An open letter to Scott Alexander

Steven Pinker writes to Scott Alexander to refuse a staged debate and to clarify his longstanding critique of apocalyptic AI scenarios. He reiterates arguments from a 2018 chapter that such dooms rely on four faulty assumptions: a crude, scalar view of intelligence; conflating intelligence with human-like drives such as dominance and self-preservation; assuming an AI would single-mindedly pursue an absurd goal (paperclip-maximizer style) without managing tradeoffs; and granting engineered systems irreversible omnipotence over physical infrastructure. He emphasizes these are substantive counterarguments, not personal attacks, and notes that many respected researchers and a 2025 RAND modeling exercise found no describable extinction scenario even under worst-case assumptions. He concedes that large language models surprised experts and that the field’s rapid, sometimes reckless deployments require updating some prior assumptions about engineering culture.

Pinker warns that doomer rhetoric is counterproductive: an apocalyptic worldview can encourage risky behavior by frontier labs convinced they must “win” a superintelligence race, and historically catastrophic warnings have produced arms races and harmful policy decisions. He accepts real, mundane AI dangers - bioterrorism, cyber-sabotage, unconstrained agents - and calls for sober safety practices: independent oversight, accident investigations, liability, humans-in-the-loop, and kill switches. He defends irreverent tone, apologizes for any inadvertent offense, and expresses admiration for Alexander while urging debate focused on evidence and policy, not spectacle.