Daron Acemoglu, David Autor, Simon Johnson · MIT · 2026-05-16
Three MIT economists — two of them 2024 Nobel laureates — argue that "pro-worker AI" is a policy choice, not a default outcome of better models. The February 2026 paper is the successor framing to Autor's earlier Applying AI to Rebuild Middle Class Jobs, and it sharpens the argument considerably.
The core claim. AI's effect on workers depends on which capabilities labs prioritize and which deployments firms reward. Both are responsive to incentives. The current default — capabilities optimised for full task automation, deployed by firms whose primary KPI is headcount reduction — tilts AI toward replacement. The authors argue this is a contingent equilibrium, not a technological inevitability.
What pro-worker AI would actually look like. The paper offers four design and deployment shifts:
- Expert-augmenting, not expert-replacing — tools that raise the floor for less-experienced workers while keeping decision authority with the human - Task-decomposable, not task-swallowing — interfaces that surface intermediate outputs so workers can audit, learn from, and improve the AI's reasoning - Workplace-deployed, not workforce-deployed — capital expenditure that augments existing workers rather than headcount reduction that displaces them - Tax-aligned, not tax-arbitraged — current US tax code subsidises capital substitution for labor; the authors argue for parity
The Autor lineage. The 2024 NBER paper (Applying AI to Rebuild Middle Class Jobs) established the empirical case that AI's wage-compression potential could narrow inequality if deployed to raise productivity in middle-wage occupations — nursing, skilled trades, K-12 teaching — rather than to displace them. The 2026 paper extends that argument into a normative and policy frame: which levers to pull, in what order.