Current AI systems lack genuine autonomy—the capacity for spontaneous, self-directed behavior that arises from internal processes rather than external prompts. We present ComPact, a functionalist architecture for personality and emotion modeling in AI agents, inspired by the traditional Chinese concept of Three Souls and Seven Spirits (三魂七魄). The system comprises four layers: (1) an embodiment layer interfacing with the physical world through sensors and actuators with interoceptive self-monitoring; (2) seven emotional value functions (Seven Spirits) that bidirectionally gate information flow—coloring inputs with emotional valence and being shaped by outputs; (3) a tripartite decision layer (Three Souls) where three independently-trained agents with distinct cognitive orientations—intuition, strategy, and desire—engage in perpetual coupled interaction through shared tensor space, producing output only upon convergence; and (4) a memory layer where selective forgetting serves as the core mechanism of personality solidification. We argue that this architecture produces structured unpredictability—behavior that is neither random nor deterministic—as an emergent property of coupled tripartite dynamics, analogous to the three-body problem in physics where deterministic rules yield chaotic yet structured trajectories.
Zichao Zhou (2026) studied this question.