Abstract
<p>People often rely on human or AI advice, but whether higher stakes improve adviceintegration remains unclear. Across two preregistered experiments (N = 99), participantscombined perceptual judgments with advice from artificial agents while stakes varied up totwenty-fold. A Bayesian model provided benchmarks for optimal advice weighting. Higherstakes prolonged deliberation but did not, on average, improve advice weighting or decisionaccuracy. Instead, their effects were uneven and varied with cognitive ability. Participantswith higher ability became more optimal and accurate under high stakes, whereas thosewith lower ability performed worse. Thus, incentives can mobilize effort without uniformlyimproving performance and may widen differences between decision-makers. These findingsshow that insufficient motivation alone cannot explain suboptimal advice integrationand identify cognitive capacity as a boundary condition on the benefits of higher stakes inassisted decision-making.</p>