Lecture Note 3

Embodied Empathic Agents

Written in 2016/11/11, Pittsburgh

This lecture is about embodied empathic agents and its application on tutoring system. It was held by Ruth Aylett, the professor of Computer Science at Heriot-Watt University.

First of all, she gave us a definition about empathy that empathy fosters emotional engagement and can result in empathic relations between a user and an agent. There are two ways to look at empathy, one is reactive empathy, which is experiencing a different emotion from others, and one is parallel empathy, which is experiencing the same emotion with others. Accordingly, there are also two types of empathy: cognitive and affective. Cognitive empathy means understanding or knowing the emotional state of another, while affective empathy means experiencing and share an emotion of another. Usually, people mediate empathy via the situation or via emotional expressions.

Then, Ruth talked about the empathic agents. There are two ways to look at empathic agents: users experiencing empathic emotion and agent showing empathy towards users and other characters. Ideally, a positive empathetic cycle can be created between users and agents, but it is very hard to achieve.

After introducing the empathic agents, Ruth moved forwards and talked about theories and architectures. She introduced the theories of affect, theories of personality, and 3 models, which are Ekman’s Primitive Emotions, Russel’s Classification and Generating Appropriate Behaviors: Cognitive Appraisal Theory. The last model was used mostly in their recent project — EMOTE. It is a model that tells how people cope with external events and internal goals and beliefs, there are basically two ways to do that: reacting to the event or suppressing the emotion inside.

EMOTE is an empathic robot tutor that can respond to the emotional states of a child empathically and form a socio-emotional bond with the child. It was built based on the fact that socio-emotional bonds between teachers and learners have been shown to motivate the learning processes. They carried out an experiment to see if there is any difference in learning gain in empathic compared to non-empathic situation, and the results showed there is no significant difference. This is probably because the subjects were asked to learn easy materials instead of complicated ones.

Finally, the conclusion drawn by Ruth is that building empathy-invoking agents is certainly feasible but building empathy-displaying agents is much harder because questions like how we establish the user affective state need to be answered.