Claire Yang


Representing Altruism in Multi-agent Theory of Minds

Research on multi-agent theory of minds assumes that each agent is motivated by the maximization of either their own utility, or their team's collective utility. There is a significant gap in our understanding of how to represent perceived unselfish behavior in social interactions with multiple people and objectives. We address this gap by utilizing a multi-agent theory of minds framework and three-player stag hunt game to simulate a situation where social cooperation can only be inferred from other agent's actions. In addition to normal game actions, each agent has actions that allow them to advertise and elicit social preferences.  Agents can declare their altruistic intentions towards any other agent, without committing to altruistic gameplay. To elicit social preferences, each agent has the ability to make any other agent play the dictator game. The dictator game allows the agents to signal whether they are purely self-interested, or actually altruistic at a cost to the agent. Preliminary results without the dictator game show that higher level agents are more likely to declare altruism, and that an agent's altruistic declarations had a notable effect on the inferred teams and the other agents' following actions. We aim to utilize this environment to conduct experiments with depressed and nondepressed participants, in order to better understand how depression affects the perception of social cooperation and altruism. 

Video file