Judging robot art
Is art painted by robots any good? Andrew Conru runs an annual competition to find out – and most of us wouldn’t mind hanging the results over the sofa
Can robots make art that humans find beautiful?
COMPOSITION, TECHNICAL ABILITY, artistic merit. The winning painting of an annual art competition, a reinterpretation of Cézanne’s Houses at L’Estaque, has all three. And while the artist behind the work is a human, Pindar Van Arman, the painting was done by a robot he designed and built.
Echoing the impressionists who set up their own salon for judging their innovative paintings in the 1800s, Andrew Conru founded RobortArt, a competition to judge the work created by human artists teaming up with robots and algorithms. “I started trying to answer the question of whether or not robots can create something beautiful,” Conru told PC Pro. The answer is yes, but humans remain heavily involved in the process, coming up with the idea, training or programming the algorithm. Only then does a robot arm “with gobs of paint” start decorating the canvas.
While a machine can be programmed to do exactly what the artist wants, algorithms can also study what we find beautiful and try to mimic it. “There’s lots of machine-learning and neural-network products that are taking a corpus of a bunch of art and then trying to figure out what elements of pixels on the screen produce something that is aesthetically pleasing,” Conru said.
The next step is getting it from digital display to canvas. That could be using a robotic arm to layer each colour on, or to develop software that analyses the idea and translates it into brushstrokes. Some projects let the robot painter have more leeway, and correct its work as it goes, teaching it via a corrective feedback loop.
Creative robots
Of the almost 100 teams that have entered the competition, “the ones that stand out tend to be the ones that use a more traditional idea of a robot arm holding a brush,” said Conru. “But we have a lot of creative ones that are using drone technology or using brainwaves for brushstroke systems, ones that are really pushing the envelope.”
Judging the work is no different than looking at art in any gallery, he notes: “If you have something created by a robot and something created by a human, can you get a similar aesthetic feedback loop? Does it look beautiful to you and did you get an emotional response? If so, we think that the artform has achieved a level of validity.” Alongside a panel of judges, Conru also opened up votes to the public, with registered users given ten votes to distribute to their favourite works.
Challenges
Three years in, the art is getting better, but challenges remain. “The sophistication of some of the art is improving, but we were definitely waiting on some of the hardware to catch up with the software side,” he said. “It’s funny when the biggest limitations that artists have is actually getting their hands on an easy-to-use, functional robot arm.” Cost isn’t the only hurdle: some of the user interfaces on the robot arms aren’t exactly easy to use – and plenty of would-be robot painters aren’t tech-savvy enough to work their way through confusing systems, Conru said.
He hopes less expensive, easier-touse robotics will spark a revolution in augmented painting. “It’s kind of like when the web came out and made it really easy for somebody to create a web page, that kind of opened the door for a lot more creativity to happen,” he said. “We’re seeing similar things happening here. There’s not yet a $5,000 robot arm that artists can use easily, but we’re hoping for it.” When it does, expect it to be a real Monet maker. (Sorry.)