P(survival|Plan C) ≠ P(survival|do(Plan C)): a causal inference objection
quetzal_rainbow · x · 2026-11-01
Replying to ApriiSR's call for Plan C planning, quetzalrainbow flags an apparent probabilistic sleight of hand: conditional on learning humanity survived, easier plans seem more likely to have succeeded—but P(survival|Plan C) is not P(survival|do(Plan C)), distinguishing observational conditioning from causal intervention.
Related event: AI Safety Community Debates Whether to Plan for 'Plan C' Scenarios(3 posts)→