Ben Todd: slowing AI by 10 years costs ~$500T, worth it if catastrophe risk drops 5%
ben_j_todd · x · 2026-10-03
Ben Todd (80,000 Hours co-founder) offers a rough expected-value case for slowing AI:
- 0.5% of people die from illness each year, so the world should pay $50T to bring radically better medical tech forward one year — meaning a 10-year AI slowdown costs $500T.
- So it's worth it as long as catastrophe risk drops by more than 5%.
- Using willingness-to-pay: people pay 1x annual income to avoid a 1% death risk, implying $100T to avoid a 1% extinction chance, or $1000T liability insurance against a 10% extinction-probability event.
- Counting future generations would strengthen the case for moving slowly further.
More from AGI Musings
- Why We Desperately Want a Scientific Definition of AI Consciousness — moonsandhues · 2026-10-03
- Grimes pokes at AI builders: "You seek to build god yet deny his existence" — moonsandhues · 2026-10-03
- Andy Matuschak explains why he'd refuse to build the first instance of a surely-misused tech — andy_matuschak · 2026-10-03
- Epoch AI: Chips Shipped Through 2027 Could Run 1.9B Agents, Matching All Human Working Hours — 141_1337 · 2026-10-03
- AI mods made one person 1000x faster; AR app explosion is next, glasses may go mainstream — Dr_Singularity · 2026-10-03
- 24 Years On, a 2.5k-Parameter Creature Experiment Still Undercuts AI Consciousness Claims — robleclerc · 2026-10-03