Challenges in Modeling the Dynamics of Emotions

Eva Hudlicka (, President of Psychometrix Associates, Inc.

Emotions represent a critical component of adaptive, intelligent behavior. Neuroscience and psychological research over the past 15 years has identified a wide variety of roles that emotions play in adaptation, homeostasis, learning, decision-making, and in social behavior. Researchers in cognitive science and artificial intelligence have developed a number of computational models of emotions, incorporating the emerging data. These are used for a variety of research and applied purposes: to elucidate the mechanisms of affective processes, to improve our understanding of cognitive-affective interactions, and to enhance the adaptive behavior and realism of autonomous agents and robots. While much progress has been made in understanding and modeling emotions, many difficult problems remain. Some of the outstanding challenges include modeling the interactions among multiple modalities of emotions (e.g., cognitive, physiological, behavioral), capturing the nature and function of the affective dynamics (the variabilities in emotion intensities over time), and understanding the relationships among multiple, co-existing emotions at a given point in time. To make further progress in understanding the nature of these complex phenomena, we need empirical data about the affective dynamics, as well as new formalisms and modeling approaches capable of representing the parallel processes underlying affective phenomena, and the interactions among them. In this talk I will first discuss some of the key recent findings about emotions from psychology and neuroscience. I will then outline some of the computational tasks necessary to model these phenomena, and describe representative models implementing these tasks. I will conclude by highlighting the key challenges in developing computational models of these complex processes, both in terms of the necessary data and the required formalisms.