《保险研究》20260402-《人工智能时代资产管理行业投研认知范式转型研究》（段国圣、迟哲）

［摘   要］在生成式人工智能技术的推动下,资产管理行业的投资研究活动正经历一场认知范式转型。本文首先从大语言模型(Large Language Models,LLMs)的技术原理和发展历程出发,探讨了其认知能力的本质、优势及局限性。其次,从认知科学视角探讨了传统投研模式的内在局限,包括有限理性、双系统偏差和图式固化等。在此基础上,重点论述了LLMs如何作为“认知引擎”重塑投研流程,以缓解人类有限理性局限,辅助纠正证实性偏见,并支持动态优化研究框架的作用机制和效果。本文提出,大模型与人类研究员在认知能力上存在非对称性互补,二者并非替代而是协同关系,这是由其认知能力的内在特点和局限性所决定的。最后,本文分析了LLMs推动的行业认知范式转型对行业生态和市场格局的影响,并提出了对资产管理机构的转型建议。

［关键词］大语言模型；资产管理；投资研究；认知论

［作者简介］段国圣,泰康保险集团执行副总裁、首席投资官,武汉大学兼职教授,研究方向:资产管理、保险投资；迟哲,泰康资产首席技术官。

A Research on the Transformation of Cognitive Paradigms in Investment Research of the Asset Management Industry in the Era of Artificial Intelligence

DUAN Guo-sheng,CHI Zhe

Abstract：Driven by generative artificial intelligence(AI) technology,investment research activities in the asset management industry are undergoing a transformation of cognitive paradigms.First,starting from the technical principles and development history of Large Language Models(LLMs),this paper explores the essence,advantages,and limitations of their cognitive capabilities.Second,from the perspective of cognitive science,it examines the inherent limitations of traditional investment research models,including bounded rationality,dual-system bias,and schema solidification.On this basis,the paper focuses on discussing how LLMs,as a “cognitive engine”,reshape the investment research process to alleviate the limitations of human bounded rationality,assist in correcting confirmatory bias,and support the dynamic optimization of the functioning mechanism and effects of research frameworks.The paper proposes that there exists an asymmetric complementarity in cognitive capabilities between LLMs and human researchers,and the two have a collaborative rather than substitutive relationship-this is determined by the inherent characteristics and endogenous limitations of their respective cognitive capabilities.Finally,the paper analyzes the impact of the industry’s cognitive paradigm transformation driven by LLMs on the industry ecosystem and market structure,and puts forward transformation suggestions for asset management institutions.

Key words：large language models;asset management;investment research;epistemology