon Kim (1993; 1998; 2002; 2005). Very roughly, the exclusion
argument states that mental properties cannot have causal powers of their
own, since the physical domain is causally closed, and all the causal work
is done by the physical properties that realize the mental properties. (I will
discuss the argument in detail in Chapter 11.) Therefore, if mental
properties are irreducible in the sense of being distinct from physical
properties, they cannot make any causal contribution to the physical world.
If Kim is right in this, there are basically two options left for the
nonreductive physicalist. The first is to insist that mental properties are
irreducible but causally impotent properties. This leads to emergent
property dualism and epiphenomenalism, and effectively means giving up
on physicalism. The second option is to accept that mental properties are
identical to physical properties (reductive physicalism), or that they do not
exist at all (eliminativism). However, this effectively means giving up on
the nonreductive part of nonreductive physicalism. Thus, if Kim is right, it
seems that nonreductive physicalism is an incoherent position.
From the point of view of the position I have defended in Part I, the
main problem with the whole debate on nonreductive physicalism is that it
has been based on unrealistic and mistaken ideas about reduction,
reductive explanation and causation. In Part III, I will argue that if we
accept explanatory pluralism and take downward-looking mechanistic
explanation to be the strongest form of reduction in the relevant sciences,
we can in fact have both: autonomous and irreducible (though not
completely independent) special sciences on the one hand, and physicalism
(or something near enough) on the other. I will also argue that the
86
6. Functional Reduction24
In recent years, the “functional reduction model” has become something
like the standard model of reduction in philosophy of mind. The model is
by no means new: its main ideas are already visible in the analytic filler-
functionalism of Lewis (1972). Lewis’ idea was that a given mental state
M is defined functionally in terms of its causal role, but M is nothing more
than the physical states that occupy this role (see also section 5.5. and
Walter & Eronen 2011). More recently philosophers like Levine (1983;
1993), Chalmers (1996), Jackson (Chalmers and Jackson 2001), and Kim
(1998; 2005) have presented somewhat varying models of functional
reduction based on this general approach. All of these authors have then
applied the supposedly general model of reduction to the problem of
phenomenal consciousness, arguing that phenomenal properties are
fundamentally irreducible, or that there is an “explanatory gap” between
phenomenal properties and the physical domain.
I will focus here on Kim’s model of functional reduction, since it is
the most explicit and detailed one. I will argue that the functional model
fails to capture the role and nature of reductive explanation in science
(particularly psychology and neuroscience). Furthermore, I will show that
if we try to revise the functional model in order to make it more
scientifically credible, it turns out that the revised model is not
significantly different from mechanistic explanation. In the next chapter, I
will consider the consequences of this for the explanatory gap argument.
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:
causes and effects. So M is now the property of having a property with such-
and-such causal potential[.] (Kim 1998, 98)
25
Esfeld and Sachse (2007) have argued that by introducing functional sub-types we
can have property identities and conservative functional reductions, multiple
realizability notwithstanding.
92
26
According to Marras (2002, 237), bridge laws are not central even in Nagel’s own
articulation of the model. See also van Riel (2011) for further criticism of the
traditional interpretation of Nagel’s theory of reduction.
94
the reducing theory; if the reduction turns out to be bumpy, many or all
mental properties are eliminated from the scientific ontology.
The New Wave model and intertheoretic models are not applicable
to the purported reduction of the mind or psychology – but for reasons that
Kim does not mention or discuss. As I have shown in Part I, they do not
work because they require that both the theory to be reduced and the
reducing theory are construable in some formal or semi-formal way, either
as sets of sentences (the “received view” of theories), or as sets of models
meeting certain set-theoretic conditions (the structuralist/semantic view of
theories). The problem is that neither psychologists nor neuroscientists are
in the business of formulating theories of this kind, and it would require
considerable philosophical violence to reconstrue the explanatory
frameworks of psychology and neuroscience in a formal or semi-formal
way.
In this sense, the model of functional reduction is prima facie
promising. It does not seem to require formal theories, since it is a model
of property reduction, not theory reduction. But is it really a realistic model
of reduction to which most cases of interlevel reduction conform?
In recent years, the functional model has been criticized from several
angles. Ausonio Marras (2002; 2005) has argued that when we analyze the
model carefully and accept certain plausible background assumptions, it in
fact leads back to Nagel reduction, which it was supposed to replace. In the
same vein but with different arguments, Max Kistler (2005) has argued that
functional reduction requires local bridge laws that are left just as
unexplained as in a Nagel reduction. John Bickle (2008; personal
communication) does not criticize the model itself, but points out that it is
based almost entirely on logical and metaphysical considerations, and that
the examples given to support it reflect an elementary school
understanding of science. In this sense, the functional model is a step
backward from intertheoretic models, which were at least based on science
(though not psychology and neuroscience).
