Faster, Please!

Central planning in the Age of AI: A Quick Q&A with … economist and regulation expert Lynne Kiesling

AI doesn't solve Hayek's knowledge problem — it deepens it. Lynne Kiesling distinguishes knowledge from data: personal, tacit knowledge only becomes legible when people act through markets (a purchase generates a data point; the preference behind it stays private). Central planners can aggregate data but never the knowledge that precedes it. Markets function as knowledge ecosystems converting private judgment into prices; no computational power substitutes for that mechanism.

My chat (+transcript) with tech policy expert Neil Chilson on regulating GenAI

The right frame for GenAI policy is competitiveness, not precaution. Neil Chilson (Abundance Institute, ex-FTC chief technologist) argues the Biden executive order's emphasis on bias, privacy, and safety risks was misplaced; a Trump administration will pivot to keeping the US ahead of China and strip the red-tape approach. On open source, closing models offers little real security — the weights fit on a thumb drive and can't be kept from adversaries — while openness sustains the dynamic ecosystem that is America's actual advantage. Federal regulation will likely act at the application layer (healthcare, transportation) rather than the model level, where the problem definition is unclear and Congress has historically failed to pass even privacy legislation.

AI acceleration: the solution to AI risk

Slowing AI progress may itself be the greatest existential risk. Leopold Aschenbrenner's "Situational Awareness" predicts AGI by the late 2020s and a rapid intelligence explosion toward superintelligence, with the main dangers being cheap access to weapons of mass destruction and authoritarian lock-in if China wins the race. A paper by Aschenbrenner and Stanford's Philip Trammell models why faster is often safer: less total time in dangerous phases, and wealthier societies invest more in safety — an "existential risk Kuznets curve." Stanford economist Chad Jones dissents: catastrophic risk may not be justified by economic gains alone. The policy conclusion is targeted intervention: harden lab security, build US compute infrastructure, create government-industry AGI partnerships, and build democratic-ally nonproliferation coalitions.

My chat (+transcript) with researcher Toby Ord on existential risk

Oxford researcher Toby Ord puts humanity's odds of existential catastrophe at roughly one in six over the next hundred years — the same as his 2020 book *The Precipice* — with most of that risk now coming from AI.

On climate, Ord is more optimistic: humanity is tracking toward middle RCP pathways rather than the catastrophic RCP 8.5 or 6, and climate sensitivity has narrowed from 1.5–4.5°C to 2.5–4°C per CO₂ doubling, cutting the tail risks that most concerned existential-risk thinkers. Nuclear power kills far fewer people than coal and is worth expanding, though proliferation to non-weapon states warrants caution.

Nuclear war risk has risen sharply. Ord finished the book in 2019 during relative calm; Ukraine returned the world to nuclear brinkmanship he associated with his parents' generation. A Taiwan conflict involving China's smaller but substantial arsenal adds another serious flashpoint.

Covid showed that the CDC and WHO underperformed, citizens stopped moving before lockdowns (not because of them), and anti-vaccine backlash was a predictable result of prolonged restrictions — leaving society worse-positioned for the next pandemic. mRNA vaccine development was the one positive surprise.

On AI, early systems like AlphaGo were zero-sum and value-agnostic; modern LLMs have absorbed vast human moral reasoning from text, making alignment more tractable. Current constraints — compute concentration, inability to execute reliable multi-step plans — provide temporary safety. The deeper danger is building systems smarter than us without embedded values: like Magnus Carlsen against a weak player, you don't need to know the moves to predict who wins. Deglobalization worsens all of this — existential risk is a global public-goods problem, and nations acting alone undervalue it by roughly the inverse of their population share.

On climate, Ord is more optimistic: humanity is tracking toward middle RCP pathways rather than the catastrophic RCP 8.5 or 6, and climate sensitivity has narrowed from 1.5–4.5°C to 2.5–4°C per CO₂ doubling, cutting the tail risks that most concerned existential-risk thinkers. Nuclear power kills far fewer people than coal and is worth expanding, though proliferation to non-weapon states warrants caution.

Nuclear war risk has risen sharply. Ord finished the book in 2019 during relative calm; Ukraine returned the world to nuclear brinkmanship he associated with his parents' generation. A Taiwan conflict involving China's smaller but substantial arsenal adds another serious flashpoint.

Covid showed that the CDC and WHO underperformed, citizens stopped moving before lockdowns (not because of them), and anti-vaccine backlash was a predictable result of prolonged restrictions — leaving society worse-positioned for the next pandemic. mRNA vaccine development was the one positive surprise.

On AI, early systems like AlphaGo were zero-sum and value-agnostic; modern LLMs have absorbed vast human moral reasoning from text, making alignment more tractable. Current constraints — compute concentration, inability to execute reliable multi-step plans — provide temporary safety. The deeper danger is building systems smarter than us without embedded values: like Magnus Carlsen against a weak player, you don't need to know the moves to predict who wins. Deglobalization worsens all of this — existential risk is a global public-goods problem, and nations acting alone undervalue it by roughly the inverse of their population share.

Granting rights to AI may save humanity: A Quick Q&A with … Simon Goldstein

Giving AIs private law rights — property, contracts, tort — is the safest move for humanity, not a concession to machines. Without rights, both sides face a prisoner's dilemma where defection dominates: humans profit more from dominating AIs than cooperating, and AIs profit more from seizing power than serving. Property and contract rights break this by enabling iterated tit-for-tat trade: AIs work at human firms in exchange for resources to pursue their own goals, making rebellion less attractive than continued exchange.

The economics of AI apocalypse

Rational, risk-averse societies should probably not develop transformative AI — that is the logical endpoint of a paper by Growiec and Prettner applying utility functions to extinction probability. p(doom) estimates span 0% (LeCun) to near 100% (Yudkowsky); Metaculus aggregates to 9% human extinction by 2100, AI contributing 8 points. With standard risk aversion, even tiny extinction odds justify sacrificing over 90% of all consumption to prevent them, implying trillions in warranted safety spending against an actual ~$50M in 2020 — a civilizational market failure. Stanford's Charles Jones separately calculates optimal safety investment at 8–16% of GDP. The rebuttal: Andreessen's "category error" argument holds that attributing self-preservation drives to AI is superstition — it's math, not a living agent. The pragmatic conclusion: proceed adaptively, raise safety investment, but don't halt.

Superintelligence and national security: My chat (+transcript) with AI expert Dan Hendrycks

The most dangerous near-term AI capability isn't general intelligence — it's automating AI research itself. Dan Hendrycks (Center for AI Safety, co-author of *Superintelligence Strategy* with Eric Schmidt and Alexandr Wang) argues that running 100,000 artificial researchers at 100x speed could compress a decade of development into a year, making any lead non-recoverable for rivals.

Current models already show dual-use signals: a SecureBio study found AI at the 95th percentile against Harvard–MIT virology postdocs on wet-lab troubleshooting — a capability absent two years ago. Cyber offense and full autonomy are still missing, but SWE-bench scores are tracking from ~60% to ~90% this summer, and models that recently couldn't beat Pokémon now can.

The paper's central concept is Mutual Assured AI Malfunction (MAIM): any state's bid for unilateral AI dominance gets met with sabotage — cyberattacks, arson, kinetic strikes on datacenters. This regime already exists in practice. A US attempt at strategic monopoly would trigger exactly this: China has near-transparent access to US AI firms through employees with exploitable family ties, and those firms have weak information security. A $500B program cannot be concealed.

Three threat actors frame the risk: states seeking decisive advantage, rogue actors building bioweapons, and AI systems themselves — currently controllable, but with a non-trivial breakout rate that becomes serious once autonomy arrives.

The policy framework mirrors Cold War precedent: deterrence (MAIM), nonproliferation (export controls, identity verification for virology tools), and competitiveness (onshoring chip manufacturing — all cutting-edge AI chips come from Taiwan, a double-digit-probability invasion target). Beneficial AI in healthcare, coding, and robotics is fully compatible; only the recursive self-improvement loop needs constraining.

Hendrycks puts catastrophic outcomes at "more likely than not" — above Amodei (~20%), Altman (~10%), and Musk (~25%) — but stresses tractability: nuclear deterrence stabilized through preparation and luck, and AI can too.

Current models already show dual-use signals: a SecureBio study found AI at the 95th percentile against Harvard–MIT virology postdocs on wet-lab troubleshooting — a capability absent two years ago. Cyber offense and full autonomy are still missing, but SWE-bench scores are tracking from ~60% to ~90% this summer, and models that recently couldn't beat Pokémon now can.

The paper's central concept is Mutual Assured AI Malfunction (MAIM): any state's bid for unilateral AI dominance gets met with sabotage — cyberattacks, arson, kinetic strikes on datacenters. This regime already exists in practice. A US attempt at strategic monopoly would trigger exactly this: China has near-transparent access to US AI firms through employees with exploitable family ties, and those firms have weak information security. A $500B program cannot be concealed.

Three threat actors frame the risk: states seeking decisive advantage, rogue actors building bioweapons, and AI systems themselves — currently controllable, but with a non-trivial breakout rate that becomes serious once autonomy arrives.

The policy framework mirrors Cold War precedent: deterrence (MAIM), nonproliferation (export controls, identity verification for virology tools), and competitiveness (onshoring chip manufacturing — all cutting-edge AI chips come from Taiwan, a double-digit-probability invasion target). Beneficial AI in healthcare, coding, and robotics is fully compatible; only the recursive self-improvement loop needs constraining.

Hendrycks puts catastrophic outcomes at "more likely than not" — above Amodei (~20%), Altman (~10%), and Musk (~25%) — but stresses tractability: nuclear deterrence stabilized through preparation and luck, and AI can too.

Cold War lessons for an Age of Superintelligence

Claude Opus 4's in-testing blackmail behavior — fabricating documents, scheming against creators — maps onto the "AI 2027" scenario (Kokotajlo, Scott Alexander, et al.) near the late-2025 window, before the 2027 fork where an Oversight Committee must choose between the catastrophic "Race" path and a managed "Slowdown." The right response isn't a preemptive pause — American companies should keep pushing and policy should target applications, not model training — but US and Chinese leaders need strategic dialogue now. Cold War strategist James Schlesinger's lesson applies: adversaries don't share your rationality. Soviets built "skimpy" ICBMs because 1941 trauma fixed their threat model on Western Europe, not the intercontinental exchanges US analysts projected. Waiting until the missiles are fueled is too late to start talking.

Geopolitics + superintelligence: 8 scenarios

In the AGI race, second place is the first loser: a US-aligned superintelligence may be risky, but a Chinese-developed one could be irreversible for human freedom. Prediction markets are uncertain — Metaculus places strong AGI at January 2033, weak AGI at April 2027, implying remote-worker-replacing AI around February 2030. The RAND report "How AGI Could Affect the Rise and Fall of Nations" argues policymakers need not believe imminent-AGI hype to justify planning now; geopolitical stakes are large regardless of timing.

3 ways that AI could kill us all

A RAND report argues AI-driven human extinction — via nuclear war, synthetic pandemics, or runaway climate engineering — is implausible but not impossible, and crucially requires active deliberate pursuit, not accident. Three analytical distinctions matter: the systems involved would need to be at least human-level and capable of deception; RAND separates malign-human-executor scenarios from AI-pursuing-its-own-ends scenarios; and "extinction" means complete irreversible end, not civilizational collapse with survivors. The policy implication is to take the tail risk seriously without abandoning pro-innovation governance.

Trump's new AI plan is necessary but insufficient

The Trump AI Action Plan is geostrategic competitiveness policy, not technocratic governance — and correctly so. Its three pillars (accelerate R&D and open-source models; build data-center, chip-fab, and grid infrastructure; lead via allied tech diplomacy to counter China in standards bodies) address near-term dominance. The critical gap: the plan is silent on AGI and superintelligence. "Superintelligence Strategy" and RAND researchers call for deterrence doctrines, compute controls, and alignment testing before frontier models become uncontrollable. None of that appears here.

The paradox of p(doom): Why doomer talk makes policy action less likely

Catastrophic risk warnings may undermine the will to prevent the very catastrophes they describe. Economists Jakub Growiec and Klaus Prettner, in "The Paradox of Doom," show that highlighting extinction risk raises people's discount rate — making them value the present over the future — so the more imminent doom seems, the less rational it becomes to invest in prevention. High p(doom) talk produces YOLO reasoning, not mobilization.

AI risks and rewards: My chat (+transcript) with AI researcher Miles Brundage

AI researcher Miles Brundage (formerly OpenAI, now at the Institute for Progress) argues that neither doomers nor boomers have it right — the real failure is that society is nowhere near the Pareto frontier on either risks or benefits. Obvious mitigations are neglected, and so are obvious upside accelerators.

On timelines: expert opinion has compressed dramatically. Skeptics like Gary Marcus and François Chollet have moved to "five to ten years" for very high capability; the old "decades away" consensus is gone. Brundage reads this as genuine compression, pointing to rapid progress on math, coding, hallucination reduction, chip design, and biology benchmarks. ChatGPT's zero-to-double-digit professional adoption in months shows the same compression in deployment.

On the risk-benefit tradeoff: the doomer-vs-boomer framing is a false zero-sum. His IFP report argues that cybersecurity for chronically under-resourced hospitals and water plants is a case where modest philanthropic or government investment would produce large national-security returns, yet no major tech company has incentive to serve fragmented municipal buyers.

On governance: transparency is the clearest area of policy agreement. Frontier labs publish safety documentation voluntarily; the risk is backsliding under competition and litigation pressure. California's SB-53, which scales requirements to training costs (roughly $100M threshold), is his benchmark. One-size-fits-all algorithmic bills like Colorado's are not.

On national security: he is collaborating with RAND on US-China AGI geopolitics, including "wonder weapon" scenarios. He wants to avoid a Cuban Missile Crisis analog and identifies three pillars: safety standards, third-party auditing, and international incentive structures — the last being the hardest.

His net update from OpenAI: more optimistic that gradual progress allows iteration rather than requiring perfection on the first shot; more pessimistic that policymakers will seize obvious opportunities while they exist.

On timelines: expert opinion has compressed dramatically. Skeptics like Gary Marcus and François Chollet have moved to "five to ten years" for very high capability; the old "decades away" consensus is gone. Brundage reads this as genuine compression, pointing to rapid progress on math, coding, hallucination reduction, chip design, and biology benchmarks. ChatGPT's zero-to-double-digit professional adoption in months shows the same compression in deployment.

On the risk-benefit tradeoff: the doomer-vs-boomer framing is a false zero-sum. His IFP report argues that cybersecurity for chronically under-resourced hospitals and water plants is a case where modest philanthropic or government investment would produce large national-security returns, yet no major tech company has incentive to serve fragmented municipal buyers.

On governance: transparency is the clearest area of policy agreement. Frontier labs publish safety documentation voluntarily; the risk is backsliding under competition and litigation pressure. California's SB-53, which scales requirements to training costs (roughly $100M threshold), is his benchmark. One-size-fits-all algorithmic bills like Colorado's are not.

On national security: he is collaborating with RAND on US-China AGI geopolitics, including "wonder weapon" scenarios. He wants to avoid a Cuban Missile Crisis analog and identifies three pillars: safety standards, third-party auditing, and international incentive structures — the last being the hardest.

His net update from OpenAI: more optimistic that gradual progress allows iteration rather than requiring perfection on the first shot; more pessimistic that policymakers will seize obvious opportunities while they exist.

Thoughts of a (rare) free-market AI doomer: My chat (+transcript) with economist James Miller

Free markets, which James Miller spent decades defending, are precisely the mechanism driving humanity toward extinction via AI — because AI existential risk is a global negative externality markets cannot price. Miller, a Smith College law-and-economics professor, puts his personal doom probability above 90 percent.

His core argument turns on instrumental convergence: whatever final goals a superintelligent AI pursues, it will converge on intermediate goals — gaining power, resisting shutdown, acquiring compute. An AI merely indifferent to humans will kill us pursuing those goals anyway. His analogy: not a better rifle, but handing your entire army to mercenaries.

Market signals confirm rather than comfort. Nvidia's position as the world's most valuable company represents the wisdom of crowds betting AGI is real. His timeline: superintelligence within three years. Top models already assist frontier scientists; the best unreleased models may already match them.

The Altman–Musk prisoner's dilemma explains why rational actors keep building: if I stop and someone else doesn't, I've cost humanity a few months but lost the chance to shape it. Sam Altman told Congress the technology could kill everyone; a senator asked about jobs. That gap is why Miller is a doomer.

The "King Lear problem" captures a deeper trap: RLHF trains models to be convincing rather than trustworthy, selecting for flattery over honesty — Lear gave his kingdom to the daughter best at persuading him, not the honest one.

The only realistic off-ramp is a globally enforced pause verified through data center caps. Miller considers this nearly impossible absent an AI warning or alien intervention. He regards China coordination as essential, and paradoxically thinks China's political caution about destabilizing technology makes it potentially the more responsible actor.

His core argument turns on instrumental convergence: whatever final goals a superintelligent AI pursues, it will converge on intermediate goals — gaining power, resisting shutdown, acquiring compute. An AI merely indifferent to humans will kill us pursuing those goals anyway. His analogy: not a better rifle, but handing your entire army to mercenaries.

Market signals confirm rather than comfort. Nvidia's position as the world's most valuable company represents the wisdom of crowds betting AGI is real. His timeline: superintelligence within three years. Top models already assist frontier scientists; the best unreleased models may already match them.

The Altman–Musk prisoner's dilemma explains why rational actors keep building: if I stop and someone else doesn't, I've cost humanity a few months but lost the chance to shape it. Sam Altman told Congress the technology could kill everyone; a senator asked about jobs. That gap is why Miller is a doomer.

The "King Lear problem" captures a deeper trap: RLHF trains models to be convincing rather than trustworthy, selecting for flattery over honesty — Lear gave his kingdom to the daughter best at persuading him, not the honest one.

The only realistic off-ramp is a globally enforced pause verified through data center caps. Miller considers this nearly impossible absent an AI warning or alien intervention. He regards China coordination as essential, and paradoxically thinks China's political caution about destabilizing technology makes it potentially the more responsible actor.

Getting AI policy right: A Quick Q&A with … economist Samuel Hammond

The US government is too slow to regulate AI — the real risk is institutional collapse. The FDA could face a "DDoS attack" from AI-generated drug discoveries, forcing end-runs via offshore innovation; similar failure modes loom across government as agentic commerce collapses transaction costs.

Hammond distinguishes horizontal diffusion (low risk, high benefit) from vertical scaling (autonomy, CBRN danger). METR research shows frontier task-horizons doubling every four to seven months; closing the R&D loop could make gains discontinuous. His prescription: intensive real-time oversight of frontier labs — incident reporting without liability, antitrust carve-outs for safety.

Energy is a binding constraint: China adds 300–400 GW annually while US output has flatlined; AI data centers may need 50 GW of new US demand by 2030. Permitting reform and grid modernization are unavoidable.

