Altruistic behavior and recursive estimation of others’ internal state

Abstract

To reveal what kind of mental model enables an agent to learn altruistic behavior from trial-and-error experiences, a series of computer simulation experiments were performed. We found that high probability of mutual cooperation is achieved by the agents that choose actions based on estimated internal states of others, compared to those that choose actions based of predicted action of others. Mutual cooperation is further promoted by introducing recursive estimation structure of others and reciprocal embedding structures.