normative
Untrained first-person AI would be more convincing evidence
The most promising criterion for AI consciousness is that an AI trained on data with all first-person point of view removed (e.g. only the Encyclopedia Britannica) that nonetheless spontaneously developed a first-person point of view would be far more convincing evidence of experience than current LLMs, which are trained on vast human-generated first-person material.
normativepending
Speaker
Ned BlockEvidence Quote
“if you successfully eliminated everything in the training data that had a first person point of view and it nonetheless developed one, you know that would have... Some convincing force”
Created: 6/13/2026, 12:30:17 AM
Your notes
Loading notes...