A 5-year window to avoid being left behind

The world of healthcare is undergoing an unprecedented technological transformation. Artificial intelligence is gradually being integrated into clinical practices, hospital information systems, medical research and the patient relationship. And this integration is going to accelerate.

According to the report Artificial Intelligence in Healthcare published by the World Health Organization (WHO) in 2021, healthcare systems around the world will, by 2030, undergo a profound transformation of their practices under the combined effect of the digitisation of health data and the massive deployment of clinical AI tools. In France, the Ministerial Delegation for Digital Health (DNS) estimates in its roadmap Ma Santé Numérique 2023-2027 that all healthcare professionals will be affected by AI tools in their daily practice by the end of the decade.

The question is therefore no longer “will I one day need to master AI in healthcare?” but “which precise skills must I acquire, and in what order, to practise my profession competently and responsibly in an environment increasingly augmented by AI?”

Why the next 5 years are decisive

Several converging signals indicate that the adaptation window for healthcare professionals is precisely that of the next 5 years.

On the regulatory side, the European AI Act, which came into force in August 2024 (Official Journal of the European Union, Regulation EU 2024/1689), requires deployers of high-risk AI systems, which include most clinical AIs, to ensure that their users have the competence, training and authority needed to oversee these systems. Concretely, using a clinical AI tool without documented training on that tool could constitute regulatory non-compliance by 2026.

On the establishments’ side, the Haute Autorité de Santé (HAS) published in 2023 an evaluation framework for AI systems in healthcare that explicitly recommends that any deployment of clinical AI be accompanied by a user-training programme.

On the studies’ side, a survey conducted in 2023 of 1,400 European doctors by Deloitte (European Health Tech Report, 2023) reveals that 78% of respondents believe AI will have a significant impact on their practice within 5 years, but that only 21% feel sufficiently trained to face this impact. The gap between the perception of the issue and the actual level of preparation is considerable and constitutes a concrete opportunity for those who act now.

Skill 1: understanding the basic functioning of clinical AI tools

What this means concretely

It is not about knowing how to program an algorithm or mastering the mathematics of deep learning. It is about understanding, at a functional level, how an AI system learns from data, what the potential sources of error are, and why an algorithm can perform differently depending on the population.

This basic understanding makes it possible to answer three essential questions when faced with any AI tool: On what data was it trained? What is its documented performance in a context comparable to mine? What types of errors does it preferentially produce?

Why it’s urgent

The European AI Act (Regulation EU 2024/1689, article 26) requires users of high-risk AI systems to be able to “understand the capabilities and limitations of the AI system.” This is not a rhetorical formula: it is a legal obligation whose application will be gradual but certain by 2026.

Moreover, a study published in The Lancet Digital Health in 2022 (Nagendran et al.) showed that clinicians who understand how an AI tool works detect its errors significantly more often than those who use it as a “black box.” Understanding is not an academic luxury: it is a patient-safety factor.

Skill 2: critically evaluating an AI tool

What this means concretely

The digital health and medical AI markets are booming, and they are not free of offerings whose advertised performance does not withstand rigorous evaluation. Knowing how to evaluate an AI tool before adopting or using it has become a professional skill in its own right.

This evaluation covers several dimensions: the quality and representativeness of the training data, the robustness of the clinical validation (retrospective or prospective study? on what population? funded by whom?), transparency about real performance under conditions of use, and regulatory compliance (CE marking for medical-device software, HAS certification where applicable).

Why it’s urgent

The HAS put in place in 2022 a specific framework for evaluating medical decision-support software incorporating AI. This framework creates an evaluation standard that healthcare professionals, as prescribers and users of these tools, have every interest in mastering.

Skill 3: mastering generative AI tools for professional use

What this means concretely

Conversational artificial intelligences, ChatGPT, Claude, Gemini, Mistral and their professional equivalents, are already used by a growing proportion of healthcare professionals for varied purposes: summarising records, drafting reports, preparing responses to patients, accelerated bibliographic research, support for clinical reflection.

Mastering these tools for professional use is not just about knowing how to ask a question. It involves knowing how to formulate precise and contextualised instructions (effective prompting), systematically verifying the information produced, knowing the situations where these tools are unreliable (complex medical reasoning, recent data, rare cases), and using them within the ethical and regulatory framework.

Why it’s urgent

A survey conducted in 2024 by Ipsos on behalf of the Observatoire de la e-santé reveals that 43% of French doctors say they have already used a generative AI in a professional context, but that 67% of them have received no training on the conditions of compliant use. This gap between actual use and training is a growing source of ethical and legal risk.

Skill 4: mastering the applicable legal and ethical framework

What this means concretely

Using AI in healthcare entails precise legal and ethical responsibilities. Understanding these responsibilities; GDPR, medical confidentiality, AI Act, medical-device regulation; is not reserved for lawyers and DPOs. It is a skill that every healthcare professional using AI tools must integrate at an operational level.

