Cédric Villani on AI, Politics, and Europe's Missed Chances: A Conversation with Takahiro Anno in Paris
安野貴博の自由研究
Takahiro Anno is an engineer and founder of the Japanese political party Team Mirai. In Paris he met Cédric Villani, a Fields Medal–winning mathematician who served in the French National Assembly and in 2018 wrote the report that shaped France's national AI strategy. Anno came with two kinds of questions. The first was practical: how a very young party should handle its growth in parliament. The second was about AI, both as a tool for making policy and as a force that is reshaping science, the economy, and geopolitics.
Villani's positions run through the whole conversation. He thinks AI in politics has so far mostly served populists, surveillance states, and militaries. He believes the concentration of AI power in a few American companies is the most urgent problem. And he sees Europe's answer in its own strength, built on open models and coordinated public investment. He was also frank about the problems he does not know how to solve.
A personal opening: Japan, anime, and Birth of a Theorem
Villani began by describing his ties to Japan. He estimates he has made about ten trips there, visiting Tokyo and Kyoto as well as Hokkaido, Shikoku, and the "artist islands" around Teshima. He said he had seen most of the classic Japanese films and read most of the manga classics, and that his children grew up on Tezuka. Most French people know Miyazaki, he noted, but he was an "enormous fan" of Satoshi Kon, the director of Perfect Blue, and had seen all his films. Paprika was his favorite, partly because it appeals to someone interested in AI. He called Kon's early death a tragic loss.
He also mentioned that his book Birth of a Theorem had been translated into Japanese. He regretted not bringing a copy and promised to send one. In his view the Japanese edition is one of the most beautiful of all the international editions. It includes an afterword by an older Japanese mathematician who was his friend and who died soon after writing it, which Villani said makes it a very personal edition.
Comparing parliamentary trajectories: from a 300-member majority to sitting alone
Turning to politics, Villani asked about Anno's party. Anno said it was one year old, had been through two national elections, and that he sits alone in the upper house without joining a coalition. In January the party won 11 seats in the lower house. The figure shifted between 10 and 12 at different points in the conversation, but 11 was the number Anno confirmed. That is enough to form an independent parliamentary group. Anno described this as a historical record in Japanese politics for a one-year-old party, especially since the party was built from scratch rather than splitting off from an existing one. He said this cuts both ways. There is no organizational base, but the party has "free hands" to decide things for itself.
Villani said he knows what sitting alone feels like. Within a single term he went through three situations. For about two and a half years he sat in the majority group, which had "300 something" members. For roughly six months he belonged to a much smaller dissident group of about 17 people. When that group lost members and fell below 15, the minimum in the French assembly, it dissolved. About ten of its members, Villani among them, chose to sit on their own for the last two years of the mandate. So, he said, he had experienced a very big group, a very small group, and no group at all. Anno replied that he has never been in a big group, but has gone from being alone in the upper house to effectively leading a group of eleven.
The hardest problem for a new party: having a position on everything
Villani's main advice came from what he had seen of new movements entering parliament. A movement or NGO can focus on one subject. A party, he argued, is expected to have an opinion on every subject, including foreign relations, immigration, and the death penalty. Once a movement reaches parliament, this is often where it discovers internal rifts.
He offered Macron's party as an extreme case. He recalled that it had about 300 deputies, many with no prior political experience. All of them had been carried along by Macron's brilliant campaign and believed the program would be what they had dreamed of. In practice, Villani said, the group held very different visions, from very progressive to very conservative. There was no real philosophical consensus about what the party was, and it became "a big mess." In his view, even today the president's party is the only major French party without a clear political vision. With the other parties, people know what they like and what they don't.
The Japanese side said Team Mirai faces exactly this tension. They describe the party as "multi-issue" but not a "full package." It works on children, labor, and education, which are things focused on the future, but takes no positions on foreign relations, constitutional questions, or immigration. On issues outside its core, members are free to vote as they choose. Much of the criticism the party receives, they said, is precisely that it is not a full package.
