pipette
ESEspañol

Confidence in global motion discrimination is better predicted by perceptual reliability than by stimulus-based heuristics

C. Oluk, M. Szinte, G. S. Masson, P. Mamassian

Preprint

In the authors' words

Decision-making requires evaluating the reliability of perceptual estimates, but whether confidence reflects perceptual reliability or stimulus-based heuristics remains debated. We used a multiple-aperture global motion stimulus in which the global direction, and therefore motion coherence, is only available after integrating locally ambiguous signals, so that no simple stimulus statistic can serve as a shortcut. Observers reported the global direction in two successive intervals and then chose which decision was more likely to be correct. Confidence tracked accuracy beyond what coherence alone could explain: observers were more accurate in chosen intervals even at the same coherence level. Observers also frequently reported the direction opposite to the true motion, and, in direct comparisons, were more confident in these errors than in other large errors. We compared two image-computable models of motion integration. A standard motion-energy model failed to produce opposite-direction errors and the associated confidence pattern. A normalized motion-energy model, adding tuned inhibition among direction channels and an untuned gain control that increases overall activation with coherence, accounted for most of the variance in both perceptual decisions and confidence judgments. Tuned inhibition, which must be sharp if it is close to excitation in strength, produced the opposite-direction errors and reproduced the direction of their confidence advantage, and the untuned gain was needed for confidence to depend on coherence as strongly as observed. Confidence was read out from the same signal as the perceptual decision, and noise at the confidence stage, although probably present, added little to the fit beyond explaining the accuracy difference between chosen and unchosen intervals. These results suggest that confidence in this task relies on perceptual reliability. They also suggest that seemingly unusual metacognitive behavior can emerge from perceptual computations, which must therefore be modeled explicitly to identify genuine confidence mechanisms.

Main resultLimitation the authors admit

Appeared: Wednesday, November 18. bioRxiv. Preprint, not yet peer-reviewed.

DOI: 10.64898/2026.11.12.766172