Connectionism and the Dynamical Approach to Cognition

We saw in chapter 1 that connectionism has its roots in the 1940s, when ideas about the nervous system, computation, and intelligence came together synergistically. We briefly traced the path from formal neurons to perceptrons and noted that events in the 1960s led to a period in which these neural network models became overshadowed by symbolic models of intelligence. With the rise of connectionism in the 1980s neural networks regained prominence, and for two decades the two approaches have each sought to dominate cognitive science. This is part of a larger story. The energetic collaborations of the 1940s had multiple, sometimes intertwined strands, of which the best known is an emphasis on feedback and control in the cybernetics of Norbert Wiener and others. This required attention to the dynamics of change in time, an aspect of the functioning of systems which has often been ignored by connectionists (though not by Stephen Grossberg and his group, who continued to advance network research during the period when symbolic models were dominant; see Grossberg, 1982). In an even broader context, the dynamics of complex physical systems have been mathematically describable at least since Newton (building on work by Galileo) formulated his three laws of motion and developed the calculus to gain the ability to predict specific planetary configurations. New geometrical approaches to dynamics by Poincaré in the late nineteenth century prepared the way for the rise of Dynamical Systems Theory (DST) in the twentieth century. Initially applied primarily to physical phenomena such as eddies in a stream (Landau, 1944), by the 1980s DST was being extended to motor coordination 1995) and by the 1990s to the development not only of coordinated activity but also more cognitive capacities (see the 1994 book by Esther Thelen and Linda Smith). Although the mathematics of nonlinear dynamical systems can be daunting, DST has spawned compelling graphics that help to provide an intuitive grasp of key concepts (among the suggested readings at the end of the chapter, see especially the copiously illustrated introduction by Abraham and Shaw, 1992). The idea that cognition is dynamic and can best be understood using the tools of DST attracted increasing attention across the 1990s but is still somewhat of a frontier outpost in cognitive science. Much of the excitement was conveyed in Mind as Motion, a 1995 book originating in a 1991 conference which brought together many