At 38, Aurélie Jean has already had several lives. This specialist in algorithms and numerical modeling has carried out several research projects in health, notably at the Massachusetts Institute of Technology (MIT) in Boston. Founder of the French-American company In Silico Veritas, specializing in strategic data and algorithmic consulting, she is also a partner with the French consulting company Altermind. Aurélie Jean notably published, in 2019, On the other side of the machine. A scientist’s journey to the land of algorithms. She responded in writing to our questions.
Aurélie Jean: This question is flawed, because there are many things that can help get out of the health crisis. Artificial intelligence (AI) is one of them, but it’s not the only one. Using AI, we can simulate deconfinement and reconfinement scenarios, contribute to the development of vaccines, or even improve logistics by optimizing the transport of materials, or even patients. That being said, AI alone cannot be successful without the support of medicine, this is a no brainer that should be emphasized and never forgotten.
Would artificial intelligence have had the possibility of preventing the pandemic or of predicting it? Or will she be able to do so in the future?
AI models are currently used to anticipate changes in seasonal flu each year, but this would have been almost impossible for the Covid-19 virus for several reasons. The AI works on explicit modeling components in which the phenomenon to be simulated is explicitly described, and on implicit components in which the logic is implicitly described by learning on data which represents scenarios from the past of the phenomenon to be simulated. That being said, we understand the limits of predicting Covid-19. Unlike seasonal flu, the mechanisms of transmission of the Covid-19 virus were not well understood. In addition, the Covid-19 pandemic being a rare phenomenon, there are no similar events in the past that an AI model could have learned from.
In your opinion, why applications like StopCovid in France or SwissCovid arouse so much mistrust, even though data is much better controlled by the state than by the tech giants?
The question is poorly worded. Putting the tech giants in the question is already underlining a shaky reason for the failure of StopCovid in France. The explanations are elsewhere. It would have been necessary to open the code much earlier to the developers of the open source community, to avoid the awkwardness on the collection of personal data that the National Commission of Informatics and Freedoms (CNIL) has identified, or to involve more technical talents. and individual private sector scientists in the development of such a tool. There has been a lack of clarity, pedagogy and technical strategy on the development of such a product. We have seen small players develop tools on their own without consideration from the political leaders responsible for StopCovid. There is a lack of ambition in the confidence we place in small industrial players, or in individual technical talents.
Today, how can we manage the issue of data confidentiality? And in the face of general mistrust, how can we convince people of the contribution of AI?
This mistrust and this loss of confidence are largely explained by the numerous scandals that have rocked the daily lives of individuals for more than two years, since the Cambridge Analytica affair more precisely. To regain this trust, private and public actors must explain how the tools they deploy work, engage consumers in the design of the tools (see the book The Inversion Factor), or explain the role of organizations such as the CNIL to understand the benefits. Our economic and political leaders also need to understand the role of data in AI models, or how algorithms work, in order to accurately explain technical points that are often too muddled in official speeches.
Do you think that an important risk linked to AI is not being able to explain how a machine made such and such a decision?
It is true that the level of explainability of algorithms decreases with learning techniques, unlike explicit mathematical modeling. That said, we still have a way to explain or interpret a minimum by showing for example what is the data on which the algorithm has learned, or by analyzing the algorithmic responses according to the input data. Scientists are currently working to enable learning on so-called “non-labeled” data. In other words, the algorithm would learn on raw data without quantitative description of the latter, which would therefore prevent explaining on which “quantities” the algorithm learns. This development will allow us to go even further in the abstraction of the problems to be solved and to go more quickly – the labeling of data takes a considerable time -, but it will also greatly reduce the level of explainability of the models.
What are the most interesting or useful examples of AI that are already in place?
My answer is necessarily biased because I am very interested in medicine, in which I still work today as part of a small research project. AI in medicine has a huge impact, both in diagnostics, optimization of treatments, as well as in medical examinations and the logistics of healthcare institutions such as hospitals. I have a lot of hope in AI applied to medicine, to provide better care, at the right time and everywhere in the world!
How do you define AI? We have the impression that it has become a marketing term used by many companies …
I don’t like this word (laughs …). To quote Professor Yoshua Bengio’s statement [ndlr: corécipiendaire du Prix Turing 2018], I would say that we must speak of “artificial stupidity” because we do not make the computer smarter but less stupid. I would also quote the book by my friend Luc Julia, director of research at Samsung, Artificial intelligence does not exist. I prefer to speak of “digital or computational science” to use the Anglo-Saxon term “computational science”. It is the set of computer simulation models that reproduce analytical and even mathematical reasoning mechanisms to solve a problem or answer a question.
Are you calling for global regulation of AI, or is it pointless or unnecessary?
I don’t call for anything at all (laughs …). Out of pragmatism, it seems to me difficult to conceive of a world regulation on anything… We are already not able to agree in the world on human rights! But I believe in exemplarity, the European General Data Protection Regulation (GDPR) has influenced a few regulatory texts around the world, such as the famous CCPA (California Consumer Privacy Act).
Do you share the fear that AI will be used for bad purposes, like social control, manipulation, or even exploitation?
I share the idea that any method, or tool, can be misused, regardless of the field or discipline. AI is obviously one of them. Captology algorithms, for example, can be used as much to capture your attention in an abusive way on a social network, in order to make you stay as long as possible, as in a children’s toy in order to make him take his drugs… Albert Einstein said: “Science is a powerful tool. How it is used depends on the man, not the tool. ”
There has been a lot of talk about the under-representation of women in tech for years. Is this problem in the process of being resolved?
Unfortunately not, everything is very (too) slow! At the same time, it is necessary to encourage girls to get started, to change the codes of technical and scientific ecosystems, to have dynamic quantified policies to achieve more quickly an exemplary diversity which will become its own catalyst. Many organizations and schools are actively working in this direction, but without the support of powerful men at the head of public and private institutions, this will not change. I worked for almost two years for a company whose boss is exemplary on the subject of diversity, Michael Bloomberg. A powerful, influential and modern man, who is a defender of women’s rights, equal opportunities for professional success, regardless of his gender, sexual orientation, age or even his skin color and religion. We need more Michael Bloomberg in the business world!