Probabilistic models of cognition: where next?

Abstract

This Special Issue surveys the state of the art of probabilistic models across a broad range of topics in cognitive science. We suggest that the present shift towards probabilistic methods has deeper origins: viz., conceptual and technical developments in probability theory and statistics that provide the machinery to engage with cognitively relevant information-processing problems. These technical developments provide a rich range of models, tools and metaphors with which to reconceptualize cognition. Moreover, the application of these probabilistic ideas to relevant engineering problems, in speech and image processing, expert systems, robotics and machine learning, has provided a rich source of insights into some of the probabilistic reasoning problems solved by the brain. Here, we highlight some of the key developments that drive current work, and consider future technical and empirical challenges. We divide our discussion into three, interlinked, domains: representation, processing and learning, before drawing conclusions concerning the prospects for the field.