AI, art, and cars are not necessarily a combination we’ve ever expected to use in the same sentence. But BMW’s exciting new project named The Ultimate AI Masterpiece managed to synthesize all three. It brings us another fascinating showcase of AI’s creativity once it has learned from real artists.
Can AI produce art on par with the world’s greatest artists? Maybe – if it learns their techniques to acquire its own creativity. To achieve this, BMW teamed up with a creative technologist Nathan Shipley of Goodby, Silverstein & Partners, and Gary Yeh, Art Collector & Founder of ArtDrunk. They utilized StyleGAN AI software fed with more than 50 thousand images that included artworks from the past 900 years. A set of 50 contemporary works was used as well. Having analyzed all these inputs, AI became a capable artist all on its own and was able to produce inventive, original work. These new artworks were projected onto a virtual rendition of BMW’s 8 Series Gran Coupe.
“We start with this huge dataset of artwork that covers more than 900 years of art history. As AI sees this, it learns to generalize what a painting looks like, and then you can use it to create new ones,” Shipley explains. “We wanted to take that a step further beyond the history of art to see what the technology what the technology could create with art from modern living artists,” he adds.
Gary Yeh stepped up to the task. One of the artists he picked was a contemporary Korean painter Lee Bae, famous for his charcoal artworks. As Shipley outlined, AI recognized something as intricate as textures and then generated an “evolving stream of new textures”. Those were “informed by his work, but also unique”. The virtual art installation premieres in conjunction with Frieze New York 2021
Is AI the future of art?
This is certainly not the first time that artificial intelligence meddled with art. It created some charmingly bizarre novels, and it’s perfect for composing a “white noise” type of music. What about painting? The first AI-generated piece of art was sold at Christie’s in 2018 for $432,500 — nearly 45 times its high estimate. Back then, it was shocking: today, with NFT jpegs selling for $69 million, not so much.
Even then, researchers played “art-historical games”, as Christie’s put it. In New Jersey, Ahmed Elgammal runs the Art and Artificial Intelligence Lab at Rutgers University, where he has experimented with a so-called CAN system – a creative adversarial network. First, CAN is trained on a dataset. Subsequently, it generates brand new images constantly evaluated against the original dataset to distinguish the “real” and “fake” artworks. This way, the network keeps improving until it produces believable new creations.
CAN art, courtesy of AI
Elgammal pointed out that CAN’s art tends to be mostly abstract, which is intriguing. “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,” he said. AI did exceptionally well, too: experiments proved that human subjects couldn’t distinguish between “real” and “fake” artworks, and sometimes even rated the generated images higher.
Christie’s suggested that it almost feels as if AI modeled the course of art history. It learned from the classics but concluded that abstraction was the way to go anyway. And while we don’t know if that’s true, it would be quite mindblowing. Especially for those who love to sum up human existence in mathematical equations.