Artificial intelligence is entering the public service! 1,000 volunteer agents will experiment with the use of generative AI in order to provide better answers to users. The public service of the future?
The release of ChatGPT created a tidal wave on the Internet, but also within tech companies. Each has begun to integrate generative AI into its services and products, embarking on a veritable race to show who has the biggest one. But after having invaded the private sector, here it is also arriving in the public sector. Indeed, the Minister of Transformation and Public Service, Stanislas Guerini, kicked off, on Thursday October 5, of an experiment with artificial intelligence in the public service. 1,000 volunteer agents – in the Family Allowance Funds (CAF), Retirement Insurance, prefectures, etc. – will be equipped with a tool using AI – more precisely Claude, an “ethical” conversational robot developed by the American start-up Anthropic (see our article) – to support them in writing online responses to users – Obviously, they will supervise them and modify them if necessary before sending them. The Government thus hopes that agents will be able to devote “more time for humans”.
AI: a more important experiment than expected
The Government had already announced last May – very discreetly on the twentieth page of the concluding document of the seventh intergovernmental committee on public transformation – that the use of generative AI would be tested in public services, as noted BFMTV. He thus authorized the “launch of an experiment with artificial intelligence tools to produce more complete responses to users while placing less burden on agents on the ‘Services Publics +’ platform and in several French services” in order to ensure the “quality of service” and the “user satisfaction”while weighing less on public sector agents.
This experiment was initially scheduled to begin in September with a sample of 200 agents from France Services houses – local counters located throughout the country which bring together different State services, such as Pôle emploi, La Poste and taxes – and was to last six months – it was therefore not a definitive implementation. The experimentation therefore takes on much greater proportions. Employees had to be able to use ChatGPT, but also Bloom, an international open source model developed by three French people, and LLaMA, the AI model developed by Meta – again, the plans were modified, since Claude was ultimately chosen.
AI: greater quality of service and relieved agents
Concretely, if we wish to obtain information – on the procedure to follow to receive unemployment compensation for example – the agent will submit our question to the AI using a “Generate response (AI)” button, integrated into the Service-public.fr platform, in order to quickly obtain a clearer or more complete response than the one he would have given under normal circumstances. He will then have to evaluate the relevance of this response to better correct it behind, and thus provide us with the most appropriate response possible. To avoid any data collection by companies and any possible leak, agents should only ask general questions without ever providing personal or confidential information. The Government promises that the data will be completely anonymized and that confidentiality will be completely preserved for the user. Everything is supervised by the National Information Systems Security Agency (ANSSI) and the National Commission for Information Technology and Liberties (CNIL). Work is also underway to develop a new sovereign model internally, based on an open source model, currently called Albert. More French civil servants will also work and be trained on the subject of generative artificial intelligence.
The Government wants to be reassuring, promising that AI will not replace agents, who will always have to check and adapt the answers provided. A legitimate fear given that the astonishing development of this technology promises to be a real revolution for the world of work and risks making certain professions obsolete, particularly in the tertiary sector (see our article). Japan has also started a similar approach since the country, to compensate for the shortage of labor in its administrations, has also experimented with generative AI at the town hall of Yokosuka, south of Tokyo, for a period of four weeks, before officially adopting the use of ChatGPT for its employees in back-office operations. The goal is to allow agents to save time by summarizing meetings or correcting communications, so they can focus on other tasks.
This news comes at a time when governments, including in France, are showing a certain distrust of AI, particularly in data collection, which could lead to interference from foreign countries. The town hall and the metropolis of Montpellier, for example, received the order, last March, to no longer use ChatGPT “as a precaution” And “as a precautionary measure”, as reported by Free Midday. Along the same lines, streaming services and mobile games have been officially banned on the professional smartphones of French civil servants (see our article). At the end of January, the CNIL set up an artificial intelligence service (SIA) intended to supervise the use of this technology and respond to the problems it raises.