Distributed Connectionist Models in Social Psychology

Distributed connectionist models of mental representation (also termed PDP or parallel distributed processing, or ANN or artificial neural networks) constitute a fundamental alternative to the associative or schematic models that have been much more prevalent in social psychology. A connectionist model is made up of a large number of very simple processing units, richly interconnected and able to send signals to each other depending on their momentary activation levels. No individual processing unit represents a meaningful concept; instead, overall patterns of activation hold representational meaning. This article emphasizes the novel properties of connectionist representation that might appeal to theorists and researchers in social psychology, including their context sensitivity and flexibility, ability to represent prototypes and exemplars within a single network, and ability to determine whether a stimulus is familiar even before the stimulus can be identified or categorized. Virtually every theory in social psychology makes reference to mental representations, including beliefs, attitudes, stereotypes, person impressions, autobiographical memories, or the self-concept. Therefore, the assumed properties of mental representations fundamentally influence the predictions that will be made by our theories. Social psychologists have often unquestioningly adopted a standard set of assumptions about the nature of representations (such as associative network or schema models; Smith, 1998), assumptions that are pleasantly straightforward, familiar, and intuitive. This article describes an alternative, distributed connectionist models of representation, and discusses some of their less intuitive properties, with the goal of encouraging social psychologists to explore the implications of these models for their theories. As I will argue, connectionist models can explode misleading dichotomies, unify seemingly distinct types of processing, and naturally generate properties (such as the flexible, contextsensitive nature of representations) that are often found in social psychological research. The properties of connectionist models (also termed PDP or parallel distributed processing, or ANN or artificial neural networks) are explored