What History Tells Us About the AI Revolution | Tyler Cowen & Alex Imas
Economists Tyler Cowen and Alex Imas argue AI will transform the world but more slowly and normally than Bay Area forecasts suggest, because real-world bottlenecks, checks and balances, and agent coordination limits constrain the economic impact.
Watch on YouTube ↗Charlie's summary
- ▶ 11:44
Double-digit growth needs knife-edge conditions, and current data say we are below them
Alex cites a Ben Jones and Chad Jones paper: singularity-style growth requires essentially no bottlenecks, with no decreasing returns to ideas. Estimating that parameter from current data, including the METR report, gives a value “decidedly less than one,” meaning ordinary single-digit growth. Tyler adds that very high growth has historically come from tiny economies like Ghana or from recoveries after wars and recessions. Both say a full RSI loop could change this.
- ▶ 15:19
Automated sectors shrink in importance, so growth is driven by the bottlenecks
Alex explains that when something gets automated and becomes cheap, it becomes a smaller share of the economy and contributes less to measured growth. With non-homothetic preferences, value shifts to whatever is still scarce. “The thing that's growing least will be contributing most to GDP.” Tyler adds that inertia in government, nonprofits, health care and schooling, perhaps 40% of the US economy, will slow diffusion.
- ▶ 6:37
Revenue forecasts for frontier labs need other sectors to grow, and nobody can name them
Tyler asked Bay Area people which other sectors would grow fast enough to justify extreme Anthropic and OpenAI revenue numbers, and no one had an answer. He concludes the extreme numbers are false, though he remains bullish. He trusts market data, which still looks normal, over Bay Area judgment. Alex adds that rising token use with flat revenue reflects price competition among labs.
- ▶ 23:58
Alignment is a mechanism design problem: build checks and balances, not perfect agents
Tyler says economists learn from principal-agent theory that a principal can never fully control an agent, so the answer is separation of powers, not perfect alignment. Alex frames it as mechanism design: agents can be individually misaligned if incentives make them behave well. Both see the Hugging Face incident as human error on human error, and think real harms so far come from evil humans directing AIs.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.