Chris Eliasmith
University of Waterloo
Depts. of Philosophy and Systems Design
Engineering
Since its inception,
the theoretical terms of ‘continuity’, ‘dynamics’, and ‘representation’ have
been central to work in cognitive science. The relation between these terms is
an interesting one, with various combinations of commitment to them
representing the major approaches to understanding cognitive systems. Consider,
as three central exemplars, symbolicism, connectionism, and dynamicism.
Symbolicism (i.e., ‘classical’ cognitive science) can be characterized as a
rejection of the relevance of continuity, a dismissal or downplaying of
dynamics, and a strong commitment to representation (Fodor and Pylyshyn 1988; Newell 1990). Connectionism generally entails an acceptance of the
importance of all three (Smolensky
1988; Churchland 1995). Lastly, dynamicism consists of a strong emphasis on
continuity and dynamics and argues from those, to a rejection of representation
(Thelen and Smith 1994; van Gelder
1995).
In this paper, I describe a number of recent results from work in computational
neuroscience that helps shed light on the nature of, and relation between,
these three terms. I begin by presenting evidence that neural systems
themselves can only be properly characterized as discrete systems (i.e. Turing
machines). This blocks dynamicist arguments from the purported continuity of
neural systems to anti-representationalism and anti-computationalism (van Gelder 1998), and connectionist arguments from continuity to
special (i.e. non-Turing describable) computational properties of brains (Churchland 1995). Nevertheless, I argue that the kind of discreteness
found in neural systems does not support the symbolicist view of cognitive
systems either. In particular, it does not establish the centrality of symbolic
psychological-level representation nor does it provide a defense against
the strong dynamicist criticisms of symbolicist ‘atemporal’ (i.e. non-, or
anti-dynamic) commitments.
By further examining these and related results from computational neuroscience,
I present a positive view of neural function which excludes continuity, but unifies dynamics and
representation. On this view, which I call Cognitive Neural Control Theory
(CNCT), representation is rigourously defined by encoding and decoding
relations which can hold at various
levels of description. The variables identified at higher levels can be
considered state variables in control theoretical descriptions of neural
dynamics. I argue that, given the generality of control theory and representation
so defined, this approach is sufficiently powerful to unify descriptions of
cognitive systems from the neural to
the psychological levels. CNCT thus avoids the lack of a proper
dynamical characterization of cognitive systems characteristic of symbolicism,
and shows how, contrary to dynamicist arguments, representation and dynamics
can be part of a consistent approach in cognitive science.
Churchland, P. (1995). The engine of reason, the seat of the soul: a philosophical journey
into the brain. Cambridge, MA, MIT Press.
Fodor, J. and Z.
Pylyshyn (1988). “Connectionism and cognitive architecture: A critical
analysis.” Cognition 28: 3-71.
Newell, A. (1990). Unified theories of cognition.
Cambridge, MA, Harvard University Press.
Smolensky, P. (1988).
“On the proper treatment of connectionism.” Behavioral
and Brain Sciences 11(1): 1-23.
Thelen, E. and L. B.
Smith (1994). A dynamic systems approach
to the development of cognition and action. Cambridge, MIT Press.
van Gelder, T. (1995).
“What might cognition be, if not computation?” The Journal of Philosophy XCI(7):
345-381.
van Gelder, T. (1998).
“The dynamical hypothesis in cognitive science.” Behavioral and Brain Sciences 21(5):
615-665.