AI governance report should also track how much AI-produced research humans actually understand
ronbodkin · x · 2026-11-19
Ron Bodkin congratulates the authors of California governor's AI governance report for its substance, praising its urgent call for transparency mandates and limits on automated AI R&D, plus risk containment protocols and a security floor.
He suggests one addition: alongside reporting the share of R&D at the highest automation levels, labs should report how much AI-produced research their people can actually explain. This "understandability share" would directly measure whether automation is outrunning oversight, and the report's IVO staff-interview provision makes it feasible.
Related event: Scholars Propose "Comprehension Audits" to Govern Recursive AI R&D(3 posts)→
More from AGI Musings
- If AI execs truly believed the risk, they'd be coordinating a stop — they aren't — wfithian · 2026-11-19
- AI Safety Community Emotionally Processing x-Risk: 'My Biggest Regret Is Not Doing This Earlier' — trevposts · 2026-11-19
- "Obsolete ideas have tenure": why AI experts anchor on outdated views — repligate · 2026-11-19
- Stanford researcher: most scientific output doesn't really need peer review — anshulkundaje · 2026-11-19
- Stanford prof Anshul Kundaje: most scientific output doesn't really need peer review — anshulkundaje · 2026-11-19
- Michael Nielsen: the real ASI danger is raw power, not a rogue misaligned system — michael_nielsen · 2026-11-19