Increased monetary incentives enhance goal-directed reinforcement learning but fail to suppress automatic learning processes
Abstract
Adaptive reinforcement learning requires assigning outcomes to the features of actions that causally determine them. Yet humans also learn reward associations with action features that are explicitly known to be outcome-irrelevant, and these associations can bias subsequent choices. Whether such maladaptive learning is regulated by motivational incentives remains unclear. Here, we tested whether increasing monetary stakes differentially modulates the influence of outcome-relevant and outcome-irrelevant learned values on choice. Participants (N = 287) completed a multi-armed bandit task in low- and high-stakes conditions while explicitly instructed that card identity predicted reward whereas card location did not. Computational modelling revealed that higher stakes increased the influence of outcome-relevant values on choice, but did not change the influence of outcome-irrelevant values. Critically, outcome-irrelevant value influences persisted even though they produced substantially greater monetary losses under high stakes. These findings show that monetary incentives selectively enhance the use of outcome-relevant learned values without improving the selectivity of credit assignment. More broadly, they suggest that only some components of reinforcement learning are sensitive to motivational control.
Related
- Effects of proactive control and reward incentives on the inhibition of sequentially linked motor actions.
- Motivational incentives improve working memory performance by enhancing memory-based representations rather than altering response strategies
- Beyond the prohibition list: redirecting negative subsidies into positive incentives could mitigate overfishing and protect coastal communities in WTO fisheries reform
- Incentives are effective but insufficient to halt smallholder deforestation in the Colombian Amazon
- BCFL-DAI-PUB: Blockchain-integrated federated learning with dynamic adaptive incentives for privacy-utility balanced credit risk modelling
BibTeX
@misc{benartzi2026increased,
title = {{Increased monetary incentives enhance goal-directed reinforcement learning but fail to suppress automatic learning processes}},
author = {Ido Ben-Artzi and Noham Wolpe and Nitzan Shahar},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/zyc8t_v2},
url = {https://osf.io/preprints/psyarxiv/zyc8t_v2/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Increased monetary incentives enhance goal-directed reinforcement learning but fail to suppress automatic learning processes
AU - Ido Ben-Artzi
AU - Noham Wolpe
AU - Nitzan Shahar
PY - 2026
DA - 2026/09/16
PB - PsyArXiv
DO - 10.31234/osf.io/zyc8t_v2
UR - https://osf.io/preprints/psyarxiv/zyc8t_v2/
AB - Adaptive reinforcement learning requires assigning outcomes to the features of actions that causally determine them. Yet humans also learn reward associations with action features that are explicitly known to be outcome-irrelevant, and these associations can bias subsequent choices. Whether such maladaptive learning is regulated by motivational incentives remains unclear. Here, we tested whether increasing monetary stakes differentially modulates the influence of outcome-relevant and outcome-irrelevant learned values on choice. Participants (N = 287) completed a multi-armed bandit task in low- and high-stakes conditions while explicitly instructed that card identity predicted reward whereas card location did not. Computational modelling revealed that higher stakes increased the influence of outcome-relevant values on choice, but did not change the influence of outcome-irrelevant values. Critically, outcome-irrelevant value influences persisted even though they produced substantially greater monetary losses under high stakes. These findings show that monetary incentives selectively enhance the use of outcome-relevant learned values without improving the selectivity of credit assignment. More broadly, they suggest that only some components of reinforcement learning are sensitive to motivational control.
ER -