Summary: People are more confident and decide faster when pursuing rewards, but they become more adaptable and less overconfident when trying to avoid losses.
Source: University of Geneva
Does the prospect of winning or losing money change how confident we are in our decisions, and does it alter how quickly we learn? Researchers at the University of Geneva, in collaboration with teams from the University of Amsterdam and ENS Paris, explored this question using a reinforcement learning task that offered monetary rewards or punishments. They found that learning to seek gains increases decision confidence—often to the point of overconfidence—while learning to avoid losses keeps people more cautious, accurate, and flexible. These findings were published in PLOS Computational Biology.
Accurate self-evaluation depends on confidence: how certain we feel about the choices we make. The research team asked whether economic framing—gains versus losses—affects both learning performance and confidence judgments during reinforcement learning tasks.
To investigate, 84 participants completed experiments in which they repeatedly chose between two abstract symbols. In the gain condition, one symbol had a 75% chance of earning 50 cents and the other a 25% chance; participants selected a symbol on each trial and then rated their confidence in that choice. In the loss condition, the task was inverted: participants chose the symbol with the lowest chance of losing money and then reported confidence. Over repeated trials, participants learned which symbol yielded better outcomes.
Confidence rises when rewards are at stake
Performance—measured by the ability to identify the better symbol—was statistically equivalent whether participants learned to gain or to avoid losing. However, confidence levels differed markedly: participants reported roughly 10% higher confidence when learning to obtain rewards than when learning to avoid losses. Because task difficulty and actual performance were the same in both contexts, this gap indicates a contextual bias in confidence judgments driven by the economic framing.
This reward-induced confidence boost is not necessarily beneficial. While confidence normally grows as people learn and make more optimal choices, the gain context amplified that effect, producing overconfidence—an overestimation of one’s actual performance of about 10%. In contrast, the loss context produced lower confidence and more cautious self-assessment, which resulted in more accurate evaluations. Researchers note that excessive doubt in negative contexts can become anxiety, potentially eroding confidence too far, but moderate caution appears to guard against overestimation.
Loss aversion promotes flexibility; reward framing reduces it
To test adaptability, half of the participants faced a reversal: the previously superior symbol became inferior. Those who had been learning in the gain condition were slower to detect and adapt to this change, persisting in their prior choice pattern. Participants in the loss condition detected the reversal more quickly and adjusted their choices. The authors suggest an evolutionary explanation: potential danger (loss) demands rapid reassessment and behavioral flexibility, whereas favorable conditions encourage maintaining a successful strategy. Practically, this implies that reward-seeking can induce inflexibility—people assume what worked before will keep working—while fear of loss preserves adaptability.
These results carry clear implications for education, training, and behavioral interventions: the learning context matters. If the goal is speed and higher subjective confidence, reward-based incentives can accelerate decisions and boost self-assurance. If the goal is flexible adaptation and accurate self-evaluation, framing tasks to emphasize avoiding losses can promote caution and responsiveness. In many real-world situations, striking a balance between encouraging confidence and preserving adaptability will be crucial.
The researchers additionally observed that choices were made faster in the reward context than in the loss context, reinforcing the link between positive framing and speed, but also highlighting the trade-off between rapid, confident choices and accurate, flexible decision-making.
Next, the team will investigate the neural mechanisms that underpin these differences: which brain regions respond to gains versus losses, which regions encode confidence, and how these networks interact to shape judgments and decisions during reinforcement learning.
Source:
University of Geneva
Media Contact:
Maël Lebreton – University of Geneva
Image Source:
The image is in the public domain.
Original Research: Open access. “Contextual influence on confidence judgments in human reinforcement learning.” Maël Lebreton, Karin Bacily, Stefano Palminteri, Jan B. Engelmann. PLOS Computational Biology. DOI: 10.1371/journal.pcbi.1006973
Abstract
Contextual influence on confidence judgments in human reinforcement learning
Estimating the probability that one’s choices are correct is essential for adapting decisions and selecting strategies. However, confidence judgments are prone to biases. This study examines how outcome valence (gains or losses) affects confidence while participants learned stimulus-outcome associations via trial-and-error. Across two experiments, participants were more confident when learning to seek gains than when learning to avoid losses, despite equal task difficulty and comparable performance. Computational modeling indicates this bias arises from a dynamically updated context-value—the average expected value of available choices—which influences confidence independently of objective accuracy. The authors show that such bias has functional consequences in volatile environments: framing outcomes as gains or losses alters learning flexibility, revealing substantial asymmetries between learning to avoid losses and learning to pursue rewards.