Human Centered AI-Institute Research Seminar and Socialization Series

Artificial intelligence (AI) is one of the fastest-growing technologies in history, offering innovative tools that can enhance the creative process. In this paper, we present a quasi-experiment that examines the influence of ChatGPT's public release on the output of adopters. We study bloggers on Medium.com and measure AI use with the least biased detection tool available at the time of writing this paper. We use Coarsened Exact Matching to balance our analysis sample with ChatGPT adopters and similar bloggers that do not adopt. We then perform a differences-in-differences analysis around ChatGPT's release with this sample. We find that ChatGPT's release increases the productivity of adopters, decreases the number of comments on their work, and increases the novelty of their content relative to other bloggers but not relative to past work. The effect on productivity is persistent over the three months following ChatGPT's release. We partition our sample by blogger following and their propensity to write in technology-based categories prior to ChatGPT to find that the popularity decline is unique to popular non-tech bloggers who use ChatGPT to alter the categories they write within we posit that the effect is driven by the disengagement of the existing audience for these bloggers with the new content rather than a decline in the quality of work produced using ChatGPT. These results imply that generative AI may boost productivity without sacrificing originality nor losing the blogger's unique voice, but bloggers using ChatGPT to deviate from their typical content may suffer popularity declines. Thus we conclude that ChatGPT expands the production function as observable productivity increases without significant quality declines.