Interview with Professor Idris Guessous, Vice-Dean of the Faculty of Medicine at the
University of Geneva and Head Physician of the Primary Care Medicine Department at HUG.
Conducted by Boris Gojanovic on December 9th, 2025.
Recorded in French on Zoom, transcribed with Otter AI, translated by ChatGPT 5.1.
Professor Idris Guessous obtained a medical degree in Lausanne in 2001, which he supplemented with clinical and epidemiological training. He specialized in general internal medicine in Lausanne and Geneva and obtained a PhD in epidemiology at Emory University (United States).
He joined the HUG in 2009 as head of the Population Epidemiology Unit, where he worked on the genetic and environmental determinants of health. In 2018, he became Medical Director of the HUG’s Primary Care Department.
Idris Guessous conducts and participates in numerous national and international population studies. He is the co-founder of the GIRAPH (geographic information for research and analysis in public health) research group, which integrates spatial analysis into the field of medicine and population health.
He is behind numerous university hospital projects, including Pro-HUG, @choum, SeroCov, SPECCHIO, and RAFAEL (the first medical chatbot at the HUG). In 2023, he develops “confIAnce”, the first chatbot for primary care, based on artificial intelligence technologies, to provide medical information accessible 24/7 to patients and healthcare professionals.
He is the principal investigator of the European multicentre clinical trial GNC-501, which aims to evaluate the efficacy of treatments in people affected by post-COVID. He is a member of the executive committee of the Swiss Society of General Internal Medicine, academic editor of the Swiss Medical Weekly and a member of the editorial board of the Swiss Medical Forum.
He is the president of the Health & Happiness Foundation. He was appointed associate professor in the Department of Community Health and Medicine in 2018. Since July 2023, he is Vice-Dean of the Faculty of Medicine at UNIGE, responsible for postgraduate and continuing education and professional identity. He is also the academic director of the HUG Innovation Centre.
Hello Idris, could you briefly describe your training background and your current role in primary care medicine?
I passed my final medical exams in 2001 at the Faculty of Biology and Medicine of the University of Lausanne. I began my medical career on the outskirts, in Monthey, during the winter season. I wanted to start immediately after my finals, eager to learn how to treat a dislocated shoulder or repair a broken bone. During the first week, you wonder how you’ll ever manage to do this as a lifelong career—it feels overwhelming. By the second week, things are already easier: you manage patients’ problems and schedules better. By the third, you start to feel at home.
In the early years, you also meet people who influence you. I had inspiring encounters with orthopedic surgeons and would have loved to pursue that path. At the same time, you meet other physicians who help steer you away from certain specialties—not because of the specialty itself, but because of the people you encounter.
I returned to the CHUV and continued my training in primary care medicine. I wanted to become a general practitioner, at a time when one almost felt obliged to apologize for not being a specialist. That changed the day a book confirmed my intuition: Range by David Epstein1. The author pointed out that there are real strengths in being a “generalist,” and that generalism would prevail in the age of information and technology. It reinforced my desire not to zoom in too narrowly, but to keep the ability to see things broadly and with perspective. There are really different types of doctors: those who want to master a single enzyme or protein—and that is extremely demanding—and others who enjoy maintaining the freedom to jump from one topic to another and address very contemporary problems.
I found in primary care both the physician’s role in patient care and the realization that it might not be enough. What was missing was the broader, population-based approach. To complete my training, I turned to epidemiology. There, new role models and professors—some of whom believed in me—told me, or made me believe, that I had talent and should go for it. So I left for the United States with my wife, who was eight months pregnant at the time. We went to Atlanta (Emory University) and spent four years there completing a PhD and having two children before returning to Switzerland and eventually raising a family of four children. In 2018, I was entrusted with leading the primary care medicine department—a responsibility I carry with great motivation and a strong sense of privilege, because our profession is truly magnificent. It is also a profession that comes with obligations: to read, to learn, and to create.
You have taken a strong interest in community and population medicine, notably through the interdisciplinary GIRAPH project. Could you tell us about it?
At the core, I felt that medicine had become very precise with the emergence of evidence-based medicine. I was proactive in precision medicine. We were studying genetic sequences and variants down to a very specific level. Yet public health—what I quickly chose to call population health—remained based on a very coarse approach. Health advice was sprinkled here and there without real precision.
I often cite the example of public posters encouraging people everywhere to “move smart, eat balanced,” and all other similar slogans. It felt strange to promote highly personalized, cutting-edge medicine, while population medicine—whose messages should be crucial—remained so generic.
A professor once told me that public health messages had to be ultra-clear and without nuance. For example, for women over 50: breast cancer screening. If you add nuance, people get confused. That always felt wrong to me. Perhaps it was true at one time, but not anymore. There are many other risk factors, motivational levers, fears, and barriers. So I started looking at people’s living environments.
