A Computational Foundation for the Study of Cognition*1

David J. Chalmers
Australian National University
New York University
chalmers@anu.edu.au

Computation is central to the foundations of modern cognitive science, but

its role is controversial. Questions about computation abound: What is it for
a physical system to implement a computation? Is computation sufficient for
thought? What is the role of computation in a theory of cognition? What is the
relation between different sorts of computational theory, such as connectionism
and symbolic computation? In this paper I develop a systematic framework
that addresses all of these questions.
Justifying the role of computation requires analysis of implementation, the
nexus between abstract computations and concrete physical systems. I give
such an analysis, based on the idea that a system implements a computation if
the causal structure of the system mirrors the formal structure of the computa-
tion. This account can be used to justify the central commitments of artificial
intelligence and computational cognitive science: the thesis of computational
sufficiency, which holds that the right kind of computational structure suf-
fices for the possession of a mind, and the thesis of computational explanation,
which holds that computation provides a general framework for the explana-
tion of cognitive processes. The theses are consequences of the facts that (a)
computation can specify general patterns of causal organization, and (b) men-
tality is an organizational invariant, rooted in such patterns. Along the way I
answer various challenges to the computationalist position, such as those put

* This paper was written in 1993 but never published (although section 2 was
included in “On Implementing a Computation”, published in Minds and Machines
in 1994). Because the paper has been widely cited over the years, I have not made
any changes to it apart from adding one footnote, instead saving any further
thoughts for my reply to commentators. In any case I am still largely sympathetic
with the views expressed here, in broad outline if not in every detail.

Journal of Cognitive Science 12: 325-359, 2011

©2011 Institute for Cognitive Science, Seoul National University
326

forward by Searle. I close by advocating a kind of minimal computationalism,

compatible with a very wide variety of empirical approaches to the mind. This
allows computation to serve as a true foundation for cognitive science.

Key words: computation; cognition; implementation; explanation; connec-

tionism; computationalism; representation; artificial intelligence

1. Introduction

Perhaps no concept is more central to the foundations of modern cognitive

science than that of computation. The ambitions of artificial intelligence
rest on a computational framework, and in other areas of cognitive science,
models of cognitive processes are most frequently cast in computational
terms. The foundational role of computation can be expressed in two basic
theses. First, underlying the belief in the possibility of artificial intelligence
there is a thesis of computational suf ficiency, stating that the right kind of
computational structure suffices for the possession of a mind, and for the
possession of a wide variety of mental properties. Second, facilitating the
progress of cognitive science more generally there is a thesis of computa-
tional explanation, stating that computation provides a general framework
for the explanation of cognitive processes and of behavior.
These theses are widely held within cognitive science, but they are quite
controversial. Some have questioned the thesis of computational sufficiency,
arguing that certain human abilities could never be duplicated computa-
tionally (Dreyfus 1974; Penrose 1989), or that even if a computation could
duplicate human abilities, instantiating the relevant computation would not
suffice for the possession of a mind (Searle 1980). Others have questioned
the thesis of computational explanation, arguing that computation provides
an inappropriate framework for the explanation of cognitive processes
(Edelman 1989; Gibson 1979), or even that computational descriptions of a
system are vacuous (Searle 1990, 1991).
Advocates of computational cognitive science have done their best to
repel these negative critiques, but the positive justification for the founda-
tional theses remains murky at best. Why should computation, rather than
some other technical notion, play this foundational role? And why should
 327

there be the intimate link between computation and cognition that the the-
ses suppose? In this paper, I will develop a framework that can answer these
questions and justify the two foundational theses.
In order for the foundation to be stable, the notion of computation itself
has to be clarified. The mathematical theory of computation in the abstract
is well-understood, but cognitive science and artificial intelligence ulti-
mately deal with physical systems. A bridge between these systems and the
abstract theory of computation is required. Specifically, we need a theory of
implementation: the relation that holds between an abstract computational
object (a “computation” for short) and a physical system, such that we can
say that in some sense the system “realizes” the computation, and that the
computation “describes” the system. We cannot justify the foundational
role of computation without first answering the question: What are the con-
ditions under which a physical system implements a given computation?
Searle (1990) has argued that there is no objective answer to this question,
and that any given system can be seen to implement any computation if
interpreted appropriately. He argues, for instance, that his wall can be seen
to implement the Wordstar program. I will argue that there is no reason
for such pessimism, and that objective conditions can be straightforwardly
spelled out.
Once a theory of implementation has been provided, we can use it to
answer the second key question: What is the relationship between compu-
tation and cognition? The answer to this question lies in the fact that the
properties of a physical cognitive system that are relevant to its implement-
ing certain computations, as given in the answer to the first question, are
precisely those properties in virtue of which (a) the system possesses mental
properties and (b) the system’s cognitive processes can be explained.
The computational framework developed to answer the first question
can therefore be used to justify the theses of computational sufficiency and
computational explanation. In addition, I will use this framework to answer
various challenges to the centrality of computation, and to clarify some dif-
ficult questions about computation and its role in cognitive science. In this
way, we can see that the foundations of artificial intelligence and computa-
tional cognitive science are solid.
328

2. A Theory of Implementation

The short answer to question (1) is straightforward. It goes as follows:

A physical system implements a given computation when the causal

structure of the physical system mirrors the formal structure of the
computation.

In a little more detail, this comes to:

A physical system implements a given computation when there exists a

grouping of physical states of the system into state-types and a one-to-
one mapping from formal states of the computation to physical state-
types, such that formal states related by an abstract state-transition rela-
tion are mapped onto physical state-types related by a corresponding
causal state-transition relation.

This is still a little vague. To spell it out fully, we must specify the class of
computations in question. Computations are generally specified relative to
some formalism, and there is a wide variety of formalisms: these include
Turing machines, Pascal programs, cellular automata, and neural networks,
among others. The story about implementation is similar for each of these;
only the details differ. All of these can be subsumed under the class of com-
binatorial-state automata (CSAs), which I will outline shortly, but for the
purposes of illustration I will first deal with the special case of simple finite-
state automata (FSAs).
An FSA is specified by giving a set of input states I1, ..., Ik, a set of inter-
nal states S1,...,Sm, and a set of output states O1,...,On, along with a set of
state-transition relations of the form (S, I) → (S’, O’), for each pair (S, I)
of internal states and input states, where S’ and O’ are an internal state and
an output state respectively. S and I can be thought of as the “old” internal
state and the input at a given time; S’ is the “new” internal state, and O’ is
the output produced at that time. (There are some variations in the ways this
can be spelled out — e.g. one need not include outputs at each time step,
 329

and it is common to designate some internal state as a “final” state — but

these variations are unimportant for our purposes.) The conditions for the
implementation of an FSA are the following:
A physical system P implements an FSA M if there is a mapping f that
maps internal states of P to internal states of M, inputs to P to input states
of M, and outputs of P to output states of M, such that: for every state-
transition relation (S, I ) → (S’, O’) of M, the following conditional holds:
if P is in internal state s and receiving input i where f(s)=S and f(i )=I, this
reliably causes it to enter internal state s’ and produce output o’ such that
f(s’)=S’ and f(o’)=O’.1
This definition uses maximally specific physical states s rather than the
grouped state-types referred to above. The state-types can be recovered,
however: each corresponds to a set {s | f(s)=Si}, for each Si ∈ M. From here
we can see that the definitions are equivalent. The causal relations between
physical state-types will precisely mirror the abstract relations between for-
mal states.
There is a lot of room to play with the details of this definition. For
instance, it is generally useful to put restrictions on the way that inputs and
outputs to the system map onto inputs and outputs of the FSA. We also
need not map all possible internal states of P, if some are not reachable
from certain initial states. These matters are unimportant here, however.
What is important is the overall form of the definition: in particular, the
way it ensures that the formal state-transitional structure of the computation
mirrors the causal state-transitional structure of the physical system. This
is what all definitions of implementation, in any computational formalism,
will have in common.

