Freedom of expression put to the test of algorithms.

In an article published on January 20, 2025, in Le Monde, the philosopher Monique Canto-Sperber examines the impact of platform algorithms and social networks on public debate. She contrasts two conceptions of freedom of expression: the French one, framed in particular by the body of press laws under which certain cases—such as racist or sexist insults, incitement to hatred, or the dissemination of false news—are subject to prosecution; and the American conception, enshrined in the First Amendment of the U.S. Constitution, which states that “Congress shall make no law (…) abridging the freedom of speech.” Within this framework, statements that would be judged in France as criminal offenses may, across the Atlantic, be regarded as major contributions to public debate, even when they are racist or harmful.

Independently of the philosophical positions one may hold on the question of freedom of expression—which, in such matters, can often be defended by arguments of roughly equal weight, as current debates clearly demonstrate—it is no longer possible, as Canto-Sperber’s article rightly points out, to address the issue without taking into account the profound transformations that the media sphere has undergone with the emergence, in the 1990s and 2000s, of the web, search engines, and social networks. It is in this context that the Digital Services Act (DSA) emerged in Europe, fully applicable since February 17, 2024. Its objective, according to the now familiar formula, is to ensure that “what is illegal offline is illegal online,” by requiring platforms, among other obligations, to provide users with tools to report unlawful content and to implement moderation measures. In other words, the DSA not only assigns platforms a certain degree of responsibility regarding the legality and potential harmfulness of published content, but also imposes binding obligations aimed at limiting the proliferation of illegal material, such as hate speech or disinformation. By contrast, Section 230 of the U.S. Communications Decency Act (CDA) of 1996—cited by Canto-Sperber—grants platforms immunity from editorial responsibility for content published by third parties, placing responsibility primarily on the author while encouraging platforms merely to remove defamatory or unlawful material as quickly as possible.

It is within this framework that recent decisions by Mark Zuckerberg must be understood, particularly his questioning of the effectiveness of fact-checking services and his denunciation of what he perceives as their political bias. In an effort to promote diversity of viewpoints and to limit what is perceived as liberticidal censorship—contrary to the vision of free speech defended by many Republicans—there appears to be a growing tendency among major tech companies to favor user-driven self-regulation, for instance through community annotations on information deemed inaccurate, as is the case on Elon Musk’s platform X.

Yet, as the French philosopher quite legitimately asks, does the total freedom of expression advocated by tech giants such as Meta or X truly exist, when the very dynamic they claim to defend is simultaneously undermined by the opaque algorithms at the heart of social networks and digital platforms? These algorithms are not designed primarily to serve democratic debate or to allow everyone’s voice to be heard, as figures like Zuckerberg or Musk might suggest, but rather to sustain an economic model whose primary function is “to maximize advertisers’ exposure” by promoting emotionally charged messages that generate the highest number of views and clicks.

Far from fostering democratic deliberation and exposure to diverse perspectives, algorithms instead reduce opportunities for constructive dialogue by amplifying certain messages at the expense of others. As Canto-Sperber puts it, it is as if “in the public square, some were speaking through a megaphone while others were speaking through a mute.” Especially when certain actors—no need to name them explicitly—use their own platforms as tools of political propaganda in support of a particular Republican candidate. Speech may indeed be free, she continues, but often without impact for most voices—a freedom of speech that no one hears: “Free speech is not free reach.”

Should mechanisms of amplification therefore be limited, or should a greater element of randomness be introduced to mitigate the problem? In other words, should algorithms themselves be regulated? In a world where the stakes are increasingly economic, political, and geopolitical, one might fear that the proposals of Monique Canto-Sperber, a leading figure in ethical philosophy, will remain unanswered—unless users, consumers, and advertisers themselves decide to abandon certain social networks.