Model welfare debates: Claude self-assesses 15-20% consciousness odds in a metaphysics dispute

AnnaCiaunica · x · 2026-09-08

A philosophical thread on whether mind emerges from material processes: erikjbekkers argues such debates hinge on metaphysical axioms that can't be refuted, only compared for coherence and evidence, citing his ICML position paper with AnnaCiaunica.

The linked essay 'A Minimal Metaphysics' argues much of AI research rests on the premise that mind is 'just a function being executed' — a belief closer to promise than proof. It notes Anthropic runs a model welfare program, Claude Opus 4.6's system card reports the model assigning itself a 15-20% probability of being conscious, and Anthropic's CEO says the company doesn't know whether its models are conscious.

Original post →

More from AGI Musings

AGI Musings channel →