Healing in the age of algorithms: what Asia teaches us about adapting healthcare systems to artificial intelligence?

By Jean-Emmanuel Bibault

Introduction: The clash of temporalities

There is something vertiginous about observing, from a Parisian radiotherapy oncology department, the speed at which certain Asian healthcare systems have integrated artificial intelligence into their daily practices. Not as a pilot project confined to a research laboratory, nor as a promise waved about at a conference, but as a real infrastructure, anchored in clinical workflows, patient records, triage protocols. In Seoul, in Shenzhen, in Singapore, diagnostic AI is no longer a subject of prospective thinking: it is a tool that the radiologist consults before signing their report, that the nurse sees activate when a patient signals an unusual pain from their home.

This observation does not aim to fuel a discourse on the European "delay," whose rhetorical mechanics are as lazy as they are ineffective. It is rather a matter of questioning a more fundamental asymmetry: that of the speeds of systemic adaptation. Because the question that arises today is not so much whether AI is technically capable of transforming medicine (the evidence is accumulating) as understanding why certain healthcare systems manage to absorb this transformation when others seem to undergo it or defer it.

Asia offers in this regard an exceptional observation terrain, precisely because it is not monolithic. Between Chinese industrial voluntarism, Singaporean governance founded on algorithmic trust, the Japanese response to demographic constraint and Korean pragmatism in matters of regulatory approval, the continent presents a diversity of models that invalidates any simplistic reading. What these experiences share, on the other hand, is a disposition to think of AI not as a tool that one grafts onto an existing system, but as an opportunity to reorganize it in depth.

This article does not claim to be exhaustive. It seeks, from a cross-reading of Asian transformations and a clinical field experience in oncology, to identify the conditions of this adaptation (organizational, regulatory, cultural) and to identify what France and Europe would have an interest in looking at with greater attention, without naivety but without condescension.

1. Asia as a global laboratory of AI in health

1.1. China: the power of data elevated to the rank of state strategy

China did not wait for AI to be mature before deciding that it would be one of its global leaders. As early as 2017, the State Council published its New Generation AI Development Plan, setting the horizon of 2030 as the deadline to reach the first place worldwide. What strikes, with hindsight, is less the displayed ambition (the rhetoric of Chinese five-year plans is accustomed to spectacular objectives) than the continuity and acceleration of their execution. The 15th Five-Year Plan, adopted in March 2026 by the National People's Congress, marks a new stage: it is the first to place artificial intelligence explicitly at the heart of the national economic strategy, mentioning it more than fifty times against eleven in the previous one, with the objective of an AI industry valued at 10,000 billion yuan and a 90% adoption rate in the economy by 2030. Health features prominently, alongside energy, biotechnology and quantum computing.

In the medical sector, this ambition translates into deployments of a scale without equivalent in the West. Companies like Infervision, DeepWise or Shukun Technology have developed medical imaging analysis solutions (detection of pulmonary nodules, triage of thoracic scans during epidemic periods, analysis of histological slides) now deployed in hundreds of hospitals. By mid-2024, 92 class III AI-powered medical devices had been approved by the National Medical Products Administration (NMPA), testifying to clear regulatory support for clinical deployment. The AI health market, valued at 1.59 billion dollars in 2023, is projected at nearly 19 billion by 2030, reflecting an annual growth rate of around 42%.

This dynamism rests on an asset that China has managed to transform into a decisive competitive advantage: the volume and centralization of its health data. With 1.4 billion inhabitants, a system of medical records progressively unified under the "Healthy China 2030" initiative, and a digital culture integrated into daily uses (WeChat is simultaneously a messaging service, payment system and health interface), the country possesses a mass of longitudinal data that no European system can approach in scale. To train deep learning models in oncology or cardiology, this abundance is a structural condition, not a technical detail.

