Orthodox cognitive science claims that situated (world-embedded) activity can be explained as the outcome of in-the-head manipulations of representations by computational information processing mechanisms. But, in the field of Artificial Life, research into adaptive behaviour questions the primacy of the mainstream explanatory framework. This paper argues that such doubts are well-founded. Classical A.I. encountered fundamental problems in moving from toy worlds to dynamic unconstrained environments. I draw on work in behaviour-based robotics to suggest that such difficulties are plausibly viewed as artefacts of the representational/computational architecture assumed in the classical paradigm. And merely moving into connectionism cannot save the received orthodoxy. If we adopt the perspective according to which neural networks are most naturally conceptualized as dynamical systems, it becomes appropriate to treat such networks as computational devices only if the network-dynamics are deliberately restricted. A different explanatory framework is required once artificial neural networks are developed both to exhibit dynamical profiles comparable to those displayed by biological neural networks, and to play the same adaptive role as biological networks, i.e., to function as the control systems for complete situated agents. I close by describing an example of a dynamical systems explanation of situated activity.
Download compressed postscript file