Mechanisms of learning through the actualization of discrepancies

Abstract

Adaptive systems, whether biological or artificial, rely on internal models to interact with their environment. This study investigates a learning mechanism driven by discrepancies between predictions and reality. A two-level computational system is analyzed: (1) passive pattern memorization and (2) active model correction. Key adaptive elements include fixed input-processing blocks (analogous to sensory channels), dynamic weight adjustments (memory-like), and a balance between model updating (learning acceleration) and stabilization. Memory plays a central role, with statistical data (*_tendency.csv) forming predictive foundations and an optimization algorithm refining them. Healthy adaptation requires equilibrium between plasticity and resilience. The framework demonstrates broad applicability, spanning AI and cognitive science. Unlike traditional views of memory as mere recall, this model emphasizes its dual role in both memorization and world-model formation, achieved through integrated memory functions. The results highlight memory’s potential as a core adaptive mechanism, bridging machine and biological learning. This approach advances AI development while offering novel insights into natural cognition, underscoring the parallels between artificial and biological adaptive systems.