95
I will develop the last line of argument in more detail, and show that
from the point of view of philosophy of science and scientific practice, the
functional reduction approach is unacceptable. I will focus on three salient
problems of the model: 1) Where do the functional definitions of the
properties to be reduced come from? (2) What is the “realization” relation
between the property to be reduced and the realizing properties? (3) What
notion of causation does the model require? These are by no means the
only problems or points that need clarification, but they suffice to show
why the model fails as a general account of reduction.
6.3.1. Functionalization
As we have seen, Step 1 in the functional model consists in defining or
redefining the property to be reduced in terms of its causal role. However,
it is not clear how we get the causal definition of the property to be
reduced. Kim seems sympathetic to the view of Chalmers and Jackson
(2001) and Levine (1993), according to which reductive explanation
requires analytic definitions grounded in (a priori) conceptual analysis (see
Kim 2005, Ch. 4). The first step of functional reduction would thus consist
in finding the analytic definition for the property to be reduced through
conceptual analysis.
However, if the functional definitions of the mental properties are to
be based on conceptual analysis that is (at least relatively) a priori, this
leads to a fundamental problem: our a priori ideas about psychological
states or processes are often simply wrong. Consider for example memory.
An armchair conceptual analysis would indicate that memory is some kind
of a simple storage, where our past experiences are waiting for retrieval –
Plato compared memory to an aviary of birds, from which we take the
correct bird when memory retrieval is successful, and the wrong bird when
it is not. However, scientific research has revealed that memories are not
just retrieved, but actively constructed, and subjectively compelling
memories sometimes turn out to be radically inaccurate (Neisser & Harsch
1992). Furthermore, memory comprises several subsystems (short term
memory, long term memory, episodic memory, visual memory, etc.),
96
27
This problem is obviously related to the issue of common-sense (analytical) vs.
empirical functionalism (psychofunctionalism), discussed in section 5.5.
97
possible that in the end we are unable to find any neuroscientific properties
playing the causal role of some psychological properties, and thus we
cannot functionally reduce them. The easiest solution in these cases would
be to revise the functional definitions of the psychological properties, but
this is not always justifiable. We might want to retain some psychological
properties more or less as they are, since they are useful in scientific
explanations. For example, Khalidi (2005) takes up the psychological
property of fear, and shows (based on empirical results in cognitive
neuroscience) that distinctions made at the neurophysiological level cross-
cut the distinctions made at the psychological level. That is, from the
vantage of neurophysiology, there is nothing playing the functional role
associated with the psychological state of fear. Importantly, this is not a
case of multiple realizability, which is a one-to-many relationship. In this
case, there is simply just mismatch: a “one-to-none” relationship.
However, we would not want to eliminate or revise the psychological
concept of fear, since it still plays an important role in research and
scientific explanations.
In this case, it seems that there are no neurophysiological states
playing the causal role of fear, and the option of redefining fear does not
seem very fruitful. Hence, Step 2 in functional reduction of fear fails. But
should we conclude from this that fear is fundamentally irreducible and
threatened by the exclusion argument? Or should we eliminate the property
of fear from our ontology? Both options seem implausible. The framework
of functional reduction seems unsuitable for dealing with situations like
this.
Certainly the basic idea that the properties to be reduced have to
specified causally is correct and in accordance with scientific practice.
However, functionalization is not just a matter of conceptual analysis, it is
not even remotely an a priori matter, and functional definitions can change
as research proceeds. Furthermore, in some cases we might not be able to
find neural realizers that play the functional role definitive of a mental
property. This does not mean that Kim’s functional model is fundamentally
wrong, but it surely is too simplified in this respect.
98
6.3.2. Realization
The second step in Kim’s account of functional reduction is finding the
‘realizers’ of the functionally defined property to be reduced. But what
makes some property a realizer of another property? How should we
understand this realization relation? And what sorts of things are the
realizers of mental properties?
The roots of talk of ‘realization’ in philosophy of mind go back to
multiple realizability (see Chapter 5, section 5.3). In the debate that
followed, very little attention was initially paid to the notion of realization
itself. However, as I have already pointed out, several philosophers (e.g.,
Polger 2004; 2007a; Shapiro 2000; 2004) have recently shown that the
realization relation is much more problematic than has been generally
assumed. One problem is that there seems to be no general realization
relation that applies to all the different cases that have been presented as
paradigmatic cases of realization. However, it might be that Kim’s account
does not need any general notion of realization, and that a more “local”
notion would suffice. In this section I will show that even if we limit the
discussion to psychological properties and their realizers, and accept that
there is no general notion of realization, Kim’s notion of realization leads
to problems.
First of all, it is not entirely clear whether Kim’s notion of realization
is “flat” or “dimensioned”.28 According to the flat view, both the realized
and the realizer properties must be instantiated in the same individual, at
the same level of composition. According to the “dimensioned” view, the
realized property may also be at a higher level of composition than the
realizer properties. Sometimes Kim seems to be explicitly supporting the
flat view: “It is evident that a second-order property and its realizers are
at the same level … they are properties of the very same objects” (Kim
1998, 82). Also one of Kim’s answers to the “generalization argument”
that threatens the exclusion argument relies on flat realization (see Walter
2008 for more).
