Increasingly, public service officials communicate with citizens using artificial intelligence, which can be worrying, given the sensitivity (Taxes, Family Allowances and Health Insurance) of the actors concerned.
Will humans end up being replaced by machines, in this case by artificial intelligence? This is obviously one of the major societal questions of recent decades, but it seems that today, we are touching on it more than ever. In its November 21 edition, the 20 Heures news on France 2 focused on the growing use of AI in public services. Although technology helps save time on certain tasks, it is far from providing complete satisfaction. Should taxpayers be worried about this?
French public officials respond to you using AI
As for the Mutualité sociale agricole de Reims, the activity obviously pushes employees to discuss very regularly with individuals. To answer their questions, the Reims MSA has a “Generate AI response” button.
For a question related to a work stoppage, simply pressing this button provides a ready-made 13-line answer, in one click and in a flash. The message is particularly convincing. “ Hello, we understand your situation regarding the processing of your sick leave (…). Without going into specific details, please note that we do everything we can to process requests as quickly as possible. (…). Your feedback is valuable to us in order to progress ”, we can read in particular in the response.
For Sonia Haigneré, service relationship manager at MSA, “ it’s completely amazing, it’s a magical tool “. She estimates that today, artificial intelligence helps her accomplish this task five times faster than before. The time saved allows him to concentrate more on files that require more support. Today, 1,000 public officials are testing the artificial intelligence platform in France.
Artificial intelligence that generates satisfaction and also raises concerns
On the side of the Interministerial Directorate for Public Transformation (DITP), we report a higher satisfaction rate of those administered in the responses provided by the AI, which seems to learn quickly, compared to those made by very human agents . The next step ? That agents can use it directly at physical counters, whether at taxes, in CAF or at Health Insurance.
However, the system today has its limits. As soon as we complicate the request a little, for example on a request for information after a driving license suspension, a slightly more complex case, the software… invents, or “hallucinates”, as is customary for say it for chatbots. The algorithm does not always have all the answers.
In Denmark, let us not forget that the government was forced to resign in 2021 for having used algorithms, for several years, to detect family benefit fraudsters, by drawing up a typical suspect profile (working-class housing, more than two children, binational couples), using statistical shortcuts. Aid was brutally cut off, without carrying out the necessary checks, with 35,000 families wrongly accused. Enough, perhaps, to temper the movements of public administration, regarding an all-out use of language models.
Source : France TV info
11