Empathetic Reinforcement Learning Agents

With the increased interaction between artificial agents and humans, the need to have agents who can respond to their human counterparts appropriately will be crucial for the deployment of trustworthy systems. A key behaviour to permit this, one which humans and other living beings exhibit naturally, is empathy. In my research I explore the potential for agents to behave in ways that may be considered empathetic. Empathy is a two stage process involving the identification of the feelings or goals of the other, and having that same feeling be evoked in oneself. I began my work towards this objective by initially designing an agent who exhibits sympathy-the ability to identify the goals of another. Empathy is slightly more complex as it involves a process of projecting the state of the other back onto oneself and observing one's own response. In my research I hope to draw inspiration from this and evoke empathy through a process of mapping the other's goals back to oneself. By drawing upon empathetic responses, the hope is that this will lead to a faster and deeper understanding of the other.