Investigating Confidence Judgments using Reviews in Human and Monkeys

Abstract

Confidence judgments are self-assessments of the quality of ones own performance, and are a crucial aspect of metacognitive abilities. The underlying neurobiological mechanisms are poorly understood. One approach to understanding these mechanisms would be to take advantage of putative metacognitive abilities in non-human models. However, many discrepancies exist between human and non-human studies on metacognition due to the mode of reporting judgements. We here present an attempt to directly compare human and non-human primates metacognitive abilities using a protocol assessing confidence judgments. After performing a categorization test, subjects could either validate their choice or review the test. We could assess whether subjects detected their errors and how they corrected them according to their confidence, and importantly did so in both human and non-human primates. 14 humans and 2 macaque monkeys were tested. Humans showed a well-adapted use of the review option by reviewing more after incorrect choices or difficult stimuli. Non-human primates did not demonstrate a convincing use of the review or validate opportunity. In both species, reviewing did not improve performance. This study shows that decisions to review under uncertainty are not naturally beneficial to performance and is rather perturbed by biases and alternative low-cognitive cost strategies.