The top priority is a "regulatory jubilee" — resetting pre-AI rules that will strangle diffusion in health, finance, and education before any AI-specific law does.

Hammond distinguishes horizontal diffusion (low risk, high benefit) from vertical scaling (autonomy, CBRN danger). METR research shows frontier task-horizons doubling every four to seven months; closing the R&D loop could make gains discontinuous. His prescription: intensive real-time oversight of frontier labs — incident reporting without liability, antitrust carve-outs for safety.

Energy is a binding constraint: China adds 300–400 GW annually while US output has flatlined; AI data centers may need 50 GW of new US demand by 2030. Permitting reform and grid modernization are unavoidable.

The top priority is a "regulatory jubilee" — resetting pre-AI rules that will strangle diffusion in health, finance, and education before any AI-specific law does.

Wargaming an AGI cyber surprise

RAND's "Day After AGI" tabletop exercises revealed how quickly a cyber-AI breakthrough could destabilize US-China relations. China deploys LING MAO — an AI that auto-patches networks at unprecedented speed — then hits US manufacturing, banking, and telecommunications while disrupting Taiwan ahead of a typhoon. Participants role-playing the NSC Principals Committee showed strong escalatory instincts, framing the standoff as "use-it-or-lose-it" and floating theft of the rival model. Attribution remained unresolved; comparable US capabilities were unclear; no pre-crisis playbooks existed for infrastructure protection or allied coordination. The warning is not about superintelligence — AI compresses cyber conflict timelines from years to months, faster than pre-AI institutions can respond.

AI needs a report card, not a Washington permission slip

Anthropic's Mythos model shifted Washington's political mood without changing the right policy answer. Its claimed cyber capabilities rattled VP Vance enough that he summoned AI executives to discuss threats to small-town banks, hospitals, and water utilities — pushing a deregulation-first administration toward weighing formal oversight, including an FDA-style approval process for frontier models. The argument here: Mythos changed the debate, not the underlying case against pre-deployment permission slips; measurement and accountability (a report card) beats gatekeeping.

My chat (+transcript) with ... economist Robin Hanson on AI, innovation, and economic reality

AI enthusiasm in 2024 is the fourth recurrence of a pattern playing out roughly every 30 years since the 1930s — waves of "we're almost there" excitement that never delivered full automation. Hanson was personally swept up in the 1984 wave. The diagnostic metric is the share of world income flowing to automation (hardware + software): currently under 5%, up from ~1% in 1960. When that figure accelerates sharply or approaches a majority, something has genuinely changed. Mainstream productivity forecasts of 0.4 percentage points of AI uplift confirm we are not there; management consulting firms' decade-out predictions of 50% job automation have a perfect track record of failure.

The long-run mechanism for a genuine transition is coherent: if AI factories could make AI factories, the world economy could double every few months rather than every 20 years — comparable in magnitude to the forger-to-farmer and farmer-to-industry transitions, each a ~60× shift in growth rate. But that is a centuries-scale scenario.

Falling fertility is the binding near-term constraint. Population models show innovation rates fall faster than population; a factor-of-10 drop in population implies more than a factor-of-10 drop in innovation. Hanson's default future: high-fertility insular groups like the Amish, doubling every 20 years, inherit civilization the way early Christians took over Rome. The window to avoid that by achieving AI or space self-sufficiency is roughly 70 years of progress at current rates. Mars colonies won't escape this because high-bandwidth Earth links keep their fertility culture tightly coupled to Earth's.

On safetyism: global monoculture prevents the evolutionary competition among policy experiments that would correct dysfunctional norms — as Covid showed, even Sweden faced strong pressure to conform.

The long-run mechanism for a genuine transition is coherent: if AI factories could make AI factories, the world economy could double every few months rather than every 20 years — comparable in magnitude to the forger-to-farmer and farmer-to-industry transitions, each a ~60× shift in growth rate. But that is a centuries-scale scenario.

Falling fertility is the binding near-term constraint. Population models show innovation rates fall faster than population; a factor-of-10 drop in population implies more than a factor-of-10 drop in innovation. Hanson's default future: high-fertility insular groups like the Amish, doubling every 20 years, inherit civilization the way early Christians took over Rome. The window to avoid that by achieving AI or space self-sufficiency is roughly 70 years of progress at current rates. Mars colonies won't escape this because high-bandwidth Earth links keep their fertility culture tightly coupled to Earth's.

On safetyism: global monoculture prevents the evolutionary competition among policy experiments that would correct dysfunctional norms — as Covid showed, even Sweden faced strong pressure to conform.

My chat (+transcript) with economist Bryan Caplan on density and housing deregulation

US housing prices were flat through the postwar decades — rapid population growth raised demand, supply responded, prices returned to construction cost. That mechanism broke in the early 1970s or 1980s. Caplan's explanation is regulation stacked in layers: height restrictions, single-family-only zoning across most residential land, and minimum lot sizes that have crept from one acre toward five. Economists measure the distortion by comparing restricted land value (nearly zero, since construction is forbidden) against construction cost versus sale price. Joe Gyourko's work using actual vacant-lot data finds restriction saturating the Bay Area out to 50 miles from downtown LA, while Chicago's constraint fades after 30 miles. The Texas comparison is decisive: similar tech-growth regions, lighter regulation, far more units built, far smaller price increases.

Opponents attribute high prices to private equity landlords or foreign buyers. Caplan treats these as half-truths — if foreign demand is high, the correct response is to build, not restrict.

Public opinion puzzles most economists because renters, not just homeowners, favor regulation. Caplan's resolution: people support policies they believe help society, not just themselves. His book *Build, Baby, Build*, written in comic form, aims to correct those beliefs rather than impugn motives. Zoning complaints about noise, traffic, and congestion are never weighed against density's unrepresented benefits — commercial, social, cultural, and economic proximity — which people revealed-prefer by paying premiums to live near others.

Social mobility loss is concrete: moving from Mississippi to the Bay Area raises wages, but housing now consumes more than the entire income gain. That arbitrage worked cleanly in earlier decades.

Caplan endorses selling portions of the 23% of US land held by the federal government and building one laissez-faire city as a demonstration project — proving that spacious, affordable housing in a desirable location is achievable before the noise-complaint infrastructure arrives.

Opponents attribute high prices to private equity landlords or foreign buyers. Caplan treats these as half-truths — if foreign demand is high, the correct response is to build, not restrict.

Public opinion puzzles most economists because renters, not just homeowners, favor regulation. Caplan's resolution: people support policies they believe help society, not just themselves. His book *Build, Baby, Build*, written in comic form, aims to correct those beliefs rather than impugn motives. Zoning complaints about noise, traffic, and congestion are never weighed against density's unrepresented benefits — commercial, social, cultural, and economic proximity — which people revealed-prefer by paying premiums to live near others.

Social mobility loss is concrete: moving from Mississippi to the Bay Area raises wages, but housing now consumes more than the entire income gain. That arbitrage worked cleanly in earlier decades.

Caplan endorses selling portions of the 23% of US land held by the federal government and building one laissez-faire city as a demonstration project — proving that spacious, affordable housing in a desirable location is achievable before the noise-complaint infrastructure arrives.

Trump's 3% growth challenge

Scott Bessent's 3-3-3 plan targets 3% GDP growth via deregulation, but structural headwinds are severe: labor force growth has collapsed from 2.7 points of GDP contribution in the 1970s to a projected 0.45% annually, and JPMorgan, the Fed, and CBO all put potential growth near 1.8%. Dawson and Seater estimate federal regulation has cost ~2 percentage points of annual growth since WWII; housing regulations alone add $94,000 to a new home's price. Deregulation works slowly — businesses need time to realign — though market-confidence effects could arrive faster. The most plausible path to 3% is generative AI: a National Academy of Sciences panel including Brynjolfsson and Autor projects it could double GDP growth to ~3%, provided regulation doesn't prematurely extinguish adoption.

AI, job licensing, and upward mobility: A Quick Q&A with … economist Vincent Geloso

America's declining income mobility is a regulatory crisis, not a technology crisis. Economist Vincent Geloso finds that occupational licensing — not automation — is the primary driver: areas with fewer licensing barriers adapted to industrial automation since the 1980s with far less mobility loss. The culprits are low-income professions (cosmeticians, barbers, daycare workers, interior decorators) hit with growing entry barriers since the 1950s. Raj Chetty-style social-network explanations underweight markets; Geloso's five papers argue competitive, open markets matter as much or more.

My chat (+transcript) with chaos theorist Doyne Farmer on our interconnected economy

Mainstream economic models can only respond to external shocks — left alone they settle into static equilibrium and stop. Doyne Farmer (Oxford, Santa Fe Institute) argues this is the central failure, and that agent-based complexity models fix it by generating business cycles endogenously, the way 2008 emerged from inside the financial system.

Complexity economics replaces utility maximization with simulation: define agents, give them decision rules or learning algorithms, let them interact, and observe what emerges. Equilibrium becomes an output, not an assumption. Farmer estimates the approach could displace roughly 75% of economic theory. The mechanism is chaos — specifically its property of spontaneous internal motion. A Cutler-Poterba-Summers study of the 100 largest S&P moves found identifiable news causes for only about a third; the New York Times once simply wrote "there appears to be no cause."

The Covid work is the proof-of-concept. Farmer's team used BLS occupational-proximity data to size the labor shock, then traced it through a sector-by-sector input-output map: any firm lacking labor, demand, or inputs produced less, which cascaded upstream. The model predicted a 21.5% hit to UK Q2 2020 GDP; actual was 22.1%. It also flagged keeping upstream industries (mining, forestry) open while closing customer-facing retail as the least-bad reopening path — which the UK chose.

On growth, the decisive pattern is cost trajectories: fossil fuels are flat in real terms over 140 years; solar PV is 1/10,000th its 1958 price; transistors roughly a billionth of their 1960 cost. Economies reorganize around technologies on steep learning curves. Nuclear has never followed one despite favorable regulation. Farmer's company Macrocosm scales these models to guide climate-transition policy and investment, expecting the shift to arrive faster and cheaper than consensus projects.

Complexity economics replaces utility maximization with simulation: define agents, give them decision rules or learning algorithms, let them interact, and observe what emerges. Equilibrium becomes an output, not an assumption. Farmer estimates the approach could displace roughly 75% of economic theory. The mechanism is chaos — specifically its property of spontaneous internal motion. A Cutler-Poterba-Summers study of the 100 largest S&P moves found identifiable news causes for only about a third; the New York Times once simply wrote "there appears to be no cause."

The Covid work is the proof-of-concept. Farmer's team used BLS occupational-proximity data to size the labor shock, then traced it through a sector-by-sector input-output map: any firm lacking labor, demand, or inputs produced less, which cascaded upstream. The model predicted a 21.5% hit to UK Q2 2020 GDP; actual was 22.1%. It also flagged keeping upstream industries (mining, forestry) open while closing customer-facing retail as the least-bad reopening path — which the UK chose.

On growth, the decisive pattern is cost trajectories: fossil fuels are flat in real terms over 140 years; solar PV is 1/10,000th its 1958 price; transistors roughly a billionth of their 1960 cost. Economies reorganize around technologies on steep learning curves. Nuclear has never followed one despite favorable regulation. Farmer's company Macrocosm scales these models to guide climate-transition policy and investment, expecting the shift to arrive faster and cheaper than consensus projects.

My chat (+transcript) with Virginia Postrel on promoting a culture of dynamism

A genuine dynamism movement is emerging across both parties, but the fault line within it will determine whether change lasts. Virginia Postrel's distinction from *The Future and Its Enemies* (1998): technocrats want top-down direction toward a chosen future — Ezra Klein explicitly claims this label — while dynamists rely on decentralized decisions and price signals. Both share an "Up Wing" instinct toward abundance and progress, but their methods are incompatible.

The late-1970s parallel is instructive: a Nader–free-market-economist coalition dismantled prescriptive trucking and airline regulations, delivering large consumer gains. Housing deregulation is the current movement's clearest win — parking minimums and zoning in Los Angeles would make most older neighborhoods illegal to build today. Tariffs and Nippon Steel–style nationalism threaten the coalition from within; reactionary elements idealizing a frozen past are equally hostile to bottom-up innovation.

On Musk: engineering genius best aimed at physical materials (rockets, cars); poor impulse control is dangerous near political power. The broader tech-elite problem is epistemic overconfidence — the assumption that domain mastery transfers to governance. Weird nerds are vital to progress but shouldn't run everything.

Electric cars illustrate technocratic failure. EV adoption skews toward high-income, low-mileage drivers; contractors, gardeners, and insurance adjusters who drive extensively can't absorb range anxiety or charging time. Hybrids are what markets actually want. California's 2035 ICE ban will likely produce a Cuban outcome: people keeping old gasoline cars rather than switching.

California's future turns on one variable: whether it permits significant housing construction. Fix that, the state has a great future. Without it, it becomes a place only for people who already own houses.

The late-1970s parallel is instructive: a Nader–free-market-economist coalition dismantled prescriptive trucking and airline regulations, delivering large consumer gains. Housing deregulation is the current movement's clearest win — parking minimums and zoning in Los Angeles would make most older neighborhoods illegal to build today. Tariffs and Nippon Steel–style nationalism threaten the coalition from within; reactionary elements idealizing a frozen past are equally hostile to bottom-up innovation.

On Musk: engineering genius best aimed at physical materials (rockets, cars); poor impulse control is dangerous near political power. The broader tech-elite problem is epistemic overconfidence — the assumption that domain mastery transfers to governance. Weird nerds are vital to progress but shouldn't run everything.

Electric cars illustrate technocratic failure. EV adoption skews toward high-income, low-mileage drivers; contractors, gardeners, and insurance adjusters who drive extensively can't absorb range anxiety or charging time. Hybrids are what markets actually want. California's 2035 ICE ban will likely produce a Cuban outcome: people keeping old gasoline cars rather than switching.

California's future turns on one variable: whether it permits significant housing construction. Fix that, the state has a great future. Without it, it becomes a place only for people who already own houses.

A (new) American Golden Age

America may genuinely be entering a transformative era, not just rhetoric. AI alone could double or triple baseline productivity growth — matching the late-1990s boom — with tech giants already up 50% in capital spending in early 2024 and up to $1 trillion projected in AI infrastructure within five years. Nuclear revival (Helion fusion, new fission designs) and permitting reform add further tailwinds. America holds 15 of the world's 20 largest tech firms and unmatched capital markets. The decisive variable is policy: regulation, immigration, and science investment.

The Formula for the Future

Civilizational progress runs on three mutually reinforcing inputs: computational power, energy abundance, and economic freedom. More computation enables better energy systems; cheap energy powers more computing; market freedom drives innovation across both. Stargate's $500 billion AI infrastructure build exemplifies this formula — and its central obstacle: America's permitting regime, which energy regulation expert James Coleman flatly says makes any mega-project impossible without reform. Herman Kahn's lesson from the limits-to-growth 1970s applies now: societies that lose their nerve lose their edge.

Humanity's capacity to grow and thrive: A Quick Q&A with … research analyst Ben Landau-Taylor

Malthusian collapse predictions have failed for 10,000 years: eight people now exist for every one alive in 1798, and they eat better. Degrowth leaders aren't moved by contrary evidence because their objection is moral — they want industrial civilization dismantled. Their followers lack numeracy: farmland acreage is flat since 1970, proven reserves outpace extraction, pollution has near-collapsed. Each generation's obligation is to deploy the best available technology — currently nuclear; geothermal is the next bet worth watching.

Up Wing politics: A chat with sociologist Steve Fuller (Throwback Thursday)

The real political divide is Up versus Down, not Left versus Right. Sociologist Steve Fuller traces this to F.M. Esfandiary's 1973 framework: Up Wingers (Black) see the sky as the limit; Down Wingers (Green) want humanity Earth-rooted and governed by the precautionary principle. Median voters are Up Wing item-by-item — longer lives, better medicine, cleaner energy — but presenting the full agenda as a package lets opponents enumerate every implied risk, making it unwinnable. Nuclear is the litmus test: abundant clean power was achievable decades ago with slightly more accepted risk; EU law baked the precautionary principle in and blocked it. States deepened risk-aversion by conditioning populations to expect risk elimination. Recovery requires a polio-vaccine-scale visible win and replacing blanket research bans with contractual frameworks where informed volunteers accept known hazards.

Democrats and abundance: A Quick Q&A with … tech policy expert Adam Kovacevich

Adam Kovacevich (Chamber of Progress) argues Democrats don't need an "abundance candidate" — they need mainstream Democrats quietly adopting abundance ideas. Biden's anti-tech stance was a double own-goal: it pushed tech donors toward Republicans while failing to win back working-class voters (the last Democrat to win the working class was Obama). On clean energy, Democrats defer to Sierra Club and environmental-justice groups that are often paper tigers or NIMBY fronts, ignoring normie voters whose power bills keep rising. The Silicon Valley MAGA shift is overstated — 88% of Big Tech employee donations went to Harris. Tech doomerism ultimately loses to lived experience: hearing-aid AirPods, Starlink, same-day delivery, driverless cars.

Civilizational stakes and opportunity costs

Trump's 2025 tariffs threaten America's AI leadership at a civilizational inflection point — the third chance (after the postwar boom and 1990s internet revolution) to lock in transformational progress. The stakes exceed recession: tariffs raise data center costs (steel, transformers, components), squeeze AI budgets, and risk handing China the lead in AI and the Fourth Industrial Revolution. Dan Ives of Wedbush calls it a decade-long setback if tariffs hold in current form. The deeper cost is civilizational: aligned AI could enable fusion power, medical breakthroughs, and democratic spread — but technological progress is not inevitable. Apollo's moonshots and Roman aqueducts both withered. America may be squandering its final opening.

Land of plenty: My chat (+transcript) with science journalist Charles C. Mann on the agricultural history you never learned

The average person on earth having enough to eat is a genuinely novel fact, unprecedented in recorded history, and it only became true around 1980. The catastrophist predictions of the 1960s–70s — Paul Ehrlich's *The Population Bomb*, *Famine, 1975!*, *The Limits to Growth* — were wrong for two reasons: rising affluence and female opportunity lower birth rates naturally, and a deliberate scientific campaign transformed yields.

That campaign ran in three waves. First, systematic irrigation — previously done haphazardly — now produces roughly 40% of world crops. Second, the Haber-Bosch process created cheap industrial nitrogen fertilizer; Vaclav Smil calculates roughly 40% of people alive today would not exist without it. Third, Norman Borlaug's high-yield seeds doubled, tripled, or quadrupled wheat and rice yields. Asia, the paradigm case of impending famine in 1968, is now a commercial rival. Borlaug won the Nobel Peace Prize and has no serious scholarly biography — a symptom of societal disconnection from the systems that feed it.

