How to improve the reception to users in the public service? Like a private company, the State intends to facilitate the entry into contact of users with its agents. This is one of the commitments included in its Public Services+ transformation program.
In this context, Stanislas Guerini, Minister of Transformation and Public Service, presented, on July 18, his “telephone plan” to the ministries and public service operators concerned. Objective: to obtain a drop-out rate of 85% by December 2024. To measure the progress made, the satisfaction of users will be evaluated following a telephone contact in the coming months.
This Telephone plan includes simple and common-sense measures such as strengthening the readability and display of telephone numbers on public services websites, in particular in order to better take into account disability situations. New features will also be implemented such as making appointments online and reminding users to limit the waiting time on the phone.
ChatGPT, LLaMA and Bloom in test
More original, the Administration intends to use artificial intelligence to improve the response provided to users. An experiment is launched in this direction by the interministerial directorate for public transformation (DITP) and the interministerial directorate for digital technology (Dinum). More than a thousand agents have volunteered to test, starting from the start of the new school year, this AI tool whose impact of use on response times and their quality will be measured.
Through this large-scale test, the agents will use, explains the ministry, ” answers generated by artificial intelligence by completing and personalizing them in order to better accompany the user and thus free up time for answers that require specific attention. »
According to BFM Business, this experiment will focus on three text-generative AIS including the famous OpenAI ChatGPT, LLaMA, the AI model developed by Meta, the parent company of Facebook and Bloom, an open source project supported by Hugging Face, a startup created by the French.
Ensuring the confidentiality of exchanges
Quoted by Public Actorss, Guillaume Coldre, Ministerial Administrator of data at Bercy, specifies the scope of their use: “ Nothing confidential or sensitive should be promptly [le texte de la requête, NDR] in a generative AIe”. Moreover, he recalls that the “ Generative generalist AI can provide erroneous answers: they are seen more as a basis for reasoning than as a basis for knowledgee”.
According to AFP, the experiment, given the sensitivity of the information exchanged between the citizen and the Administration, will be supervised by the Cnil and Guillaume Avrin, national coordinator for artificial intelligence. AI is ” a help “for civil servants,” not a substitute “who would make it possible to trim the workforce tempers, moreover, the Government.
Improving the quality of service would not be the only field of application. AI must help public service IT specialists by automatically detecting and correcting bugs in their code. Other avenues envisaged: the drafting of job descriptions or job offers or ensuring the completeness of a file within the directorate general of public finance (DGFiP).
Propose an alternative to online procedures
Users will have their say on the Phone plan and this AI experimentation. In mid-June, a new version of the Observatory for the quality of online procedures was launched. Public services are evaluated according to different criteria such as their availability, the simplicity of the language or the consideration of the disability.
According to the principle of “tell us once”, these steps must not repeatedly request the same supporting documents from users, a fortiori when the administration already has it. Moreover, to avoid any form of exclusion, these approaches must systematically propose a non-digital, physical or telephone alternative. Provided you can reach an agent on the phone…