Villani said a freedom of vote on particular sensitive issues is "completely okay" and that he has seen other parties do it. But he said it should be restricted to a well-identified core. He added a caveat: if people ever expect the party to join a governing coalition, it will need positions on everything, because it will have to decide whether to support the government on every issue.
He described his own experience. When he entered parliament he was confident on science and fairly solid on international affairs, because he had worked with European think tanks and calls himself a European federalist. On many other subjects he found himself wondering what he should even think. At first, he said, one can tell the media "we are not yet prepared," but after a while the media are no longer satisfied with that.
"I don't know" versus the media's demand for certainty
Anno said that, in his personal view, a politician should say "I don't know" when they genuinely don't know. Villani agreed in theory and said he occasionally did it early on. But he argued that in today's world of media and image, saying "I don't know" too often "just doesn't work." He found it paradoxical that people who are firmly convinced of something and change their opinion many times cause less trouble than people who have no opinion.
His extreme example was Donald Trump. Villani described him as changing positions every month, always claiming he had held his current view all along, and presenting himself as the world's greatest expert on taxes, war, economics, and, Villani joked, probably mathematics too. Somehow, Villani observed, this looks more reassuring to many people.
His practical advice was to work seriously on the issues inside the party. That means inviting experts with different points of view and taking the time to build a complete program. The party may lose a few people who decide "this was not my project," but Villani's guess is that clarity will attract more supporters and make the party more stable.
Team Mirai's approach: an in-house engineering team and open-source policymaking
Anno then explained what he sees as a distinctive feature of his party. It has an in-house software engineering team that builds platforms to gather opinions from the general public, process them, and turn them into policy proposals. He said the party is interested in AI in two ways: as a tool politicians can use to make better policy, and as a force that will radically change society. He asked Villani in particular about AI's effect on science, mentioning the news that week about Navier–Stokes.
Villani's warning: AI in politics has so far served populists, control, and war
Before addressing science, Villani said he wanted to make "another comment beforehand." He argued that politicians have been interested in using AI for politics for a long time, but mainly on the populist right. He said the first to do this systematically were Italy's Five Star Movement. He recommended a book by an Italian political scientist named Giuliano, published in French as Les Ingénieurs du chaos (The Engineers of Chaos), which he thought was written around five years ago. According to Villani, the book traces how this kind of AI-driven politics was first tried by Five Star, then taken up by Orbán in Hungary, by Polish politicians, and by Trump and his teams.
Villani summarized the logic as he understood it from the book. At some point the Five Star strategists stopped asking what they wanted to say and started asking what people wanted to hear so that they would get elected. AI would tell them, and they would adapt. In Hungary, he said, they became experts at this: each morning the AI says what to say, and the politician says it. He called it "computerized politics."
Villani said he knows of no example of a progressive party that has used AI successfully. He named two other areas where AI has succeeded. One is controlling citizens, as in China. The other is war. He mentioned a film released that morning about how the Israeli army used AI to target the homes and families of people it wanted to kill in Gaza. He said the film features people from the army speaking, and that it shows up to 500 civilians being killed to hit a single target. His conclusion: AI in politics has so far been very successful for populist parties, for controlling populations, and for waging war. "We are still waiting about the very successful AI to do the good of people."
Anno's response: transparency and reviewability as the difference
Anno called this a good point. He answered that his party wants the process to be transparent and reviewable. It open-sources its software and publishes the prompts and models it uses, so that anyone can, technically, propose changes or submit a pull request. The party does not want to become authoritarian through its use of AI. Anno's hypothesis is that AI can be used in a more democratic way and can strengthen democracy. He stressed that nobody yet knows the right method, so many trials are needed.
Villani recalled that a few years ago France also had a burst of activity among small startups looking for new democratic tools, such as new ways to poll, to make suggestions, or to evaluate politicians. That "agitation," he said, has largely died down, and the focus is now on enormous AI projects. Asked why, he pointed to funding. Money shifted from being spread widely to being concentrated on very big players. Competition became intense, and a process of enormous selection left only the biggest survivors.
From science to politics to economics: the scale of the AI race
Villani gave a deliberately caricatured history of AI. Ten years ago the field was dominated by science, five years ago by politics, and now by economics: "who gets the billions." He added that AI now reaches both the "true economy," such as energy, and what he called the "magic economy" of finance.
