Researchers debate human-indistinguishable AI avatars, propose 'Anthroporealism'

dioscuri · x · 2026-10-04

Anil Seth argues there are no upsides to creating AI avatars indistinguishable from real humans — only major downsides on top of already dubious anthropomimetic AI — so anthropomimesis should be minimized.

Dioscuri pushes back with a finer distinction: perceptual vs practical indistinguishability. Agents that look, sound and move like humans could have valuable use cases, just as photorealistic graphics do, provided watermarking and disclosure requirements let people reliably tell what they're interacting with.

He also proposes coining Anthroporealism (by analogy with photorealism) for the concept, while admitting "human-indistinguishable" may be clearer if blander. The exchange crystallizes the core ethics question: technical feasibility of human-like agents doesn't settle whether we should build them — reliable identifiability is the baseline.

Related event: Anil Seth Argues Against AI Avatars Indistinguishable From Real Humans(4 posts)→

Original post →

More from AGI Musings

AGI Musings channel →