Startups around the world are competing for his young graduates. But Thierry Coulhon, president of the Paris Polytechnic Institute, which brings together six of the most prestigious engineering schools (Ecole Polytechnique, Ensta, Ponts et Chaussées, ENSAE, Télécom Paris and Telecom Sudparis), does not want to fall asleep on its laurels. Faced with the dazzling progress of the United States and China, he advocates “optimistic concern”. According to him, there is “no reason to think that we could not compete”. But we must not satisfy ourselves with our current place. Interview.
L’Express: the polytechnic ratio in start-ups generative is high. What distinguishes them in this area?
Thierry Coulhon: The creation of the School was the subject of a great debate between Monge and Laplace. The first thought that a school had to be created for the applied sciences, the second for high sciences. It was Laplace who won. Hence the cliché today of the polytechnician in the moon which cannot unscrew a bulb. But the school has today completely exceeded this tension. It certainly gives more room than others to pure mathematics, statistics and data analysis – which precisely constitutes the base of artificial intelligence. But the fact that it belongs to a set of six grandes engineering schools [NDLR : les cinq autres étant l’Ensae, l’Ensta, les Ponts et Chaussées, Télécom Paris et Télécom SudParis] Who excels for example in robotics and mechanics, widens his vision beyond the fundamental sciences towards the applied sciences.
The American giants praise the talents of French AI engineers which they recruit with appetite. Is it a good thing that so many high-flying French graduates go to work abroad?
It is very healthy that they acquire an international experience. But it is important that a sufficient number returns to France, setting up their business or working for local players. This is the case. We are no longer victims of brain flight that we could still observe ten years ago. It is also crucial that we would attract brains from abroad. To promote this, we locate brilliant students who wish to make a master’s degree with us and offer them a “PHD Pack”, easy access to a scholarship for a doctorate. To date, 47 % of our doctoral students are of foreign origin.
The United States has a considerable advance in AI. With Deepseek, China seems to fill its delay at full speed. Where is France located?
At the IA summit held in Paris, we had no problem attracting the heights of higher rank from around the world. They do not come out of pure friendship, but because they know that Paris is an important place in the world scene of AI. There is no reason to lament or think that we could not compete. But we must not satisfy ourselves with our current place.
Faced with the geopolitical challenge, I advocate optimistic concern. An attitude to adopt, in general, in research. Large international institutions know that their rank is never definitively acquired. Science evolves, international politics can hit them hard, as we saw during Brexit. Up to question permanently is the essence of our profession. This is what we also transmit to our students.
The quality of French AI graduates is recognized, but we don’t form enough. How to get upside down?
Yes, you need strong actions to meet future needs. Within the Polytechnic Institute in Paris, we will multiply by three the number of doctoral students and ten the number of Undergraduates specializing in AI. And those following other types of courses will have, at least, an exposure to AI.
The United States and China have a big lead in AI. What do they do in terms of training that France should be inspired?
Before placing a rather dark geopolitical reading grid, you have to keep in mind that the academic world is a very interconnected world. The eggs of French AI have no trouble bringing people from everywhere. We do the same science, the same math. And AI remains the product of fairly pure mathematics. So this sphere remains very interconnected. To this is added an entrepreneurial geopolitics, with very deregulated American models on one side and on the other, very state Chinese models. Europe is looking for a more democratic model, and we are only at the dawn of the IA revolution. It is too early to say what side the cards will fall.
When they see the colossal means deployed by the Americans and the Chinese to accelerate in AI, do French researchers remain confident in the future?
They are generally enthusiastic: something extraordinary happens in their field. And a mistral shows that we are in the battle. Lamento is not the style of the house. An example? We often hear criticism of French centralism, but in AI, it is a considerable advantage. This offers us in particular a well -organized and well anonymized health data system, which we could take advantage of to develop expert AI in the field.
The ENSAE has been improving advanced know-how in the analysis of large databases for years. And it can rely on INSEE data for this, to which it accesses via a center specially designed to allow secure access to data. We have a tradition of state statistical services and access to very well protected and very well regulated data. With, in addition, a very strong tradition of cryptography. There remains of course the question of financial means, which are not the same as in the United States. But Mistral or Deepseek show that we can do better with less.
Science advocates the transmission of knowledge, but policies do not always have an interest in it. Does the border between what states share and what they keep for them in research has moved in recent years?
These questions arise in a slightly different way. The know-how continues to circulate. Generally, fundamental knowledge is widely shared, in conferences where you will find both Russians and Chinese. And the know-how that will be more protected is one that has an economic value. What has changed is that we go faster from the first to the second.
You said above, IA technologies are pure mathematics. However, France is poorly rated in this area. Is this a problem?
The average mathematical level does not reflect that of research. It is the latter that matters most in AI, and in this area, we start from very high. We must remain vigilant, but there is no reason to be outrageously worried.
“” “What enthuses me is the way in which AI can change teaching in France»
Regarding the teaching of this discipline at school, the issues are different. An important action to be carried out is to orient girls more to math. We must also assume our responsibilities in these dysfunctions. If the grandes écoles make it endorse to universities, which postpones it on the school, the college, and so on, we will not go. Everyone has an action to be carried out at their level.
What are the peculiarities of the French approach in mathematics?
There is a very strong tradition, an extraordinary line. What is called “the French school in mathematics” is distinguished by a very strong taste for abstraction is the influence of the Bourbaki collective. However, there was a rebalancing between applied mathematics and pure mathematics. The link between math and physics has been anchored for a long time. But discipline has opened up to others, such as biology, social sciences and engineering. She also discharged internally. Before, either you were making probabilities, or you made partial derivative equations. Today, we see that these two worlds are integrated.
Can artificial intelligence help research in mathematics?
I’m going to hide behind a big man, Terry Tao, probably the biggest living mathematician. On his bloghe describes the way he uses it and he explains that it helps him save time on a little routine things. He entrusts him with calculations he could entrust to a medium level license student. The question is to see what it will bring later. This will make the sciences deeply evolve in any case, in particular the experimental sciences, for example those where you must make calculations on millions of molecules. I remember Jean-Louis Krivine who, at the time when I started my thesis, told me that he wanted to teach a machine to make math. We are probably not very far from it. What enthuses me is the way it can change teaching in France.
That each student has the right to a particular teacher in the form of an AI is exciting, but she is sometimes mistaken. How to take advantage of this despite this big flaw?
Let’s not be naive: a probabilistic AI will, by nature, sometimes be mistaken. But it also happens to humans! The AI will help us surpass ourselves, but we must obviously not go there with your eyes closed.
.