While he is “happy” to be eternally portrayed as the creator of Siri, computer scientist Luc Julia has moved on. With a critical--sometimes very harsh--eye when it comes to announcements worth hundreds of billions on both sides of the Atlantic. On Monday 24 March, the engineer will be one of the headliners at the , organised by the ministry of the economy, the Luxembourg Chamber of Commerce, the Idea Foundation and Fedil, in collaboration with PWC Luxembourg [].
Thierry Labro: What do you think of all these announcements about artificial intelligence, which are multiplying in Europe, the United States and China?
Luc Julia: There are a lot of announcements, which is good, but some of them are more or less lies... In the United States, it’s essentially about infrastructure, data centres that are going to destroy the planet! In Europe, there will also be data centres between the €109bn announced by Emmanuel Macron and the . Lots of data centres too, but perhaps with a greater ecological conscience than in the United States. And as for China, true to form, we know nothing... We’re seeing AIs coming out, with very interesting price/cost ratios. They’re clever, they do things with fewer resources than the big guys demand. I’m pleased because this is in line with what I’ve been saying for years, which is that there’s no point in developing these big ‘things’ like OpenAI.
We need to take on a European, African or Asian bias, because our AIs need to respond to us as Europeans, Africans and Asians!"
How can these slightly more frugal AIs, working on smaller models, be as effective and interesting as OpenAI and co?
The big guys are pursuing an impossible goal: generative artificial intelligence. Something that will do everything, all the time and very well. God, in fact. They’re looking for God. Unfortunately, in the 70-year history of artificial intelligence, every time we look for this ideal, we go back to things that are much more reasonable but that work. By narrowing the focus, we’ll end up developing specific AIs for specific tasks. As a result, it will be more frugal.
So there’s no real reason to spend billions of dollars running these computers at the moment?
No. We need a bit. Especially in Europe, because we’re too dependent on what’s happening in the United States. We need sovereign clouds, of which there are only two or three models in certain European countries. We need a European sovereign cloud. It wouldn’t do us any harm to create models.
Another thing we should be talking about, I think, is that we need to assume the biases of these AIs. At the moment, these are American biases. We need to assume European, African or Asian biases, because our AIs need to respond to us as if we were Europeans, Africans or Asians. Let’s take an emblematic example of what these AIs are and why we have to assume our biases. If you ask any AI who invented aviation, it will tell you that it was the Wright brothers in 1903. If you ask any Frenchman, he’ll tell you it was Clément Ader, 13 years earlier, in 1890. And that’s the truth. It’s the truth that the French have accepted. It’s the accepted truth of the Americans... Culturally, AIs are based on the Internet and the Internet has an Anglo-Saxon, West Coast bias... We need an infrastructure to restore a bit of balance to the internet.
Yes, on the side of cloud providers, there are often layers of technology that are American or Chinese. But in the European Union’s communication on AI Gigafactories, what the Europeans didn’t really say was how they were going to get access to 100,000 latest-generation AI chips to run them. And if I add the GDPR, which splits up the data sets on which models can be trained, European sovereignty is not for tomorrow...
On chips, yes, we have a problem, but Deepseek’s frugality has shown us that we can potentially do things with less. That’s the one thing the Chinese didn't lie about. We decided, I don’t know, a few years ago to stop developing chips... It was a monumental mistake. The Chinese are now making their own chips. It took them four or five years to set up the whole industry. We, like frightened virgins, are going to prevent ourselves from copying things, so it’s going to take us longer. But some of the European billions will have to be used to redevelop the chips. We’ll need ten years.
Facial recognition, if it’s to enable a disabled person to open their door, is pretty good!
As for the GDPR, it’s your fault, you Europeans who live in Europe. Luxembourgers, French, Belgians... It was interesting but it served no purpose whatsoever. Apart from annoying users who have to click on dozens of pop-ups without reading anything at all... I’m pro-regulation! The GDPR has only had an educational virtue. It has explained to people all the dangers associated with the Internet. We still don’t know why people are still on Facebook, but they are still on Facebook. It’s an aberration. The problem is that the EU has implemented an AI Act. When AI started to arrive, the Europeans rushed to draft an AI Act, , which makes me smile. For one thing, it contradicts the GDPR. Good luck to the lawyers applying it! Take a look at the battle going on right now between Apple and Facebook. Meta is asking for algorithms in the name of transparency and Apple is saying no because it’s contrary to the GDPR. Europeans are shooting themselves in the foot with these texts. As a result, neither Meta nor Apple are deploying things in Europe, like Apple AI intelligence--maybe that’s just as well, I don’t know.
