Is it possible to regulate artificial intelligence (AI)? Which countries will become major powers tomorrow in the field of AI? Can Europe succeed in imposing quality AI faced with giants like China and the United States? As part of the sixth edition of the international SophI.A Summit from November 22-24 at Sophia Antipolis Technology Park on the French Riviera, we interviewed Olivier Humbert, a professor of medicine and head of the AI & Health Chair at the Interdisciplinary Institute for AI at the University Côte d’Azur.
MedicalExpo e-magazine: What developments have you seen in recent years regarding the development of artificial intelligence in healthcare at the SophI.A Summit?
Olivier Humbert: Firstly, we see year after year developments in terms of applications in the sense that, in the early days of AI in medicine, there were very few applications that made it to the patient’s bedside. Research is advancing very quickly, but before projects are implemented in hospitals, it takes a long time. This gap is explained by all the regulatory, ethical and clinical validation aspects that remain lengthy and cumbersome.
However, today, we observe that there are more and more AI solutions offered by companies for patient care with CE marking or others. So even though there is still a gap between research and the arrival of algorithms in hospitals, we have seen a profound evolution in this area in just four or five years.
“Another visible evolution, beyond the technical aspect, is the emergence of increasingly in-depth reflections on how we want to use AI. What kind of AI do we want to use to treat our patients? And what kind of AI do we not want?”
Another visible evolution, beyond the technical aspect, is the emergence of increasingly in-depth reflections on how we want to use AI. What kind of AI do we want to use to treat our patients? And what kind of AI do we not want? These reflections are becoming more and more present and during this 2023 edition, there will be discussions on all these ethical aspects.
On this subject, there is a lot of talk about the concept of “human guarantee,” meaning that a human—whether a doctor or a biologist—must retain control, validation and verification of what is produced by the algorithm. These are the big current topics; they may not have been as prominent three or four years ago, but today, these questions are at the heart of the European debate.
ME e-mag: Is it also about working to regulate AI? Is it possible to do so?
Olivier Humbert: Technology is advancing rapidly, tools arrive first and then we think about how to use them and what regulations to put in place. We see this today with ChatGPT, for example, which ultimately remains a societal media buzz. In healthcare, it may have specific interests, but today it is not yet fine-tuned or at the core of patient care.
There are currently AI regulations, but they must be able to adapt very quickly to technological developments and slow them down if necessary. At the SophI.A Summit, there are many presentations of technological innovations, both from academic research and research in companies, with a strong focus on ethics this year.
We have a presentation from a Frenchman heavily involved in the European program, David Gruson, founder of Ethik-AI, who is thinking about the rules to apply. All of his proposals have greatly influenced the emerging European proposal: the EU AI Act. This regulatory project concerns how the EU will invest in and regulate artificial intelligences, especially those at high risk, including health and medicine. The aim is to reach an agreement by the end of this year.
ME e-mag: The Sophia Summit also involves speakers from outside Europe with a focus on India this year. Why did you choose this country?
Olivier Humbert: The spotlight is on India this year because the University Côte d’Azur has established a partnership with several major Indian universities, including the University of New Delhi. These are universities that not only have a strong focus on research in technologies such as AI but also experience significant growth and funding. They are emerging leaders at a rapid pace.
India is poised to be one of the major powers in the field of AI that we’ll need to reckon with in the future. It is an immense country with substantial investments where cutting-edge research is rapidly emerging. For instance, the University of New Delhi is particularly high-tech.
For us, this scientific and educational partnership is very interesting because the work they do in India complements what we do here. When working with AI, access to large databases of populations is crucial. We would like to study the commonalities and differences among these populations.
For example, in India, they focus a lot on significant public health issues such as access to primary care (emergencies, general practitioners) or in the context of infectious diseases. These might not necessarily be the themes prevalent in Europe, where we focus more on cancer-related issues. So, the objective of the partnership is to compare these themes, the various approaches and see what we can achieve together.
ME e-mag: AI knows no borders, it is everywhere. Is there international competition?
Olivier Humbert: Yes, research in AI involves both collaborations and competitions. If in Europe, especially in medicine, we want to differentiate ourselves with ethical AI that respects patients’ rights, we need to be able to propose and push for suitable AI models.
