Weak AI safety regulations: valuable or a distraction from extinction risk?

In a new LessWrong post, MichaelDickens systematically examines whether weak AI safety regulations are valuable for reducing the existential risk from superintelligent AI. He outlines several potential benefits: direct risk reduction (e.g., GPU tariffs slowing foreign AI development, mandatory safety tests catching catastrophic issues), empowering future efforts (monitoring AI developers creates infrastructure for later bans), revealing warning shots (mandatory test publications and whistleblower protections make dangerous AI behaviors visible), and shifting the Overton window (light regulation may pave the way for stronger laws). However, the evidence for the latter is inconclusive, citing a meta-analysis on the foot-in-the-door effect that showed mixed results.

On the downside, Dickens highlights three major concerns: opportunity cost (time spent on weak regulation might be better spent on strong bans, though supporting bills like SB 53 or the RAISE Act may not crowd out stronger alternatives), regulation fatigue (existing rules could reduce appetite for further regulation), and slowing beneficial technological progress. He concludes that while strong international agreements banning ASI until proven safe are the ideal, weak regulations may still be worth pursuing if they don't impede stronger efforts and help build the political and monitoring infrastructure needed for future action.