However, the “flat” or intralevel view of realization is hard to
reconcile with Kim’s (2005) later exposition of the functional model,
28
This distinction was introduced by Gillett (2002; 2003).
99
that laws underlie causal relations, and that theories are sets of laws (both
assumptions are controversial, but commonly accepted in philosophy of
mind), the implication is that Kim’s model comes very close to theory
reduction: in order to reduce a psychological theory, we need to find in (or
derive from) the neuroscientic theory a structure that is isomorphic to the
psychological theory. This is not so different from the New Wave model of
reduction, where a psychological theory is reduced by deriving from
neuroscience an “analogue” or “equipotent image” that is isomorphic to the
psychological theory.
Marras (2002; 2005) makes a similar point with a somewhat
different reasoning: in a closer analysis, Kim’s model turns out to be a
model of intertheoretic reduction. If this is the case, the functional model
only appears to be an advance over the intertheoretic models, and faces
exactly the same problems (see Part I, Chapter 1).
Another fundamental problem with Kim’s notion of realization was
already mentioned at the end of section 6.1: if we accept multiple
realizability, the realized properties have to be either identical to the
disjunction of the realizers, or just concepts (or predicates or designators).
Kim rejects the first option for philosophical reasons and accepts the
second one. However, in the context of realization, the problem with the
second option is that it seems to leave no room for the idea that neural
properties realize mental properties. According to the second option, the
mental concepts simply (non-rigidly) designate different neural properties
in different species, just like in Lewis’ (1972) filler-functionalism. If this is
true, there is no realization relation here. Mental properties cannot be
realized, since there are no mental properties, just mental concepts (or
property designators) that group physical properties in interesting ways.30
Mental concepts cannot be realized, since concepts in general are not the
30
This distinction between real properties and “mere concepts” is problematic in
general. One of the points of the pluralistic approach I am defending in Part III is that
such a distinction does not make sense: a concept that groups together properties in a
scientifically useful way is not “just a concept;” it picks out a real property. Perhaps
one further solution more congenial to Kim’s approach would be to argue that mental
properties are some special kind of “abstract” properties. However, Kim does not
appear to seriously consider such a solution. In any case, it would require developing
or spelling out the metaphysics for such properties, which is no easy task.
101
sorts of things that are realized. But if this is the case, the whole talk of
realization has been misleading, and the claim that the functional model
can accommodate multiple realizability turns out false.31
Perhaps, however, there are yet other ways of understanding
realization. As Polger and Shapiro (2008) have pointed out, one
problematic assumption that underlies many of these issues is the
assumption that the realizers have to be properties. Particularly in more
recent writings, Kim himself has been less strict and allows the realizers to
be mechanisms: “Find the properties (or mechanisms) in the reduction base
that perform the causal task C” (Kim 2005, 102, my emphasis).
If we (unlike Kim) take this idea of mechanistic realization seriously,
it leads to a more complicated picture of mental realization than the one the
functional model presents. The idea is that a functionally (causally) defined
psychological state, property, or capacity is realized by a neural
mechanism that plays that functional role. A crucial aspect of this kind of
mechanistic realization is the multilevel nature of the mechanisms: on any
reasonable understanding of neural mechanisms, they have to be
hierarchically organized into levels. Therefore, instead of a simple two-
level model with the mental property and its neural realizers, we have a
more complicated picture where the realizer is also organized into levels.
An often-cited example of a psychological property or capacity that is
realized by a (multilevel) neural mechanism is memory consolidation (see
Part I, Chapter 2). As Wilson and Craver (2007) point out, this comes close
to how the term ”realization” is used in the cognitive sciences: when
scientists state that they are looking for, say, the neural realization of
memory consolidation, what they typically mean is that they are looking
for the neural mechanism of memory consolidation.
31
In fact, Kim sometimes seems ready to reject the multiple realizability of mental
properties and argues for “species-specific identities,” such that “multiply realized
properties are sundered into diverse realizers in different species and structures” (Kim
1998, 105). This leads to problems if there is also multiple realizability within species
or structures: it seems to follow that mental properties are spliced into properties
restricted to very specific neural or physical structures, and it is hard to see how such
properties could be relevant in scientifically explaining human behavior. See also
section 5.3 for more on multiple realizability.
102
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
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)
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
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
35
I thank Max Seeger for comments and discussions that helped clarify the ideas
presented in this chapter.
110
molecular kinetic energy” are a posteriori identities: they are true in all
possible worlds, but we get to know them a posteriori.
However, as Kripke acknowledged, this is prima facie implausible,
since in these a posteriori identities there seems to be an element of
contingency that is in contrast to their supposed necessity. It does not seem
that heat is identical to average molecular kinetic energy (from now on
MKE) in every possible world. We can conceive of worlds where heat is
something else than MKE. If there are indeed such possible worlds, then
the identity seems to be contingent (not true in every possible world). But
there are no contingent identities: if an identity is true, it is true in all
possible worlds. Therefore, the conclusion would be that heat is not
identical to molecular kinetic energy at all.
