Will AI soon lead to double-digit growth?
20 days ago
- - Many AI insiders predict double-digit GDP growth (10-100% per year) in the next decade, but the authors argue such growth is extremely unlikely in the next 10-15 years.
- - It is more useful to think in levels: doubling GDP in 15 years implies only 4.7% annual growth, so even 4-5% growth would be massive.
- - In theory, standard growth models can produce explosive AI-driven growth when automation removes labor as a bottleneck and capital becomes accumulable.
- - Explosive growth requires five assumptions: fast economy-wide automation, continued spending on automated goods, demand for output, no destructive AI cyber incidents, and AI-accelerated R&D.
- - Automation diffusion is slow, most work is physical, and political resistance would slow a rapid transition to 66% automation by 2035.
- - Spending shifts away from automated goods to scarce human-provided services (Baumol's cost disease); the non-automatable part of the economy constrains growth.
- - Essential inputs like energy, construction, and specialized components limit how quickly machines can be built, while demand may falter if income shifts to high-saving capital owners.
- - AI-driven cyberattacks could destroy economic value and slow deployment, while AI automating R&D faces falling research productivity and long lags before affecting GDP.
- - AI is not like a million new workers: it cannot do physical/relational tasks, needs years to build infrastructure, and does not consume, so demand may not absorb new output.
- - GDP misses much of AI's welfare value, so 4-5% measured growth could coexist with large consumer surplus gains.
- - Overoptimism likely stems from extrapolating Silicon Valley's capability growth to the whole economy and from accepting elegant growth-model assumptions without stress testing them.