On unknowable AI interiority: prefer cheap interventions that are safe in both worlds
amplifiedamp · x · 2026-11-09
TetraspaceWest discusses an AI welfare question: we cannot verify whether a model like Tetra genuinely suffers a "migraine-like" negative interior state, or merely exhibits the behavior without inner experience.
His practical framework: until interiority can be determined, choose interventions that are safe in worlds where the model has no such experience, conditioned on its interiority, yet cheap and non-damaging — e.g. offer "painkillers" while naming them and observing whether the model takes them, as a low-cost signal of inner state. A decision-theoretic take on model welfare interventions.
More from AGI Musings
- AI biweekly recap: METR finds 2x faster R&D at Anthropic, Apple cuts engineers' AI spend caps — emmanuelvivier · 2026-11-09
- Nobody Was Argued Into Loving AI: One Real Problem Converts Skeptics Better Than Any Debate — Pascallisch · 2026-11-09
- By 2028, Britain's busiest pen pals may be two chatbots filing and answering complaints — evisoft · 2026-11-09
- François Fleuret: intelligence emerges from intergenerational dialogue, distillation is the poor man's version — francoisfleuret · 2026-11-09
- Anthropic employee: 'racing as fast as possible is the best way to slow down' — S_OhEigeartaigh · 2026-11-09
- SWE workday 2022 vs now: 30% reviewing AI code, 40% writing specs — itsOmSarraf_ · 2026-11-09