However, according to Kripke, in cases like this we can explain
away the appearance of contingency. When we think we are conceiving of
heat not being identical to MKE, what we are conceiving is that someone is
having a sensation of heat, but that this is caused by something else than
molecular kinetic energy. For example, we can conceive of a planet where
there are creatures whose sensations of heat are not caused by MKE but by
something else. But this does not amount to conceiving that heat itself is
not MKE. Heat itself cannot be anything else than MKE.
The gist of Kripke’s argument against the identity theory of mind is
that this strategy does not work when it comes to identities involving
phenomenal properties. Let us consider the standard example “Pain = C-
fiber firing.”36 Again, there is felt contingency: it seems that there could be
pain without C-fiber firing. For example, it seems possible that there are
creatures that experience pain but have a neurological system different
from ours and no C-fibers. However, in this case we cannot explain the
contingency away by saying that we are in fact thinking of something else
than pain itself. We cannot make the distinction between the sensation of
pain and the way it appears to us, because the way it appears to us just is
the sensation of pain. Thus, pain is not necessarily identical to C-fiber
firing, and because there are no contingent identities, pain is not identical
36
This statement is in fact empirically false, but we can ignore this for sake of
simplicity and take the statement as a placeholder for some true phenomenal-physical
identity statement involving rigid designators.
111
37
This comes close to Broad’s (1925) views of reductive explanation (Chapter 2).
According to Broad (in contemporary terminology), a property of a system is
reductively explained by deducing it from the properties of the components of the
system, and in the case of phenomenal properties, this is not possible. See Stephan
112
is different, since the identity does not make it intelligible why pain should
feel the way it does. No matter how much we know about
neurophysiology, the argument goes, the qualitative character of pain is not
explained, and there remains an explanatory gap.
According to Levine, reductive explanations like that of water to
H2O proceed roughly in the following way: First, we give a our
pretheoretic concept of water a conceptual analysis in terms of its causal
role, and then we find out that it is H2O that in fact plays this causal role.
This yields a two-stage picture of reductive explanation:
Stage 1 involves the (relatively? quasi?) a priori process of working the concept
of the property to be reduced “into shape” for reduction by identifying the
causal role for which we are seeking the underlying mechanisms. Stage 2
involves the empirical work of discovering just what those underlying
mechanisms are. (Levine 1993, 132)
The reason why this does not work in the case of phenomenal properties is
that causal role is not all there is to concepts of phenomenal properties:
(2004) for a more detailed analysis and comparison of different approaches to the
reductive explanation of phenomenal properties.
113
38
Seeger (unpublished manuscript) argues that Levine (1993) cannot be claiming that
only properties that can be given exhaustive causal analyses can be reductively
explained, since if this was the case, much of the discussion in Levine’s paper would
be superfluous. On this interpretation, what Levine claims is that what is required for
example in the water case is that the microphysical properties of H2O
“epistemologically necessitate” all the superficial properties or macroproperties (be
they causal or not) of water. I briefly consider this approach to reductive explanation
below.
114
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:
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:
[A] reduction should explain what is reduced, and the way we tell whether this
has been accomplished is to see whether the phenomenon to be reduced is
epistemologically necessitated by the reducing phenomenon, i.e. whether we
can see why, given the facts cited in the reduction, things must be the way they
seem on the surface. (Levine 1993, 129, emphasis mine)
The idea is that, in the case of “Water = H2O,” the microphysical properties
of H2O necessitate or entail the superficial or macroscopic properties of
water. It is inconceivable that water should not behave the way it does,
given the microphysical properties of H2O. For instance, it is inconceivable
39
One appealing way of pumping intuitions regarding the explanatory gap is the
following. Let us assume we have built a robot with humanlike mental capacities. We
understand completely the mechanics of the robot’s “brain.” Still, it seems to be a
perfectly legitimate philosophical question to ask whether the robot has subjective
phenomenal experiences like we do. In this sense, there does seem to be a “gap”
between phenomenal consciousness and the physical domain, although intuitions may
differ. However, I do not want to defend any position regarding the explanatory gap –
the point of this chapter is merely to show certain problems in the arguments for the
gap.
117
that water should not boil at 100 °C at sea level, given the microphysical
properties of H2O. However, the neurophysiological properties arguably do
not necessitate or entail the phenomenal quality of pain. It remains
conceivable that there is no subjective experience at all, given the
neurophysiological properties. This argument does not require that the
property to be reduced be given an exhaustive causal analysis – all that is
required is that microphysical properties necessitate or entail the
macroscopic or “superficial” properties of the phenomenon to be reduced.
Of course, if we further assume that macroscopic properties in general are
exhaustively defined by their causal role (adopting a causal theory of
properties), this argument for the explanatory gap is indeed very close to
the other argument, but as should be clear from the above discussion, I do
not think that all macroscopic properties are exhaustively causal properties.
