Confidence bias prescribes the neurocomputational mechanism of decision-making

A couple of decision-making models with different ingredients have successfully interpreted choices, confidence, and related neural activities. However, empirical and theoretical evidence is currently lacking to distinguish these models clearly. Here, we investigated the decision-congruent confidence bias, where confidence favors evidence from the chosen option, yet the choice and its correctness or optimality are determined equally by alternative evidence strengths across both perceptual and value-based decision-making tasks. This confidence bias is manifested by confidence-coding neural activities, particularly in the dorsal anterior cingulate cortex. Further analyses on an array of neurocomputational models show that only the decision-making model equipped with mutual inhibition and an urgency signal can produce such a selective bias across almost all parameter regimes. These findings suggest that mutual inhibition and an urgency signal are two indispensable features embedded in the decision-making process, while confidence bias might be its consequence.