Scott Alexander: Handoverism isn't successionism — AI rule might serve human interests
Astral Codex Ten · rss · 2026-10-21
Scott Alexander (Astral Codex Ten) pushes back on critiques of "AI successionism," arguing critics conflate two distinct positions:
- True successionism: AIs should take over, steamrolling human interests — which he firmly opposes and calls the target of AI safety.
- Handoverism: Humans should prioritize their own interests, but those interests might be best served with AIs in positions of control — a position he and some in AI safety find plausible, now being smeared by conflation.
He argues from a liberal outlook: the state is a service provider dispensing coordination, order, and redistribution; if robot legislators deliver these better than humans, automating governance is no different from automating trash collection. Two escalating visions:
- Kemalist AI: Once AI replaces laborers and soldiers, the traditional democratic safeguard (buy-in of workers/soldiers) collapses, and elites — even "democratic" ones — could seize permanent power. The only fix is military AIs aligned to democracy itself, which is already a form of handover.
- President Claude: If a perfectly aligned, smarter, honest Claude 10 ran in 2028 against human politicians and beat them on every measurable quality, what principled reason would you have not to vote for it?
Alexander concedes the exercise is speculative, but insists the labeled researchers are discussing a different proposition: that human interests could be better served by AI in charge.
More from AGI Musings
- Anthropic taps law firms to teach AI agents how to negotiate deals — jkubicki · 2026-10-21
- komorama's weekly: the AI Peter Principle—first 90% never faster, second 90% never slower — lennysan · 2026-10-21
- a16z talks health AI with OpenAI's Karan Singhal: 300M people weekly ask ChatGPT health questions — a16z · 2026-10-21
- Gartner unveils Top 10 Strategic Technology Trends for 2027 as AI talk shifts to governance — brucemacv · 2026-10-21
- 'Training your own models is an incredibly stupid idea' — why RAG beats fine-tuning pitches — rseroter · 2026-10-21
- HeyGen Founder Joshua Xu: Agents Should Explain with Video, Not Text Reports — toolstelegraph · 2026-10-21