Grégory Chatonsky, a Realism without Reality
What’s meant by artificial intelligence? It’s actually a question of three different phenomena that overlap. On the one hand we’re talking about fiction: science fiction films and novels, a whole imaginary world that influences both the notion that people have of artificial intelligence (AI)—even before they know anything about it—and the researchers in Silicon Valley.
We’re also talking about a technology the foundations of which go back to Frank Rosenblatt’s creation of the first artificial neural network in 1957: Perceptron, a character recognition system. In other words, teaching a machine to read. This technology, after a revival of interest in the 1990s, exploded culturally in June 2015 when a Google engineer, Alexander Mordvintsev, created Deep Dream to explain what happens inside an AI. This creation was made possible by the convergence of a model of statistics—AI is statistics, induction or generalisation—and big data, i.e. the enormous stocks of data accumulated like no other civilisation before us. In this case, the computer has images of dogs and molluscs in its memory. Whatever image it is shown, it obsessively sees dogs and molluscs. And through no fault of the engineer, the images it generates resemble psychedelic hallucinations on LSD. Finally, we’re talking about systems of domination that use this technology to set up an automated surveillance system that signals to human operators whenever something abnormal happens.
NEURAL NETWORKS
What is it technically? The term “Artificial intelligence” immediately scares people, but the only word to retain is “artificial”.The word “intelligence” is ambiguous and confusing. In concrete terms, we’re talking about software that, in general, feeds on a lot of numerical data and derives statistics from it. For example, if I give it 5,000 pictures of birds, the software won’t understand what a bird is at all. It has no idea. On the other hand, it’ll stupidly calculate the proximity between all the pixels and all the colours and, from a banal induction, it obtains what’s called a latent space, this statistical space that means that if it draws a pixel of a given colour somewhere at random, it knows the probability that another pixel of a given colour is next to it. Then it does something disturbing, which for me is as important as the invention of photography: it produces new birds, which don’t exist, but which we, human beings, recognise as such. It’s therefore the appearance, in the aesthetic sense, of a new realism without reality, which uses photography, but is no longer photography. Now, since the industrial revolution, realism had been largely defined by photography. To see this new realism appear, which is to my knowledge the only one equal to big data, is a fundamental moment in history, and not only in the West. The way AI works is very simple. Its perceptive and hermeneutic effects, or understanding, are strictly incalculable. It’s this abyss between cause and effect that remains untapped.
These are techniques that you’ve been using for about ten years. What’s your relationship with them? For me, technique in art isn’t a means to an end, it’s an end in itself: a medium. I have a certain appetite for experimenting with all kinds of techniques, asking myself questions that aren’t those of technicians. So I don’t use any particular technique, but it’s always about neural networks. People talk about GANs (generative adversarial networks), RNNs (recursive neural networks) and so on. I imagine these software programmes to be like characters. The GANs fight against themselves, and are a bit schizophrenic. While one part of the networks generates results, the other part judges whether these are credible, accepting some and rejecting others. It is the Kantian court of reason that criticizes itself by splitting into two. RNNs, on the other hand, allow for recognition, classification and generation. They anticipate by remembering what’s gone before and what they’ve learned. They’re predictive machines, in the manner of clairvoyants or David Hume’s enquiry. But overall, it’s the conceptual simplicity of induction (I have a large series, I draw a generality from it) and the fact that it doesn’t produce an identical repetition, but the air of a resemblance, of a déjà vu.
TECHNOLOGY WORLD
You say you want to do everything yourself. But appropriating these techniques can’t be easy. I hesitate like any artist about their medium, I seek, which means I don’t subject the technique to a preconceived idea. On the other hand, by working with software, codes, and by trying to change the parameters, ideas emerge. Nothing’s more interesting than setting up a mechanism that surprises, not being in control or mastering. I’m always amazed at the production of differences that you can get with AI. I try to “de-instrumentalise” it. Many artists have a very critical point of view which, for me, is problematic because, as we’ve known since Guy Debord, criticism contributes to domination. Nor do I see AI as something autonomous, which is the case most of the time in the mass media: people wonder whether it’ll replace artists—thus presupposing that the artist is—autonomous and, beyond that, whether the machines won’t kill us all—except that you just have to pull a plug to stop everything.
On the contrary, I position myself in a relationship of fragile co-dependency. For me, AI is like an otherness. My imagination as an artist is affected by what happens in the software, but also by the cultural baggage it digests; and conversely, I influence the software by coding. It’s a heteronomy, a loop between the software, our cultural context and my imagination. For example, I recently co-wrote a science fiction novel with an AI. I would write a sentence and, when I ran out of inspiration, I’d ask this AI, which had a library of books to my taste (Philip K. Dick, Pessoa, Beckett, etc.), to continue. It’d make several suggestions, I’d retain one and integrate it, then I’d continue, and so on.