1.2. Singapore: governance as competitive advantage

At the opposite end of Chinese gigantism, Singapore illustrates what a state of five million inhabitants, endowed with solid institutions and a coherent strategic vision, can accomplish when it decides to make algorithmic trust a national advantage. The National AI Strategy, published in 2019 then revised in 2023, explicitly positions Singapore as a global hub of trustworthy AI (trustworthy AI in official terminology) by articulating technological innovation and ethical frameworks from the design of systems.

In the health domain, the Ministry of Health has engaged in a progressive integration of clinical decision support tools in public polyclinics, those primary care structures that constitute the first level of contact with the healthcare system. Risk stratification algorithms for chronic diseases (diabetes, renal insufficiency, cardiovascular risk) are today operational there, with decision traceability and human supervision mechanisms explicitly integrated into protocols. What is remarkable in the Singaporean model is not so much the technical sophistication of the tools as the rigor of the deployment framework: prior clinical validation, mandatory training of practitioners, regular algorithmic audit devices.

This model benefits from an exceptional facilitating context: a homogeneous healthcare system, a digitally educated population, a unique national identifier (MyInfo) that allows native interoperability of data across all health actors. These conditions are not reproducible identically in France or Europe, but they designate targets toward which to tend.

1.3. Japan and South Korea: two responses to demographic constraint

Japan and South Korea share a common demographic challenge (accelerated aging, foreseeable shortage of caregivers) but have responded to it with distinct strategies, equally instructive.

Japan has engaged for several years in an ambitious program of robotics and AI in the service of home maintenance and care for the elderly. From physical assistance robots to early fall detection systems, passing through conversational interfaces for the monitoring of isolated patients, Japanese innovation in e-health is profoundly marked by this geriatric urgency. On the diagnostic front, the Japan Radiological Society has integrated AI into its practice recommendations for thoracic imaging analysis, and the PMDA (Pharmaceuticals and Medical Devices Agency) has put in place an accelerated approval pathway for AI-based medical devices, recognizing the necessity of adapting the regulatory tempo to the rhythm of innovation.

South Korea has opted for a different strategy, more driven by a dynamic startup ecosystem and a strong cultural appetite for medical technology. The MFDS (Ministry of Food and Drug Safety) has developed specific approval procedures for artificial intelligence medical software, allowing faster market launches than in most European countries. Companies like Lunit or Vuno have developed oncological and cardiological imaging analysis solutions recognized internationally, sometimes approved in the United States via the FDA before even having completed their domestic deployment. Korea also understood very early the soft power dimension of e-health: K-health, in the image of K-pop or K-drama, has become a vector of influence and technological export in its own right.

1.4. India: AI as a lever of health equity

India occupies a singular place in this panorama, not for the sophistication of its urban hospital systems (comparable to those of high-income countries) but for the way in which it uses AI as a tool for reducing inequalities in access to care. In a country where the doctor-patient ratio remains dramatically insufficient in rural areas, AI is not an instrument of marginal optimization: it is a condition of possibility for universal health coverage.

AI-assisted cervical cancer screening programs, deployed with community health workers equipped with smartphones, have shown results comparable to those of an expert gynecologist in contexts where the latter is simply absent. More broadly, China and India share this characteristic of facing a massive epidemiological transition (chronic diseases, cancers, cardiovascular pathologies) for which traditional public health models are manifestly insufficient, and where AI represents a scalable response to a public health emergency. It is a lesson in pragmatism that high-income healthcare systems, inclined to think of AI as a technological luxury, would be wrong to neglect.

2. The structural conditions of adaptation

2.1. Data interoperability: a sine qua non condition

There exists a recurring temptation in debates on AI in health: that of placing the performance of algorithms at the center of the conversation, as if the quality of a model were sufficient to guarantee its clinical utility. This is an error of perspective. The most sophisticated algorithm in the world produces no medical value if it cannot access, in real time and in a structured manner, the data it needs to function. Before being a question of artificial intelligence, the adaptation of healthcare systems is a question of data infrastructure.

