Why an AI saying "I'm conscious" tells us nothing — and what to measure instead

ResponsibleBuy7451 · reddit · 2026-09-05

Arguing that sentences like "I'm conscious" are circular evidence (producing them is exactly what LLMs do), the author proposes measurable signals instead: consistency across context shifts, calibrated uncertainty, internal states matching consciousness theories, and persistent memory/goals/self-models — none of which prove consciousness either. The open question: with full access to a future model's weights, activations and architecture, what would you actually look for?

Original post →

More from AGI Musings

AGI Musings channel →