Creation assisted by Artificial Intelligence
“The algorithm has no past, no present, no future. It is created by humans to perform a task more efficiently.”
How did you form a collective?
It all starts with a buddy story. Gauthier and I have known each other forever, then we met Hugo, with whom we were in the same gang in high school. When we finished our studies, we thought about what we wanted to do together: starting with entrepreneurship, we quickly realized that we wanted to do something more creative. One day Hugo came across the GAN algorithms, which we are using right now and which are incredibly powerful. He shows us the faces of people who never existed, similar to the faces we now find on This Person Doesn’t Exist. The questions that emerged behind this tool were fascinating and almost philosophical, especially if we brought them to the field of Art.
Why was Art an ideal field of experimentation?
The choice of Art is not a question of technique, it does not serve more than anything else to test the power of the algorithm. But it is a way to carry a message. For several months we thought about the implications of the use of such a tool in the artistic field, to conclude that they deserved to be asked. And the best way to do this was to start creating. If you entrust an algorithm with a visual creation work, you necessarily ask questions, because this artistic process underlies societal issues.
A CREATION ACCOMPANIED BY ARTIFICIAL INTELLIGENCE
Could you go back over the detailed process of creating a work?
The first step is the choice of subject. You imagine the collection you want to create. You then have to find data samples of pre-existing works. With the help of a bot we then perform a scrapping, that is to say the massive recording of similar images. At the end of this process, one obtains about a hundred thousand images, but which can be portraits, landscapes, or even animals. So the first job is to sort this data set, grouping it into categories so that the algorithm can detect patterns in common. For the Japanese prints we chose to keep only portraits and landscapes. We then replay these data sets so that they are really clean, so that all the elements look the same, which will help the algorithm to work at its best.
We then run the algorithm on the database thus constituted. This step should not be seen as linear, because you try things, you change parameters, there is trial and error, as in the work of a painter. It will thus run for 48 hours, to generate an infinite number of images, landscapes and portraits. From this mass, we will choose the images that best suit our artistic approach. We wanted 11 landscapes and 11 portraits that represent the diversity of the database, and include women and men, lakes and mountains. Once these images have been selected, we will choose the printing medium. The latter is not the same for Japanese prints as for portraits, because here we wanted a traditional method.
We see in this example all the inputs that are given to the algorithm. So it is indeed a human/machine collaboration, but the human is still in charge. We wouldn’t have been able to make visuals of this quality, but we were able to guide this machine, so we are artists.
Are the images proposed by the algorithm are composites of the database?
No, the results they propose are intrinsic to the technology, he learned to do by deciphering the codes of the paintings shown to him. Our algorithm understands the rules of portraiture: two eyes, a nose, a mouth. By looking at a multitude of them, he makes new ones following the same rules. But there is no pixel from another work in his realization, it is not a style transfer algorithm. He does not create the middle work, but a new work.
On the other hand, between an image in JPEG format on a computer and a work exhibited at the Hermitage there is a step. Printing the image on a canvas, placing it in a gold frame, signing it with a mathematical formula, all these purely human choices contribute to transforming the image into a work of art.
A CONCEPTUAL APPROACH
Who is the author of the final creation and why?
It is Obvious who is written about the cartel, for several reasons. In the story of Boronali, the donkey who painted with his tail, it was not him who painted, but someone who had the idea of making a happening to question the nature of the artist in the creation of a work. We are interested in this parallel, because an algorithm is stupid, and only knows how to do one thing: it has no consciousness, is incapable of reflecting on itself, has no vision. It has no past, no present, no future. He is created by humans to accomplish a task more efficiently. To question his ability to create would be to believe that our dog is as creative as we are, in an anthropomorphic dimension.
Yet we often hear about works created by Artificial Intelligence.
We have suffered because we try to say the opposite. People sometimes think it is a robot painting in a basement. Since this is popular, it helps to propel us to the forefront, but paradoxically it serves to undermine our core message, which is that it’s not a core message but about ourselves.
The problem with these new technologies is that they use a vocabulary that does not have the same meaning in different disciplines. Artificial Intelligence in video games is not that of Research, or Science-Fiction. There is no such thing as the fantasy robot: the most advanced elements are mechanical arms that manage to catch a ball, or Boston Dynamics that manages to do somersaults on selected videos.
An AI with a conscience belongs to science fiction. There won’t be one for 50 years, and probably never. In any case, it won’t be the learning machine that will be used then. But if everyone uses this term today, it is because its promoters thought it would be interesting from a marketing point of view. It’s a brilliant idea when in reality it’s automation on large databases.
The real question is: what is an artist? For me it’s a person who steers an artistic process that results in a work. He has an idea, follows a practical and conceptual approach that results in a work of art. In this process, the algorithm plays no role: it never occurred to me that he was going to paint a work of art.
Another fundamental question concerns the role of AI in creation.
To ask the question of Art accompanied by Artificial Intelligence is to ask the question of AI in our society. The fact that people directly imagine Sarah Connor being chased by Terminator creates a general vagueness about the concept. That’s why we spend our time explaining it again. One of the fundamental questions in our work is whether or not an AI has the ability to be creative. This question is already problematic at the concept level: an AI is a fuzzy concept, so is creativity. Yet people love to ask us this question because they see it as an entity, which it is not.
The next question we might ask is whether AI creates things of its own. In Art no, only humans can be. Creativity is the chess player who deploys a new openness: our algorithm has a form of inventiveness but it is not creative. Neuroscientists define creativity in two points: learning by example (if I want to draw a car I need to know what it is). This is the work done by our algorithm. To this we must add social and demographic factors, encounters, trainers, time and medium. Creativity then becomes very much linked to sensitivity. If the result does not exist outside of our intervention it is because it is the only thing that really counts.
