Building Pro-Worker Artificial Intelligence
Card Grid View — Acemoglu, Autor & Johnson (2026)
1. Defining Pro-Worker AI
- Core definition
- AI that makes human skills and expertise more valuable
- Expands what workers can do, rather than commodifying their labor
- Technology that complements rather than replaces human capabilities
- Pro-worker AI vs automation
- Automation replaces human tasks; collaboration augments them
- Pro-worker AI raises productivity AND expands worker capabilities
- Focus on human-AI collaboration
2. Five Categories of Technological Change
- Capital-augmenting
- Makes capital more productive, displaces labor
- Automation
- Replaces human tasks with machines — ambiguous for workers
- Labor-augmenting
- Makes workers more productive in existing tasks
- Expertise-leveling
- Reduces skill gaps — ambiguous (helps some, hurts others)
- New task-creating (unambiguously pro-worker)
- Creates entirely new tasks and roles for humans
- The ONLY category considered unambiguously pro-worker
3. Market Underinvestment in Pro-Worker AI
- Why markets underinvest
- Private returns favor automation over augmentation
- Firms capture more value from replacing workers than empowering them
- Labor savings are easier to monetize than productivity gains
- Systemic failure
- Market incentives systematically biased against pro-worker AI
- Even when automation reduces overall productivity, firms still adopt it
- AGI ideology creates skepticism about pro-worker alternatives
4. Automation vs Collaboration
- Automation
- Replaces human judgment entirely
- Must be infallible to be useful
- Reduces wages and employment
- Devalues human expertise
- Collaboration
- AI as collaborator, not replacement
- Does not need to be perfect — human provides oversight
- Raises productivity and expands worker capabilities
- Preserves human agency and autonomy
5. Real-World Examples
- USPTO "More Like This" tool
- Helps patent examiners find prior art more efficiently
- Augments rather than replaces examiner expertise
- Hearing aid voice chatbot (China)
- AI tool improved job performance of hearing-impaired gig workers
- Customer review scores dramatically improved
- AMT Liftoff Assistants
- "AMA Automator" harms workers vs "AMT Assistant" helps
- Contrast shows design choices determine worker outcomes
6. Expertise-Leveling Technologies
- What they are
- Reduce the gap between expert and novice performance
- Make specialized knowledge more accessible
- Ambiguous impact
- Help less experienced workers catch up
- But may commodify expert skills and reduce their premium
- Not clearly pro-worker — depends on context
- Example
- AI tools that let novices perform tasks previously requiring experts
7. Policy Recommendations
- Shift incentives
- Tax automation, subsidize pro-worker AI development
- Change corporate incentives to favor augmentation
- Public investment
- Fund R&D in collaborative AI technologies
- Support education and training for AI-augmented work
- Regulatory framework
- Require human oversight for AI-driven decisions
- Ensure workers benefit from productivity gains
- Promote new task creation over mere automation