AI, the web, social networks and more generally digital technologies are transforming society, for the better, but also for the worse. They facilitate daily activities for searching for information or writing texts, automate painful or tedious tasks. They also contribute to the progress of knowledge in science, health or space. On the risk side, we can fear the disappearance of certain jobs and an increase in unemployment . But what about the threats to our freedoms and our democracies and how can we protect ourselves from them?

Regulation to the aid of individual freedoms

Since February 2, 2024, the first part of the European regulation on artificial intelligence, or “AI Act” has created a regulation with a view to protecting our fundamental freedoms.

The European Union now bans three types of AI practices:

• subliminal techniques,
• social credit ( which assesses the solvency of individuals for the granting of credit),
• real-time identification of biometric techniques .

The AI ​​Act complements other regulations that aim to protect individuals from abuse by states or large companies, while promoting a vigorous, sovereign and independent European industry. However, one may be surprised by the fictitious nature of some potential bans on artificial intelligence. In this respect, the first of them, which concerns the use of subliminal techniques, is revealing.

Let us recall that the term subliminal was introduced in 1957 by an American publicist, James Vicary , who claimed that by inserting images containing messages into a film every fiftieth of a second, these messages influenced our behaviour without us being aware of it. However, nothing has highlighted the tangible effects of subliminal manipulations , except in a very marginal way. Nevertheless, the European institutions, very concerned about the autonomy of individuals, are already concerned about it, which may seem excessive.

Fake news and citizenship

Beyond the protection of fundamental freedoms – freedom of opinion and expression, freedom of thought and religion, freedom of assembly, freedom of movement – ​​it is important to emphasize that democracy resides above all in the ability of individuals to take part in political life, to deliberate, to exchange freely in the public space. However, AI, by contributing to the fabrication and selective dissemination of false information, harms this collective deliberation.

The spread of so-called “fake news” interferes with electoral processes by introducing confusion into people’s minds, which destabilizes democratic institutions. And AI contributes to this destabilization in three ways:

• by producing huge masses of fake news,
• by creating images, sounds and videos that create illusions,
• by profiling individuals and targeted dissemination of information to specific segments of the population.

Mass production and distribution

Libels, canards, propaganda, disinformation: the creation of misleading opinions is not new. However, the weight they have acquired with AI is increasing disproportionately. In the past, writing a convincing text took time. And drawing a caricature even more. As for disseminating it, this required either subtle messengers responsible for instilling the poison, or expensive infrastructures, press organs, radio, television, etc. Now, anyone, with generative AI techniques, produces texts, images and sounds at will and transmits them freely to the whole world on the web, almost for free, without asking any permission.

Image and sound generation

These contents produced by AI techniques look very much like photographs or videos. They are then presented, to use the semiotic categories introduced by Charles Sanders Peirce (1839-1914) at the turn of the 19th ^and  20th ^centuries  , as “badges”, in other words as signs pointing to the realities they would designate.

They give the impression of being its trace, in the same way that photographs are imbued with the luminous trace of things. But where photographs attest to the things they show, the images produced by generative AI aim to counterfeit reality, which gives them a particularly pernicious character.

Profiling and targeting

Finally, by profiling individuals using machine learning techniques and then screening the population, social networks send each segment of the public the counterfeit information most likely to provoke a reaction in order to increase the probability of retransmission and, therefore, the circulation of information.

Thus, during the last presidential campaign in the United States, fake videos showing Kamala Harris promoting the positions of Benjamin Netanyahu were sent to Democratic voters while other videos showing her alongside the Palestinians were sent to Jewish and conservative circles, which caused a lot of emotion and exchanges. This is what Bruno Patino calls, in a figurative way, the “goldfish civilization”: we each live in a very comfortable informational fishbowl, where we exchange with others who receive the same information as us and share more or less the same ideas.

End of citizenship?

The consequence is that society is becoming increasingly fragmented, that public space is gradually disappearing since the different components of the population no longer share the same information. From then on, collective deliberation cannot take place, since there is no longer agreement on the facts. The exercise of citizenship, in the sense that philosophers like Hannah Arendt understand it, namely the participation of free individuals in collective debate through the exchange of arguments, is becoming increasingly difficult. Ultimately , AI, made available to major digital players, creates a risk of the disappearance of citizenship .

Is this, however, inevitable? AI, if used well, could just as well contribute to restoring to citizens their power of control over the information that comes to them and thus to rebuilding a space for collective deliberation where “men of good will” could work.

So, we can try, on a given subject, to identify using AI the different pieces of information that are circulating and to compare them, as a journalist does who must, on each subject, look for the sources.

Indeed, AI is not only used to automatically generate texts: it calculates the proximity of use of each word (or each part of a word, what we call tokens ) with others. This makes it possible to compare a word with its synonyms, a paragraph with a second that has the same meaning, in order to know what is said on a given theme. It is therefore possible to compare the different points, by identifying the oppositions.

In short, if used well, AI techniques may allow everyone to take part in collective deliberation, without being deceived, and thus to fully exercise their citizenship. All we need is the will and not to equate democratic ideals with fundamental freedoms alone.

To this end, European legislation must not be content with protecting fundamental freedoms. It must establish the conditions for exercising citizenship in a common public space, which involves combating fake news. The purpose of the DSA ( Digital Services Act ) should have been to meet this need. The latter merely denounces violations of common law (for example, hate speech ) in cyberspace, but this remains largely insufficient. It would be appropriate to go further in order to combat lies, while protecting freedom of expression.

Author Bio: Jean-Gabriel Ganascia is Professor, Artificial Intelligence, Cognitive Sciences at Sorbonne University