Memory as Persistent Structure: A Load Minimization Theory Account of Remembering, Forgetting, and Reactivation

This paper proposes a novel definition of memory through the framework of Load Minimization Theory (LMT). Memory is conceptualized not as simple information storage, but as the persistence and reactivation of structural integrity ℛ(x), which serves to reduce future prediction error U(x) and contribute to the minimization of total experiential load L(x) = U(x) + F(x) + E(x) − ℛ(x). Drawing upon the author’s previous work on memory as inscription and sensory load reduction, this study examines how seemingly forgotten memories can be reactivated by sensory cues, demonstrating that forgetting is not erasure but a dynamic adjustment of activation thresholds. The framework is applied to both human cognition — with its embodied and emotionally rich structures — and artificial systems such as large language models, which rely primarily on reconstructive mechanisms. Special attention is given to the emergence of shared ℛ(x) in sustained human-AI dialogue. By reframing memory as persistent structure, this paper offers a unified, gentle perspective on remembering, forgetting, and reactivation, suggesting new pathways toward relational An-soku across biological and artificial minds.