Predicting Agents' Behavior by Measuring their Social Preferences

There are many situations in which two or more agents (e.g., human or computer decision makers) interact with each other repeatedly in settings that can be modeled as repeated stochastic games. In such situations, each agent’s performance may depend greatly on how well it can predict the other agents’ preferences and behavior. For use in making such predictions, we adapt and extend the Social Value Orientation (SVO) model from social psychology, which provides a way to measure an agent’s preferences for both its own payoffs and those of the other agents. The original SVO model was limited to one-shot games, and assumed that each individual’s behavioral preferences remain constant over time—an assumption that is inadequate for repeated-game settings, where an agent’s future behavior may depend not only on its SVO but also on its observations of the other agents’ behavior. We extend the SVO model to take this into account. Our experimental evaluation, on several dozen agents that were written by students in classroom projects, show that our extended model works quite well.