| 摘要: |
| 目的 旨在深入探索人智协作概念设计的优势和局限，结合神经科学、生理学和社会科学的三重研究范式，测量大语言模型智能体对设计师的认知过程、设计结果(绩效)及协作体验的影响。方法 使用GPTs开发一款概念设计智能体，将30名参与者随机分为人-智组和人-人组，进行一项10 min的概念设计任务对照。结果 认知过程中，与智能体合作，显著提高了参与者大脑认知资源的协调性，降低了大脑认知负荷和生理皮肤电激活程度。在输出结果中，参与者与智能体进行协作时表现出更高的思维流畅性，而与人协作时则展现出更强的原创性。参与者普遍认为智能体具备出色的信息检索和处理能力，与其协作时压力更小。但智能体需要被参与者主导决策，且在沟通效率和信任度上不及人类协作者。结论 当前，智能体在提升设计师大脑认知效率、提高创意产出数量，以及稳定任务压力和情绪等方面优势明显，但在原创性、沟通效率和信任度方面仍需要提升。 |
| 关键词: 概念设计 大语言模型智能体 人-智协作 人-人协作 功能性近红外光谱 皮肤电 |
| DOI：10.19554/j.cnki.1001-3563.2025.10.004 |
| 分类号: |
| 基金项目:北京市教育科学“十四五”规划2022年度重点课题(CDAA22036) |
|
| Cognition and Performance of Conceptual Design Based on Human-Agent Collaboration |
|
GE Shijun1, CUI Yin2, QI Kaize3, SUN Yuanbo1, LI Jiahua4
|
|
(1. Beijing Institute of Technology, Beijing 100081, China;2. Shenzhen Technology University, Shenzhen 518118, China;3. National University of Singapore, Singapore 119077, Singapore;4. Beijing Everloyal Technology Co., Ltd, Beijing 100029, China)
|
| Abstract: |
| The work aims to explore the advantages and limitations of human-agent collaborative conceptual design and measure how LLM-based agent influences designers' cognitive processes, design outputs (performance) and collaboration experience by integrating neuroscience, physiology, and social science paradigms. A conceptual design agent was developed with GPTs. 30 participants were randomly assigned to human-agent and human-human groups for a 10-minute conceptual design task comparison. In terms of cognitive processing, the collaboration with the agent significantly enhanced participants' brain cognitive resource synchronization while reducing cognitive load and physiological skin conductance levels. The human-agent collaboration demonstrated higher ideation fluency, whereas human-human collaboration exhibited stronger originality in design outputs. Participants generally acknowledged the agent's exceptional information retrieval and processing capabilities, reporting lower stress levels during collaboration. However, the agent required participant-led decision-making and showed limitations in communication efficiency and trust compared to human collaborators. In conclusion, LLM-based agent currently demonstrates clear advantages in enhancing the cognitive efficiency and creative output of designers, as well as in stabilizing their workload and emotional well-being. However, there is still room for improvement in terms of originality, communication efficiency and trust. |
| Key words: conceptual design LLM-based agent human-agent collaboration human-human collaboration fNIRS EDA |
