Synopsis

In 2015, a journalist for Slate and Libération explained how the photograph from his Wikipedia entry, published under a Creative Commons license, had been reused by various websites to illustrate articles that had absolutely nothing to do with him. Over the years, Vincent Glad — that was his name — had thus found himself turned into one of the “5 guys you date while waiting for Matt to be single again,” one of the “4 sexiest guys despite their pendant,” a “man with a degree in gender studies [who] causes panic at a party,” a representative of “brown-haired men,” or even an advertising illustration. The journalist concluded his article as follows: “Having your photo in Creative Commons on the Internet makes you enter another dimension. This photo is no longer myself; it is Riley Johnson, the friendly Australian praising the virtues of a real-estate ebook; it is Sandor, the genial Hungarian explaining the benefits of a computer course in Budapest; it is Brendan Parker, the gender-studies graduate. It is so many other brown-haired men with awkward smiles.”¹

One may assume that such images would today be generated by AI. The discourse would no longer need Vincent Glad — or any other “brown-haired (and white) man with an awkward smile.”² Generative AI systems thus illuminate, retroactively, the brutal truth of the economy of signs: that whoever exposes himself within it constitutes, in the order of their meaning, only an unfortunate accident — the correlate of a technical limitation now lifted. We could, of course, regret those archaic times when discourse still required that something like Vincent Glad exist, in order to illustrate with his face a story that, in fact, had no need of him. We can also acknowledge that this constraint has now been lifted, that the order of signs has gained its autonomy from the somewhat credulous humans who still believed they were communicating themselves through it, and ask, then, what this autonomy tells us about ourselves, about our present, and about the form that something like a political intervention might still take within it.

Overturning the Obvious

Foucault opens the first volume of his History of Sexuality, The Will to Knowledge, with a curious analytical reversal. We supposedly live, he says, under the sway of a stubborn belief according to which the Victorian age gagged sexuality, imposing silence and restraint upon it, submitting it to law, prohibition, and censorship. Foucault calls this belief the “repressive hypothesis,” and immediately shows how misleading it is. Never, in fact, has so much been said about sexuality as since the moment when this speech was supposedly forbidden. Confession, medicine, psychiatry, the tribunal: so many dispositifs that multiplied occasions not for silence, but for enunciation, explanation, and the ceaseless production of a discourse of truth about sex.

In this book, we would like to perform an analogous reversal. The thesis it will defend is that digital discursive infrastructures have not so much freed a speech otherwise supposedly held captive by the old discursive monopolies, as they have freed us from having to speak; that, on the Internet, it speaks — enough to spare us from having to do so ourselves. The age, stubbornly, likes to repeat that social networks restored to individuals the use of their voice, the admissibility of their testimony, the truth of their experience. Vincent Glad’s testimony, however anecdotal it may seem, says something else. It reminds us that the experience of the sign and the image, on digital infrastructures, is first of all a confrontation with their irreducible exteriority — that is, also, an experience of desubjectivation: a status one rereads years later without recognizing oneself in it, a stereotyped vignette of indignation one reposts without having to risk oneself in it, a format of enunciation or a series of hashtags always already available, repeated endlessly in order to pretend to enjoy a life from which one mostly expects rest. From this point of view, digital networks may be understood as a privileged site for the exercise of a certain interpassivity³: a space in which one lets signs speak, denounce, enjoy, sympathize in our place, without having to invest oneself in them any further than by pretending they are ours — while intimately knowing that they no longer are.

Once reversed in this way, the obvious unsettles the whole small metaphysics of contemporary AI and the partition around which it is organized: on one side, a human speech supposedly full, embodied, authentic; on the other, a machinic speech perceived as empty, mechanical, statistical, incapable of saying anything because it would be carried by no one. The consolation is old but desperate: to plead that, at the end of every utterance, every sentence, someone must still be standing, someone who, by his or her presence, could guarantee its origin and preserve its truth. It is precisely this partition that the book would like to make less comfortable. It will not be a matter of opposing to the statistical coldness of machines the supposed warmth of a human speech finally restored to itself, but of beginning from the opposite suspicion: that this speech has never fully belonged to those who pronounce it.

Digital infrastructures did not invent this exteriority of the sign; they made it daily, massive, almost impossible to deny. They installed a milieu in which our sentences, our affects, our indignations, our images very quickly cease to be the fragile expressions of an interiority and become traces to be taken up, compared, displaced, offered up to circulation. The success of social networks may have less to do, then, with their having allowed each of us to enter the great bustle of the world than with this: they organized enough speech around us to spare us from having truly to speak — enough signs for us still to pretend to be their subjects.

Generative AI systems will no longer appear, then, as an empty speech come from outside to profane the full speech of human beings, but as the culmination of a world in which signs had already learned to do without those who signed them. They do not destroy the authenticity of expression; they retrospectively compromise its legend. Where social networks had organized the immense circulation of speech, affects, and experiences in the reassuring form of human profiles, AI removes the scenery: it is no longer even necessary for a face, a name, a biography, or a small interiority to come and guarantee what is said. The machine does not silence the subject; rather, it reveals that the subject often served to cover the impersonal circulation of discourses with a supplement of soul. What we took to be our voice was perhaps already, most often, only a fragment momentarily inhabited by a chatter much vaster than ourselves.

