Yiran Ge¹, Siyuan Cheng¹, Nihong Chen¹; ¹Tsinghua University, Beijing, China

People can introspect on their performance through confidence ratings. However, this process is not flawless. The degree to which individuals’ confidence ratings align with the accuracy of their own perceptual decisions reflects metacognitive efficiency. In this study, we trained subjects on a basic visual motion task. Before and after training, we asked them to report their confidence on a three-point scale following each perceptual decision. To quantify metacognitive efficiency, we computed the ratio of the slopes of psychometric functions associated with high and low confidence. Surprisingly, while five days of training (3000 trials in total) improved sensory coding and refined their decision-making process, metacognition did not keep pace with these improvements. A model (Boundy-Singer et al., 2023) assuming confidence as a noisy estimate of decision reliability indicates that subjects’ uncertainty about their decision at the matched sensory just-noticeable-difference actually increased. Error-related neural signals in the dorsolateral prefrontal cortex and anterior cingulate cortex also declined for the trained motion stimuli. Together, our findings reveal neural substrates underlying perceptual learning and highlight the dissociation between perceptual decision-making and confidence judgment in the human brain.

Acknowledgements: This work was supported by MOST 2021ZD0203600 and NSFC 31971031, 31930053.