Tyler Cowen — The #1 bottleneck to AI progress is humans

Tyler Cowen argues AI will add ~0.5 percentage points to annual growth over decades, not explosive acceleration, because human institutions and diffusion are the binding constraints.

• Cowen estimates AI boosts growth by ~0.5 percentage points per year over 30-40 years — transformative in aggregate but imperceptible year-to-year.
• Government, healthcare, education, and nonprofits together comprise roughly half the US economy and will adopt AI slowly, capping headline GDP gains.
• Cost disease logic still applies: adding more intelligence tightens all other constraints — energy, regulation, institutions — raising their marginal value instead.
• Cowen now writes his books primarily for AI models, not human readers, and argues anyone not doing this is missing a large and growing audience.
• He sees increasing variance in young people: the top and bottom are improving, but a thick middle band is declining — explaining both alarming anecdotes and broadly stable test scores.
• On EA: Cowen told an EA gathering at its peak that the movement would collapse, citing rapid growth, weak institutional crystallization, and cult-like dynamics — before SBF.
• Silicon Valley systematically overvalues intelligence because the people there are unusually smart, blinding them to diffusion research showing tech adoption is always slow.
• Cowen’s main concern about progress is its interaction with war: new technologies historically become weapons, and rare but increasingly destructive conflicts may be the key risk.

2025-01-09 · Watch on YouTube