Building pro-worker AI
S · Shapes the field Daron Acemoglu, David Autor and Simon Johnson, three of the most cited labour economists of the period, set out a framework for which AI raises the value of human expertise, written for the Brookings Hamilton Project.
Summary based on the full text·Original: about 90 minutes to read
In short
NBER Working Paper 34854 (February 2026) by MIT economists Daron Acemoglu, David Autor and Simon Johnson, written for the Hamilton Project at Brookings. It defines "pro-worker AI" as technology that makes human skills and expertise more valuable by expanding what workers can do. The authors argue AI is better suited to collaboration than to full automation, yet markets underinvest in the collaborative direction because of misaligned incentives. They close with nine policy directions.
Key numbers
- 52%US workers surveyed in 2024 who are worried about how AI will affect their jobs
- 42% vs 30%AI users vs non-users who believe AI will reduce their future job opportunities
- almost 80%USPTO patent examiners using AI-based search features as of June 2024
- 74% to 46%labour share of value added in US manufacturing, 1981 to 2016
As reported by the source.
English·PDF·opens nber.org
What it reports
- Five categories of technological change: labour-augmenting, capital-augmenting, automating, expertise-leveling and new task-creating. Only new task-creating change is unambiguously pro-worker.
- Survey context: 52% of US workers (2024) are worried about AI and their jobs; 42% of AI users expect fewer future job opportunities vs 30% of non-users.
- Labour share in US manufacturing fell from 74% to 46% between 1981 and 2016; labour share of national income fell from 58% to 52%.
- Case contrast in aviation maintenance: an "automator" tool that aims to de-skill technicians vs an "assistant" that extends their expertise.
- USPTO "More Like This" search tool (introduced 2021, built with Accenture Federal Services): queries answered in seconds vs hours before; almost 80% of examiners had used AI search features by June 2024. The authors call it pro-worker only if examiners use the freed time for deeper judgment.
- Ride-hailing example: US taxi and chauffeur employment rose by around 170%, from 288,971 in 2000 to 780,371 in 2018, while the new competition likely lowered incumbent drivers' wages.
- Market failures named: firms deploying AI to cut dependence on labour or weaken bargaining, developers focused on automation, path dependence, and a pro-automation ideology.
- Nine policy directions include public investment in health care and education AI, government AI expertise, grant-making, a tax code that treats hiring and software investment symmetrically, antitrust, worker voice, worker ownership of expertise and licensure reform.
How they got it
A conceptual essay built on task-based models of technology and labour, illustrated with hypothetical tools (aviation maintenance, electricians, custodial work) and real cases (patent examination, teacher assistants, gig delivery). Figures come from cited surveys, BLS and Census data and prior work by the authors. No new dataset is collected.
Limits
- Classification of real tools as pro-worker is tentative and depends on how firms use the freed time, which the paper does not measure.
- Policy proposals are US-specific (tax code, USPTO, licensure).
- At least one co-author disclosed relationships of potential relevance; Autor acknowledges Google and Schmidt Sciences fellowship support.
Why it matters for companies
The framework gives managers a test for AI investments: does the tool make the job's expertise more valuable, or does it only make the same work cheaper? Automation-first deployments tend to commoditise skills and weaken internal talent pipelines. Collaborative tools tolerate imperfect accuracy because a human keeps the judgment, which lowers deployment risk. The tax and bargaining arguments also explain why business cases default to headcount reduction even when augmentation would pay more.
Summary based on the full text. Read the original for the full argument and data.Open PDF on nber.org