Gualtiero Piccinini’s Research
NB: Comments on my work are always
welcome
Updated: February 2011
My research covers several areas, such
as computational theories of mind, cognitive neuroscience, intentionality, and
consciousness. The goal is to understand the mind, so all the pieces are
connected in various ways. My work may be approximately organized within
the following categories.
·
Background
of Computationalism
·
The
Mechanistic Account of Computation
·
Reformulating
Computationalism
·
From
Cognitive Science to Cognitive Neuroscience
·
Intentionality
·
Consciousness
and Introspection
Background of Computationalism
I have devoted
considerable energy to understanding the original motivation of computational
theories of mind. The motivation is to find a mechanistic explanation of
mental capacities. It is important that we identify the relevant notion
of mechanistic explanation. I think the notion that serves our purposes
is roughly the one worked out by Carl Craver and other
philosophers of science. But in order to show that mechanistic
explanation in the relevant sense is what computationalists
are after, we also need to dispel a number of confusions that have permeated
the literature.
In Alan Turing
and the Mathematical Objection, I reconstruct Turing’s (usually
misunderstood) reply to the mathematical objection. (Jack Copeland
invited me to present this paper at a workshop on hypercomputation
in London; the proceedings were published in Minds and Machines,
2003.) According to the mathematical objection, machines cannot think
because they can only follow a finite set of rules, whereas minds can invent new
rules. This objection was initially raised by Kurt Gödel and was later
made famous by John Lucas and Roger Penrose. I argue that Turing’s views
on intelligence were subtler than hitherto recognized. He did not believe
that a Turing machine could reproduce the human mind, but he did believe that a
digital computer might, because a digital computer might have the ability of
changing its programs in ways that correspond to the invention of new rules by
human beings. Although previously unknown or unappreciated, Turing’s
reply to the mathematical objection is the most effective that I know of.
In Allen Newell (in the New
Dictionary of Scientific Biography, Thomson Gale, 2007), I briefly
reconstruct the main ideas and contributions of one of the greatest and most
sophisticated champions of the computational theory of mind.
In The
First Computational Theory of Mind and Brain: A Close Look at McCulloch and
Pitts’s ‘Logical Calculus of Ideas Immanent in Nervous Activity’ (in Synthese, 2004), I argue that contrary to popular
belief, the first computational theory of mind and brain is not due to Turing
but to McCulloch and Pitts. I investigate the enormous impact their
theory had, both positively, in offering a mechanistic
explanation of mental capacities, and negatively, in introducing the mistaken
idea that appealing to computation automatically explains the intentionality of
mental states. The mixture of computationalism and semantic language that
one finds in McCulloch and Pitts carried over in the philosophical literature,
where most authors believe computational states are individuated by their
semantic properties. I call this the semantic view of computation.
I look at the
sources of the semantic view of computation in Functionalism,
Computationalism, and Mental Contents (in Canadian Journal of Philosophy,
2004), where I argue that the semantic view of computation is untenable and
question begging, and should be replaced by a non-semantic view of computation.
But a
non-semantic view of computation is workable only if we don’t conflate functional/mechanistic
descriptions and computational ones, as is often done in the literature.
In Functionalism,
Computationalism, and Mental States (in Studies in the History and
Philosophy of Science, 2004), I identify the origin of this conflation and
attempt to eradicate it. One of my conclusions is that computational
descriptions are only one kind of functional/mechanistic descriptions among
others. In this paper, I also discuss and reject the popular view that
there is a useful sense in which everything performs computations.
The Mechanistic
Account of Computation
I have developed
a novel account of computation, the mechanistic account, which
integrates conceptual resources from computability theory, computer design, and
philosophy of science. According to the mechanistic account, computing
systems are mechanisms that have the function of manipulating
medium-independent vehicles in accordance with a general rule that applies to
all vehicles and depends on the inputs for its application. A medium-independent vehicle is a vehicle
defined simply in terms of differences between different portions of the
vehicles along a relevant dimension, and not in terms of any of its more
specific physical properties. Thus,
medium-independent vehicles can be implemented in different physical
media. The vehicles of digital
computations are digits, which are one kind of medium-independent vehicle, but
there are other kinds of medium-independent vehicles (e.g., continuous
variables), which give rise to other kinds of computation. A key term
here in the mechanistic account is “function”: computing mechanisms are
individuated by their functional (non-semantic) properties, and functional properties
are specified by mechanistic explanations.