1
I take it that something like this is the “standard” definition of implementation
of a finite-state automaton; see, for example, the definition of the description of a
system by a probabilistic automaton in Putnam (1967). It is surprising, however,
how little space has been devoted to accounts of implementation in the literature
in theoretical computer science, philosophy of psychology, and cognitive science,
considering how central the notion of computation is to these fields. It is remarkable
that there could be a controversy about what it takes for a physical system to
implement a computation (e.g. Searle 1990, 1991) at this late date.
330

2.1 Combinatorial-state automata

Simple finite-state automata are unsatisfactory for many purposes, due to
the monadic nature of their states. The states in most computational formal-
isms have a combinatorial structure: a cell pattern in a cellular automaton,
a combination of tape-state and head-state in a Turing machine, variables
and registers in a Pascal program, and so on. All this can be accommodated
within the framework of combinatorial-state automata (CSAs), which differ
from FSAs only in that an internal state is specified not by a monadic label S,
but by a vector [S1, S2, S3, ...]. The elements of this vector can be thought of
as the components of the overall state, such as the cells in a cellular automa-
ton or the tape-squares in a Turing machine. There are a finite number of
possible values S ji for each element Si, where S ji is the jth possible value for
the ith element. These values can be thought of as “substates.” Inputs and
outputs can have a similar sort of complex structure: an input vector is [I1, ...,
Ik], and so on. State-transition rules are determined by specifying, for each
element of the state-vector, a function by which its new state depends on
the old overall state-vector and input-vector, and the same for each element
of the output-vector.
Input and output vectors are always finite, but the internal state vectors
can be either finite or infinite. The finite case is simpler, and is all that is
required for any practical purposes. Even if we are dealing with Turing
machines, a Turing machine with a tape limited to 10200 squares will cer-
tainly be all that is required for simulation or emulation within cognitive
science and AI. The infinite case can be spelled out in an analogous fashion,
however. The main complication is that restrictions have to be placed on
the vectors and dependency rules, so that these do not encode an infinite
amount of information. This is not too difficult, but I will not go into details
here.
The conditions under which a physical system implements a CSA are
analogous to those for an FSA. The main difference is that internal states of
the system need to be specified as vectors, where each element of the vec-
tor corresponds to an independent element of the physical system. A natural
requirement for such a “vectorization” is that each element correspond to
a distinct physical region within the system, although there may be other
 331

alternatives. The same goes for the complex structure of inputs and outputs.
The system implements a given CSA if there exists such a vectorization of
states of the system, and a mapping from elements of those vectors onto
corresponding elements of the vectors of the CSA, such that the state-tran-
sition relations are isomorphic in the obvious way. The details can be filled
in straightforwardly, as follows:

A physical system P implements a CSA M if there is a vectorization of

internal states of P into components [s1, s2, ...], and a mapping f from
the substates s j into corresponding substates S j of M, along with similar
vectorizations and mappings for inputs and outputs, such that for every
state-transition rule ([I1, ..., Ik], [S1, S2, ...]) → ([S’1, S’2, ...], [O1, ..., Ol])
of M: if P is in internal state [s1, s2, ...] and receiving input [i1, ..., in]
which map to formal state and input [S1, S2, ...] and [I1, ..., Ik] respec-
tively, this reliably causes it to enter an internal state and produce an
output that map to [S’1, S’2, ...] and [O1, ..., Ol] respectively.

Once again, further constraints might be added to this definition for various
purposes, and there is much that can be said to flesh out the definition’s var-
ious parts; a detailed discussion of these technicalities must await another
forum (see Chalmers 1996a for a start). This definition is not the last word
in a theory of implementation, but it captures the theory’s basic form.
One might think that CSAs are not much of an advance on FSAs. Finite
CSAs, at least, are no more computationally powerful than FSAs; there is a
natural correspondence that associates every finite CSA with an FSA with
the same input/output behavior. Of course infinite CSAs (such as Turing
machines) are more powerful, but even leaving that reason aside, there are
a number of reasons why CSAs are a more suitable formalism for our pur-
poses than FSAs.
First, the implementation conditions on a CSA are much more constrained
than those of the corresponding FSA. An implementation of a CSA is
required to consist in a complex causal interaction among a number of
separate parts; a CSA description can therefore capture the causal organiza-
tion of a system to a much finer grain. Second, the structure in CSA states
can be of great explanatory utility. A description of a physical system as a
332

CSA will often be much more illuminating than a description as the corre-
sponding FSA.2 Third, CSAs reflect in a much more direct way the formal
organization of such familiar computational objects as Turing machines,
cellular automata, and the like. Finally, the CSA framework allows a uni-
fied account of the implementation conditions for both finite and infinite
machines.
This definition can straightforwardly be applied to yield implementa-
tion conditions for more specific computational formalisms. To develop an
account of the implementation-conditions for a Turing machine, say, we
need only redescribe the Turing machine as a CSA. The overall state of a
Turing machine can be seen as a giant vector, consisting of (a) the internal
state of the head, and (b) the state of each square of the tape, where this
state in turn is an ordered pair of a symbol and a flag indicating whether
the square is occupied by the head (of course only one square can be so
occupied; this will be ensured by restrictions on initial state and on state-
transition rules). The state-transition rules between vectors can be derived
naturally from the quintuples specifying the behavior of the machine-head.
As usually understood, Turing machines only take inputs at a single time-
step (the start), and do not produce any output separate from the contents of
the tape. These restrictions can be overridden in natural ways, for example
by adding separate input and output tapes, but even with inputs and outputs
limited in this way there is a natural description as a CSA. Given this trans-
lation from the Turing machine formalism to the CSA formalism, we can
say that a given Turing machine is implemented whenever the correspond-
ing CSA is implemented.
A similar story holds for computations in other formalisms. Some for-
malisms, such as cellular automata, are even more straightforward. Others,
such as Pascal programs, are more complex, but the overall principles are
the same. In each case there is some room for maneuver, and perhaps some
arbitrary decisions to make (does writing a symbol and moving the head
count as two state-transitions or one?) but little rests on the decisions we
make. We can also give accounts of implementation for nondeterministic
and probabilistic automata, by making simple changes in the definition of

2
See Pylyshyn 1984, p. 71, for a related point.
 333

a CSA and the corresponding account of implementation. The theory of

implementation for combinatorial-state automata provides a basis for the
theory of implementation in general.

2.2 Questions answered

The above account may look complex, but the essential idea is very simple:
the relation between an implemented computation and an implementing
system is one of isomorphism between the formal structure of the former
and the causal structure of the latter. In this way, we can see that as far as
the theory of implementation is concerned, a computation is simply an
abstract specification of causal organization. This is important for later
purposes. In the meantime, we can now answer various questions and
objections.
Does every system implement some computation? Yes. For example,
every physical system will implement the simple FSA with a single internal
state; most physical systems will implement the 2-state cyclic FSA, and so
on. This is no problem, and certainly does not render the account vacuous.
That would only be the case if every system implemented every computa-
tion, and that is not the case.
Does every system implement any given computation? No. The condi-
tions for implementing a given complex computation — say, a CSA whose
state-vectors have 1000 elements, with 10 possibilities for each element and
complex state-transition relations — will generally be sufficiently rigorous
that extremely few physical systems will meet them. What is required is not
just a mapping from states of the system onto states of the CSA, as Searle
(1990) effectively suggests. The added requirement that the mapped states
must satisfy reliable state-transition rules is what does all the work. In this
case, there will effectively be at least 101000 constraints on state-transitions
(one for each possible state-vector, and more if there are multiple possible
inputs). Each constraint will specify one out of at least 101000 possible con-
sequents (one for each possible resultant state-vector, and more if there are
outputs). The chance that an arbitrary set of states will satisfy these con-
1000
straints is something less than one in (101000)10 (actually significantly less,
because of the requirement that transitions be reliable). There is no reason
to suppose that the causal structure of an arbitrary system (such as Searle’s
334

