How AI Helps and Hampers Junior Workers

Summary

A three-month field experiment by Autor et al. examined how a non-public AI tool affected 133 US patent lawyers. Lawyers with access to AI produced patent documents that ranked around the 65th percentile, compared with the 50th percentile for the non-AI group. The average improvement was concentrated in drafting: junior lawyers produced better documents, but did not improve on a later review exercise completed without AI. Senior lawyers showed little difference in drafting quality, yet outperformed non-AI colleagues when reviewing documents without the tool. Their reviews focused more on commercial scope, legal liabilities, and legal principles, while junior lawyers were more likely to make broad or stylistic comments. AI-use outcomes among junior lawyers also became more dispersed, with more very low and above-average results. The author argues that prior knowledge may determine how effectively employees use AI, while noting that the study does not prove AI caused junior lawyers to become worse. Suggested workplace practices include allowing experienced employees to use AI, having them share their methods, and assessing junior employees' knowledge and skills rather than evaluating output alone.