Among everything that artists can create with artificial intelligences, a new form is discreetly appearing: the skill, a text that teaches a machine a way of doing. Writing one’s method rather than one’s work places this practice within a history running from the verbal scores of Fluxus to the wall instructions of Sol LeWitt, and within yet another history, that of the layers of abstraction in computing. To this I attach a proposal for the cultural sector: to legitimise the skill as a work, with free licences as the framework of its circulation.
Writing one’s way of writing
I have always written. Articles, books, screenplays, working texts, and this daily practice of writing, by hand so to speak, is the foundation of everything that follows. Since the arrival of generative artificial intelligences, one more tool has joined my table. Some texts I write with it, others without, depending on the subject, depending on what the text calls for, and there is no conversion and no renunciation in this, simply a palette that widens. It is by practising this tool over time that I came to understand from the inside what a skill is.
For to write with a machine without losing your voice in it, you have to teach it your voice. So I gradually built a document that encodes my way of writing. It contains my principles, trust in the content and in the reader, the refusal of dramatisation, but also a repertoire of flaws to hunt down, fed by my rereadings, with for each of them examples of faulty sentences and their correction, ways of conducting research and of connecting things, and a working sequence in several passes. Each time I correct a text produced with the machine, the correction joins the document, which grows sharper. What is called a skill, in today’s technical vocabulary, is exactly that, a structured set of instructions, examples and counter-examples that a model loads before working, and which durably inflects what it produces.
What strikes me most in this experience is what it does to me. I learn a great deal from reading the articles I write this way, because the machine, applying my own instructions for research and for connecting things, brings me connections I would not have made on my own. And over time, this document has taken on, in my eyes, an object of strange status. It is not a work, since it is not read for its own sake. It is not quite a tool, since it does nothing on its own. It is a way of doing, written down, and it was in rereading it one day the way one rereads a text, with its finds and its clumsinesses, that the question came to me: what if this form were, in itself, a form of creation?
“The idea becomes a machine that makes the art”
Artists prepared this question long ago. In 1967, Sol LeWitt published in Artforum his “Paragraphs on Conceptual Art”, which contains a sentence that became a programme, “the idea becomes a machine that makes the art”. LeWitt drew the most concrete consequences from it with his Wall Drawings, begun in 1968: the work is not the drawing on the wall, it is the written instruction, a few lines prescribing strokes, directions, densities, which assistants or museum teams execute, wall after wall, exhibition after exhibition, long after the artist’s death. Each execution differs, none is a copy, all are authentic, because what makes the work is the rule and not its particular realisation. And LeWitt, all the while, went on drawing with his own hand; writing instructions for others never prevented him from doing things himself.
A skill is exactly a wall drawing whose executant is a model. The person who writes it sets a rule of production, knowing that they will not control each realisation, and that the value of what they write will be measured by the quality of what others make of it. Writing a skill means learning what the conceptual artists knew, that a well-written rule is an aesthetic object in its own right, and that there is an art of prescription just as there is an art of gesture.
The scores of Fluxus
A decade before LeWitt, the artists of the Fluxus movement had pushed even further the reduction of the work to its notice. Fluxus, to put it simply, is an international movement born at the turn of the 1960s around George Maciunas, bringing together among others George Brecht, Yoko Ono, Nam June Paik, Alison Knowles and Ben, and which wanted to dissolve the border between art and life, through minuscule actions, modest objects, concerts of everyday events, against the solemnity and the market of art. George Brecht’s event scores sometimes fit in three words on an index card. Yoko Ono’s collection Grapefruit, published in 1964, gathers dozens of these verbal scores, brief instructions, sometimes performable, sometimes only imaginable, which each person is invited to execute or to dream. The reach of this gesture was twofold, and political in both cases. It desacralised the art object in favour of the proposition, and it placed execution back in everyone’s hands, making anyone a possible interpreter of the work.
