Over-reliance on AI in information tasks: Unraveling the roles of reasoning visibility and task complexity

Abstract

Recent large language models expose their reasoning processes to users, an interface design intended to improve explainability and support calibrated trust. Yet whether visible reasoning calibrates reliance or amplifies over-reliance remains unclear. This study examined how reasoning visibility and task complexity influence users' over-reliance on AI in information processing tasks, along with the underlying psychological mechanisms and boundary conditions. Three online experiments (total N = 657), spanning the health, environment, and technology domains, showed that both visible reasoning and task complexity increased over-reliance, with no interaction between them. Mediation analyses indicated that both factors increased over-reliance through perceived AI competence, while task complexity additionally did so by reducing perceived unaided self-efficacy. AI literacy, particularly awareness literacy, buffered the effect of reasoning visibility but not that of task complexity. These findings demonstrate that interface features designed to enhance transparency can backfire when their mere presence is interpreted as a cue of AI competence, amplifying over-reliance above an existing baseline. They carry implications for understanding over-reliance in human-AI collaborative information behavior and for designing responsible AI interfaces.