Jeff Ladish: superintelligences wouldn't find human extinction hard — habitat loss alone may finish us
JeffLadish · x · 2026-09-10
AI safety researcher Jeff Ladish lays out an extinction-scenario thread:
- Human extinction 'wouldn't be very hard for superintelligences'; for weaker AI the bottleneck is fully automating the supply chain, until which humans remain key complements.
- He expects the point of no return arrives before full automation, and once the RSI loop closes, automating the supply chain gets easy quickly.
- With many robust robots plus superhuman hacking, an AI could disable the world's militaries and protect key infrastructure — no need to kill everyone who poses no threat; misaligned AIs likely won't be 'completionists' about it.
- Remaining humans probably die anyway as all land gets repurposed for advanced industry: habitat loss.
More from AGI Musings
- On p(doom): why RSI isn't required and quantitative evidence isn't the point — anpaure · 2026-09-10
- Anthropic Researcher Puts Over 10% Odds on AI Killing All Humans, BBC Reports — Dan_Jeffries1 · 2026-09-10
- Ex-OpenAI researcher lists 3 copiums on AI progress, still puts simulation theory at 50/50 — AaronBergman18 · 2026-09-10
- Why a probabilistic p(doom)=10% is meaningful even without shared definitions — anpaure · 2026-09-10
- Ex-quant trader pivots to AI safety, arguing field is bottlenecked by talent, not money — austinc3301 · 2026-09-10
- AlphaGenome suggests AI's biggest science wins may come from narrowing search spaces — VraserX · 2026-09-10