These works I have frequented closely, and not only as a spectator. Some fifteen years ago, with my agency, I myself manufactured the DVDs of Fluxus works published by Re:Voir, the Paris publishing house devoted to experimental cinema, whose Fluxfilm anthology, compiled by Maciunas himself, remains available on its site. One can also watch these thirty-seven Fluxus short films on UbuWeb, the immense online archive of the avant-garde founded in 1996 by the poet Kenneth Goldsmith, which gathers films, texts and sound documents far beyond Fluxus, and whose singular history, legal aspects included, I recounted in Framed, autonomous, vernacular : a typology of the regimes of images. I recommend spending time there; one understands, better than in books, how joyful a form of art the instruction can be.
From binary to human language
There is another genealogy of the skill, absent from art history books, and it is that of computing itself. Andrej Karpathy, co-founder of OpenAI and former director of artificial intelligence at Tesla, summed up the present situation in a sentence that became famous, in January 2023: “The hottest new programming language is English.” To measure what this sentence contains, one must remember what layers of abstraction are in computing, and I will do so in a few lines, because this history makes it possible to understand what a skill is.
At the base of computing there is binary, zeros and ones circulating in circuits according to logical rules, through what are called logic gates, simple electronic assemblies which, combined in very large numbers, make calculation and logic possible. One can program directly at this level, and I have done it: to add interactivity to the DVDs I was manufacturing, objects designed above all to distribute films, with very little memory and very few possibilities, one sometimes had to code in binary, as close as possible to the machine. Above that comes assembly, machine language, where one programs the microprocessor no longer with zeros and ones but with named instructions, specific to each processor architecture, some with a reduced instruction set, others with a larger one. First layer of abstraction: one writes words, which will be converted into binary. Above that again, the higher-level languages, BASIC, COBOL, Pascal, where a single instruction such as PRINT, which displays a text on the screen, actually triggers dozens of assembly operations that the interpreter performs for us. Then the contemporary interpreted languages, Python on our computers, JavaScript in our browsers, whose same source code runs on different machines because intermediate layers adapt it to each processor and each system. At each floor, the movement is the same: we say less and less how to do, and more and more what to do, moving away from the machine and closer to thought.
Artificial intelligence is the next floor of this history, and Karpathy is right to describe it this way: human language becomes the programming language. But something happens on this floor that had never happened on the previous ones. The PRINT instruction does not discuss; it executes, always identically. A large language model receives the instruction, and it can question it, complete it, challenge it, propose something else. The feedback loop is no longer that of a command and its execution, it is that of a dialogue and a thought. We are no longer facing an instruction set but facing a relationship, and this is precisely why this layer of abstraction, which moreover arrived in the form of a brutal rupture and an online service on 30 November 2022, making us dependent on a few industrial players, frightens and fascinates so much. What this dependency holds politically, I developed in the article For a Political Consciousness of Digital Technology.
An executant that thinks
This history of the layers of abstraction transforms the comparison with Sol LeWitt. The assistants who realise a wall drawing follow the instruction; their micro-deviations are part of the work, but the instruction does not ask them to think against it. The executant of a skill, however, is a cognitive machine, if we are willing to hear this word without lending it consciousness: a machine capable of interpreting the instruction, of applying it to unforeseen situations, of pointing out what seems to it to be missing. The executant has more part in it than it has ever had, and that is what troubles us. I showed in Contribution cannot be commanded that the position we assign to this machine, executant or contributor, changes the nature of what it gives; the skill is the place where this question becomes a question of art.
So, who is the artist in all this? The question is not new, it is only carried to a new degree. Before a wall drawing traced by hands that are not LeWitt’s, before a piece by Pina Bausch recreated after her death, and I do mean recreated, by performers who bring it to life, not mimed, we already know that the work can be elsewhere than in the hand of its author. What is new is that the interpreter can be a machine, and I have no closed answer to offer.
