07/08/2021, by Guerric Poncet
INFO LE POINT. The space imagery of the French armies will now be analyzed by algorithms that the CIA wanted to appropriate in 2020. Explanations.
This article is a translation of the article from Le Point, available here
The French Ministry of the Army will entrust the analysis of all its satellite images to the artificial intelligence (AI) of the French company Preligens. This system, already used by the Directorate of Military Intelligence (DRM), is capable of automatically detecting armored vehicles, planes or ships, and especially of monitoring the evolution of a strategic site.
The images taken by the ministry's historical satellites will be integrated, as well as those of the most recent constellation, CSO, whose second satellite out of three, CSO-2, entered service at the end of 2020. The AI can alert a human analyst whenever a nuclear submarine leaves a military port, or when an unusual concentration of armor is detected.
Between 50 and 100 times too many images for analysts
"This tool allows armies to increase their surveillance and intelligence capabilities by processing the growing volume of data coming from new sensors and by relieving experts of repetitive tasks," explains the Ministry of the Armed Forces, assuring that "the army thus has greater efficiency in situation assessment and decision-making.
The CIA wanted to take over Preligens
"Analyzing images is a crucial issue, and our AI generates alerts as soon as predefined scenarios are observed," continues Arnaud Guérin, whose company underwent a takeover attempt in 2020 by investors close to the CIA, before being secured in extremis by French and European funds, including the fund of the Ministry of the Armies, Definvest. "We will now process 100% of the images acquired by the French armed forces," he says with satisfaction.
Le Point: Can your AI detect so-called deception maneuvers, i.e. tricks, when the enemy tries to deceive the satellite?
Arnaud Guérin, CEO of Preligens: Our tool is not magic, it detects what we teach it to detect! Eventually, we could cross optical imagery with infrared and radar to confirm that tanks visible on a satellite image are not made of cardboard, for example. And if it's a power plant that's being monitored, infrared data will tell us whether it's in production. The same goes for a uranium enrichment facility... In any case, the result is always dependent on the input data, and our AI allows the analyst to scale up, because we handle very large volumes of data very well.
In our industry, patenting is not a strategy that works. Protecting our intellectual property would require us to be able to monitor what our competitors are doing, and to act if there is an infringement. But we can't knock on the doors of American intelligence agencies or industrialists! Instead, we intend to protect ourselves through discretion and by staying ahead of the game. We have the largest team in the western world today in AI for intelligence: we have more than 60 engineers and doctors in deep learning, 100% French. For the last recruitment, we received 140 applications in two weeks.
We obviously follow the rules set by the embargoes and by the French government. But the restrictions that we decide internally are much stronger. We draw up a green list of countries based on NGO data on democratic or non-democratic regimes, as well as on our own analyses. Our priority development targets are Europe, North America and Southeast Asia, and this is already a very good playground.