Chalmers and Jackson (2001) have elaborated in detail on this way
of arguing for the explanatory gap. They argue that reductive explanation
of a phenomenon requires that it be a priori entailed by the facts of physics,
and that if there is no such a priori entailment in the case of phenomenal
properties, this constitutes an explanatory gap. There is an a priori
entailment from P to Q if and only if “PQ” is a priori, that is, when it is
possible to know that “PQ” with justification independent of experience.
In the case of “Water = H2O,” Chalmers and Jackson claim, there is such a
priori entailment, since the microphysical facts about H2O imply the non-
basic facts about water (such that it is transparent, boils at 100°C, etc.), and
this implication is knowable a priori. In this sense, an ideal cognizer who
knows all the relevant microphysical facts about H2O could deduce all the
non-basic facts about water without further a posteriori knowledge.
According to Chalmers and Jackson, any account of reductive explanation
that is weaker than this would not yield the desired metaphysical necessity
of the truth of “Water = H2O” (if the connection between water and H2O is
weaker, one has to deal with the Kripke argument). In cases where there is
no reductive explanation in the sense of a priori entailment, there is an
explanatory gap.
Needlessly to say, this approach to reductive explanation is very
different from the approach defended in this thesis. As I have shown in
Part I, actual scientific explanation (including reductive explanation) has
118
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):
In his latest book, Jaegwon Kim (2005, Ch. 5) has presented (at least) four
distinct objections against the explanatory argument. The first three of
these concern the strategy of inference to the best explanation, both
generally and as it is applied to the problem of phenomenal properties. I
will not discuss them further here, since Bates (2009) and McLaughlin
(2010) have presented quite detailed and convincing replies to these
objections. It might well be that inference to the best explanation is
problematic either generally or in this context, but Kim has not succeeded
in showing this, and it is not my intention to attempt it here.
Instead, my focus is on Kim’s fourth objection, which he apparently
also considers as his main weapon against the explanatory argument. This
objection states that psychophysical identities do not explain
psychophysical correlations, and hence the argument fails.
In fact, this objection consists of many parts that are at least partly
independent from each other. One of Kim’s claims is that the whole point
122
light waves, and why (3) the refractive indices in materials are exactly the same
for light waves and electro-magnetic waves. (McLaughlin 2010, 282)
I grant that this example shows that (contra Kim) identities do play a role
in scientific explanation. However, it does not show that identities are a
way of explaining correlations, understood in McLaughlin’s sense (as the
kind of nomological copresence expressed in the correlation thesis). The
crucial part in Maxwell’s argument was showing that electromagnetic
waves and light waves propagate at the same velocity (speed of light).
After considering the velocity of propagation of “magnetic disturbances”,
Maxwell concluded that “[t]this velocity is so nearly that of light, that it
seems that we have strong reason to conclude that light itself (including
radiant heat, and other radiations if any) is an electromagnetic disturbance
in the form of waves propagated through the electromagnetic field
according to electromagnetic laws” (Maxwell 1865, 466).
Importantly, the fact that light waves are present when and only
when electromagnetic waves are present did not play a big role in making
(or justifying) the “bold hypothesis”. The hypothesis was not presented to
explain this correlation (and such a correlation alone would never have
been enough to justify the bold hypothesis). It was presented to explain the
fact that light waves and electromagnetic waves share a crucial property
(the speed of propagation). The hypothesis was then later confirmed by
empirical results, which showed that light waves and electromagnetic
waves also have other properties in common, such as their refractive
indices (as McLaughlin correctly points out).
Let us consider another example from a field somewhat closer to
philosophy of mind, namely vision research. About 40 years ago, what
came to be known as “luminance units” were discovered somewhat
accidentally while recording extracellularly from the cat retina (Barlow
and Levick 1969). These units were extremely rare (less than 1% of the
retinal ganglion cell population) and responded to light stimuli in an
unusual way: the response was sluggish, relatively straightforwardly
related to the light intensity, and increased monotonically with increasing
light intensity (normally the responses of retinal ganglion cells are much
more complex).
125
(2009) present (Bill Sikes = the burglar, water = H2O, Tully = Cicero, light
waves = electromagnetic waves, etc.). However, what is at issue in the case
of psychoneural correlations and type physicalism is the identity of
properties. The claim is that the correlation of properties is explained by
the fact that the properties are identical. In order to support this claim,
Type Physicalists would be expected to provide examples from science
where the correlation of two properties is explained by these properties
being identical, but so far they have failed to do so.
In fact, I believe such examples will be hard to find, since it is not a
common explanatory strategy in science to explain the copresence of
properties by hypothesizing that they are identical. In contrast, it is often
the case that correlated properties are not identical. Consider for example
the property of having a heart and the property of having a circulatory
system. An organism has a heart if and only if it also has a circulatory
system. However, explaining this correlation with the hypothetical identity
“the property of having a heart = the property of having a circulatory
system” would be deeply mistaken. If type physicalists want to stick to the
claim that their hypothesis is empirical and scientific, the burden of proof
is on them to present evidence from science where the identity of
properties explains the correlation of properties.