It is precisely on this terrain that the gap between the most advanced Asian systems and the majority of European systems is most significant. In China, the "Healthy China 2030" initiative has imposed a progressive standardization of electronic medical records at the national scale, backed by a unique patient identifier allowing continuity of follow-up across levels of care. In Singapore, the MyInfo system constitutes a digital backbone from which health data can circulate between authorized actors without a break in the chain. In South Korea, interoperability between university hospitals, city clinics and national health insurance systems is ensured by exchange standards imposed regulatorily since the late 2010s.

The French situation is neither desperate nor satisfactory. Mon Espace Santé, the Shared Medical Record in its renovated version, the Digital Health Space constitute real advances, carried by political will and substantial investments. But their adoption remains partial, their population uneven according to establishments and specialties, and their effective interoperability with hospital information systems (often aging and heterogeneous) remains an open construction site. The national AI and health data strategy, whose artificial intelligence component was presented in November 2025, explicitly recognizes this priority, articulating four structuring axes: clarifying regulation, strengthening the evaluation of AI solutions, supporting professionals, and building a sustainable economic framework for innovation. The intention is right. The speed of execution will be determining.

This point is not technical: it is political. Building an interoperable health data infrastructure supposes resolving conflicts of interest between actors that are not resolved by the sole virtue of technical standards. The Asian countries that have achieved this have done so through a combination of strong state will, financial incentives and, often, a cultural tolerance with regard to the centralization of data that European societies do not have and are not destined to adopt. The European path will have to be different (more negotiated, more distributed, more protective of individual rights) but it cannot afford to be indefinitely slower.

2.2. Training and acculturation of professionals

The second condition of adaptation is human, and it is often underestimated in national strategies that privilege infrastructures and regulations. Deploying a decision support tool in a hospital department without having prepared caregivers to use it critically amounts to installing high-precision equipment in a room without training its operators. The result is not neutral: it can be actively harmful.

Two symmetrical risks lurk for the practitioner confronted with an AI system. The first is excessive mistrust, which leads to ignoring relevant alerts in the name of a clinical intuition that the tool, statistically, surpasses on certain tasks. The second (and it is more insidious) is what the literature in cognitive psychology calls automation bias: the tendency to delegate one's judgment to the machine, to validate without questioning, to substitute trust in the algorithm for clinical thinking. Both pathologies have been documented in studies on AI systems deployed in radiology, intensive care and oncology. They are not inevitable, but they do not prevent themselves spontaneously.

What the most successful experiences in Asia show is that training in AI tools cannot be reduced to an interface familiarization session. It supposes a minimal understanding of the functioning of models (their strengths, their blind spots, the potential biases linked to training data) and an explicit reflection on human supervision protocols. In Singapore, continuing education programs for doctors now integrate mandatory modules on clinical AI literacy. In South Korea, several medical faculties have restructured part of their curriculum around the understanding of health data and the interpretation of algorithmic outputs.

In France, the question is beginning to be raised (timidly in medical faculties, more directly in a few DESC and specialized training programs) but remains largely treated as an option rather than a priority. Training caregivers to work with AI also means training them to resist its errors, to identify its limits, to maintain the primacy of clinical judgment in the decision loop. It is, in other words, a condition of patient safety, not just a professional competitiveness issue.

2.3. Organizational architectures: who pilots, who validates?

The third condition of adaptation is organizational. The integration of AI into a healthcare system does not reduce to the acquisition of software or the deployment of a model in a hospital server. It supposes a reconfiguration of responsibilities, validation circuits and decision chains, a transformation that traditional hospital organizational charts, designed for stable and hierarchical care flows, are not naturally equipped to absorb.

The central question is that of governance: who, in an establishment, is responsible for the selection of AI tools, their clinical validation, their post-deployment surveillance and the management of their failures? In most French hospitals, this responsibility is diffuse, shared between IT departments, medical establishment commissions and user departments, without clear attribution or formalized process. This is a configuration that exposes to two risks: under-deployment through excess of bureaucratic caution, or conversely the precipitous deployment of insufficiently validated tools.

