Modeling of Emotional Effects on Decision-making by Game Agents

Abstract

Game AI agents today do not reflect the affective aspects of human behavior. In particular, game agents do not reflect the effects of human emotional state on an agent's decision-making behavior. In rare instances when emotional aspects are addressed in game agent architectures, such behavior tends to be ad hoc and not informed by an underlying theory of emotion, nor validated using actual data. This paper presents a new emotional game agent architecture that is based on an underlying theory of emotion and validated by limited experiments. This architecture manifests a range of emotional effects on game agent behaviors. The overall approach is informed by both appraisal and dimensional theories of emotion. The combination of these theories as underpinnings ensures that emotionally appraised concepts in memory are reflected in the emotional state of the agent, and that such correspondence produces realistic emotional effects on the agent's decision-making behavior. The approach is validated through a series of increasingly more sophisticated experiments, in terms of scenario complexity and methods employed. The results are correlated with human data from previous cognitive science experiments. The results show that “lightweight” intelligent agents based on the new game agent architecture can exhibit realistic emotional behavior in real-time decision-making situations encountered in games across various domains.