ISSN 1671-3710
CN 11-4766/R
主办:中国科学院心理研究所
出版:科学出版社

• •    

人-生成式AI协作模式对创造力的影响:基于创造性认知功能分布框架

钱晨, 郑晓明   

  1. 北京科技大学经济管理学院, 北京 100083 中国
    清华大学经济管理学院, 北京 100084 中国
  • 收稿日期:2026-07-23 修回日期:2026-09-30 接受日期:2026-10-14
  • 基金资助:
    动态视角下的人机协作创新研究:基于互动文本大数据的探索(72502127)

How Do Human–Generative AI Collaboration Modes Shape Creativity? A Creative Cognitive Function Distribution Framework

  1. , 100083, China
    , 100084, China
  • Received:2026-07-23 Revised:2026-09-30 Accepted:2026-10-14
  • Supported by:
    Research on Human-Al Collaborative Innovation from a Dynamic Perspective: An Exploration Based on Interactive Text Big Data(72502127)

摘要: 生成式人工智能在提升个体创意质量的同时,可能导致群体创意趋同,且其即时绩效增益未必转化为使用者的独立创新能力,形成“AI创意悖论”。现有研究多比较是否使用AI或比较不同协作配置的总效应,尚未揭示AI承担何种创造性认知功能及其如何影响人的认知发展。本研究提出创造性认知功能分布框架,将AI相对于使用者当前认知状态的边际贡献区分为启发式协作与验证式协作,并开展3项递进研究:研究1开发两类协作的功能性构念,并借助LLM辅助的交互文本大数据分析,识别AI相对于人类前置认知状态的边际功能,将动态协作过程转化为可比较的认知功能分布指标;研究2揭示其分别经由认知增强与认知替代影响短期创意绩效的双刃剑机制及边界条件;研究3考察反复协作如何通过创造策略内化与替代性认知卸载,影响AI撤除后的独立创新能力。在此基础上,本研究建构人机创造功能分布的动态理论,将研究焦点从“是否使用AI”推进到“如何分配创造功能”,并从AI在场时的联合绩效拓展至AI撤除后的能力遗留,为实现人机协作的绩效增益与人的创造能力发展提供理论依据。

关键词: 生成式人工智能, 人机协作模式, 创意悖论, 认知功能分布, LLM文本分析

Abstract: Generative artificial intelligence (GenAI) can improve the quality of individual creative outputs while reducing collective creative diversity. Moreover, its immediate performance benefits may not persist as users’ unaided creative capability after AI support is withdrawn. We conceptualize these tensions as the “AI creativity paradox.” Existing research primarily compares AI use with non-use or examines the total effects of bundled collaboration configurations, leaving unclear which creative cognitive functions AI performs and how their distribution shapes human cognitive development. To address this gap, we develop a creative cognitive function distribution framework. Based on AI’s marginal contribution relative to a user’s current cognitive state, we distinguish between inspiration-oriented and validation-oriented collaboration. Three studies are proposed. Study 1 develops and validates the two functional constructs and uses LLM-assisted analysis of large-scale interaction logs to measure their occurrence, intensity, and temporal dynamics. Study 2 examines how the two collaboration modes influence short-term creative performance and collective diversity through the competing pathways of cognitive augmentation and cognitive substitution, as well as how these pathways vary across technological, task, and user conditions. Study 3 investigates how repeated collaboration shapes unaided creative capability after AI withdrawal through creative strategy internalization and substitutive cognitive offloading. Together, these studies develop a dynamic theory of creative cognitive function distribution, shifting the research focus from whether AI is used to how creative functions are allocated and from joint performance with AI to the development of human creative capability.

Key words: generative artificial intelligence, human–AI collaboration modes, AI creativity paradox, creative cognitive function distribution, LLM-assisted text analysis