Marshall L. Green¹, Dobromir Rahnev¹; ¹Georgia Institute of Technology

A central question in visual metacognition is whether confidence is computed strictly after or together with the perceptual decision. On one hand, several prominent models such as 2DSD (Pleskac & Busemeyer, 2010) postulate that confidence is based exclusively on information accrued after the perceptual decision (i.e., confidence is a post-decision process). On the other hand, many other models posit that confidence is computed simultaneously with the perceptual decision (i.e., confidence is an online process). Here we aim to adjudicate between these two theories. We developed a novel task that required continuous report of both choice and confidence. Specifically, participants (N = 25) estimated the global motion direction of dot motion stimuli by centering a confidence interval on the perceived motion direction and provided confidence by adjusting the width of the confidence interval. We recorded angular error and confidence every 16 ms. We found that for the first 500 ms, participants exhibited chance performance and extremely low confidence. This initial period was followed by a rapid increase in both performance and confidence, with both reaching a plateau around 1.5-2 seconds after stimulus onset. Critically, the confidence-accuracy correlation was near zero during the first 500 ms but then rapidly increased at the same point in time that performance and confidence changed, suggesting that confidence did not exhibit any post-decision delay. This pattern was the same regardless of motion coherence, except that stimuli with higher coherence led to steeper increases in performance, confidence, and the confidence-accuracy correlation. These time courses were fit well by a circular drift-diffusion model (cDDM) where decision and confidence are simultaneously made based on a noisy 2D accumulator. Altogether, our results strongly suggest that confidence is an online process that is computed simultaneously with the perceptual decision and challenge models where confidence is a strictly post-decision process.

Acknowledgements: This work was supported by the National Institute of Health (award: R01MH119189) and the Office of Naval Research (award: N00014-20-1-2622).