Answers are grounded in this paper and cite their sources.
Memory is a fundamental capability for maintaining continuity, retaining experience, accumulating knowledge, and supporting intelligent decision-making. In large language model (LLM)-based systems, memory mechanisms have increasingly evolved beyond temporary conversational context to include persistent individual memory, external memory repositories, multi-agent memory sharing, and human-assisted knowledge retention. However, these mechanisms are commonly investigated through separate architectural and operational perspectives. Individual model memory, shared memory among multiple models, and human memory participation are not necessarily integrated into a general framework that determines when each memory configuration should be used, how configurations should transition, and how multiple memory sources should be coordinated. This paper introduces the Adaptive Self–Shared–Human Memory Framework (ASSHMF), a unified theoretical architecture for integrating three fundamental memory dimensions: Self-Memory, in which an individual artificial intelligence model retains, retrieves, organizes, and updates its own historical information; Shared Memory, in which multiple artificial intelligence models access, contribute to, maintain, and utilize a common or coordinated memory infrastructure; and Human Memory, in which human individuals or groups contribute their own remembered experiences, knowledge, contextual understanding, and recollections to memory-dependent processes. The framework extends the conceptual foundation of the Universal Self-Mutual-Human Model (USMHM) by applying its general participant organization and adaptive operational principles specifically to memory. Its central contribution is not simply the coexistence of three memory types, but the establishment of a mechanism-independent architecture supporting Universal Adaptive Memory Selection, Adaptive Memory Switching, Adaptive Memory Combination, and Adaptive Memory Reconfiguration. An intelligent system can therefore determine whether a particular task should rely on its own memory, retrieve information from shared memory, request human recollection, or combine these sources. It can dynamically change memory configurations according to information relevance, reliability, uncertainty, freshness, accessibility, privacy, task requirements, participant capabilities, and resource constraints. The framework further distinguishes memory ownership from memory accessibility, memory retrieval from memory integration, memory updating from memory synchronization, and memory sharing from unrestricted information disclosure. A formal representation, adaptive decision model, memory lifecycle, transition architecture, governance principles, theoretical propositions, and evaluation methodology are developed. To the best of our knowledge, this paper is the first to explicitly formulate a unified, mechanism-independent Self–Shared–Human memory architecture that integrates adaptive selection, switching, combination, and reconfiguration across artificial intelligence models and human participants. The framework establishes a general theoretical foundation for designing and evaluating memory systems that can dynamically move beyond dependence on a single memory source.
No takes yet. Share an insight, caveat, or question.
Jincheng Zhang (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: