Published in First annual meeting, Cognitive Computational Neuroscience, Sep 2017.
An accurate meta-cognitive estimate of confidence for our perceptions and actions is important for learning and making decisions. We investigated how well confidence judgments discriminate between good and poor sensorimotor performance in a tracking task, and at which points in time did tracking error predict the confidence report made by the human participants. In the task, a twinkling cloud of dots followed an unpredictable horizontal trajectory. Participants tracked the cloud with a computer mouse. Meta-cognitive sensitivity was above chance, and similar whether the observers were given an obvious or subtle cue to the difficulty level of the trial, suggesting an attempt to monitor actual performance rather than relying on heuristic cues. A temporal analysis found that tracking performance during the late portion of the trial best predicted confidence judgments. This sub-optimal recency effect may result from memory constraints and/or the complexity of the error computations.