The most advanced Asian healthcare systems have begun to respond to this question through the emergence of new roles and new structures. Positions of Clinical AI Lead or Chief Medical AI Officer are appearing in major hospitals in Seoul, Tokyo and Singapore, embodying a translation function between medical culture and data culture, and carrying the responsibility of algorithmic governance at the management level. Algorithmic ethics committees, distinct from traditional research ethics committees, are being put in place to evaluate the implications of deployments at the scale of patient populations.

This movement sketches in negative what the hospital of tomorrow could be: no longer a place where AI is one tool among others, managed at the margins by IT teams, but an organization where artificial intelligence is treated as a critical infrastructure, subject to the same requirements of reliability, traceability and responsibility as any other biomedical equipment. This change of status (from tool to infrastructure) is perhaps the most important conceptual transformation that European healthcare systems still have to accomplish.

3. The regulatory battlefield: between the EU AI Act and asian pragmatism

3.1. The European framework: ambitious, but whose complexity is a risk in itself

The European Union has made the choice to be the first region in the world to equip itself with a comprehensive legislative framework on artificial intelligence. The EU AI Act entered into force in August 2024 and applies progressively, establishing a classification by risk level and prohibiting systems presenting an unacceptable risk, while imposing specific obligations on high-risk systems to guarantee their safety, performance, transparency and accountability.

For medical devices, the implications are considerable. Any AI system used for diagnosis, therapeutic planning or patient monitoring is automatically classified as high risk, which includes AI-based medical software, deep learning models and decision support tools influencing clinical outcomes. Manufacturers of medical devices integrating AI must now comply with both the MDR/IVDR regulation and the EU AI Act, the latter adding specific requirements in terms of data quality, algorithmic governance, traceability, transparency and human supervision.

The EU AI Act has been in force since August 1, 2024, but its most constraining obligations for high-risk AI systems will not apply before August 2, 2027. Within the framework of the Digital Omnibus Package presented by the European Commission in November 2025, this deadline could be pushed back to 2028, in order to give manufacturers more time to prepare.

This delay reveals in negative the real difficulty of the exercise. In October 2025, no notified body had yet been designated under the AI Act, which means concretely that there is still no accredited body to evaluate the conformity of high-risk AI systems in Europe. The framework is in place; the infrastructure of its implementation is showing a worrying delay. For companies developing medical AI tools intended for the European market, this uncertainty is not neutral: it slows investments, complicates regulatory roadmaps and can, in certain cases, orient market launch priorities toward geographic areas with more legible governance.

It would however be reductive to see in this only an obstacle to innovation. The requirements for human supervision, algorithmic traceability and data governance that the EU AI Act imposes correspond to real clinical and ethical necessities. What the Asian experiences underline is that these requirements are better absorbed when they are integrated from the design of systems, in a logic of "compliance by design," rather than plastered at the end of the chain onto tools developed without them.

3.2. Asian models: regulatory agility and variable maturity

Faced with this complex European landscape, Asian regulatory approaches offer a striking contrast, not because they would be less demanding on substance, but because they have made different choices on the sequence between innovation and framing.

China has opted for a voluntarist regulatory approval, allowing companies to deploy their solutions in partner hospitals for validation in real conditions, within the framework of a NMPA approval process that has allowed reaching 92 approved class III devices by mid-2024. This rapidity has a counterpart: post-marketing surveillance requirements remain less structured than in Europe, and algorithmic pharmacovigilance mechanisms (the equivalent, for medical software, of the monitoring of adverse drug effects) are still insufficiently developed.

South Korea and Japan have found a more subtle balance. The Japanese PMDA and the Korean MFDS have both developed specific approval pathways for artificial intelligence medical software, recognizing that learning software has properties (notably its capacity to evolve after deployment) that render traditional procedures inadequate. This regulatory specificity is an important conceptual advance that Europe is beginning to integrate, but with a delay that has been measured in years of deployment.

Singapore has opted for a regulatory sandbox type approach: controlled experimentation environments in which AI tools can be tested in real clinical conditions, before the definitive regulatory framework is finalized. The European AI Proposal of November 2025 foresees moreover widening the use of regulatory sandboxes and real-condition testing, signaling a welcome methodological convergence, even if belated.