Regulation is a problem. Because the texts were drawn up without asking the real AI specialists for their opinion! We don’t really know why Thierry Breton and his friends did this in such a small committee."
Regulation is a problem. Because the texts were drawn up without asking the real AI specialists for their opinion! We don’t really know why Thierry Breton and his friends did this in such a small committee. We have regulated too much at the level of the technologies themselves instead of at the level of their application. The granularity that goes right up to the ban is incredibly stupid, because we don’t yet know what applications might emerge. If they do, there will be some fantastic applications.
Take facial recognition. No, if it’s used for social scoring like in China, that’s no good. But if it’s to enable a disabled person to simply open his or her door, that’s pretty good! Today, those who want to develop good applications and who would not be able to do so under the AI Act will have to go through European justice and case law, and that will take ten years. What annoys me is that back in 1984, France made the same kind of mistake with genetics and the ban on genetic manipulation out of fear of the development of eugenics. But people suffering from genetic diseases are still waiting for solutions.
Your report on Arte, Silicon F***ing Valley, reminds us, if we had forgotten, that you are both an expert on Silicon Valley and a true European. What is lacking in the schemes announced in recent weeks on one side of the Atlantic to enable us to hope to compete with what has been announced on the other?
I’m trying to take a positive view of things, to explain the dangers, to stop talking nonsense about these technologies like Elon Musk and Sam Altman, who have only one objective: to make money for their companies. OpenAI has already taken $10bn but needs at least ten times more. But these are extraordinary tools if you use them wisely. You have to understand them, you have to educate yourself. For this to work in Europe--and yes, I’m a convinced European, but a 1950s European--you need money. That’s what we wrote in a report for the French president in February 2024 with dozens of specialists. We have experts and geniuses in Europe, well trained in mathematics but also multi-educated. Our Enlightenment heritage enables us to understand maths, but also philosophy and many other things. We said we needed €10bn. We’ll have to use them properly if we are to make the transition from startup nation in 2013 to scaleup nation. In Europe, it’s very complicated to put €100m into a company.
The money arrives. What are we going to do with it right now?
We’re financing data centres and infrastructure. Take a look. The French supercomputer, Jean Zay, has tripled its capacity and we’ve given it €40m. It is 1,000 times less powerful than Facebook’s computers. We need €40bn! Then we need to give these companies the capacity to go out and find a market. I’m in favour of the European market of 1953. But today, it doesn’t exist. There are 35 countries and as many languages, borders and regulations... They say we have a captive market of 400m people, but we don’t. That’s a lie. That’s a lie. Here in the United States, we have a market of 330m people who all speak the same language, and there are no borders between states. We have to look to other markets. In Europe, but also in China, where it’s complicated, or in the United States, and there again, you need money.
And you, who for 30 years have been eternally hailed as the father of Siri, what are you up to these days?
I work at Renault. I’m lucky enough to be able to do what I want, where I want and how I want. I spent ten years in research, ten years in a startup and ten years in a big company in the United States, and three years ago I decided to spend ten years in a big French company to give something of myself back to France. I do AI for this big company. I’m doing AI for the people in the factories, the white-collar workers, to be more productive, to do better, to improve the quality of what they do. And we’re also putting AI into cars, in the face of the Chinese and Koreans who are very comfortable with these technologies. We need a little avatar, Reno, for the Renault 5, who talks to you and explains how the car works.
On the subject of productivity and quality, it is often said that AI will improve productivity, but others say that the productivity gains will be eaten up by the processes required to implement it correctly. Where do you stand?
The history of AI and technology has shown this dozens of times. You have to be very careful what you promise. It offers productivity gains in certain sectors and under certain conditions. Take the environment. If you adopt some of these technologies and have to draw up a CSR report, there’s a good chance you'll run into some problems... Here again, education is key. Some things can be good. With generative AI, the quality of what humans produce increases.
This article was originally published in .