The Minister of the Armed Forces makes artificial intelligence a strategic national defence priority

On the occasion of the visit made by the minister to present the ministry’s artificial intelligence strategy (AI), several partners from the Institut DATAIA presented innovative technology based around 3 themes: 

Photo V. BESNARD - Ministère des Armées

• Artificial intelligence and information management
• Artificial intelligence and interactions
• Artificial intelligence and embedded systems

Information management

L’IA et la science des données permettent de représenter les connaissances humaines, en créant des modèles du langage humain et en reliant des entités qui apparaissent associées dans de grands corpus de textes. Ceci permet de tester automatiquement l’authenticité d’une affirmation.

• ConnectionLens: Data and AI against “fake news”, by Ioana Manolescu, Inria and LIX - the IT laboratory at École Polytechnique

How do we know whether or not to trust a piece of information? How do we do this in real life? In our daily lives, trust is based on “who is talking to us” and “what we already know” about a given source of information. In collaboration with the fact-checking team from the newspaper Le Monde, the Cedar team designed ConnectionLens, a system that can be used todiscover and analyse links (social, collaborative, etc.) between individuals and organisations, which are gathered through a wide range of data sources. This type of analysis makes it possible, for example, to identify links or conflicts of interest between different stakeholders, or to interpret their words through the prism of their affiliations and economic interests.

• YAGO: one of the largest public knowledge bases, by Fabian Suchanek, from the Institut Mines-Telecom

YAGO is a knowledge base, i.e. a collection of data about the world that a computer can “understand”.It contains 10 million entities (including towns, people in the public eye, organisations, etc.) and more than 100 million facts about these entities (who was born where, which actor appears in which film, etc.). YAGOwas one of the first knowledge bases, which saw it awarded the “Test of time award” at the WWW 2018 conference (the world's largest conference on the internet).

Nowadays, knowledge bases are used in search engines as well as in AI applications. YAGO, for example, is used in the famous IBM Watson system.   

Interactions

Designing enhanced modes of interaction between humans and artificial intelligence systems is a fundamental issue. But how can we improve the way in which we use these interfaces? And are there any risks of human manipulation?

• Robotics and prompting by Laurence Devillers, CNRS-LIMSI and the University of Paris-Sud

Connected objects, and voice assistants in particular (such as Google Home, for example), have brought a new dimension to interactions between humans and systems and could soon have an influence on individuals. For the time being, however, they are neither regulated nor evaluated, and remain highly opaque.

Focusing on the study of“nudges”, techniques studied so as to be able to modify individuals’ behaviour, Laurence Devillers’ research team “Social and affective dimensions in spoken interactions” shine a light on the threat posed by these techniques for vulnerable groups, including children or elderly individuals. 

• AI used for speech analysis in the defence sector by Jean-Luc Gauvain, CNRS-LIMSI and the University of Paris-Sud

The goal of this demonstration is to present the latest developments in relation to automatic multilingual processing of conversational speech in the defence sector, whether this is in relation to security, intelligence or human-computer interaction. A significant amount of progress has been made in recent years in terms of the quality of speech recognition, thanks in large part to the implementation of deep neural networks.

Defence applications have not been able to reap the benefits of this progress as a result of difficult environments (noise, stress, accents, rare languages, specific phraseology, etc.). Developments in AI mean it is now possible to tackle these issues relating to speech recognition. 

Embedded systems

The world of AI is not restricted to large computational platforms: many applications require decentralised computation, carried out on lightweight devices. This might involve calculating the position of an agent without the use of GPS, for example, or analysing flows of images on compact devices.

• Embedded systems and Artificial Intelligence by Marc Duranton, CEA

CEA List develops AI solutions capable of ensuring that embedded systems (self-driving vehicles, smart sensors, etc.) are able to function to high performance levels (processing power, low energy use, low volume). N2D2, for example, is a tool developed at the CEA List which, in a development context for a deep learning system enabling a vehicle to analyse its environment by compiling a 3D reconstruction, helps cut power usage by generating an optimised code for embedded architectures. These tools can also be used to carry out real time (120 images per second) labelling (detecting vehicles, boats, swimming pools, etc.) of aerial images.

In the interests of efficiency, and in order to retain a degree of sovereignty, CEA List are able to develop hardware accelerators that consume 200 times less energy than embedded GPUs used to carry out image recognition tasks (“PNeuro”), or even to make use of the advanced technology of the CEA Leti by developing, for example, an intelligent retina capable of processing up to 5,000 images per second. These various different solutions have already been implemented by CEA’s industrial partners.

• BlueForce for soldiers: civilian geotracking in GPS-denied environments, particularly inside buildings, by Thomas Bonis, Inria

When it comes to access in tactical situations and rescue operations, for defence applications, it is critically important to be able to locate civilians in GPS-denied environments. First responders such as soldiers or firefighters must be able to locate individuals immediately without having to install any infrastructure and without hampering the work being carried out by the agent . A device called an inertial measurement unit fulfils this role for large-scale defence operations.

Thanks to a ground-breaking innovation - magneto-inertial navigation, developed by Sysnav - and movement recognition algorithms developed by Inria’s DataShape team, the Sysnav team were able to model a unique localisation solution using low-cost, space-saving sensors capable of meeting the operational needs of first responders . The work undertaken to arrive at this solution was carried out as part of MALIN, a contest organised by DGA and the ANR and won by Sysnav and DataShape / Inria. The launch market selected by the company for the deployment of this technology was the medical market, with the measurement of trajectory parameters in the context of clinical tests used to evaluate effectiveness during the process of developing new molecules.