Type physicalists have one obvious answer to the concerns raised in
this chapter. They might grant all of the above, but still claim that in some
sense identities do explain correlations. For example, it does seem
intuitively quite plausible that the fact that water is present when and only
when H2O is present is explained by the hypothesis that water is H2O.
Perhaps this could also be translated to talk of properties: the fact that the
property of being water is present when and only when the property of
being H2O is present is explained by the hypothesis that “the property of
being water = the property of being H2O”. However, claims like this are
not convincing as long as they only rely on intuitions or philosophical
models of explanation and are not backed up by actual scientific cases. The
history of the discovery of water is a long story, but it is safe to say that the
fact that water is present when and only when H2O is present played
absolutely no role in the process that led to the hypothesis that water is
H2O. Type physicalists might insist that the fact that water is present when
128
and only when H2O is present is in some sense explained by the fact that
water is H2O, but even if this is the case, such explanations are only
shallow, philosophical, or ad hoc explanations, not real scientific
explanations.
To repeat, the role of identities in science is not to explain
correlations. What leads to the identity claim (initially a hypothetical one)
in cases like “light waves = electromagnetic waves” or “water = H2O” or
“luminance units = melanopsin-containing ganglion cells” is that things on
either side of the identity share important properties. The hypothetical
identities immediately suggest ways of empirically testing themselves. In
the case of phenomenal state C and the correlated neuroscientific state, no
relevant properties are shared. The claim is that the correlation between
physical state P and a phenomenal state C is explained by the identity “P =
C”. This hypothetical identity claim does not indicate any further
hypotheses that could be empirically tested. Therefore, the identity claim
of the type physicalists is of a different kind than identity claims in science,
and the former is not supported by the success and significance of the
latter.
Type Physicalism might still be preferable to other metaphysical
positions for purely philosophical reasons, but the claim that it is on a par
with other scientific hypotheses is not supported by the history of science.
Type physicalists have tried to present the best explanation argument as an
empirically supported argument, while it is in fact pure metaphysics.
Furthermore, the above considerations also provide general reasons
to question the best explanation strategy of arguing for Type Physicalism. I
have argued that the identity thesis explains the correlation thesis, if at all,
only in a very shallow or ad hoc way. How important is it then that the
identity thesis is currently the best explanation for the correlation thesis? If
we have theories that do not explain phenomenon P at all and one theory X
that explains it in a shallow and ad hoc manner, does this provide
justification for holding X true? I believe the answer must be no.
Therefore, the second premise of the explanatory argument (“P2: One
possible explanation for the correlation thesis is type physicalism”) is
either false or too weak to support the argument.
129
PART III:
A New Framework for
Philosophy of Mind
134
135
Introduction
In Parts I and II, I have argued against strong forms of reductionism in
philosophy of mind (most importantly intertheoretic reductionism,
metascientific reductionism, and functional reductionism) and defended
explanatory pluralism and the interventionist approach to causation. But
what does explanatory pluralism exactly consist in? What are the
ontological implications of explanatory pluralism and the interventionist
account of causation? Is pluralism threatened by the dreaded causal
exclusion argument? In this part, I will give answers to these questions. I
will argue that if we adopt explanatory pluralism and the interventionist
approach to causation, our understanding of physicalism has to change, and
this leads to what I call pluralistic physicalism. I will then show that this
pluralistic physicalism is not endangered by the causal exclusion argument,
and that it is compatible with a kind of (weak) reductionism.
The structure of this part is as follows. In Chapter 9, I will give a
more precise characterization of explanatory pluralism, and discuss four
theses that form its core. In Chapter 10, I will argue that, under some very
plausible assumptions, explanatory pluralism leads to pluralistic
physicalism. I will also consider how this kind of physicalism relates to
more traditional forms of physicalism. In Chapter 11, I will discuss the
causal exclusion argument and argue that it does not pose a fundamental
problem for the kind of pluralism I am defending, if causation is
understood in the interventionist sense. In Chapter 12, I will argue that
pluralistic physicalism is in fact compatible with many reductionist ideas,
and discuss some largely neglected ways of understanding the reductionist
aspects of scientific explanation. Finally, I conclude this part and the whole
book with a summary of the results and an outlook on future work.
136
137
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
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
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
about the experiences the system undergoes, from the perspective of the system. If
phenomenal properties are real, they are real in a very different sense than other
objects of scientific inquiry, and we certainly should not expect same models of
reductive explanation to apply to them, or draw any conclusions from the impossibility
of such reductive explanation. However, this would be a topic for a treatise of its own.
44
This dictum is named after Samuel Alexander (1920), who wrote that
epiphenomenalism “supposes something to exist in nature which has nothing to do, no
purpose to serve, a species of noblesse which depends on the work of its inferiors, but
is kept for show and might as well, as undoubtedly would in time, be abolished”
(1920, vol. 2., 8).
