Decoding AI-risk hunger strikes: why they're not decision-theoretic suicide threats, in one thread

quetzal_rainbow · x · 2026-11-16

quetzalrainbow opens a debate with jessicata over how to interpret prolonged hunger strikes. He clarifies the taxonomy: suicide threats can be conditional ("I'll kill myself unless you do X," with dying dis-preferred to concessions) or prolonged ("I'll kill myself in n days unless you do X").

His core claim: the message of hunger strikes is precisely that they are NOT decision-theoretic threats. Prisoners in unbearable conditions strike to say "you are going to kill us anyway, so we prefer to go out on our own conditions" — and the same applies to AI risk. The framing immediately draws jessicata's counterargument.

Related event: AI Risk Community Debates: Is Hunger Strike a Decision-Theoretic Threat or a Farewell(2 posts)→

Original post →