Elsewhere in the AI world, researchers are playing other art-historical games. Ahmed Elgammal, director of the Art and Artificial Intelligence Lab at Rutgers University in New Jersey, is working with a system that he calls CAN — a ‘creative’ rather a ‘generative’ network. The basic binary hokey-cokey is the same — maker and judge, artist and critic — but CAN is specifically programmed to produce novelty, something different from what it sees in the data set, which in this case consists of all manner of paintings from the 14th century on.
‘I am surprised by the output every time I run it,’ says Elgammal. ‘An interesting question is: why is so much of the CAN’s art abstract? I think it is because the algorithm has grasped that art progresses in a certain trajectory. If it wants to make something novel, then it cannot go back and produce figurative works as existed before the 20th century. It has to move forward. The network has learned that it finds more solutions when it tends toward abstraction: that is where there is the space for novelty.’
This raises the intriguing notion that AI algorithms do not merely make pictures, they also tend to model the course of art history — as if art’s long progression from figuration to abstraction were part of a program that has been running in the collective unconscious for half a millennium, and the whole story of our visual culture were a mathematical inevitability.
No AI researchers are claiming that much just yet. They are still addressing the fundamental question of whether the images produced by their networks can be called art at all. One way to do that, surely, is to conduct a kind of visual Turing test, to show the output of the algorithms to human evaluators, flesh-and-blood discriminators, and ask if they can tell the difference.