This Monday, May 2 at 9:10 p.m. will be broadcast the first issue of “l’Hôtel du temps”, presented by Thierry Ardisson. This biopic 2.0, revives disappeared personalities through artificial intelligence. A new kind of program that questions researchers from the University of Limoges.
On the night of May 2 to 3, 1987, Dalida disappeared. 35 years later, this mythical singer is about to reappear on television screens, in a program of a new genre entitled “l’Hôtel du temps”.
Based on the technique of deepfake or hyperfaking – used until now to divert videos very often in a comic way -, this program consists in bringing back to life disappeared personalities like Dalida, Coluche, or even Jean Gabin.
Behind this technique: the concept of artificial intelligence, well known to master’s degree in computer science, image synthesis and graphic designISICG, University of Limoges.
This unique master in France trains students to create and transform images mainly for video games and cinema. “The training is oriented towards the fields of image synthesis and graphic design”explains Djamchib Ghazanfarpour, responsible for this master for 20 years and specialized in image synthesis since 1982.
“In particular, our students are trained in the technique of deep learning, which consists of developing calculation models to model data. This technique is the basis of DeepFakes”, he adds.
The term deepfake or hyperfaking is indeed a combination of “deep learning” and “fake”. Based on artificial intelligence, this technique makes it possible to simulate real characters.
In this master, Frédéric Claux, teacher-researcher in images is in charge of giving lessons around this technique. He explains how it works:
It’s an automatic program that can practically run in real time if you have an extremely powerful machine. At first, it will analyze the video to be processed, isolate the face, it ignores everything around it. This step is very important, since we only want to modify the face.Frédéric Claux, teacher-researcher in images at the University of Limoges
Next, “on another video you have a second face which interests you, it is the target face. In the same way, the program will isolate the face. Then, when you have the two videos just with the face, you will pass them through what is called an auto-encoder which is a program that detects the pose for the two videos learned and that allows you to go from the source face to the target face. For this to work, the two faces must be very close”.
The trailer of the program broadcast on France 3, Monday May 2 at 9:10 p.m.:
“We can go very far in terms of realism. There are not many technological constraints, IT has no limit”adds, Djamchib Ghazanfarpour.
For those computer nerds, this show is a “technical prowess, I’m going to be glued to my television”, confides with laughter Frédéric Claux.
But others, on the other hand, do not see the concept very favorably. This is the case of Nicole Pignier, professor of information and communication sciences at the University of Limoges.
In recent years, she has conducted extensive research on this deepfake technique.
In my opinion, this phenomenon of deepfakes, fake images or fake videos raises the question of what to state means. When someone speaks, sings, or draws, he is expressing himself or expressing something. This is no longer the case here. We make him say something.Nicole Pignier, professor of information and communication sciences at the University of Limoges
Before adding, “Dalida is not asked for her opinion, if she agrees. We are going to have fun, to make her sing, to make her say things, but behind this arises the question of respect for the person who has disappeared”.
Thierry Ardisson defends himself, “there is no image rights for deceased persons. But each time I contacted the heirs, out of respect, out of ethics and because they also help me with the screenplay. To write the texts, I gives themes (origins, childhood, studies, etc.) to librarians who bring together all the sentences actually spoken or written”.
The presenter of the show directed by Serge Khalfon and produced by 3ème Œil Productions, also said in an interview with France 3, “I am the first to make a positive and cultural use of deepfake. VSis a tool, and a tool has no ideology”.
Statements contradicted by Nicole Pignier: “Cis totally wrong. Behind all this, there is the ideology of neoliberalism designed to tame human activities, to frame them. Thierry Ardisson forgets that to produce an image, to produce a song, to play a role is both to signify, to mean something but to weave a human bond in a society”.
“Instead of offering singers and actors to put themselves in each other’s shoes to play the other, we will rely on technological prowess which will seem more real than real, but is it do we really need that to pay homage to someone? It’s not a revolution, it’s one more step towards this above-ground world, we fly towards flows of words that have no much to do with our reality.”, she adds.
Behind this concept, a final question, that of deception. Is it possible to detect videos made with this technique?
“Sure, explains Frédéric Claux, you use the same principle, you take a faked video, you isolate the face, you tell the program, that’s faked and you do the same thing with a video that is not faked. You reproduce the exercise thousands of times, it will be able to detect fake videos at the end”.
As much criticized as praised, the deepfake does not leave anyone indifferent. Tonight, more than ever, the words, and words and words of Dalida, have not finished intoxicating us.
We want to give thanks to the author of this post for this incredible content
Ardisson revives Dalida: the prowess and limits of artificial intelligence pointed out by researchers
Our social media profiles here and other pages related to them here.https://www.ai-magazine.com/related-pages/