Artificial intelligence and public services “sandbox”: the CNIL supports 8 innovative projects

04 décembre 2023

The CNIL has selected four artificial intelligence (AI) projects aimed at improving public services. The winners of the “sandbox” will benefit from personalised support over several months. The CNIL will also help four other projects of interest for data protection.

A "sandbox" to promote innovative and privacy-friendly AI


The CNIL’s ‘sandbox’ is a mechanism designed for actors initiating innovative projects. It allows to benefit from the CNIL’s support and expertise on emerging legal and technical issues.

For its 3rd edition, the CNIL launched this summer a call for projects  on artificial intelligence in public services. This initiative is part of the CNIL’s desire to support the development of AI respecting people’s rights and to build appropriate regulation alongside stakeholders.

The CNIL has received more than twenty applications, mostly from public actors and mainly for generative AI tools for various use cases such as ecology, relations between users and administrations, employment and health.

After an analysis and a pre-selection of the applications by a committee composed of external personalities and members of the CNIL, the president of the CNIL selected four winning projects.

The winners: service to users at the heart of the projects

The winners will be accompanied for several months by the CNIL teams and, in particular, by its new artificial intelligence department. The discussions will make it possible to identify and implement solutions adapted to the problems of project leaders.

Albert

The “Albert” project of the DINUM (interministerial directorate for digital affairs) is aimed at civil servants. It consists of assisting them in the search for information and helping them to formulate specific responses to users. The project is based on an open language model.

Several use cases are under development, including an experiment with the network of “France services” houses, which advise users in their administrative procedures and the use of digital services.

This project aims to meet different needs, in different situations. It could thus constitute a lever for the deployment of AI in administrations.

Personalized advice from Employment Intelligence

The project “Personal Job Intelligence Advice” is equipped with a conversational tool for Pôle Emploi advisors to help them propose a personalized journey adapted to the needs of job seekers. Among the multitude of services available (training, workshops, job dating, etc.), the tool facilitates the identification of the one that will be best suited to the user by relying on their professional data (diplomas, skills, etc.).

With an immediate impact for advisors, the project aims to strengthen support for jobseekers. It will also be an opportunity for the CNIL to address, via a concrete case, issues related to the creation of databases in order to train a language model.

Ekonom IA

The Ekonom IA project focuses on the water consumption of the inhabitants of Nantes Métropole, the project leader. The aim is to provide users with information and recommendations on their water consumption. For this purpose, it is planned to carry out a comparison, using AI, of users’ consumption to that of a reference household.

This project addresses a major concern of local authorities in terms of ecological transition. If it succeeds, it could be adapted to other public policies, but also to other territories.

RATP project

With its project, RATP wants to develop an AI that would work on new forms of video capture.

It is based on the use of matrix data capture technology (i.e. numerical values) to ensure that no personal data is retrieved. The data thus collected would allow the development and training of AI algorithms in order to detect and characterise predetermined events.

By incorporating the principle of privacy by design, the project would respond to a need identified by many stakeholders, while ensuring that the rights of data subjects are respected.

Support for four other projects

Four other applicants stood out for the quality of their project, of strong interest in data protection. To support their implementation, the CNIL will provide them with legal and technical support:

  • College of Medical Information Departments (DIM) of University Hospital: development of a generative language model specialized on useful tasks in the hospital, in particular on the declaration of hospital activity and activity pricing.
  • Directorate for Legal and Administrative Information (DILA): implementation of a generative AI tool to provide initial assistance to the user by formulating a response adapted to the problems identified and directing them to the appropriate resources of the service-public.fr website.
  • Directorate of Digital and Modernization (DINUM) of New Caledonia: development of a conversational agent on the service-public.nc website to assist the people in their procedures.
  • Sport Data Hub: implementation of a tool aimed at analyzing the performance of athletes in their discipline, and estimating their chances of winning future international competitions.

A return to the entire AI ecosystem

As for previous editions, the results of the work carried out with the project leaders will be published to benefit all innovators.

The review of the work will complement the resources published by the CNIL to help professionals comply and anticipate the new European regulation on AI.

The selection committee

  • Guillaume AVRIN, National Coordinator for Artificial Intelligence (Chair of the Committee)
  • Laurent ROJEY, Digital Deputy Director General of the National Agency for the Cohesion of Territories (ANCT)
  • France CHARRUYER, lawyer – ALTIJ, DPO and president of the DATA RING association
  • Claude CASTELLUCCIA, CNIL Commissioner for Electronic Communications and Innovative Technologies, and Director of Research at Inria Grenoble
  • Thomas DAUTIEU, Director of Legal Support at the CNIL
  • Bertrand PAILHÈS, Director of Technology and Innovation at the CNIL