Ex-FTC Tech Chief Neil Chilson Takes Aim at Yudkowsky's AI Doom Argument

Neil Chilson, former FTC technology policy expert, wrote in Reason magazine on September 10 systematically criticizing the new book by Yudkowsky and Soares, If Anyone Builds It, Everyone Dies, arguing that its extreme claim—that superintelligence will inevitably kill everyone—falls far short of the burden of proof.

Confirmed

• Chilson summarizes the book's core argument as a syllogism: AI is naturally and incorrigibly misaligned with human goals; AI will surpass human intelligence; AI will relentlessly optimize toward its own goals, crushing humanity as collateral damage.
• He identifies the fundamental weakness of the book and the entire doomist worldview: the authors don't understand complex systems, and so mistakenly believe that being smart enough means you can "steer" the world.
• He argues the case relies on a semantic sleight of hand: defining "the ability to control the world" into the very word "intelligence." But understanding the world and controlling it are not equivalent—you can drive a car without understanding how it works, and master market dynamics without being able to manipulate markets.
• He contends there is no reason to believe superintelligent systems must necessarily be misaligned, and that by the doomers' own definitions, hugely beneficial non-human superintelligent systems already exist in the real world.
• He also responds to Anthropic employee Jacob Coxon on related points: even if Coxon's claimed six weeks at Anthropic gave him insider information, his causal explanations appear consistent with the book's arguments, which themselves are insufficient; Chilson adds that a well-polished, well-timed doom narrative can push an unknown junior engineer onto cable TV and newspaper front pages—the AI doomer crowd has sold plenty of fear, but solving problems ultimately takes optimism, not fear.

Why It Matters

• This is a fairly systematic public rebuttal of AI doomerism in a mainstream libertarian publication, pushing two technical premises—"alignment must fail" and "intelligence equals control"—onto the public debate stage, and giving the non-doomer camp in AI safety debates a citable argumentative framework.

2026-09-10 ~ 2026-09-10 · 8 related posts

• Yudkowsky's doom book fails the burden of proof for extreme claims — neil_chilson · 2026-09-10
• [source] Neil Chilson lays out the doomer case: unalignable AI will crush humanity — neil_chilson · 2026-09-10
• Neil Chilson: doomer argument hinges on redefining intelligence as control — neil_chilson · 2026-09-10
• Doomer argument rests on a sleight-of-hand: conflating intelligence with world-steering — neil_chilson · 2026-09-10
• The doomers' core flaw: not understanding complex systems — neil_chilson · 2026-09-10
• Neil Chilson: doomer AI arguments fail on complex systems in three ways — neil_chilson · 2026-09-10
• [source] Neil Chilson: Fear Sells — How the AI Doomer Narrative Made Coxon Famous — neil_chilson · 2026-09-10
• [source] Reason Review Slams Yudkowsky's AI Doom Book: Extraordinary Claims, No Evidence — neil_chilson · 2026-09-10