There is currently in Europe intense competition with other major AI leaders, notably China and the United States, to develop the applications we want, the ones we need, without necessarily just accepting what comes from abroad. Europe has a significant role in structuring itself at this level.
“There is currently in Europe intense competition with other major AI leaders, notably China and the United States, to develop the applications we want, the ones we need, without necessarily just accepting what comes from abroad.”
ME e-mag: There are also significant financial stakes involved…
Olivier Humbert: Yes, everyone is getting into AI and when we see the financial power that companies like Google or others of this kind have, it’s undoubtedly an unmatched firepower. But I would say there are also quality issues with AI because an AI trained on American or Asian populations may not necessarily perform the same on European populations and vice versa.
That’s why we want to develop, verify and validate AI in Europe for our populations but also collaborate with other countries to create AIs that are fair and equitable for different global populations.
As part of our partnership with India, we are trying to develop algorithms on European databases and see how these algorithms work on Indian patient databases. Will the algorithms have the same performance, or, on the contrary, will they show many more failures because these are different populations and diseases, and the algorithms are not necessarily generalizable?
The generalizability aspect of AI is crucial today; it’s a significant technical challenge.
ME e-mag: Would quality AI developed by Europe have a chance to compete against the AIs of giants like the United States and China?
Olivier Humbert: Yes, I truly believe so because. At the European level, we are currently undergoing a genuine reflection and structuring in the field of healthcare. For an AI to be good and effective, it is necessary to have high-quality data. It’s not just about quantity.
While China is often mentioned for its easily accessible data due to its population size, quantity in AI is not everything; quality matters too. Therefore, it is essential to establish, and currently, this is not happening enough, consortiums involving AI companies and researchers, as well as healthcare professionals. This collaboration will enhance the performance and quality of algorithms.
There is a real challenge, and ultimately, China and the United States are not more advanced than us in structuring these networks. So, nothing is lost; it’s not just about money, but also about interdisciplinary structuring and Europe must stay actively involved in this.
“Quantity in AI is not everything; quality matters too. Therefore, it is essential to establish consortiums involving AI companies and researchers, as well as healthcare professionals. This collaboration will enhance the performance and quality of algorithms.”
ME e-mag: Have you observed a change in attitudes towards AI in recent years?
Olivier Humbert: I am responsible for teaching AI to healthcare professionals and in society in general and in medicine in particular, we see a clear evolution. On the patient side, we have always been able to rely on them for research, especially in oncology. Even in difficult situations, they are always willing to contribute to the advancement of projects by providing their data; they do not block access and are willing participants. This is a constant an important point to mention.
Then, at the level of healthcare professionals, we see a real evolution. Three or four years ago, they were still in a kind of fear of being replaced by AI due to a strong lack of understanding of the subject. Today, training programs are being established to educate healthcare professionals about AI.
This is leading to the emergence of a new generation of leading physicians who understand how AI works, have no fear of it, are involved in research projects and see the tremendous potential it can generate to better treat patients. But on the condition that humans are still involved in the process and verify whether what is done is truly for the benefit of the patient, for aiding doctors and not for misguided AI purposes. So, doctors have understood the potential and power of AI to improve care.
“The third phase will be to say: AI is wonderful but there are risks with biases. Therefore, there is a strong need to communicate, not to step back, but to better channel things. While AI has many strengths, it is also crucial to remind people of its weaknesses. We are gradually moving towards this third phase, which is essentially a realistic phase.”
There is also a great need for healthcare professionals and doctors to be involved in these projects, whether in academic research or in companies, to ensure the proper use of AI and avoid misuse that harms everyone. Although there is still much information and training work to be done, this aspect is well understood today compared to four or five years ago. I think this trend holds true beyond Europe.
Moreover, with a shortage of doctors today and many medical deserts, an increasing number of people understand that AI can offer solutions. So, there was an initial stage of misunderstanding, lack of knowledge and fear. Today we are in a second stage where we are dispelling these fears and highlighting the potential of AI to help doctors and address real public health and patient care challenges.
The third phase will be to say: AI is wonderful, but there are risks with biases. Therefore, there is a strong need to communicate, not to step back, but to better channel things. While AI has many strengths, it is also crucial to remind people of its weaknesses. Knowing the weaknesses allows us to anticipate them and be vigilant. We are gradually moving towards this third phase, which is essentially a realistic phase.