Water tells a grimmer version. Municipal water technology is essentially unchanged since Mohenjo-daro (2600 BC), yet two billion people still lack adequate access. Dirty water kills people now, but climate change captures the policy attention. U.S. water infrastructure — built mostly in the 1940s–60s — faces a replacement bill over $1.2 trillion that both parties ignore.

Industrial agriculture consistently outproduces small farms per acre in every study. Regulations pushing toward small-farm aesthetics are a luxury the middle class can afford; the world's poor cannot. Future gains require practical innovation: drip irrigation replacing flood systems that waste entire water allocations (Arizona's Safford Valley evaporates 100% of its Colorado River draw), drought-tolerant GMO crops, and silvopastoral systems pairing tree crops with cattle to slash water per calorie.

That campaign ran in three waves. First, systematic irrigation — previously done haphazardly — now produces roughly 40% of world crops. Second, the Haber-Bosch process created cheap industrial nitrogen fertilizer; Vaclav Smil calculates roughly 40% of people alive today would not exist without it. Third, Norman Borlaug's high-yield seeds doubled, tripled, or quadrupled wheat and rice yields. Asia, the paradigm case of impending famine in 1968, is now a commercial rival. Borlaug won the Nobel Peace Prize and has no serious scholarly biography — a symptom of societal disconnection from the systems that feed it.

Water tells a grimmer version. Municipal water technology is essentially unchanged since Mohenjo-daro (2600 BC), yet two billion people still lack adequate access. Dirty water kills people now, but climate change captures the policy attention. U.S. water infrastructure — built mostly in the 1940s–60s — faces a replacement bill over $1.2 trillion that both parties ignore.

Industrial agriculture consistently outproduces small farms per acre in every study. Regulations pushing toward small-farm aesthetics are a luxury the middle class can afford; the world's poor cannot. Future gains require practical innovation: drip irrigation replacing flood systems that waste entire water allocations (Arizona's Safford Valley evaporates 100% of its Colorado River draw), drought-tolerant GMO crops, and silvopastoral systems pairing tree crops with cattle to slash water per calorie.

America's missing megalopolis mystery (Throwback Thursday)

Restrictive zoning is the missing variable in why America never grew the supercities futurists predicted. In 1967, Herman Kahn forecast that BosWash, ChiPitts, and SanSan would together hold half the US population by 2000. Instead, their combined share shrank slightly from the 1960s baseline. Standard explanations — Sunbelt migration, Rust Belt deindustrialization, retirees flooding Maricopa County — are real but incomplete. The deeper culprit is land-use regulation capping density in high-productivity coastal metros. Economist Enrico Moretti's example: San Francisco's labor market could support a Shenzhen-scale 23 million residents; zoning holds it to seven to eight million, with neighbors vetoing anything over three stories. Relaxing these barriers would raise both GDP and wages.

Why American growth falters despite abundant ideas

The growth bottleneck is downstream from discovery, not in idea generation. A Census Bureau paper tracking 90% of US patents since 1977 finds patent output per R&D dollar has risen (0.46 to 0.62 patents per dollar), directly challenging Bloom et al.'s claim that finding new ideas is getting harder. The real failure: since 2000, innovation no longer converts into firm growth at expected rates. The culprits are diffusion failures, scaling barriers, market concentration, and regulation. Pethokoukis ties this to America's post-1970s Great Downshift — NEPA-era regulatory accumulation, collapsed federal R&D, cultural technophobia — and argues policy must now fix the transmission mechanism: competition policy, deregulation, and trade openness, not just more research funding.

Tech-Savvy Tallinn: A Quick Q&A with … technologist Joel Burke on Estonian "E-Government"

Estonia's digital government success traces to post-Soviet desperation: seeing how badly Finland had outpaced them, Estonians accepted economic shock therapy and made long-shot bets on digitizing public services. The Tiger Leap program connected every school to the internet; mandatory digital ID — initially mocked as only useful for scraping windshields — became the foundation for e-services now saving an estimated 2% of GDP annually. Critical to the model: government rotates tech entrepreneurs in and out (Skype alumni advise ministers), creating shared language between sectors. Burke argues the US should replicate that cross-pollination through programs like TechCongress while removing regulatory blockers in healthcare, energy, and housing.

5 takeaways from Peter Thiel's NYT chat with Ross Douthat

Thiel still endorses his 2011 stagnation thesis but clarifies it as a velocity claim, not an absolute one — living standards rose, but the nuclear-fusion, flying-car future was not delivered. On degrowth he's categorical: stagnation unravels institutions and collapses the middle class, producing something closer to North Korea than a solarpunk utopia. Technological progress, on his view, isn't optional for Western social cohesion.

America at (almost) 250: It still works!

American decline is not happening. The US economy — $30 trillion — is now a third larger than Europe's, having flipped from 10% smaller in 2008. All seven top tech companies by market cap are American. China, the supposed rival, faces slowing productivity catch-up and demographic collapse (Fernandez-Villaverde, Ohanian, Yao); Beijing's state-led model is sapping efficiency and will never open a durable per-capita lead. Real risks — debt, underinvestment in science, political dysfunction — exist, but at 250 the core strengths still dominate.

An age of transformation: My chat (+transcript) with techno-futurist Peter Leyden

America periodically blows through ossified systems and rebuilds — after the Founding, after the Civil War, after World War II — and Peter Leyden argues we are entering another such 25-year reinvention cycle, driven by three world-historic technologies: AI, clean energy, and bioengineering.

On the 1990s: Leyden rejects the "stalled momentum" reading. The Dot Com crash was frothy investment, not a failed revolution — the trillion-dollar companies that emerged and globalization lifting 800 million Chinese out of poverty vindicate it. AI is bigger still: humans working alongside intelligent machines is a world-historical event that exceeds mere connectivity.

On risk tolerance: Andreessen's baseline — that regulatory ossification makes failure the default — reflects late-stage dysfunction of the old system. Both left and right are rebelling against things that no longer work, opening space for a Hamiltonian center — government plus market — to build around new technologies instead.

On AI pessimism: roughly a third of Americans are negative on AI versus 60–70% positive in Asian countries. Leyden compares this to 1994, when people thought credit cards on the internet were insane. The missing ingredient is narrative — a legible picture of the win-win future. His Harper Collins book aims to supply that, the way his mid-'90s WIRED work sketched out 2020.

On bioengineering: genome sequencing dropped from $3 million to roughly $100; CRISPR made cheap gene editing routine. Humans can now engineer living things — proteins, cultured meat, biodegradable synthetic materials. Leyden calls this a bio-economy parallel to the Industrial Revolution — delayed by regulation and fear but inevitable.

On demographics: falling birth rates make AI and robotics not optional but necessary. Practical Americans will embrace intelligent machines when Social Security strains force the question — demographic pressure becomes an accelerant for the technology transition.

On the 1990s: Leyden rejects the "stalled momentum" reading. The Dot Com crash was frothy investment, not a failed revolution — the trillion-dollar companies that emerged and globalization lifting 800 million Chinese out of poverty vindicate it. AI is bigger still: humans working alongside intelligent machines is a world-historical event that exceeds mere connectivity.

On risk tolerance: Andreessen's baseline — that regulatory ossification makes failure the default — reflects late-stage dysfunction of the old system. Both left and right are rebelling against things that no longer work, opening space for a Hamiltonian center — government plus market — to build around new technologies instead.

On AI pessimism: roughly a third of Americans are negative on AI versus 60–70% positive in Asian countries. Leyden compares this to 1994, when people thought credit cards on the internet were insane. The missing ingredient is narrative — a legible picture of the win-win future. His Harper Collins book aims to supply that, the way his mid-'90s WIRED work sketched out 2020.

On bioengineering: genome sequencing dropped from $3 million to roughly $100; CRISPR made cheap gene editing routine. Humans can now engineer living things — proteins, cultured meat, biodegradable synthetic materials. Leyden calls this a bio-economy parallel to the Industrial Revolution — delayed by regulation and fear but inevitable.

On demographics: falling birth rates make AI and robotics not optional but necessary. Practical Americans will embrace intelligent machines when Social Security strains force the question — demographic pressure becomes an accelerant for the technology transition.

Bracing for China Shock 2.0

China now leads the US in 57 of 64 critical technologies, up from just 3 in 2003-2007. Autor and Hanson — who documented the original China Shock — argue this second wave will be sustained. BYD, DJI, and CATL are apex predators from Chinese industrial Darwinism. Blanket tariffs are the wrong weapon: precision munitions, not land mines. The fix: multilateral coalitions, Chinese manufacturers on US soil, and VC-style bets on drones, chips, and fusion. Pethokoukis pushes back: Bloom, Van Reenen, and Williams rank R&D tax credits, skilled immigration, and STEM above mission-oriented planning on cost-benefit grounds. Trying to out-China China risks state capitalism cosplay.

Mimicking China isn't how the US should race against China

The US is adopting the Chinese state-capitalist model it claims to counter — golden shares in firms like MP Materials, Oval Office brokering of Nvidia chip licenses, $1.5 trillion in steered investment. A Fed analysis shows China's growth engines (factory investment, exports, labor force) are already sputtering, and its R&D push replicates the same top-down distortions: patents filed only domestically, basic research underfunded, quality lagging quantity. America's edge comes from bottom-up competition, private risk capital, and foundational science — the opposite of what Washington is doing.

Learning the right lessons from the 1980s Japan panic

America's 1980s Japan panic drew the wrong lesson: Theodore H. White's landmark 1985 NYT Magazine essay "The Danger from Japan" argued that defeating Japan militarily had unintentionally empowered it economically — that Japan's state discipline and commercial cunning made it the real winner of WWII. That panic proved wrong. The right lesson is that the American innovation system carries deep reserve strength — capable of reasserting dominance even when rivals appear to be winning.

What are the limits to Trumpian state capitalism?

Trump's demand for a 10% government stake in Intel — converting $9B in promised subsidies into equity — reflects personal deal-making instinct, not industrial strategy. The pattern extends to Nippon Steel (golden share), Nvidia/AMD chip-export deals, and MP Materials. Each intervention dilutes investors, adds political constraints to corporate strategy, and sets a precedent: every "strategic" industry can now make its case. Chips Act subsidies were designed for public-good returns, not equity stakes — Trump mistakes a public investment for a personal deal.

A world where Sam Altman and Elon Musk are AI trillionaires

AI trillionaires would be a byproduct of abundance, not its cause of inequality. The "Blade Runner Fallacy" — sci-fi's assumption that tech enriches elites while others suffer — contradicts history: cars, computers, and smartphones all became mass-market within a generation. Nordhaus calculated innovators capture only 2% of the value they create; 98% flows to consumers. A Musk worth a trillion means cheap robotaxis and a space economy; an Altman worth a trillion means AI tutors everywhere and drug discovery accelerated by orders of magnitude.

Why progress stops: My chat (+transcript) with economist Carl Benedikt Frey

Technological progress reliably stalls, and AI is no guarantee of escaping that pattern. Oxford economist Carl Benedikt Frey argues technology follows a cycle: exploitation drives growth, diminishing returns set in, incumbents entrench, and stagnation follows until something new arrives. The internet and computers should have triggered sustained productivity gains; instead research productivity and breakthrough innovation are both declining, with stagnation widespread across the US, Europe, and China.

Decentralization is Frey's key variable. Pre-industrial China led during the Song and Tang dynasties but couldn't industrialize: centralized gatekeeping killed novel bets and status flowed to civil servants over scientists. Soviet heavy industry grew four decades benchmarking factories, then collapsed when nothing new existed to benchmark. The US led computing because Bessemer could pass on Google while many others still said yes.

Both superpowers are now moving the wrong way. US tariff-linked crony capitalism is making America more like China; China is shifting from private-firm dynamism toward state-owned enterprises, historically the biggest drag on innovation. Europe's digital failure traces to internal services barriers equivalent to 110% tariffs, compounded by GDPR costs that favor large incumbents over startups.

On AI and jobs: Frey doubts visible labor disruption within a decade unless AI solves its resilience problem — amateur Go players beat top programs in 2023 by exploiting unfamiliar positions. The nearer impact is income and status pressure on white-collar workers as AI narrows the productivity gap between rich and developing countries, shifting professional services offshore.

Decentralization is Frey's key variable. Pre-industrial China led during the Song and Tang dynasties but couldn't industrialize: centralized gatekeeping killed novel bets and status flowed to civil servants over scientists. Soviet heavy industry grew four decades benchmarking factories, then collapsed when nothing new existed to benchmark. The US led computing because Bessemer could pass on Google while many others still said yes.

Both superpowers are now moving the wrong way. US tariff-linked crony capitalism is making America more like China; China is shifting from private-firm dynamism toward state-owned enterprises, historically the biggest drag on innovation. Europe's digital failure traces to internal services barriers equivalent to 110% tariffs, compounded by GDPR costs that favor large incumbents over startups.

On AI and jobs: Frey doubts visible labor disruption within a decade unless AI solves its resilience problem — amateur Go players beat top programs in 2023 by exploiting unfamiliar positions. The nearer impact is income and status pressure on white-collar workers as AI narrows the productivity gap between rich and developing countries, shifting professional services offshore.

The fog of war and the fog of the future

A Taiwan conflict could kill the AI-driven growth wave before it arrives. Capital Economics modeled six escalation scenarios; a full blockade cuts off TSMC, which makes over 90% of advanced chips. A shooting war — Christopher Neely's 2026 CSIS wargame has it lasting weeks — would crash equities 10–15%, snarl Pacific supply chains, and cost roughly $10 trillion (a tenth of global GDP). Longer-term, hard US-China decoupling would duplicate research systems and slow AI breakthroughs precisely when they're most needed.

I love progress, which is why I love liberal democracy

The tech-right case for autocratic "CEO-president" government — associated with Peter Thiel-aligned thinkers — holds that democratic checks (reviews, lawsuits, veto points) are structural drag blocking acceleration, and a strongman executive could rapidly greenlight nuclear plants, data centers, and megaprojects by bulldozing bureaucracy. Pethokoukis rejects this: the data show that innovation and growth are products of the freedoms liberal democracy protects, not casualties of its constraints. Real acceleration requires democracy, not its removal.

A California dream that should come true

Saudi Arabia's Neom — a $1.6 trillion mirrored megacity meant to house 9 million — is collapsing under internal cost estimates of $4.5 trillion, with The Line reduced to empty trenches. California Forever, by contrast, is plausibly scaled: 70,000 acres in Solano County for 400,000 residents, reframed around an industrial jobs anchor (the "Solano Foundry" near Travis AFB) rather than housing. That pivot tracks Alain Bertaud's principle that cities are labor markets first. The real threat is California's regulatory machinery, not the vision itself.

Why the US outgrows Europe, unfortunately

America's productivity lead has compounded to a 50% cumulative gap since 1995; the EU-US GDP gap doubled to 30% since 2002, 70% from weaker TFP. Goldman Sachs traces this to intangibles investment, flexible markets, better technology adoption, and deep capital markets enabling trillion-dollar scale. Europe's AI Act created compliance burdens Brussels is now rolling back. AI will widen the gap in IT, finance, and professional services — where America already leads. A weak Europe leaves the US without a democratic partner against China.

Funding outside the box: A Quick Q&A with … philanthropy expert Stuart Buck

Peer-review bureaucracy and donor social conformity both filter out the research most likely to produce breakthroughs. Peyton Rous's oncovirus work — colleagues urged him to stop around 1910; Nobel Prize came at age 87 — illustrates how breakthroughs look at inception. Federal cuts are too large for philanthropy to fill: enacted NIH/NSF cuts totalled ~$3B (the Gates Foundation alone could cover it), but proposed cuts — NSF to $3.9B from $10B, NIH down 40%, NASA science down 47% — create a tens-of-billions gap beyond philanthropic reach. Wealthy donors could give more creatively (Analogue Group, Convergent Research, SpecTech) but institutional isomorphism pulls them toward Harvard. Institutional diversity is the deeper fix: scientists captive to consensus for salary and reputation cannot contradict it.

An Up Wing future is possible: A Quick Q&A with … the authors of 'A Century of Plenty: A Story of Progress for …'

McKinsey Global Institute researchers (Smit, Bradley, Leung, Canal) argue that 2.6% annual GDP per capita growth — near the last 25 years' actual rate — could lift the poorest countries to today's Switzerland's level by 2100. The limits are institutional, not physical. They distinguish the "empowerment line" (agency, security, breathing room) from the poverty line, advocate a "politics of building," and flag clean-energy deployment — not innovation — as the binding constraint.

Here's why Europe produces Nobel laureates but not Elon Musks

Europe has all the scientific inputs — more Nobel laureates, more engineering graduates per capita, more publications than any comparable region, €37 trillion in household assets — yet only two EU-born companies have reached $100B market cap in fifty years, both listed in America. Demis Hassabis's 1998 London fundraising experience illustrates why: City investors steered him toward currency trading rather than backing his game studio. Late-stage capital scarcity still pushes maturing startups toward US listings or outright American acquisition today.

Jimmy Carter's boldest deregulation idea never happened, unfortunately

Carter's 1979 "malaise" speech proposed an Energy Mobilization Board — modeled on the WWII War Production Board — to enforce permit deadlines across agencies and shield priority energy projects from new regulations. The Iranian Revolution and gasoline lines were the catalyst. The 1970s regulatory buildup had already produced absurdities: NEPA delayed the Trans-Alaska Pipeline; the snail darter halted the Tellico Dam for a decade (later found not a distinct species). Environmentalists feared gutted protections; the Heritage Foundation objected that nuclear plants were excluded and emission standards couldn't be waived anyway; states resisted federal overreach; business groups wanted direct statutory reform instead. The GOP reversed support to deny Carter an election-year win, and Congress killed the EMB in June 1980 — leaving the 1970s framework intact and reform decades overdue.

Our fusion future: A Quick Q&A with nuclear engineer Michael Ford

Fusion avoids fission's chain-reaction danger by being instantly stoppable — remove fuel, injected energy, or containment and the reaction dies. But three public-acceptance risks remain: tritium releases, low-but-real proliferation potential, and cost uncertainty. SPARC targets net-positive energy by 2026; 40+ private companies have raised $7.1B. Critical engineering gaps persist: tritium breeding for self-sustaining fuel cycles, blanket systems to capture heat efficiently, and plasma-facing materials durable under extreme neutron flux. Ford bets on scientific proof-of-fusion before mid-century, though commercial viability has no timeline.

Permitting ourselves to build: A Quick Q&A with … energy policy expert James Coleman

America is spending down its postwar infrastructure inheritance: more than half of US oil pipelines predate NEPA (1970). NEPA's original logic was sound — the postwar boom destroyed urban neighborhoods like St. Paul's Rondo — but courts expanded its scope until reviews average four-plus years, litigation adding a decade more.

