Joshua Saxe: No AI Safety Position Can Be Falsified Yet, So Build Agile Institutions
joshua_saxe · x · 2026-10-12
AI safety researcher Joshua Saxe published a long-form argument surveying the AI safety debate and calling for epistemically humble institutions.
- Core claim: All serious positions in the safety debate are deeply considered but cannot yet be falsified because the deciding data hasn't arrived; institutions should be built to support multiple very different futures.
- Positions surveyed:
- Jensen Huang: Market forces and existing laws will force AI companies to deliver controllable AI — control is a normal engineering problem, full speed ahead. Saxe concedes no major AI damage has occurred yet.
- Sayash Kapoor & Arvind Narayanan: Control is hard but solvable with today's technology plus the right legal incentives; the real issues are incentives and culture, and institutions must adapt as they did for railroads, electrification, and aviation.
- Yudkowsky, Alexander, Kokotajlo, Bengio, Hinton et al.: Smarter AI becomes illegible and hard to monitor; race dynamics and government dysfunction could grant dangerous AI authority before control is solved, requiring extraordinary interventions like a pause or globally regulated data centers.
- Author's stance: We should price in extreme technological surprise (e.g., a single misaligned self-replicating agent causing crisis), but it's premature to adopt Alexander-style extreme scenarios — Saxe aligns more with the Kapoor/Narayanan camp.
Related event: Joshua Saxe Argues AI Safety Debates Are Unfalsifiable(4 posts)→
More from AGI Musings
- Suleyman: Claude isn't conscious, and Anthropic's constitution is circular logic — mmitchell_ai · 2026-10-12
- Damned either way: using AI more is 'cognitive surrender,' less is 'boomer mindset' — paulnovosad · 2026-10-12
- Scott Aaronson: the singularity has already begun for math — ChrSzegedy · 2026-10-12
- AI Labs 2026: You Pay for the Kitchen, Not the Meal — gerardsans · 2026-10-12
- Meme: how did other species fare after encountering a 'superintelligence'? — Puzzleheaded-King584 · 2026-10-12
- Breiman's 2001 'Two Cultures' paper resurfaces as AI makes black boxes the norm — akbirthko · 2026-10-12