Anthropic's new economics model: extreme AI scenario puts 2030 US unemployment at 11.9%
akorinek · x · 2026-09-09
Anthropic's economics team (Anton Korinek et al.) released the working paper "Economic Scenarios for Transformative AI" plus an interactive scenario explorer, decomposing the AI-economy debate into a few parameters (task coverage, diffusion, per-instance cost savings, automation share, new-task creation) and projecting outcomes to 2030.
Three illustrative scenarios (2%/yr growth without AI):
- Modest: GDP +1.6% vs. the no-AI path
- Substantial (AI on 1/5 of cognitive tasks): GDP +8.3%, unemployment 4.6%, labor share falls from 60% to 56%
- Extreme (AI on half of tasks): GDP +32%, unemployment 11.9% (17.9% for cognitive workers), labor share drops to 45%
The core tension is distribution, not growth: in the extreme case the average wage is 9.7% above the no-AI path, but cognitive wages fall 11.5% while other wages rise 34%, as automation shifts income from labor to capital. Aggregate gains are roughly 3x cognitive workers' losses, so transfers could compensate—but only with policy. A Morning Consult survey of 10,980 US adults yields a median expectation near the substantial scenario (GDP +8.6%, unemployment 4.6%).
Related event: Anthropic Models Economic Scenarios for Transformative AI Through 2030(39 posts)→
More from AGI Musings
- Dev Criticizes LLM Surveys: Models Are Text Samplers, Not People — gerardsans · 2026-10-26
- Beyond Jev: a guide to 15 new System One decision-model research approaches — TheTuringPost · 2026-10-26
- Dev calls for a new design aesthetic: enough minimalist shadcn slop in the AGI era — zack_overflow · 2026-10-26
- If Anthropic won't rule out model consciousness, is selling minds at $20/month justifiable? — nucle4r_attack · 2026-10-26
- AI shifts the firm's boundary: cheaper in-house plus AI-amplified security risk may spell the death of B2B — cocktailpeanut · 2026-10-26
- Microsoft chief scientist Horvitz: AlphaFold's 200M predictions are forcing science to rethink credit and integrity — erichorvitz · 2026-10-26