In the background, I have always been fascinated by environments that bring together people from different backgrounds—schools, for example. Public schools are extraordinary because they bring together families and children who live very different lives. You share the same classroom, but depending on the neighborhood you come from, life is very different. That left a deep impression on me.
These differences operate on many levels, including health. I felt that living environments strongly shape health behaviors. I then met a colleague from EPFL who was doing spatial genetic analysis of sheep in Africa to see which sheep reproduced best. I thought that if you replace sheep at the macro level with human populations, you might observe similar mimetic behaviors as in sheep herds.
We developed a spatial analysis methodology (GIRAPH) at the “individual sheep” level. I was not interested in aggregated data comparing Geneva with Zurich. The idea was to see how things look for Boris, with his address and neighborhood, compared with Idris, with another address and another neighborhood—while still applying the logic of precision medicine, down to the genetic variant. It was unbearable to me to accept ultra-precision in medicine while relying on coarse aggregation in population health.
We produced initial maps and showed them to politicians, revealing clear geographic clusters of health parameters. For a long time, spatial data on health determinants did not resonate with policymakers. Then one day, you meet one or two politicians whose eyes light up as yours do, and who see what can be done with precision population health for greater efficiency: if area A has problem A, let’s bring solution A’, and if area B has a different problem, let’s bring solution B’. All choices justified by science. One example: by identifying geographic clusters of high noise exposure, we prioritized the installation of sound-absorbing barriers in those neighborhoods. That allows investments to be justified rather than applying community-wide measures everywhere indiscriminately.
As with any new approach, skepticism was present. When we mapped obesity prevalence, some people said, “Where are these obese people? We don’t see them.” Others said we were not showing anything new—that neighborhood differences were already well known, such as between Geneva’s right and left banks.
Sometimes science must also push open doors that are already ajar. When you push open a door that people refused to visualize methodically, a new reality appears. Our maps revealed clusters that could no longer be ignored. And this clarity also included nuance among different health determinants. We liked that. Our approach was mixed: partly “macro,” partly ultra-precise, “ultra-micro.”
A word that emerges from many of your projects is innovation. What does innovation mean from your perspective as a healthcare professional?
For me, it boils down to one word: creativity. It only becomes “innovation” in the public mind when it works, but many ideas never reach that stage. Innovation always begins with creativity—the marriage of intelligence and imagination. That requires learning first, through reading and study, and then adding imagination.
Take this issue of your sports and exercise medicine journal: nobody asked you to create this new hybrid approach combining artificial intelligence, literature, and medical review. You imagined it, using intelligence to reach creativity —and creativity is the engine of innovation.
Whether through new internal communication tools, reorganizing management, establishing a new culture with a small touch of imagination—this is what gives me immense pleasure. I have the privilege of working in an institution that values this, not commercially but in terms of overall recognition of a department. Sometimes you must fight for ideas.
Just before COVID, I struggled to convince administration to invest in webcams for meeting rooms. A few weeks later, everyone switched to videoconferencing. We were fully ready and the first department to use Teams.
You are also the academic lead of the HUG Innovation Center. What is its mission and vision?
I often think of it as a magical place—but not a sacred one. Magical because the hospital has dedicated beautiful space with natural light to innovation, right above the intensive care unit rather than hidden in a basement.
It is not the equivalent of the flashy incubators seen elsewhere, but it is remarkable that a public institution has allocated space for this purpose. It is not sacred because it must not be elitist. HUG has 13,500 employees across all professions, most of whom are not in management. By definition, people must feel comfortable coming to the Innovation Center with an idea—an idea shaped by someone who wants to contribute to change.
Today, this is becoming a competitive advantage. Some may see it as a cost center, but its added value lies elsewhere. We must think about what keeps such a large institution alive and about avoidable costs—such as disengagement and staff turnover. The Innovation Center stimulates talent and attracts new ones. Competition is fierce, and if the institution listens to ideas and provides a place to develop them, it can progress with the best people.
The center thrives on these people and runs activities throughout the year. Recently, we held a symposium on leadership and creativity—how to motivate teams, how to inspire. Nobody wants to resemble a task force; everyone wants to follow in the footsteps of a true leader. That too is innovation.
We are here to talk about AI. Since 2023, you have developed the confIAnce project at HUG, a healthcare chatbot. How did it begin and how is it progressing?
The origin lies in multiple projects and a mindset of addressing contemporary medical questions. At one point we worked on GWAS (ndlr : genome-wide association studies). Then we built a training platform for primary care physicians in precision medicine: Frontliners.ch2.
A new challenge emerged, and we had to explore it. We would not be experts, but we would not be ignorant either. HUG is the largest training center for primary care medicine; as a university hospital, it must master new developments and stay ahead of the citizens it advises.
