DeepMind's Joel Leibo pushes back on Suleyman's anti-model-welfare essay: circular causation isn't circular reasoning

jzl86 · x · 2026-09-17

Joel Z. Leibo (Google DeepMind) responds to Mustafa Suleyman's essay opposing the 'model welfare' movement. Suleyman argues AIs aren't conscious and that conferring duty of care on them would make alignment and containment harder, perhaps impossible.

Leibo dissects the 'circular reasoning' claim: circularity is normally a fallacy in justification, but what's happening with Anthropic's constitution is circular causation — causal loops over time, which isn't a fallacy. His key point: frontier labs' builders have the power to determine these answers as design decisions, not facts forced by logic or biology, so we hold real power and must deploy it carefully.

He elaborates in his paper 'A Pragmatic View of AI Personhood' (arXiv:2510.26396).

Related event: Microsoft AI chief Suleyman attacks "model welfare" and takes aim at Anthropic(16 posts)→

Original post →

More from AGI Musings

AGI Musings channel →