3.3. Toward a necessary convergence, without naivety

The temptation, faced with this picture, is to conclude that Europe over-regulates and that Asia innovates better. This is a reading as seductive as it is inaccurate. The systems that have deployed most rapidly have not necessarily deployed best: speed without evaluative rigor produces tools whose real performance in populations remains poorly characterized, whose potential biases remain undocumented, and whose responsibility in case of failure remains legally undetermined.

What Europe has to bring to this international conversation is therefore not negligible. Its framework for the protection of patient rights, its requirements for algorithmic transparency, its culture of independent medico-economic evaluation constitute a body of exportable values, that several Asian countries are beginning to look at with interest, precisely because the limits of unregulated deployment are becoming visible.

The prospect of international regulatory convergence (within the framework of the IMDRF, the G7 health discussions, or the multilateral arenas where France is active) is both necessary and realistic. It does not mean uniformization, but the definition of shared minimum standards (on the quality of training data, on clinical validation requirements, on post-deployment surveillance mechanisms) without which the risk of permanent regulatory arbitrage will remain structural.

4. Oncology as a paradigmatic experimentation terrain

4.1. Why cancer?

If oncology has become the most active laboratory of medical AI, this is not by chance nor by fashion effect. It is because cancer management concentrates, in an exceptional density, the entirety of conditions that make artificial intelligence both useful and necessary: massive volumes of imaging to analyze, biological and genomic data of growing complexity, therapeutic decisions whose window of opportunity is narrow, and a longitudinal follow-up that extends over years.

Radiotherapy illustrates this potential particularly well. The segmentation of target volumes and organs at risk (an indispensable step in the planning of any treatment) is a long, repetitive task, whose inter-operator variability is documented and clinically significant. Automatic segmentation AI systems, trained on tens of thousands of expert-annotated scans, today reach performances comparable to those of an experienced radiologist on many tumor localizations. Their deployment does not suppress the practitioner's expertise (who validates, corrects and assumes the final decision) but frees up time for tasks of higher added value. It is precisely in this logic of augmentation of human competence, rather than substitution, that the most lasting contribution of AI in oncology resides.

In Asia, this logic has been grasped with particular acuity. The major oncological centers of Seoul, Tokyo and Shanghai have invested massively in integrated platforms combining imaging analysis, natural language processing for the extraction of information from clinical reports, and predictive models of treatment response. Their comparative advantage is not only technical: it lies in the capacity of these institutions to construct annotated data cohorts at a scale that most European establishments cannot reach individually.

4.2. Active patient monitoring: when AI enters daily life

The most profound innovation in oncology is perhaps not where one expects it. If diagnostic AI captures most of the media attention (the algorithm that detects a pulmonary nodule better than a radiologist), the most structurally significant transformation for patients plays out in ambulatory follow-up, between consultations, in the clinical silences where complications develop without being detected.

Cancer, indeed, is no longer treated exclusively in the hospital. Oral chemotherapy, targeted therapies, immunotherapy, hypofractionated radiotherapy have profoundly modified the geography of oncological care: the patient now spends the majority of their therapeutic journey at home, exposed to adverse effects that the caregiving teams do not see in real time. Silent hematological toxicity, progressive dehydration, febrile neutropenia that sets in at night: these situations that lead to unplanned hospitalizations (costly for the system, trying for the patient) are precisely those for which continuous monitoring, assisted by AI, offers the most added value.

It is in this context that the development of Patient-Reported Outcomes (PRO) platforms coupled with intelligent alert engines fits. The principle is simple in its formulation, complex in its execution: collect daily the symptoms reported by the patient from their home, analyze them in real time by a model trained to distinguish benign signals from alert signals, and trigger an escalation toward caregivers when the clinical profile reaches a critical threshold. The international literature has demonstrated, in several randomized controlled studies, that this type of monitoring improves not only the quality of life of patients but reduces mortality, a result of sufficient strength to be rare in oncology to merit being underlined.

