Akash Raj G P¹ (), Zoe Boundy-Singer¹, Corey Ziemba¹, Robbe Goris¹; ¹University of Texas at Austin

Observers are aware of the fallibility of perception. When we feel confident in a perceptual interpretation, it is more likely to be correct. However, such metacognitive judgements are not flawless. In general, performance in difficult perceptual, cognitive and behavioural tasks improves with practice. Might metacognitive abilities improve with practice as well? Previous investigations yielded inconclusive results, potentially because the confidence judgments were not difficult enough, leaving little room for improvement in metacognitive ability. To test this idea, we leveraged insights offered by CASANDRE, a process model of perceptual confidence in which confidence reflects a subject’s noisy estimate of the reliability of their perceptual decisions. The quality of this estimate is limited by the subject’s uncertainty about the variable that informs their decision (‘meta-uncertainty’). This meta-uncertainty can be manipulated experimentally by increasing the number of levels of stimulus reliability within a single experiment or by making the stimuli stochastic. We conducted a series of psychophysical experiments in which 38 subjects judged ambiguous visual stimuli and additionally reported their confidence. Each experiment consisted of 3000 trials and was completed over 2 sessions. We fit CASANDRE to each subject’s data and studied the temporal evolution of meta-uncertainty. For the majority of the subjects (27 out of 38), meta-uncertainty decreased over the course of the experiment. The median fractional change in meta-uncertainty was a decrease by 83% across all subjects (Wilcoxon signed-rank test, p < 0.001). Meta-cognitive learning appears to be a general phenomenon: it was evident in both orientation- and texture discrimination tasks. As hypothesized, this learning was most prominent in experiments that involved many levels of stimulus reliability and/or stimulus stochasticity. Together, these results demonstrate that metacognitive ability can improve with practice, provided that the confidence judgments are difficult enough.