ChatGPT decreases idea diversity in brainstorming

Abstract

arising from B. C. Lee & J. J. Chung Nature Human Behaviour https://doi.org/10.1038/s41562-024-01953-1 (2024)

Lee and Chung¹ explore how ChatGPT augments human creativity in brainstorming. In a series of experiments, they randomized participants to complete various creative challenges, either with or without the help of ChatGPT. These challenges included tasks such as coming up with gift ideas, designing toys from everyday objects and repurposing household items. Each participant submitted one idea, which external evaluators rated on multiple creativity dimensions, including innovativeness and usefulness. Across tasks, instructing participants to use ChatGPT enhanced the average creativity of ideas, outperforming web searches and unaided human intuition, with creativity measured as the average of aggregated scores of originality (original, innovative, creative) and appropriateness (practical, effective, useful). These results strengthen the empirical evidence for the effectiveness of large language models in idea generation, contributing to a rapidly growing literature^2,3,4. However, as we demonstrate analytically, reliance on ChatGPT for idea generation comes with a trade-off: while enhancing individual ideas’ creativity, it reduces the diversity of ideas in a pool of ideas—a critical element for effective brainstorming.