Economists Debate Framing of GenAI Wage Studies

Economist joseazar and co-authors updated their paper on GenAI and the labor market, triggering a public methodological debate with paulnovosad. The core finding: occupations with high GenAI exposure have lower wages relative to low-exposure ones, yet real wages may be rising across all groups—and how the paper words this result is the crux of the dispute.

Confirmed

• The paper's three main findings (posted by joseazar): whether using Bureau of Labor Statistics data or Revelio data, GenAI automation exposure has a negative effect on wages; GenAI exposure also reduces employee turnover and job-switching rates.
• joseazar responded to the statistical misinterpretation controversy: after 2022, occupations with high GenAI exposure earned about 10% less than low-exposure ones (difference-in-differences estimate); this indicates a negative relative effect, but the absolute rise in real wages among low-exposure groups cannot be directly attributed to AI.

Unconfirmed

• paulnovosad argues the paper's percentile-of-exposure difference-in-differences design can only show correlation between exposure and wage changes, and cannot establish overall wage or employment losses; charts in an early version show real wages rising for all groups, with AI-exposed occupations gaining the least. He suggests the paper should state that "real wages rose for all groups, but the exposed group gained the least," and says he would not conclude that "workers with high AI exposure are worse off." This challenge to the paper's interpretive power has itself not been adjudicated by a third party.

Why it matters

• paulnovosad stresses the need to distinguish two kinds of "AI": the effect of jobs being automated by AI versus overall labor market effects; hidden assumptions in the difference-in-differences design can lead to misreading.
• The debate is a reminder for media and policymakers: the same empirical result can be framed as "wages fell in high-AI-exposure occupations" or "low-exposure occupations gained more"—two opposite narratives—so wording and framing must be checked carefully when reading such papers.

2026-09-16 ~ 2026-09-16 · 10 related posts

• [source] Study: GenAI automation exposure cuts US wages and locks workers in place — joseazar · 2026-09-16
• [source] Economists clash: GenAI automation cuts US wages, Danish data shows no effect — paulnovosad · 2026-09-16
• Economist challenges AI labor paper: diff-in-diff can't support wage-loss claims — paulnovosad · 2026-09-16
• [source] Paper author clarifies: GenAI-exposed workers saw ~10% relative wage losses, not absolute declines — joseazar · 2026-09-16
• Critics push back on GenAI wage paper's framing: wages rose least in exposed jobs, not declined — paulnovosad · 2026-09-16
• Economist: Don't Overread 'Negative Wage Effects of AI' Papers — paulnovosad · 2026-09-16
• Economist cautions AI job-loss studies may overstate harm due to occupation vs. person mismatch — paulnovosad · 2026-09-16
• Brynjolfsson et al. 2026: 11% net job loss in most AI-exposed jobs, 10% gain elsewhere — paulnovosad · 2026-09-16

2 near-duplicate retellings: paulnovosad · paulnovosad