NB: In the
following four papers, I used the term “computation” for what I now call
digital computation. A more updated,
general, and adequate account of computation is in Section 3 of “Information
Processing, Computation, and Cognition” (written with Andrea Scarantino and
published in 2011 in Journal of
Biological Physics).
In Computation
without Representation (in Philosophical Studies, 2008), I formulate
the mechanistic account with respect to digital computational states, defend it
on the grounds of computer science’s explanatory practices, and criticize
existing arguments for the semantic account of computation.
In Computing Mechanisms (in Philosophy
of Science, 2007), I formulate the mechanistic account more generally in terms
of digital computing mechanisms and their mechanistic explanation, and argue
that it is superior to other accounts of digital computing mechanisms because
it has a number of desirable features.
In Computers (in Pacific
Philosophical Quarterly, 2008), I argue that contrary to what many
philosophers have maintained, there is a principled distinction between digital
computers properly so called and other digital computing mechanisms, and I use
the mechanistic account to draw such a distinction in terms of their functional
properties. I also exhibit the fruitfulness of the mechanistic account by
analyzing in some detail some of the theoretically important functional
differences between computers, such as programmability vs. program control,
special purpose vs. general purpose, parallel vs. serial, and analog vs.
digital. There are quite a few surprises for some orthodox views here.
In Some
Neural Networks Compute, Others Don’t (in Neural
Networks, 2008), I look more closely at connectionist systems and the sense
in which they perform (digital) computations. Connectionist systems are
often invoked as the most plausible model of neural and cognitive
mechanisms. Yet there is considerable confusion as to whether
connectionism is a computational framework. I show how the mechanistic
account naturally accommodates paradigmatic connectionist systems and sheds
light on the way in which they perform (digital) computations. I also
draw a distinction between connectionist systems that perform (digital)
computations in a classical way, connectionist systems that perform (digital)
computations in a non-classical way, and connectionist systems that don’t
perform (digital) computations at all. I believe brains fall into the
last class.
In The Physical
Church-Turing Thesis: Modest or Bold? (forthcoming in British Journal for the Philosophy of Science), I try to shed light
on the Physical Church-Turing thesis, not to be confused with the original
thesis put forward by Church and Turing. I argue that existing
formulations of the Physical Church-Turing thesis are too strong and too little
concerned with practical usefulness to be relevant to the notion of
computability that motivates the (original) Church-Turing thesis and the
discipline of computer science. Such formulations should be replaced by a
more modest but more pertinent formulation, according to which anything that
can be physically computed (in a sense specified in terms of usability) can be
computed by some Turing machine. Besides being more pertinent to
computability, this is a more plausible thesis than traditional formulations,
although its truth value remains unknown.
Reformulating
Computationalism
With the
mechanistic account in hand, we can go back to computationalism and begin to
reformulate and resolve some old disputes. A first step is to appreciate
that computational theories of mind and brain have great elegance and
explanatory power. According to the mechanistic account, computational
theories are bona fide mechanistic theories, explaining the capacities of a
system in terms of their components and their functional organization.
Different versions of the computational theory differ in the kinds of
mechanisms they postulate, and their explanatory power is a function of the
capacities that can be exhibited by the mechanisms that they postulate.
In this respect, for instance, paradigmatic “connectionist” theories are less
explanatorily powerful than paradigmatic “classical” theories.
In
The Mind as the
Software of the Brain? Revisiting Functionalism,
Computationalism, and Computational Functionalism (in Philosophy and
Phenomenological Research, 2010),
I clarify the relationship between functionalism, computationalism, and
computational functionalism. I give clear content to the idea that functionalism
and computationalism are logically independent views. Functionalism is
the view that the mind is the functional organization of the brain;
computationalism is the view that the functional organization of the brain is
computational; and computational functionalism is the conjunction of the
two. I explicate all the relevant notions, including functional
organization, in mechanistic terms. Thus, this paper offers a novel
formulation of functionalism, which I call mechanistic functionalism.