wall) will satisfy these constraints. It is true that while we lack knowledge
of the fundamental constituents of matter, it is impossible to prove that
arbitrary objects do not implement every computation (perhaps every pro-
ton has an infinitely rich internal structure), but anybody who denies this
conclusion will need to come up with a remarkably strong argument.
Can a given system implement more than one computation? Yes. Any
system implementing some complex computation will simultaneously be
implementing many simpler computations — not just 1-state and 2-state
FSAs, but computations of some complexity. This is no flaw in the current
account; it is precisely what we should expect. The system on my desk is
currently implementing all kinds of computations, from EMACS to a clock
program, and various sub-computations of these. In general, there is no
canonical mapping from a physical object to “the” computation it is per-
forming. We might say that within every physical system, there are numer-
ous computational systems. To this very limited extent, the notion of imple-
mentation is “interest-relative.” Once again, however, there is no threat of
vacuity. The question of whether a given system implements a given com-
putation is still entirely objective. What counts is that a given system does
not implement every computation, or to put the point differently, that most
given computations are only implemented by a very limited class of physi-
cal systems. This is what is required for a substantial foundation for AI and
cognitive science, and it is what the account I have given provides.
If even digestion is a computation, isn’t this vacuous? This objection
expresses the feeling that if every process, including such things as diges-
tion and oxidation, implements some computation, then there seems to be
nothing special about cognition any more, as computation is so pervasive.
This objection rests on a misunderstanding. It is true that any given instance
of digestion will implement some computation, as any physical system
does, but the system’s implementing this computation is in general irrel-
evant to its being an instance of digestion. To see this, we can note that the
same computation could have been implemented by various other physi-
cal systems (such as my SPARC) without it’s being an instance of diges-
tion. Therefore the fact that the system implements the computation is not
responsible for the existence of digestion in the system.
With cognition, by contrast, the claim is that it is in virtue of implement-
 335

ing some computation that a system is cognitive. That is, there is a certain
class of computations such that any system implementing that computation
is cognitive. We might go further and argue that every cognitive system
implements some computation such that any implementation of the com-
putation would also be cognitive, and would share numerous specific men-
tal properties with the original system. These claims are controversial, of
course, and I will be arguing for them in the next section. But note that it is
precisely this relation between computation and cognition that gives bite to
the computational analysis of cognition. If this relation or something like it
did not hold, the computational status of cognition would be analogous to
that of digestion.
What about Putnam’s argument? Putnam (1988) has suggested that on
a definition like this, almost any physical system can be seen to implement
every finite-state automaton. He argues for this conclusion by demonstrat-
ing that there will almost always be a mapping from physical states of a
system to internal states of an FSA, such that over a given time-period
(from 12:00 to 12:10 today, say) the transitions between states are just as the
machine table say they should be. If the machine table requires that state A
be followed by state B, then every instance of state A is followed by state
B in this time period. Such a mapping will be possible for an inputless FSA
under the assumption that physical states do not repeat. We simply map the
initial physical state of the system onto an initial formal state of the compu-
tation, and map successive states of the system onto successive states of the
computation.
However, to suppose that this system implements the FSA in question is
to misconstrue the state-transition conditionals in the definition of imple-
mentation. What is required is not simply that state A be followed by state
B on all instances in which it happens to come up in a given time-period.
There must be a reliable, counterfactual-supporting connection between the
states. Given a formal state-transition A → B, it must be the case that if the
system were to be in state A, it would transit to state B. Further, such a con-
ditional must be satisfied for every transition in the machine table, not just
for those whose antecedent states happen to come up in a given time period.
It is easy to see that Putnam’s system does not satisfy this much stronger
requirement. In effect, Putnam has required only that certain weak material
336

conditionals be satisfied, rather than conditionals with modal force. For this
reason, his purported implementations are not implementations at all.
(Two notes. First, Putnam responds briefly to the charge that his system
fails to support counterfactuals, but considers a different class of counter-
factuals — those of the form “if the system had not been in state A, it would
not have transited to state B.” It is not these counterfactuals that are relevant
here. Second, it turns out that Putnam’s argument for the widespread real-
ization of inputless FSAs can be patched up in a certain way; this just goes
to show that inputless FSAs are an inappropriate formalism for cognitive
science, due to their complete lack of combinatorial structure. Putnam gives
a related argument for the widespread realization of FSAs with input and
output, but this argument is strongly vulnerable to an objection like the one
above, and cannot be patched up in an analogous way. CSAs are even less
vulnerable to this sort of argument. I discuss all this at much greater length
in Chalmers 1996a.)
What about semantics? It will be noted that nothing in my account of
computation and implementation invokes any semantic considerations,
such as the representational content of internal states. This is precisely as
it should be: computations are specified syntactically, not semantically.
Although it may very well be the case that any implementations of a given
computation share some kind of semantic content, this should be a conse-
quence of an account of computation and implementation, rather than built
into the definition. If we build semantic considerations into the conditions
for implementation, any role that computation can play in providing a
foundation for AI and cognitive science will be endangered, as the notion of
semantic content is so ill-understood that it desperately needs a foundation
itself.
The original account of Turing machines by Turing (1936) certainly had
no semantic constraints built in. A Turing machine is defined purely in
terms of the mechanisms involved, that is, in terms of syntactic patterns and
the way they are transformed. To implement a Turing machine, we need
only ensure that this formal structure is reflected in the causal structure of
the implementation. Some Turing machines will certainly support a system-
atic semantic interpretation, in which case their implementations will also,
but this plays no part in the definition of what it is to be or to implement
 337

a Turing machine. This is made particularly clear if we note that there are
some Turing machines, such as machines defined by random sets of state-
transition quintuples, that support no non-trivial semantic interpretation.
We need an account of what it is to implement these machines, and such an
account will then generalize to machines that support a semantic interpreta-
tion. Certainly, when computer designers ensure that their machines imple-
ment the programs that they are supposed to, they do this by ensuring that
the mechanisms have the right causal organization; they are not concerned
with semantic content. In the words of Haugeland (1985), if you take care
of the syntax, the semantics will take care of itself.
I have said that the notion of computation should not be dependent on
that of semantic content; neither do I think that the latter notion should
be dependent on the former. Rather, both computation and content should
be dependent on the common notion of causation. We have seen the first
dependence in the account of computation above. The notion of content has
also been frequently analyzed in terms of causation (see e.g. Dretske 1981
and Fodor 1987). This common pillar in the analyses of both computation
and content allows that the two notions will not sway independently, while
at the same time ensuring that neither is dependent on the other for its anal-
ysis.
What about computers? Although Searle (1990) talks about what it
takes for something to be a “digital computer,” I have talked only about
computations and eschewed reference to computers. This is deliberate, as it
seems to me that computation is the more fundamental notion, and certainly
the one that is important for AI and cognitive science. AI and cognitive sci-
ence certainly do not require that cognitive systems be computers, unless
we stipulate that all it takes to be a computer is to implement some compu-
tation, in which case the definition is vacuous.
What does it take for something to be a computer? Presumably, a com-
puter cannot merely implement a single computation. It must be capable of
implementing many computations - that is, it must be programmable. In the
extreme case, a computer will be universal, capable of being programmed
to compute any recursively enumerable function. Perhaps universality
is not required of a computer, but programmability certainly is. To bring
computers within the scope of the theory of implementation above, we
338

could require that a computer be a CSA with certain parameters, such that
depending on how these parameters are set, a number of different CSAs can
be implemented. A universal Turing machine could be seen in this light, for
instance, where the parameters correspond to the “program” symbols on the
tape. In any case, such a theory of computers is not required for the study
of cognition.
Is the brain a computer in this sense? Arguably. For a start, the brain can
be “programmed” to implement various computations by the laborious
means of conscious serial rule-following; but this is a fairly incidental abil-
ity. On a different level, it might be argued that learning provides a certain
kind of programmability and parameter-setting, but this is a sufficiently
indirect kind of parameter-setting that it might be argued that it does not
qualify. In any case, the question is quite unimportant for our purposes.
What counts is that the brain implements various complex computations,
not that it is a computer.

