RobotArt Competition Highlights Advances in AI-Generated Painting
The second annual competition features machines competing for $100,000 in prizes.
Depending on how you define your terms, machines have been making art for decades — even centuries. From sticks to brushes to styluses, artists have deployed increasingly sophisticated tools and technology since the first documented cave painting, found in the famous Lascaux caves.
In recent years, rapid advances in robotic technology and artificial intelligence have arguably initiated an entirely new era of artistic expression — one in which machines are no longer just tools, but collaborators, or even artists in their own right. Algorithms and dedicated AI programs are generating original ideas, imagery, and music — check out Google's Magenta initiative for some recent and occasionally hallucinatory examples.
In the realm of traditional visual art, the RobotArt competition takes things in yet another direction. Now in its second year, the competition requires robots and AI systems to use actual paints and brushes to physically create art on canvas — no printers allowed. This year, 39 robot artists have submitted more than 200 paintings to the competition.
Most of the entrants represent team efforts — systems developed by research groups at robotics labs or university engineering departments. A few of the robots have been designed by entrepreneurs, or even traditional working artists that have drifted into the technical end of things.
Of the 39 entrants this year, no two teams took the exact same approach, but the goal was the same: to produce artwork physically painted by robots, with real brushes and canvas. Or snakes. More on that in a minute.
This is where the RobotArt competition gets really interesting. While a few of this year's entrants are essentially mechanical in nature — robotic arms programmed or controlled by a human artist — most incorporate some kind of generative function within the machine itself. These robots are designed to collaborate with human artists, or in some cases, create art all by themselves.
For instance, one of the bots in competition creates original imagery by tracking the movements of dancers and transposing the choreography into lines and colors. Another uses electrodes and custom algorithms to turn the electrical activity of the brain into art. Then there's the Anguis system, in which a snake-shaped soft robot is cut loose on a horizontal canvas, generating abstract imagery as it slithers around.
The robot artists and their human handlers will compete for more than $100,000 in awards and prizes in this year's competition. Winners will be determined based on a combination of public voting via Facebook, plus assessments by a panel of judges including working artists, critics, and technologists.
The selection process is overseen by RobotArt founder Andrew Conru, who is funding the competition personally and plans at least a five-year run. If that name sounds familiar, it’s might be because Conru was one of the pioneers of the early internet. He founded FriendFinder, one of the earliest experiments in social network and dating services, as well as the digital advertising platform Adknowledge. Conru is currently CEO of FriendFinder Networks.
“I made a lot of money on tech and I thought I wanted to give something back,” Conru said from his home in Seattle. “I thought, what's the best way to leverage it? What about a competition to encourage creativity in tech people all over the world.”
Another goal is to get more students to consider STEM disciplines — science, technology, engineering, and math — as a path to a creative career, Comru said.
As to the question of whether machines can actually create art, that's a question for philosophers and metaphysicists. The RobotArt website, in fact, has a page dedicated to the debate. But Conru said that the wide range of artistic approaches in this year's competition suggests robotic art, whatever it might be, is evolving along with the rest of the AI world.
“Some of the art this year is what I call reinterpreted; it's copied or sourced from an original photograph or artwork,” Conru said. “But with about half of this year's entries, the original source material was generated in collaboration with some kind of AI or deep learning system. In some cases, the AI is creating its own art out of other kinds of input — brainwaves, for instance. These are all stepping stones.”
WATCH: The Freaky Way Robots Are Becoming More Human