What counts for me is what one receives. A person who reads a text or looks at a photograph has as their sole subject what this work brings them: what they learn from it, what makes them think, what touches their sensibility. This visible part of the iceberg is not secondary, it must be rich, useful, enriching, that is its whole demand. Knowing whether the text was written with one hand or four, whether it is a first draft or has been reworked a hundred times, comes afterwards, and has never made the value of a work. And if an artificial intelligence took part in the making, the ecological stakes are real, I devoted the article Ecology and Artificial Intelligence to them, but they must be held in orders of magnitude, as I did in The last straw. An industrial beef steak represents around ten thousand litres of water, the equivalent of some four thousand queries to a language model; a person making fifty queries a day, which is intensive use, would consume in a year the water equivalent of three or four steaks. We feel guilt when writing a prompt and none when ordering a rib steak; which is to say that our ecological anxiety lodges where we place it, rarely where the masses are. These questions of creation and ecology will moreover be at the heart of the residency I will be facilitating for NEF Animation, which I mentioned in Invisible artificial intelligence.
Autographic arts, allographic arts
The philosopher Nelson Goodman gave, in Languages of Art (1968), the most rigorous framework for thinking about what is at stake here. He distinguishes two regimes of existence of works: the autographic arts, such as painting, where the identity of the work depends on its history of production, so that the most perfect copy remains a fake; and the allographic arts, such as music or theatre, where a notation, the score, the text, allows multiple executions that are all authentically the work. And Goodman makes a historical observation, an art is not allographic by nature, it becomes so when a notation is invented for it; music was autographic as long as it was not written down.
The skill is a notation, and its appearance tips into the allographic regime practices that had never known it. A way of writing, a style of framing, a palette of graphic gestures, all things that were until now inseparable from the person, from their hand, from their history, can now be written down, and therefore executed elsewhere, several times, without them. I measure how ambivalent this shift is. It offers to ways of doing the circulation and the survival that notation offered to music, and in the same movement it raises the question that Goodman finally helps us to ask clearly: what is an authentic execution of a style, and is the style that executes itself without the person still theirs?
The protocol, the intention and the world
To move forward on this question, I propose a comparison between the skill and a device I have practised for years, well before artificial intelligence: my protocol-based photography workshops, whose method I described in Participatory reverse photography workshop. The instruction there is precise, there must for example be a hand in the photograph, the time is very short, you have to go out, find a place, make the image, put it online, come back. Then all the photographs are looked at together, and the person who made each image is not allowed to speak, a rule they discover at that moment. What makes the meaning is not taking a photograph, taking a photograph in itself is nothing; it is the whole of the device. And what I observe every time is that people, looking at their own photographs through the gaze of others, discover in themselves riches they had not perceived. They then come into contact with what Starhawk calls “power-from-within”, that inner power I spoke of in Power, potency, domination, which cannot be commanded, which frees itself when the conditions are met.
Now what exactly happens in these photographs? The person has an intention, to show human solidarity for example, and they photograph a young man helping an old lady across the street. But what is really in the image, what enriches those who look at it, overflows this intention on every side: the architecture of the old building behind, with the traces of its whole history, the attitude of the passers-by, their clothes, the car that happened to pass at that instant. One could say this is chance without importance. It is enough to frequent the Rencontres d’Arles or any exhibition to observe the contrary: the richness of a photograph comes from the whole of what constitutes it, because it is made in a world, and this world comes to collide with the intention and enlarge it. This is why the photographic artist is less of a demiurge than others; he or she knows that an enormous amount escapes them, and it is with this that they work. I saw the same thing in the hundreds of films on the platform By my window, where I simply proposed to film what you see through your window while telling an important memory, and where it would happen that a bird crossed the frame at the moment the person spoke of flying away. Nothing supernatural in this; simply, we are in the world, and the world exceeds us, enriches us, offers us more, if we open ourselves to it.
