This paper evaluates the impact of Generative Artificial Intelligence (GenAI) on productivity and work inequality. I run a Randomized Controlled Trial in a university debating competition, in which I randomly assign GenAI support to students. This setting allows me to directly measure the effects of GenAI in a task that involves a unique combination of higher-order cognitive and social skills that mimic the skills required in high-earning jobs of different types. Contrary to most of the early findings in the GenAI literature, I find that high-skilled students benefit significantly more from GenAI than their lower-performing counterparts. Analysis of mechanisms suggests that high-ability students enjoy stronger complementarity with the AI in sub-tasks that require good judgment to extract the most relevant information from the AI under time pressure. The survey evidence also shows that students experience greater improvements in their perception of the time required to prepare debates. I suggest a possible explanation to reconcile these results with the previous literature. LLMs provide good and structured information available at a very low cost, but they are not yet a good substitute for good judgment in real-time knowledge-intensive interactions where AI cannot be used as a substitute.
Antonio Roldan
18 June 2024 Paper Number POIDWP096
Download PDF - When GenAI increases inequality: Evidence from a university debating competition