A toot-summary of our recent article!
We often use our confidence to gauge the reliability of our perception.
But what about the confidence... in our confidence?
Sometimes, we can be certain we are uncertain.
With Samuel Recht, Ljubica Jovanovic, and Pascal Mamassian, we had fun testing the limits of meta-metacognition in a classic visual task.
We found surprising accuracy of confidence up to the fourth order (confidence in confidence in confidence, or meta-meta-meta-cognition).
Confidence was better than predicted if we assumed all the variability disrupting metacognition would be passed on to the next re-evaluation: relative performance could improve from one re-evaluation to the next.
This is not just a party trick. We can use nested cognition to better understand metacognition. For example, we show confidence in this context must rely on a fine-grained representation of the perceptual decision-evidence, compared to discrete bins.
More details in the manuscript: https://academic.oup.com/nc/article/2022/1/niac014/6761526#.Y2Fl1I8D6c0.twitter