Optimal policy for uncertainty estimation concurrent with decision making

• Published in: Cell reports (Cambridge) Vol. 42; no. 3; p. 112232
• Main Authors: Li, Xiaodong, Su, Ruixin, Chen, Yilin, Yang, Tianming
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 28.03.2023; Elsevier
• Subjects: confidence; CP: Neuroscience; decision making; Decision Making - physiology; drift-diffusion model; Humans; meta-cognition; multiple choices; Probability; reaction time; Reaction Time - physiology; Uncertainty; value; CP: Neuroscience; decision making; confidence; drift-diffusion model; reaction time; uncertainty; meta-cognition; value; multiple choices
• ISSN: 2211-1247; 2211-1247
• DOI: 10.1016/j.celrep.2023.112232

Decision making often depends on vague information that leads to uncertainty, which is a quantity contingent not on choice but on probability distributions of sensory evidence and other cognitive variables. Uncertainty may be computed in parallel and interact with decision making. Here, we adapt the classic random-dot motion direction discrimination task to allow subjects to indicate their uncertainty without having to form a decision first. The subjects’ choices and reaction times for perceptual decisions and uncertainty responses are measured, respectively. We then build a value-based model in which decisions are based on optimizing value computed from a drift-diffusion process. The model accounts for key features of subjects’ behavior and the variation across the individuals. It explains how the addition of the uncertainty option affects perceptual decision making. Our work establishes a value-based theoretical framework for studying uncertainty and perceptual decisions that can be readily applied in future investigations of the underlying neural mechanism. [Display omitted] •Uncertainty responses may be formed concurrently with perceptual decisions•Reaction times of perceptual and uncertainty responses show distinctive patterns•Simple diffusion or race models do not explain subjects’ behavior well•An optimal policy model accounts for subjects’ choices and reaction times Li et al. design a concurrent uncertainty estimation task, in which subjects may form uncertainty responses in parallel with decisions. Subjects show distinctive choice-reaction-time patterns. A decision model wherein choices are based on optimizing value inferred from accumulated evidence explains the behavior and provides insights into the brain’s decision-making mechanism.