The relationship between AI dependency and critical thinking disposition in nursing students: The mediating role of metacognitive ability and the moderating role of academic self-efficacy.
BACKGROUND Artificial intelligence tools are increasingly integrated into nursing education. They bring great potential to improve learning efficiency and access to information. However, whether over-reliance on artificial intelligence affects the development of critical thinking remains to be verified. Critical thinking is a core competency for clinical decision-making and safe nursing practice. Although existing evidence suggests a negative association between AI dependency and higher-order cognitive abilities, the underlying mechanisms remain unclear. In particular, the roles of internal cognitive processes and individual psychological characteristics are not well understood. Existing research has largely focused on external factors related to technology adoption. Less attention has been paid to how AI dependency affects metacognitive functioning or the moderating role of academic self-efficacy. There is a particular lack of in-depth research on nursing students, a group that faces multiple cognitive demands from theoretical study, clinical practice, and professional preparation. OBJECTIVE This study aimed to explore the relationship between artificial intelligence dependency and critical thinking disposition among nursing students. It examined the mediating role of metacognitive ability and the moderating role of academic self-efficacy. The goal was to reveal the underlying mechanism of this relationship. METHODS A questionnaire survey was conducted among nursing students using the Artificial Intelligence Dependence Questionnaire (assessing independent variable: AI dependency), the Metacognitive Ability Scale (assessing mediator: metacognitive ability), the Critical Thinking Disposition Scale (assessing dependent variable: critical thinking disposition), and the Academic Self-Efficacy Scale (assessing moderator: academic self-efficacy). Moderated mediation analysis was performed using Model 5 in the SPSS macro program PROCESS. Structural equation modeling and simple slope analysis were used to examine the path relationships and moderating effects among the variables. RESULTS The direct predictive effect of artificial intelligence dependency on critical thinking disposition was not significant (β = 0.021, P = 0.325). However, the indirect effect through metacognitive ability was significant. This indicated that metacognitive ability played an indirect-only mediating role in this sample between the two. Academic self-efficacy played a negative moderating role in the path from artificial intelligence dependency to metacognitive ability (β = -0.008, P < 0.01). It also played a negative moderating role in the path from metacognitive ability to critical thinking disposition (β = -0.003, P < 0.05). Simple slope analysis showed that for students with low academic self-efficacy, the positive predictive effect of artificial intelligence dependency on metacognitive ability was stronger (β = 0.218, P < 0.01). Similarly, the positive predictive effect of metacognitive ability on critical thinking disposition was also stronger for this group. For students with high academic self-efficacy, the positive effects in both paths were weakened. CONCLUSION Artificial intelligence dependency was indirectly associated with critical thinking disposition in nursing students through the indirect-only mediating role of metacognitive ability. Academic self-efficacy played a negative moderating role in both the first and second halves of the model. This model revealed a differentiated mechanism through which artificial intelligence dependency affects critical thinking. For students with low academic self-efficacy, AI dependency may serve as a form of cognitive compensation by activating metacognition. For students with high academic self-efficacy, the positive effects of AI dependency were relatively limited. Rather than simply encouraging or restricting AI use, nursing educators should adopt differentiated guidance strategies based on students' self-efficacy levels-supporting metacognitive engagement for low-efficacy students while fostering critical evaluation of AI outputs for high-efficacy students. The goal is to promote reflective AI use that strengthens metacognitive skills, rather than passive dependence.