147
Recently Ladyman and Ross (2007) have also argued for ontological
pluralism, but in somewhat different terms. First of all, they show that
prominent philosophers of mind have misunderstood or neglected
contemporary physics. The idea (implicit also in Kim’s work) that
everything comes down to “microbangings” amongst elementary things at
the fundamental level makes absolutely no sense from the point of view of
current physics. There is no causation in fundamental physics, no
“fundamental level,” and not even elementary things in the sense of self-
subsistent individuals. Therefore the thesis that everything that is real is
ultimately composed of fundamental level microphysical things (e.g., Pettit
1993) is simply false. For similar reasons, Ladyman and Ross also reject
general (mereological) reductionism, type identity theory and traditional
forms of physicalism. Instead, they defend a form of ontological structural
realism conjoined with the idea of real patterns (Ladyman & Ross 2007,
Ch. 2 and 4).
The talk of real patterns goes back to Dennett’s (1991) well-known
paper. The idea that Dennett, Ladyman, and Ross defend is that to be real
is to be a real pattern. However, what real patterns exactly are is a rather
complicated matter. Dennett (1991, 34) gives only a rough and weak
definition: there is a real pattern in some data if there is a description of the
data that is more efficient than a verbatim bit map description. Ladyman
and Ross provide a very technical definition based on information theory,
and going through it here would lead too far astray. In any case, the basic
idea is similar to that of Dennett’s: a pattern in a structure of events S is a
real pattern if describing S as a bit map would be information-theoretically
less efficient than describing it in terms of the pattern, and there are aspects
of S that cannot be tracked without the pattern (Ladyman & Ross 2007,
233). Real patterns are not real just relative to human capabilities and
limitations: any computational system that is efficient will make use them,
and real patterns are real even if there is no one to observe them.
Since there are real patterns all over the place, also in the domains of
all the special sciences, pattern-realism leads to Rainforest Realism: “Ours
is thus a realism of lush and leafy spaces rather than deserts, with science
regularly revealing new thickets of canopy” (Ladyman & Ross 2007, 234).
It is interesting that Ladyman and Ross end up using the same rainforest
148
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
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
48
This was originally pointed out to me by Dan Brooks, for which I am very grateful.
155
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
51
The idea that pluralism and reductionism are compatible is not entirely new. One
author that has explicitly argued for it is Daniel Steel (2004), and Wimsatt (2007) is
also a pluralist who emphasizes the importance of reductive explanation.
162
and so on. These are different kinds of explanations, and thus non-
competing. This idea is reflected in the thesis (1) of my definition of
explanatory pluralism, which states that for fully understanding the human
mind we need explanations of different kinds (see Chapter 9). The MEM
thesis is not in conflict with this, since MEM as such does not rule out non-
mechanistic explanations for mental phenomena.52
Another reductionist thesis that can be accommodated to pluralism is
related to the special nature of fundamental physics. The history of science
shows that there appears to be the following fundamental asymmetry:
higher-level explanations are constrained by physical explanations, but the
converse does not hold. This is captured in the Primacy of Physics
Constraint (PPC) already presented in Chapter 10: “Special science
hypotheses that conflict with fundamental physics, or such consensus as
there is in fundamental physics, should be rejected for that reason alone.
Fundamental physical hypotheses are not symmetrically hostage to the
conclusions of the special sciences” (Ladyman & Ross (2007, 43). That is,
physics sets constraints for the theories of special sciences. As Ladyman
and Ross (2007, 210) suggest, this asymmetry can perhaps be extended to
apply more broadly: for instance, computer science asymmetrically
constrains psychology in the sense that what is computationally impossible
is also impossible for human brains, but not vice versa.53
In Chapter 10, I defended robustness as a criterion for considering
something real. Importantly, robustness is a notion that comes in degrees:
some things are more robust than others. For instance, photons are
obviously more robust than cognitive dissonance (i.e., the condition of
conflict or anxiety that arises from holding two conflicting ideas or beliefs
52
I have not argued that the MEM thesis is true, and I do not know whether it is – the
point is to show that it amounts to a form of reductionism that is compatible with the
pluralism I have defended.
53
Some authors (e.g., Steel 2004) have also argued for corrective asymmetry. The idea
is that resources from the lower levels are often necessary to explain exceptions in
generalizations at the higher levels, but not vice versa. For example, exceptions to
Mendelian laws of inheritance are explained by resources from molecular genetics,
and not by any higher-level or same-level sciences. However, it is not clear whether
this asymmetry holds generally – sometimes exceptions in generalizations are
explained also by resources from the same level or higher levels (see Raerinne 2011).
163
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)
To take a clear example from physics, the van der Waal’s force law ([P +
a/V2][V - b] = RT) is invariant under a wider range of interventions than the
ideal gas law (pV = nRT). This is because it includes the additional
variables a (the attraction between the particles) and b (the average volume
excluded by a particle). In this sense, the van der Waal’s force law can
answer more what-if-things-had-been-different questions, and provides
deeper explanations.
