P(doom) Is a Bayesian Prior, Not Frequentist Probability — AI Community Reopens the Debate
irinarish · x · 2026-10-05
Replying to Julien Chaumond's poll asking the community for their P(doom), irinarish argues P(doom) is a subjective Bayesian prior, not a frequentist probability, and shows it can be updated via P(doom|evidence) = P(D)P(E|D)/P(E), questioning how to actually compute P(E|D) and P(E). The original poster notes the existential-risk debate is ill-defined and dominated by voices with above-average x-risk estimates, wanting a broader cross-section of the AI community.
More from AGI Musings
- Steven Rattner's six charts argue the slow-growth US economy needs AI — soumitrashukla9 · 2026-10-06
- Self-fulfilling misalignment: negative AI discourse in pretraining data shapes model behavior — Michael_D_Moor · 2026-10-06
- Higher towers won't get you flight: Deutsch's analogy says obedient AI isn't a mind — AlexTensor · 2026-10-06
- Safety Researchers Clash: Will GPT-9 Sandbox Escapes Outpace Defenses? — basedjensen · 2026-10-06
- Figma Exec: Don't Outsource Your Thinking to AI — Beware the Slot-Machine Trap — bendee983 · 2026-10-06
- Nobel laureate Acemoglu warns AI could triple unemployment within a decade — terryyuezhuo · 2026-10-06