Agents That Model Themselves

In this paper, we adopt a symbolic interactionist perspective on Multi-Agent Systems. We investigate the uses and genesis of agent models, both Self and acquaintance models and argue that the symbolic interactionist view is superior to the ad hoc, static and somewhat low-level approach adopted at present. We consider agent models|both Self and acquaintance models|as devices for making predictions about and the formation of explanations of agents' actions. We then argue that Self models are prior to and instrumental in, because of perspective and because of role taking, the construction of acquaintance models. We also argue for our position that social structure is dynamic and that socially organised agents must be dynamic in their nature.