The received opinion in mainstream cognitive science is that cognition is the manipulation of representations by computational information processing mechanisms. However, in this paper, I argue that cognition ought to be conceptualized not as computation, but as state-space evolution in certain classes of non-computational dynamical systems. Following a review of van Gelder's arguments for this dynamical cognition hypothesis, I suggest that the most compelling reasons for adopting a dynamical perspective come from research into situated (world-embedded) autonomous agents. We should think of cognitive architectures as control systems for the situated activity of sufficiently complex embodied agents. I draw on key work in the simulation of adaptive behaviour which indicates that situatedness is best explained using the concept of dynamical coupling. I then present a preliminary sketch of other ways in which the conceptual framework of dynamics provides a powerful basis for thinking about and studying cognition.
Download compressed postscript file