Models of social simulation based on artificial intelligence represent a significant advancement in the ability to generate plausible agentic behaviors in complex social contexts. However, the present analysis argues that these systems operate at a fundamentally different level from that of genuine social prediction. The crucial distinction does not lie in the computational quality of the models, but in the epistemic nature of what such models can and cannot access: the internal generative structure of human systems. Drawing on the theoretical frameworks of predictive processing, embodied cognition, and complex systems theory, this paper argues that behavior-based simulation grounded in observable data is not, within current modeling approaches, capable of adequately representing phenomena such as identity transformation, psychological discontinuity, and the non-linear reorganization of the self. The body as a dynamic physiological constraint, interoception as a generative signal of identity, and the possibility of rupture in the narrative frame constitute dimensions of human psychology that are not fully accessible within current architectures based on observable behavior. The paper concludes by introducing the IRAR model (High-Resonance Relational Trigger) as a theoretical framework operating at the level of structural psychological transformation, thereby conceptually addressing the epistemic limitation identified here. It follows that social prediction, insofar as it aims to account not for behavioral regularity but for the transformations that disrupt and reconfigure it, cannot rely exclusively on the modeling of observable behavior, but requires access to the processes of transformation operating at the level of identity structure.
Ginevra Pigliacelli (Tue,) studied this question.