Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session B: Wednesday, August 13, 1:00 – 4:00 pm, de Brug & E‑Hall

Experimental Assessment of Choice-Congruent Confidence Bias in Human Reinforcement Learning

Alexandre Lietard¹, Kobe Desender¹; ¹KU Leuven

Presenter: Alexandre Lietard

Confidence constitutes a fundamental signal which aids adaptative processes in both learning and decision-making. Yet, it often deviates from optimality. In the domain of perceptual decision-making, one such deviation, known as the choice congruent bias, reflects the tendency to overweight evidence that supports the chosen option. As a result, confidence tends to increase with the amount of overall evidence, even when accuracy remains unchanged. Here, we examined whether the same biased computation occurs in value-based decisions, by testing whether greater overall evidence leads to higher confidence in a reinforcement learning task. The results demonstrate that increasing the average reward (considered as evidence) successfully elevates confidence levels. Our computational modelling indicates, however, that this effect does not necessarily reflect a biased confidence computation. Therefore, while manipulating average reward may be a useful method for dissociating confidence from accuracy, it cannot serve as a direct test of the choice-congruent bias.

Topic Area: Reward, Value & Social Decision Making

Extended Abstract: Full Text PDF