Don't Believe the AI Hype

Summary

Acemoglu argues that neither economic theory nor empirical data support the exuberant AI productivity forecasts from Goldman Sachs (7% global GDP boost) and McKinsey (3-4% annual growth). Using a task-based model, he estimates AI will contribute only 0.53-0.66% to total factor productivity over the next decade, translating to roughly 1.1-1.6% GDP growth.

Key Points

• Only about 4.6% of total work tasks are likely to be cost-effectively automated by AI within the next decade
• Task-level labor cost savings average about 27%, yielding only ~0.05% annual productivity growth
• Generative AI excels at “easy-to-learn tasks” with clear success criteria, but harder tasks yield diminishing returns
• AI is deployed too much for automation and not enough for complementing workers — yielding “so-so technology”
• AI-driven gains are unlikely to lead to sizable wage increases and may widen the capital-labor income gap