When generative AI increases inequality: experimental evidence from a university debate competition

→ This research brief summarizes the results of research evaluating the impact of generative artificial intelligence (AI) on productivity and inequalities through a debate tournament experiment (RCT), in which students are randomly assigned ChatGPT support to prepare for a series of one-on-one debates.

→ Contrary to results so far in the literature that point to a reduction in productivity inequalities due to the improvement of lower-skilled workers on a variety of written tasks as a consequence of using generative AI, the results of this experiment show that high-skilled individuals benefit significantly more from it than their lower-skilled peers.

→ The author proposes a possible explanation to reconcile these results with previous findings: in unpredictable verbal interactions requiring critical thinking and social skills, when responses cannot be copied and pasted, high ability workers benefit more from these tools.

Translated with DeepL.com (free version)