Elsevier Medical Education presents an interview with Juan José Beunza Nuin, specialist in Internal and Tropical Medicine, Professor of Public Health and Interprofessional Education at Universidad Europea de Madrid, and Director of the AI and Health research group (IASalud), exploring how to integrate AI into healthcare education in a structured, ethical, and sustainable way.
From an institutional perspective, how should medical schools begin integrating AI into their curricula in a structured and sustainable way?
Simply incorporating technology has never been a good strategy on its own. We shouldn’t start by buying a tool or creating a standalone course. We need to begin with an institutional reflection: what kind of healthcare professional do we want to train for a clinical system where AI is already involved in diagnosis, research, management, and patient care?
Defining the doctor of the “AI-era” first
The first step would be to define what the doctor “of the future” or “of AI” looks like, mapping out both the technological competencies they will need as well as the human competencies that AI risks undermining in their professional development. Outsourcing human skills to AI is a risk that must be taken seriously.
The fundamentals could be introduced in the early year courses: what is AI, how models are built, what it means to train them on data, and what their limitations are. From there, this knowledge should be applied to clinical cases, research, diagnostic imaging, health documentation, public health, or decision-making. Finally, during clinical placements, students should learn to use, monitor, and critically question real or simulated systems – including AI agents. Also, there will almost certainly need to be periods of training where AI is banned entirely, in order to foster the development of personal skills.
Five pillars for sustainable integration
For the process to be sustainable, I consider five key elements essential:
- Institutional leadership
- Faculty training
- Collaboration across medicine, engineering, bioethics, and law
- Secure technological infrastructure
- Ongoing evaluation of results
The goal is not to teach a specific platform – which will likely become obsolete – but to develop transferable skills. Implementation must be people-centered, interdisciplinary, transparent, equitable, and subject to ongoing evaluation. Many of these elements are absent from the most impactful commercial products today. That’s why it’s essential to design curricula with the needs of our own clinical environment in mind, rather than around whatever the commercial market currently offers.
What are the main challenges medical schools face when adopting AI in medical education, and how can they overcome them?
Challenge 1 – Faculty unpreparedness
The first major challenge is faculty unpreparedness. Many educators have deep expertise in their clinical discipline but have received no training on algorithms, generative models, data quality, or critical evaluation of AI systems. The answer is not to turn every professor into an engineer, but rather to offer them practical literacy training, support, and safe spaces for experimentation.
Challenge 2 – The pace of technological change
The second challenge is the pace of technological change. A curriculum built exclusively around specific tools can become outdated within months. That’s why the curriculum must focus on enduring principles: critical thinking, validation, safety, ethics, governance, communication, and professional responsibility.
Challenge 3 – Data protection
Another challenge is data protection. Students and faculty should not be allowed to enter identifiable clinical information into commercial tools without institutional safeguards. Schools need clear policies covering privacy, intellectual property, consent, traceability, and permitted uses.
Challenge 4 – Rethinking assessment
Finally, we must address the challenge of assessment. When a student can generate a text, a differential diagnosis, or an academic response using AI, evaluating only the final product is no longer sufficient. We must assess the process: how they frame the problem, what sources they use, how they verify the response, what errors they detect, and how they justify their decision.
These difficulties are not purely technological. The European report on digital health competencies highlights training gaps, funding shortfalls, time constraints, infrastructure issues, and inequalities among professional groups. It also highlights the need for cooperation between the health and education sectors and for regular curriculum review.
"We don’t need healthcare professionals who obey the algorithm – we need healthcare professionals who know when to trust it, when to question it, and when to discard it."
Juan José Beunza Nuin, Professor of Public Health, Universidad Europea de Madrid | Elsevier Faculty Hub, July 2026
What key AI-related competencies should medical schools prioritize to prepare future healthcare professionals for an increasingly digital clinical environment?
The first is AI literacy. Every healthcare professional should understand, at least conceptually, how a model learns, what distinguishes correlation from causation, what training data is, and why a system can produce seemingly convincing errors.
The second is data literacy: knowing how to interpret the quality, representativeness, and origin of data. A model is only as good as the data and human decisions that went into building it.
The third is critical evaluation of clinical tools. The future physician must ask: Is this system validated in a population comparable to my patient’s? What are its sensitivity and specificity? How does it integrate into the clinical workflow? What happens when it fails? Does it actually improve clinical outcomes?
