Tyler Cowen argues that AI will not produce explosive economic growth (20%+ annually) because bottlenecks and diminishing returns constrain how fast any economy can expand, even with transformative technology. He expects AI to boost growth by roughly half a percentage point per year—enormous over decades but not revolutionary in any single year. The core insight is that making one factor of production (intelligence) abundant makes other constraints more binding, not less.
Why AI won’t cause explosive growth
Cost disease generalizes: Even if AI makes some sectors hyper-productive, large parts of the economy—government (~18%), healthcare (~20%), education (~6-7%), nonprofits—are structurally slow to adopt new technology. These sectors won’t disappear; they’ll persist for decades, dragging down aggregate growth rates.
The classic cost disease mechanism (wages rise everywhere because labor is limited) is just one instance of a broader problem: when you flood the economy with more intelligence, the other scarce factors—regulatory capacity, energy, institutional adaptability, human attention—become more binding, and the marginal value of additional IQ drops.
Historical precedent supports slow diffusion: Economists who study technology diffusion (printing press, electricity, industrialization) consistently find slow, uneven adoption. Tyler trusts these domain experts over AI researchers, who he thinks are naive about diffusion.
Population models don’t predict well: Chad Jones’s model suggests that doubling effective population (via AI) should produce explosive growth. Tyler disagrees:
Renaissance Florence had ~60,000 people but produced outsized value because of institutional and cultural factors, not headcount.
Sub-Saharan Africa still lacks reliable clean water despite abundant global intelligence—showing that IQ alone doesn’t solve problems.
The Romer model (more people → more ideas → more growth) hasn’t been validated empirically; many populous regions became less innovative until recently.
Specific bottlenecks Tyler identifies:
Regulation and institutional inertia: Clinical trials, FDA processes, educational bureaucracies. AI might halve drug development time from 20 to 10 years, but the regulatory apparatus remains.
Energy supply: Expanding energy infrastructure is slow and politically difficult; nuclear power projects take 10+ years even in the best cases.
Human resistance: People trained for and attached to the current world will resist rapid change—not necessarily from doomsday fears, but from legitimate preference for stability. This will be a “massive fight.”
Fragility of high-level talent: The “bundles” that make people exceptional (IQ + determination + multiple skills at 8-10/10) are scarce and don’t scale linearly. AI can be smart and conscientious, but replicating the full bundle of top human performers is unclear.
What Tyler’s view implies
Slow takeoff is the default: Even strong AI faces diminishing returns because it interacts with a world full of non-AI constraints. The optimistic long-run view (AI transforms everything) and the expert consensus on slow diffusion are both correct—they operate on different timescales.
Half a percentage point per year is transformative over 30-40 years: Compounded, this would reshape the world, but year-to-year it would feel incremental—like a drug that took 20 years now taking 10.
The progress studies tension: Progress studies emphasizes fixing institutions and regulations to unlock growth. Tyler agrees this matters but argues that even flooding the economy with genius-level AI agents produces only modest growth gains because of the bottlenecks above. There’s no contradiction—both perspectives reflect diminishing returns to any single factor.
Increasing variance in human performance
Tyler disagrees with Patrick Collison’s “competency crisis” framing. He sees increasing variance, not uniform decline:
The top is getting better: Young chess players, NBA players, scientists, and internet writers are more impressive than their predecessors. Measured performance at the peaks is rising.
The bottom is also improving: Youth crime has fallen since the 1990s.
The middle is declining: A thick band around the median shows worsening outcomes—more mental health issues, flimsy excuses, need for accommodations. This drives anecdotal concerns (e.g., Stanford professors saying students are worse) but doesn’t reflect the tails.
PISA score declines are modest: Much of the measured decline comes from pushing more students into taking the tests. Adjusted scores are roughly constant—not great, but not catastrophic.
Selection effects explain anecdotes: The Stanford students Tyler meets are a highly selected group and remain impressive. Complaints about declining talent often reflect salience bias, not representative data.
Founder mode and talent clusters
Founders economize on courage: Founders have the authority and willingness to push through big changes (e.g., Meta’s pivots). This is scarce and valuable—it’s not just about intelligence but about the ability to act decisively.
The Beatles as a case study: Their greatness came from creative tension between two leaders (John then Paul), not stable equilibrium. They broke up, but those 7-8 years of tension produced extraordinary value. Four founders (Ringo included) with complementary skills.
Talent clusters are real but recursive: Patrick Collison hired Greg Brockman because Patrick is Patrick—talent recognizes talent. This doesn’t diminish the cluster’s value; it just pushes the explanation back a step. The Beatles, Stripe, and Nobel laureate labs all reflect this dynamic.
AI’s limits at the extreme: Even strong AI may not help much with the scarcest human achievements—the Beatles-level creative bundles. These remain “extreme human bottlenecks.”
Effective Altruism and Progress Studies
Tyler called peak EA before SBF’s collapse: He saw common patterns from observing movements since the 1960s—rapid growth, cult-like tendencies, secular/semi-religious tension, fragile institutional incentives. He predicted the movement would collapse while its best ideas would endure.
Progress studies should be decentralized: Tyler and Patrick designed it to be an ethos, not a formal organization. He hopes it has a “gentler but more enduring trajectory”—reflected in better science policy across countries.
Adverse selection risk: As progress studies becomes more popular, marginal proposals get worse. Tyler is raising his bar for “capital P, capital S” applications to Emergent Ventures.
What AI changes for Tyler personally
Writing for AIs: Tyler’s last book (Goat) and his next book are written primarily for AI systems, not human readers. He believes AIs will trawl and internalize this content, shaping how they see him. “You’re an idiot if you’re not writing for the AIs.”
Shift from content producer to connector: With AI handling more content production, Tyler sees his role shifting toward networking, connecting people, and developing intangibles that AIs can’t easily replicate.
The 75% AIs can’t capture: In Emergent Ventures interviews, Tyler can judge applicants in minutes based on vibe, priorities, and unspoken signals. A transcript captures maybe 25% of the value; the rest is intangible. He thinks AI will eventually learn some of this from recorded interviews, but not soon.
DC vs. SF vs. EU
Bay Area: Overvalues intelligence; thinks in terms of infinities (creative and destructive). Dynamic and ambitious but naive about bottlenecks and diffusion.
Washington DC: Thinks at the margin; much wiser about constraints but terrible for growth. If everyone thought like DC, “our world would end.”
EU (especially France): Cultured, wise, knowledgeable about history and art—but if they ruled the world, growth would be negative 1%.
The US balances these: The cultural mix of dynamism and constraint has worked well. The UK used to do this but has lost the balance.
War as the main risk from progress
Tyler’s primary concern with progress is its interaction with war. New technologies become instruments of war, and history shows terrible outcomes when this happens (17th-century England, WWI, nuclear weapons).
There may be a ratchet effect: wars become rarer but more destructive when they occur.
Steven Pinker’s relative peace is real but fraying at the edges; the numbers are moving in the wrong direction.
The optimistic note: Human agency matters. The fact that people in 1000 AD couldn’t imagine today’s world—even in poorer countries—shows how much progress is possible. The task is to take this heritage and do more with it.
