Confidence: Its Measurement, Relation to Behavior, and Neural Representation

Confidence (certainty/uncertainty about one's own decision) has attracted considerable attention in psychology, neuroscience, and artificial intelligence. Yet its relation to behavior remains unclear: Some treat confidence as a cause of behavior whereas others view it as a state controlled by environmental variables. The unobservable nature of confidence makes its measurement validation and analyses difficult. To tackle these issues, we developed a novel procedure combining a decision-making task, risk choices, and a post-task open-ended questionnaire. Point-gain/loss amounts for correct/incorrect decisions varied depending on choices between high-risk/high-return and low-risk/low-return. In the questionnaire, 90% of 16 participants reported feelings interpretable as certainty/uncertainty when making those risk choices despite no instruction about confidence, whereas 8% did so when the choices led to identical outcomes/returns. This suggests confidence is a state controlled by environmental contingencies rather than a cause of behavior. We also decoded changes in the state from electroencephalography and functional magnetic resonance imaging.