Summary
The preprint introduces the Xzistor Mathematical Model of Mind (XMMM), a unified control-theoretic architecture that explains how emotion, cognition, and adaptive behavior emerge from a single principle: the brain (or artificial agent) minimizes homeostatic and allostatic deprivation, which is experienced as body-located somatosensory emotions. The architecture consists of five interoperating algorithms—Sensing, Drive, Reflex, Association, and Motion—coordinated by a Linking Algorithm acting as the executive controller.
Raw physiological error signals are transformed into valenced Deprivation or Satiation Emotions via a Body Map, which serves as the sole motivational input to the executive. This design enables intrinsic reinforcement learning (Reward-based Backpropagation), goal-directed navigation, fear conditioning, social bonding, and flexible problem-solving through “Directed Threading” — all without externally specified reward functions or pre-programmed behaviors.Implemented and tested in both a virtual agent and a physical robot, the XMMM demonstrates emergent behaviors such as navigation gradients, daydreaming, and inductive inference.
A parametric lesion study cleanly dissociates reactive associative behavior from deliberative cognition. The model shows strong post-hoc convergence with biological systems, including the insular cortex (Body Map), hypothalamus (Drive Algorithm), hippocampus (Association), and prefrontal-basal ganglia circuits (Linking Algorithm). It also offers five specific, falsifiable predictions and outlines a scalable neuro-symbolic hybrid architecture for achieving artificial general intelligence while maintaining alignment by construction through its homeostatic drive system.