PSYCHOPHYSIOLOGICAL MEASURES OF THE IMPACT OF LEARNING AND CONFIDENCE ON EVIDENCE ACCUMULATION

Trust is central in human-AI teaming. To understand how trust towards automated systems emerges, confidence at individual level needs to be deepened. While the emergence of confidence can be regarded as inherent to the decision-making process using metacognition and evidence accumulation models, its variation can be regarded as a time varying-variable that can be studied through learning theories. Here, we aim to assess cognitive markers of confidence and the effect of learning on confidence judgment during decision making. We had 28 participants perform a two steps perceptual decision-making task in which uncertainty and implicit learning were varying throughout blocks. For each trial they had to identify the brighter circle in a set of three circles, followed by a rating of their confidence and a feedback. Subjective (d’, meta-d’), behavioral (accuracy, reaction times) and electroencephalographic (ERPs based on response, feedback and confidence judgement) measures were recorded during the task. Mixed linear models revealed a significant decrease of d’ and meta-d’ with increased uncertainty as well as a significant increase in confidence with implicit learning. At the neurophysiological level, we identified the Centro-Parietal Positivity (CPP), a neural correlate of evidence accumulation that predicts confidence, and error/feedback prediction potentials. Potentials were significantly modulated by accuracy and confidence levels. Building on the interpretation of the CPP, we further investigated models of evidence accumulation through a 3-choice model Linear Balistic Accumulator to unravel the dynamics of decision making. Finally, we assessed the extent to which learning acts as a mediating factor of confidence.