David Autor

persona · active · confidence: medium · reviewed 2026-09-21 · tier: quarterly

Who he is

Affiliation: Head, MIT Department of Economics (effective 2026-07-01) and Daniel (1972) and Gail Rubinfeld Professor in Economics. Role check run 2026-09-21: the department-head appointment was announced 2026-06-26, succeeding Jon Gruber (MIT News); it post-dates v1's May review, so this is the first review to carry it. Two v1 credentials were stale against his own CV (2026-01-15 PDF): the chair — he was Ford Professor 2016–24 and has held the Rubinfeld chair since 2024 — and the editorial role — v1 said "co-editor, Journal of Economic Perspectives [TBD: confirm 2026]"; he was Editor in Chief 2008–14, so the role ended twelve years ago (current editors are Heidi Williams and Jeffrey Kling). Faculty co-director, James M. and Cathleen D. Stone Center on Inequality and Shaping the Future of Work (2025–): the Shaping the Future of Work Initiative folded into the Stone Center on 2025-07-01, with Acemoglu as director and Autor and Simon Johnson as co-directors (MIT News, 2025-05-13). Co-director, NBER Labor Studies Program (2017–, confirmed by the same 2026-06-26 announcement). New to the file and load-bearing for the incentive map: Google Technology and Society Visiting Fellow (2025–27) and Digital Fellow, Stanford Digital Economy Lab (2022–, Brynjolfsson's lab). Also on the CV: Foreign Affairs Policy Board, U.S. State Department (2022–, listed as current in January 2026; status under the 2025 administration not verified this cycle); Board of Economic Advisors, Congressional Budget Office (2020–); Standing Committee, One Hundred Year Study on AI (2024–25). Honors in the review window: 2025 Clarivate Citation Laureate with Lawrence Katz "for seminal analysis of wage structure, earnings inequality, educational advance, and technological change" (2025-09-25); Distinguished CES Fellow 2025 (Munich); Prospect Top 25 Thinkers for 2025 (2024-12-04); TIME100 AI 2026 (2026-08-27). MIT faculty since 1999; PhD Public Policy, Harvard Kennedy School, 1999; BA Psychology (with computer science), Tufts, 1989.
One-line position: AI is the first technology in roughly forty years with a realistic shot at narrowing wage inequality — by codifying expert judgment (turning know-how into something a machine can carry) and letting less-credentialed workers do higher-stakes work — but only if it is built to collaborate rather than automate, and his own 2025 "Expertise" model says the same technology can cut either way: remove a job's inexpert tasks and wages rise while headcount shrinks; remove its expert tasks and headcount grows while wages fall. His 2026 register: "All this talk of everyone being replaced has been debunked. It hasn't happened yet." (TIME, 2026-08-27).

Discipline & technical bet

Labor economist — the principal author of the task framework — whose instrument is eight decades of occupation-level Census, DOT and O*NET microdata (government job-content catalogs), not model benchmarks. Two live technical wagers, both new since v1. (1) The expertise framework (Autor & Thompson, Journal of the European Economic Association 23(4), 2025; NBER w33941, June 2025; delivered as the Schumpeter Lecture, 2024-08-29): a content-agnostic way to measure how expert a job's remaining tasks are, plus a model in which automation's wage and employment effects run in opposite directions depending on which tasks it removes. Exhibit: accounting clerks and inventory clerks faced the same computerization; over 1980–2018 accounting clerks' wages rose 39% while employment fell 32%, and inventory clerks' wages fell 13% while employment grew 175% (Digitalist Papers, 2025-12; restated in CNBC, 2026-09-13). (2) A field-experiment program on whether AI builds or erodes expertise. First result (NBER w35720, September 2026, with six Google co-authors, Google-funded, pre-registered AEARCTR-0015823): a three-month RCT (randomized trial) with 133 patent lawyers at eleven US firms found the AI assistant raised work quality (+0.34 SD at 10 days, +0.38 SD at 90 days, larger gains for juniors) — but when the tool was taken away, the durable judgment gain sat entirely with senior lawyers (+0.45 SD); juniors showed no average gain and a bifurcated distribution. Side bets: "Misaligned by Design" (NBER w34504, 2025-11, with Caplin, Martin, Marx) — an incentive-theory result that training a classifier on the human's asymmetric loss function can weaken its learning; and "Resolving the automation paradox" (arXiv 2601.06343, 2026-01, with Kausik) — a competitive-economy model in which a falling labor share (workers' slice of national income) can raise wages, estimating that falling labor share accounted for 16% of US real wage growth 1954–2019. Non-technical on models themselves; after Brynjolfsson, the most empirically instrumented labor voice on the bench, and unlike Brynjolfsson he now runs experiments as well as observational data.

