Computers Are Learning About Art Faster than Art Historians
An algorithm took just a few months to draw connections between artists that scholars have been working on for years
/https://tf-cmsv2-smithsonianmag-media.s3.amazonaws.com/filer/7d/bd/7dbd1b15-74e2-4858-9f38-0c979186b8b6/42-30876037.jpg)
Computers are getting better at some surprisingly human tasks. Machines can now write novels (though they still aren’t great), read a person’s pain in their grimace, hunt for fossils and even teach each other. And now that museums have digitized much of their collections, artificial intelligence has access to the world of fine art.
That makes the newest art historians on the block computers, according to an article at MIT Technology Review.
Computer scientists Babak Saleh and Ahmed Egammal of Rutgers University in New Jersey have trained an algorithm to look at paintings and detect the works’ genre (landscape, portrait, sketch, etc.), style (Abstract Impressionism, Baroque, Cubism, etc.) and artist. By tapping into the history of art and the latest machine learning approaches the algorithm can draw connections that had only been made by human brains before.
To train their algorithm, researchers used the more than 80, 000 images from WikiArt.org, one of the largest online collections of digital art. The researchers use this bank of art to teach the algorithm how to key in on specific features, such as color and texture, slowly building a model that describes unique elements in the different styles (or genres or artists). The end product can also pick out object within the paintings such as horses, men or crosses.
Once it was schooled, the researchers gave their newly-trained algorithm paintings it had never seen before. It was able to name the artist in over 60 percent of the new paintings, and identify the style in 45 percent. Saleh and Elgammal reported their findings at arXiv.org.
The algorithm could still use some tweaking — but some of the mistakes it made are similar to those a human might make. Here’s MIT Technology Review:
For example, Saleh and Elgammal say their new approach finds it hard to distinguish between works painted by Camille Pissarro and Claude Monet. But a little research on these artists quickly reveals both were active in France in the late 19th and early 20th centuries and that both attended the Académie Suisse in Paris. An expert might also know that Pissarro and Monet were good friends and shared many experiences that informed their art. So the fact that their work is similar is no surprise.
The algorithm makes other connections like this one—connecting expressionism and fauvism, and mannerism with the Renassance styles that were borne out of mannerism. These connections themselves aren’t new discoveries for the art world. But the machine figured them out in just a few months of work. And in the future the computer could uncover some more novel insights. Or, in the nearer future, a machine algorithm able to classify and group large numbers of paintings will help curators manage their digital collections.
While the machines don’t seem to be replacing flesh-and-blood art historians in the near future, these efforts really are mearly the first fumbling steps of a newborn algorithm.