MIT study: AI lifts patent-drafting quality, but junior lawyers' unaided judgment lags
A not-yet-peer-reviewed working paper by MIT economist David Autor reports a three-month experiment with 133 patent lawyers at 11 US intellectual property law firms, about two-thirds of whom were randomly assigned to first use a custom AI patent-drafting assistant and complete patent drafts from simulated invention materials at 10 and 90 days. AI use raised draft scores, with a 0.38 standard-deviation improvement at 90 days. But on a final test that barred AI and asked lawyers to redline a hypothetical patent document seeded with errors, gains were concentrated among lawyers with seven or more years of experience, who averaged 0.45 standard deviations higher, while less-experienced lawyers showed no average improvement. Autor told Fortune that AI is more a performance equalizer that may act as a skill disequalizer, warning young professionals about a capability illusion because real skill gains are still measured without AI assistance.
The findings indicate AI assistance can raise output quality without building less-experienced workers' independent judgment, and Autor warned that firms cutting junior training for efficiency may sever the apprenticeship pipeline that produces future senior talent.


麻省理工學院(MIT)經濟學者大衛·奧托(David Autor)最新研究顯示,人工智慧或許能讓工作成果看起來更好,卻不一定能幫助年輕工作者真正累積能力。這項針對美國專利律師(patent lawyers)的 3 個月實驗發現,AI 確實提升整體草稿品質,但當研究人員拿掉工具、改為測試獨立判斷能力時,只有資深律師的平均表現進步,初階律師則沒有明顯收穫。
這份尚未經同儕審查的工作論文,納入 133 名來自 11 家美國智慧財產權律師事務所的專利律師。研究團隊隨機讓其中約三分之二的人先使用一款客製化的 AI 專利撰寫助理,並在 10 天與 90 天後,要求他們根據模擬的發明資料完成專利草稿。結果顯示,使用 AI 的律師草稿評分明顯較高,90 天後的提升幅度達 0.38 個標準差。