In Computationalism,
the Church-Turing Thesis, and the Church-Turing Fallacy (in Synthese, 2007), I address the main arguments to the
effect that the Church-Turing thesis entails computationalism. Although
these arguments turn out to be unsound, I do point out some important
connections between computation, the Church-Turing thesis, and theories of mind
and brain.
In Computational Modeling vs.
Computational Explanation: Is Everything a Turing Machine, and Does It Matter
to the Philosophy of Mind? (in Australasian J
Phil, 2007) I discuss in detail pancomputationalism—the
view that everything performs computations. Believe it or not, pancomputationalism has been defended or at least endorsed
in some form by a large number of distinguished philosophers (and scientists),
including Hilary Putnam, David Chalmers, and John Searle. This is largely
because until now, no one has produced an adequate account of the difference
between things that compute and things that don’t. To remedy this, I
offer an account of the distinction between computational explanation
(explaining a phenomenon in terms of the computations performed by the
mechanism that is responsible for the phenomenon) and computational modeling
(describing a phenomenon using computations). By using this distinction,
I dismiss the view that everything performs computations as either trivial or
false.
In two papers written with Andrea
Scarantino--“Computation
vs. Information Processing: How They Are Different and Why It Matters” (in
Studies in Hist. and Phil. of Sci., 2010) and “Information
Processing, Computation, and Cognition” (in J. Biol. Phys., 2011)—we
explore the relationships between information, computation, and cognition. The first paper puts more emphasis on
history, but otherwise the second paper is a revised, improved, and expanded
version of the first, so I recommend reading the second one. Bottom line:
computation is not the same as information processing, computing does
not entail processing information, but processing information does entail
computing (in the generic sense). The notions of computation and information have
different histories, are associated with different bodies of mathematical
theory, and have different roles to play in a theory of cognition. Furthermore,
cognition requires both information processing (in several senses) and
computation in the generic sense, although it remains to be seen which type of computation
is involved in natural cognition.
In Computationalism
in the Philosophy of Mind (in Philosophy Compass, 2009) and Computationalism (forthcoming in the Oxford Handbook of Philosophy and Cognitive
Science), I review the recent literature on computationalism. Among
other things, I connect some of the dots between some of my previous
papers. The second paper is more complete,
updated, and improved relative to the first, so I recommend the second one even
though it’s a bit longer. Notice that
Section 6 of the paper posted here will be cut from the published version for
reasons of space.
From Cognitive Science
to Cognitive Neuroscience
The received
view about psychological explanation is that it amounts of functional analysis,
which is an explanatory style distinct and autonomous from mechanistic
explanation. In “Integrating
Psychology and Neuroscience: Functional Analyses as Mechanism Sketches”
(with Carl Craver, forthcoming in Synthese), we argues that contrary to the received view, functional
analyses are mechanism sketches (i.e., elliptical mechanistic explanations),
and therefore functional analyses are neither distinct nor autonomous from
mechanistic explanations, and therefore psychological explanations can be
seamlessly integrated with neuroscientific
explanations.
The conceptual relations between computational
explanations of mind and neighboring topics, such as modeling the mind,
explaining the mind mechanistically, and understanding the brain, are not
always carefully and correctly laid out. I clarify some of these
relations in Computational
Explanation in Neuroscience (in Synthese,
2006). The same essay also introduces some relevant articles in the same
issue of Synthese, which I guest-edited.
The articles are by Carl Craver, Frances Egan and Robert Matthews, Oron Shagrir, and Rick Grush.
In “The
Resilience of Computationalism,” (in Philosophy
of Science, 2010), I review arguments against computationalism with
emphasis on arguments from differences between neural processes and
computations (which are not discussed either in my review articles or just
about anywhere else in the philosophical literature), why they don’t work as
they stand, and a promissory note on how they can be improved upon by employing
the mechanistic account of computation.
(This paper is an expansion of a section of “Symbols, Strings, and
Spikes”.)
The mechanistic account of computing mechanisms
can be used to formulate computational theories sufficiently precisely as to
make them testable on the grounds of evidence from neuroscience. When
this is done, I believe there is no (nontrivial) digital computational theory
that survives the empirical test. I give a brief analysis of the notion
of computational explanation in neuroscience in Computational
Explanation and Mechanistic Explanation of Mind (in Cartographies of the
Mind: The Interface between Philosophy and Cognitive
Science, edited by de Caro et al., 2007). A more complete discussion
of this topic is in Neural Computation and the
Computational Theory of Cognition, co-authored with neuroscientist Sonya Bahar. The conclusion is that digital forms of
computationalism do not fit the scientific evidence. Since digital
computational theories have been the mainstream over the last decades, we need
to consider the possibility that generally speaking, neural computation is sui
generis and requires its own mathematical theory.
