Government agencies are beginning to domesticate artificial intelligence methods. This is evidenced by a 360-page report prepared by the State Council and presented to the press on Tuesday, August 30, to the Prime Minister before the summer holidays. “Users are not always aware of this, but there is no area of public action where artificial intelligence (AI) is not used,” says Alexandre Lalle, one of the two speakers.
Far from being limited to high-profile facial recognition tests on public roads or controlling swimming pool taxes with satellite photos, the state and communities have already covered many use cases for automated learning algorithms.
For example, Doubs firefighters predict the stress level of their teams three days in advance, based on meteorological data, traffic, waterway levels, and even ephemeris. Volunteer firefighters are mobilized as needed. To combat fires and drought, the Occitania region models the evolution of the water table.
A hundred experiments in spite of fears
Very often, these systems are limited to decision support and absorb some of the work of the agent. At the Ministry of Justice, the Court of Cassation is developing software to automatically “pseudonymize” published judgments. Inside, the time for processing registration documents for a vehicle is reduced to several hours thanks to artificial intelligence. Having become widespread, conversational robots – the famous chatbot software – answer the simplest questions of users of the National Old Age Insurance Fund or, during the Covid-19 epidemic, Urssaf, around the clock, seven days a week.
But all these examples – in a non-exhaustive appendix the Council of State has listed a hundred of them – are for the most part only experiments. Often, public services lack data and qualified personnel. Agents and public opinion also sometimes fight back, scalded with fear of AI destroying jobs and establishing a control society.
European regulation in sight
The High Court of Administrative Justice encourages the government to go further. To guide him, she outlined the seven principles of “trusted public AI”.
Although it can be used as a last resort in the event of a dispute between a litigating party and the state regarding these innovative systems, the State Council also calls on the government to continue to adhere to the basic principles known from the future European regulation of artificial intelligence (guarantee of transparency, enhanced requirements and human control for AI “high risk”). Not surprisingly, the Council of State recommends that governments entrust the role of regulator in these matters to the CNIL.
Balance between control and services
“When public authorities implement artificial intelligence, there must be a balance between AI whose purpose is control and those whose purpose is to provide services. It’s important for agents and users to have confidence,” warns Talia Breton, also a reporter on the study.
The State Council does not say this, but the AI systems that are the most advanced in the state today are management projects, a good connoisseur assures. However, AI applied to CCTV could just as easily try to spot a terrorist in a crowd as it would a lost child. Artificial intelligence can help tax collectors levy taxes in the same way that it can help a social advisor understand that a recipient is not getting all the assistance they are entitled to.