Mode
Text Size
Log in / Sign up

Generative AI in medical education poses risks of deskilling and automation bias in clinical trainingArtificial intelligence poses new risks for medical student training

AI-generated summary of the cited source, checked by automated accuracy review. How we work

Key Takeaway
Recognize that generative AI may cause deskilling and automation bias, necessitating a focus on human-centric reasoning.

This narrative review explores the integration of generative artificial intelligence (AI) in medical education, focusing on its impact on cognitive development and clinical reasoning. The authors synthesize information regarding how AI tools may influence the training of undergraduate, graduate, and continuing medical education students.

The review identifies three primary cognitive vulnerabilities: deskilling (the loss of diagnostic abilities), never-skilling (the failure to develop foundational mental models), and automation bias (the uncritical acceptance of machine-generated recommendations). Furthermore, the authors argue that traditional memory-based assessments, such as multiple-choice questions, are becoming insufficient for evaluating clinical readiness in an AI-integrated environment.

To mitigate these risks, the authors suggest a shift toward educational methodologies that AI cannot easily replicate, such as Case-Based Learning (CBL) and Problem-Based Learning (PBL). They propose a division of labor where AI manages data processing while humans focus on patient interaction, emphasizing pathophysiological reasoning and human-AI collaboration. The review notes that these findings are conceptual risks identified in the literature rather than empirical data on the frequency of these occurrences.

Medical students today face a new challenge: how to learn the art of healing while using powerful AI tools. While these tools can process data quickly, experts warn that they might create a trap called deskilling. This happens when students rely too much on technology and lose their own ability to diagnose patients or think through complex problems.

There is also a risk of never-skilling, where students fail to build the basic mental models needed for medicine. Another concern is automation bias, which is the tendency to accept what a machine says without questioning it. These issues make it harder for students to develop the deep reasoning and human connection required to care for real people.

Because of these risks, experts suggest a shift in how we teach. Instead of just memorizing facts, students should focus on methods AI cannot replicate, like problem-based learning. The goal is a partnership where AI handles the data while humans focus on the patients, ensuring that the human touch remains at the heart of medicine.

What this means for you:
Over-reliance on AI in medical training may lead to deskilling and a loss of critical clinical reasoning.

Common questions

What are the risks of using AI in medical education?

There are three main risks identified: deskilling, which is the loss of diagnostic abilities; never-skilling, which is the failure to develop basic mental models; and automation bias, which is the uncritical acceptance of machine-generated advice. These issues can make it harder for students to develop the reasoning needed to treat patients.

How is AI changing how medical students are tested?

Traditional memory-based tests, like multiple-choice questions, are becoming less effective at showing if a student is truly ready to practice medicine. Because of the rise of AI, educators are looking for ways to better test clinical readiness and reasoning.

How can medical schools balance AI use with student learning?

Experts suggest a division of labor where AI handles data while humans focus on patients. They recommend teaching methods that AI cannot replicate, such as problem-based learning and a focus on the underlying reasons for diseases, to ensure students maintain their humanistic skills.

Study Details

Study typeSystematic review
EvidenceLevel 1
PublishedSep 2026
View Original Abstract ↓
BackgroundGenerative artificial intelligence (AI) is rapidly transforming medical education. While AI enhances personalized learning, its integration raises concerns regarding “cognitive outsourcing,” automation bias, and erosion of independent clinical reasoning.Aim & methodsThis narrative review critically evaluates the cognitive and educational impacts of generative AI and proposes structural realignment across the medical education continuum. We synthesized literature from major databases (PubMed, Scopus) focusing on AI's intersection with clinical reasoning and humanistic competencies.Main findingsWe identify three cognitive vulnerabilities: deskilling (loss of diagnostic abilities among advanced learners), never-skilling (failure to develop foundational mental models in junior trainees), and automation bias (uncritical acceptance of machine-generated recommendations). Traditional memory-based assessments, such as multiple-choice questions (MCQs), are increasingly insufficient for evaluating clinical readiness.Key recommendationsMedical education needs to pivot towards methodologies AI cannot replicate, acknowledging that strategies like case-based learning (CBL) and problem-based learning (PBL) are not novel but are now urgently necessitated by AI. We propose a division of labor where “AI treats the chart” while “humans treat the patient.” Undergraduate medical education (UME) is encouraged to prioritize pathophysiological reasoning to counter never-skilling; graduate medical education (GME) would benefit from emphasizing human-AI collaboration and deliberate reflection to counter deskilling and automation bias; continuing medical education (CME) may consider facilitating lifelong adaptation and structured “unlearning” of obsolete heuristics.ConclusionRather than competing with algorithmic memory, medical education needs to cultivate “augmented clinicians” equipped with high AI literacy and profound humanness, ensuring technology enhances rather than displaces relational patient care.
Free Newsletter

Clinical research that matters. Delivered to your inbox.

Join thousands of clinicians and researchers. No spam, unsubscribe anytime.