Critic: AI Safety's Existential-Risk Framing Sets a Needlessly High Burden of Proof

jjvincent · x · 2026-09-10

Writer Eric Levitz argues AI safety advocates overemphasize "existential risk": it's extremely hard for anything to kill literally all humans—nothing has yet—while "could kill a huge number of people" justifies precautions just as well in nearly every context. In the quoted exchange, Osita Nwanevu counters that nukes have clear extinction mechanisms, whereas an LLM killing everyone relies on speculative stories like "it convinces someone to build a bioweapon that reaches all humans." A representative pro/con clash in the AI safety debate.

Original post →

More from AGI Musings

AGI Musings channel →