arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2502.04259cs.HCcs.CY
收藏

认知AI框架2.0:人类思维模拟的进展

Cognitive AI framework 2.0: advances in the simulation of human thought

Albizu University(阿尔比祖大学);AGM University School of Professional Studies and Cyberpsychology(AGM大学专业研究与网络心理学学院)

Rommel Salas-Guerra

AI总结:

认知AI框架2.0通过统一的记忆架构和受控的知识更新,提升人机交互的个性化与适应性,解决可扩展性、偏见缓解和伦理合规等挑战。

摘要:

人类认知模拟框架提出了一种受控的认知AI架构,旨在提高人机交互中的个性化、适应性和长期一致性。该框架整合了短期记忆(对话上下文)、长期记忆(交互上下文)、认知处理模块和受控的知识持久性,形成一个统一的架构模型,确保跨会话的上下文连续性和相关信息的受控积累。一个核心贡献是受显式治理机制监督的统一记忆架构,包括算法相关性验证、选择性持久性和可审计性。该框架整合了针对逻辑、创造性和类比推理的不同处理模块,使系统既能执行结构化任务,又能进行复杂的上下文推断。通过动态和选择性的知识更新,系统在不修改大型语言模型内部参数的情况下增强其能力,依靠检索增强生成和受控的外部记忆。所提出的架构通过在认知循环中嵌入操作安全机制,解决了可扩展性、偏见缓解和伦理合规等关键挑战。这些机制为未来持续学习、可持续性和多模态认知交互的研究奠定了基础。本文是之前发布的预印本(DOI:10.48550/arXiv.2502.04259)的大幅修改和扩展版本。

英文摘要:

The Human Cognitive Simulation Framework proposes a governed cognitive AI architecture designed to improve personalization, adaptability, and long-term coherence in human AI interaction. The framework integrates short-term memory (conversation context), long-term memory (interaction context), cognitive processing modules, and managed knowledge persistence into a unified architectural model that ensures contextual continuity across sessions and controlled accumulation of relevant information. A central contribution is a unified memory architecture supervised by explicit governance mechanisms, including algorithmic relevance validation, selective persistence, and auditability. The framework incorporates differentiated processing modules for logical, creative, and analogical reasoning, enabling both structured task execution and complex contextual inference. Through dynamic and selective knowledge updating, the system augments the capabilities of large language models without modifying their internal parameters, relying instead on retrieval augmented generation and governed external memory. The proposed architecture addresses key challenges related to scalability, bias mitigation, and ethical compliance by embedding operational safeguards directly into the cognitive loop. These mechanisms establish a foundation for future work on continuous learning, sustainability, and multimodal cognitive interaction. This manuscript is a substantially revised and extended version of the previously released preprint (DOI:10.48550/arXiv.2502.04259).

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