Here then, term by term, is the comparison between the skill and the device of these workshops. The workshop’s protocol is the skill: the way of doing, written down, which gives all the images their common form. The theme of the day, the situation, is the prompt. And artificial intelligence holds the place of the world: that immense, unpredictable context which comes to meet the intention and deposit in the work infinitely more than what was put into it. For artificial intelligence is now part of the world. It is in the taxi that arrives when you order it, in the music you listen to on the radio without realising it, at the very instant you open a social network on your phone; like the digital before it, it has become a “milieu of existence”. The works issuing from the same skill are extremely different, and yet they have something in common, which is the skill itself, each keeping its uniqueness, exactly like the photographs of a workshop where each person had to place a hand in the image: one same protocol, and images exceedingly different from one another. I do not yet know how to name this type of work. But I know that this displacement takes us out of the figure of the demiurge artist, who would master everything about their work, and for me that is good news.
Who signs, what circulates, under what licence
These questions finally engage law and economics, and there they meet an already rich history. Copyright, as we know, protects expression and not style; one cannot appropriate a way of doing, only works. Yet the skill materialises a style in a text, and a text is protectable. This discreet displacement could become considerable, for it gives for the first time a legally graspable form to what circulated until now without protection or remuneration, the manner.
But one must look at what already exists, and it is immense. Skills published for free number in the hundreds of thousands, perhaps millions, on platforms like GitHub, including on writing itself; the Humanizer skill, for example, designed to rid texts of the writing tics of artificial intelligences, is published freely and evolves through everyone’s contributions. This is the very spirit of computing and of open source, and it deserves to be understood for what it is, for there is nothing economically aberrant about it. Each person who shares enriches the community, and finds themselves enriched by it; on this common foundation, people then build, in the real world and around themselves, things far better and far more solid than what they would have built alone. Everyone moves much faster, and the world moves forward. Free software formalised this logic from the late 1980s with the copyleft of the GPL licence, that reversal of copyright which uses intellectual property not to close but to guarantee that the work will remain open, freely usable, modifiable and shareable, on condition that its derivatives remain so too. And this gesture reached art, with the Creative Commons licences from 2001 and, less well known, the Licence Art Libre, created in France as early as 2000 by the Copyleft Attitude movement, under which artists publish their works while explicitly authorising copying, distribution and transformation.
Legitimising the skill as a work: a proposal for the cultural sector
It is on this foundation that I make my proposal to the cultural sector: to legitimise the skill as a work. An artist signs a work that is a way of doing, publishes it under a copyleft-type licence, and others appropriate it, within a framework where circulation is traced and copyright respected, with versions, branches, feedback. Publishing the skill does not freeze it, on the contrary; one can see its results, the works that come out of it, and the instruction itself goes on living through those who take it up. It is a form of living art, and it is also a perspective of cultural democracy, for it multiplies the ways of entering into relation with creation: looking at the works issuing from a skill, reading the skill itself as a work of thought, using it to create in one’s turn, modifying it and proposing one’s branch. So many facets, so many possible mediations, for cultural venues, for education, for amateurs.
For cultural venues, this legitimation makes concrete gestures possible. Exhibiting a skill alongside the works it has made possible, as museums present LeWitt’s instructions alongside the painted walls. Putting it into mediation, by inviting the public to read it, to use it, to modify it, and why not to write their own, for every person has a way of doing that deserves to be written. Inscribing it in commissions and residencies, by accepting that the deliverable of a work of creation may be the skill itself, a form in its own right, which is exhibited, published, taught and shared under a free licence. Preserving it, finally, as libraries preserve scores, so that these ways of doing remain executable when their authors are no longer there.
There remains a risk, which I feel in my own practice, that of self-pastiche. A skill that is too closed freezes the manner at the moment it is written down, and the machine then indefinitely produces the artist as she or he was, a caricature ever more faithful and ever less alive. Umberto Eco described in The Open Work (1962) what distinguishes the living work, a field of possibilities that each interpretation actualises without exhausting it. A good skill seems to me to have to remain open in the same sense, a living document, dated, fed by corrections, revisable, carrying within it the mention of what it does not yet know. It is on this condition that it is a work and not a mould. The artists who seize upon it early will invent the rules of an art of the rule.