3. Computation and cognition

The above is only half the story. We now need to exploit the above account
of computation and implementation to outline the relation between compu-
tation and cognition, and to justify the foundational role of computation in
AI and cognitive science.
Justification of the thesis of computational sufficiency has usually been
tenuous. Perhaps the most common move has been an appeal to the Turing
test, noting that every implementation of a given computation will have a
certain kind of behavior, and claiming that the right kind of behavior is suf-
ficient for mentality. The Turing test is a weak foundation, however, and
one to which AI need not appeal. It may be that any behavioral descrip-
tion can be implemented by systems lacking mentality altogether (such as
the giant lookup tables of Block 1981). Even if behavior suffices for mind,
the demise of logical behaviorism has made it very implausible that it suf-
fices for specific mental properties: two mentally distinct systems can have
the same behavioral dispositions. A computational basis for cognition will
require a tighter link than this, then.
Instead, the central property of computation on which I will focus is one
 339

that we have already noted: the fact that a computation provides an abstract
specification of the causal organization of a system. Causal organization is
the nexus between computation and cognition. If cognitive systems have
their mental properties in virtue of their causal organization, and if that
causal organization can be specified computationally, then the thesis of
computational sufficiency is established. Similarly, if it is the causal organi-
zation of a system that is primarily relevant in the explanation of behavior,
then the thesis of computational explanation will be established. By the
account above, we will always be able to provide a computational specifica-
tion of the relevant causal organization, and therefore of the properties on
which cognition rests.

3.1 Organizational invariance

To spell out this story in more detail, I will introduce the notion of the
causal topology of a system. The causal topology represents the abstract
causal organization of the system: that is, the pattern of interaction among
parts of the system, abstracted away from the make-up of individual parts
and from the way the causal connections are implemented. Causal topology
can be thought of as a dynamic topology analogous to the static topology
of a graph or a network. Any system will have causal topology at a number
of different levels. For the cognitive systems with which we will be con-
cerned, the relevant level of causal topology will be a level fine enough to
determine the causation of behavior. For the brain, this is probably the neu-
ral level or higher, depending on just how the brain’s cognitive mechanisms
function. (The notion of causal topology is necessarily informal for now; I
will discuss its formalization below.)
Call a property P an organizational invariant if it is invariant with respect
to causal topology: that is, if any change to the system that preserves the
causal topology preserves P. The sort of changes in question include: (a)
moving the system in space; (b) stretching, distorting, expanding and con-
tracting the system; (c) replacing sufficiently small parts of the system with
parts that perform the same local function (e.g. replacing a neuron with a
silicon chip with the same I/O properties); (d) replacing the causal links
between parts of a system with other links that preserve the same pattern
of dependencies (e.g., we might replace a mechanical link in a telephone
340

exchange with an electrical link); and (e) any other changes that do not alter
the pattern of causal interaction among parts of the system.
Most properties are not organizational invariants. The property of flying
is not, for instance: we can move an airplane to the ground while preserving
its causal topology, and it will no longer be flying. Digestion is not: if we
gradually replace the parts involved in digestion with pieces of metal, while
preserving causal patterns, after a while it will no longer be an instance of
digestion: no food groups will be broken down, no energy will be extracted,
and so on. The property of being tube of toothpaste is not an organizational
invariant: if we deform the tube into a sphere, or replace the toothpaste by
peanut butter while preserving causal topology, we no longer have a tube of
toothpaste.
In general, most properties depend essentially on certain features that are
not features of causal topology. Flying depends on height, digestion depends
on a particular physiochemical makeup, tubes of toothpaste depend on
shape and physiochemical makeup, and so on. Change the features in ques-
tion enough and the property in question will change, even though causal
topology might be preserved throughout.

3.2 The organizational invariance of mental properties

The central claim of this section is that most mental properties are organiza-
tional invariants. It does not matter how we stretch, move about, or replace
small parts of a cognitive system: as long as we preserve its causal topology,
we will preserve its mental properties.
An exception has to be made for properties that are partly supervenient
on states of the environment. Such properties include knowledge (if we
move a system that knows that P into an environment where P is not true,
then it will no longer know that P), and belief, on some construals where
the content of a belief depends on environmental context. However, mental
properties that depend only on internal (brain) state will be organizational
invariants. This is not to say that causal topology is irrelevant to knowledge
and belief. It will still capture the internal contribution to those properties
— that is, causal topology will contribute as much as the brain contributes.
It is just that the environment will also play a role.
The central claim can be justified by dividing mental properties into two
 341

varieties: psychological properties — those that are characterized by their

causal role, such as belief, learning, and perception — and phenomenal
properties, or those that are characterized by way in which they are con-
sciously experienced. Psychological properties are concerned with the sort
of thing the mind does, and phenomenal properties are concerned with the
way it feels (Some will hold that properties such as belief should be assimi-
lated to the second rather than the first class; I do not think that this is cor-
rect, but nothing will depend on that here).
Psychological properties, as has been argued by Armstrong (1968) and
Lewis (1972) among others, are effectively defined by their role within an
overall causal system: it is the pattern of interaction between different states
that is definitive of a system’s psychological properties. Systems with the
same causal topology will share these patterns of causal interactions among
states, and therefore, by the analysis of Lewis (1972), will share their psy-
chological properties (as long as their relation to the environment is appro-
priate).
Phenomenal properties are more problematic. It seems unlikely that these
can be defined by their causal roles (although many, including Lewis and
Armstrong, think they might be). To be a conscious experience is not to per-
form some role, but to have a particular feel. These properties are character-
ized by what it is like to have them, in Nagel’s (1974) phrase. Phenomenal
properties are still quite mysterious and ill-understood.
Nevertheless, I believe that they can be seen to be organizational invari-
ants, as I have argued elsewhere. The argument for this, very briefly, is
a reductio. Assume conscious experience is not organizationally invari-
ant. Then there exist systems with the same causal topology but different
conscious experiences. Let us say this is because the systems are made of
different materials, such as neurons and silicon; a similar argument can
be given for other sorts of differences. As the two systems have the same
causal topology, we can (in principle) transform the first system into the
second by making only gradual changes, such as by replacing neurons one
at a time with I/O equivalent silicon chips, where the overall pattern of
interaction remains the same throughout. Along the spectrum of intermedi-
ate systems, there must be two systems between which we replace less than
ten percent of the system, but whose conscious experiences differ. Consider
342

these two systems, N and S, which are identical except in that some circuit
in one is neural and in the other is silicon.
The key step in the thought-experiment is to take the relevant neural cir-
cuit in N, and to install alongside it a causally isomorphic silicon back-up
circuit, with a switch between the two circuits. What happens when we flip
the switch? By hypothesis, the system’s conscious experiences will change:
say, for purposes of illustration, from a bright red experience to a bright
blue experience (or to a faded red experience, or whatever). This follows
from the fact that the system after the change is a version of S, whereas
before the change it is just N.
But given the assumptions, there is no way for the system to notice these
changes. Its causal topology stays constant, so that all of its functional states
and behavioral dispositions stay fixed. If noticing is defined functionally
(as it should be), then there is no room for any noticing to take place, and
if it is not, any noticing here would seem to be a thin event indeed. There is
certainly no room for a thought “Hmm! Something strange just happened!”,
unless it is floating free in some Cartesian realm.3 Even if there were such a
thought, it would be utterly impotent; it could lead to no change of process-
ing within the system, which could not even mention it (If the substitution
were to yield some change in processing, then the systems would not have
the same causal topology after all. Recall that the argument has the form
of a reductio). We might even flip the switch a number of times, so that red
and blue experiences “dance” before the system’s inner eye; it will never
notice. This, I take it, is a reductio ad absurdum of the original hypothesis:
if one’s experiences change, one can potentially notice in a way that makes
some causal difference. Therefore the original assumption is false, and phe-

3
In analyzing a related thought-experiment, Searle (1991) suggests that a subject
who has undergone silicon replacement might react as follows: “You want to cry
out, `I can’t see anything. I’m going totally blind’. But you hear your voice saying
in a way that is completely out of your control, `I see a red object in front of me’”
(pp. 66-67). But given that the system’s causal topology remains constant, it is
very unclear where there is room for such “wanting” to take place, if it is not in
some Cartesian realm. Searle suggests some other things that might happen, such
as a reduction to total paralysis, but these suggestions require a change in causal
topology and are therefore not relevant to the issue of organizational invariance.
 343

nomenal properties are organizational invariants. This needs to be worked

out in more detail, of course. I give the details of this “Dancing Qualia”
argument along with a related “Fading Qualia” argument in (Chalmers
1995).
If all this works, it establishes that most mental properties are organi-
zational invariants: any two systems that share their fine-grained causal
topology will share their mental properties, modulo the contribution of the
environment.