However, it is plausible that there is not just one dimension of
explanatory depth, but several, as has been recently argued by Petri
Ylikoski and Jaakko Kuorikoski (Ylikoski & Kuorikoski 2010). An
explanation A of a phenomenon P can be better than explanation B in some
ways, but worse in some other ways. The main point of the authors is that
there are several distinct dimensions of explanatory power, and that they
do not go hand-in-hand: in fact some of them are systemically in conflict.
Ylikoski and Kuorikoski define the following five dimensions of
explanatory power: non-sensitivity (with respect to background
conditions), precision, factual accuracy, degree of integration (to a larger
body of knowledge), and cognitive salience (i.e., how easily the
explanation can be grasped). The account is very provisional, and it is not
clear why exactly these five dimensions are selected, but this is not crucial
here: the important point is that explanatory power comes in degrees, and
that it has several dimensions.
This has significant consequences for understanding reductionism.
One widely shared intuition that makes reductionism appealing is that
lower-level generalizations or explanations are more fundamental than
higher-level ones. I believe this intuition is not entirely mistaken: it is
plausible that lower-level explanations are better than higher-level
165
54
Related to this, Wimsatt (2007, Ch. 5 and 12) has discussed at length “reductive
biases” and the dangers of being too eager to apply reductive heuristics, especially in
situations where they are not warranted.
55
McCauley (2009) uses scope as one criterion to define higher vs. lower levels. See
Part I, Chapter 4 for more.
166
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
I have now come to the end of the third and final part of this thesis. In this
part, I have defended explanatory pluralism and pluralistic physicalism as a
framework for philosophy of mind, and pointed out ways in which
reductionistic ideas and intuitions can be accommodated to this framework.
Now it is time to take stock of the results of this thesis in general, to
summarize the consequences for philosophy of mind, and to consider
topics that need to be addressed by further research.
First of all, from an epistemological point of view, it is clear that the
prospects of strongly reductive or eliminative programs do not look very
good. Both the history of science and considerations based on the
philosophy of causation and scientific explanation show that explanations
of different kinds and at different levels will be needed for fully
understanding human behavior and the mind. This is not just due to the
incompleteness of our current scientific knowledge or our cognitive
limitations: it is due to the overwhelming complexity of the world we live
in. The strongest form of explanatory reduction warranted by scientific
practice is downward-looking mechanistic explanation. Its claim to the title
“reduction” can be disputed, but it does satisfy some pretheoretic intuitions
about reduction, and is close to what many scientists mean when they talk
about reduction.
From an ontological point of view, matters are more complicated. I
am not very sympathetic to the type identity theory, but I have not
explicitly argued against it. Instead, I have argued that the main argument
for the identity theory, the causal exclusion argument, loses its force when
causation is understood in interventionist terms. For the same reason, a
large part of the motivation for being a traditional physicalist dissolves.
Therefore, I have suggested replacing traditional physicalism with what I
call pluralistic physicalism: properties of different kinds and at different
levels are real, but fundamental physics still has a special status among the
sciences, as it asymmetrically constrains the theories and explanations of
other sciences. Regarding the mind-body problem, this position implies
that psychological properties (as discovered by empirical psychology) are
170
real and not in any strong sense reducible to neural properties. But on the
other hand, psychological properties are not strongly emergent or
completely independent from neural properties: they are integrated into
neural mechanisms, and can be seen as their higher-level properties.
This pluralistic framework leaves room for many approaches that
have been traditionally associated with reductionism. The idea that
explanations of lower levels are more powerful than explanations of higher
levels can be accepted, as long as we understand that explanatory power is
a multidimensional thing, and that lower level generalizations fare better
than higher level generalizations in some (but not all) of these dimensions.
Furthermore, the thesis that all mental phenomena can be mechanistically
explained is more than enough for many reductionists, and it is compatible
with pluralistic physicalism.
There are obviously several unresolved issues that need to be
addressed by future research. First of all, I have critically analyzed several
problems in philosophy of mind in light of the results of this thesis, but
there are many more potential targets that I have left untouched. These
include intentionality and representation, the extended mind, self-
consciousness and self-models, emotions, embodiment, enactivism,
perception, free will, and many more.
One thing that this thesis has hopefully made clear is that there is a
huge gap between formal approaches to reduction (e.g., New Wave
reductionism, functional reduction) and non-formal approaches that are
closer to scientific practice (e.g., mechanistic explanation). What is still
unclear is what exactly are the fields of application of non-formal and
formal approaches, and what are the theoretical and practical limitations of
each. For instance, what exactly are the domains where mechanistic
explanations are the norm, and what kinds of phenomena generally resist
mechanistic explanations?
Another cluster of important open questions concerns the idea of
levels, a topic that has received comparatively little philosophical attention.
Does it make sense to talk of general levels of organization or are “levels
of mechanisms” sufficient? What are the criteria for identifying levels and
171
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