Economists may botch AI labor research: no solid evidence yet on jobs, wages or productivity
paulnovosad · x · 2026-10-01
A threaded discussion on how AI is reshaping labor-market research. alexolegimas argues economists may do a poor job studying frontier AI by fixating on conventional metrics (employment counts, wages, productivity) while workers experience far more instability in the very concept of work. paulnovosad adds that better data now exists on how jobs are changing—but it mostly sits inside closed labs, though teams like Peter McCrory's and Ronnie Chatterji's are starting to externalize it.
Quoted in the thread, Rishi Bommasani offers an off-the-cuff evidence ledger: no evidence of aggregate effects on employment or wages; some disputed localized evidence for AI-exposed early-career workers; thin evidence on macro productivity (TFP) gains. The upshot: solid empirical proof of AI's macro labor impact is still missing, and how to measure it is actively contested.
More from AGI Musings
- Grove Research launches Delvetown, a multi-agent society for humans and AI agents — lfschiavo · 2026-10-01
- David Sacks: Calling AI agents autonomous deflects responsibility from the companies building them — kevinnbass · 2026-10-01
- Pedro Domingos: progress per AI researcher has never been slower — pmddomingos · 2026-10-01
- Startup Speed Advantage Has Inverted: The Future Is Now Measured in Weeks, Says soleio — soleio · 2026-10-01
- The Industrialization Bar Is Now 4,000 kWh Per Capita — Two-Thirds of Countries Haven't Cleared It, and AI Raises the Stakes — RichardsonDx · 2026-10-01
- Artemis Consortium: LLMs Are Already Enough for AGI Collectively — Humans Aren't AGI Individually Either — SydSteyerhart · 2026-10-01