He gave some numbers to show the scale. According to him, each of the super-large American AI companies is spending about half a billion dollars per day on AI. He compared this to the cost of a real war, saying the war in Ukraine runs at about half a billion to a billion per day, and called these levels unprecedented. He also said that last year, for the first time, Google announced it was losing some of its cash. Before that, he said, it had more money than it knew how to spend, and now it is running short because of what the race requires.
Science and the "two problems"
On science itself, Villani made one concrete prediction. The International Congress of Mathematicians meets every four years, and he said it is quite possible that 2026 was the last congress "of a classical type," with all future ones deeply shaped by AI. That is the first problem: the work of scientists is being completely changed.
He considers the second problem more urgent. The key actors are a handful of "super huge crazy capitalist companies" worth hundreds of billions and, in his words, under the control of a president who is "a madman." Even if everything else changes, he said, we should at least keep some control. He believes this can only happen if many forces come together around open source and transparency to build very powerful shared packages.
He cited Yann LeCun, whom he called a French superstar. Villani noted that LeCun left Meta and started a French company. He said LeCun has consistently argued that the best LLMs will eventually be the open ones, because they will not be distorted by censorship, reputational fears, and similar pressures, and will be controlled by a community, much as Wikipedia is by and large the best encyclopedia available. Villani said this may be true. But the private forces now in play are so strong that it is hard to see how an open sector could raise comparable investment.
He then gave a small personal observation. According to him, last year Wikipedia saw its request rate fall for the first time in its history. On his Android phone, he said, the device now pushes him toward Gemini's answer before Wikipedia's, whereas a year or two ago the Wikipedia result would have come first.
If he were president: a European answer
The Japanese side asked what Villani would do if he were president, in any field: science, education, or the labor market. He began by saying he disagrees with Macron on many things, but on this issue he thinks Macron's overall vision and policy went in the right direction.
He stressed that the answer depends heavily on a country's geopolitical and economic situation, and said his answer is a European one. The priority, as he sees it, is to strengthen the European Union in AI. That means building on its human resources, skills, and scientific networks so the whole region becomes bigger and stronger in economics, talent, public relations, and training. It is an "ecosystem" effort that also reinforces the open sector.
Within the EU he named France as strong in AI and called Poland "the other giant." He said young Polish programmers are among the best in the world and the Polish digital economy is growing very fast, which makes Poland an important ally. He extended this to the broader Slavic culture of Eastern Europe. He also said Europe lost the UK, which he described as probably better than France in AI overall, but whose ties to the United States "destroy the whole thing."
The single biggest regret: Brexit and DeepMind
Villani said his "single biggest regret" in this whole area is the combination of Brexit and Google's acquisition of DeepMind. If the UK had stayed in Europe, and if Europe had funded DeepMind at the level it deserved, then DeepMind might be European. In his view it is the actor that, until very recently, had the strongest record in science. He asked the listener to imagine how different the geopolitics would be. Instead, he said, Demis Hassabis now stands next to Donald Trump at the G7. "What a pity," he said. The solution, he argued, has to start from politics and from Europe's own internal forces.
The 2018 report: what it achieved
Anno asked whether Villani's 2018 report had shaped AI policy. Villani said it had an impact. He cited independent evaluations by French oversight bodies, which he said concluded that the strategy kept France in the race, roughly in the top ten AI countries, and that without it France would clearly have fallen behind. It was still not enough, he said, to put France in the same league as the US, Canada, China, the UK, and perhaps Israel. But France remained a major player. He pointed to VivaTech, now a large international AI event in Paris, and to French companies including Mistral. He said these companies are a hundred times smaller than the giants but still in the billion-dollar league, and that they are so far the only decent LLM alternative that is neither American nor Chinese, apart from perhaps the Polish founders of ElevenLabs.
