Can we use confidence ratings to assess conscious perception? I address this question in a new paper, forthcoming in WIREs Cognitive Science (philpapers.org/rec/MICCIC-3). My answer is "Yes", but it’s not straightforward! A thread.

(2) You may have seen studies assessing whether participants perceive stimuli consciously or unconsciously by asking them to provide confidence ratings. That might seem weird at first. Consciousness and confidence aren’t the same thing.

(3) We can nevertheless (validly) interpret metacognitive indicators as indicating consciousness, if we make some assumptions. I start by providing a full, non-technical introduction to indicators like meta-d' and the M-ratio.

(4) I then lay out the conditions under which variations in M-ratio can be interpreted as indicating differences in consciousness. The most important is what I call "Consciousness-Selectivity": unconscious sensory evidence isn’t available for optimal metacognition.

(5) I (hesitantly) provide a philosophical argument for this condition: If you accept that consciousness is relevant for perceptual justification, then confidence might be a good way to assess consciousness (since it presumably tracks justification).

(6) But there's also converging empirical evidence indicating that metacognition is "consciousness-selective", including evidence from blindsight patients who seem unable to correctly assess their own perceptual decisions. I conclude that these procedures are valid.

(7) This doesn’t mean that confidence-based procedures are accurate. Here, I review the relevant literature comparing results from different scales. Conclusion: assessing consciousness with confidence is, most of the time, accurate enough for our epistemic and practical purposes.

(8) Furthermore, confidence-based procedures are probably better than alternative ways of assessing consciousness, such as the perceptual awareness scale. The reason is that the former is more likely to assess consciousness of task-relevant features.

(9) You can learn more about this last point here too: philpapers.org/rec/MICHNT.

(10) The last section of the paper answers eleven (yes, eleven) objections! Some of these are missing the point. But other objections do succeed in showing that confidence-based procedures aren’t entirely accurate. For instance, M-ratio isn’t entirely independent of bias.

(11) This is not a good reason to reject these procedures entirely, just a good reason to improve them! The field of perceptual metacognition is making a lot of progress, leading to better models, and better indicators. These procedures will only get better from there.

(12) Hopefully, improvements in models and indicators of metacognition can pave the way for better measures of conscious perception as well. I hope you enjoy the paper and find it useful! /END

I thank @smfleming, @meganakpeters, @hakwanlau, @jorgemlg, @attninaction, and many others for discussing these issues with me!