Anthropic researcher uses aviation fatality history to frame how much AI harm society should tolerate
robleclerc · x · 2026-09-16
Rob LeClerc's long-form thread (with a correction: the compute figure should be Quadrillion, not Quintillion) argues that while he doesn't expect existential risk from AI, AI won't be risk-free either — the real question is how much harm society is willing to tolerate.
- He points to aviation fatality history as sober guidance: aviation was permitted to kill thousands of people a year for half a century because its benefits outweighed the risks (the big spike in the 1940s was WW2).
- Zero risk was never the standard; AI governance should similarly balance benefits against an acceptable level of harm.
A substantive big-picture take on AI risk framing from a notable figure.
Related event: Researcher Uses Aviation Fatality History to Frame AI Risk Tolerance(2 posts)→
More from AGI Musings
- Turing warned in 1951 machines could take control; RT mocks execs who ignored warnings — DKokotajlo · 2026-09-16
- eigenrobot mocks AI safety researchers who equate alignment with being a decent person — eigenrobot · 2026-09-16
- Zoom is to X what early telephony was to the internet: mapping AI-mediated sensor networks — curious_vii · 2026-09-16
- The Myth of the Rogue AI: Why Superintelligence Doom Collapses Under Real Engineering — deadgirlrevvy · 2026-09-16
- Recruiter: 30/60/90-Day Plans Are Dead, Everything Has Accelerated — morganb · 2026-09-16
- Sara Hooker calls 10% AI extinction risk claims highly irresponsible — sarahookr · 2026-09-16