By Marine Corniou
The GANs are based on a confrontation between two neural networks: a generator and a discriminator. The first produces images by imitating the actual works gathered in a database. The second must “guess” which works are from the data bank and which are synthetic pieces, from the generator. Feedback from this “judge” improves imitations of the “forger” network until they blend into the original style. Above: members : of the Obvious collective and some of their works, including (left, completely at the top) and the (right). to do so,” explains Pierre Fautrel,who works with his two childhood friends Gauthier Vernier and Hugo Caselles-Dupré(a doctoral student in machine learning).
“Everyone got agitated with Belamy’s s’est portrait, but the use of algorithms in the arts was around long before,” says Nathalie Bachand. Last February, at a day organized in Montreal by the Quebec Council for Media Arts (CQAM), this independent curator recalled that the first computer graphic experiments took place as early as the 1960s. Then,the 1990s saw the flowering of many interactive installations and automate artistic productions. “The word is scary, but algorithms have no real autonomy or free will,” she said. What is new is the GAN, which has become increasingly accessible over the past five years.»
Far from being mere copyists, these neural networks can also inspire their masters. Like Montrealer Marc-André Cossette, who composes electronic music using a system that deciphers the movements of dancers and l’instar creates real-time sounds “inspired” by the position of bodies. “The dissonances and mistakes made by AI have a lot of influence on me, including when I compose without it,” he told the CQAM forum. What if the machine actually increased human creativity?