Science

INTERVIEW. “Barriers to AI are more human than technical” — Sciences et Avenir

After the publication of “Health and Artificial Intelligence” together with Cedric Villani in 2018, Bernard Nordlinger and he edited a new book more specifically dedicated to the application of AI in the field of medicine. In the publication “Medicine and Artificial Intelligence”, published by CNRS on March 24, 2022, more than sixty experts spoke about the main problems that will arise when using algorithms in medical practice. During a symposium at the Medical Academy on May 3, 2022, Bernard Nordlinger, Head of the Department of General Surgery and Oncology of the Digestive Tract Ambroise-Pare, researcher and member of the Academy, answered our questions.

Bernard Nordlinger. © College of Bernardines

Sciences et Avenir: How does France position itself in the development of artificial intelligence (AI) in medicine and is it possible to imagine a European or international database?

Bernard Nordlinger: France is quite well positioned in the field of artificial intelligence, this was desired by the Villani commission in 2018 and the state that allocated funds for its development. We are probably not the first country in this area, but we, for example, organized a symposium with the Massachusetts Institute of Technology (MIT), which considers us good partners and wishes to resume the meeting this year. In the healthcare field, we are fortunate to have an accessible vital map database available to researchers, subject to confidentiality. This kind of access to health data is unique in the world, others envy us. During the Covid-19 pandemic – a chapter of the book is devoted to it – artificial intelligence was not widely used, which was not the case with digital technologies that were used to create programs for predicting epidemiological evolution. It was this that allowed health authorities to anticipate the influx of patients in intensive care, the extent to which barrier gestures were used, or even manage vaccination outcomes. I think that this crisis was handled better in France than in other countries. All the information collected will be very important if, unfortunately, but it is likely that we are faced with another health crisis of this magnitude. We know that reliable databases are the basis for the development of artificial intelligence.

There is a body that tries to manage health data at a European level, but it is particularly at risk. In France, this data is collected by the National Health Data System (SNDS) from the social security service, through the vital card, and through hospitals and death registers. Increasingly, they are also stored in the Health data centre, the organization responsible for representing France at European level. To gain access, certain conditions must be met. The data must be anonymous, with a guarantee that the person cannot be re-identified. They should also be used judiciously and not used, for example, by insurance companies to adjust premiums or premiums according to risk, or to direct medical visitors to places where pharmaceutical companies believe this is for marketing purposes. Obviously, this development should be adapted at the European level, but certain rules make international cooperation very difficult, especially with the United States, which discourages research laboratories. However, the Covid-19 epidemic did not stop at the borders of France or Europe, and international cooperation was required to develop effective vaccines.

Sciences et Avenir: Have some medical disciplines been affected by the emergence of artificial intelligence more than others, and what are the main problems associated with its application in practice with medical personnel?

Bernard Nordlinger: These applications are being developed gradually and are not currently in practice. According to some estimates, in 10-15 years the use of artificial intelligence will become effective. Of course, there are areas where the contribution of digital technologies and AI is more obvious, in others less, but eventually it will affect all medical disciplines. Artificial intelligence can quickly manage huge amounts of data much more efficiently than humans if we know how to use it. The first area in which he will provide significant assistance is image processing, of course, radiology, as well as dermatology, pathological anatomy and even ophthalmology. Other disciplines less known to the general public are also touched upon, such as psychiatry, where artificial intelligence is beginning to take an important place. Machines analyze images much more precisely than radiologists can, where sometimes some elements of a few pixels are too small to be clearly identified with the naked eye.

But, a priori, and Cedric Villani recalled this during the conference, the obstacles to the progress of artificial intelligence in medicine are more human than technical. Physicians will need to be trained and consider AI as an aid in their daily practice to support them in their profession with a wealth of information at their fingertips without fear of their work being taken away from them. This is especially true for radiologists, who will have to work closely with artificial intelligence and train with this tool. This training will also apply to general practitioners. In China, some places where there is a shortage of doctors have apps that can refer patients based on the severity of symptoms. In France we will see the emergence of this type of diagnostic help application, of course with a person who confirms or does not confirm this opinion. But the goal is not to replace the doctor with an algorithm that feeds his symptoms in order to make a diagnosis. When we are sick, we need a person to explain the decision and let us know, especially if it is a serious illness.

There have been attempts to apply on the topic of cancer using the IBM Watson Health program, but they have not worked very well. The goal was to help diagnose cancer, and several American and French hospitals bought these algorithms, but they turned out to be not very effective, in part because the data on which the AI ​​was trained came from only one medical center. But other attempts like that will see the light, Internet giants are on the move, for example the American group Mayo Clinic, which offered to work with Google. Health today is one of the main areas of development of AI, we will see the emergence of various applications that will undoubtedly work more or less well. However, we reminded in the book that artificial intelligence should not be fantasized as the solution to everything. The term itself is not very suitable, and as a result it is scary. It’s a bad name that makes people think that we’re going to steal their brains, that the computer will take over our decisions. Artificial intelligence exists not to deprive people of their intelligence, but to make us more efficient.

Sciences et Avenir: Given the many questions that artificial intelligence poses from an ethical perspective, how can we guarantee a relationship of trust with artificial intelligence and with the companies that develop it, and ensure that data is not manipulated for bad purposes? ?

Bernard Nordlinger: One can imagine a situation where a general practitioner working with an algorithm agrees with the diagnosis proposed by the machine. In this case, the doctor explains to the patient why he agrees and confirms the AI ​​proposal, which implies that the doctor is trained to understand how it works. But people ask the question, “What happens if the doctor doesn’t agree with the algorithm?” So far this has not happened, but these issues will have to be raised gradually. Generally speaking, when a doctor uses artificial intelligence, it is he who is responsible for the resulting medico-legal problems, except in cases of a characteristic malfunction of the device.

Hospitals are also prime targets for cyberattacks that hold medical data hostage for ransom or steal it to sell to private labs. There is also the more insidious risk of manipulating the brain with data or hacking the brain to force people to change their judgments by evaluating some data at the expense of others. An example that is known to many is when we are asked to accept cookies when we visit a website. The “accept cookies” button is highlighted much better than the “continue without accepting” button that helps you make a decision. If you want to be a “citizen of new technologies”, you must be aware of what they are, know the risks and know what you can get from them. Therefore, it is important to explain these issues in order to gain the trust of medical professionals as well as the citizens who benefit from these AIs.

Unfortunately, the attitude of the media towards artificial intelligence sometimes inspires fear, and does not provide objective information. In this book, we tried to be reasonable and called on the best French experts in their field to show where the research is, what the risks are, and how to use it to your advantage. Our goal is to explain and inform in order to gain people’s trust. I asked all authors that the texts be not only perfect for specialists, but also understandable in their basic principles to the public. We’re not asking people to read in one go, but those interested in the topic of trust, for example, might want to read sociologist Gerald Bronner’s chapter on trust and AI’s sense of dispossession. There are syntheses on visualization or cancer. Once again, the goal is to explain that AI is a tool and that you should know how to use it.

Back to top button