3.3 Justifying the theses

To establish the thesis of computational sufficiency, all we need to do now
is establish that organizational invariants are fixed by some computational
structure. This is quite straightforward.
An organizationally invariant property depends only on some pattern of
causal interaction between parts of the system. Given such a pattern, we can
straightforwardly abstract it into a CSA description: the parts of the sys-
tem will correspond to elements of the CSA state-vector, and the patterns
of interaction will be expressed in the state-transition rules. This will work
straightforwardly as long as each part has only a finite number of states
that are relevant to the causal dependencies between parts, which is likely
to be the case in any biological system whose functions cannot realistically
depend on infinite precision. (I discuss the issue of analog quantities in more
detail below.) Any system that implements this CSA will share the causal
topology of the original system. In fact, it turns out that the CSA formalism
provides a perfect formalization of the notion of causal topology. A CSA
description specifies a division of a system into parts, a space of states for
each part, and a pattern of interaction between these states. This is precisely
what is constitutive of causal topology.
If what has gone before is correct, this establishes the thesis of computa-
tional sufficiency, and therefore the the view that Searle has called “strong
artificial intelligence”: that there exists some computation such that any
implementation of the computation possesses mentality. The fine-grained
causal topology of a brain can be specified as a CSA. Any implementation
of that CSA will share that causal topology, and therefore will share organi-
zationally invariant mental properties that arise from the brain.
344

The thesis of computational explanation can be justified in a similar way.

As mental properties are organizational invariants, the physical properties
on which they depend are properties of causal organization. Insofar as men-
tal properties are to be explained in terms of the physical at all, they can be
explained in terms of the causal organization of the system.4 We can invoke
further properties (implementational details) if we like, but there is a clear
sense in which they are not vital to the explanation. The neural or electronic
composition of an element is irrelevant for many purposes; to be more
precise, composition is relevant only insofar as it determines the element’s
causal role within the system. An element with different physical composi-
tion but the same causal role would do just as well. This is not to make the
implausible claim that neural properties, say, are entirely irrelevant to expla-
nation. Often the best way to investigate a system’s causal organization is to
investigate its neural properties. The claim is simply that insofar as neural
properties are explanatorily relevant, it is in virtue of the role they play in
determining a systems causal organization.
In the explanation of behavior, too, causal organization takes center stage.
A system’s behavior is determined by its underlying causal organization,
and we have seen that the computational framework provides an ideal lan-
guage in which this organization can be specified. Given a pattern of causal
interaction between substates of a system, for instance, there will be a CSA
description that captures that pattern. Computational descriptions of this
kind provide a general framework for the explanation of behavior.
For some explanatory purposes, we will invoke properties that are not
organizational invariants. If we are interested in the biological basis of cog-

4
I am skeptical about whether phenomenal properties can be explained in wholly
physical terms. As I argue in Chalmers 1996b, given any account of the physical or
computational processes underlying mentality, the question of why these processes
should give rise to conscious experience does not seem to be explainable within
physical or computational theory alone. Nevertheless, it remains the case that
phenomenal properties depend on physical properties, and if what I have said
earlier is correct, the physical properties that they depend on are organizational
properties. Further, the explanatory gap with respect to conscious experience is
compatible with the computational explanation of cognitive processes and of
behavior, which is what the thesis of computational explanation requires.
 345

nition, we will invoke neural properties. To explain situated cognition, we

may invoke properties of the environment. This is fine; the thesis of compu-
tational explanation is not an exclusive thesis. Still, usually we are interested
in neural properties insofar as they determine causal organization, we are
interested in properties of the environment insofar as they affect the pat-
tern of processing in a system, and so on. Computation provides a general
explanatory framework that these other considerations can supplement.5

3.4 Some objections

A computational basis for cognition can be challenged in two ways. The
first sort of challenge argues that computation cannot do what cognition
does: that a computational simulation might not even reproduce human
behavioral capacities, for instance, perhaps because the causal structure
in human cognition goes beyond what a computational description can
provide. The second concedes that computation might capture the capaci-
ties, but argues that more is required for true mentality. I will consider four
objections of the second variety, and then three of the first. Answers to most
of these objections fall directly out of the framework developed above.
But a computational model is just a simulation! According to this
objection, due to Searle (1980), Harnad (1989), and many others, we do
not expect a computer model of a hurricane to be a real hurricane, so why
should a computer model of mind be a real mind? But this is to miss the
important point about organizational invariance. A computational simula-
tion is not a mere formal abstraction, but has rich internal dynamics of its
own. If appropriately designed it will share the causal topology of the sys-
tem that is being modeled, so that the system’s organizationally invariant
properties will be not merely simulated but replicated.
The question about whether a computational model simulates or repli-

5
Of course there is a sense in which it can be said that connectionist models
perform “computation over representation”, in that connectionist processing
involves the transformation of representations, but this sense is to weak to cut the
distinction between symbolic and subsymbolic computation at its joints. Perhaps
the most interesting foundational distinction between symbolic and connectionist
systems is that in the former but not in the latter, the computational (syntactic)
primitives are also the representational (semantic) primitives.
346

cates a given property comes down to the question of whether or not the
property is an organizational invariant. The property of being a hurricane is
obviously not an organizational invariant, for instance, as it is essential to
the very notion of hurricanehood that wind and air be involved. The same
goes for properties such as digestion and temperature, for which specific
physical elements play a defining role. There is no such obvious objection
to the organizational invariance of cognition, so the cases are disanalogous,
and indeed, I have argued above that for mental properties, organizational
invariance actually holds. It follows that a model that is computationally
equivalent to a mind will itself be a mind.
Syntax and semantics. Searle (1984) has argued along the following
lines: (1) A computer program is syntactic; (2) Syntax is not sufficient for
semantics; (3) Minds have semantics; therefore (4) Implementing a com-
puter program is insufficient for a mind. Leaving aside worries about the
second premise, we can note that this argument equivocates between pro-
grams and implementations of those programs. While programs themselves
are syntactic objects, implementations are not: they are real physical sys-
tems with complex causal organization, with real physical causation going
on inside. In an electronic computer, for instance, circuits and voltages push
each other around in a manner analogous to that in which neurons and
activations push each other around. It is precisely in virtue of this causation
that implementations may have cognitive and therefore semantic properties.
It is the notion of implementation that does all the work here. A program
and its physical implementation should not be regarded as equivalent —
they lie on entirely different levels, and have entirely different properties. It
is the program that is syntactic; it is the implementation that has semantic
content. Of course, there is still a substantial question about how an imple-
mentation comes to possess semantic content, just as there is a substantial
question about how a brain comes to possess semantic content. But once
we focus on the implementation, rather than the program, we are at least in
the right ball-park. We are talking about a physical system with causal heft,
rather than a shadowy syntactic object. If we accept, as is extremely plau-
sible, that brains have semantic properties in virtue of their causal organiza-
tion and causal relations, then the same will go for implementations. Syntax
may not be sufficient for semantics, but the right kind of causation is.
 347

