This column criticizes calls for the total rejection of generative AI, deemed essentialist and counterproductive, because they ignore the diversity of infrastructures, uses and possible alternatives.
This text offers a response to several recent forums, which call for the large-scale rejection of any form of generative artificial intelligence (AIg): “Faced with the hurricane of generative AI, we have two or three years left to act” (Éric Sadin, Libération, October 2025); “Faced with generative AI, conscientious objection” (Ecopol Workshop, November 2025); “Faced with the disaster of generative AI, an abrupt, radical critique is necessary” (Abel Quentin, Le Monde, December 2025). Examining their main arguments here, we posit that the issues raised by AIg deserve better than an essentialist critique, which is not only ineffective, but above all leaves the way wide open to the perils it intends to combat.
Ecology
The argument for the overconsumption of AIg energy is due to the lack of separation between the production of models (LLM) and their uses, and above all the lack of distinction between several infrastructures. A query on ChatGPT, for example, has an environmental impact 50 times greater than an equivalent query on a local and open source model like Llama. This fundamental difference invalidates the amalgam made by Ecopol between platform capitalism and decentralized offers.
Sovereignty
If the geopolitical importance of controlling AIg models is considered by Ecopol, this idea is immediately rejected due to financial and industrial difficulties. This position amounts, in fact, to abandoning the field and therefore to further strengthening the positions of the major US and Chinese players since a ban on AI on an international scale is inapplicable. The Swiss AIg model Apertus, for example, was only trained on royalty-free data and each stage of its development was documented to make the whole thing auditable. It shows that other paths are possible.
Stupidity
Eric Sadin’s text is not afraid to proclaim that “AI is the Trojan horse of self-denial” and to worry about “hordes of aphasic beings who are coming and (the) dismal standardization of the world which is coming.” The Ecopol text even brings the Web closer to AIg, both of which would contribute to a democratic peril and a reduction in knowledge. This dramatic story ignores the long history of the Web, which includes dynamics of emancipation and expression of minoritized, disadvantaged and threatened populations. In other words, cutting off access to such infrastructures, imperfect and toxic in many respects, but also complex and nuanced, would amount to consolidating the privileges in place.
Forfeiture
Above all, forums like those of Eric Sadin reproduce a very old gesture, that of moral conspiracy in the face of supposed decadence. For centuries, the collapse of culture, of civilization, even of humanity, has been predicted. Writing, says Socrates in the Phaedrus, destroys memory and weakens authentic thought. The novel corrupts youth and dissolves social values. Cinema reduces the literary imagination. The Web destroys the reliability of information by allowing as many people as possible to express themselves. Video games are nothing but violence and stupidity. Without being totally erroneous, hence their convincing force, these formulations implicitly refer to the idea of a lost purity. They erase variations, alternative configurations, possible appropriations.
Disconnect
The decisive choice is not between the naive acceptance of transhumanist AI (“pro-AI” speeches must also be considered) and the disconnectionism that these forums offer us. This binary dialectic assumes that there is a flow to be accepted or stopped, a direction that should either be celebrated or prohibited. But reality is infinitely more complex and less directional. Technologies never advance in a straight line. They create spaces of possibility that open and close. They generate resistance, diversions, unpredictable uses. Generative AI is neither the technological messiah that some proclaim nor the civilizational disaster that others denounce. It is an unstable assembly of technical configurations, investment choices, social practices, political appropriations.
Creation
This is why the only relevant path is neither naive acceptance nor moral refusal. It’s experimentation. Concretely experience what AIg does to thinking, writing, language. Not to judge it in advance, according to established moral principles, but to transform oneself in this encounter – like the “heteronyms” of Fernando Pessoa who, already, posed the question of who, or what, writes. The absence of a stabilized subject does not mean the absence of creation, but another form of creation, an acceptance of what we only half control. Language never belongs to a single being. When we write with a language model, this multiplicity becomes tangible, impossible to deny.
Transformation
The objection will not disconnect AI: it will disconnect its authors from the possibility of directing its development towards emancipatory goals. Because this is where true radicalism lies: taking the problem to its roots in the acceptance of risk, in overcoming what threatens us, in the self-transformation that results from it. It is a difficult, uncertain path, which offers no guarantees. But it is the only path that remains alive because, by disconnecting from AI, we will not disconnect it and we will then close a possibility that we have not even half-opened.
Authors
Grégory Chatonsky (artist)
Anthony Masure (research manager, HEAD – Geneva, HES-SO)