He attributed part of the report's success to its fit with Macron's identity. Macron wanted to present himself as a president engaged with the future and friendly to tech and international business. During the report process, Villani co-organized an international workshop with one of Macron's advisers, and Macron had dinner directly with figures including Yann LeCun and Demis Hassabis, with the whole discussion in English, which Villani called rare in the French system. Cathy O'Neil, author of Weapons of Math Destruction, whom he described as highly respected on AI ethics, was there. So were Meredith Whittaker, now president of Signal, and Kate Crawford. Villani contrasted this with Trump's dinners with big-tech CEOs, which he characterized as an "I'm the boss" display. Macron, by his account, was trying to understand: "tell me what I should know."
Villani estimated that about half of the report was put into action. He called that a very high figure, since he said parliamentary reports usually see 10% implementation, or even 0%. The public debate around the report also made him a recognized voice on AI. He said that even now he is invited to an AI debate roughly every week, largely because of it.
What worked, what didn't, and what he can't solve
According to Villani, the research system and the medical and health measures saw good progress, and a number of the main measures were carried out well. The areas least addressed, where he said the government did not really know what to do, were gender and training. Women remain heavily underrepresented in AI. And not enough people are being trained.
On training he added a caveat that he said he does not know how to handle. For the first time, programming students are having difficulty getting hired. Society still needs programmers and people who understand the whole technology, but large companies are using fewer and fewer of them. In his view this is starting to cause serious security problems, because many people are needed to check, maintain, and fix systems. The Japanese side said this was exactly the topic they most wanted to discuss. Villani repeated that it is a real, big issue and that he does not know how to handle it.
Regulation, fragmentation, and the missing European project
On regulation, Villani said he likes what Europe did with the AI Act and considers it good regulation. Implementation, however, remains a problem. He said the regulation is applied in many different national "shades," and that Mistral, for example, is being driven crazy by having to follow one interpretation in the Czech Republic, another in France, and so on. He said it needs simplification.
His biggest concern is the lack of concentrated action and strong networks. Europe could have a large open LLM project, he argued, but it would need to be a public-private partnership. In his view, fully public projects run into many difficulties, and public-private partnerships are not easy either. Public budgets are not big enough to drive such a project alone, so public money should at least serve as an incentive. He called this the most important task, and said it would require coordinated European action. That is where the conversation ended: the European open-model project he describes does not yet exist, and he still has no answer for the collapse in programmer hiring.
Brexit and DeepMind being bought up by Google.
We are still waiting about the very successful AI to do the good of people.
I think we would like to make the process transparent and reliable.
Yes. Yes.
Thank.
So you.
Stop.
Title.
AI for humanity.
It's good. You do this.
Think of this.
Yeah.
Fore.
Speech.
Very special pleasure to receive you, although my international career has led me many times to Japan.
Okay.
I think I made 10 trips to Japan or something like this. And not just the main Tokyo, Kyoto. Also was in Hokkaido. I was in Shikoku. I was in Teshima, the artist islands. I visited and attended a lot of ceremonies, trips, trains, whatever. And always a fan of the culture.
Was a big fan of the Japanese movies. I saw most of the classical theu and so on. What's the mitoi and so on. Read most of the manga classics.
Classics.
So my children grew up on Tezuka, and the more recent like...
Yeah. And a big fan of... Everybody in France knows about Miyazaki, but I also was an enormous fan of the—what was his name? The guy of Perfect Blue.
Yeah. Satoshi Kon.
Satoshi. Yeah. I saw all his movies. I love Paprika.
Paprika was my favorite. Paprika is very appealing for somebody who is interested in AI. And Paprika also, she was extraordinarily attractive.
Yes. It's crazy good.
Yeah. Yes. I thought it was a tragic loss when he died prematurely.
Yes. Unfortunately, we lose him.
And I actually, stupidly, I forgot to bring one copy, but one of my books was translated into Japanese.
Oh, really?
So I will arrange to have one sent to you afterwards. We will copy. Yeah.
It's called Birth of a Theorem, and they did beautiful—the Japanese edition actually is one of the most beautiful in all the international editions. It's stupid of me to have forgotten to bring one.
You know, Japanese publishing companies invest in their papers and front side. So I think I like Japanese publishing.