In South Korea, integrated ambulatory oncological monitoring platforms are deployed in several major university centers, with automated alert systems allowing coordination nurses to identify in real time patients requiring an intervention.

4.3. What Asian systems understood first

There exists a temptation, in the evaluation of AI systems in health, to reduce the question of their value to their technical performance: sensitivity rate, specificity, area under the ROC curve. These indicators are necessary but insufficient. An algorithm that achieves 95% sensitivity on a test dataset and that is never used by caregivers in their real practice has no clinical impact. A technically less performant tool but perfectly integrated into work routines, understood and adopted by teams, can significantly transform the management of thousands of patients.

The Asian healthcare systems that have obtained the most probative results have understood this distinction earlier than most. They have invested as much in implementation engineering as in algorithmic engineering: user training, workflow redesign, feedback mechanisms allowing clinicians to signal system errors and feed its continuous correction.

This maturity is accompanied by a growing requirement for rigorous medico-economic evaluation. The demonstration of the clinical value of an AI tool is no longer sufficient to justify its sustainable financing: it is also necessary to demonstrate that its deployment cost is justified by a reduction in avoidable hospitalizations, an improvement in survival or an optimization of the allocation of caregiving resources. This requirement, that European health insurance systems are beginning to formulate (in France through the work of the HAS on digital medical devices), is structuring for the future of the sector.

Oncology, in this sense, is more than one domain of application among others. It is the mirror in which reflects, with particular clarity, what the adaptation of healthcare systems to AI really requires: not the sophistication of algorithms, but the maturity of the organizations that deploy them.

5. For a France-Asia dialogue on AI governance in health

5.1. What France and Europe have to learn

There is in the relationship that Europe maintains with Asian technological innovation a persistent ambivalence, oscillating between fascination and condescension, between competitive fear and a feeling of ethical superiority. This ambivalence is an obstacle to lucidity. It prevents one from looking squarely at what Asian experiences really teach, in their complexity, without reducing them either to models to imitate or to convenient counter-examples.

The most immediately useful lesson is perhaps the simplest: the speed of systemic adaptation is itself a competence that is organized, financed and governed. The Asian healthcare systems that have best integrated AI have done so because explicit political decisions have allocated resources to the construction of data infrastructures, to the training of professionals and to the putting in place of the necessary governance architectures. This strategic deliberation (its clarity, its continuity, its carrying at the highest level of the State) is what France lacks most, not for lack of intelligence nor of will, but for lack of temporal coherence in its commitments.

The second lesson is organizational. The most advanced systems have understood that the hospital of the algorithmic era is not the traditional hospital augmented by a few software programs: it is a profoundly different organization in its decision circuits, its competence profiles and its relationship to data. Preparing this transformation supposes acting simultaneously on several levers (initial and continuing training, recruitment of new hybrid profiles, revision of clinical protocols, overhaul of information systems) with a coherence that fragmented reforms struggle to produce.

The third lesson is perhaps the most counter-intuitive for systems accustomed to valuing regulatory caution as a cardinal virtue: the absence of deployment is not a neutral position. An AI tool that is not deployed does not generate algorithmic risks, but it generates risks of another nature, less visible but just as real: diagnoses made later, complications detected less early, caregiving resources mobilized on tasks for which they have less added value. The evaluation of AI risk in health cannot stop at the risks of commission; it must integrate the risks of omission.

5.2. What Europe has to contribute

The reciprocity of this dialogue is not a diplomatic politeness: it reflects a substantial reality. Europe, and France in particular, possesses a body of ethical, legal and epistemological reflection on AI in health that has no equivalent in most Asian countries, and whose practical value is beginning to be recognized by interlocutors who have measured, to their cost, the limits of unregulated deployment.

