General Seminar - Pei-Cheng Yu

Optimal Taxation in the Age of AI Uncertainty

This paper examines the impact of non-insurable technological shocks on labor markets and optimal tax policy. Using survey data, we document a substantial dispersion in artificial intelligence (AI) pessimism that is negatively correlated with worker income. Motivated by these findings, we build a dynamic Mirrlees income taxation model with ambiguity-averse workers. Workers form endogenous worst-case beliefs about the impact of new technologies, that vary by skill level. This endogenous heterogeneity in expectations enables the formulation of government policies that indirectly influence workers’ subjective beliefs to enhance welfare. We calibrate the model to the empirical findings on AI expectations and solve for optimal tax schedules and AI investment. We find that uncertainty over AI motivates a more progressive labor income tax schedule, combined with a mostly flat savings subsidy of about 12%. The optimal policy reverses the sign of the empirical relationship between AI pessimism and worker income.