Concretely, this means knowing how to answer practical questions: is this software a medical device subject to CE marking? Is my patient data processed in compliance with the GDPR when I use this tool? Must I inform my patient that an algorithm took part in their diagnosis? What is my liability if the AI is wrong and I followed it without exercising my judgement?

Why it’s urgent

The National Council of the Order of Physicians (CNOM) recalled in its 2022 report that the continuing-education obligation of doctors explicitly extends to digital technologies and AI. Failing to train in the applicable framework constitutes an ignorance that does not exonerate one from professional liability in the event of a breach.

Skill 5: interpreting and contextualising the results produced by an AI

What this means concretely

An algorithm produces scores, probabilities, suggestions, alerts. These results are not interpreted in isolation: they must be contextualised within the patient’s overall clinical picture, weighted by the known limits of the tool, and integrated into a medical reasoning that remains the exclusive responsibility of the healthcare professional.

This critical-interpretation skill involves understanding what a 73% risk score means, knowing when an automatic alert deserves to be escalated and when it can be contextualised, and recognising the situations in which the clinical picture must take precedence over the algorithmic suggestion.

Why it’s urgent

Studies published in JAMA and The BMJ have documented the phenomenon of automation bias, the tendency of clinicians to over-weight the results of an automated system relative to their own clinical judgement (Goddard et al., Journal of the American Medical Informatics Association, 2012). This bias increases with the confidence placed in the system and decreases with the understanding of its limits. Training in the critical interpretation of AI outputs is therefore directly linked to patient safety.

Skill 6: contributing to the design and deployment of AI tools

What this means concretely

This skill is less immediate than the previous ones, but it is becoming strategic as healthcare establishments multiply their AI projects. Healthcare professionals who know how to actively participate in an AI deployment project; by expressing real clinical needs, evaluating the relevance of the proposed use cases, identifying risks for patients, co-building the protocols of use; bring considerable value to their establishments.

Why it’s urgent

The report Harnessing AI for Health published by the OECD in 2023 emphasises that AI projects in healthcare fail significantly more often when healthcare professionals have not been involved in their design. Clinicians trained in this skill are today in a position of real influence over the trajectory of their institutions.

Where to start: a progressive roadmap

Faced with these six skills, a practical question naturally arises: where to start, given that the time available for training is scarce for most healthcare professionals?

Phase 1: Fundamentals (0 to 6 months)

Understanding the basics of how AI works (skill 1), mastering generative AI tools for immediate professional use (skill 3), acquiring the essential reference points of the legal and ethical framework (skill 4). These three skills are the most immediately actionable and the most useful day to day.

Phase 2: Deepening (6 to 18 months)

Developing the ability to critically evaluate tools (skill 2) and the contextual interpretation of algorithmic results (skill 5). These skills require more practice and clinical situational application.

Phase 3: Leadership (18 months and beyond)

Getting involved in the deployment and design projects of AI tools in your establishment or structure (skill 6). This phase builds on the previous ones and is part of a gradual increase in responsibility.

Training now means choosing to fully practise your profession tomorrow

Healthcare professionals who acquire these skills over the next 5 years do not become medical-IT specialists. They become augmented clinicians, able to practise their profession with the tools of their time, knowingly and in full responsibility.

Conversely, those who wait for training to come to them, by regulatory obligation or institutional mandate, risk finding themselves passive users of tools they do not understand, unable to evaluate their limits or to question their deployment.

The good news is that these skills can be acquired. They require neither years of study nor a prior technical background. They require method, teaching adapted to the medical field, and the decision to start.

Would you like to acquire these skills in a structured way, at your own pace, with use cases directly drawn from your specialty? Our training courses for healthcare professionals cover all six of these skills, from understanding the basics to actively participating in the AI projects of your establishment. Discover the programme suited to your profile.

Sources and references

1. WHO. _Ethics and governance of artificial intelligence for health_. World Health Organization, 2021.
2. Regulation (EU) 2024/1689 of the European Parliament and of the Council — Artificial Intelligence Act. _Official Journal of the European Union, August 2024_.
3. Ministerial Delegation for Digital Health (DNS). Roadmap _Ma Santé Numérique 2023-2027_, 2023.
4. HAS. “Good-practice framework: AI in healthcare.” _Haute Autorité de Santé_, 2023.
5. HAS. “Certification framework for medical decision-support software.” _Haute Autorité de Santé_, 2022.
6. Deloitte. _European Health Tech Report_. 2023.
7. Nagendran, M. et al. (2022). “Artificial intelligence versus clinicians: systematic review of design, reporting standards, and claims of deep learning studies.” _The Lancet Digital Health_.
8. Ipsos / Observatoire de la e-santé. “E-health professionals barometer.” 2024.
9. CNOM. “Doctors and patients in the world of data, algorithms and AI.” Report, 2022.
10. Goddard, K., Roudsari, A. & Wyatt, J. C. (2012). “Automation bias: a systematic review of frequency, effect mediators, and mitigators.” _Journal of the American Medical Informatics Association_, 19(1), 121–127.
11. OECD. _Harnessing AI for Health_. 2023.