Decision Under Risk in Multi-Agent Systems

Abstract

Expected Utility Theory (EUT) is the standard decision-making methodology used under uncertain conditions, but it is incapable of fully capturing actual human decision-making behaviors when under risk. Kahneman and Tversky's prospect theory is an alternative to EUT to deal with uncertainty when under risk. However, incorporating it into multi-agent decision-making requires two significant extensions to the present theory. First, rather than immediately making a choice directly from the set of alternatives, an agent should use the prospect theory's value function to initially form an attitude concerning each possible alternative and likewise solve the corresponding decision problem with a preference order. Correspondingly, this preference order should not be constructed solely on the basis of the individual agent's own attitudes but with the attitudes of all agents which are relevant. This paper presents our work on these two extensions. Finally, we present several experiments that explore risk management behaviors of agents in social settings.