Artificial intelligence, a danger for democracy?

Several projects are already aiming to entrust part of the legislative work to artificial intelligence (AI). While for the time being, these are mainly tools for transcribing parliamentary debates, translating or analyzing archives, drafting tools or legislative texts are also being considered. The use of AI may seem interesting: it can improve readability or facilitate standardization. But, depending on how it is used, it may not spontaneously respect the fundamental principles of the legislative process, such as legality, proportionality or equality. These requirements, established by the Constitutional Court, are the subject of numerous case law decisions. As part of a doctoral thesis, Aline Nardi, is sifting through them to see whether the use of AI is possible in legislative processes.

His finding? "For csome more critical tasks, which go beyond simple editing, problems arise." In particular, she points to the opacity of the reasoning, since "it is difficult to trace why the AI suggests such and such a lead from the query submitted to it". Yet, in a state governed by the rule of law, the legislative process must be transparent and comprehensible if it is to be challenged by parliamentarians and citizens.

There are also issues of democratic legitimacy: "The company that develops the system used by parliamentary assemblies can influence decisions that are extremely important for a nation. Is a parliamentarian still useful if he or she is only there to ratify decisions made by a machine?" asks the jurist.

On the citizens' side, AI has been transforming election campaigns for some years now. Micro-targeting, automated disinformation, manipulation of emotions via social networks: these are all formidable tools that can skew public debate. The 2016 US presidential campaign clearly illustrates this problem."Cambridge Analytica, for example, targeted and influenced undecided voters whose positions could be tilted in Donald Trump's favor. Around 20 million undecided voters thus received tailored messages, such as memes, videos, blog posts in favor of Trump", recalls Aline Nardi. Combined with social networks, AI therefore calls into question the right of voters to form an opinion based on pluralist and reliable information.

More recently, the Romanian Constitutional Court invalidated the 2024 presidential election on the grounds of misuse of AI and lack of transparency in digital campaigns. "This affected the freely expressed nature of the citizens' vote and in particular their right to be properly informed, explains Aline Nardi.

The European Union is trying to take action against these abuses. Several regulations have been put in place. They aim to frame digital technologies and limit their harmful effects on the information space, particularly during electoral processes. These regulations include the Digital Services Act (DSA), the AI Regulation and the Transparency and Targeting of Political Advertising Regulation (TTPA). These regulations have prompted Google, and more recently Meta, to stop allowing political, electoral and social issue advertising in the European Union.

"However, questions remain as to the applicability, concrete implementation and effectiveness of certain standards stemming from bodies of digital law", alerts Aline Nardi, who calls for "sortir de ce flou juridique".