The emergence of defence artificial intelligence (AI) has been gradual. When did it begin?

Bertrand Rondepierre: In France, the subject began to gain ground in 2015-2016. I nonetheless think that the breakthrough came in 2019 when Florence Parly, then the French Defence Minister, presented the nation's major priorities and ambitions in terms of AI. On the other hand, it is more difficult to date its arrival on a global scale. In the United States, for example, the White House published a national AI strategy shortly before 2018. But it was above all the creation of the CDAO (Chief Digital and Artificial Intelligence Office) in June 2022 that really launched the country down this path. More generally, over the last few years we have seen a clear and very significant acceleration in technological progress, with an AI maturity increasing sharply in 2023. I therefore believe that AI is no longer in the realm of innovation. It should be deemed as a technology incorporated into operational applications.

In what way does the Russo-Ukrainian war illustrate the acceleration of the shift to AI to modern conflicts?

Without necessarily delving into the Ukrainian conflict, AI in warfare is not limited to the battlefield, but extends from everyday life to command and operations. Actually, there are two main areas of application for AI. The first is command assistance: AI in reflective time. The aim here is to shorten the decision-making loop. How can this be achieved? By better understanding the volume and mass of data. Let's take the example of a drone that sends back images of the terrain. In this particular case, AI can process the information by analysing and interpreting it much more comprehensive and faster way than a human operator. The second area of application is in real-time embedded systems. For example, with pilot assistance, or even the automatic piloting of a combat aircraft.

What have you learnt from the fighting in Ukraine?

There are related issues to artificial intelligence that are also very important. I'm particularly thinking of the omnipresence of data. Its understanding is therefore fundamental, because the battlefield is completely digitised. As a result, we need to be able to react and adapt in real time. A drone in Ukraine flies in a totally jammed environment. If you don't have jamming or intelligent bypassing systems, then no machine can move in the air. Therefore, one of the lessons of the conflict is that it is imperative to be able to wage war in a deteriorated situation, namely in the electromagnetic field. AI is thus used to acquire a certain degree of autonomy, for example so that a drone reaches its target whatever the environment.

In March 2024, Sébastien Lecornu presented the ministerial strategy for defence AI. What are the key points of it?

Sébastien Lecornu did remind us that AI is a "game changer" for the Ministry and the armed forces. So we must adapt to meet the circumstances straightaway. This includes, among other things, the creation, in the first half of 2024, of the Ministerial Agency for Defence Artificial Intelligence (Amiad). Its mission is to professionalise the use of AI in the armed forces by accelerating its integration for the benefit of operations and organic processes. The result will be enhanced operational superiority and greater efficiency in the day-to-day work of the ministry's employees. In recent years, our ability to put AI technologies into production for the benefit of weapons systems was quite limited. Amiad will address this shortfall through a number of levers and means of action. One of these relates to human resources. The agency is namely tasked with building up a pool of expertise, so as not to rely solely on manufacturers for any particular project. But also, so as to take back ownership of some of the digital and AI expertise within the Ministry, both for the most sensitive applications and to be as close as possible to operational needs in real time. So this brings us to the question of sovereignty.

Precisely, why is it important for France to be sovereign in terms of defence AI?

This sovereignty implies mastery of the technologies, through people who also know how to develop them. Internalising some of these skills is therefore part of our action plan. To fulfil its mission, which is to put AI into production, Amiad must obviously recruit experts in many different fields. I'm thinking in particular of full stack[1] developers, or those working in information system security. For the record, Amiad comprises 75% technical professions and will employ 300 people by 2026. More broadly, Amiad must facilitate collaboration within the entire AI ecosystem, from academics to manufacturers, including start-ups.

Is this an achievable goal?

It all depends on the overall ambition. Sovereignty can be applied at both very low and very high levels. For example, it is not realistic to want to produce everything ourselves. But by taking back ownership of skills internally, the French Ministry of Defence has been developing a certain mastery of the technologies it needs to produce AI. The roadmap that has been drawn up is therefore perfectly achievable and relevant. To sum up, it's not about reinventing all technologies, but rather starting by mastering the existing ones. If one day we were to lose access to some of them, for whatever reason, we would thus be able to do the work ourselves.

How does France compare with other countries?

It's always difficult to compare ourselves with others, because there isn't really any tangible elements to rely on. Every nation talks about AI without ever saying too much about it, and sometimes the words are more or less faithful to reality. Nevertheless, in Europe, France is very clearly in the lead. Besides, we are aiming for the top spot on the Old Continent, an ambition that is very close to being achieved. Internationally, things are much more complex. The United States is probably in the lead. Then there's the question of the exact positioning of the various players.

The French Ministry of Defence should also have its own classified supercomputer dedicated to AI by 2025. In practical terms, what will this change?

The Minister wants to have this supercomputer in place by the end of 2025. A deadline that we are well within at the moment. What will it allow us to do? It will give us a fairly massive 'secret' level capacity. On a daily basis, we work with a mass of data that is often highly sensitive. To process it, we need machines with significant computing power, capable of handling classified information. This capability does not currently exist. Located at Mont-Valérien in Suresnes (Hauts-de-Seine), this supercomputer will also allow us to guarantee our sovereignty in the long term.

In the longer term, what role will AI play in structuring programmes such as the Future Air Combat System (Scaf) or the New Generation Aircraft Carrier (PA-NG)?

This is precisely one of Amiad's missions. The agency must indeed think about this and ask itself the following question: from a capability and operational standpoint, what does it mean to have AI? The Future Air Combat System and the New Generation Aircraft Carrier are two programmes that are still quite distant[2]. So we need to be a bit visionary. Nobody really knows what the AI will be like in the next aircraft carrier. What we are sure of, however, is that it will indeed be there. As we can't foresee the technologies that will come out by then, the challenge is not so much to define the future AI algorithms of the ship. We can, however, determine the concepts of operational use and the link between humans and machines. In short, we can anticipate the overall architecture so that once it is assembled, it is not already obsolete.

[1] A technician or engineer, the full stack developer performs all the technical functions of a website or web application.

[2] The successor to the Rafale fighter jet should be operational by 2040. The new aircraft carrier, meanwhile, should see the light of day in 2038.