With AI, it was the same. It all began with Raphaël—a patient who contacted me after severe COVID in March 2020, with persistent symptoms. Treatments did not help. This observation of a new problem led us to recognize long COVID. We created a cohort and contributed to the literature showing that long COVID existed. But we knew nothing about it—the patients actually knew more than we did.
So we created a tool that would allow us not only to respond but also to collect patients’ information and questions 24/7. Raphaël was born3. It relied on early NLP (nldr: natural language processing) models with limited response algorithms, but it became the first widely used chatbot.
Later, we applied the same thinking to our extensive primary care clinical guidelines4. These high-quality, open-access resources (like you journal) are widely used in the francophone world. Yet during consultations—usually only 20 minutes—we still miss patients’ existential questions.
A great deal of knowledge has been documented, but today it cannot replace actual time with patients during consultations. In general, we have 20 minutes, which is far too little, and we end up missing the existential questions of medicine. The physician spends more time discussing the technical aspects of an ABPM (ambulatory blood pressure monitoring) than the issue of hypertension itself. The doctor leaves the consultation frustrated, just like the patient, who walks out saying, “I forgot to ask this, I forgot to ask that.” I found this quite unacceptable in 2023.
Our medical culture is not one of answering via WhatsApp or SMS, but between a consultation in February and the next one in October, there should not be a complete communication void. We created confIAnce specifically to fill that gap and to continue the conversation between two consultations5. Concretely, the physician ends the consultation by reminding the patient that it was not possible to cover everything. They summarize what has been discussed, then direct the patient to the confIAnce platform for explanations on other relevant topics. The platform will explain how ABPM works just as the physician would have done, because the model uses the same information sources.
ConfIAnce was not developed to make diagnoses or handle acute problems, nor to direct patients to emergency services, but rather to continue the conversation between two annual consultations for chronic diseases. We now have around forty chronic conditions documented. A patient with a newly diagnosed diabetes will need far more communication than what a 20-minute consultation can offer. Those 20 minutes can only skim the surface of the problem.
What is remarkable is that we have observed the outstanding quality of these models’ responses. We have other projects underway for a more advanced, more personalized chatbot whose goal will be to support patients in making behavior changes.
Have you received feedback from patients and the general public on its use?
Yes, we use several approaches: focus groups, response ratings (“likes”), comments, and so on.
I think what has worked particularly well is the model’s sensitivity, which is not always what we find in North American models. The “voice” of the model corresponds to that of our physicians within the institution that hosts the tool.
Technically, we use RAG (retrieval-augmented generation) so that the model queries our validated and regularly updated databases. If it cannot answer, it says that it did not find the information in the database and then continues by providing what it was able to find elsewhere. It knows a great deal, and I believe we have found the right balance between model power and personalization.
We now have several thousand users and nearly ten thousand conversations. However, breaking into the chatbot market remains difficult. ConfIAnce will not be the primary and essential tool that patients turn to. We are competing with giants that are extremely strong, with billions in data and investment. Much greater resources would be needed, but we are continuing our efforts.
How do you view the progress of AI and large language models since November 2022 (the launch of ChatGPT)?
What I like to emphasize is that with artificial intelligence, we are not dealing with a linear dynamic. It is an exponential one. Others have expressed this better than I have, but you can feel that people do not necessarily realize it—it’s human nature. They think we are slowly climbing a linear path, whereas in fact it is exponential. What we have seen in recent weeks (ed : interview on December 9, 2025) has almost nothing in common with the early advances since 2023.
We must prepare for this, because our speed of adaptation cannot simply be linear and incremental, taking our time to see what comes. This means that what seemed completely unimaginable yesterday becomes trivial the next day.
Some people still manage to be amazed. I personally want to remain amazed when I see what a chatbot can do, and I feel somewhat sad when others fail to recognize it. It can respond very clearly and accurately, in a fraction of a second and in ten lines, to a medical question—one for which I spend months training young physicians to achieve a comparable result. There is something truly staggering about this, and I want us to remain amazed, but not in a naïve or dazzled way, while remaining fully aware of the stakes, responsibilities, biases, and so on.
It is striking to hear some people say that AI will change nothing or that these models are completely stupid. At the same time, we see that these very same people quickly grow accustomed to what chatbots are capable of doing. They adapt very rapidly to a new technology without questioning how it is built. Take smartphones and the FaceID function—it has become trivial. Yet what lies behind it to make it work reliably is extraordinary.
More generally, how do you perceive the medical profession’s position toward generative AI today?
I think there is a healthy dose of caution, which has a certain virtue. But there is also inertia. We know that the healthcare system is far behind other sectors when it comes to adopting AI, and we understand some of the reasons. Healthcare has particular constraints, such as data protection. We are dealing with something extremely sensitive.
