An Algorithm Is Turning Nude Paintings Into Museum-Worthy Surreal Art

The field of AI research and its applications is so expansive and in its early days that there really is something for everyone, even those who believe it’s impossible to be excited by algorithms.

Using a set of nude paintings as his source material, researcher Robbie Barrat recently fed some visual data into a Generative Adversarial Network (GAN) — essentially the term for AI that uses two neural networks that work together (“generative” and “discriminative”) to carry out its unsupervised learning.

Because of the way the technology is built, in theory it should be constantly generating more and more accurate, lifelike nudes as it’s exposed to more data. But in the case of Barrat’s research, the GAN keeps producing “fleshy blobs that fool the discriminator pretty well, so overall they stop getting better at painting.”

Translation: Instead of some tasteful TnA, the robot is spitting out trippy, humanish renderings that give Salvador Dalí run for his money. And plenty of surprises too, like the fact that the GAN paints faces in yellow and purple globs despite that not being part of the training set input into the algorithm.

Could the Surrealists have predicted their art would be perfectly realized by an artificially sentient creative? They would probably have delighted in it.