Associative Memory and Cumulative Learning in Artificial Intelligence

Abstract

Conceptual research paper — not peer reviewed. Persistent artificial intelligence should be evaluated by how accumulated experience changes later judgment. This paper develops that proposition through a critical synthesis of associative memory, evolving agent memory, feedback-based adaptation, and emotional context. The central proposal is to retrieve linked histories of experiences, claims, corrections, and consequences while preserving the distinction between personal significance and evidential reliability. Established research already addresses each major component, so the paper claims neither discovery of associative memory nor a demonstrated replacement for existing architectures. It specifies a candidate retrieval policy and a comparative pilot that could determine whether the proposed integration adds value beyond similarity retrieval and temporal memory. The pilot isolates correction handling, retrieval from indirect cues, resistance to unsupported updates, and contextually appropriate responses. No implementation, experimental results, or superiority claims are reported. The scientific contribution offered here is a reasoned synthesis, an explicit design specification, and a falsifiable evaluation plan whose architectural novelty remains subject to further review. AI assistance was used for literature discovery, development of the proposed specification and evaluation plan, and drafting and editing. The author retains responsibility for the final text.