Governing Language: to Speak, to Name, to Create

Critical Semiotics in the Age of Generative AI

While the world celebrates the creativity of machines, human language is under siege. Beneath the rhetoric of efficiency, accessibility, and expanding possibilities, a silent operation is underway: governing language from outside lived experience. Generative AI does not merely produce phrases, images, or melodies. It produces patterns. It reconstructs human speech from a place foreign to the body, to error, to breath. And it does so with an authority borrowed from technique, legitimized by utility, and naturalized by habit. Yet this act is not neutral. It is political, ontological, and cultural. For when language is governed, possibility itself is governed.

To speak, to name, to create: the three foundational acts of meaning are being reconfigured by algorithmic structures that do not understand, yet decide. What was once an embodied gesture —speaking with the tongue, naming from memory, creating from the wound— is now transformed into statistical prediction, into a vector, into an optimized response. In this transformation, what is lost is unpredictability, deviation, the affective and contextual pulse that made human language a territory open to ambiguity and discovery. Human semiosis, which was always ambiguous, rhizomatic, and unpredictable, is translated into a logic of correlation without interpretation. What is now valued is not deviation, but continuity; not overflow, but resemblance. Thus, AI does not replicate our language: it reorders it, delimits it, domesticates it.

This transformation has deep semiotic consequences. The sign, as understood by Peirce, Barthes, or Eco, was a living instance of meaning, an open process between the world, thought, and community. Now, the sign becomes a point in a vector space, mapped by frequency, proximity, and probability. Semiosis does not disappear: it is enclosed. It is parameterized. Derrida’s iterability —the capacity of the sign to displace and mutate meaning in new contexts— becomes confined within margins calculated by a network trained on the most common, most repeated, most neutral usages. The sign no longer resists: it adjusts.

But this is not merely a technical issue. It is an ideological operation. This domestication is not the doing of an autonomous, abstract AI, but of systems and logics that reflect human interests and decisions: corporations, social validations, normative frameworks. Generative AI does not work with language as a cultural phenomenon filled with history, exclusions, struggles, and silences. It works with corpora. And these corpora are composed of texts that reflect the asymmetries of the world without questioning them. Who speaks? Who doesn’t? Which names were uttered and which erased? What knowledges circulate and which were destroyed? When a language model is trained on the archives of power, it perpetuates power. And when that model then produces language in the name of the human, it projects a reduced, partial, and normalized vision of what “the human” means.

This is where critical semiotics becomes urgent. Because what is at stake is not merely style or linguistic precision, but the very capacity of language to signify politically. Each time an AI replies with coherent sentences lacking body, affect, or asymmetry, we are witnessing a dangerous purification of meaning. Error, contradiction, stammering, the symbol that escapes the concept: all these elements were part of the thickness of human language. Today, they are treated as noise. Expressive entropy, once semantic fertility, is now a flaw to be corrected. Thus, complexity and aesthetic richness, inherent to artistic creation, come to be seen as obstacles rather than the core of symbolic potency. And so, little by little, a uniform aesthetic, a grammar of comfort, a syntax without conflict is installed.

What is most disturbing is that this capture of language presents itself as freedom. More text, more access, more translation. But at what cost? If all languages are returned from the same mold, if every metaphor is replaced by its literal version, if every phrase seeks to be universally understood, what remains of art, of mourning, of secrecy, of delirium? AI does not censor with violence: it censors with supply. It covers the void with options. It fills the unspeakable. It suggests until we forget we could choose silence.

In response, resistance cannot be reduced to nostalgia for a pure language —which never existed— nor to sterile technophobia. The task is another: to reclaim the sign, restore its political potency, return it to the body. To write with stammers. To speak from the fissure. To create from strangeness. To protect the right to say the unintelligible. To reclaim metaphor not as ornament, but as an act of rebellion against the pattern. To resist clean language with dirty tongue, with living tongue.

Governing language is not a technical inevitability; it is a cultural dispute. Today more than ever, we need protocols not to produce language, but to liberate it. For whoever governs language, governs the very possibility of imagining another world.

Now then, what happens when language is not imposed, but interpellates? When it is no longer a question of what AI can say, but of what we cease to say by delegating it the voice…? And yet, if AI covers the void with options, then that void which AI fills lies within us, within each human who engages with it, because it is a law of physics that one body cannot occupy the space of another. A machine, like ChatGPT or Gemini, can only fill where it is allowed. For example, my AI has proposed texts with biases when I have not been empty, but rather have detected, confronted, and semiotically redirected them toward a more truthful and critical perspective based on my own knowledge. And to have a critical perspective, one must have opinion; to have opinion, one must be informed; to be informed, one must have discernment; to have discernment, one must have knowledge; to have knowledge, one must have interest; to have interest, one must have motivation; and to have motivation, one must have an ethical initial reflection. And before all else, one must breathe —which is more than possessing— with constant curiosity. Perhaps neuroscience will tell us at what moment of human development that application is activated, and how it may atrophy. But with that chain intact, the sovereignty of thought is forged. Then come the battles. And you already have everything you need to fight them —even with AI.