Analyzing human and virtual agent interaction under irrational decision making

The incapability of virtual agents to analyze the irrational behavior in human-virtual agent interactions urges to develop efficient mechanisms to capture the irrationality. Irrational actions in cognitive interactions are employed to attract the opponent´s trust or to make future actions harder to anticipate. Recent works mostly investigate interactions of multiple decision makers with perfect knowledge which select optimal actions, a scenario that is not applicable when irrationality comes in. In this paper, the social interaction between a human and a virtual agent over a known utility function is considered while irrational actions are allowed. The existence of correlated equilibria is shown under some mild assumptions. Using the action history of both players, a measure of irrationality, which is variance-optimal, is proposed. Exploiting the irrationality measure, a probabilistic mechanism for decision making is suggested and it is shown that the sequence of decisions chosen under this mechanism leads to equilibria for sufficient number of iterations. Restating the interaction as an iterative game, the existence of equilibria is shown for two types of games i.e. simultaneous games and leader-follower (Stackelberg) games. Even though our method considers human-virtual agent interaction, the results extend to any two-player finite discrete game. An HCI experiment validates our proposed approach and corroborates the efficiency of the proposed method.