Key claims (Says)

Notable predictions — with falsifiable checks

Revealed behavior (Does)

Feels

Wants the middle-class thesis to be true and has built the instruments that could disprove it — the visible tension in 2026 is a man whose own model, own experiment, and own colleague's payroll data all say "it depends," while his public register still says "debunked." Underneath: the China Shock scar — he watched a benign consensus be wrong about who pays, and does not intend to be on that side twice.

Hears

NBER Labor Studies and the Stone Center seminar loop (Acemoglu, Johnson, Restrepo, Salomons, Katz), Brynjolfsson's payroll data as a Stanford Digital Fellow, Google's Technology & Society people through the fellowship and the RCT, the CBO and Fed advisory circuit, and the elite-op-ed market (NYT, Atlantic). Thin on frontier-lab capability claims except as filtered through Google; almost nothing from the alignment or x-risk world.

Sees

Forty years of occupation-by-task microdata that nobody else has assembled the same way — he can say which tasks left which job and what happened to pay afterward, so he reads AI's early effects against the accounting-clerk/inventory-clerk base rate that others lack. New since 2026: inside a law-firm RCT, what happens to a junior's judgment when the tool is taken away. Blind side: model capability jumps; his instruments measure deployed effects, not what ships next quarter.

Incentive map

Re-derived this cycle, and the v1 line — "no commercial stake in AI forecasts — credibility weight high per CONVENTIONS rule 20" — does not survive the CV. Paid by MIT (now with a department budget to defend and an institutional interest in MIT's AI positioning). Research money from Google (Technology and Society Visiting Fellow 2025–27; a 2022 grant, "Where Are Technology Leaders Leading the Labor Market?"; the 2026 RCT, Google-funded with six Google co-authors and the tool supplied by Google), Schmidt Sciences (AI2050 Senior Fellowship 2024, grants 2021 and 2024), NOMIS, Hewlett, Smith Richardson, and earlier Accenture and IBM. His flagship public essay (Atlantic, 2025-08-24) was co-written with James Manyika, Google's Senior Vice President and President for Research, Labs, Technology & Society (role since 2023-04) — the executive whose organization funds the fellowship. The alignment is with the thesis, not a portfolio: "AI should collaborate and amplify expertise, not automate it" is simultaneously Autor's scholarly position and the message a company selling AI tools into professions wants an MIT economist to carry. Sells no product, holds no equity found; sells books, keynotes (Leigh Bureau exclusive) and policy authority. What he cannot easily say: "AI is a wholesale replacement technology" (contradicts both his brand and his funder's premise), and "the entry-level payroll data confirm substitution" (he built the framework that predicts it, but has stayed on attribution caution). What he did say against interest: the juniors-don't-retain-skill result, and the labor-collapse scenario. Rule-25 discount: low-to-moderate on "collaboration over automation" advocacy and on any Google-adjacent claim; low on the historical scholarship, which predates the money and is peer-reviewed.

Theories aligned with

What he's reacting against

Where he overlaps / splits (with Rich)

Track record

Empirical vs normative

Weak spots / open questions

Rich's take

Delta log

2026-09-21 — v1→v2 migration + validation (bulk run)

Sources

migrated v1→v2 2026-09-21 · backup: _archives/david-autor.html.bak-20260921 · converts-from: people/david-autor.md · AI & Society domain