This paper investigates the fundamental question of how an AI system "knows." We propose a "3 by 5" memory architecture that integrates three distinct storage mechanisms—implicit, explicit, and agentic—with the five functional modules of the Wuxing framework (Exploration, Reasoning, Knowledge, Rules, and Adaptation). By drawing on Polanyi’s tacit knowledge and Ryle’s distinction between "knowing-how" and "knowing-that," we demonstrate how this architecture manages the essential trade-offs between static knowledge storage and dynamic, adaptive action. 本文探讨了人工智能系统“如何知晓”的基础性问题。通过引入“隐性(Implicit)”、“显性(Explicit)”与“代理(Agentic)”三种存储机制,并将其与五行功能模块相结合,构建了一个“3乘5”的内存架构。该研究借鉴了波兰尼的“默会知识”与赖尔的“知其然/知其所以然”理论,为 AI 在固定知识存储与动态环境适配之间的权衡提供了结构化方案。
Rui Chai (Tue,) studied this question.