Not yet human-reviewed
This source record was written automatically by the research pipeline and has not yet been reviewed by the maintainer or integrated into a topic page. Treat it as a provisional summary and verify claims against the original source.
AI-induced Deskilling in Medicine: A Mixed-Method Review and Research Agenda for Healthcare and Beyond
Citation. Natali, C., Marconi, L., Dias Duran, L. D., & Cabitza, F. (2025). AI-induced deskilling in medicine: A mixed-method review and research agenda for healthcare and beyond. Artificial Intelligence Review, 58, 356. https://doi.org/10.1007/s10462-025-11352-1
Summary. This mixed-method review examines AI-induced deskilling and upskilling inhibition in medicine, combining a PRISMA-guided systematic review with a narrative synthesis. It identifies risks to clinical examination, differential diagnosis, clinical judgment, communication, autonomy, and training opportunities while distinguishing skill loss from changes in the distribution of expertise. The authors argue for longitudinal and in-situ research, skill-retention measures, and AI designs and governance arrangements that keep clinicians meaningfully engaged.
Key claims
- The systematic review identified 22 eligible sources and organized 17 deskilling concerns against PACES clinical-skill domains, with recurring vulnerabilities in physical examination, differential diagnosis, clinical judgment, and physician–patient communication.
- The review distinguishes deskilling, the degradation of previously acquired competence through reduced practice, from upskilling inhibition, the loss of opportunities to acquire competence when AI bypasses challenging supervised work, particularly for trainees.
- The reviewed literature links clinician over-reliance on AI to automation bias, reduced diagnostic reasoning and critical thinking, and a shift from active decision-making toward supervisory work, but also recognizes that well-designed hybrid intelligence and decision aids can support learning and clinical capability.
- The authors propose monitoring clinician–AI interaction and outcomes longitudinally, extending evaluation beyond accuracy and efficiency to skill retention, cognitive engagement, autonomy, and failure analysis.
Evidence & limitations
- The systematic component searched PubMed, Scopus, and Web of Science, screened 60 initial records, and included 22 sources after screening and cross-citation; the narrative component synthesized 62 conceptually selected articles across human–AI interaction and adoption/integration.
- The paper is a peer-reviewed mixed-method review rather than a longitudinal primary study, and its broad narrative corpus includes empirical studies, reviews, conceptual analyses, and opinion pieces, so it maps concerns and research needs more directly than it estimates the causal prevalence or rate of clinical deskilling.
- The authors note unresolved conceptual boundaries between deskilling, upskilling inhibition, and reskilling, plus a lack of clear metrics and structured empirical approaches for measuring these phenomena in practice.