At this point, some
philosophers will be tempted to reply that my attempt to use neuroscience to
test computational theories is wrong-headed, because computational processes
are at a different level than neural mechanisms. According to this reply,
computational theories are psychological descriptions, and psychology is
autonomous from neuroscience. I agree that psychology is autonomous from
neuroscience in many respects, but it doesn’t follow that computational theories
are not testable by neuroscience. In fact, any nontrivial computational
theories of cognition are committed to appropriate mechanisms that realize the
computations. If those mechanisms are not found in the brain, then
computationalism is refuted, at least for biological organisms. For more on how psychological explanations
are just sketches of mechanisms, see “Integrating
Psychology and Neuroscience: Functional Analyses as Mechanism Sketches”
(with Carl Craver, forthcoming in Synthese).
Intentionality
I’m very interested in
intentionality. One of the limitations of many current theories of mental
content is that they are conjoined with computational theories of mind, where
computation is understood as in the semantic view of computation. But as
I’ve argued in some of the papers mentioned above, computation per se does
relatively little to advance our understanding of intentionality, and the
computational theory of mind needs to be rejected anyway. I think freeing
our thinking about mental content from the baggage of the computational theory
of mind paves the way for making progress in our theory of content. We
can find useful resources in cognitive neuroscience.
As a preliminary step, I have written Splitting
Concepts (co-authored with Sam Scott, in Philosophy of Science,
2006, followed by Edouard Machery’s
response, How to
Split Concepts), where I argue that the traditional notion of concept might
need to be split into several different notions, each of which explains
different phenomena. (There is not really any cognitive neuroscience in
this paper, though.)
A follow-up to
the above paper is “Two Kinds of
Concept: Implicit and Explicit,” forthcoming in Dialogue in a symposium on Edouard Machery’s book Doing
without Concepts. In this paper I
revise, articulate, and defend the view (originally proposed in the paper just
above) that concepts split into two kinds, which I now call implicit and
explicit.
I have also written Recovering What Is Said with Empty Names (co-authored
with Sam Scott, in Canadian Journal of
Philosophy, 2010). Empty names, such as ‘Santa Claus’, are names that
lack a referent. The paper offers robust evidence, based on semantic
intuitions elicited under appropriate circumstances, to the effect that empty
names have meaning. Such evidence refutes the many semantic theories (Millianism) that deny meaning to empty names.
Consciousness and
Introspection
I’m also interested in
consciousness. Here too, I think many philosophers have underestimated
the potential for progress that would come from paying serious attention to
neuroscience. For now, most of what I’ve written in this area pertains to
the methodology of using introspective reports in science and the related topic
of whether scientific methods should be public.
In Epistemic
Divergence and the Publicity of Scientific Methods (in Studies in Hist and Phil of Sci, 2003),
I revisit the venerable methodological principle that scientific methods ought
to be public by offering a plausible formulation of the principle and defending
it against a recent attack by Alvin Goldman. Goldman’s attack was based
in part on the view that the use of introspective reports in science
constitutes a private method.
In Data from Introspective Reports: Upgrading from Commonsense to Science (in J Consciousness Studies, 2003), I articulate and defend a middle way between Daniel Dennett’s “heterophenomenology” about introspective reports (scientists should be neutral as to their truth value) and Alvin Goldman and David Chalmers’s “first-person science” (scientists should trust introspective reports). My alternative is that scientists can and should validate introspective reports by publicly available evidence and assumptions, in ways that I spell out in my paper. Curiously, both Goldman and Dennett now claim to agree with me. Goldman cites my article in his “Epistemology and the Evidential Status of Introspective Reports: Trust, Warrant, and Evidential Sources,” Journal of Consciousness Studies, 2004, 11.7-8, pp. 1-16. Goldman’s position in that article appears to be closer to mine than his previously published one, and he has told me in conversation that our views are now close. Dennett cites my article in his “Heterophenomenology