Can robots make art Yes - but don't ask them to write a poem

19.05.2016
Robots can paint, but when it comes to writing, they shouldn't quit their day jobs. That's the combined conclusion from results of two contests announced this week.

On the upside, artificial intelligence created some pretty impressive works for RobotArt.org's first annual $100,000 competition, the results of which were announced Tuesday.

The contest challenged artists and engineers to create a robot that painted like a real artist. Essentially, the aim was to get "as many teams as possible to set up a robot that can do any sort of painting," the contest site explains. Fifteen teams from seven countries responded with more than 70 robot-created paintings made with a variety of techniques.

Robots were judged through a combination of public voting, professional opinion and how well the team met the spirit of the competition. First place and $30,000 went to TAIDA, a robotic arm from National Taiwan University that painted "in a manner very similar to a classical painter," the contest's organizers said.

"Once it achieved the desired color, it would begin by first laying down and underpainting before continuing on to a refinement layer," they said. "In the refinement layer, the robot would repeatedly compare the picture drawn on the canvas with what it was attempting to paint to find the most different area, and then try to make it less different, similar to how a classical painter goes about completing a canvas."

Second place and $18,000 were awarded to cloudPainter of the United States. Third place and $12,000 were awarded to NoRAA of Italy.

"The results of this competition show a significant step in the advancement of robotics and artificial intelligence to create beauty," said RobotArt.org founder Andrew Conru.

It doesn't look like the capacity for creating beauty extends to the literary world, though -- at least not yet.

Organizers of Dartmouth College’s “Turing Tests in Creative Arts” announced on Wednesday that there were no winners in their competition for AI algorithms to create “human-quality” short stories, sonnets and dance-music sets.

The goal was to show whether human judges -- in the form of literary readers and party-goers on a dance floor -- could distinguish which creations were generated by machines and which by humans. The contest’s organizers expected dozens of entries in each category but received far fewer submissions than they expected -- "a testament to the difficulty of writing sophisticated code that creates another dimension of AI -- creative intelligence," they said.

In the poetry contest, there were two entries that each generated two sonnets; in the literature contest, there were three entries that each generated a short story. Some human-generated entries were included as well.

Passing the contest's Turing Test required that a majority of the judges rate the entries as “human.” None of the machine-generated sonnets or short stories fooled any of the judges, with the exception of one short story that fooled one judge.

"Our algorithms seem not yet able to imitate human kinds of poetry, but the code that was submitted was still amazing,” says Dan Rockmore, a professor of mathematics and computer science who was one of the contest's organizers.

AI fared better in creating dance-music sets: Two of the eight algorithmic entries submitted fooled about 40 percent of the human voters.

You can test out your own ability to distinguish AI-created from human-made dance sets in a poll on the Dartmouth site.

Although there were no winners, the Dartmouth contest organizers said several of the entries in each category were sophisticated enough to deserve a portion of the prize money.

A full account of the results, as well as a prize-winning sonnet, can be found on the project's site.

Katherine Noyes

Zur Startseite