On the nature of explanation: A PDP approach

Abstract

Neural network models of sensory processing and associative memory provide the resources for a new theory of what explanatory understanding consists in. That theory finds the theoretically important factors to reside not at the level of propositions and the relations between them, but at the level of the activation patterns across large populations of neurons. The theory portrays explanatory understanding, perceptual recognition, and abductive inference as being different instances of the same more general sort of cognitive achievement, viz. prototype activation. It thus effects a unification of the theories of explanation, perception, and ampliative inference. It also finds systematic unity in the wide diversity of types of explanation (causal, functional, mathematical, intentional, reductive, etc.), a chronic problem for theories of explanation in the logico-linguistic tradition. Finally, it is free of the many defects, both logical and psychological, that plague models in that older tradition.