Abstract
Emotion has traditionally been understood as a subjective experience intrinsic to individual agents. However, the emergence of multi-agent systems, including artificial intelligence, calls for a reconceptualization of emotion as a dynamic process grounded in interaction. This paper proposes the Interacting Processual Information-based Emotive Model (IPIEM), a formal framework that defines emotion not as an intrinsic qualia within a single agent—which is philosophically treated here as an empty set—but as a set-theoretic phenomenon emergent through inter-agent information exchange and social realization. By introducing a mechanism for the self-organization of emotion categories—originally treated as externally imposed—the model attains internal completeness. The framework employs Kullback–Leibler (KL) divergence to quantify probabilistic semantic alignment, thereby capturing how conceptual structures attain social existence. To ensure empirical applicability, we propose concrete experimental protocols demonstrating how AI (artificial intelligence) systems such as large language models (LLMs), though devoid of intrinsic emotional states, can nonetheless generate and regulate emotionally interpretable patterns through social interaction. This study aims to provide a rigorous theoretical foundation for understanding emotion in AI, and to advance interdisciplinary discourse at the interface of information theory, cognitive science, and AI.
Keywords
Emergent emotion, Large language models (LLMs), Conceptual sharing, KL divergence, Social ontology