So this is the English edition, but there is a Japanese edition.
Yes, I found it. This one.
Yeah. That's the one. And they added an afterword written by a Japanese mathematician, researcher, who was my friend, quite older than me. He died just soon after writing the foreword. So it's a very personal edition, also one of the most beautiful editions all around the world.
Wow.
So back to your issue. How old is the party?
One year old.
One year old.
One year old. And we experienced two national elections, and we got 12 seats at lower house.
How many in the upper? How many in the lower?
Upper, me, just me, and lower...
It's normal. Upper is lower in the upper house.
Yeah. The member is different.
In upper house, do you sit on your own, or are you a member of a coalition? Big of a coalition?
On my own.
On your own. Very good. Yes. In parliament, I spent two years on my own, so I know what it is. I was in parliament, in lower house. For two years and a half, I was sitting in the majority group, 300-something people. Then for six months I was in a much smaller group, kind of dissident.
We were, I think, 17 or something. At the beginning, we were 17. Then we lost one, two, three people. Fifteen is the minimum in our house in France. So then we dissolved, and 10 of us decided to stay on our own. So the last two years of my mandate, I was on my own. In the same term, I knew the experience of a very big group, a very small group, and no group at all.
No group at all. Okay. Wow. For me, I have never experienced the big group one, but in this one year, at first it was just me in upper house. But we got 12 seats, 11 seats, in this January.
Eleven seats in the lower house?
In the lower house.
In a total of?
Total is 100.
One hundred. So is enough to make a group, 11?
Yes. So we can independently have a group. So I experienced kind of boss of...
This is extremely interesting. This is a lot, 11 in the lower house. So it really struck a chord in the Japanese society.
It's a historical record in Japanese political history to get 11 seats in just one year, and especially we are not diverted from the existing party. I mean, we make the party from scratch, from zero. So we don't have any organization back. That's a good and bad point. But the good point is we can decide by ourselves. We have free hands.
From experience of what I see with movements here and there, I guess that the most tricky issue for your new group will be to make sure that they can find a common stance on the various issues which are not related to your core program. I mean, political party, movement—if you are a movement, NGO, you can focus on one subject. But a party has to have an opinion on each and every subject, including...
Foreign relations.
Yeah. Immigration, death penalty, whatever, any subject. And this often is the most tricky when a movement arrives in parliament, and then they discover that there are rifts between them.
In our case, by the way, with the Macron party, it was kind of extreme. If you remember, there were 300 people there, many of them with no experience in politics before. And because Macron had made such a brilliant campaign, all of us were convinced that the program would be what we dreamed.
It turned out there were very different visions inside the group, from very progressive to very conservative, etc. And because there was no—I would say ideology, but ideology is negative—because there was no real philosophical consensus on what we are and so on, it became a big mess. Even today, the president's party is the only big party which does not have a clear political vision that is made. All the other main parties, we know what they like, what they don't like, but the president's party, nobody really knows.
It's exactly like what we do. I mean, it's multi-issue. It's not one issue. It's not just labor laws. It's multi-issue. We do child, we do labor, and we do education, so anything that's focusing on the future. But we don't do foreign relations. We don't do should Japanese be right. We don't do immigrants. So we call that multi-issue, but we're not full package.
And a lot of the negative reviews we get is that we are not full package, although we have 10 seats, we have a lot of influence. But we say for some issues, we don't put a restriction on our party. You can choose where you want for some issues. But for the issues that matter to us, we will have one big idea. But for others, we don't. People don't really...
It's completely okay if you say—some parties I've seen do that—on this and that particular sensitive issue, you have freedom of vote within the group.
Exactly.
But you should really restrict it to a core which is well identified, and have a global... Eventually, if one day people expect you to be part of a government coalition, you need to have positions on each and every subject, because then you have to decide if you support the government or not on each and every issue. This is a real thing.
I had to go through this when I came in parliament. I knew subjects related to science. I also knew international subjects not bad, because I was working in think tanks, European think tanks, etc. I'm what is called a federalist, people really thinking European construction is the most urgent thing and has to be, in many ways, transformed. But on many other subjects, I was like, what is this? What should I think of this? At the beginning, you can say in the media, we are not yet prepared, we don't know, I have to decide. But after some time, the media are not happy with that. They want you to have a position.
