---

TALK: “The emotional underpinnings of self-motivated learning”

Abstract
While the importance of emotions in humans cannot be underestimated, their potential in artificial agents (AA) has not been realised in every respect. Many studies of emotions in AA provide merely an interface to communicate emotional expressions in human-agent interaction. In other cases, emotions in AA are conceptually indistinguishable from the evaluative signals that are used in standard reinforcement learning. We will argue that the interaction of an emotional dynamics with learning processes can provide AAs with the function of an intrinsic and essential emotion-equivalent subsystem which can enable additional capabilities such as active learning, transfer learning, theory of mind, and meta-learning. This proposal will be discussed based on current work on affective computing and companion robots. We will conclude that, although emotions in AA are promising as models for human emotions, a general theory of the latter and thus a proper evaluation of the former is, however, still beyond reach.

---

Short Bio

Dr. J. Michael Herrmann is a Lecturer in the School of Informatics at the University of Edinburgh. He studied mathematics and obtained his PhD in computer science at Leipzig University, focusing on neuronal networks. Over the years, he has held research positions in Germany, Israel, Denmark, and Japan, including appointments at the Max Planck Institute for Fluid Dynamics, NORDITA, and RIKEN. His research spans nonlinear dynamics, neural computation, and information representation. Since 2008, he has been with the University of Edinburgh, where he continues to work at the intersection of informatics, complex systems, and artificial intelligence.