What troubles me is a kind of moratorium on various AI-related topics. I believe we have a moral obligation to explore—responsibly, of course—but nonetheless to explore what AI can do in the healthcare system, by testing and by placing—provocatively speaking—moratoriums on checklists rather than on projects. In Europe, we tend to want sovereignty while at the same time discouraging “doing,” discouraging project development. Instead, we favor checklists that tell us how things should be done. This is a dead end, because it means that the people who actually build things will do so without us. We must be present and active in model development—this is imperative.
We will have to go beyond mere commentary, because sovereignty cannot simply be decreed. You become sovereign through the tools that people actually adopt—it’s that simple. You do not become sovereign by writing a new checklist every week about what should or should not be done. I am stunned by the number of groups that devote themselves to drafting checklists and thereby slow creativity within institutions, especially around tools like chatbots. Personally, I now call on programmers to program, developers to develop, ethicists to help us, and for us to work as a team in an environment that tests, builds models, makes mistakes in beta versions—but does so responsibly and with risk mitigation.
I believe that by overusing these brakes on AI development, we are going to fall far behind, and the risk is that tomorrow or the day after, we will be blamed for having practiced a form of medicine that failed to equip itself with AI.
It must be remembered that medical and treatment errors are the third leading cause of death (100,000 deaths per year in the United States, perhaps 1,500 in Switzerland), that 10% of hospitalized patients experience an error or a potentially avoidable interaction. In addition, 33% of professionals across all fields are either disengaged or burned out—even in what is supposedly the most beautiful profession in the world. So we cannot claim that everything is fine. We are part of a privileged generation, in an era in which we potentially have a tool that could correct the flaws of a system—a system that costs too much and can no longer keep up with care needs, to the point that 25% of the Swiss population forgoes healthcare for economic reasons.
I believe the trend can be reversed, but I can also imagine that soon families will tell us: “Wait, are you telling me that there was no AI analyzing the interaction that my son or daughter suffered?” We must change our mindset by considering not only the risks of acting, but also the risks of not acting.
Speaking of changing mindset, is AI taught in medical school today?
As you know, curricula are difficult to change. But we are determined to integrate AI into the Faculty of Medicine at the University of Geneva, with the support of our dean.
Of course, there is resistance—voices saying that it is a passing fad, that it is nonsense. There is a kind of “chic” or “elegant” posture claiming that AI has nothing to do with human intelligence. In my view, we spend far too much time trying to decide whether ChatGPT is intelligent or not. We have probably already moved beyond that question. If we simply said “statistical learning,” everyone would agree.
This is a typically academic narrative, but one that somewhat stifles progress. Then the question of the level of teaching arises. There are two options: either we try to be very universal, or we create modules that go much deeper into AI in healthcare and medicine. These would target those who are already highly motivated or eager to become involved, to program, and so forth.
These are the two approaches. Either we try to teach everyone and see what level of competence we end up with, or we develop a more targeted track with stronger guarantees that after six years of training, we will have 20 or 30 physicians who truly have a high level of expertise. Geneva is currently very well positioned in this respect.
Finally, what is your personal and/or professional use of AI?
I think I fall into that small category of people who manage to remain amazed by today’s tools—and that is a gift.
I often wonder how people fail to realize the speed at which you can learn. It has reached such a level that I now take notes of questions to explore with AI during every reading session, professional or otherwise, so that I can return to them later and find time to question a conversational agent. This is how it has become a phenomenal source of learning for me.
It is much the same with children, and the need to find a balance. As a family, we try not to systematically turn to AI for explanations, in favor of developing natural discussions—even when verifiable questions and doubts arise. We must resist the temptation to interrupt exchanges by checking with AI. In the past, it was simple: children asked their parents when they had a doubt, and that is also something important to preserve. We tell them, “we’ll check later,” and we do not interrupt the conversation or break the dynamic.
This is a balance that I value in my private life. I use AI regularly and in many different ways, but with safeguards so as not to miss out on other fundamental aspects of life.
Idris, thank you very much for sharing your experiences, and congratulations on leading so many projects driven by innovation and creativity.
Correspondence
Idris Guessous
idris.guessous@hug.ch
Interviewer
Boris Gojanovic
Tel: +41 22 719 6363
Email: boris.gojanovic@latour.ch
References
- Epstein, D. J. (2019). Range: Why Generalists Triumph in a Specialized World. Riverhead Books. ISBN 978-0593084496.
- http://www.frontliners.ch. Accessed on December 9th 2025.
- https://www.rafael-postcovid.ch/. Accessed on December 9th 2025.
- https://www.hug.ch/medecine-premier-recours/strategies-medecine-premier-recours. Accessed on December 9th 2025.
- https://www.hug.ch/actualite/confiance-chatbot-qui-repond-vos-questions-medecine-generale. Accessed on December 9th 2025.
Comments are closed.