Clean energy bears this disproportionately. Oil improvises routes to market; solar, wind, and nuclear depend on pipelines and power lines requiring permits. The Endangered Species Act and Clean Water Act add further vetoes. The decisive bottleneck is judicial injunctions: presidents can instruct agencies to narrow reviews, but prior-administration judges can halt projects indefinitely.

A late-2024 DC Circuit ruling stripped the Council on Environmental Quality of authority to issue binding NEPA rules; the Trump administration reinforced that decision, opening space for Trump-appointed judges to revisit expansive precedent. Coleman's fix: cap injunctions at three to four years from review start. Without reform, nuclear expansion and high-speed rail are simply not buildable — Texas's solar and wind dominance shows what easy permitting unlocks.

Clean energy bears this disproportionately. Oil improvises routes to market; solar, wind, and nuclear depend on pipelines and power lines requiring permits. The Endangered Species Act and Clean Water Act add further vetoes. The decisive bottleneck is judicial injunctions: presidents can instruct agencies to narrow reviews, but prior-administration judges can halt projects indefinitely.

A late-2024 DC Circuit ruling stripped the Council on Environmental Quality of authority to issue binding NEPA rules; the Trump administration reinforced that decision, opening space for Trump-appointed judges to revisit expansive precedent. Coleman's fix: cap injunctions at three to four years from review start. Without reform, nuclear expansion and high-speed rail are simply not buildable — Texas's solar and wind dominance shows what easy permitting unlocks.

Time for a hard reset for environmental policy: A Quick Q&A with … Emily Domenech

Environmental activists have weaponized permitting law to kill projects financially: NGOs file 70% of legal challenges against forest and energy projects; 80% ultimately fail, but delays alone drain developer capital. The bipartisan Manchin-Barrasso bill (EPRA 2024) collapsed because its jurisdiction excluded House Republican priorities — NEPA and Clean Water Act reform — and after the Republican November sweep the core trade (more oil and gas production for transmission reform) no longer made sense for either side.

Our nuclear problem: A Quick Q&A with … economist Eli Dourado on NRC overreach

NRC compliance costs have made nuclear irrelevant — only three US reactors built in 28 years, all over budget. The NRC inherited rules that ignore a 1954 congressional mandate requiring carve-outs for small, safe reactors; Texas and Utah are now suing to enforce it. State-level SMR regulation would create an escape valve: both states have enough market demand to scale SMRs potentially below solar's cost. Deregulation would enable iterative design — cheap test reactors to validate and simplify designs — which is how nuclear actually gets cheap. Nuclear's downstream effect is catalytic: new materials, desalination, faster transportation. Advanced geothermal fills gaps nuclear can't reach, particularly in proliferation-risk countries where SMRs won't be permitted.

AI's Need for Nuclear: A Quick Q&A with … energy writer and historian Emmet Penney

A ChatGPT query uses 2.9 watt-hours versus 0.3 for a Google search — synthesis demands far more compute. China has 30 nuclear plants under construction plus 10 newly commissioned for $27B, targeting 200 GW by 2040 — double the US fleet — building America's own AP-1000 design faster and cheaper. Jevons's paradox means AI efficiency gains expand demand rather than cap it. SMRs promise on-site power for data centers, but Georgia proved modularity costly. Trump backing both coal and nuclear is simply grid triage.

Washington and the New Energy Revolution: How's it going?

The House reconciliation bill guts most IRA clean-energy credits but gives nuclear a partial pass: new plants have until 2029 to start construction, existing plants keep credits through 2031. Trump's executive orders to streamline NRC rules and back ten new reactors lifted nuclear stocks 10–30%. But expert Tyler Norris and Duke Energy both warn that without the ITC and a functioning Loan Programs Office, new nuclear at $186/MWh simply can't pencil. Fusion and geothermal get nothing — shut out while Washington rewards lobbying muscle over technological promise, the exact failure a carbon tax would avoid.

The rise and (maybe) fall of NEPA: A Quick Q&A with … law professor Nicholas Bagley

The Abundance movement's core complaint isn't regulatory excess writ large — it's that too many procedural layers have made government itself unable to function. Michigan law professor Nicholas Bagley distinguishes this from libertarian deregulation: the problem is NEPA-style rules letting NIMBY groups hijack permitting, strangling new infrastructure, housing, and nuclear energy. The Supreme Court's *Seven County Infrastructure Coalition v. Eagle County* ruling gives agencies more control over NEPA review scope — potentially clearing that bottleneck.

Our fission-powered future: My chat (+transcript) with nuclear scientist and author Tim Gregory

Nuclear fission is already as safe per unit of power as wind and solar, and orders of magnitude safer than fossil fuels. The half-century stall traces almost entirely to Three Mile Island and Chernobyl. Nobody died at Three Mile Island; Chernobyl killed roughly 30 immediate victims plus a few hundred thyroid-cancer cases in exposed children. The huge death tolls sometimes cited rest on the linear no-threshold model, one of the most contested hypotheses in medical science, with little physical evidence behind it. Radiophobia — radiation as cultural horror motif — amplified a manageable record into existential panic. The US built more reactors in the five years before Three Mile Island than in the fifty years since.

The revival is now driven more by AI data-center demand than climate policy. Energy consumption projections already called for a doubling or tripling; AI wasn't priced in when those estimates were made. Google, Microsoft, Amazon, and Meta have all placed SMR orders. SMRs scale from one megawatt (shipping-container-sized) to hundreds of megawatts, with 90 percent of components factory-buildable — unlocking the learning-curve cost reductions that industrialization always delivers. Reactors also produce surplus heat directly useful for steelmaking, ammonia synthesis, and other hard-to-electrify industries.

Waste is smaller than intuition suggests: a lifetime of nuclear-powered electricity produces waste that fits in a wine glass. Finland operates the world's first functioning geological repository. Breeder reactors can recycle 95 percent of spent fuel, reducing radioactive half-lives from a million years to a few centuries.

France is the template: 55 reactors built in 25 years, 80 percent of electricity from nuclear by 2005, at below-EU-average prices — grid decarbonization achieved before climate was a political issue.

Commercial fusion is probably decades away, possibly a century. Fission is the proven bridge.

The revival is now driven more by AI data-center demand than climate policy. Energy consumption projections already called for a doubling or tripling; AI wasn't priced in when those estimates were made. Google, Microsoft, Amazon, and Meta have all placed SMR orders. SMRs scale from one megawatt (shipping-container-sized) to hundreds of megawatts, with 90 percent of components factory-buildable — unlocking the learning-curve cost reductions that industrialization always delivers. Reactors also produce surplus heat directly useful for steelmaking, ammonia synthesis, and other hard-to-electrify industries.

Waste is smaller than intuition suggests: a lifetime of nuclear-powered electricity produces waste that fits in a wine glass. Finland operates the world's first functioning geological repository. Breeder reactors can recycle 95 percent of spent fuel, reducing radioactive half-lives from a million years to a few centuries.

France is the template: 55 reactors built in 25 years, 80 percent of electricity from nuclear by 2005, at below-EU-average prices — grid decarbonization achieved before climate was a political issue.

Commercial fusion is probably decades away, possibly a century. Fission is the proven bridge.

Hot rocks, new politics

Enhanced geothermal systems — borrowing oil-and-gas fracking and drilling techniques — are breaking geothermal out of its volcanic-geography constraint. Fervo Energy's Utah project cut per-well costs from $9.4M to $4.8M and drilling time by 70%. Gas-plant capital costs tripling to ~$3,000/kW narrows geothermal's price gap. Republicans see it as oil-and-gas-adjacent; Democrats and Big Tech want carbon-free 24/7 power. Tax credits survived the "One Big Beautiful Bill." Remaining blockers: NEPA permitting delays, irregular federal lease sales, and an undersized DOE R&D budget.

Tracking a 200-year energy wager

Herman Kahn's 1976 bet that humanity would master cheap, inexhaustible energy by 2176 is technically on track but politically stalled. Solar capacity doubles every three years and already supplies 6% of global electricity. Geothermal, redeployed from oil-and-gas drilling, could hit 100 GW by 2050 at $45–65/MWh. Nuclear revival is real but slow — Goldman Sachs sees only 380 to 575 GW by 2040. Fusion may yield commercial reactors in the 2030s. The constraint is permitting and politics, not physics.

FP! Week In Review, Briefly #18

Nuclear power's costs are not intrinsic — they reflect missing demand and policy support, and that is now changing. Nuclear policy researcher Jessica Lovering argues the US pause was overdetermined: the industry scaled reactor size too fast in the 1960s–70s, outrunning project-management experience; oil prices fell; inflation made capital-intensive projects unviable; deregulation rewarded optimizing existing assets over building anything new; and NRC regulatory instability added uncertainty. France and Sweden succeeded by pacing buildouts slowly. Three Mile Island accelerated the decline but was not the root cause.

Today's revival rests on more durable foundations. Climate goals, retirement of roughly 100 GW of US coal, and surging demand from AI data centers, EVs, and industrial electrification create sustained need for clean firm power. The IRA and ADVANCE Act now provide tax credits and deployment incentives that replicate what solar and shale gas received — the mechanism that drove their cost declines. Meta, Google, and Microsoft nuclear commitments follow from those incentives meeting data center demand.

On reactor design, large AP1000s suit regulated southeastern utilities; SMRs could serve smaller utilities on a turnkey basis, though none has been built in the US yet. Lovering's realistic target: 30–40% nuclear, up from 20% today. Europe's pivot since Russia's 2022 gas cutoff has been sharper.

Shell's 2026 Energy Security Scenarios finds that across all three modeled futures, electricity's share of final energy rises at multiples of its historical rate. America becomes an electrostate by economic logic regardless of climate policy; the only variable is whether it builds the supply chain first or imports it later.

A 166-country study finds democratic trust closely tracks lifetime economic growth. Poorly performing democracies fail to build legitimacy even with long exposure — recent performance is what matters. When growth slows, trust erodes and populism follows, making innovation-driven growth a democratic survival condition.

Today's revival rests on more durable foundations. Climate goals, retirement of roughly 100 GW of US coal, and surging demand from AI data centers, EVs, and industrial electrification create sustained need for clean firm power. The IRA and ADVANCE Act now provide tax credits and deployment incentives that replicate what solar and shale gas received — the mechanism that drove their cost declines. Meta, Google, and Microsoft nuclear commitments follow from those incentives meeting data center demand.

On reactor design, large AP1000s suit regulated southeastern utilities; SMRs could serve smaller utilities on a turnkey basis, though none has been built in the US yet. Lovering's realistic target: 30–40% nuclear, up from 20% today. Europe's pivot since Russia's 2022 gas cutoff has been sharper.

Shell's 2026 Energy Security Scenarios finds that across all three modeled futures, electricity's share of final energy rises at multiples of its historical rate. America becomes an electrostate by economic logic regardless of climate policy; the only variable is whether it builds the supply chain first or imports it later.

A 166-country study finds democratic trust closely tracks lifetime economic growth. Poorly performing democracies fail to build legitimacy even with long exposure — recent performance is what matters. When growth slows, trust erodes and populism follows, making innovation-driven growth a democratic survival condition.

A nuclear step forward—maybe

TerraPower's Natrium plant in Wyoming — a 345 MW sodium-cooled reactor paired with molten-salt storage — has cleared NRC construction approval, the first non-light-water reactor greenlit in 40 years, targeting 2031 operation. The economics remain unproven: SMR costs are estimated at $200–$400/MWh, against a competitive benchmark near $125–$130 and gas at $55–$85. Smaller reactors sacrifice the scale economies that drove down traditional nuclear costs. The real verdict, per TerraPower's own CEO, arrives at reactor ten, not reactor one.

The axis of US decline: anti-data center, anti-AI, anti-nuclear

Anti-data center, anti-AI, and anti-nuclear activism now form a single coalition — opposing one means opposing all three. Data centers will grow from 4.4% of US electricity in 2023 to 12% by 2028, making SMR economics viable: Oklo signed 1.2 GW with Meta, Amazon put $300M into X-Energy with a 5 GW order, TerraPower received a construction permit in March 2026. Constellation's TMI restart for Microsoft AI crystallized the backlash, now extending to SMRs before they've operated. Decline is a choice.

The AI Paradox: When will artificial intelligence boost economic growth?

AI's productivity gains may already be happening but invisible in official statistics — the same trap Robert Solow identified in 1987 and Alan Greenspan navigated in 1995. Greenspan argued ICT productivity was real but mismeasured (services especially) and that physical-tonnage GDP metrics miss "conceptualization" of output. He resisted premature rate hikes; the 1990s boom vindicated him. The same question now applies to AI: when will it show up in the stats?

The Fed (maybe) sees a glimpse of the Age of AI

Establishment institutions never lead with AI optimism, which makes the November 2024 FOMC minutes notable: some Fed participants explicitly linked recent strong productivity growth to AI adoption in the workplace. But the evidence is contested. JPMorgan and Moody's attribute the upturn to better post-pandemic employer-employee matching; the San Francisco Fed sees a typical cyclical bounce-back from pandemic layoffs. Total factor productivity — where genuine transformation would show — is up only 0.4 percentage points above the 2005-2019 baseline. The Fed's glimpse is real but tentative, and how much AI explains remains uncertain.

My chat (+transcript) with journalist Nicole Kobie on why the future of tech still hasn't arrived

Technology timelines repeatedly disappoint not because of villains but because every technology passes through four distinct phases — academic idea, laboratory science, physical engineering, and commercialization — each demanding different people, money, and problem-solving. Journalists compound the illusion by covering lab breakthroughs while skipping the grinding middle stages, then writing the same "it's almost here" story on a fresh cycle because newsrooms have short memories.

Nicole Kobie, futures editor at PC Pro and a Wired contributing editor, draws the clearest case from driverless cars: "imminent" for over a decade. Processing power, not regulation, is what held AI back — Geoffrey Hinton had the core ideas in the 1990s and couldn't execute them. The instant capability arrived, critics called for a pause, which is exactly the failure mode her book diagnoses: we wait, then we stop.

The villain-versus-innovator framing is itself the problem. Regulators asking OpenAI for specifics is the system functioning, not a battle. Aviation's real dysfunction: certification cycles run so long hardware is outdated before approval clears, requiring re-certification from scratch. China deploys autonomous vehicles broadly because government and commercial interests are fused — faster, not better. The Oxford-AstraZeneca vaccine came from UK government-funded academia, showing the American market-test model isn't universal.

AI is not one thing: an LLM drafting routine emails is uncontroversial; one generating NHS letters that incorrectly deny benefits is a specific, addressable risk — and must be treated separately from the global "should we use AI?" debate.

On 2025–2035: change will be real but smaller than the internet transition. Autonomous technology will improve public transit more than personal vehicles. CRISPR-era medicine is the highest-upside bet. Climate mitigation is the most underinvested domain. AI will not eliminate work. We overestimate two-year change and underestimate ten-year change — we notice dramatic events and miss cumulative drift.

Nicole Kobie, futures editor at PC Pro and a Wired contributing editor, draws the clearest case from driverless cars: "imminent" for over a decade. Processing power, not regulation, is what held AI back — Geoffrey Hinton had the core ideas in the 1990s and couldn't execute them. The instant capability arrived, critics called for a pause, which is exactly the failure mode her book diagnoses: we wait, then we stop.

The villain-versus-innovator framing is itself the problem. Regulators asking OpenAI for specifics is the system functioning, not a battle. Aviation's real dysfunction: certification cycles run so long hardware is outdated before approval clears, requiring re-certification from scratch. China deploys autonomous vehicles broadly because government and commercial interests are fused — faster, not better. The Oxford-AstraZeneca vaccine came from UK government-funded academia, showing the American market-test model isn't universal.

AI is not one thing: an LLM drafting routine emails is uncontroversial; one generating NHS letters that incorrectly deny benefits is a specific, addressable risk — and must be treated separately from the global "should we use AI?" debate.

On 2025–2035: change will be real but smaller than the internet transition. Autonomous technology will improve public transit more than personal vehicles. CRISPR-era medicine is the highest-upside bet. Climate mitigation is the most underinvested domain. AI will not eliminate work. We overestimate two-year change and underestimate ten-year change — we notice dramatic events and miss cumulative drift.

What DeepSeek means for faster AI-driven economic growth

DeepSeek's $5.6M training cost doesn't shrink AI's macroeconomic upside — it likely accelerates it. Goldman Sachs's prior forecast stands: GenAI adds ~1.5 percentage points to US labor productivity over a decade and 7% to global GDP. Cheaper models shift who captures value — hardware makers lose share, as software firms took only 25% during 1981–2012 — but GDP measures total output. Lower costs remove the platform-buildout bottleneck, pulling GS's 2027 US adoption inflection earlier. Geopolitical rivalry pushes governments to cut barriers and fund domestic AI. Adoption is when, not whether.

The Age of AI, an update: A Quick Q&A with … economist Michael Strain

AEI economist Michael Strain argues generative AI will diffuse through the economy faster than electricity or the internet — but hasn't yet, and won't be measurable in 2025. Invention and diffusion run on separate tracks; we're early on both. Workers facing displacement should identify which parts of their job AI can replace, lean into the parts it can't, and use the technology to amplify their own productivity rather than treating it as a signal to switch industries.

Why China might lead the Robot Revolution

China is poised to dominate humanoid robotics the same way it dominated EVs — through manufacturing scale, dense supplier networks, and government backing. JPMorgan's Hong Kong industrials team sees at least 34 humanoid models nearing production in 2025, with a total addressable market of 5 billion units driven by shrinking workforces across China, Japan, South Korea, and Europe. China's demographic pressure adds urgency. Tesla's Optimus is America's clearest counterweight, targeting commercial production in 2025 and a tenfold increase by 2026-27. The US advantage is AI brains — Nvidia, Google, OpenAI — while China supplies the brawn. The decisive question is which matters more.

Back to the Nineties? My chat (+transcript) with economist Skanda Amarnath on the 2020s productivity outlook

The 1990s — especially 1996–2000 — achieved high prime-age employment, falling inflation, and productivity growth not replicated since. Skanda Amarnath of Employ America identifies three drivers: a fully employed labor market where accumulated human capital let output scale without proportionate hours growth; a fixed-investment surge in IT and telecom (the 1996 Telecommunications Act catalyzed the fiber boom); and falling healthcare costs through HMOs. Janet Yellen called that healthcare dynamic underrated. When HMO cost controls collapsed in the 2000s, healthcare inflation helped end the boom.

Since 2019 Q4, productivity has run ~1.9% annually — about 0.5 percentage points above the pre-pandemic trend of 1.4%. Two of the three '90s legs are currently intact: labor markets remain historically strong (~160,000 monthly payrolls), and AI-driven capital deepening — data centers, energy infrastructure, software — is building the investment leg. The missing leg is cost stability. Broad-based tariffs raise input costs and investment hurdle rates across the same sectors where productivity upside is concentrated. The 2018–19 tariff round, far smaller in scope, had identifiable drag on manufacturing; current threats are much wider, hitting exporters like Caterpillar and John Deere and complicating transformer and energy infrastructure buildout.

