AI, Redistribution, and the Size of the Economy

Summary

Alex Tabarrok examines a paper by Anthropic’s economic team, including Anton Korinek and Chad Jones, that models how AI capability, adoption speed, and task automation could affect growth, wages, and unemployment. In the paper’s extreme scenario, GDP is 32.4% higher by 2030 than it would be without AI, while labor’s share falls from 60% to 45.2%. Because the larger economy offsets the smaller share, total labor income is almost unchanged from the no-AI baseline. Tabarrok says his own updated, 10-year analysis suggests that only the modest scenario requires a net labor transfer for AI to be Pareto improving in aggregate. The aggregate result masks distributional effects: cognitive occupations lose income while other occupations gain, and restoring the affected occupations’ wage bill in the extreme scenario would cost about 9% of GDP. He argues this adjustment burden may be overstated because workers can reallocate through retirement and new labor-market entry, although those channels have limited effect by 2030. He also proposes shifting taxation from labor to consumption: with a 45% labor share, shifting taxes equal to 5% of GDP would reduce labor’s net burden by 2.75% of GDP without raising total revenue. Past episodes, including U.S. unemployment benefits reaching about 2.5% of GDP in 2020 and Britain’s roughly 5%-of-GDP compensation to slaveowners after abolition, show that governments have mobilized substantial support. Tabarrok concludes that AI growth could make compensation affordable, even though stable aggregate labor income would not protect every worker.