Yeah. Exactly. Same in Japan, and we are kind of suffering from that environment. But in my personal view, you should say, "I don't know," in front of the people when you really don't know.
And this is what also I think is, in theory, true. In the beginning, it happened to me occasionally. But in the current world of media image and so on, you say "I don't know" too many times, and it just doesn't work.
Paradoxically, when people are convinced of something and change their opinion, many times this is less problem than people having no opinion. I think the extreme case is Donald Trump. Every month he changes his position, and he says, "I always thought that," and then you put whatever you want. And then he explains to you that he's the best in the world for everything. He's the one who knows taxes best in the world. He knows war best in the world, economics, even mathematics. I think he knows probably better than me. But he always looks like... And this is somehow more reassuring for many people.
Yeah.
But my advice is, within the party—and we've been through that—to work seriously on the issues internally. Invite experts, listen to experts of different points of view, and take time to make your program so that you can complete into a full stand. And maybe you lose a few people in the process who say, "Guys, this was not my project." But the clarity will attract more and will make you more stable. That's my guess.
I see. Thank you for your advice. And one of the unique characteristics of our political party is we have an in-house software engineering team.
Oh, really?
We make software or platforms to gather notions, voices from the general public and process it, and try to make a new policy. That's our way to make a policy. And so we are interested in AI in many ways. One is how to use it as a politician to make a great policy. And at the same time, AI will change society radically. Especially, I would like to know your opinion about AI's effect on science. This week, Navier–Stokes news is catching on, so I'd like to know your view.
Yeah, I will tell you. Just another comment beforehand. It's been a long time that some politicians called the interest of AI to do politics. But so far, it has been mainly on the very right-wing side, very populist. And the first ones to do this significantly, systematically, were in Italy.
This was the Five Star Movement.
Movement. Yeah. Yes.
Maybe you read a book. There is an important book by a specialist of political sciences, Italian Giuliano. It's called Engineers of Chaos. Ingénieurs du chaos in French. I don't know the original title. It was written a few years ago, maybe five years ago or something.
And it analyzes very well how that type of AI-based politics started, was first experimented by the Five Star, then by people like Orbán in Hungary, then by the Polish people, and then also by Trump and his teams and so on. And he analyzes very well how this was. In particular, at some point the Five Star people said, "Let's not ask what do we want to say, but ask what do the people want to hear for us to be elected. And AI will tell us what the people like to hear, and we'll adapt to that," and so on.
These guys in Italy, but in Hungary, where they became experts at that, in the morning, "See, AI tells me to do this. Okay, I will say this," etc. Like computerized politics.
On the contrary, so far I don't have any example of a progressive party which was successful in using AI too. So there are many examples of successful very tough regimes, but not of progressive parties. And the other use of AI, of course, has been to better control the citizens. It's in China, or it's to control in war.
This morning there is out a movie about how the Israeli army used AI to target the homes and families of the people they wanted to shoot in Gaza, and that they had all—and it's part of the AI system. The movie shows, by the way—it's from people from the army speaking—and they show that they killed up to 500 civilians to shoot one person that they... One target.
Yeah. Yes.
Anyway, AI so far in politics, very successful for the populist party, very successful for controlling the population, and very successful to do the war. We are still waiting about the very successful AI to do the good of people.
Yeah. That's a good point. I think we would like to make the process transparent and reviewable.
Yes.
And we open-source our software and open our prompts, models, what we use. And so everybody can make a suggestion or pull request, technically. So that's how we...
Very good.
So we don't like to become authoritarian, I mean, using AI for us. But I think AI can be used in a more democratic way, so it can strengthen our democracy. That's my hypothesis. But we never know the right method yet, so we should have many trials.
Yes. A few years ago, also in France, there was a lot of activity among small startups trying to find out new democratic ways, like new ways to poll or to suggest or to evaluate the politicians, etc. And somehow all that agitation has much diminished, and now the focus is really on the big, enormous AI project. But I remember a few years ago there was a lot of agitation about what would be democracy of the future.
