The test will last six months. The government will launch at the beginning of September an experiment with artificial intelligence, he said Thursday, confirming information from BFM Business. The objective: to respond more quickly and efficiently to users of public services.
"AI will inevitably transform interactions between users and administrations," says the entourage of the Minister of Public Transformation Stanislas Guerini. "The administration must not suffer, it must appropriate" AI and "transform it so that it meets the expectations of agents and users".
Hence this experiment, which will train three generative AI models (ChatGPT, LLaMA and the French Bloom) to answer users' questions, before a hypothetical generalization to the entire administration in 2024. "Content generation models will inevitably make mistakes, we will have to train them little by little, we will have to perfect them, to have the most accurate answer possible," explains the ministry.
"A help, not a substitute"
Each answer will be subject to the control of an administrative agent before being communicated to the user. To conduct this experiment, the government hopes to find 200 volunteer agents. AI is "an aid" for civil servants, "not a substitute" that would make it possible to cut the workforce, insists the government.
"The idea is not to shake up morals," still seeks to reassure the Ministry of Public Transformation, stressing that AI is already used at Pôle emploi or in the tax administration, for applications other than the relationship with users.
Read also"ChatGPT, prepare me my holidays": here are 30 concrete uses of AI
Given the confidentiality of certain data exchanged by citizens with the administration, the experiment must finally make it possible to "lift" the problems posed by AI "in terms of regulation and protection of personal data". It will therefore be conducted in partnership with the CNIL, the French data protection policeman, and the national coordinator for AI Guillaume Avrin.
Of great technical complexity, artificial intelligence systems, capable of writing in a few seconds essays, poems or translations, fascinate as much as they worry. Scientific figures have called for a moratorium on the development of the most powerful systems, until they are better regulated by law.