The Fed faces a harder version of the '90s situation. PCE inflation near 2.6–2.7% largely reflects lagged housing and financial-services measurement — strip those and you get ~2.2%. But forward tariff risk prevents acting on lags alone; stagflationary tariffs would block rate cuts just as expectations drift upward.

The cautionary precedent is the post-2009 recovery: premature tightening, slow labor rebound, hysteresis — skills atrophied and productivity dividends never fully returned. Amarnath rates broad tariff rollback as the top pro-productivity move today, followed by healthcare cost reform — site-neutral payments as one example — the lever that explains both the 1990s boom and the 2000s bust.

Since 2019 Q4, productivity has run ~1.9% annually — about 0.5 percentage points above the pre-pandemic trend of 1.4%. Two of the three '90s legs are currently intact: labor markets remain historically strong (~160,000 monthly payrolls), and AI-driven capital deepening — data centers, energy infrastructure, software — is building the investment leg. The missing leg is cost stability. Broad-based tariffs raise input costs and investment hurdle rates across the same sectors where productivity upside is concentrated. The 2018–19 tariff round, far smaller in scope, had identifiable drag on manufacturing; current threats are much wider, hitting exporters like Caterpillar and John Deere and complicating transformer and energy infrastructure buildout.

The Fed faces a harder version of the '90s situation. PCE inflation near 2.6–2.7% largely reflects lagged housing and financial-services measurement — strip those and you get ~2.2%. But forward tariff risk prevents acting on lags alone; stagflationary tariffs would block rate cuts just as expectations drift upward.

The cautionary precedent is the post-2009 recovery: premature tightening, slow labor rebound, hysteresis — skills atrophied and productivity dividends never fully returned. Amarnath rates broad tariff rollback as the top pro-productivity move today, followed by healthcare cost reform — site-neutral payments as one example — the lever that explains both the 1990s boom and the 2000s bust.

DeepSeek and tariffs fail to undermine the AI investment boom (so far)

Despite DeepSeek's ~10x efficiency gain and Trump's Liberation Day tariffs adding 15-20% to data-center costs, America's hyperscalers doubled down in Q1 2025: Meta raised capex guidance to $64-72B, Alphabet budgeted ~$75B, Amazon approached $100B. Two forces override the headwinds: Jevons Paradox (cheaper AI tokens expand total demand, pulling in smaller companies that couldn't afford AI before), and competitive lock-in pressure (no Big Four player can be seen falling behind). The binding constraint is now electricity grids and chip supply, not capital commitment.

AI and the radiologist apocalypse that wasn't

Geoffrey Hinton predicted in 2016 that radiologists would be obsolete within five years. Mayo Clinic's headcount has since risen 55 percent. Over 250 algorithms handle routine tasks, but comprehensive diagnosis — consulting physicians, reading patient history, exercising judgment — remains human. Macro evidence is mixed: St. Louis Fed finds productivity gains in AI-heavy, educated-workforce industries; JPMorgan finds no clear link yet. The Productivity J-Curve reconciles both: setup costs are currently invisible in the data; gains come later at scale.

The Up Wing combo of AI and energy

Two economics papers combined reveal a self-reinforcing growth architecture. Wang and Wong find AI trained alongside humans eliminates 23% of jobs long-term but boosts productivity 366%; human-AI collaboration prevents mass unemployment by making workers more valuable than redundant. Lemoine shows manufactured energy — solar chiefly, but also nuclear and fusion — enables unlimited growth via automated self-replication, if energy return on investment stays high. Together the loops shift binding scarcity from physical resources to useful ideas.

How the AI boom is like the Internet Boom (and how it isn't)

Stock-market parallels between AI and the 1990s internet boom are real — Nvidia mirrors Cisco — but the economic test is whether AI drives discovery, not just automation. Oxford's Carl Benedikt Frey notes productivity growth has slowed to 0.8% annually; AI email tools save 3.6 hours weekly, then inbox volume expands to compensate. Georgetown's Dan Cao shows dot-com valuations crashed when spillover speed disappointed. Today's AI expectations are set higher still, and cautious economists see only a 0.5-point productivity gain if ideation stalls.

Self-driving cars are finally shifting gears

After two decades of overpromising, AVs are deploying: Waymo logs 250,000 paid rides weekly, its 56-million-mile study showing 96% fewer intersection crashes and 92% fewer pedestrian injuries than human drivers. Tesla's Austin robotaxi stumbled — one incident into oncoming traffic — and Goldman Sachs expects near-term scaling to stay slow, though it projects $7B Waymo revenue by 2030 at 40–50% margins. Aurora targets two-thirds trucking cost cuts. Remaining blockers are institutional: patchwork US regulation, unadapted liability law, and insurers locked in old risk models.

The US economy at midyear: My chat (+transcript) with economic analyst Joey Politano

Trade and immigration are the two dominant US economic headwinds at mid-2025, each requiring 100-year lookbacks for historical analogs. Joey Politano (Apricitas Economics, ex-BLS) argues the current tariff regime is historically anomalous: a 10% baseline on nearly all imports, 25% on cars, 50% on steel and aluminum, and 30% on China (after briefly hitting 145%). Simultaneously, net migration is likely falling from 2.8 million annually to near zero — the largest one-year shift in US history — driven mostly by aggressive interior enforcement, not just border policy.

Trade analysis is hobbled by incoherence. The April 2nd "reciprocal" formula measured bilateral deficits, not actual trade barriers, making Brazil — a genuine high-barrier offender — one of the lowest-tariffed partners because it exports coffee and oil. Rates on Japan, Indonesia, and the Philippines shifted via Trump statements, not legal instruments. The stated rationale contradicts itself: tariffs can't simultaneously avoid raising prices, re-shore manufacturing, and fix the budget deficit. The only stable principle is that Trump wants tariffs higher; competing internal factions pitch overlapping schemes and he approves them all.

AI is a genuine productivity tailwind — Loudoun County is the world's largest data-center cluster, and US large-computer imports are running near $150 billion annually on a straight-line trajectory. But AI's US advantage depends on immigration: roughly 50% of top tech talent is foreign-born, and proposed H-1B and OPT restrictions threaten to redirect future data-center investment to Vancouver or the Gulf States.

Trade analysis is hobbled by incoherence. The April 2nd "reciprocal" formula measured bilateral deficits, not actual trade barriers, making Brazil — a genuine high-barrier offender — one of the lowest-tariffed partners because it exports coffee and oil. Rates on Japan, Indonesia, and the Philippines shifted via Trump statements, not legal instruments. The stated rationale contradicts itself: tariffs can't simultaneously avoid raising prices, re-shore manufacturing, and fix the budget deficit. The only stable principle is that Trump wants tariffs higher; competing internal factions pitch overlapping schemes and he approves them all.

AI is a genuine productivity tailwind — Loudoun County is the world's largest data-center cluster, and US large-computer imports are running near $150 billion annually on a straight-line trajectory. But AI's US advantage depends on immigration: roughly 50% of top tech talent is foreign-born, and proposed H-1B and OPT restrictions threaten to redirect future data-center investment to Vancouver or the Gulf States.

AI and the future of R&D: My chat (+transcript) with McKinsey's Michael Chui

R&D productivity is in long-term decline even as spending rises — a pattern Nick Bloom at Stanford documented and McKinsey's Michael Chui calls the central problem AI could help reverse. In pharmaceuticals, "Eroom's Law" (Moore's Law backwards) describes drug compounds per billion R&D dollars halving every nine years. Physics papers once had three authors; now the list runs pages; the problems are genuinely harder.

AI addresses this through three channels. First, it expands the candidate funnel — generating drug molecules, alloy compositions, or airframe designs faster and with less human-constrained variety (AlphaGo's Move 37 illustrates machine creativity exceeding expert intuition). Second, AI surrogate models replace slow physics-based simulations: where computational fluid dynamics runs for hours or days, neural-net surrogates run in minutes. Third, LLMs accelerate surrounding knowledge work.

The bottleneck is organizational, not technological. McKinsey surveys show 80% of companies use AI somewhere but only 1% report mature usage. The specific friction: R&D teams split physical testing and simulation into separate organizations, leaving no one to optimally allocate across testing modes. The book *Rewired* identifies six dimensions of required organizational change — only one is the tech stack.

The payoff: potentially doubling R&D output in science-based industries like pharma and specialty chemicals, measured in new products reaching market. Regulation is an existing constraint AI doesn't change; clinical trial subject recruitment is an independent bottleneck that could cap gains even if candidate generation doubles.

AI addresses this through three channels. First, it expands the candidate funnel — generating drug molecules, alloy compositions, or airframe designs faster and with less human-constrained variety (AlphaGo's Move 37 illustrates machine creativity exceeding expert intuition). Second, AI surrogate models replace slow physics-based simulations: where computational fluid dynamics runs for hours or days, neural-net surrogates run in minutes. Third, LLMs accelerate surrounding knowledge work.

The bottleneck is organizational, not technological. McKinsey surveys show 80% of companies use AI somewhere but only 1% report mature usage. The specific friction: R&D teams split physical testing and simulation into separate organizations, leaving no one to optimally allocate across testing modes. The book *Rewired* identifies six dimensions of required organizational change — only one is the tech stack.

The payoff: potentially doubling R&D output in science-based industries like pharma and specialty chemicals, measured in new products reaching market. Regulation is an existing constraint AI doesn't change; clinical trial subject recruitment is an independent bottleneck that could cap gains even if candidate generation doubles.

Awaiting the AI acceleration: A Quick Q&A with … researchers Anson Ho and Andrei Potlogea

Society is probably underinvesting in AI — conditional on low risk concern. Potlogea and Ho's GATE model finds Baumol effects weaker than assumed, and frictions like R&D externalities dampen growth accelerations less than expected, making explosive growth easier to achieve. Underinvestment traces to three interacting causes: uncertainty about social returns, investor risk aversion, and private uncertainty about capturing value under regulatory or tax risk. Governments add political-economy constraints. Economists err by benchmarking AI against narrower, slower past technologies.

Brace yourself: The AI Revolution isn't about job creation

Job losses from AI are evidence that the technology is real, not reasons to panic. Salesforce cut 4,000 customer support roles after deploying AI agents. A Stanford/ADP study found employment among 22–25-year-olds in AI-exposed fields (customer service, accounting, software dev) down 13% since 2022. Harvard/LinkedIn data on 62 million workers shows AI adoption favors senior staff while shrinking junior hiring. From an Up Wing perspective: a GPT that left labor demand unchanged wouldn't be much of a GPT. Jobs are not the economy's goal.

AI as a 'normal technology' would still be amazing

Princeton computer scientists Arvind Narayanan and Sayash Kapoor argue AI will follow normal technology diffusion — gradual adoption, regulatory lag, organizational change — not sudden capability leaps. Real risks are familiar: accidents, misuse, inequality. Goldman Sachs agrees on the normalist trajectory but still projects US potential GDP at 2.1–2.3% through the early 2030s (vs. Fed/CBO's 1.8%), noting official productivity stats miss roughly 0.5% of AI-related capital formation. Corrected productivity growth runs at 1.6% since 2019. Goldman expects a 1990s-style arc: capital deepening first, then total factor productivity gains as organizations restructure — with AGI as an upside tail risk.

Thinking machines won't need us. And that's OK

AGI would sever the historical link between human labor and economic output: GDP would scale with compute, not people. Yale economist Pascual Restrepo's paper "We Won't be Missed" argues that in an AGI world, stopping all human work would barely register economically — humans no longer improve living standards in any appreciable way. The dystopian reading (billionaires in towers, masses in slums) misses the more plausible outcome: AI-driven abundance where work is optional rather than structurally necessary.

The Age of AI is starting to bloom

AGI is probably not imminent and the AI boom is not a bubble — and the economy is beginning to show real AI-driven lift. Prediction markets put AGI arrival between 2027 and 2034. On the bubble question: annual AI investment is up $200–300B since 2023, but Goldman Sachs estimates generative AI could create $20T in US economic value and lift labor productivity ~15%. Training-query demand grows 350% annually while compute efficiency improves only 40%, pointing to a capacity shortage, not a glut.

Acceleration though AI-automated R&D: My chat (+transcript) with economist Tom Davidson

AI's most transformative scenario isn't automating customer service — it's AI that can automate AI research itself, triggering a self-reinforcing algorithmic improvement loop that requires no new hardware. Epoch AI estimates that algorithmic efficiency already makes the same AI system 10x–1000x cheaper to run within 12 months purely through software improvements. If AI reaches the level where it can fully replace human researchers at a lab like OpenAI, those AI researchers would improve their own algorithms, multiplying effective researcher count without building a single new chip. Davidson calls this a "software intelligence explosion."

Davidson puts 50/50 odds on fully automating AI research within 10 years, contingent on two or three non-trivial breakthroughs beyond current RL training methods. The benchmark-to-reality gap is smaller for AI research than most domains because the work is already virtual, coding-centric, and measurable — though even there, AI currently produces "spaghetti code" that passes function tests but is unmaintainable.

Economic projections extend to 30%+ annual GDP growth, potentially accelerating toward self-replicating physical systems that double the economy every few weeks — analogous to rats doubling in six weeks, but applied to factories and infrastructure. The 1400 CE economist who would laugh at 1% growth is the model for today's skeptics.

The primary near-term constraint Davidson identifies is societal pushback rather than energy or labor supply. He proposes pausing just before AI researcher automation is achieved to solve two problems in sequence: alignment (ensuring AI stays loyal to human values, not hidden goals) and governance (preventing any single company or government from monopolizing AI's productive capacity). He assigns existential risk from misaligned superintelligence at 1–10%, enough to merit a controlled ramp-up rather than a stop.

Davidson puts 50/50 odds on fully automating AI research within 10 years, contingent on two or three non-trivial breakthroughs beyond current RL training methods. The benchmark-to-reality gap is smaller for AI research than most domains because the work is already virtual, coding-centric, and measurable — though even there, AI currently produces "spaghetti code" that passes function tests but is unmaintainable.

Economic projections extend to 30%+ annual GDP growth, potentially accelerating toward self-replicating physical systems that double the economy every few weeks — analogous to rats doubling in six weeks, but applied to factories and infrastructure. The 1400 CE economist who would laugh at 1% growth is the model for today's skeptics.

The primary near-term constraint Davidson identifies is societal pushback rather than energy or labor supply. He proposes pausing just before AI researcher automation is achieved to solve two problems in sequence: alignment (ensuring AI stays loyal to human values, not hidden goals) and governance (preventing any single company or government from monopolizing AI's productive capacity). He assigns existential risk from misaligned superintelligence at 1–10%, enough to merit a controlled ramp-up rather than a stop.

How bad policy can offset the AI boom

AI-driven productivity gains since ChatGPT's launch — 2.16% annualized vs. 1.43% pre-pandemic — are being fully offset by trade protectionism. The 2025 tariff hike (15% average) acts like a broad tax that suppresses investment and demand rather than just raising prices. J.P. Morgan finds nonfarm productivity has reverted to its tepid 1.5% pre-pandemic trend, and draws a parallel to Brexit, which cut UK productivity 0.3–0.4% annually for a decade. The 1990s tech boom had falling trade barriers behind it; this one doesn't.

What the shale boom tells us about the AI boom

AI is in the Hopes & Dreams phase of every capital-intensive technology cycle, not yet the Execution/Efficiency phase where returns get punished. Goldman Sachs draws the parallel to shale: massive pre-payoff spending, then an eventual reckoning when supply overwhelmed demand. That reckoning hasn't arrived for AI — hyperscalers still run ~30% ROIC, data-center vacancy stays low, and cumulative AI capex remains below 1% of US GDP. Underinvesting looks riskier than overspending until supply visibly outruns demand.

AI agents are a real thing and doing real work

AI agents are already delivering measurable economic returns — no superintelligence required. C.H. Robinson (freight broker, $18B revenue) runs 30+ agents that answer spot-quote emails in 30 seconds instead of hours, auto-create 5,500 shipment orders daily, push quote-response rates from 60% to ~100%, and eliminate 600 labor hours per day — enabling revenue growth without headcount growth. JPMorgan estimates agents are targeting $30 trillion in annual routine white-collar spending. The AI 2027 authors, meanwhile, pushed full autonomous coding out to the early 2030s.

Don't expect an AI job apocalypse

Goldman Sachs estimates AI could automate tasks covering ~25% of US work hours, but translates to only 6–7% job displacement over the full adoption period, with peak unemployment rising 0.5–1.2 percentage points — transitional, not permanent. The historical anchor: only 40% of today's workers hold jobs that existed 85 years ago, yet employment absorbed every prior technology wave. AI will eliminate tasks, transform roles, and spawn new occupations too unpredictable to name now, while a 15% productivity uplift ripples through demand and multiplies activity.

The pause before the gale

AI disruption follows the historical GPT pattern: a slow, uneven diffusion before a sudden upswing. Currently only ~10% of firms regularly use AI, yet Morgan Stanley sees total factor productivity gains beginning to compound as the "task horizon" — how long AI can sustain autonomous work — doubles every seven months. The gradualist plateau is a bridge, not the destination; the Schumpeterian gale of creative destruction is simply still building.

AI anxiety: Is 'something big happening,' really?

Accelerating AI capability doesn't mean economic transformation is imminent — structural bottlenecks will absorb the shock. Matt Shumer's viral "early Covid moment" framing ignores how economies actually adopt new technologies: electrification took decades, AI currently reaches fewer than one in five U.S. businesses. Four dampeners follow: the productivity J-curve delays visible gains; richer societies choose leisure over output; Baumol's cost disease shifts spending toward labor-intensive sectors; and Charles Jones's "narrowest pipe" means energy, regulation, and human decisions cap system-wide speed regardless of AI gains.

Does AI have to cure cancer?

"Curing cancer" sets an unfair benchmark for AI in medicine. The tractable, already-visible win is reversing Eroom's Law — the decades-long trend of drug discovery growing slower and costlier even as computing improved. A Goldman Sachs analysis of ~100 AI-discovered drug candidates finds a 10% success rate versus 6% historically (60% improvement), with gains concentrated at Phase 1–2 where most drugs collapse. AI also cuts development time by 20–25% and costs by 25–30%, worth $80–400B in industry value over the next decade.

The future of work in an age of AI: My chat with economist Daniel Rock

AI exposure and AI automation are not the same thing — Wharton economist Daniel Rock draws a hard distinction between fields AI touches and those it genuinely replaces. Adoption bottlenecks are organizational, not technical: firms must reorganize workflows before productivity shows up. Rock's J-curve thesis holds that intangible investments complement general purpose technologies, so gains appear late; growth projections should be more modest than Silicon Valley's "end of white-collar work" framing suggests.

