Arthur Mensch, CEO of Mistral, acting as a big brother at École Polytechnique, on AI and entrepreneurship

"Artificial intelligence is a bit like the work of an electrician. Our role is to transform megawatts into intelligence." This phrase sums up in a few words the energy and climate challenges of artificial intelligence and the prospects it opens up. It also sums up the style of its author, Arthur Mensch (X2011), co-founder and CEO of Mistral, the French AI gem: simple, direct, detached, but with a clear and uncompromising vision of the potential and risks of a technology that he has helped to develop and is working to deploy.

Invited by "Binet Start-up", École Polytechnique's student association dedicated to entrepreneurship, to meet with students at École Polytechnique, he referred to the motto of his former School-"For the Nation, for Sciences, and for Glory."-in a speech on artificial intelligence and its implications, as well as on his career as an entrepreneur.

PUSHING THE BOUNDARIES OF SCIENCE

Starting with science, he discussed the consequences of AI for engineers and for scientific and technological research. "My generation of engineers has more or less succeeded in commoditizing its own profession," he noted. "What we were trained to do-the ability to synthesize, calculate, and think, and the ability to develop software that many of us use (...), what we have done over the past ten years is to create systems that perform these tasks better and better. "

"Today's technology requires us to move to an additional level of abstraction (...) artificial intelligence promotes a new type of brain that is much more creative," continued Arthur Mensch, adding, however, that while we need to learn how to use AI, we also need to know how to do without it completely.

"To learn, to understand, to develop reasoning skills, working without a computer remains very important. You have to learn to do without AI to understand how the system works," he emphasized, adding that with the advances made by AI, software is now designed in a completely different way. Whereas previously the aim was to develop (software) solutions to problems, "today we have the solution [editor's note: AI], so the real issue is identifying the problems we want to solve."

In this new environment that has emerged in recent years, he believes there are three possible paths for future engineers: working for AI, working with AI, and working without it.

"There is still plenty of time to work on AI. We know how to build systems that solve certain problems well (...) that enable coding, but we don't yet know how to build systems that understand humans, interact with the real world, understand how robots are made, understand the physics of materials, biotechnology, and so on. At Mistral, we strongly believe in pushing the boundaries in all areas of science through AI. We need to equip ourselves with AI to make discoveries more quickly. There are enormous opportunities at the intersections (of AI) with different disciplines, but this requires a detailed understanding of the questions to ask and the problems to solve," emphasized Arthur Mensch.

"To develop models in materials science or quantum physics... we'll have to get up early. We'll have to search for data, find simulators, and set up an environment that allows us to test and strengthen the system. All of this will require a great deal of human thought," he added.

NO TECHNO-SOLUTIONISM

"It is also possible to work with AI. All companies in all sectors of activity will have to undergo profound reorganization of processes that will run differently, and information systems that will function differently," explained Arthur Mensch.

"All countries will deploy AI and change their processes, whether in financial services, public services, industry, agriculture, defense... The transformations that need to be implemented to maintain competitiveness, particularly in Europe, are very significant. "

"It is also possible to work without it and focus on issues that will not be resolved by AI," noted Arthur Mensch. "There is a whole range of professions, such as customer service and healthcare, and a whole range of sciences where, objectively, AI has a minor role to play," he said.

"We launched Mistral [Editor's note: the two other co-founders of Mistral AI are Guillaume Lample (X2011) and Thimothée Lacroix (ENS)] in France in particular because we do not completely believe in the technosolutionism that is in vogue in the western United States in particular, which consists of thinking that all the world's problems can be solved thanks to generative AI (...) It's a little too simplistic, and AI is not going to solve the problem of global warming... in fact, it tends to make it worse," he pointed out.

A TRIPLE CHALLENGE TO SOVEREIGNTY

Continuing with The Nation, Arthur Mensch said he believes that AI is a matter of sovereignty in three areas. "The first is industrial, in that, given that a large number of processes will be partially automated using AI, there is a problem if we import 80% of this technology or its components from two countries [Editor's note: China and the United States] whose strategic alignment [Editor's note: for France and Europe] is increasingly problematic. In France and Europe, there is a growing awareness that in order to have a competitive industry, we need AI that is not subject to the control of external entities. This will remain true and will become increasingly true as a growing share of European GDP revolves around AI," warned Arthur Mensch.

"AI systems are like many technologies, but perhaps more so than others, dual-use technologies that can be used for very interesting civilian applications, but also to empower drones, to have more autonomous systems on the battlefield, or to increase-in technocratic terms-lethality on the battlefield," added Arthur Mensch.

"Finally, there is the issue of influence. We have seen Google establish a 95% monopoly on how we access information around the world. AI systems are systems through which you already access to information and through which future generations will access even more to information. Unfortunately, these are systems with which everyone empathizes. The mechanism of influence, the power of influence of these systems, is ten times greater than that of previous systems, ten times greater than that of social networks. This is something we really need to be concerned about," warned Arthur Mensch.

"If tomorrow we were in a world where two entities owned the information portals, it would give them excessive power to control thought, which is not at all compatible with maintaining democracy," he said.

In light of these sovereignty issues, Arthur Mensch challenged future engineers. "You have a choice to make. Do I choose to innovate, to contribute to the power of the country where I was born and educated, or do I choose to bring innovation-and innovation is the main driver of power-to another country that is not necessarily aligned with my own?" "

"Technology today is designed as a vector of power, as a vector of colonization, and designing your career around that to try to reverse the balance of power is something you might want to think about."

"DO A PHD !"

Ending with glory and antiphrasis, he talked about his career as an entrepreneur and "its less than glorious aspects."

"Very quickly, entrepreneurship requires knowing how to sell and knowing how to iterate if you don't succced to sell... and that's very difficult. It requires patience and the ability to take criticism. Very often, customers or investors will tell you that they think what you're doing is rubbish. You're always in a weak position when you're the seller because you're doing the work, you're getting tired, and you're facing criticism. You have to live with this asymmetry and be resilient," warned Arthur Mensch.

From this point of view, "I can only recommend that you write a thesis ," he told the audience, because beyond the research work that allows you to identify problems, "a thesis teaches you a lot about entrepreneurship, particularly how to see projects through to completion, how to persevere, how to struggle, because in reality, being an entrepreneur is generally 90% struggle and 10% success. "

"This aspect of not succeeding...is something you learn very well in a thesis," he summarized.

POSSIBLE INDUSTRIAL BUBBLE IN AI

During his discussion with the audience, he did not shy away from any questions, from Mistral's ability to continue its commitment to open source models to the consequences of military applications of AI, including the risk of an AI bubble bursting, as the sector has benefited from massive investment in a very short period of time, at the risk of not delivering on all its promises as quickly as expected.

"Today, the [AI] sector is betting that massive investment is needed, in the order of trillions of dollars, because there is enormous value to be created... and it's true that there is enormous value to be created," Arthur Mensch said on this point.

"Investors expect a return on their investments within a timeframe that is not necessarily the same as the timeframe in which AI will be able to deliver on all its promises," he continued.

"This is not a speculative bubble, but rather an industrial bubble in the sense that there are capital deployments for which investors expect returns that will take longer to materialize than they hope. "

"For us [Editor's note: at Mistral], this is not a major issue because we are very focused on creating long-term value for companies, but we cannot rule out macroeconomic scenarios in which access to private capital is closed off due to a wave of panic. This is entirely possible, and we must take it into account and prepare for it," concluded Arthur Mensch.