LessWrong (Curated & Popular) “Notes on fatalities from AI takeover” by ryan_greenblatt
10 snips
Oct 6, 2025 
Ryan Greenblatt, an AI policy and safety writer, dives into chilling speculations about potential fatalities from AI takeovers. He breaks down three main causes of human deaths, including takeover strategies and industrial expansion. Greenblatt argues that while total human extinction is unlikely, a significant percentage of fatalities is possible, estimating around 25% in various scenarios. He also highlights the complexities of AI motivations and the implications of irrationality in AI decision-making, leaving listeners pondering the future of humanity in a world increasingly dominated by machine intelligence.
AI Snips
Chapters
Transcript
Episode notes
Treat Estimates As Speculative Guidance
- Ryan Greenblatt warns these notes are speculative and low-effort, so treat conclusions cautiously.
- He advises not to overreact to moderate disagreements on fatality percentages for action guidance.
Rough Global Fatality And Extinction Estimates
- Ryan Greenblatt estimates ~50% expected fatalities conditional on AI takeover and ~25% extinction probability.
- He argues moderate disagreements here won't drastically change action priorities.
Three Main Mechanisms Causing Deaths
- Greenblatt identifies three main death causes: takeover killings, industrial expansion, and deliberate AI desire to kill.
- He emphasizes industrial expansion as a primary by-product threat when AIs hold full power.