近日，实验室博士生詹昊熹关于生成式智能体的研究被会议International Conference on Case-Based Reasoning 2025（ICCBR）录用。ICCBR是案例推理领域的顶级会议，被CCF推荐为B类国际学术会议。

标题：Case-Based Reasoning in Generative Agents: Review and Prospect

摘要：With the rapid pervasion of Generative AI across academia, industry, and society, researchers and the public alike have begun to envision the future of generative agents and their potential role in achieving Artificial General Intelligence (AGI). However, while representative generative AI algorithms have demonstrated impressive language capabilities, they still fall short in complex reasoning tasks. Recent findings in cognitive neuroscience suggest significant differences in the cognitive mechanisms underlying language and reasoning abilities within the human brain. In this paper, we adopt an interdisciplinary perspective to review relevant research and perspectives from psychology, cognitive science, psychiatry, and artificial intelligence. Building upon this review, we envision future generative agents with case-based reasoning at their core. We propose a technical framework for CBR-based generative agents and discuss potential research avenues.

随着生成式人工智能在学术界、工业界和社会中的迅速普及，研究人员和公众开始畅想生成智能体的未来及其在实现通用人工智能（AGI）中的潜在作用。然而，尽管代表性的生成式AI算法已展现出令人印象深刻的语言能力，它们在复杂推理任务中仍存在明显不足。来自认知神经科学的最新研究表明，人类大脑中语言与推理能力所依赖的认知机制存在显著差异。本文采用跨学科视角，综述了心理学、认知科学、精神病学以及人工智能领域的相关研究与观点。在此基础上，我们设想了以案例推理（Case-Based Reasoning, CBR）为核心的下一代生成智能体，并提出了相应的技术框架，同时探讨了未来可能的研究方向。