Thales designs an AI like no other for the military

Within Thales, more than 600 experts are developing artificial intelligence for the defense and security fields, with constraints that are out of proportion to mainstream AI.

“This critical environment brings us duties that do not exist in a general public environment. The constraint is injected from the design of the AI ​​algorithm which cannot function as a black box and where humans are absolutely essential,” explains Philippe Keryer, director of strategy, research and technologies at Thales, the leading patent holder. in AI of critical systems in Europe.

Aerial anti-drone system tested during the Olympic Games in Paris, intelligent sensors to detect underwater mines, piloting swarms of drones, optimizing flight trajectories or preventing identity fraud: ahead of the Summit for the action on AI which will be held in Paris on February 10 and 11, Thales opened its research laboratories this week in Palaiseau, in the Paris region.

“We have the responsibility to fundamentally rethink the functioning of AI as well as learning models,” believes Philippe Keryer.

It employs “ethical hackers” to anticipate threats, invent the most sophisticated attacks and subject software to a “resilience crash test” before being validated. A “sword and shield” principle already applied for weapons systems (drones and anti-drone systems). “It is by thinking about evil with the most perfidious attacks that we will create good,” affirms Philippe Keryer.

Another challenge: on a battlefield, “we are constrained in size, weight, power, but also by the type of network to which we are connected,” underlines Fabien Flacher, head of cybersecurity at Thales. And if artificial intelligence is generally trained on “long-frozen” data, this cannot work for modern conflicts. “We instantly relearn the AI ​​to be more relevant” after each mission.

“We judge AI more harshly than humans,” says Christophe Meyer, technical director of CortAIx Labs, responsible for AI for Thales. But the crucial decision always comes down to humans. “If there are drones with shooting capabilities, there would be a human decision to say ‘I validate this suggestion that you are making to me, with my criteria which are human criteria’,” he underlines.

The solutions proposed by this type of AI also contain a rational explanation. The calculations it provides allow the operator to relieve their cognitive load. Thus, an intelligent radar “will recognize the size of hundreds of targets in a few tens of seconds, where we needed tens of minutes before,” explains Nicolas Léger. The algorithms help “accelerate classification and evaluate the relevance of producing identification and neutralization operations,” explains Benoît Drier de Laforte, advisor in the fight against mines. This technique allows for only “1% to 2%” of false alarms, according to him.

However, algorithms are not yet ready to replace human “big ears”. “If the algorithm has not been trained on a new threat, it risks lacking performance,” underlines the expert.