The fourth is recognizing bias, inequity, and generalization problems. Students must understand that an algorithm may perform well in one hospital but poorly in another, or yield different results depending on a patient’s age, sex, background, or socioeconomic conditions.
Privacy, cybersecurity, ethics, regulation, and professional accountability must also be prioritized. Using an algorithmic recommendation does not eliminate the clinician’s responsibility. Healthcare decisions must remain comprehensible, justifiable, and overseen by a human being.
I would add one especially important competency: patient communication. Healthcare professionals must be able to explain when AI was used, what role it played, and what its limitations are, without delegating the therapeutic relationship to technology.
Finally, students must learn how to collaborate with AI systems: formulating good questions, verifying results, cross-checking sources, and documenting their use. We don’t need healthcare professionals who obey the algorithm – we need healthcare professionals who know when to trust it, when to question it, and when to discard it.
Is the use of AI in medical education already a widespread reality today, or is there still a long way to go?
It is already a reality but not yet a uniform, mature, or fully institutionalized one. AI is being used to generate cases and questions, create virtual patients, personalize activities, provide feedback, support research, design simulations, and track student progress. In the United States and Canada, the percentage of medical schools reportedly integrating AI into their curricula rose from 53% in 2023 to 77% in 2024.
However, incorporating some AI-related activity is not the same as having a comprehensive program. In many institutions, we find initiatives driven by individual faculty members, voluntary workshops, or pilot experiences – but there are not always structured curricula, governance policies, trained faculty, or assessment frameworks. Some say we are in the age of "pilotitis."
There is also a gap between the pace of clinical adoption and the pace of educational adoption. In 2026, the WHO reported that nearly three-quarters of EU countries were already using AI in diagnostics, while continuing to highlight the need to accelerate professional training.
Therefore, we can say that we have moved past the question of whether AI should be introduced into medical schools. The question now is how to integrate it with rigor, safety, and clear educational purpose. There is still a long way to go — especially when it comes to transforming isolated experiences into sustainable institutional strategies.
Finally, in your expert opinion, what are the main trends that will shape the future of AI in healthcare education?
1. Personalized learning
The first major trend will be the personalization of learning. Systems will be able to adapt cases, questions, and explanations to each student’s level, identify specific difficulties, and offer immediate feedback. The challenge will be preventing this personalization from locking students into overly automated learning pathways.
2. Multimodal virtual patients
The second will be the development of multimodal virtual patients. Students will be able to train with patients capable of holding conversations, displaying symptoms, providing diagnostic test results, and reacting to different decisions. This will allow them to practice rare, complex, or high-risk scenarios without putting any real person at risk.
3. Assessment
The third will be a profound transformation of assessment. Tasks focused solely on reproducing information will carry less weight, while problem-solving, argumentation, evidence verification, and the ability to detect AI-generated errors will become increasingly important.
4. AI as a clinical co-pilot
The fourth trend will be the incorporation of AI as a clinical and educational co-pilot. Students will learn not only medicine, but also how to work with systems that summarize medical records, structure information, generate hypotheses, and help prepare for patient encounters. The differentiating skill will no longer be the ability to produce a quick response, but rather the ability to oversee its quality.
We will also see greater integration between AI, simulation, virtual reality, and monitoring devices – especially in surgery, emergency medicine, nursing, and interprofessional training.
5. Governance
Governance will be another key factor. Schools will establish committees, secure technological environments, catalogs of approved tools, and mechanisms to audit for bias, privacy, educational impact, and safety. Innovation without governance will no longer be acceptable. I also believe that the deployment of local solutions – using open-source models adapted to local languages, contexts and tasks – will gain significant traction. While this requires significant initial investment in hardware, it provides absolute guarantees of data privacy, both in educational and clinical settings.
6. Human competencies
Finally, the importance of human competencies will grow. The greater the technical capabilities of AI systems, the more valuable clinical judgment, empathy, listening skills, communication, ethical deliberation, and accountability will become. AI should not lead us toward a less human form of medicine. When used well, it should free up time and cognitive capacity so that professionals can practice medicine that is more attentive, more personalized, and more humane. That’s why it’s important to orient the development and deployment of AI toward people and not solely toward the commercial interests of a handful of developer companies.
Ultimately, the future will not be a competition between the healthcare professional and AI. The difference will be between those who know how to use it critically, ethically, and responsibly – and those who have not been prepared to do so.
Leave a Reply