The Simple Macroeconomics of AI

Summary

Using a task-based economic framework and Hulten’s theorem, Acemoglu estimates that AI’s total factor productivity gains will be no more than 0.66% over the next decade, with corresponding GDP gains of roughly 0.93-1.56%. He argues that existing experimental evidence comes from “easy-to-learn” tasks and cannot be extrapolated to harder, context-dependent tasks where AI’s cost savings will be much lower.

Key Points

• Only about 19.9% of tasks are exposed to AI, and only 23% of those are cost-effective to automate — yielding 4.6% of tasks actually impacted
• Average labor cost savings per task estimated at 27%, yielding at most 0.66% TFP gain over 10 years (~0.064% annually)
• Distinguishes “easy-to-learn” tasks (clear outcome metrics, like writing subroutines) from “hard-to-learn” tasks (context-dependent, like medical diagnosis)
• AI will likely widen the capital-labor income gap; capital share rises by ~0.31 percentage points
• Improving low-skill productivity doesn’t necessarily reduce inequality — general equilibrium effects can paradoxically increase wage gaps
• Some AI-generated activities (deepfakes, manipulative algorithms) may increase GDP while reducing actual welfare
• Largest potential gains require reorienting AI toward worker complementarity rather than automation