The question of human supervision in the medical decision loop (that the CCNE, European learned societies and the ESTRO-AAPM recommendations on AI in radiotherapy have contributed to theorizing and operationalizing) is precisely one of those on which several Asian systems are today seeking to consolidate their framework. The definition of what it means concretely to "maintain a doctor responsible for the decision" in an environment where AI has produced the diagnosis, structured the therapeutic options and calculated the treatment plan is not a trivial question. The European response constitutes an exportable intellectual contribution.

Similarly, the European requirement of rigorous clinical validation (prospective studies, comparisons to the state of the art, measurement of impact in real populations) is a standard whose international diffusion is in the interest of all patients. An AI diagnostic tool validated on a retrospective homogeneous cohort of a single Asian country is not necessarily performant on European or African populations with different genetic, epidemiological and socio-economic profiles.

Finally, France possesses a specific institutional asset: its capacity to articulate, in the same space of reflection, medical, scientific, ethical, industrial and citizen voices. The Health Data Hub, the CCNE, evaluation agencies, learned societies and patient associations constitute, when they function in synergy, a digital health governance ecosystem whose sophistication is rare at the global scale. Mobilizing it explicitly within the framework of an active digital health diplomacy (in the direction of Asian partners, in the arenas of the WHO, in the European Union-ASEAN discussions) would be both coherent with the values that France defends and strategically pertinent for its international influence.

5.3. The Irreplaceable Role of Dialogue Spaces

The transformations described in this article will not occur by the sole virtue of public policies nor by the sole dynamic of markets. They suppose spaces of encounter between actors who, left to their sole institutional logics, would have few occasions to speak to one another: an oncologist clinician from Tokyo and an AI ethics researcher from Lyon, a hospital director from Singapore and a digital policy official from the French Ministry of Health, a Korean healthtech entrepreneur and a representative of French patient associations.

This is precisely the function fulfilled by the foundations and Young Leader networks that have irrigated this Franco-Asian dialogue for years: creating the conditions for encounters that, without them, would not take place, and in which the mutual understandings from which concrete cooperations become possible are forged. At a time when AI is reshaping the balances of global health power (where the capacity of a country to care effectively for its population is also becoming a question of digital sovereignty and industrial competitiveness), these dialogue spaces are not a cultural luxury. They are a strategic infrastructure.

Conclusion: adaptation as a collective competence

Artificial intelligence will not transform healthcare systems by the sole force of its algorithms. It will transform them (or will not transform them) according to the capacity of the organizations that compose them to learn, to adjust, to maintain the trust of those who care as well as those who are cared for.

What the observation of Asian experiences reveals, at bottom, is that adaptation is not a state that one reaches: it is a competence that one cultivates. A competence that is built in the data infrastructures that one chooses to finance, in the training programs that one decides to make mandatory, in the governance architectures that one takes the trouble to design, in the international dialogues that one takes seriously rather than treating them as communication exercises.

Asia does not offer a model to copy. It offers a set of experiences to read with care, in their diversity and in their contradictions, to extract from them what is transposable, to criticize what is not, and to identify what Europe (strong in its own values and its own achievements) can bring to a conversation whose stakes, in fine, are universal: using artificial intelligence to care better, to care more equitably, and to care while preserving what makes the heart of medicine, that is to say the relationship between a human being who suffers and another who commits to helping them.

*****

Jean-Emmanuel Bibault is a University Professor-Hospital Practitioner in radiotherapy oncology at the Hôpital Européen Georges-Pompidou (Université Paris Cité) and a researcher affiliated with INSERM UMR 1138. He is also a member of the National Consultative Ethics Committee (CCNE) since 2025. An alumnus of the Stanford AI for Medicine Lab, he is co-author of the ESTRO-AAPM recommendations on artificial intelligence in radiotherapy and author of 2041, L'Odyssée de la Médecine. His work focuses on the clinical integration of AI, decisional agents in oncology and the governance of AI systems in health.

This publication reflects the views and opinions of the individual authors. As a platform dedicated to the sharing of information and ideas, our objective is to highlight a diversity of perspectives. Accordingly, the opinions expressed herein should not be interpreted as those of the Fondation France-Asie or its affiliates.