What Reinforcement Learning can tell us about Emotions

Date:
Friday January 26th, 2024, 16:00 CEST

Author:
Joost Broekens, Leiden University, the Netherlands

Abstract:
Emotions are tied to appraisal of personal relevance, motivation and adaptation of behavior. Many animals show signs of emotion in their behavior. Therefore, emotions must be related to mechanisms that aid survival, and emotions must be evolutionary continuous phenomena. I propose that emotions are manifestations of Temporal Difference Reinforcement Learning (TDRL) error assessment. The TD error reflects the estimated gain or loss of utility – well-being – resulting from new evidence. I propose a TDRL Theory of Emotion, and discuss recent computational findings to investigate this.

Bio:
Joost Broekens is associate professor and head of the Affective Computing and Human-Robot Interaction lab at the Leiden Institute of Advanced Computer Science (LIACS), Leiden University. He is president emeritus of the Association for the Advancement of Affective Computing (AAAC). He is co-founder of Interactive Robotics and co-founder of Daisys. His research focuses on affective computing, in particular computational modelling of emotions in reinforcement learning and computational models of cognitive appraisal and on human-robot interaction.