What’s the place of these practices within contemporary art? Are they isolated from it as net art was? Net art, which is where I started, was somehow right too soon. We were taken for geeks. The situation’s definitely changed. For example, visitors understood the Hito Steyerl exhibition at the Centre Pompidou this year, (1) which was quite specialised, quite immediately. AI has become a cultural phenomenon.The fact that two of the world’s most important artists, Pierre Huyghe (2) and Hito Steyerl, have each taken it on in their own way, naturally and intelligently, testifies to the fact that AI is much more than a technology, and constitutes a world. Complex ecosystems are drawn from it. For example, for the group exhibition La Vie à Elle-Même [Life to Itself] at the Vassivière art center this summer, I used sensors to connect a living organism that was a bit zombie-like—a dead tree with insects and plants on it, since the characteristic of the dead is that the living colonise it—to a film which, through interaction, generates in real time the natural counter-history of a branch of evolution that would have been possible if the Earth had developed differently. AI allows artists to explore and invest this world of possibilities in a new way, by proposing different alliances between humans, inhumans and ahumans. Every living species needs allies. When Hito Steyerl shows Boston Dynamics robots being beaten by humans, we’re moved. We can only side with the robot that says nothing and gets back up, unperturbed. Here, it’s the inhumanity of the human race that’s denounced, not the robots. These alliances between technology and the living, including humans, thus give rise to new experiences of feelings.
I PERCEIVE THAT I PERCEIVE
In your installation Terre Seconde [Second Earth, 2019], the AI does indeed ask itself existential questions. Hearing it is very disturbing, like hearing one’s own recorded voice. You also talk about the mistakes made by machines that lead us to put ourselves in their place. What experiences do your works offer? This difference between the voice we hear in our own skull and the one we hear outside when we’re recorded, which we don’t recognise, is exactly what AI does. In Terre Seconde, a machine that creates a second Earth, there is this AI that I have fed with texts on consciousness and technical texts to make it ask questions about itself. And it wonders whether, because of reboots, it mightn’t be repeating a text like a parrot, thinking that it is the first time. We believe it, we’re touched. It makes us wonder if our own existential questions are just the surface effect of a calculation. It is a transcendental aesthetic experience (I perceive and I perceive that I perceive) and the introduction of a Cartesian doubt without resolution.
At the heart of an AI is this cultural data (digitised texts, images and sounds). By regurgitating them, since it is a matter of metabolisation, the software sends us back to this heritage, in a non-identical, hybrid way, never seen before, but which we recognise— which could be a definition of beauty. This is the great power of metamorphosis. In the installation Complétion 1.0 (2021), a screen scrolls 14 million images feeding an AI that generates new ones. Another AI, which I have learned to speak to like texts of photographic aesthetics, describes these images. The whole point isn’t that it does this, but that when you listen to it, you wonder why the software, with all its context, saw this or that in this way. By putting ourselves in the place of the machine, we put ourselves in the place of our cultural baggage. I also reverse the process by generating images directly from texts, in particular Jean-François Lyotard’s Libidinal Economy, the introduction to which hallucinates the body. The results are disturbing, as if ultra-surreal. I’m not trying to denounce a system of domination, which is indeed in place and catastrophic, I want to move on to the next stage by generating new possibilities, of course with a critical device, but one that gives the technique cultural resonances by creating different degrees of displacement. We’ve produced mechanisms that are beyond us: the web is an accumulation of data that we are unable to navigate. By digesting these stocks of information, AI gives them a form, and with it, a fundamental access to ourselves and new modes of reflexivity.
Hypermnesia plays an important part in your work. Since the end of the 1990s, with net art, the question for me has been this completely crazy accumulation of memory: a headlong rush that I think should be compared with the evolution of the feeling of finitude, which would no longer be only on the scale of the individual (my own death) but of the whole species (our extinction). My hypothesis, a fictional one, is that by trying to record everything, we’re trying to create a kind of pyramid, traces that would remain after our passing, our disappearance. This is a very interesting vision because the data and the induction are used to generate alternative or counterfactual versions. This gives a very surprising sense to the story, and highlights the reappearance of an unexpected theme: resurrection. Transhumanism is, for example, about resurrection: transferring the software of the mind into the hardware of a new body. Since Terre Seconde, in the tradition of 19th century Russian cosmism, in particular Nikolai Fedorov, who believed that a museum should be the complete resurrection of the world, I’ve been trying to think of AI as a resurrection, of memory, not of bodies.This is the beautiful true story of two young Russian friends, Roman and Eugenia. He died suddenly at 25. She fed an AI with all of Roman’s exchanges, emails, chats, which now allows her to chat with his chatbot.
My project Complétion uses techniques to complete deficient or damaged documents, sculptures that are missing pieces, unfinished or lost texts: to take all the history of the past and repair it, which is obviously an impossible project. At the same time, I’m pursuing my projects of alternative histories. I believe in the power of simulacra, this divergent resemblance. My relationship with neural networks allows me to explore these questions in quantities that I’m humanly incapable of accessing, even though our era demands it.
Translation: Chloé Baker
1 See our interview with Hito Steyerl, artpress no. 467, June 2019, and our exhibition review, artpress no. 488, May 2021. 2 See the article on Pierre Huyghe in this issue.