The Double-Edged Roles of Generative AI in the Creative Process: Experiments on Design Work

• 成果类型: Article; Early Access
• 署名作者: Hou, Jinghui (Jove); Wang, Lei; Wang, Gang; Wang, Harry Jiannan; Yang, Shuai
• 署名单位: University of Houston System; University of Houston; Indiana University System; Indiana University Bloomington; IU Kelley School of Business; University of Delaware; Donghua University
• 刊物名称: INFORMATION SYSTEMS RESEARCH
• ISSN/ISSBN: 1047-7047; 1526-5536
• DOI: 10.1287/isre.2024.0937
• 发表日期: 2025-10-03
• 关键词: generative AI; cocreative process; ideation; implementation; cognitive fixation; DIVERGENT THINKING; convergent thinking; Einstellung effect; designer expertise; Fixation

摘要

Generative artificial intelligence (GenAI) mimics human creativity by producing novel and complex content solutions, which redefine human-AI collaboration across various creative domains. Despite this transformative potential, the existing GenAI literature largely examines creativity as an end product, overlooking the intricate dynamics of the human-GenAI cocreative process. Addressing this gap, our research conceptualizes creativity as a process encompassing two distinct stages: an ideation stage and an implementation stage. Drawing on theories of creativity and expertise fixation, we theorize that GenAI influences the two stages differently, depending on the expertise level of the human creators. Our Study 1 (a laboratory experiment) demonstrates that GenAI tremendously enhances work creativity in the ideation stage by mitigating cognitive fixation for all designers. However, during the implementation stage, the impacts diverge: low-expertise designers with GenAI continue to experience improvements in work creativity, whereas high-expertise designers with GenAI show no gains in their work creativity and suffer a significant reduction in their work efficiency. Further video analyses reveal that expertise fixation underpins these impacts. That is, as GenAI introduces work approaches that deviate from high-expertise designers' established approaches in implementation, cocreation with GenAI leads to counterproductive outcomes. Building on these findings, Study 2 (a field experiment) further validates the impact of GenAI and the role of expertise fixation among professional designers, and rules out alternative explanations. This study also employs a cutting-edge GenAI model to ensure the robustness of our findings against technological advancements. Our findings provide a nuanced understanding of GenAI's role in the cocreative process and elucidate its heterogeneous effects based on creators' expertise levels. This research advances the literature on human-AI collaboration and offers actionable insights for optimizing the use of GenAI in creative work.