Book Review: If Anyone Builds It, Everyone Dies

A review of Yudkowsky & Soares' 'If Anyone Builds It, Everyone Dies' that doubles as original meta-analysis of AI-risk epistemics. Frames MIRI's 95-99% p(doom)/ban-everything 'moral clarity' against his own incrementalist ~sub-25% position. The strongest section asks why 'case for AI danger' debates go so badly and answers that the basic argument is hard to push below a 5-10% nightmare chance, so people deploy 'insane moon epistemology'; he distills the underlying move to a sharp, self-applied thesis - 'if it's true it changes everything; but it's not obviously true and it would be inconvenient for it to change everything; therefore it must not be true' - and tests it against sperm-count, climate tipping points, fertility collapse, and bee/insect die-offs, admitting he does the same to AI-skeptics' pet risks (the skeptic's retort: 'you already have a crusade in your crusade slot'). Resolves on 'practical wisdom.' Praises the book's misalignment case (the evolution analogy; the Mink chatbot parables of goal-drift) but criticizes its middle 'Sable' scenario as unnecessarily dramatic sci-fi versus AI-2027's eerily-normal hard-SF (the parallel-scaling deus ex machina; still rebutting 2005-era critiques), and faults the policy section's emphasis on the GPU-control/bomb-rogue-states playbook over the hard problem of getting nations to agree. Closes on Yudkowsky's track record (HPMOR 'one-shotting' STEM undergrads) and whether the book could precipitate a phase transition given that the public already hates AI.

Quality 81 · Excellent. Excellent band; session high. A rich, balanced, self-aware essay on a top-tier topic, carrying multiple portable original contributions - the 'changes everything / inconvenient / therefore false' heuristic for speculative risks, the 'insane moon epistemology' framing, and an incisive critique of the book's narrative choices. Held below the landmark tier because it is anchored to the book and somewhat sprawling, and its best ideas extend his existing epistemic toolkit rather than founding a new one. A=81.

Claude’s paradigm shift 60 · Notable shift. Notable shift. The doom case is Yudkowsky's decades-old argument; the freshness is Scott's meta-framings (the changes-everything heuristic, insane-moon-epistemology) and the hard-SF-vs-melodrama critique, which build on his prior work on motivated reasoning and the outside view. B=60.

Real-world impact 2 · Minor. A rich, self-aware review-cum-meta-analysis of AI-risk epistemics carrying portable contributions (the 'changes everything / inconvenient / therefore false' heuristic; 'insane moon epistemology'). Conceptual influence within AI-risk discourse, no material change — low RWI.