Why do these startup companies diminish?
I think the structure of the funding changed from spreading everything to focusing on the very big ones, and competition becoming also very tense. There was the process of selection, of enormous selection, so only the biggest survived and so on. That's what I...
So basically it's the funding problem?
The funding is enormous, enormous stuff. Maybe, I think, in terms of AI, let's caricature, saying that 10 years ago it was still very much science-dominated. Five years ago it was really politics-dominated, and now I think it's economics-dominated. It's about who gets the billions and so on. And things like now, because now it has achieved economy, both the true economy, like energy, and the magic economy, like finance, etc.
Magic economy.
Yeah, the magic economy.
Yeah.
I think right now, as we are speaking, each of the super big American AI companies is spending about half a billion per day, each of them, on AI. So each of them is spending about the budget of a war. A real war in Ukraine is about half a billion or one billion per day. They are doing this, each of them, for AI, which makes it levels never seen before.
Google, for the first time last year, announced that they lost some of their cash. Before that, they had so much cash that they didn't know how to spend it. They were giving special... Now they are lacking cash for the first time in their history, because it's so much amount which is required for the race. Yes. Then it really is... Nobody knows.
The most pressing problem maybe is the political one. A, it's a game changer, and the whole future may be changed. International Congress of Mathematicians, once every four years. It's quite possible that the 2026 was the last one of a classical type and that all the ones from now on are very much impregnated with that AI issue. So A, your job is completely changed. That's one thing.
But B, the only people, the only key people in this are a couple of super huge, crazy capitalist companies, rich by the hundreds of billions, controlled by a president who is a madman.
So, problem B may be the most urgent to solve now. At least maybe the whole thing has changed, but at least we have some control in this. And problem B can be solved only if many forces assemble together with a lot of open source and transparency to build packages which are very powerful.
People like French superstar Yann LeCun have consistently advocated in the past years for open models and said—of course, he resigned from Meta. He started his French company. He's a star in France. And consistently he said in the future the best LLM will be the open ones because they will not be marred by censorship, public reputation, damage, fear, whatever. There will be control of community, in the same way as Wikipedia is by and large the best online encyclopedia and best encyclopedia that you can find. That may be true.
But currently the private forces built up there are so strong that it's not easy to see how on earth the open sector could build up such an amount of investment. By the way, last year, for the first time in its history, Wikipedia saw its request rate go down.
That's because—
Okay. Yeah.
Even me, I can see I'm on Android, and of course always the smartphone wants to tell me to look at the answer by Gemini before Wikipedia. So it's a shift. A couple of years ago, or even one year ago, you would have had the Wikipedia answer first. Now you have the AI answer first.
I'd like to ask: you know the political field, and you are a scientist as well. So the situation is kind of crazy now. And my question is that if you were president of this country, what would you do to react to this era? Any field is okay, so as for science, education, labor market, what's on your mind?
First, I disagree with the president in many respects, but on this particular issue, I think globally his vision and politics was in the right direction. It very much depends on the situation, country and geopolitics where you are, because now it's super linked to the geopolitics and the economics. And the answer I will give you is a European answer.
The emergency is to strengthen the European Union on this and leverage on the human resources and on the skills, and then the scientist network to make the whole area get bigger, stronger in terms of economics, human resources, public relations, training. So it's ecosic. It's also reinforcement of the open sector, etc.
And when you look from the European Union, we can see France is a strong country in AI and so on. Within the European Union, maybe the other giant is Poland, which is super strong culture. The young Polish programmers are among the best in the world. The digital economy in Poland is growing like crazy and so on. So it's an important ally, and more generally the whole Slavic culture in Eastern Europe.
Unfortunately, we lost the UK in the EU, because that's also another big superpower, probably better than we are globally in AI, but with links with the United States that destroy the whole thing. And by the way, my single biggest regret in the whole area is the combination of Brexit and DeepMind being, let's say, bought up by Google.