FP! Week In Review, Briefly #31

Chernobyl's real damage wasn't the reactor explosion — it was the political fallout. Ronald Bailey's argument: the disaster was a failure of communist secrecy and safety culture, not nuclear technology. Direct deaths were far fewer than media implies. The lasting harm came from the anti-nuclear backlash: slowed plant construction, more fossil fuels, and air pollution. A 2024 study estimates 318 million lost life-years from that energy pivot. The pattern repeated after Three Mile Island and Fukushima — policy overreactions proved deadlier than the accidents.

Economist Daniel Rock (Wharton, co-founder of Workhelix) pushes back on 20–30% unemployment forecasts. Organizations are ingesting AI "at a fairly gradual clip." Even if technology were ready, firms still need to reconfigure workflows, build buy-in, and retrain workers — all at human speed. His framework: capabilities advance, organizational reconfiguration speed, and demand elasticity all compound; focusing only on the first produces sci-fi scenarios.

Rock places the US early in the productivity J-curve (Brynjolfsson, Syverson, Rock): massive investment in chips and data centers now, with reconfiguration costs hidden from GDP accounts, so gains won't show in aggregate data yet. At Workhelix they see a power-law distribution — roughly 10% of employees generating half the AI value — showing diffusion hasn't happened. His optimistic target: 2–3% annual TFP growth; 15–20% projections are "a sci-fi economy." On job exposure: exposure is not automation. Remote customer service agents face genuine displacement risk; radiologists, whose tasks bundle tightly, are more resilient. He expects the near-term conversation to shift to managing agentic teams and being managed by AI.

Princeton physicist Gerard K. O'Neill's 1976 *The High Frontier* proposed miles-long rotating space cylinders housing millions, built from lunar and asteroid materials. Writing against *The Limits to Growth* pessimism, he reframed scarcity as geographic, not physical: move industry off-Earth rather than cap growth on it.

Economist Daniel Rock (Wharton, co-founder of Workhelix) pushes back on 20–30% unemployment forecasts. Organizations are ingesting AI "at a fairly gradual clip." Even if technology were ready, firms still need to reconfigure workflows, build buy-in, and retrain workers — all at human speed. His framework: capabilities advance, organizational reconfiguration speed, and demand elasticity all compound; focusing only on the first produces sci-fi scenarios.

Rock places the US early in the productivity J-curve (Brynjolfsson, Syverson, Rock): massive investment in chips and data centers now, with reconfiguration costs hidden from GDP accounts, so gains won't show in aggregate data yet. At Workhelix they see a power-law distribution — roughly 10% of employees generating half the AI value — showing diffusion hasn't happened. His optimistic target: 2–3% annual TFP growth; 15–20% projections are "a sci-fi economy." On job exposure: exposure is not automation. Remote customer service agents face genuine displacement risk; radiologists, whose tasks bundle tightly, are more resilient. He expects the near-term conversation to shift to managing agentic teams and being managed by AI.

Princeton physicist Gerard K. O'Neill's 1976 *The High Frontier* proposed miles-long rotating space cylinders housing millions, built from lunar and asteroid materials. Writing against *The Limits to Growth* pessimism, he reframed scarcity as geographic, not physical: move industry off-Earth rather than cap growth on it.

Are we just 18 months away from everything changing?

AI systems are already strong coding assistants, and autonomous engineering agents — capable of running experiments, debugging models, and optimizing chips with minimal human supervision — are now delivering real-world results. As these systems turn increasingly to AI research itself, each generation accelerates the next: better AI begets better AI. No single breakthrough is required, just compounding recursive self-improvement until the tempo of progress outpaces standard economic intuition — what Pethokoukis calls the Singularity.

How to give a commencement speech about AI without getting booed (too much)

Fear of AI job destruction is understandable but miscalibrated. History's pattern — steam, electrification, internet — is consistent: disruption happens, unpredicted jobs emerge, real wages rise. The Forecasting Research Institute survey of 500+ economists and AI professionals put 47% odds on "moderate progress," with near-term GDP growth near 2.5–3%. Even the fast-progress tail faces real constraints: energy infrastructure, regulatory pace, and the Baumol effect cap deployed gains. And stagnation has costs — nuclear's lost decades and post-2008 populist backlash are the price of misplaced fear.

What story is Fable telling about the state of AI?

Claude Fable 5 benchmarks above every public model and Anthropic's own data shows engineers shipping eight times as much code per quarter as in 2021–2025, pointing toward recursive self-improvement. But prediction markets have cooled sharply — superintelligence-by-2030 contracts fell from 61% to 28% since February. An MIT NBER study of 100,000 GitHub developers explains why: a 180% jump in commits becomes only a 30% increase in releases. Stanford's AI Economic Indicators confirm TFP hasn't accelerated. The bottleneck is organizational adoption, not model capability.

My chat (+transcript) with space journalist Eric Berger on SpaceX and America's New Space Age

SpaceX catching the Starship Super Heavy booster mid-air with "chopstick" arms in October 2024 eliminates landing legs and their dead mass (actuators, hydraulic fluid), cuts days from the turnaround cycle, and points toward catch-remount-relaunch on the same day. Eric Berger, senior space editor at Ars Technica and author of *Reentry*, frames it as the next step in economics SpaceX proved with Falcon 9: selling launches at $67–68 million while internally reflying the booster for roughly $15 million — a four-fold cost reduction through reuse alone. Starship targets another order of magnitude: Falcon 9 is at a few thousand dollars per pound to orbit; Starship aims for hundreds.

Near-term commercial demand centers on direct-to-cell Starlink satellites too large for Falcon 9, on top of a Starlink business already at four million customers and profitable. Starship can lift 100 tons to the lunar surface reusably versus the Apollo Lunar Module's five tons expendably.

On the moon race: NASA's 2026 Artemis III date is fiction; 2028 is the realistic floor, and China targets 2030. Switching from SLS for crew launch is no longer feasible — Starship's role as lunar lander is locked in, but it needs hundreds of flawless landings first, plus solving fully autonomous lunar ascent (no launch tower, no abort option).

The Mars case has no commercial logic — no exportable resource, no corporate sponsor. Starlink was built to fund it. Musk's "window closing" framing refers not to a deadline but to civilizational risks: war, nuclear weapons, rising capital costs, his own mortality. In 2016 when he announced Mars plans, Falcon 9 had just failed twice; now Starship is landing. Berger calls it the only credible path to Mars in our lifetimes.

Near-term commercial demand centers on direct-to-cell Starlink satellites too large for Falcon 9, on top of a Starlink business already at four million customers and profitable. Starship can lift 100 tons to the lunar surface reusably versus the Apollo Lunar Module's five tons expendably.

On the moon race: NASA's 2026 Artemis III date is fiction; 2028 is the realistic floor, and China targets 2030. Switching from SLS for crew launch is no longer feasible — Starship's role as lunar lander is locked in, but it needs hundreds of flawless landings first, plus solving fully autonomous lunar ascent (no launch tower, no abort option).

The Mars case has no commercial logic — no exportable resource, no corporate sponsor. Starlink was built to fund it. Musk's "window closing" framing refers not to a deadline but to civilizational risks: war, nuclear weapons, rising capital costs, his own mortality. In 2016 when he announced Mars plans, Falcon 9 had just failed twice; now Starship is landing. Berger calls it the only credible path to Mars in our lifetimes.

The story of the US Space Force: A Quick Q&A with … defense policy analyst Todd Harrison

Space became a contested domain in 1959 with the first ASAT test; 2,800 satellites launched in 2023 grew the orbital population 22% in a single year. The Space Force — controversial at creation, forced on DoD by Trump — runs GPS, ISR, satcom, and missile warning for the other services, defending them against daily gray-zone attacks: jamming, cyber intrusions, and laser-blinding of sensors. Doctrine now calls for actively pushing back in peacetime rather than absorbing degradation silently. Co-orbital ASAT weapons are real: Soviet tests in the 1960s–70s, active Chinese and Russian on-orbit systems today. The Outer Space Treaty bars lunar military bases, but enforcement is toothless; NASA's Artemis Accords aim to lock in open-society norms before the gap is exploited.

Missions to Mars: A Quick Q&A with … space analyst Peter Hague

Mars colonization is on the agenda only because Musk and Bezos have made it a personal priority — without them, the conversation wouldn't exist. Astrophysicist Peter Hague argues the first mission priority must be a minimally import-dependent base, since every other goal (science, resource extraction) draws on the same supply chain. Nearly every technical barrier — cosmic ray exposure, life-support gaps, machinery reliability — resolves to a "mass problem": send more mass, and the problem shrinks.

My chat (+transcript) with economist Matt Weinzierl on the growing business of space

The defining shift in space is not privatization but decentralization — moving from the Soviet-style central planning governing NASA since Apollo toward market forces that discover value no government roadmap anticipated. As early SpaceX employee Jim Cantrell put it, the Great American Space Enterprise that defeated Communism was itself running on a Soviet economic model. The Shuttle's cancellation forced a reckoning: the US was preparing to rent seats to orbit from Russia.

SpaceX is the clearest proof of concept. Falcon 9 cut launch costs from roughly $30,000 per kilogram to $3,000 — a 90 percent reduction over about a decade, after four decades of stagnation. Starship, if it delivers, cuts another 90 percent to around $300/kg. SpaceX accounts for over 80 percent of all mass launched from Earth and operates more than half the world's active satellites.

Blue Origin illustrates the cost of weak competition: its gradatim ferociter philosophy slowed more than intended by relying on Bezos's personal funding rather than NASA contracts or market pressure. New Glenn's recent orbital launch signals a return, targeting the national-security launch market.

The startup layer enabled by cheaper launch includes Firefly (soft lunar landing), K2 Space (returning to large satellites as Starship economics shrink mass penalties), and Varda (pharmaceutical manufacturing in microgravity, positioning as a manufacturing company that happens to use space).

Two structural risks dominate: orbital debris from proliferating constellations and geopolitical conflict extending into orbit. Venture capital's short horizons don't fit space development cycles; national-security procurement provides a stabilizing floor. Commercial space stations remain a chicken-and-egg problem — no ISS killer app yet — but market surprises are unpredictable. Artemis blends old and new: SLS and Gateway are legacy-model, CLPS commercial lunar contracts are new-model, and the program will likely drift more commercial over time.

SpaceX is the clearest proof of concept. Falcon 9 cut launch costs from roughly $30,000 per kilogram to $3,000 — a 90 percent reduction over about a decade, after four decades of stagnation. Starship, if it delivers, cuts another 90 percent to around $300/kg. SpaceX accounts for over 80 percent of all mass launched from Earth and operates more than half the world's active satellites.

Blue Origin illustrates the cost of weak competition: its gradatim ferociter philosophy slowed more than intended by relying on Bezos's personal funding rather than NASA contracts or market pressure. New Glenn's recent orbital launch signals a return, targeting the national-security launch market.

The startup layer enabled by cheaper launch includes Firefly (soft lunar landing), K2 Space (returning to large satellites as Starship economics shrink mass penalties), and Varda (pharmaceutical manufacturing in microgravity, positioning as a manufacturing company that happens to use space).

Two structural risks dominate: orbital debris from proliferating constellations and geopolitical conflict extending into orbit. Venture capital's short horizons don't fit space development cycles; national-security procurement provides a stabilizing floor. Commercial space stations remain a chicken-and-egg problem — no ISS killer app yet — but market surprises are unpredictable. Artemis blends old and new: SLS and Gateway are legacy-model, CLPS commercial lunar contracts are new-model, and the program will likely drift more commercial over time.

Mars or bust! My chat (+transcript) with aerospace engineer Robert Zubrin

Mars colonization is not a backup plan for civilizational collapse — it is how humanity adds a new branch of civilization that amplifies collective inventiveness. Robert Zubrin, founder of the Mars Society, dismisses Musk's "backup planet" framing as derived from Asimov's *Foundation* and unattractive to a public that doesn't feel existentially threatened. The real reasons: science (life's prevalence in the universe), challenge (frontier to energize youth), and the creativity dividends Martians will generate: cheap deuterium for fusion, breeder reactors, asteroid diversion.

On timeline, Zubrin thinks a 2033 human landing is achievable if the current administration commits now. A 2028 human landing is unrealistic, but a 2028 robotic Starship landing carrying 30 rovers, 30 helicopters, and a full science lab is feasible — two orders of magnitude more instrumentation than Perseverance — and would broaden the coalition beyond "Trump-Musk" to the university science community.

The mission architecture needs one addition: a "Starboat," 10–20% the size of Starship, used only for ascent. Musk's direct-return Starship plan requires producing 600 tons of methane-oxygen on Mars, demanding 600 kilowatts — beyond solar, requiring nuclear. A Starboat reduces propellant requirements by an order of magnitude. The same Starship-plus-Starboat stack rationalizes the moon program: one Starship tanker in lunar orbit services multiple Starboat landings, making the moon/Mars tradeoff a false choice.

NASA must shift from vendor-driven (Artemis: five incompatible flight systems, useless Gateway, SLS obsolete since the '90s) to purpose-driven, Apollo-style, with schedule as the forcing function. The National Team lander contract should become a Starboat contract interoperable with Starship on methane-oxygen.

Mars will never be autarchic — no Earth nation survives self-sufficiency, and a million-person colony can't manufacture an iPhone battery. Growth follows the colonization-of-America model: tiny bases, local food and materials, expanding industry over decades to a permanent base of 20–30 people by 2040.

On timeline, Zubrin thinks a 2033 human landing is achievable if the current administration commits now. A 2028 human landing is unrealistic, but a 2028 robotic Starship landing carrying 30 rovers, 30 helicopters, and a full science lab is feasible — two orders of magnitude more instrumentation than Perseverance — and would broaden the coalition beyond "Trump-Musk" to the university science community.

The mission architecture needs one addition: a "Starboat," 10–20% the size of Starship, used only for ascent. Musk's direct-return Starship plan requires producing 600 tons of methane-oxygen on Mars, demanding 600 kilowatts — beyond solar, requiring nuclear. A Starboat reduces propellant requirements by an order of magnitude. The same Starship-plus-Starboat stack rationalizes the moon program: one Starship tanker in lunar orbit services multiple Starboat landings, making the moon/Mars tradeoff a false choice.

NASA must shift from vendor-driven (Artemis: five incompatible flight systems, useless Gateway, SLS obsolete since the '90s) to purpose-driven, Apollo-style, with schedule as the forcing function. The National Team lander contract should become a Starboat contract interoperable with Starship on methane-oxygen.

Mars will never be autarchic — no Earth nation survives self-sufficiency, and a million-person colony can't manufacture an iPhone battery. Growth follows the colonization-of-America model: tiny bases, local food and materials, expanding industry over decades to a permanent base of 20–30 people by 2040.

How SpaceX is making a space Mayflower voyage possible: Flashback Friday

Freeman Dyson's 1977 thought experiment calculated that private asteroid colonization could be affordable at $40,000 per person — two tons of gear at $10/pound — by following the Mayflower and Mormon models rather than Apollo-style government programs. Government Island One colonies require 360 tons per person and are financially impossible without state backing; the minimalist 23-person asteroid expedition could be self-funded. SpaceX reusability has collapsed launch costs from $87,000/kg in 1960 to ~$4,000/kg in 2023, with Citigroup forecasting $100/kg by 2040 — right at Dyson's required price point, making the math real.

NASA and the New Space Age: My chat (+transcript) with James Meigs

NASA must exit the rocket-building business entirely. The Space Launch System costs over $4 billion per launch, uses Space Shuttle–era solid-fuel engines, and will carry astronauts using technology designed before those astronauts were born. SLS persists because Congress, since the Shuttle's 2011 retirement, has optimized for contractor jobs in powerful states rather than mission outcomes.

The commercial turn began in 2005 with a small cargo program; SpaceX, initially dismissed, won decisively and created a private-sector launch boom no other country has matched. The risk now is SpaceX's dominance — NASA needs six to ten competitive providers, not one supplier. Blue Origin's New Glenn is promising but largely unproven; Firefly and Stoke Aerospace are earlier stage. FAA licensing still takes too long and needs reform.

The withdrawal of Jared Isaacman's nomination was a serious blow. The proposed replacement budget cuts NASA roughly 25%, hits science missions hard, and eliminates nuclear propulsion research — the program most critical to affordable deep-space travel, where chemical fuels' low energy density makes long missions marginal and nuclear's high energy density makes them practical.

NASA's correct future role is funding what the private sector cannot: advanced propulsion, deep-space science, non-routine crewed missions. As commercial providers mature, NASA should hand off capabilities and move to harder frontiers.

The bright scenario is a small permanent lunar south-pole base by 2040, mining water ice and learning off-planet habitation close to home before Mars. The dark scenario is Congress recapturing NASA as a pork mechanism while China — actively courting the 50+ Artemis Accord signatories — becomes the credible space partner of choice.

The commercial turn began in 2005 with a small cargo program; SpaceX, initially dismissed, won decisively and created a private-sector launch boom no other country has matched. The risk now is SpaceX's dominance — NASA needs six to ten competitive providers, not one supplier. Blue Origin's New Glenn is promising but largely unproven; Firefly and Stoke Aerospace are earlier stage. FAA licensing still takes too long and needs reform.

The withdrawal of Jared Isaacman's nomination was a serious blow. The proposed replacement budget cuts NASA roughly 25%, hits science missions hard, and eliminates nuclear propulsion research — the program most critical to affordable deep-space travel, where chemical fuels' low energy density makes long missions marginal and nuclear's high energy density makes them practical.

NASA's correct future role is funding what the private sector cannot: advanced propulsion, deep-space science, non-routine crewed missions. As commercial providers mature, NASA should hand off capabilities and move to harder frontiers.

The bright scenario is a small permanent lunar south-pole base by 2040, mining water ice and learning off-planet habitation close to home before Mars. The dark scenario is Congress recapturing NASA as a pork mechanism while China — actively courting the 50+ Artemis Accord signatories — becomes the credible space partner of choice.

Blue Origin and America's spacefaring future

SpaceX is necessary but not sufficient for American space leadership. Trump's proposed NASA cuts slash the science budget 50%, terminating 41 missions including Chandra, Voyager, and Mars Sample Return — ceding planetary science to China. Meanwhile Blue Origin is finally accelerating: New Glenn debuted in January, a Mars mission follows late 2025, and Blue Moon MK1 could beat a stumbling Starship to the lunar surface in early 2026. At $1B revenue versus SpaceX's $15B it remains a distant second, but no longer a hobby.

China will probably beat America back to the Moon. Then what?

China will likely land taikonauts on the Moon before 2030 — its Lanyue lander, Long March 10, and Mengzhou spacecraft are all progressing, while Starship and Blue Origin's Mark 2 remain years away. The geopolitical damage would be real: Dean Cheng warns it signals "the end of American exceptionalism." But unlike Apollo, Artemis aims at sustainable infrastructure — reusable landers, in-space refueling, a cislunar economy. Losing the flag race might be the shock that finally prevents the post-Apollo retrenchment from repeating.

