16880: When ChatGPT is Gone: Individual Creativity Fails to Sustain While Homogeneity Keeps Climbing (session 26201)

ChatGPT has been proved enhancing human creativity with the cost of content diversity. Yet, what occurs when individuals, who have developed a dependency on it, find ChatGPT inaccessible? In this study, we examine the impact of both the presence and absence of ChatGPT on creative output and content homogeneity, leveraging two complementary methodologies: a natural experiment (Study 1) and a controlled laboratory experiment with extended follow-ups (Study 2). Study 1 analyzed 419,344 academic papers published before and after ChatGPT-3.5’s release across all subjects categorized by Web of Science (i.e., Physical Science, Life Sciences & Biomedicine, Technology, Social Science, Arts & Humanities). The results indicate an enhanced productivity and quality but a reduced content diversity post-ChatGPT integration (compared to pre-ChatGPT). Study 2, with a dataset of 3,585 original ideas and 427 solutions, repeatedly found that ChatGPT group (compared to non-ChatGPT group) performed better on daily creative tasks at the expense of content diversity. Notably, when ChatGPT was unavailable, participants’ creativity dropped to baseline levels, while the trend of content homogeneity endured even two months later. These insights suggest that generative AI like ChatGPT may temporarily elevate creative output but ultimately constrain the sustained innovation by fostering homogeneity.

view paper (if available)