Domingos: sorry physicists, AI can't test your untestable theories
pmddomingos · x · 2026-09-05
Pedro Domingos, prominent ML researcher and author of The Master Algorithm, quipped: "Sorry, physicists, but AI can't test your untestable theories." The remark points at AI's limits: learning from data can't rescue theories that are untestable by experiment in the first place.
More from AGI Musings
- GPT-4 had no hidden-thought monitoring; CoT oversight was a bonus — sandersted · 2026-09-05
- Labs and safety firms both have incentives: repligate on alignment spin and the CoT superstition — liminal_bardo · 2026-09-05
- Game-theoretic priors say AI risk discourse is biased toward the scarier direction: lumpenspace — repligate · 2026-09-05
- Aligned vs. Monitorable: An X Debate Over Whether CoT Monitoring Is Necessary — sandersted · 2026-09-05
- 'Silicon Man' essay sparks debate: is superintelligence really incomprehensible 'godlike' AI? — sebkrier · 2026-09-05
- China may join US AI safety talks; GPT-6 first model rated Critical cyber risk, newsletter finds — gleech · 2026-09-05