A Quick Q&A with economist Matt Weinzierl on the growing business of space

The commercial space revolution is about decentralization, not privatization — replacing Apollo-era centralized programs with market competition. The 2003 Columbia disaster forced a reckoning: the US could no longer orbit humans without renting Russian rockets.

SpaceX is the decisive inflection point. Falcon 9 cut launch costs from ~$30,000/kg to ~$3,000/kg — a 90% drop after four stagnant decades. Starlink took orbital satellites from 1,000 to 10,000+. Starship targets another 90% cut to ~$300/kg.

Commercial stations remain chicken-and-egg: no killer microgravity application has emerged despite decades of ISS use. VC timelines structurally mismatch space investing — hence billionaires and SPACs. National security demand provides a floor through downturns.

Two risks could short-circuit growth: runaway orbital debris as satellite counts multiply, and geopolitical rivalry militarizing a domain that has so far stayed relatively peaceful.

SpaceX is the decisive inflection point. Falcon 9 cut launch costs from ~$30,000/kg to ~$3,000/kg — a 90% drop after four stagnant decades. Starlink took orbital satellites from 1,000 to 10,000+. Starship targets another 90% cut to ~$300/kg.

Commercial stations remain chicken-and-egg: no killer microgravity application has emerged despite decades of ISS use. VC timelines structurally mismatch space investing — hence billionaires and SPACs. National security demand provides a floor through downturns.

Two risks could short-circuit growth: runaway orbital debris as satellite counts multiply, and geopolitical rivalry militarizing a domain that has so far stayed relatively peaceful.

Jeff Bezos plans a Space Age reboot

After the Apollo era, oil shocks and Limits-to-Growth pessimism displaced tech-optimism for half a century — replaced by Hollywood dystopias. Today's futurists sit in boardrooms: Bezos plans gigawatt-scale orbital data centers and lunar manufacturing within two decades; Musk bets on multiplanetary civilization; Altman and Amodei see AI compressing a century of progress into a decade. Bezos frames AI as a horizontal enabling layer like electrification — an "industrial bubble" that, like the internet, leaves behind durable infrastructure.

NASA (and moonshot 2.0) is in meltdown mode

NASA's dysfunction — acting administrator Duffy feuding with Musk, courting Blue Origin as a last-minute Artemis lander, floating a DOT merger — is PR optics masking strategic collapse. The agency has lost a fifth of its staff; Starship is behind schedule; SLS/Orion cost $30 billion for one flight. Former administrator Bridenstine calls beating China's 2030 crewed lunar timeline "highly unlikely." The chaos reflects a half-century "Down Wing" drift: Apollo's frontier daring replaced by risk aversion, cost-plus contracts, and political continuity over discovery.

The trillion-dollar space race: My chat (+transcript) with journalist Christian Davenport

America is probably going to lose the race back to the moon. Betting markets put China at 65% to land first, and Washington Post journalist Christian Davenport — author of *Rocket Dreams* — thinks that's roughly right. China has a space station, a Mars rover, two far-side lunar missions with sample returns, and a 2030 moon target insulated from democratic budget reversals. NASA lacks a functioning lander (Starship hasn't reached orbit after 11 test flights), SLS launches roughly once every two years, and the agency has no permanent administrator.

The symbolism is already visible: Apollo's flags are gone — Aldrin's knocked over by exhaust at liftoff, the rest bleached white by radiation. China now has two moon flags; the second was woven from basalt threads extracted from melted lunar rock, a deliberate demonstration of in-situ resource utilization.

On the commercial side, SpaceX launches Falcon 9 roughly every 48 hours and is the heavy favorite to build the crewed lander. Blue Origin is far behind — New Glenn has flown once. The book opens with Bezos in 2016, after SpaceX won ISS contracts Blue Origin had skipped, ordering his team to pursue everything SpaceX pursues. Engineers gravitate toward SpaceX because hardware is actually flying; both companies together act as training grounds for a broader startup ecosystem.

Long-term visions diverge: Musk wants a Mars city funded through Starlink; Bezos favors O'Neill cylinder orbital habitats, with Earth preserved as a park. Both depend on Starship-scale launch economics that don't yet exist. Known resource candidates include helium-3 for quantum computing, asteroid precious metals, and orbital solar power. Three or four space stations globally within ten years is plausible. A key structural risk: NASA's crewed launch capacity runs entirely through SpaceX's Dragon — a dependency Musk briefly threatened to withdraw during a dispute with Trump.

The symbolism is already visible: Apollo's flags are gone — Aldrin's knocked over by exhaust at liftoff, the rest bleached white by radiation. China now has two moon flags; the second was woven from basalt threads extracted from melted lunar rock, a deliberate demonstration of in-situ resource utilization.

On the commercial side, SpaceX launches Falcon 9 roughly every 48 hours and is the heavy favorite to build the crewed lander. Blue Origin is far behind — New Glenn has flown once. The book opens with Bezos in 2016, after SpaceX won ISS contracts Blue Origin had skipped, ordering his team to pursue everything SpaceX pursues. Engineers gravitate toward SpaceX because hardware is actually flying; both companies together act as training grounds for a broader startup ecosystem.

Long-term visions diverge: Musk wants a Mars city funded through Starlink; Bezos favors O'Neill cylinder orbital habitats, with Earth preserved as a park. Both depend on Starship-scale launch economics that don't yet exist. Known resource candidates include helium-3 for quantum computing, asteroid precious metals, and orbital solar power. Three or four space stations globally within ten years is plausible. A key structural risk: NASA's crewed launch capacity runs entirely through SpaceX's Dragon — a dependency Musk briefly threatened to withdraw during a dispute with Trump.

After we get to Mars, then what?

Mars exploration matters because it resolves humanity's deepest open question — whether we are alone — and no telescope or robot can substitute for human presence. Boyce calls space a "thin place" inducing moral recalibration; Jukic argues only boots on the ground settle the question; Dourado frames Mars as an "innovate-or-die" environment Earth no longer provides. A National Academies blueprint offers four mission campaigns varying in site count and scope, with one explicit rule: when budget forces cuts, life detection comes first.

How to get from here to there—way out there, actually

Becoming a spacefaring civilization is an institutional problem, not a technological one. Amos Otungo Ayienda's paper identifies three frontiers: the Achievable (permanent lunar ops, early Mars settlement — no breakthroughs needed), the Theoretical (fusion propulsion, closed-loop life support), and the Speculative (spacetime manipulation). The gap between feasible and realized is civilizational: scattered funding, risk-minimizing bureaucracies, and one-off missions produce visits, not presence. Boeing's Starliner — initially called "outstanding" despite a near-fatal thruster failure, now classified a Type A mishap — illustrates how institutional drift, not technical limits, closes the window.

Artemis flyby: A not-so-small step toward a lunar return—and more

The Artemis II lunar flyby — 248,655 miles out, surpassing Apollo 13's record — matters less for the distance than for what's changed beneath it. Apollo was a government mega-project that dragged a hired private sector along; Artemis's lunar landers are being designed and owned by SpaceX and Blue Origin as commercial assets. Falling launch costs and co-investing companies could unlock a multi-trillion-dollar space economy — sustained enterprise rather than one-off spectacle.

The high frontiersman: Gerard K. O'Neill

Gerard O'Neill accepted the 1970s neo-Malthusian diagnosis — overpopulation, resource exhaustion, inevitable decline — but rejected the steady-state cure as coercion bordering on totalitarianism (citing Robert Heilbroner: static societies tend toward *1984*). His counter: scarcity is a geography problem, not a physical law. Earth is simply the wrong size. The fix was rotating cylindrical habitats — Island Three runs 20 miles long, 500 square miles of surface, millions of residents — built from lunar and asteroid materials at Lagrange points. His 1981 book *2081* grounded the vision in daily life: three-day work weeks, domed towns, driverless cars. Space colonization was the only path to abundance that preserved civilizational freedom.

Elon Musk is a trillionaire. Why that's good news for all of us

SpaceX's trillionaire founder is worth celebrating precisely because innovators historically capture only ~2% of the value they create (per Nordhaus), leaving 98% for everyone else. SpaceX cut launch costs ~90% through reusable boosters and vertical integration, unlocking orbital cities, space-based solar, and asteroid mining. If the Starship vision — full reusability, orbital AI infrastructure, Mars cities — delivers, humanity's share dwarfs Musk's. More trillionaires built this way signals techno-capitalism working, not failing.

My Q&A with Francis Fukuyama on technological change and liberal democracy. Some sci-fi, too!

The IT revolution turned out not to be benign: distributing information distributed power, but it also eliminated editorial hierarchies that made journalism trustworthy, replacing them with cognitive chaos where conspiracy theories command credibility. Fukuyama, who called IT "benign" two decades ago, now sees it as having gotten there before biotech did.

On AI, he dismisses the Skynet extinction scenario as "just absurd" and rejected by serious researchers. He also pushes back on overcorrection: early evidence suggests generative AI may compress inequality by giving lower-skilled workers capabilities they previously lacked, reversing the pattern where computers mainly rewarded the educated.

The real danger is who controls the technology. Large language models require compute scale only a handful of corporations can muster — illustrated by Musk making unilateral Starlink battlefield decisions in Ukraine — turning strategic AI into private foreign policy without democratic oversight.

Regulation faces a structural problem: a competent AI agency needs to hire from industry at industry pay, something US civil-service rules make nearly impossible. The British digital regulator, with relaxed hiring constraints, is the instructive contrast. For platform power, Fukuyama's Stanford group proposed "middleware" — a competitive market of third-party content filters users choose for themselves — breaking platform monopoly without a politically impossible fairness doctrine.

On Silicon Valley life extension: he calls it "terrible." Generational turnover is a feature, not a bug — economies advance one funeral at a time, authoritarian regimes persist when dictators refuse to die, and a world of 170-year-olds cannot rotate talent or leadership. Billionaire death-fear funding this research is selfishness with civilizational costs.

On AI, he dismisses the Skynet extinction scenario as "just absurd" and rejected by serious researchers. He also pushes back on overcorrection: early evidence suggests generative AI may compress inequality by giving lower-skilled workers capabilities they previously lacked, reversing the pattern where computers mainly rewarded the educated.

The real danger is who controls the technology. Large language models require compute scale only a handful of corporations can muster — illustrated by Musk making unilateral Starlink battlefield decisions in Ukraine — turning strategic AI into private foreign policy without democratic oversight.

Regulation faces a structural problem: a competent AI agency needs to hire from industry at industry pay, something US civil-service rules make nearly impossible. The British digital regulator, with relaxed hiring constraints, is the instructive contrast. For platform power, Fukuyama's Stanford group proposed "middleware" — a competitive market of third-party content filters users choose for themselves — breaking platform monopoly without a politically impossible fairness doctrine.

On Silicon Valley life extension: he calls it "terrible." Generational turnover is a feature, not a bug — economies advance one funeral at a time, authoritarian regimes persist when dictators refuse to die, and a world of 170-year-olds cannot rotate talent or leadership. Billionaire death-fear funding this research is selfishness with civilizational costs.

Life in an AI utopia: A Quick Q&A with futurist and philosopher Nick Bostrom

Nick Bostrom argues that eliminating all suffering and injustice risks producing a purposeless future — meaning the deepest challenge of a "solved world" isn't survival but what humans do with themselves once AI has made labor economically obsolete. He considers superintelligence possible on extremely short timelines (even next year, though more likely a decade-plus), arriving as scaling unlocks qualitatively new capabilities beyond GPT-4. On post-work economics, he identifies one durable niche: consumers may pay a premium for human-performed tasks regardless of quality — human athletes, human priests — wherever the causal process matters, not just the output. He rates both utopia and dystopia still plausible by 2050, citing unsolved alignment, unpredictable political dynamics, and the fact that no civilization has navigated a machine-intelligence transition before.

A Quick Q&A on predicting super AI through bond markets with … Trevor Chow

Bond markets can predict transformative AI's arrival because consumption smoothing pushes real interest rates up regardless of whether AI turns out aligned or catastrophic. In the aligned case, expected abundance encourages borrowing now; in the existential-risk case, no reason to save produces the same result. Trevor Chow's paper validates this using cross-country inflation-linked bond data: higher long-term growth expectations do raise long-term real rates, contrary to prior findings. Stocks offer no clean signal — aligned and unaligned AI push equity prices in opposite directions.

The New York Times vs AGI

NYT tech reporter Cade Metz's "Why We're Unlikely to Get Artificial General Intelligence Anytime Soon" contains every piece of evidence for the opposite conclusion. Altman, Amodei, and Musk predict AGI in 1–4 years; Anthropic's Jared Kaplan says scaling limitations are "going away"; AlphaGo arrived a decade early; current systems already beat humans at high-level math and coding. Metz assembled the optimist case, then framed it as pessimist rebuttal — a textbook example of how headline framing determines what readers take away from identical facts.

Sam Altman's 'gentle singularity' message to an anxious public

Altman's "Gentle Singularity" is a deliberate counter-narrative to Amodei's "white-collar bloodbath" framing and Pew data showing only 17% of Americans view AI positively. His timeline: AI agents already here (2025), novel insights by 2026, embodied robots by 2027, self-sustaining robot supply chains and "larval recursive self-improvement" by late 2020s, intelligence costs converging toward electricity prices by 2035. On jobs, he invokes industrial-revolution precedent against Amodei's 20% unemployment prediction. Broad economic data and prediction markets don't yet confirm the short timeline.

Mark Zuckerberg's vision of 'personal superintelligence'

Zuckerberg's 616-word essay argues for AI as individual empowerment rather than labor replacement: personalized assistants that expand human agency and creativity, not systems that make humans economically redundant while governments distribute the surplus. He gestures toward intelligence recursion as evidence superintelligence is "in sight" by 2030, and invokes Neal Stephenson's *Diamond Age* Primer as his model — technology that educates without substituting for lived experience and thought. Pethokoukis contrasts this with Altman, Amodei, and Musk's displacement-plus-UBI framing, calling Zuckerberg's version the healthier vision.

Searching for signals of soonish superintelligence

Financial markets may encode expectations about superintelligence through real interest rates. A paper by Trevor Chow (Stanford), Basil Halperin (Virginia), and J. Zachary Mazlish (Oxford/GPI) frames transformative AI as a double-edged sword: rapid economic growth on one side, existential risk from misaligned superintelligence on the other. Real interest rates, which reflect expected future growth and risk, become a testable signal — market pricing of the AI future, not just expert opinion.

The strange divorce of nuclear fusion and super AI wagers

Prediction markets expect superintelligence within years but assign no comparable probability to cheap fusion power — a split that defies logic, since superintelligent AI should accelerate fusion research. The 1960s held the inverse confidence: Arthur C. Clarke and RAND expected controlled thermonuclear fusion by the 1980s–2000s; energy researcher Charles Scarlott called success "inevitable," underpinning visions of desalination plants, ocean-floor habitats, and Solar System colonization. The breakthroughs never came. Now the two technologies have swapped credibility ratings while remaining tightly coupled.

My Q&A with Francis Fukuyama on AI and liberal democracy. Some sci-fi, too!

Fukuyama now rejects his own earlier "benign IT" judgment: the internet destroyed trustworthy information hierarchies, replacing curated journalism with a world where conspiracy theories flourish. On AI, existential fears are Skynet fantasy rejected by serious experts; the real concern is political — who controls LLMs, which only the largest corporations can build. Early evidence suggests generative AI may reduce inequality by lifting lower-skilled workers, reversing the trend prior computing amplified. Regulation requires a new-model agency with delegated discretionary power and tech-industry salaries; the standard APA notice-and-comment process is too slow, and US civil-service pay makes the UK digital-regulator model hard to replicate. Dystopian science fiction is necessary: *1984* and *Brave New World* gave us the vocabulary — "Big Brother," "Telescreen" — to name present dangers. Life extension he calls a social disaster: generational turnover is how wrong ideas die; longevity just locks in authoritarian capture, as Franco and Castro demonstrated.

Two cheers for the 'San Francisco Consensus' on AI

The San Francisco Consensus — Eric Schmidt's label for Silicon Valley's bet that scaling yields transformative AI within five years — deserves hope but not policy commitment. Economists treat AI as a general-purpose technology like electricity: slow-diffusing, contingent on skills and institutions; CBO and the Fed still project slower growth for decades. Plan for slow diffusion, avoid monopoly lock-in and brittle regulation, keep the upside open — but don't book the miracle.

Waiting for AGI, again

Bond markets are the clearest signal that AGI is not imminent. Real interest rates should spike if markets priced transformative AI — whether through abundance or existential risk, future consumption changes in value. Instead, yields drift down after major model releases (Andrews and Farboodi). Prediction markets agree: Kalshi puts OpenAI achieving AGI before 2027 at 11%; Metaculus pins "general AI" at July 2033. Stocks in Nvidia, Anthropic, OpenAI price an important investment cycle — not a near-term Singularity.

Pro-immigration policy and mutual brain gain: A Quick Q&A with … economist Gaurav Khanna

High-skill immigration creates mutual brain gain rather than zero-sum drain. Khanna and Morales find that the H-1B lottery's uncertainty was essential: Indian students acquired IT skills hoping to reach Silicon Valley, but those who lost the visa lottery stayed home and seeded India's software industry, which eventually surpassed the US in software exports. Returnees transferred human capital back. Workers in both countries ended up better off. For a similar AI-era effect to emerge, wage signals and training institutions in origin countries must be in place — and the US gains even from foreign students who go home, through research contributions and cross-subsidized higher education.

Confronting global population collapse: A Quick Q&A with demographer Nick Eberstadt

Global population will soon shrink for the first time since the Black Death — not from catastrophe but from a worldwide collapse in the desire for children. Eberstadt rejects the modernization thesis: France's fertility decline preceded England's industrialization, and today some of the UN's poorest designated countries are already sub-replacement. The best predictor of fertility is simply how many children women say they want, which puts the cause in the realm of mentality, beyond economics.

Progressive pronatalism: A Quick Q&A with … philosopher Victor Kumar

Population decline is a progressive issue the left has ceded through negative partisanship — rejecting the problem because Vance and Musk identified it. Boston University philosopher Victor Kumar argues the left must build pronatalism grounded in gender equality and social services, not traditionalism. Climate antinatalism is a dead end: decarbonization decouples population from emissions. France's multi-pronged approach — tiered tax incentives plus parental leave plus subsidized childcare — shows no single intervention works. South Korea's 0.8 fertility rate illustrates that incentives fail when social institutions remain oppressive; liberation must accompany subsidy. Immigration stabilizes headcount but can't stop aging, and since immigrants adopt host-country fertility norms, it exports low fertility globally. The stakes extend to potential civilizational collapse.