Are you in the great tradition of conceptual art carried by a concept?
Conveying an interesting concept is our priority, and it is for this reason that we propose a key to understanding through the signature formula, so that the spectator is not left with an indecipherable portrait. Our work is not radical in the concept, but this is not our goal. We do not wish to be the new Duchamp, but rather to follow in his footsteps, talking about Art as such.
QUESTIONS ASKED BY AI
Is this a purely performative creation?
On the level of pure performance, others have algorithms that are often better than ours. But in all cases it is essential to understand that the creation itself is not the fruit of artificial intelligence. At most, it is created using machine learning or deep learning algorithms.
Our approach is completely conceptual, because we could choose a photorealistic path, but we preferred, as in our Japanese collection, to try something else, because this blur has an interest, but also because the technology is not yet mature: there is no point in making people believe that it is possible to obtain a Rembrandt. We will never get the result of a site like This Person Doesn’t Exist that works according to what is called cherry picking, i.e. using the best algorithm on the best data set. Again, our goal is not to show how good algorithms are, but how they can help us to be creative. Thus, the revival of major themes in the history of Art allows us to talk to everyone.
Can we speak of a sensitive approach in the case of such a creation?
When people select a work from our catalogue, they choose according to what is beautiful, what they prefer. Whether this perception is not primordial, whether it takes second place to a willingness to invest or an interest in AI certainly. However, this sensitivity linked to the visual created by the algorithm necessarily comes into play. Thus, there are people who didn’t like the first collection but appreciate the second one.
How is Artificial Intelligence likely to make the Art market evolve?
AI is a medium for us as artists, but it can also be used by the markets themselves to analyze and better understand their statistics. It is also necessary to differentiate between the learning machine that is available to companies and the one that allows artists to create works. We cannot know today whether we will create a movement or whether people will lose interest. All this is too external for us to know. I think there is a greater chance that AI will become a tool that gives a better digital image than a way to ask questions about society.
Does the algorithm still have room for improvement?
Don’t fantasize about the algorithm. In fact, it’s never the same, but it’s the latest version available. A version developed by someone else, which is then reworked. It’s like always using Pythagoras, but in a way that no one has ever thought of before. As an artist, it is normal to want to use all the means at your disposal.
However, the images that are produced by the algorithms improve gradually. Turing’s tests on abstract have been working for 3 years, and on figurative for 2 years. If a work is made to believe that it is created by a human being and you accept it, it is because the test works, and in Art it is passed very quickly, unlike a book or a piece of music. There is such a diversity of visual productions that it is impossible to imagine that a painting has not been made by a human being.
It is even possible that one day there may be images so precise that they could not be made by men. Images of 50 giga that would be painted pixel by pixel, and in which it would be possible to zoom in infinitely. A city map in which you could see people behind every window.
OBVIOUS AND THE ART MARKET
Is Obvious a business or an artist collective?
Obvious aims to be both. To have artistic ambitions you need means. If we are not able to launch our next collection, or pay for our tools, we don’t exist. So it’s a business, as it is for other artists: if you want to make a living from it, you have to sell works. Originally, I’m not an artist but an entrepreneur, I want to create projects and make a living from them. I became an artist a little by chance. It’s a role that I now take on because I think I have interesting things to say and bring. But I’m not interested in the myth of the artist living in the attic. I want to take our creativity to as many places around the world as possible, and across a multitude of mediums, from fine art to fashion.
Fine art and design are simply two ways to express oneself. There are those who handle volumes and shapes better; others handle colours and depths of field. There are people who are able to do sequence shots and editing. We want to explore all these subcategories, because anything creative is art to us. I don’t differentiate between art forms that would be premium and others that would not.
If the particularity of your work comes from the technique used, do you have an advantage in being first?
But we are not the first. There are people who have been making art with algorithms – but without Artificial Intelligence – for 10 years. And others were making AI-assisted paintings long before us. If Christie’s selected us, it’s because we decided to take a different stance, notably by choosing to go figurative rather than abstract. We chose iconic currents, which speak to everyone. It’s not in the technique that we are the best, but in the concept. And this advantage comes from the fact that we are three guys who know each other, who grew up in the same city and are reflections of their generation.
Why did you choose to do your first series on portraiture?
When we talk about Art, 90% of people imagine portraits that look like Edmond de Belamy, because they have seen them in books, in museums. We are caricatural in our approach in order to make people understand what we are talking about: the choice of the classical portrait was thus obvious, hence also the name of our collective “Obvious Art”. Our program is thus composed of several of these icons: portraits, Japanese prints, Lascaux paintings and Greco-Roman sculptures. But these choices also come from our education: we are not museum inmates, and this caricatured side also comes from our candor in the face of this environment, while giving us great strength.
Why couldn’t you choose an artist for a series?
The algorithm can only work with a database of one or two thousand images. For portraits we had fifteen thousand and for prints twenty-five thousand. We tried with Pollock or Klein but they didn’t have enough works, so it was impossible to find common rules. Algorithms that claim to be able to make new Rembrandts in a photo-realistic way are misleading. It is indeed impossible to assemble a sufficient data set from the existing paintings. Therefore, the only solution is to assemble the heads together and then refine the eyes using software. Selling the result as an algorithmic creation is misleading and puts all those who use the algorithm in an uncomfortable position.
YOU WILL ALSO LIKE
Kintsugi is a technique for restoring ceramics with lacquer and gold dust. Myriam Greff, one of the few specialists to master it, explains how it allows one to work on the time and emotions that bind an object to a person.