The Chinese room. There is not room here to deal with Searle’s famous
Chinese room argument in detail. I note, however, that the account I have
given supports the “Systems reply”, according to which the entire system
understands Chinese even if the homunculus doing the simulating does not.
Say the overall system is simulating a brain, neuron-by-neuron. Then like
any implementation, it will share important causal organization with the
brain. In particular, if there is a symbol for every neuron, then the patterns
of interaction between slips of paper bearing those symbols will mirror pat-
terns of interaction between neurons in the brain, and so on. This organiza-
tion is implemented in a baroque way, but we should not let the baroque-
ness blind us to the fact that the causal organization — real, physical causal
organization — is there (The same goes for a simulation of cognition at
level above the neural, in which the shared causal organization will lie at a
coarser level).
It is precisely in virtue of this causal organization that the system pos-
sesses its mental properties. We can rerun a version of the “dancing qualia”
argument to see this. In principle, we can move from the brain to the Chi-
nese room simulation in small steps, replacing neurons at each step by little
demons doing the same causal work, and then gradually cutting down labor
by replacing two neighboring demons by one who does the same work.
Eventually we arrive at a system where a single demon is responsible for
maintaining the causal organization, without requiring any real neurons at
all. This organization might be maintained between marks on paper, or it
might even be present inside the demon’s own head, if the calculations are
memorized. The arguments about organizational invariance all hold here —
for the same reasons as before, it is implausible to suppose that the system’s
experiences will change or disappear.
Performing the thought-experiment this way makes it clear that we
should not expect the experiences to be had by the demon. The demon is
simply a kind of causal facilitator, ensuring that states bear the appropriate
causal relations to each other. The conscious experiences will be had by the
system as a whole. Even if that system is implemented inside the demon
by virtue of the demon’s memorization, the system should not be confused
with demon itself. We should not suppose that the demon will share the
implemented system’s experiences, any more than it will share the experi-
348

ences of an ant that crawls inside its skull: both are cases of two computa-
tional systems being implemented within a single physical space. Mental
properties arising from distinct computational systems will be quite distinct,
and there is no reason to suppose that they overlap.
What about the environment? Some mental properties, such as knowl-
edge and even belief, depend on the environment being a certain way. Com-
putational organization, as I have outlined it, cannot determine the environ-
mental contribution, and therefore cannot fully guarantee this sort of mental
property. But this is no problem. All we need computational organization
to give us is the internal contribution to mental properties: that is, the same
contribution that the brain makes (for instance, computational organization
will determine the so-called “narrow content” of a belief, if this exists; see
Fodor 1987). The full panoply of mental properties might only be deter-
mined by computation-plus-environment, just as it is determined by brain-
plus-environment. These considerations do not count against the prospects
of artificial intelligence, and they affect the aspirations of computational
cognitive science no more than they affect the aspirations of neuroscience.
Is cognition computable? In the preceding discussion I have taken for
granted that computation can at least simulate human cognitive capacity,
and have been concerned to argue that this counts as honest-to-goodness
mentality. The former point has often been granted by opponents of AI (e.g.
Searle 1980) who have directed the fire at the latter, but it is not uncontro-
versial.
This is to some extent an empirical issue, but the relevant evidence is
solidly on the side of computability. We have every reason to believe that
the low-level laws of physics are computable. If so, then low-level neuro-
physiological processes can be computationally simulated; it follows that
the function of the whole brain is computable too, as the brain consists in a
network of neurophysiological parts. Some have disputed the premise: for
example, Penrose (1989) has speculated that the effects of quantum gravity
are noncomputable, and that these effects may play a role in cognitive func-
tioning. He offers no arguments to back up this speculation, however, and
there is no evidence of such noncomputability in current physical theory (see
Pour-El and Richards (1989) for a discussion). Failing such a radical devel-
opment as the discovery that the fundamental laws of nature are uncomput-
 349

able, we have every reason to believe that human cognition can be compu-
tationally modeled.
What about Gödel’s theorem? Gödel’s theorem states that for any con-
sistent formal system, there are statements of arithmetic that are unprov-
able within the system. This has led some (Lucas 1963; Penrose 1989) to
conclude that humans have abilities that cannot be duplicated by any com-
putational system. For example, our ability to “see” the truth of the Gödel
sentence of a formal system is argued to be non-algorithmic. I will not deal
with this objection in detail here, as the answer to it is not a direct applica-
tion of the current framework. I will simply note that the assumption that
we can see the truth of arbitrary Gödel sentences requires that we have the
ability to determine the consistency or inconsistency of any given formal
system, and there is no reason to believe that we have this ability in general
(For more on this point, see Putnam 1960, Bowie 1982 and the commentar-
ies on Penrose 1990).
Discreteness and continuity. An important objection notes that the CSA
formalism only captures discrete causal organization, and argues that some
cognitive properties may depend on continuous aspects of that organization,
such as analog values or chaotic dependencies.
A number of responses to this are possible. The first is to note that the
current framework can fairly easily be extended to deal with computation
over continuous quantities such as real numbers. All that is required is that
the various substates of a CSA be represented by a real parameter rather
than a discrete parameter, where appropriate restrictions are placed on
allowable state-transitions (for instance, we can require that parameters are
transformed polynomially, where the requisite transformation can be condi-
tional on sign). See Blum, Shub and Smale (1989) for a careful working-out
of some of the relevant theory of computability. A theory of implementation
can be given along in a fashion similar to the account I have given above,
where continuous quantities in the formalism are required to correspond
to continuous physical parameters with an appropriate correspondence in
state-transitions.
This formalism is still discrete in time: evolution of the continuous states
proceeds in discrete temporal steps. It might be argued that cognitive orga-
nization is in fact continuous in time, and that a relevant formalism should
350

capture this. In this case, the specification of discrete state-transitions

between states can be replaced by differential equations specifying how
continuous quantities change in continuous time, giving a thoroughly con-
tinuous computational framework. MacLennan (1990) describes a frame-
work along these lines. Whether such a framework truly qualifies as com-
putational is largely a terminological matter, but there it is arguable that the
framework is significantly similar in kind to the traditional approach; all
that has changed is that discrete states and steps have been “smoothed out”.
We need not go this far, however. There are good reasons to suppose that
whether or not cognition in the brain is continuous, a discrete framework
can capture everything important that is going on. To see this, we can note
that a discrete abstraction can describe and simulate a continuous process to
any required degree of accuracy. It might be objected that chaotic processes
can amplify microscopic differences to significant levels. Even so, it is
implausible that the correct functioning of mental processes depends on the
precise value of the tenth decimal place of analog quantities. The presence
of background noise and randomness in biological systems implies that
such precision would inevitably be “washed out” in practice. It follows that
although a discrete simulation may not yield precisely the behavior that a
given cognitive system produces on a given occasion, it will yield plausible
behavior that the system might have produced had background noise been
a little different. This is all that a proponent of artificial intelligence need
claim.
Indeed, the presence of noise in physical systems suggests that any given
continuous computation of the above kinds can never be reliably imple-
mented in practice, but only approximately implemented. For the purposes
of artificial intelligence we will do just as well with discrete systems, which
can also give us approximate implementations of continuous computations.
It follows that these considerations do not count against the theses of
computational sufficiency or of computational explanation. To see the first,
note that a discrete simulation can replicate everything essential to cogni-
tive functioning, for the reasons above, even though it may not duplicate
every last detail of a given episode of cognition. To see the second, note that
for similar reasons the precise values of analog quantities cannot be relevant
to the explanation of our cognitive capacities, and that a discrete descrip-
 351

tion can do the job.

This is not to exclude continuous formalisms from cognitive explana-
tion. The thesis of computational explanation is not an exclusive thesis. It
may be that continuous formalisms will provide a simpler and more natural
framework for the explanation of many dynamic processes, as we find in
the theory of neural networks. Perhaps the most reasonable version of the
computationalist view accepts the thesis of (discrete) computational suf-
ficiency, but supplements the thesis of computational explanation with the
proviso that continuous computation may sometimes provide a more natu-
ral explanatory framework (a discrete explanation could do the same job,
but more clumsily). In any case, continuous computation does not give us
anything fundamentally new.

