The self-contradiction in assuming a level-n superintelligence civilization is near-optimal
jessi_cata · x · 2026-11-18
jessicata poses a thought experiment on relative intelligence advantage: if you intuit that dropping a level-n superintelligence onto 2026 Earth would outright win, why wouldn't a 10n superintelligence stomp a civilization of level-n superintelligences?
She argues the standard rebuttal — that the level-n civ would have self-improved to 10n — concedes there are returns to intelligence beyond level n. And if they stayed at level n because they 'found optimal solutions to all problems', yet would lose to a 10n arrival, the hypothetical is self-contradictory. Her conclusion: if the 10n intelligence would win, then a level-n civilization would never have stayed at level n. A substantive alignment-adjacent debate on returns to scale in intelligence and what 'near-optimal' could even mean.
More from AGI Musings
- AI assistant products are rapidly commoditizing as the barrier to entry hits zero — signulll · 2026-11-18
- FAU professor replies to DIE ZEIT: AI makes intellectual work scalable, not universities obsolete — maier_ak · 2026-11-18
- Stanford's Anshul Kundaje: scientific deliverables will diversify, peer review will become 'a complete farce' — anshulkundaje · 2026-11-18
- Youth unemployment at 7.5% is better than any point from 1974-1994, analyst argues — binarybits · 2026-11-18
- Epoch and Anthropic capability indices both show a clean trend break to 1.7-2x progress after coding automation — tszzl · 2026-11-18
- US hiring rate back at 2013 levels as economists clash over the jobs slowdown — binarybits · 2026-11-18