Comparison of affect-control theoretic agents and humans in the prisoner's dilemma

Abstract

Symbolic interactionist principles of sociology are based on the idea that human action is guided by culturally shared symbolic representations of identities, behaviours, situations and emotions. Shared linguistic, paralinguistic, or kinesic elements allow human to co-ordinate action by enacting identities in social situations. Structures of identity-based interactions can lead to the enactment of social orders that solve social dilemmas (e.g., by promoting co-operation). Our goal is to build an artificial agent that mimics the identity-based interactions of humans, and to compare networks of such agents to human networks. In this paper, we take a first step in this direction, and describe a study in which humans played a repeated prisoner's dilemma game against other humans, or against one of three artificial agents (bots). One of the bots has an explicit representation of identity (for self and other), and attempts to optimise with respect to this representation. We compare the human play against bots to human play against humans, and show how the identity-based bots exhibits the most human-like behaviour.