Abstract
"Confidence" commonly refers to the judgment of one's performance in general knowledge (memory), motor performance (action), or sensory faculty (perception). However, in most laboratory studies of perceptual confidence, observers make confidence judgments on individual trials or responses, which does not involve the judgment of overall task performance. Here, we are interested in general confidence, the way to measure it, and its characteristics in comparison to single-trial confidence. Observers (N=25) were presented with two Gabor patches simultaneously, and were asked to discriminate their orientations (which Gabor was more tilted) or spatial frequencies (which Gabor had higher frequency). For each observer, the two tasks were randomly assigned as the "global" and the "local" task, respectively. Observers completed the global task in blocks of 40 trials, with each block delivered at a different, calibrated difficulty level. After each block, observers performed 80 trials of the local task. Each local-task trial was immediately followed by a confidence-comparison task, in which observers indicated whether they were more confident 1) in their overall responses across the 40 trials in the global-task block, or 2) in the response they had just given in the current local-task trial. We found that, across all global-task difficulty levels, the proportion of "more-confident-in-local-task" responses monotonically increased with the performance in the local task. Critically, observers adjusted their criterion for the confidence comparison according to the global task difficulty. Finally, we found a reliable confidence bias in favor of the local task. These results were present regardless of the global task, be it over orientation or spatial-frequency. Our results suggest that humans can build a global confidence for overall task performance within a relatively short time, and use it for comparison with the traditional, single-trial confidence. Furthermore, as shown using our "local-vs-global" paradigm, humans tend to be overconfident in the local task.
Meeting abstract presented at VSS 2014