Toward a Theoretical Foundation for Multi-Agent Coordinated Decisions

Second International Conference on Multiagent Systems

Volume

Second International Conference on Multiagent Systems

Contents

Epistemic utility theory, originally developed to model single-agent cognitive decision-making, is extended to the multi-agent case and applied to practical coordinated decision-making. For an M-agent society, individual and multiple agent coordination information is contained in a 2M-dimensional coordination space. The coordination function is a joint probability mass (or density) function defined over the coordination space that characterizes: (a) each individual agent’s goals and standards; (b) each individual agent’s abductive considerations, such as cost, hazard, or resource consumption; and (c) information pertaining to the interactions within the society, such as cooperation, competition, exploitation, compromise, tolerance, and indifference. The coordination function is used to derive two joint utilities, termed accuracy and rejectability, and a joint version of Levi’s rule of epistemic utility is applied to identify the satisficing set of decision vectors that represent jointly rational behavior for the society.