Dependencia cognitiva de la inteligencia artificial en estudiantes de ciencias de la salud: una revisión de alcance

Introduction. The rapid incorporation of generative artificial intelligence (GenAI) into university education has shifted the debate from academic integrity toward a deeper question: what happens to the cognitive processes of learners who habitually delegate reasoning tasks to conversational systems. In the health sciences, where professional competence is built upon clinical reasoning and judgement under uncertainty, this question carries clinical and not merely pedagogical relevance. Objective. To map and critically synthesise the available evidence on cognitive dependence on artificial intelligence among health sciences students, identifying its prevalence, the proposed explanatory mechanisms, its cognitive and affective correlates, and the moderating factors that condition its effects. Methods. Scoping review conducted according to PRISMA-ScR guidance. PubMed/MEDLINE, Scopus, Web of Science, SciELO, Redalyc and ERIC were searched, together with the academic meta-search engines Consensus, Elicit and SciSpace, covering literature published between January 2022 and October 2026 in Spanish and English. Empirical studies, systematic and scoping reviews, and conceptual works on university students’ use of GenAI were included, with preferential analysis of those conducted in health sciences populations. Extraction was organised into eight thematic domains through narrative synthesis. Results. GenAI adoption among health sciences students is virtually universal, with use prevalences between 83.9 % and 96.3 %, and self-reported dependence exceeding 60 % in the largest series. Quantitative evidence converges on a small-to-moderate inverse association between AI dependence and critical thinking. The proposed mechanisms include substitutive cognitive offloading, cognitive debt, metacognitive laziness, cognitive fatigue and automation bias, and all of them share a common core: the transfer of epistemic agency from the learner to the system. In the clinical domain, three stage-specific risks are described: never-skilling, deskilling and mis-skilling. Dependence is further associated with anxiety and depressive symptoms, and with high uncritical acceptance of fabricated content and references. The effect is not deterministic: mode of use, AI literacy, metacognitive scaffolding and assessment design consistently moderate the direction of the outcome. Conclusions. Cognitive dependence on GenAI is a plausible and measurable educational risk, but a conditional one. The evidence supports neither prohibition nor uncritical adoption, but rather curricular and assessment redesign aimed at preserving students’ epistemic agency. Longitudinal studies with objective measures of unassisted competence are needed, as is an Ibero-American body of research that remains scarce.