Empathy for Artificial Agents

Abstract

The paper has three goals. First, it introduces into different notions of empathy and related capacities such as emotional contagion, affective empathy, cognitive empathy, and sympathy. Second, it presents a case in point of an intelligent tutoring system, Affective AutoTutor, whose affect-sensitive behavior seems to further and enhance the outcome of its interactions with its students. Affective AutoTutor appears to behave empathically within a well defined learning environment. Third, attention is directed towards the requirements to be met by artificial empathizers to be judged as empathizers tout court by their social interactants, even when acting in unspecified social situations. To be a convincing empathizer, the artificial agent would not only need to grasp the emotional states of its interaction partners and understand their situation with respect to an adequate world model, but also communicate its own affective states. Eventually, an artificial empathizer should be ready to react appropriately to its interaction partner's reciprocal empathy.