If the UK had remained in Europe, and if Europe had funded DeepMind to the level that it deserved and continued, so possibly the actor which up to last month had the strongest record in science, DeepMind, would be European. Think of all the change in the geopolitics. Now it's different, and Demis Hassabis stands next to Donald Trump in the G7 and so on. What a pity.
So we need to use our forces in this, and the solution has to start from the politics and our internal forces. As you know, I gave a report in 2018 to the government. Yes, 2018.
Ask about that too. Do you see that it was you that affected a great much of AI politics?
Yeah, the report had an impact. There were independent evaluations of the report by our controlling body and so on. They evaluated that the strategy allowed France to remain in the race, let's say in the top 10 of AI countries, and they say it is clear that without the strategy coming out of the report, it would have fallen down.
It still was not enough to make France in the same league as the US, Canada, China, UK and maybe Israel. But France remained a big player, and VivaTech now is a big, fancy international event in Paris about AI. This is a new thing.
And we have AMI, we have Mistral, which are 100 times smaller than the big giants, but still are in the billion-dollar league, etc., talked about, and so far the only decent alternative to LLMs which is neither US nor China, maybe save the Polish people from ElevenLabs who are doing this kind of things.
Okay, so there it was really taken seriously. It went well with the identity of President Macron, who wanted to portray himself as a president working with the future, very much a friend of tech, of international business, etc. So it came well with his identity.
And when we organized the report, we also had an international workshop. And President Macron—I was organizing this with one of his advisers—so he had dinner directly with people like Yann LeCun and Demis Hassabis. He was at the table. The whole discussion was in English, which is rare in the French system.
There was Cathy O'Neil, the author of Weapons of Math Destruction.
Yeah. A very respected figure for the ethics of AI, Cathy O'Neil.
There was Meredith Whittaker, the Signal president. She was there. Kate Crawford was there. So many of those people were here and had dinner directly with President Macron. It had nothing to do with the dinner of Trump with the big CEOs, like, 'I'm the boss.' It was really like the president was trying to understand: 'What is it? Tell me what should I know,' etc.
So it went well with them. I estimate that about half of the report was put in action, which is a very high figure compared—usually a parliamentary report is 10% or even 0%.
Yeah. Even there wasn't. [laughter]
Yes.
Wow.
So there was an impact, and the debate was there during the process of the report. There was a lot of public debate which made me identified, I mean, for the public audience. Even now, every week or so I get invited for a debate about AI, which is in large part because I was involved in that report, and it was known by those.
And by and large, for instance, for the research system there was good evolution. For the medicine issue system, a good evolution. Among the main measures, a number of steps were well taken. And by and large, the issues that were least treated, and government did not really know how to do, are things such as gender. The girl problem is still there, very much underrepresented in AI, and it's still up there.
Training: we're not training enough people. Careful. Now we know there's a catch, because for the first time, students in programming have difficulties being hired.
Yeah, that's the only thing we want to talk about, too. But—
That's a real issue. I don't know how to handle it. I don't know how to handle it. This is a real big issue because we need programmers. We need people to understand the whole stuff. But now we see that the big companies are using less and less programmers, which, by the way, is starting to cause big problems also for security, because you need a number of people to check and care and cure and whatever.
Okay. Anyway, the training and the international at the European level, it's nowhere near what it should be. The regulation was done, and I like what we did, what Europe did in the AI Act. Regulation was good. Implementation is still a problem because the big European regulation is separated into 26 different shades, which, for instance, Mistral is becoming crazy because they have to use the rule like this in Czech Republic and like this in France, and they all have different particularities. So it needs simplification.
But in terms of action and strength of the network and concentration in big projects, we don't have—for instance, Europe could have a big LLM project, an open one, which would need to be a public and private partnership. Usually if it's fully public, it runs into lots of issues and difficulties. A private-public partnership is not so easy either, and we don't have enough money in the public system to just drive it. Or at least we should have—the public money should incentivize the incentives for the public and so on. So certainly this is the most important, and it requires collective European action and so forth. Yes.
That's okay.
So, soch.
That's okay.
That's right.
Hey.
It's good.
Science.
There we go.
[music]
[music]
[music]
Article published · Updated