How is AI shaping economic research?

Generative AI may either narrow or widen productivity gaps in economic research, according to a recent study. By examining data from three sources—NBER working papers, LLM-assisted writing measures, and a survey of 227 economists—the researchers found that the AI's impact on productivity varies among researchers. Authors without research assistants (RAs) saw an increase in the number of papers published in top-tier journals following ChatGPT's release, while researchers from smaller teams experienced growth in working-paper output and publication speed.

However, when examining individual AI use through the survey, the results were less conclusive. While there was no significant effect on average working-paper output, there was substantial variation in the impact. Greater AI exposure tended to close the output gap between researchers with and without RAs, but widened the gap between senior and junior researchers.

Selection into AI use also differed, with junior researchers utilizing AI more intensively and being more likely to employ agentic tools. The researchers suggest that these differences in productivity effects stem from varying returns to AI rather than selection into its use. A model incorporating diminishing returns to execution, increasing returns to conceptual capacity, and complementarity between the two factors provides a potential explanation for how AI can help alleviate support limitations while amplifying the rewards of accumulated expertise.

Over time, however, early convergence may transition into divergence as execution constraints diminish and AI's conceptual capabilities continue to improve. The study was conducted by Tianyu Jiang, who is a job market candidate at the University of Michigan this year, and co-authored with Yuqing (Claire) Wang.