Whether it’s to assist in the fight against fraud, or to help France Services agents answer citizens’ questions, artificial intelligence is increasingly being integrated into public services. This irruption is part of the wider digital transformation of the State (transformation numérique de l’État), which aims to make the State’s actions more efficient.
Given the risks of AI – be they existential, or more mundane and immediate – and the stakes involved in their use, the focus today is on developing “trustworthy” AI systems (« dignes de confiance »)
The use of artificial intelligence in public services touches on a huge number of areas. During the Paris 2024 Olympic Games, artificial intelligence was deployed to identify dangerous situations (identifier des situations dangereuses ) from video images.
Artificial intelligence also helps to recognize images of drivers not wearing seatbelts (conducteurs ne portant pas de ceinture de sécurité) or to recognize the faces of travelers presenting themselves at a PARAFE device for passport control (contrôle de passeport. A generative AI called Albert helps France Services agents answer citizens’ questions about the steps to take. An AI helps Caisse Nationale des Allocations Familiales (CNAF) inspectors prioritize fraud investigations (Caisse nationale des allocations familiales (CNAF) à prioriser les enquêtes pour fraude). Another analyzes satellite images to detect undeclared constructions. An AI algorithm is at work in the Parcoursup system. Some tools help to counter cyber-attacks, while others analyze connection data to detect risks of terrorism (détecter des risques de terrorisme).
In 2022, the Conseil d’État has identified eight major areas for AI deployment by the State, ranging from national defense to personalized support for jobseekers.
Each of these areas requires an individualized approach to understanding what risks are associated with an AI deployment and how to manage them in practice.
The term “AI” covers a wide range of technologies, uses, and therefore risks
Terrorism risk detection tools and the Parcoursup algorithm rely on “old-school” ( à l’ancienne ) AI based on logical rules created by humans, while the algorithms used for image processing rely on supervised machine learning algorithms. Generative AI relies on a new reinforcement learning technology called “transformers” « transformeur ».
In addition to this technological diversity, the uses of AI, and the associated risks, are extremely varied. An error in predicting the emptying of a dumpster does not have the same consequences as an error in predicting the risk of terrorism. Generative AI feeds fantasies about the end of the world. The French Human Rights Ombudsman (Défenseur des droits) highlights the more down-to-earth risks of discrimination.