yunxuan zheng¹, Medha Shekhar¹, Kai Xue¹, Dobromir Rahnev¹; ¹Georgia Tech

Perceptual decisions are subject to sensory, decisional, and metacognitive noise. However, dissociating between these different types of noise has proven challenging within conventional paradigms where the different types of noise can mimic each other. Here, we isolated each of these types of noise using an external noise paradigm where the same stimulus value could be generated from two categories. Subjects judged whether the number of dots presented on the screen was generated from a distribution with a higher or lower mean. These judgments are corrupted with a combination of sensory and decisional noise. In addition, subjects rated their decision confidence, with these judgments being corrupted by a combination of sensory and metacognitive noise. Subjects also performed a separate 2-alternative forced choice (2AFC) task where they identified which of two squares had more dots. 2AFC tasks involve minimal decisional noise and therefore allow one to estimate an upper bound of the sensory noise induced by our stimuli for each subject. Having estimated the level of sensory noise, we built a computational model that calculates the level of decisional and metacognitive noise in the external noise task. We found evidence for substantial decisional and metacognitive noise that, in some cases, exceeded the sensory noise. Moreover, metacognitive noise was higher than decisional noise, suggesting that confidence ratings may largely inherit the decisional noise of the initial perceptual judgment. Importantly, metacognitive noise became larger for confidence criteria further away from the decision criterion. This finding supports the recent lognormal meta-noise model of metacognition, which postulates that metacognitive noise is signal-dependent, such that it increases for more extreme evidence values (Shekhar & Rahnev, 2021). Overall, our study successfully dissociated sensory, decisional, and metacognitive noise, enabling quantification of each factor’s impact on the corruption of perceptual decisions.

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