Artificial social reasoning: computational mechanisms for reasoning about others

Abstract

With a view to supporting expressive, but tractable, collaborative interactions between humans and agents, we propose an approach for representing heterogeneous agent models, i.e., with potentially diverse mental abilities and holding stereotypical characteristics as members of a social reference group. We build a computationally grounded mechanism for progressing their beliefs about others’ beliefs, supporting stereotypical as well as empathic reasoning. We comment on how this approach can be used to build finite-state games, restricting the analysis of possibly large-scale problems by focusing only on the set of plausible evolutions.