Art History in 2 minutes as seen by AI
Strolling along the latent space of classical paintings through surrealism
100,000 classical paintings have been used to train a machine to understand the different classes of artistic styles and patterns encompassing their totality.
One of the dominating patterns of the 512 patterns identified is a classical portrait facing the left, another may be of trees around the center with a clear sky and green bottom and another of a calm sea…etc
Even more interesting than these Dimensions, Eigenvectors or latent representations of the 512 main ingredients of Classical Paintings is what happens if we mix some of these ingredients together.
Somewhere between pine trees, the portraits of the duchesses and a turbulent sea, we come across Salvador Dali’s and Yve Tanguy’s surrealism as if it were the roads or bridges connecting between all classical artwork. I am still wrapping my mind around its implications but I am certain that if we were to take 512 of the most representative dimensions of high-level patterns and cut these dimensions with a lower dimension space such as a plane in the case of a 3D space, we would most likely end up with a Daliesque sort of surrealism.
To dig deeper into the mysteries of what the machine sees or hides, I recommend NVIDIA’s article https://arxiv.org/pdf/1812.04948.pdf and the following summary about GANs.
One of the coolest and most exciting advances in Artificial Intelligence (AI) is Generative Adversarial Networks (GANs) that I have used to generate faces of non-existing people in 2016 and later on in 2019 NVIDIA used to generate hi-res hyper realistic faces: www.thispersondoesnotexist.com.
As creepy as this may sound, with all the ideas of things we can bring to life from scratch that I discuss in a previous article such as cats, fake airbnb postings, fake cvs, fake presidents, fake celebrities…etc, we can also generate new styles of art, music, architecture and cinema.