A Speech Without Anyone

One may then recognize that, between the photographic little scene evoked in the introduction and the advent of generative AI, the relation is perhaps less one of rupture than of continuity. A genealogical continuity, first, since AI does not proceed from the progress of computational reason alone. It also owes itself to the immense accumulation of traces that everyone, for decades, has deposited on the Internet: images, sentences, gestures, confessions, poses, fragments of self; traces that then only needed to be absorbed, indexed, captured, and compressed into more or less sophisticated neural architectures. The existence of large language models presupposes that the Internet had already brought human speech to its point of excess; that everything, or almost everything, had already been said, said again, contradicted, commented on, repeated, displaced. The machine did not learn to speak: it taught us to inhabit the ruins of language in order to discover there, bitterly, that everything had already been said and that everything, or almost everything, could now be said again without us.

This genealogical continuity is then doubled by a more strictly logical continuity. For, from within these ruins, AI announces less the advent of a new intelligence than it reveals something about the old structures of language — and about the ways in which something like a subject may be constituted there. Its power does not only lie in producing texts, images, or voices; it lies in making visible an older truth — and no doubt a more wounding one: that the sign has never needed fully to belong to the one who emits it in order to circulate, signify, repeat itself, operate.

This exteriority of the sign, we know, is not a defect of language; it is, as Derrida showed, the very condition of every mark. A sign is valid as a sign only because it can function in the absence of the one who produced it: be cited, diverted, repeated in contexts that no presence can guarantee. The iterability of the mark is not what corrupts a speech originally full and present; it is what constitutes it. Lacan is perhaps the one who drew from this reflection the most accomplished conclusions: the subject does not sovereignly precede language, like an intact interiority that would subsequently come to express itself there; it is caught within it, divided, represented by signifiers that precede and displace it. It speaks through us.

The scandal is therefore not that machines speak; it is that, in speaking, they keep open the internal division constitutive of every subject: that they expose us to it, daily, stubbornly, even in those ordinary forms of speech — writing an email, sharing a post, preparing a class — in which we had grown used to taking refuge in the small fiction of our sovereignty. Social networks still offered us the benefit of the illusion: letting us believe that others nevertheless believed in the reality of our interiority — even though our words could in fact appear to them only as the umpteenth repetition of the same statement, the same indignation, the same desire with which their feed was already saturated. Generative AI denies us even this politeness; repetition no longer needs us there at all, and thus appears in its brutal truth: as the impersonal place from which we had already too often been speaking without wanting to know it.

Politics of Desubjectivation

TO BE REWORKED

This fiction of the subject will therefore run through this book as it runs through the age: less as a theoretical error than as a stubborn, almost administrative demand. Everywhere, there would still need to be voices, opinions, testimonies, wills, acts of speech; everywhere, someone would still need to recognize themselves in what is said, assume it, sign it, answer for it. But this demand now comes too late, or too badly. The same infrastructures that demand speaking subjects never cease to undo them, converting their words into traces, their affects into formats, their gestures into series, their styles into models. The problem, then, is not that we have lost a full subject that would need to be restored. The problem is that our political language continues to speak as if this subject were self-evident, at the very moment when the dispositifs that organize speech make its unity less credible every day.

Nor will it be a matter, however, of celebrating desubjectivation as a deliverance. That would only invert the old sovereign fiction. Not every release from the self amounts to emancipation; not every dissolution of the subject produces the common. Certain forms of desubjectivation may open a way out of the obligation to be oneself, out of the imaginary ownership of speech, out of the small theater of personal expression. They may make room for another way of speaking, listening, being affected, composing with what passes through us. But other forms do no more than administer this release: they relieve the subject of having to speak, to feel, to answer, until experience itself is taken over by the automatic circulation of signs. The political question begins exactly there: in the necessity of distinguishing a desubjectivation that opens from a desubjectivation that exempts.

This book will therefore seek less to defend the human against machines than to displace the very categories from which we still believe we can defend it. If signs circulate without us, if it speaks before us, through us, sometimes better than us, then we must stop grounding politics in the reassuring fiction of a subject who owns its voice. We must ask which dispositifs still make room for a gap, a void, a possibility of address or reprise, and which saturate in advance the space in which something like subjectivation might occur. A politics of desubjectivation would therefore consist neither in mourning lost sovereignty nor in rejoicing at its disappearance; it would seek to think, within the very collapse of that sovereignty, the fragile conditions of an intervention that is not immediately taken up, spoken, digested by the impersonal flow of discourses.

Conclusion

This book will therefore not seek first of all to decide whether AI systems are intelligent, creative, or dangerous in the sense in which public debate ordinarily understands those terms. Rather, it will propose to understand them as revelators: revelators of the exhaustion of a certain economy of speech, of the fragility of the subject supposedly owner of its signs, and of the way in which digital infrastructures have made thinkable — and then technically possible — a speech without a speaker. Through Foucault, Lacan, Derrida, Pfaller, Žižek, and the recent history of social networks, the question will be to understand what becomes of politics when signs no longer need us in order to circulate.

Footnotes

1. http://an-2000.blogs.liberation.fr/2015/11/13/comment-je-suis-devenu-la-cible-du-gorafi-feministe-americain/ The article has since been deleted. ↩

2. Vincent Glad has since been implicated in the “Ligue du LOL” affair. We will therefore be careful not to regret too quickly the technical possibility of his disappearance. ↩

3. In the sense given to this term by Robert Pfaller and then Slavoj Žižek. ↩