Posted: February 5, 2018 -
It’s difficult to determine what is more disconcerting: that an exhibit purposely displayed a fake Rembrandt, or that an Artificial Intelligence (A.I.) program painted it – or, rather, generated it on a 3D printer. The effort was the result of an 18-month-long collaboration between several organizations — including Microsoft, ING Bank and, J. Walter Thompson, a Dutch advertising firm. The presentation revealed that machines can be as creative as humans.
The quality of the portrait and the “intelligence” with which the A.I. went about deciding the painting it should create are eye-opening. No matter the style of the painting, however, A.I. technologies take the same approach to learning and then mastering a medium.
How Artificial Intelligence Technology Creates Art
The Rembrandt that the A.I. painted was a new composition the A.I. created in the style of the Master himself. The A.I. software analyzed 346 of Rembrandt’s paintings and decided the world needed (yet another) portrait of Rembrandt in his 30’s. The A.I. used about 170,000 fragments of paint from Rembrandts to develop a work of nearly 150 million pixels. Instead of painting the portrait on a canvas, however, the program manufactured the entire painting through an additive process of layers of materials.
The process for tutoring A.I. to paint involves researchers feeding millions of images to the software’s “artificial neural networks”. The networks mimic the way the brain learns. After it has made millions of associations between the data points, the A.I. can learn enough to look at a picture, then recognize and classify specific features in it, according to Mentalfloss.
The Art and A.I. lab at Rutgers University has been using similar algorithms to enable software to learn about and recognize styles of painting as varied as Baroque, Pointillism, Color Field, Rococo, Fauvism, and Abstract Expressionism. The laboratory modified the system known as Generative Adversarial Networks (GAN) to create and then critique its works. The researchers at Rutgers settled on Abstract Expressionism to expand their A.I.’s creative palette.
A.I. Projects Push the Boundaries of Art
After its success with its GAN algorithms, the Lab at Rutgers then adjusted its software to generate paintings that fall outside the boundaries of the art categories in which scientists had schooled it. “The images generated by CAN do not look like traditional art, in terms of standard genres (portrait, landscapes, religious paintings, still life, etc.),” according to a paper the Lab produced about the study. The tech giant Google took a more mathematical approach to teaching its A.I. how to create art.
Google researchers let their A.I. algorithms create their own art. They named the A.I. program Deep Dream, to reflect the perception that the graphics the A.I. created were dreams from the machine itself. The A.I. works by scanning a photo or some other image into the computer. The mathematical modeling upon which developers built Deep Dream looks for patterns in the picture. It enhances the patterns, the repeats the designs. “This creates a feedback loop: if a cloud looks a little bit like a bird, the network will make it look more like a bird,” Google said in a blog post about the project. “This in turn will make the network recognize the bird even more strongly on the next pass and so forth, until a highly detailed bird appears, seemingly out of nowhere.”
Google put its art on auction in 2016 to fund the Gray Area Foundation for the Arts in the San Francisco Bay Area. Some pieces sold for as much as US$8,000, according to the Wall Street Journal.
But is A.I. Art Good?
The Rutgers Lab invited groups of people to view their CAN- and GAN-generated art. They also asked viewers whether a machine or a human had generated specific Abstract Expressionist pieces that real-life artists had created. The fourth set of art involved non-figurative works produced by humans and shown at Art Basel in 2016.
The Abstract Expressionist works rated the highest. Eighty-five percent of respondents correctly identified them as the work of a human artist. Survey participants also thought people made 53 percent of the CAN images. They also felt human artists had made 35 percent of the GAN images. Viewers believed human artists had created 41 percent of the Art Basel works, according to Artnet.
So while the jury is still out about whether or not machines are creative, developments in artificial intelligence are forcing people to question humanity’s exclusive claim to creativity. Whatever the answers, though, economics will become a factor in whether people display A.I.-generated art or art created by a human being.
A Look Over the Horizon
While A.I. technology does not pose a near-term existential threat to artists, A.I. may provide opportunities to develop new media, expositions, and markets.
For example, as A.I. programs become more accessible to the public, they could be used to reduce the cost of art in commercial spaces. For instance, condominium property owners who would like to adorn their properties with art may find human artists unaffordable. A new generation of artists armed with A.I. algorithms may meet the esthetic goals of property investors at a fraction of the cost and time it would otherwise take.
Architects could keep an A.I.-artist on staff to design and 3D-print art to adorn the interiors of buildings they’ve developed. The works could transform the intent of structures beyond that in which humans merely live and work.
Another intriguing job opportunity for A.I. could involve “painting” the portraits of individuals and families in styles the sitters choose: Pollack, Picasso… even Rembrandt. The costs for such customized styling would become affordable even to families on modest incomes.
Art markets will continue to auction the Pollacks, Picassos, and Rembrandts of their time for millions of dollars. However, art generated by artificial intelligence may provide value to ordinary citizens in ways that only machines can dream.
Image and video credit: https://www.nextrembrandt.com