Dall-E, MidJourney, StableDiffusion, Point-E, ChatGPT… So-called generative artificial intelligence (AI) will have marked the year 2022. To the point that some observers are more and more convinced: these tools which make it possible to generate texts, computer code, images, or 3D objects with simple sentences, will change the face of the world. For good, but also perhaps for bad. As a convinced technophile, Raphaël Doan fell in love with the subject. The senior civil servant, associate professor of classics and successful essayist, has developed a website thanks to the ChatGPT platform, in order to host a game there. The goal: to recognize, among a succession of images, paintings by grand masters generated by AI. What reserve some surprises, even for the most informed. Feedback and perspectives.
L’Express: We tested the quiz and unsurprisingly, we made a lot of mistakes… Beyond the personal affront, does the machine’s ability to deceive us so easily offer a dystopian vision of our future ?
Raphaël Doan: It is certain that between a painting by Rembrandt and an AI, the boundaries are dissipating a little… But to be completely honest, I have deliberately selected images on my site that are confusing: good generations by artificial intelligence, handpicked, mixed with real works of art less known than the great classics. Otherwise it would be easier. But the goal was elsewhere: it was a question of proving to those who believe that AI is bad, and will never be able to do the equivalent of what an artist achieves with brushes and paints, that it is not is not that obvious. It’s like a kind of Turing test, but for the image. And indeed, we can see the thing from a dystopian point of view by saying to ourselves – because we thought that artistic creation was going to remain definitively a human specificity – that it is ultimately replaceable like any technical activity. This explains, I think, a whole movement of artists and illustrators who are already trying to ban this type of AI or in any case to regulate it very strongly. They actually find themselves in the same situation as textile workers in the 19th century, or painters at the dawn of the 20th century, and that was something that was unthinkable three years ago.
Do you think they are right?
They believe that the images used to train the artificial intelligence are under copyright. And therefore that the works generated are illegal. This is a legal question to be decided. Proponents of these AIs claim that they are legal because they are only inspired by existing works. They don’t plagiarize them, they don’t copy them, and they don’t use parts of those images to create their own. It’s a bit like an artist learning to draw by looking at existing paintings and illustrations; neither does he steal them. To tell the truth, a Picasso, too, gathered in his “biological database” all the old paintings before designing his unique works.
Is there still a way to distinguish real works from AIs anyway?
For now yes. The eyes are, for example, quite poorly reproduced by the AI. Human beings are so accustomed to scanning those of their congeners that we spot these defects much more easily than anything else. Same thing for the hands. Small artefacts which I think, at the rate at which technology advances, will be settled in a few months or a few years. I find, in any case, that it is essential to know how these tools work in order to understand how it is still different from what a human being achieves. The AI, for example, doesn’t really make a hierarchy between the different parts of an image. A pixel is a pixel. A painter or a photographer, conversely, when he makes a portrait, will give more importance to the eyes, the mouth, the face than to the chair behind or a piece of wall. Let us finally reassure: programs make it possible to check whether an image has been made by AI or not. There will therefore always be a place for authentic images, including paintings and drawings. The two will coexist in parallel.
In addition, these AIs seem to have difficulty in innovating, in generating style themselves. Is this, somewhere, a bit reassuring for us?
On platforms like MidJourney and StableDiffusion, there is still a kind of style that emerges and that differs from what we are used to seeing in art. But actually, AIs do nothing on their own. They only act on instructions. It’s not as if they were artists per se. It would be interesting, however, to imagine a system where an AI is trained both on a database of images and on hundreds of thousands of art history books in order to refine its knowledge. . To integrate the fact that in the 18th century, for example, one paints only with such pigment or such type of brush, and thus to lead in the works generated to something more innovative in the style… I think that the we will get there one day.
What are the advantages that can be drawn from these artificial intelligences?
They will give enormous productive power to individuals who never had it before. One can imagine that there are plenty of film geniuses who would have been extraordinary directors but who in their life did not have the opportunity to direct a film. Because obviously, that requires making a career, working there for a long time to have the right connections… This access, which has become so easy to the means of creation, will turn things upside down and that’s rather exciting. It will be the same process for imagining 3D virtual worlds, metaverses.
AIs like chatGPT, for text, will revolutionize access to knowledge. Steve Jobs, in the 1980s, evoked a future in which a robot or software would be able to make us discuss with Aristotle. We are there, in a way. A tool like chatGPT will be able to act as a personalized tutor: an advantage that only children from very wealthy families have access to today. AI will also change the way books are made. There will always be stylistic writers who will hold the pen themselves, of course, but there will also be many others, journalists, essayists for whom the work of writing will be closer to that of conductor and will consist of describing to the AI which paragraphs it must write, with which argument, which style, which emotion. Even if it means retouching the text afterwards… Or having it retouched by AI. I don’t believe in replacing human authors, like artists, since you always need someone to lead artificial intelligence. It’s as if everyone now has a very intelligent robot secretary but without any spirit of initiative.
And the disadvantages?
Naturally, this democratization of access to creation will make it more personalized for the consumer. In short, this will generate works tailored to our tastes. We are thus going to lose common cultural references. It’s something that’s already happening, though. A series like Game of Thrones and others among the most popular of the moment, do not have as many viewers connected at the same time in front of their television sets as the American series broadcast 50 years ago. There has been a fragmentation, an archipelization of culture, including mass, and I think this is a phenomenon that is already underway and that will increase, even multiply with AI. The impact will be significant on living together and everything that makes up the social fabric.
Essayist Raphaël Doan
© / YouTube screenshot
Technology moves very fast. Won’t it cause tensions in society more quickly than we think?
Yes, and that explains the epidermal reaction of certain illustrators who are organized to fight this, which I explained previously. Some didn’t even know these technologies existed a few months ago and in six months they may not have a job. At the time of the arrival of photography, anonymous painters had a little more time, either to finish their career, or to train elsewhere… There will perhaps be limits that we did not have not foreseen and which, in a year, will lead to a new “winter” of AI as there has already been before. A stagnation. Although many economists often repeat that technological innovation never really destroys jobs but redirects them, we are in a phase where the speed of technological change can create a new type of problem for public authorities.
Beyond art and text, many industries are interested in generative AI. Can we trust the machine to this extent, not always reliable, crippled by the biases specific to our social representations?
I think this problem can be solved with an AI capable of constantly checking and updating its knowledge. But it is clear that we will have to get more and more used to putting our lives back in the hands of machines. What we already do a little, it must be said. I even think that at some point, perhaps in a few decades, this shift will occur where we will look back and say to ourselves that it was incredible, barbaric and absurd to imagine performing operations with human surgeons. Or to let ourselves drive cars. Because AI will outperform us in these areas, and many more.
Can we take the problem in the other direction and ask ourselves: what human activities do we want to preserve at all costs?
Fortunately, there will always be people who technically know how to do things the traditional way. Today, there are plenty of trades that still continue to exist, such as book binders, even though we know how to proceed in an industrial way. But they will be very romantic niches from a social perception point of view. This is also the argument of some artists who protest against AI: why don’t we replace painful and difficult or dangerous tasks rather than replacing an activity like drawing and painting, which are things that give people pleasure? But in truth, no one has really decided. It’s paradoxically easier to make imaginary paintings than to automate the job of deliveryman. It will happen with robotics, probably, but the progress is slower than in software. Companies like OpenAi (which designed Dall-E and ChatGPT, among others), still believe that AI will one day be able to perform all human tasks. This is what they call artificial general intelligence. And it will perhaps feed our domestic robots of tomorrow…