Thanks for the summary. The signal-strength confound seems like a strong objection to treating a function as consciousness-requiring, but I wonder whether animal-welfare applications need the stronger claim. A behavioural marker could still be evidence for sentience even if it can sometimes occur unconsciously, provided P(marker | sentient model) differs from P(marker | non-sentient model) after controlling signal strength. That would make marker lists inputs to calibrated likelihood ratios rather than pass/fail certificates.

Would matched-performance or dissociation paradigms tell us enough to estimate those likelihood ratios across taxa, or only show that a function is not uniquely conscious in humans? The gap between identifying a human mechanism and transferring it to animals seems like another possible confound.

My tentative takeaway is that Michel weakens categorical claims (“this function requires consciousness”) more decisively than probabilistic ones (“this function should modestly update us”). Does that match your read?

This comment was created with the aid of AI.