4. Other kinds of computationalism

Artificial intelligence and computational cognitive science are committed

to a kind of computationalism about the mind, a computationalism defined
by the theses of computational sufficiency and computational explana-
tion. In this paper I have tried to justify this computationalism, by spelling
out the role of computation as a tool for describing and duplicating causal
organization. I think that this kind of computationalism is all that artificial
intelligence and computational cognitive science are committed to, and
indeed is all that they need. This sort of computationalism provides a gen-
eral framework precisely because it makes so few claims about the kind of
computation that is central to the explanation and replication of cognition.
No matter what the causal organization of cognitive processes turns out to
be, there is good reason to believe that it can be captured within a computa-
tional framework.
The fields have often been taken to be committed to stronger claims,
sometimes by proponents and more often by opponents. For example, Edel-
man (1989) criticizes the computational approach to the study of the mind
on the grounds that:

An analysis of the evolution, development, and structure of brains

makes it highly unlikely that they could be Turing machines. This is so
352

because of the enormous individual variation in structure that brains

possess at a variety of organizational levels. [...] [Also,] an analysis of
both ecological and environmental variation, and of the categorization
procedures of animals and humans, makes it highly unlikely that the
world (physical and social) can function as a tape for a Turing machine.
(Edelman 1989, p. 30.)

But artificial intelligence and computational cognitive science are not com-
mitted to the claim that the brain is literally a Turing machine with a mov-
ing head and a tape, and even less to the claim that that tape is the environ-
ment. The claim is simply that some computational framework can explain
and replicate human cognitive processes. It may turn out that the relevant
computational description of these processes is very fine-grained, reflect-
ing extremely complex causal dynamics among neurons, and it may well
turn out that there is significant variation in causal organization between
individuals. There is nothing here that is incompatible with a computational
approach to cognitive science.
In a similar way, a computationalist need not claim that the brain is a von
Neumann machine, or has some other specific architecture. Like Turing
machines, von Neumann machines are just one kind of architecture, par-
ticularly well-suited to programmability, but the claim that the brain imple-
ments such an architecture is far ahead of any empirical evidence and is
most likely false. The commitments of computationalism are more general.
Computationalism is occasionally associated with the view that cogni-
tion is rule-following, but again this is a strong empirical hypothesis that
is inessential to the foundations of the fields. It is entirely possible that the
only “rules” found in a computational description of thought will be at a
very low level, specifying the causal dynamics of neurons, for instance, or
perhaps the dynamics of some level between the neural and the cognitive.
Even if there are no rules to be found at the cognitive level, a computational
approach to the mind can still succeed. Another claim to which a computa-
tionalist need not be committed are “the brain is a computer”; as we have
seen, it is not computers that are central but computations).
The most ubiquitous “strong” form of computationalism has been what
we may call symbolic computationalism: the view that cognition is compu-
 353

tation over representation (Newell and Simon 1976; Fodor and Pylyshyn
1988). To a first approximation, we can cash out this view as the claim that
the computational primitives in a computational description of cognition
are also representational primitives. That is to say, the basic syntactic enti-
ties between which state-transitions are defined are themselves bearers of
semantic content, and are therefore symbols.
Symbolic computationalism has been a popular and fruitful approach
to the mind, but it does not exhaust the resources of computation. Not all
computations are symbolic computations. We have seen that there are some
Turing machines that lack semantic content altogether, for instance. Perhaps
systems that carry semantic content are more plausible models of cognition,
but even in these systems there is no reason why the content must be car-
ried by the systems’ computational primitives. In connectionist systems, for
example, the basic bearers of semantic content are distributed representa-
tions, patterns of activity over many units, whereas the computational prim-
itives are simple units that may themselves lack semantic content. To use
Smolensky’s term (Smolensky 1988), these systems perform subsymbolic
computation: the level of computation falls below the level of representa-
tion.6 But the systems are computational nevertheless.

6
[Note added 2011.] In order to make them compatible with the views of
consciousness in Chalmers 1996b, the thesis of computational sufficiency and the
claim that mental properties are organizational invariants must be understood in
terms of nomological rather metaphysical necessity: the right kind of computation
suffices with nomological necessity for possession of a mind, mental properties
supervene nomologically on causal topology. These claims are compatible with
the metaphysical possibility of systems with the same organization and no
consciousness. As for the thesis of computational explanation: if one construes
cognitive processes to include arbitrary intentional or representational states,
then I think these cannot be explained wholly in terms of computation, as I think
that phenomenal properties and environmental properties play a role here. One
might qualify the thesis by understanding “cognitive processes” and “behavior” in
functional and nonintentional terms, or by saying that computational explanation
can undergird intentional explanation when appropriately supplemented, perhaps
by phenomenal and environmental elements. Alternatively, the version of the thesis
most directly supported by the argument in the text is that computation provides
a general framework for the mechanistic explanation of cognitive processes and
354

Note that the distinction between symbolic and subsymbolic computa-

tion does not coincide with the distinction between different computational
formalisms, such as Turing machines and neural networks. Rather, the dis-
tinction divides the class of computations within each of these formalisms.
Some Turing machines perform symbolic computation, and some perform
subsymbolic computation; the same goes for neural networks. (Of course it
is sometimes said that all Turing machines perform “symbol manipulation”,
but this holds only if the ambiguous term “symbol” is used in a purely syn-
tactic sense, rather than in the semantic sense I am using here.)
Both proponents and opponents of a computational approach have often
implicitly identified computation with symbolic computation. A critique
called What Computers Can’t Do (Dreyfus 1972), for instance, turns out
to be largely directed at systems that perform computation over explicit
representation. Other sorts of computation are left untouched, and indeed
systems performing subsymbolic computation seem well-suited for some of
Dreyfus’s problem areas. The broader ambitions of artificial intelligence are
therefore left intact.
On the other side of the fence, Fodor (1992) uses the name “Computational
Theory of Mind” for a version of symbolic computationalism, and suggests
that Turing’s main contribution to cognitive science is the idea that syntac-
tic state-transitions between symbols can be made to respect their semantic
content. This strikes me as false. Turing was concerned very little with the
semantic content of internal states, and the concentration on symbolic com-
putation came later. Rather, Turing’s key contribution was the formalization
of the notion of mechanism, along with the associated universality of the
formalization. It is this universality that gives us good reason to suppose
that computation can do almost anything that any mechanism can do, thus
accounting for the centrality of computation in the study of cognition.
Indeed, a focus on symbolic computation sacrifices the universality that
is at the heart of Turing’s contribution. Universality applies to entire classes
of automata, such as Turing machines, where these classes are defined syn-
tactically. The requirement that an automaton performs computation over

behavior. That is, insofar as cognitive processes and behavior are explainable
mechanistically, they are explainable computationally.
 355

representation is a strong further constraint, a semantic constraint that plays

no part in the basic theory of computation. There is no reason to suppose
that the much narrower class of Turing machines that perform symbolic
computation is universal. If we wish to appeal to universality in a defense
of computationalism, we must cast the net more widely than this.7
The various strong forms of computationalism outlined here are bold
empirical hypotheses with varying degrees of plausibility. I suspect that
they are all false, but in any case their truth and falsity is not the issue here.
Because they are such strong empirical hypotheses, they are in no position
to serve as a foundation for artificial intelligence and computational cogni-
tive science. If the fields were committed to these hypotheses, their status
would be much more questionable than it currently is. Artificial intelligence
and computational cognitive science can survive the discovery that the brain
is not a von Neumann machine, or that cognition is not rule-following, or
that the brain does not engage in computation over representation, precisely
because these are not among the fields’ foundational commitments. Compu-
tation is much more general than this, and consequently much more robust.8

7
It is common for proponents of symbolic computationalism to hold, usually
as an unargued premise, that what makes a computation a computation is the
fact that it involves representations with semantic content. The books by Fodor
(1975) and Pylyshyn (1984), for instance, are both premised on the assumption that
there is no computation without representation. Of course this is to some extent a
terminological issue, but as I have stressed in 2.2 and here, this assumption has no
basis in computational theory and unduly restricts the role that computation plays
in the foundations of cognitive science.
8
Some other claims with which computationalism is sometimes associated
include “the brain is a computer”, “the mind is to the brain as software is to
hardware”, and “cognition is computation”. The first of these is not required, for
the reasons given in 2.2: it is not computers that are central to cognitive theory but
computations. The second claim is an imperfect expression of the computationalist
position for similar reasons: certainly the mind does not seem to be something
separable that the brain can load and run, as a computer’s hardware can load and
run software. Even the third does not seem to me to be central to computationalism:
perhaps there is a sense in which it is true, but what is more important is that
computation suffices for and explains cognition. See Dietrich (1990) for some
related distinctions between computationalism, “computerism”, and “cognitivism”.
356

5. Conclusion: Toward a minimal computationalism

The view that I have advocated can be called minimal computationalism.

