"Applying AI to Rebuild Middle-Class Jobs" is a February 2024 NBER working paper (No. 32140) by David Autor (MIT). It argues that AI, unlike prior waves of automation and computerization, could supply expert judgment to less-credentialed workers and so help reverse the hollowing-out of the middle of the US labor market.
Summary of argument
Autor argues that earlier waves of automation and computerization codified expertise into the hands of a narrow set of elite experts, whereas AI can run in the other direction by supplying contextual, rule-based judgment to less-credentialed workers, enabling broader participation in high-stakes decision-making. The paper's core proposition is quoted directly:
"AI, if used well, can assist with restoring the middle-skill, middle-class heart of the US labor market that has been hollowed out by automation and globalization."
The conditional "if used well" carries the argument: the paper presents this as a proposition about potential rather than a forecast that the outcome will occur.
Key claims
- Previous computerization concentrated decision-making among elite experts rather than democratizing it, contrary to expectations common in the 1990s. Autor presents this as part of his established empirical line of work (high confidence).
- AI differs categorically because it combines information, rules, and acquired experience in ways that can be handed off to less-credentialed workers. This is an argument rather than a measured result (medium confidence).
- Domains where this mechanism could operate include medical care, document production, software coding, and undergraduate education. These are offered as illustrations (medium confidence).
- The opportunity is policy-contingent: how deployment is shaped matters more than the raw capability (presented as a high-level framing claim).
Autor's position within the academic labor-AI debate is one of conditional optimism centered on deployment choices rather than on technology alone.
Relation to Acemoglu
The paper is a counterpart to Acemoglu's "Simple Macroeconomics of AI" (2024). The two authors have co-authored elsewhere; here their emphasis diverges. Autor advances conditional optimism about labor-market structure, holding that AI can re-widen the middle if deployment is shaped toward augmentation. Acemoglu (NBER 32487) advances bounded pessimism about aggregate total factor productivity (TFP) and distribution, arguing that macro gains are small and that the capital-labor split widens regardless.
The two positions are not strictly contradictory. Autor's mechanism (AI as an extender of expertise) could deliver distributional improvements even under Acemoglu's modest TFP ceiling. The disagreement is over which framing policy should center: Acemoglu warns against overclaiming and emphasizes the inequality baseline, while Autor urges active shaping of AI toward augmentation.
The paper also relates to Generative AI at Work, which finds large productivity gains concentrated among lower-skilled workers, an empirical result compatible with Autor's thesis.
Relationships
- supports: AI Labor Disruption, AI and Productivity
- contradicts: The Simple Macroeconomics of AI (framing only)
- related: Generative AI at Work, Generative AI and the Nature of Work, (Source: epoch.ai), Labor Disruption Timelines: Who Predicts What and Why