Trusting AI too much? Understanding judgment attenuation in Human–AI decision support systems
Artificial intelligence (AI)-based decision support systems are increasingly shaping organizational decision-making by influencing how humans engage in cognitive tasks. While prior research has largely emphasized the performance benefits of AI adoption, less attention has been given to its association with human judgment. Drawing on Automation Bias theory and Human–AI collaboration research, this study examines the relationships among trust in AI-based decision support systems (AI-DSS), reliance intention, judgment attenuation—defined here as a perceived reduction in independent evaluative effort—and accountability pressure. Survey data from 400 organizational employees were analyzed using covariance-based structural equation modeling. The results show that trust in AI-DSS is positively associated with users’ reliance intention, which in turn is associated with greater self-reported judgment attenuation. Trust also shows a direct association with judgment attenuation, indicating that AI use co-occurs with both behavioral and cognitive correlates of reduced independent evaluation. Furthermore, the positive association between trust and judgment attenuation is weaker under conditions of high accountability pressure, a pattern consistent with, though not a direct test of, greater cognitive engagement. Because judgment attenuation is measured by self-report at a single time point without a performance criterion, these findings should be interpreted as evidence of perceived rather than behaviorally verified erosion of judgment, and the cross-sectional design precludes strong causal claims. This study extends Automation Bias theory to contemporary Human–AI collaboration by showing that trust in AI is associated with both the facilitation and the perceived constraint of organizational decision-making, and it highlights accountability as a candidate governance mechanism for preserving self-reported human judgment in AI-assisted environments, pending behavioral validation.