It is defined by the twin theses of computational sufficiency and computa-
tional explanation, where computation is taken in the broad sense that dates
back to Turing. I have argued that these theses are compelling precisely
because computation provides a general framework for describing and
determining patterns of causal organization, and because mentality is rooted
in such patterns. The thesis of computational explanation holds because
computation provides a perfect language in which to specify the causal
organization of cognitive processes; and the thesis of computational suf-
ficiency holds because in all implementations of the appropriate computa-
tions, the causal structure of mentality is replicated.
Unlike the stronger forms of computationalism, minimal computational-
ism is not a bold empirical hypothesis. To be sure, there are some ways that
empirical science might prove it to be false: if it turns out that the funda-
mental laws of physics are noncomputable and if this noncomputability
reflects itself in cognitive functioning, for instance, or if it turns out that our
cognitive capacities depend essentially on infinite precision in certain ana-
log quantities, or indeed if it turns out that cognition is mediated by some
non-physical substance whose workings are not computable. But these
developments seem unlikely; and failing developments like these, computa-
tion provides a general framework in which we can express the causal orga-
nization of cognition, whatever that organization turns out to be.
Minimal computationalism is compatible with such diverse programs as
connectionism, logicism, and approaches focusing on dynamic systems,
evolution, and artificial life. It is occasionally said that programs such as
connectionism are “noncomputational”, but it seems more reasonable to
say that the success of such programs would vindicate Turing’s dream of a
computational intelligence, rather than destroying it.
Computation is such a valuable tool precisely because almost any theory
of cognitive mechanisms can be expressed in computational terms, even
though the relevant computational formalisms may vary. All such theories
are theories of causal organization, and computation is sufficiently flexible
 357

that it can capture almost any kind of organization, whether the causal rela-
tions hold between high-level representations or among low-level neural
processes. Even such programs as the Gibsonian theory of perception are
ultimately compatible with minimal computationalism. If perception turns
out to work as the Gibsonians imagine, it will still be mediated by causal
mechanisms, and the mechanisms will be expressible in an appropriate
computational form. That expression may look very unlike a traditional
computational theory of perception, but it will be computational neverthe-
less.
In this light, we see that artificial intelligence and computational cogni-
tive science do not rest on shaky empirical hypotheses. Instead, they are
consequences of some very plausible principles about the causal basis of
cognition, and they are compatible with an extremely wide range of empiri-
cal discoveries about the functioning of the mind. It is precisely because
of this flexibility that computation serves as a foundation for the fields in
question, by providing a common framework within which many different
theories can be expressed, and by providing a tool with which the theories’
causal mechanisms can be instantiated. No matter how cognitive science
progresses in the coming years, there is good reason to believe that compu-
tation will be at center stage.

References

Armstrong, D.M. 1968. A Materialist Theory of the Mind. Routledge and Kegan
Paul.
Block, N. 1981. Psychologism and behaviorism. Philosophical Review 90: 5-43.
Blum, L., Shub, M., and Smale, S. 1989. On a theory of computation and complexity
over the real numbers: NP-completeness, recursive functions, and universal
machines. Bulletin (New Series) of the American Mathematical Society 21(1):
1-46.
Bowie, G. 1982. Lucas’ number is finally up. Journal of Philosophical Logic 11:
279-85.
Chalmers, D.J. (1995). Absent qualia, fading qualia, dancing qualia. In (T.
Metzinger, ed) Conscious Experience. Ferdinand Schoningh.
Chalmers, D.J. (1996a). Does a rock implement every finite-state automaton?
Synthese.
358

Chalmers, D.J. (1996b). The Conscious Mind: In Search of a Fundamental Theory.

Oxford University Press. Press.
Dietrich, E.S. 1990. Computationalism. Social Epistemology.
Dretske, F. 1981. Knowledge and the Flow of Information. MIT Press.
Dreyfus, H. 1972. What Computers Can’t Do. Harper and Row.
Edelman, G.M. 1989. The Remembered Present: A Biological Theory o f
Consciousness. Basic Books.
Fodor, J.A. 1975. The Language of Thought. Harvard University Press.
Fodor, J.A. 1987. Psychosemantics: The Problem of Meaning in the Philosophy of
Mind. MIT Press.
Fodor, J.A. and Pylyshyn, Z.W. 1988. Connectionism and cognitive architecture.
Cognition 28: 3-71.
Fodor, J.A. 1992. The big idea: Can there be a science of mind? Times Literary
Supplement 4567: 5-7 (July 3, 1992).
Gibson, J. 1979. The Ecological Approach to Visual Perception. Houghton Mifflin.
Harnad, S. 1989. Minds, machines and Searle. Journal of Experimental and
Theoretical Artificial Intelligence 1: 5-25.
Haugeland, J. 1985. Artificial intelligence: The Very Idea. MIT Press.
Lewis, D. 1972. Psychophysical and theoretical identifications. Australasian Journal
of Philosophy 50: 249-58.
Lucas, J.R. 1963. Minds, machines, and Gödel. Philosophy 36: 112-127.
MacLennan, B. 1990. Field computation: A theoretical framework for massively
parallel analog computation, Parts I-IV. Technical Report CS-90-100. Computer
Science Department, University of Tennessee.
Nagel, T. 1974. What is it like to be a bat? Philosophical Review 4: 435-50.
Newell, A. and Simon, H.A. 1981. Computer science as empirical inquiry: Symbols
and search. Communications of the Association for Computing Machinery 19:
113-26.
Penrose, R. 1989. The Emperor’s New Mind: Concerning computers, minds, and the
laws of physics. Oxford University Press.
Penrose, R. 1990. Precis of The Emperor’s New Mind. Behavioral and Brain
Sciences 13: 643-655.
Pour-El, M.B., and Richards, J.I. 1989. Computability in Analysis and Physics.
Springer-Verlag.
Putnam, H. 1960. Minds and machines. In (S. Hook, ed.) Dimensions of Mind. New
York University Press.
Putnam, H. 1967. The nature of mental states. In (W.H. Capitan and D.D. Merrill,
eds.) Art, Mind, and Religion. University of Pittsburgh Press.
Putnam, H. 1988. Representation and Reality. MIT Press.
 359

Pylyshyn, Z.W. 1984. Computation and Cognition: Toward a Foundation for

Cognitive Science. MIT Press.
Searle, J.R. 1980. Minds, brains and programs. Behavioral and Brain Sciences 3:
417-57.
Searle, J.R. 1984. Minds, brains, and science. Harvard University Press.
Searle, J.R. 1990. Is the brain a digital computer? Proceedings and Addresses of the
American Philosophical Association 64: 21-37.
Searle, J.R. 1991. The Rediscovery of the Mind. MIT Press.
Smolensky, P. 1988. On the proper treatment of connectionism. Behavioral and
Brain Sciences 11: 1-23.
Turing, A.M. 1936. On computable numbers, with an application to the
Entscheidungsproblem. Proceedings of the London Mathematical Society,
Series 2 42: 230-65.