Machines might one day replace human laborers in a number of professions, but surely they won't ever replace human artists. Right?
Think again. Not even our artists will be safe from the inevitable machine takeover, if a new development in artificial intelligence by a team of researchers from Rutgers University and Facebook’s A.I. lab offers a clue of what's to come. They have designed an A.I. capable of not only producing art, but actually inventing whole new aesthetic styles akin to movements like impressionism or abstract expressionism, reports New Scientist.
The idea, according to researcher Marian Mazzone, who worked on the system, was to make art that is “novel, but not too novel.” It's such an effective system that the art produced by it is already being given the thumbs-up by human critics when presented in public.
The algorithm at play is a modification of what's known as a generative adversarial network (GAN), which essentially involves two neural nets that play off against each other to get better and better results. The model used in this project involved a generator network, which produces the images, and a discriminator network, which "judges" whether it's art. The discriminator is programed with knowledge of 81,500 examples of human paintings that either count as art or don't, as well as knowledge of how to categorize art into known styles, and it uses these benchmarks to carry out the judging process.
This may seem overly simplistic, but there's a twist. Once the generator learns how to produce work that the distributor recognizes as art, it's given an additional directive: to produce art that doesn't match any known aesthetic styles.
“You want to have something really creative and striking — but at the same time not go too far and make something that isn’t aesthetically pleasing,” explained team member Ahmed Elgammal.
The art that was generated by the system was then presented to human judges alongside human-produced art without revealing which was which. To the researchers' surprise, the machine-made art scored slightly higher overall than the human-produced art.
Of course, machines can't yet replace the meaning that's infused in works by human artists, but this project shows that artist skillsets certainly seem duplicatable by machines.
What will it take for machines to produce content that's infused with meaning? That might be the last A.I. frontier. Human artists can at least hang their hats on that domain ... for now.
“Imagine having people over for a dinner party and they ask, ‘Who is that by?’ And you say, ‘Well, it’s a machine actually’. That would be an interesting conversation starter,” said Kevin Walker, from the Royal College of Art in London.