MIT's David Autor and Google ran an RCT on AI at work: juniors gained ~3x seniors, and heaviest AI users learned the least
_akpiper · x · 2026-09-21
David Autor, the world's most-cited labor economist at MIT, partnered with Google on a rare randomized controlled trial of AI at work.
- 133 lawyers across 11 US firms were randomly assigned: two-thirds used AI to draft patent documents for 3 months, one-third didn't; output was blind-graded by specialist lawyers.
- With AI on, everyone improved, but juniors' quality gains were nearly 3x those of seniors.
- The kicker: when AI was removed and everyone revised a document unassisted, those who benefited most from AI learned the least. Prior expertise appears to be a prerequisite for converting AI assistance into durable learning.
Related event: NBER trial: AI drafting tool benefits senior patent lawyers most(2 posts)→
More from AGI Musings
- OpenAI's Noam Brown: aligned AI workers could hand the edge to incumbents over startups — victor_explore · 2026-09-21
- Measuring AI's impact: focus on final outputs, not lines of code — soumitrashukla9 · 2026-09-21
- If your p(doom) > 0, why are you accelerating frontier lab research? — _arohan_ · 2026-09-21
- AI assistants don't remove decisions, they multiply them — and nobody wants that — menhguin · 2026-09-21
- Researcher Calls Out p(doom) Contradiction: Why Accelerate Labs If Doom Is Nonzero? — _arohan_ · 2026-09-21
- Daniel Mac: It's amazing we can even debate whether current AI is AGI — daniel_mac8 · 2026-09-21