Confidence responses in global motion discrimination task are well predicted by visual reliability

Poster Presentation 23.462: Saturday, May 18, 2024, 8:30 am – 12:30 pm, Pavilion
Session: Decision Making: Perceptual decision making 1

Can Oluk1 (), Martin Szinte2, Guillaume S. Masson2, Pascal Mamassian1; 1Laboratoire des Systèmes Perceptifs, École Normale Supérieure, Université PSL, CNRS, Paris, France, 2Institut de Neurosciences de la Timone (UMR 7289), Centre National de la Recherche Scientifique and Aix-Marseille Université, Marseille, France

Recent studies on perceptual confidence judgments have shown that human observers can be very accurate in estimating their own performance. However, these studies primarily focused on simple perceptual decisions about a single sensory feature. Here, we aim to understand the mechanisms of confidence formation in a global motion direction discrimination task where motion information must be integrated across a large number of local motions. Stimuli were composed of 468 drifting Gabors of identical size and spatial frequency, but random orientations (Amano et al., 2009, JOV). Motion coherence was manipulated as the fraction of Gabors whose velocity is consistent with a common global motion. In an 8-AFC task, participants reported the global motion direction of the multiple Gabors pattern. After two successive decisions, they judged which of them was more likely to be correct in a confidence forced-choice paradigm. We found that perceptual discrimination performance increases with motion coherence, and participants are consistently accurate in judging their own performance. We also found that when they make errors, participants often choose the direction directly opposite to true motion direction rather than random directions. Interestingly, participants are also more confident about these opposite direction reports. To model these results, we extended a classical motion integration model by incorporating strongly tuned inhibition that accounted for the opposite-direction effect. We also included a gain-like mechanism, ensuring the model's total response to stimulus (summed across all directions) increases with motion coherence. We next showed that our model can account for the observed variance in both perceptual and confidence judgments. We concluded that confidence responses are strongly tied to perceptual reliability and predominantly inherit characteristics from the associated visual mechanisms. These findings can serve as a foundation for understanding the mechanisms of confidence formation in global motion perception.

Acknowledgements: ANR grant no. ANR-19-NEUC-0003