Modeling social intelligence through attachment-based learning

Publication Type:

Conference Paper


Proceedings of the Japanese Society for Artificial Intelligence Workshop on Social Intelligence Design, JSAI Press, Matsue, Shimane, Japan (2001)


artificial intelligence


Augmenting social intelligence is particularly challenging because it requires a nuanced, multidisciplinary understanding of human social dynamics. Unfortunately, building a comprehensive model of human interaction is an "AI complete" problem. Nevertheless, it is possible to develop representations of particular types of interaction that can serve as a theoretical foundation for a broad range of social intelligence systems.

This paper focuses on one class of interactions: social learning. Social activity is central to learning (see Vygotski, 1980). We develop and expand a theory of social learning based on a series of seminars given by professor Marvin Minsky at MIT. Professor Minsky argues that human attachments are essential to learning, a theory he will present in his forthcoming book, The Emotion Machine. We position his theory here with the goal of providing social intelligence researchers with a theoretical model on which to base new system designs.

A model of attachment-based learning is particularly germane to this task because it illuminates the social interactions that promote learning. Often models of mind (see Minsky, 1985, Piaget, 1990) focus on the self-organization of the mind, and do not provide a theory of how people (teachers, parents, role-models, etc.) can influence and enhance the mental restructuring of others. Minsky's theory of attachment-based learning begins to address this problem.