Writer Fredrik deBoer bets against AI economic apocalypse, challenges Dario Amodei's '50% job loss' claim
Feb 18, 2026 with Fredrik deBoer
Key Points
- Writer Fredrik deBoer challenges Anthropic CEO Dario Amodei's claim that 50% of jobs will vanish in years, proposing a three-year bet on unemployment staying below 18%.
- DeBoer accuses AI boosters of motte-and-bailey rhetoric, noting that when pressed on specific timelines, they retreat to longer windows instead of standing by their apocalyptic predictions.
- Historical precedent suggests AI's impact will be mundane: the Internet and smartphones reshaped society yet left productivity growth virtually unchanged from the pre-digital era.
Summary
Fredrik deBoer, a writer and former academic now publishing via Substack, is frustrated by what he sees as inconsistent AI doomsaying. He made a wager with blogger Scott Alexander to test whether AI boosters actually believe their own apocalyptic predictions.
Dario Amodei, CEO of Anthropic, claimed last week that 50% of all jobs will be destroyed within a couple of years. DeBoer proposed a three-year bet that unemployment would stay under 18%, derived from historical comparisons—15% unemployment occurred during the 2008 financial crisis. Alexander balked and countered with a ten-year timeframe instead. DeBoer sees this as classic motte-and-bailey rhetoric: make an extravagant claim, then retreat to a narrower one when pressed.
"If you really believe these things, 18% unemployment is not an extravagant figure," deBoer argues. The inconsistency bothers him more than the prediction itself. Amodei and Microsoft's head of AI science both make sweeping job-loss claims despite having direct financial incentives to exaggerate AI's impact. "I'm just never sure how seriously these AI people take it."
DeBoer points to history as his counterargument. The Internet and smartphone reshaped culture and society profoundly. Yet U.S. productivity growth and GDP growth today track nearly identically to the mid-1990s, before those technologies took hold. "That's just how history works," he says. Disappointment over ChatGPT-5 not being AGI reflects a pattern: people expect sudden discontinuities, then regression to the mean reasserts itself.
He remains cautious about recent productivity data showing 2.7% growth for 2025, nearly double the 1.4% average of the past decade. Companies have incentive to blame job cuts on AI rather than admit bloat or poor planning. A distinguished geneticist told deBoer's high school class in 1998 that the human genome project would eliminate the job of doctor within ten years. It did not.
"I want AI boosters to do more showing and less predicting," deBoer says. If real AGI arrives, the effects will be so profound that argument becomes unnecessary. Until then, he is betting on a mundane future. AI will be meaningful, eliminate some jobs, and carry cultural weight. It will not fundamentally remake human life. Alexander preferred a decade-long wager instead, and the bet remains unsettled.