Autor NBER trial: AI lifts patent lawyers, splits juniors

Give 133 patent lawyers at eleven U.S. intellectual property firms an AI drafting assistant for three months, then take it away and score them blind. That is the setup for NBER Working Paper 35720 by David Autor, Tanya Rodchenko, Josh Martin, Zanna Iscenko, Scott Strand, David Pearl, and Melissa Ferere.

During AI use, quality rose: 0.34 standard deviations at 10 days (p=0.03) and 0.38 SD at 90 days (p=0.01), with larger gains for junior lawyers. The more interesting number came after. When everyone completed a core patent task without the tool, treated lawyers still outperformed controls by 0.32 SD (p=0.04).

The gain did not split evenly. Senior lawyers held a sustained 0.45 SD edge (p=0.02). Junior lawyers, per the paper, showed "no average gain; their scores instead bifurcated, with sharply fewer mediocre scores offset by more poor and more good ones."

The authors' line: "foundational expertise may be a prerequisite for extracting durable skill from AI-assisted practice." Or, put more bluntly in the paper, "the largest gains from AI thus accrued to the lawyers who retained the least."

Two researchers on our tracker were passing the paper around this week.