Economists Clash Over GenAI's Impact on Wages and Jobs: Same Data, Two Narratives

Economist joseazar and coauthors updated their paper on GenAI and labor markets, triggering a public methodological debate with paulnovosad. The core fact: high-exposure occupations earn less relative to low-exposure ones, but early-version charts suggest all groups' real wages may be rising, making the paper's wording the crux of the dispute. The same day, paulnovosad discussed Brynjolfsson et al.'s 2026 paper showing clear employment reallocation across exposure levels. No third-party resolution has emerged.

Confirmed

• joseazar's three main findings: GenAI automation exposure negatively affects wages in both BLS and Revelio data; exposure also lowers quit and job-switching rates.
• joseazar's response to the misreading controversy: high-exposure occupations earn about 10% less relative to low-exposure ones after 2022 (diff-in-diff estimate); the relative effect is negative, but absolute wage gains among low-exposure workers cannot be attributed to AI.
• Brynjolfsson et al. 2026: the top 40% most AI-exposed occupations saw roughly 11% net employment losses, while the least-exposed 60% gained about 10%. Paulnovosad views this as a normal technological-change pattern with retraining options.
• Paulnovosad cautions on occupation-level vs. individual-level distinctions: job-title mismatch can exaggerate measured losses.
• Paulnovosad highlights hidden assumptions in diff-in-diff (e.g., missing intercept) and warns against reading dynamic DD plots without checking group means.

Unconfirmed

• Paulnovosad argues percentile-exposure diff-in-diff only shows correlation between exposure and wage changes, not aggregate wage or employment losses; he suggests the paper be worded as "real wages rose for all groups but least for the exposed." This critique itself remains unresolved.

Why it matters

• One must separate automation exposure effects from overall labor-market effects; hidden DD assumptions can cause systematic misreading.
• The same empirical result can support opposite narratives—"AI-exposed wages fell" vs. "low-exposure wages rose more"—so media and policymakers must scrutinize wording, framing, and underlying data.

2026-09-16 ~ 2026-09-16 · 12 related posts

• [source] Study: GenAI automation exposure cuts US wages and locks workers in place — joseazar · 2026-09-16
• Economists clash: GenAI automation cuts US wages, Danish data shows no effect — paulnovosad · 2026-09-16
• [source] 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

4 near-duplicate retellings: paulnovosad · paulnovosad · paulnovosad · paulnovosad