Beyond Optimization: Non-Optimal Internal State Dynamics as a Basis for Artificial Personality

This paper proposes a theoretical framework that conceptualizesartificial personality not as a set of optimized traits, symbolic representations, or learned parameters, but as an emergent property arisingfrom non-optimal, time-varying internal state dynamics. We introducethe concept of the Machine Endocrine System (MES)—a persistentregulatory layer that biases cognitive execution at runtime without encoding explicit goals or utilities. Unlike conventional AI architecturesthat pursue task-specific optimality, the MES introduces structured,history-dependent variability that may reduce short-term performancewhile enabling long-term behavioral coherence. Crucially, this framework does not perform automated personality recognition or aestheticjudgment. Instead, it externalizes traces of internal state transitions,allowing human observers to infer individuality as the “shadow” castby these dynamics on perception and action. [Update: 2026-01-19] Added "Supplementary Note: On Demonstration Design for Evaluating Execution Dynamics in Musical Performance". This note formalizes the conditions for meaningful demonstrations of the MES framework, emphasizing the role of human inference and information loss.