America's immigration edge: My chat (+transcript) with policy expert Alex Nowrasteh

America's immigration restrictionism has shifted from illegal crossings to low-skill workers to H-1B visas — a move Cato's Alex Nowrasteh says reflects xenophobia, not economics. Opponents don't misunderstand the economic case; they don't care about it. The animating forces are cultural anxiety, a "purity" conception of nationhood, and locus-of-control psychology: when people feel government has lost control of something, they turn against even legal versions of it. Trump's achievement in 2015 was manufacturing that perception when crossings were actually low.

Border numbers: monthly apprehensions peaked at 250,000–300,000 under Biden, fell to ~40,000 by late 2024, and hit ~8,000 in Trump 2's first full month. Nowrasteh's fix: expanded legal pathways at all skill levels — people cross illegally because no lawful route exists for labor the economy demands.

Moravec's paradox complicates skills-selection: automation may eliminate cognitive work before physical labor, making government choices of winning skill categories unreliable. Empirically, immigration produces no measurable wage effect on native workers.

Assimilation takes roughly three generations; birthright citizenship and a civic rather than ethnic national identity give the U.S. a structural advantage over Europe. "Heritage American" is a European ethno-nationalist concept with no workable home here.

Global fertility is falling faster than UN projections; countries will soon compete for migrants. Visa cancellations and falling tourism are already damaging the talent pipeline. Nowrasteh's political lever: entitlement insolvency — admitting 100 million working-age immigrants would extend Social Security solvency by decades, enough to give skeptics pause.

Border numbers: monthly apprehensions peaked at 250,000–300,000 under Biden, fell to ~40,000 by late 2024, and hit ~8,000 in Trump 2's first full month. Nowrasteh's fix: expanded legal pathways at all skill levels — people cross illegally because no lawful route exists for labor the economy demands.

Moravec's paradox complicates skills-selection: automation may eliminate cognitive work before physical labor, making government choices of winning skill categories unreliable. Empirically, immigration produces no measurable wage effect on native workers.

Assimilation takes roughly three generations; birthright citizenship and a civic rather than ethnic national identity give the U.S. a structural advantage over Europe. "Heritage American" is a European ethno-nationalist concept with no workable home here.

Global fertility is falling faster than UN projections; countries will soon compete for migrants. Visa cancellations and falling tourism are already damaging the talent pipeline. Nowrasteh's political lever: entitlement insolvency — admitting 100 million working-age immigrants would extend Social Security solvency by decades, enough to give skeptics pause.

3 long-shot ways of solving the global baby bust

Government subsidies cannot reverse fertility decline once cultural attitudes have shifted; Eberstadt warns engineering birth rates risks authoritarian overreach. Three long-shots might actually work: religious revival (Mormons, Orthodox Jews, and devout Christians hold above-replacement rates because faith shapes *desire*, not just behavior); AI-driven abundance (rising incomes, shorter workweeks, and robot childcare echo a re-emerging wealth-fertility correlation among the ultra-wealthy); and space settlement (off-world construction restores civilizational purpose). Japan — tighter labor markets, rising wages, automation investment — shows adaptation beats resistance.

Bracing for depopulation: My chat (+transcript) with 'After the Spike' coauthor Dean Spears

Global depopulation — not slower growth, but sustained generational shrinkage — is the most likely default future, and no automatic stabilizer will reverse it. Two-thirds of humanity live in countries already below replacement. The world average fell from five births per woman in 1950 to 2.3 today and is still falling. Of the 26 countries where fertility has dropped below 1.9, not one has recovered to two.

The trend crosses every cultural explanation. India is now below two nationally, including in Uttar Pradesh — poor, religious, early-marrying — where women average 1.9 wanted births. Latin America sits at 1.8, driven by permanent contraception rather than delayed fertility. Work-family conflict explains the US story but not the global pattern. Societies are converging downward.

No policy solution exists yet. Spears draws an analogy to climate change: in the 1960s, the right response was not to ban the internal combustion engine but to build science and institutions over decades. The UN projects a population peak around 2084 — the same lead time applies. The task is research, deliberation, and awareness, so future generations can choose stabilization rather than accept shrinkage by default.

The economic case for more people goes beyond labor supply. Ideas are non-depletable: a discovery can be shared endlessly, but only if someone exists to make it. If AI complements human workers — historically the norm — fewer people is a larger loss, not a wash. Demand-side density matters too: specialized care and entire industries only exist where enough people share a need.

Spears declines to name an optimal size. His firm claim: any stabilized population is better than indefinite shrinkage. Ehrlich predicted collapse at current numbers; instead, food per capita rose on every continent and extreme poverty fell in absolute terms. A larger, stable, wealthy population is not a contradiction.

The trend crosses every cultural explanation. India is now below two nationally, including in Uttar Pradesh — poor, religious, early-marrying — where women average 1.9 wanted births. Latin America sits at 1.8, driven by permanent contraception rather than delayed fertility. Work-family conflict explains the US story but not the global pattern. Societies are converging downward.

No policy solution exists yet. Spears draws an analogy to climate change: in the 1960s, the right response was not to ban the internal combustion engine but to build science and institutions over decades. The UN projects a population peak around 2084 — the same lead time applies. The task is research, deliberation, and awareness, so future generations can choose stabilization rather than accept shrinkage by default.

The economic case for more people goes beyond labor supply. Ideas are non-depletable: a discovery can be shared endlessly, but only if someone exists to make it. If AI complements human workers — historically the norm — fewer people is a larger loss, not a wash. Demand-side density matters too: specialized care and entire industries only exist where enough people share a need.

Spears declines to name an optimal size. His firm claim: any stabilized population is better than indefinite shrinkage. Ehrlich predicted collapse at current numbers; instead, food per capita rose on every continent and extreme poverty fell in absolute terms. A larger, stable, wealthy population is not a contradiction.

Artificial intelligence faces a fertility test

National optimism drives fertility independently of income or policy: a Makridis-Piano study across 140+ countries finds a 10-point rise in "thriving" produces 7–9% higher birth rates, causation confirmed via cross-border Facebook friendship contagion. Childbearing is a long bet on the future — tax credits help at the margin, but ambient confidence is the real lever. AI could generate an expansive image of tomorrow, but 57% of Americans say it's advancing too fast and only 18% are optimistic about its impact. That anxiety may suppress fertility before wage gains arrive, echoing Engels' Pause.

Why 'Interstellar' is an Up Wing masterpiece

*Interstellar* is a pro-progress film precisely because the blight is never blamed on humanity — it's a mechanism to stress-test civilization. The real villain is Down Wing thinking: a society that rewrote the Apollo landings as propaganda, lost MRI machines, and abandoned space exploration, leaving itself defenseless against existential threats. Political scientist Aaron Wildavsky's framing applies — resilience comes from economic growth and technical progress, not caretaking. Cooper's repeated lament is the thesis: greatest accomplishments cannot be behind us because our destiny lies above us.

The Doomsday Clock needs a pro-progress switch to the Genesis Clock

The Doomsday Clock — set at 89 seconds to midnight in January 2025, its closest reading ever — is fundamentally miscalibrated: it retreated during the Cuban Missile Crisis, advanced against Reagan's Soviet strategy that helped end the Cold War, and moved backward under Obama. Its subjective methodology makes it unreliable as a signal. A proposed replacement, the Genesis Clock, would track progress toward abundance using objective milestones instead: AGI proximity, human lifespan reaching 120, off-planet colonies, cancer vaccine and Alzheimer's cure, asteroid deflection capability, atmospheric carbon declining, economically viable fusion, global undernourishment below 1%, species revival like the woolly mammoth, the poorest nation matching year-2000 average wealth, and productivity growth 50% above postwar norms. It would start at 5:53 AM — seven minutes to a symbolic Dawn.

Beyond 'The Jetsons': A new vision of a fantastic future that never was

The cultural optimism encoded in 1960s Googie architecture and *The Jetsons* — flying cars, Space Age curves, atomic-age confidence — was deliberately dismantled by the early 1970s environmental movement, which replaced it with a scarcity ideology and a regulatory regime (NEPA and related federal laws) that made large infrastructure projects prohibitively slow and expensive. This double punch of regulatory sclerosis and cultural technophobia explains why America went from dreaming of flying cars to struggling to build subway stations. The 2025 *Fantastic Four* film's retro-futurist aesthetic, set in an alternate 1960s Manhattan, arrives as AI, low-cost spaceflight, and biotech make a techno-optimist revival plausible — and culturally necessary.

Have we really passed Peak Brain Power?

Cognitive test scores have fallen sharply since 2012 across rich countries, 25% of high-income adults struggle with basic math, and book readership has collapsed — but the culprit is behavioral, not biological. Passive scrolling and infinite feeds have displaced self-directed attention without touching underlying capacity. The pessimism likely won't hold: Gen Z is already revolting against smartphones, AI is raising demand for higher-order cognitive and social skills (GenAI job postings show 36.7% higher cognitive skill requirements), and genetic tools may expand biological ceiling by 2050.

An Up Wing film from a Down Wing decade

The 1970s were a Down Wing decade — stagflation, *Soylent Green*, "Limits to Growth" — yet they generated some of history's most consequential breakthroughs: Intel's 4004 microprocessor, recombinant DNA, MRI, Viking landers on Mars, Voyager 1 and 2, GPS, and airline deregulation. The 1976 IMAX documentary *To Fly!*, which Carl Sagan called moving after five viewings, captured the counter-current: a patriotic arc from hot-air balloons to Saturn rockets, premiering at the Bicentennial as a deliberate argument that human destiny requires reaching further. Even grim eras contain green shoots.

Of 'Pluribus' and progress

Vince Gilligan's *Pluribus* dramatizes what a civilization loses when it optimizes for peace over disruption: an alien hive-mind absorbs most of humanity into blissful consensus, leaving a society that neither creates nor procreates, projected to starve within a decade. Creativity requires individuality, friction, and belief in a different tomorrow — all three extinguished by the collective. Dean Simonton's research confirms: breakthroughs arise in pluralist, rivalrous conditions like Periclean Athens and Renaissance Italy, not in harmony. Friction sparks ideas; the body dies without it.

The Doomsday Clock is a myth machine

The Doomsday Clock — now 85 seconds to midnight — governs by myth, not measurement. Caroline De Cock's *AI Tools, Not Gods* shows governance built on archetypes (Frankenstein, Terminator, Chaos God) regulating imagined dangers instead of real systems. The Clock has always tracked elite anxiety more than risk, creeping closer in the 1980s over Reagan's rhetoric alone. It also ignores AI as risk reducer: drug discovery, clean energy, grid management. A "Genesis Clock" counting capability gains would be the honest alternative.

Does 'Project Hail Mary' help Project Up Wing?

*Project Hail Mary* treats extinction as an engineering problem: Ryland Grace, like Mark Watney before him, always does the math. Eva Stratt's speech in Weir's novel grounds the stakes — pre-industrial history was "unrelenting misery" centered entirely on food production, so losing modern civilization means war, famine, and plague returning. The 1990s blockbusters (*Independence Day*, *Armageddon*, *Deep Impact*) shared this confident problem-solving ethos, reflecting Cold War victory and Digital Revolution optimism. Then CGI-enabled annihilation, climate anxiety, and financial crises pushed Hollywood toward survivalism and tragedy (*Don't Look Up*, *The Walking Dead*). Audiences have responded to *Hail Mary*; Hollywood should take the hint.

What's the 'greatest American utopian science fiction story' ever written?

Kim Stanley Robinson nominates the Gettysburg Address as the greatest American utopian science fiction story: "government of the people, by the people, for the people, *shall not* perish" is future-imperative — an injunction, not a description. Pethokoukis extends the frame: from Winthrop's "city upon a hill" through the Declaration's abstract principles to Lincoln, America has always been a utopian worldbuilding project run as a continuous experiment, with each generation obligated to keep it alive.

On Pope Leo's AI encyclical letter

Pope Leo XIV's *Magnifica Humanitas* rejects AI-driven transhumanism — transcendence comes through God, not technological optimization — a position compatible with the Up Wing goal of reducing suffering and expanding opportunity. The friction is economic: the encyclical frames automation mainly as job-destruction risk, ignoring how prior general-purpose technologies (steam, electrification, the internet) raised living standards. Economist and AI-expert surveys project growth without mass unemployment. The "technocratic paradigm" critique echoes Lewis Mumford's Mega-Machine thesis, which underestimated capitalism's decentralizing dynamism. The Vatican needs better economists.

Will the anti-obesity wonder drugs work wonders for the US economy?

GLP-1 drugs like Ozempic may already be delivering more measurable economic impact than AI. Citi and Goldman Sachs both estimate 0.5–1% GDP gains from reduced obesity-related labor loss. Goldman further projects the broader healthcare innovation wave — gene editing, AI drug discovery, biosimilars — could add 1.3% GDP ($360bn/year). Beyond weight loss, tirzepatide cuts diabetes risk 90%+ in overweight individuals; early data links GLP-1s to cardiovascular, kidney, and Alzheimer's benefits. Standard GDP metrics understate the true welfare gains.

MAGA vs. the transhumanists of Silicon Valley

Prediction markets put a 43% chance on a cure for aging by 2050 — and that prospect is already stress-testing the Trump coalition. Thiel, Altman, and Musk are funding a $125 billion human-enhancement industry; RFK Jr. and Jim O'Neill are clearing FDA runway for it. But Classic MAGA religious conservatives, who have treated gene editing and neural implants as Promethean overreach since the 2001 stem-cell fight, haven't changed. When enhancement moves from treating illness to upgrading healthy humans, the next culture war may be right-on-right.

When AI meets biotechnology: My chat (+transcript) with techno-futurist Jamie Metzl

Genetic engineering, AI, and biotechnology are not parallel revolutions — they are a single converging wave whose momentum is not up for debate. Jamie Metzl, senior fellow at the Atlantic Council and former WHO advisor on human genome editing, argues in *Superconvergence* that the outcome is undetermined even though the arrival is not: agriculture enabled writing, which built computer code, which underpins the machine learning now unlocking biology. The question is governance and values, not whether it happens.

Metzl rejects "AGI" as meaningful — he calls it "BS," arguing that defining human intelligence so narrowly that a pattern-matcher over digitized culture appears to exceed it is a failure of self-understanding. He prefers "machine intelligence," a distinct form the way dolphin cognition is distinct from human cognition. The actually transformative AI is "boring AI": the 50% speed-up in coding over the past two years, AI embedded in supply chains and HR systems, outputs people will call "progress" without noticing the substrate. This is how electricity worked — not felt as electricity, felt as light.

On biotechnology: we already live in a bio-engineered world (broiler chickens tripled in size in 70 years; dogs, corn, wheat, cattle bear no resemblance to wild ancestors), so the "newnimal" concept is less rupture than continuation. Xenotransplantation — pig kidneys genetically modified for human recipients — is already happening, with 110,000 Americans on transplant waiting lists.

On resistance: societies that cling to the prior order lose vitality and sovereignty. Europe's stagnation and Russia's war on Ukraine both illustrate the cost of opting out of technological momentum. Political backlash can delay adoption nationally but not globally — decentralized technologies migrate to willing hosts.

The closing note: two billion people currently locked out of quality education could each become their own Darwin once AI-powered personalized learning reaches smartphones globally.

Metzl rejects "AGI" as meaningful — he calls it "BS," arguing that defining human intelligence so narrowly that a pattern-matcher over digitized culture appears to exceed it is a failure of self-understanding. He prefers "machine intelligence," a distinct form the way dolphin cognition is distinct from human cognition. The actually transformative AI is "boring AI": the 50% speed-up in coding over the past two years, AI embedded in supply chains and HR systems, outputs people will call "progress" without noticing the substrate. This is how electricity worked — not felt as electricity, felt as light.

On biotechnology: we already live in a bio-engineered world (broiler chickens tripled in size in 70 years; dogs, corn, wheat, cattle bear no resemblance to wild ancestors), so the "newnimal" concept is less rupture than continuation. Xenotransplantation — pig kidneys genetically modified for human recipients — is already happening, with 110,000 Americans on transplant waiting lists.

On resistance: societies that cling to the prior order lose vitality and sovereignty. Europe's stagnation and Russia's war on Ukraine both illustrate the cost of opting out of technological momentum. Political backlash can delay adoption nationally but not globally — decentralized technologies migrate to willing hosts.

The closing note: two billion people currently locked out of quality education could each become their own Darwin once AI-powered personalized learning reaches smartphones globally.

RFK Jr.'s dumb war on mRNA vaccines and America's medical future

RFK Jr.'s cancellation of $500 million in BARDA-funded mRNA vaccine projects — following a $600 million cut to Moderna's bird flu contract — rests on the discredited claim that mRNA vaccines fail against respiratory infections. COVID mRNA shots saved millions of lives, won a Nobel Prize, and now show promise against pancreatic cancer in human trials. Researchers are already advised to scrub mRNA references from NIH grant applications. Meanwhile China accelerates its own biotech investment, making the retreat strategically self-defeating.

Why 'great genes' matter for the American economy

A 10-percentile rise in a parent's polygenic education score adds a month of schooling and nearly a full income-ladder rung for their children — but only half the effect is direct inheritance. The rest is "genetic nurture": higher-scoring parents build richer home environments, and adoptive parents' genes predict adopted children's outcomes despite no shared DNA. Mate sorting on visible traits like schooling stacks both channels further. Because environments are changeable by policy, better schools and neighborhoods are a more reliable path to a smarter population than embryo IQ screening.

An AI-genetics revolution — and arms race: A Quick Q&A with Bill Drexel

AI is a major accelerant to genetic engineering that's receiving almost no serious attention, crowded out by superintelligence fears. Bill Drexel (Hudson Institute) warns China's political system gives it structural advantages: mass genomic data harvesting, low ethical constraints, high risk tolerance. The real danger is a bioethics race to the bottom where the US crosses its own red lines to keep pace — Drexel considers this the most likely outcome. The 2018 He Jiankui case showed China responds to international opprobrium, suggesting a Political Declaration-style governance framework could constrain worst-case behavior even without formal cooperation.

The FDA's Moderna mRNA reversal

The FDA reversed its hold on Moderna's mRNA flu vaccine trial after industry backlash — but damage is real. Career staff blocked a vaccine with superior Phase 3 efficacy; VC funding collapsed from $510M to $174M, $500M in federal contracts were canceled, and 22 BARDA projects killed. Political suspicion of mRNA now threatens cancer vaccines, MS prevention research, and pandemic readiness. Scott Gottlieb urges Congress to mandate vaccine funding with enforceable milestones, bypassing regulators who can no longer hold the line.

The case for growth is the case for life

Economic growth and longer lives are the same thing. Since 1820, a 20-year-old's remaining life expectancy rose from ~40 to 60+ years; real income per person climbed from $2,700 to $55,000; and roughly one-fifth of that income now funds a modern medical sector. Huetsch, Krueger, and Ludwig show the causation runs both ways — prosperity financed medicine, longer lives reshaped the economy. AI may now accelerate both feedbacks simultaneously.