What would you say if someone told you that artificial intelligence is now capable of creating actual works of art? What would you say if you found out that artificial intelligence can create new art styles as well?
Move over, Picasso—or any other human artist out there. Scientists have developed an artificial intelligence program that can produce works of art in new styles. The program was designed to create art with a “twist”: that the work won't fall into any existing art style. However, people were still able to recognize the program's work as art, in spite of the existence of this “twist”.
Marian Mazzone, an art historian at the College of Charleston in South Carolina who was also part of this project, says that the idea was to create art that was “novel, but not too novel”. But how can artificial intelligence create art?
A team of researchers at Rutgers University and Facebook's AI lab, as well as Mazzone, developed the AI program by modifying an algorithm. This algorithm is a generative adversarial network (GAN) in which two neural nets interact to get better results from each interaction. The interactions go like this: one neural net creates a solution and the other one modifies it in some way. This goes on in loops until the algorithm reaches the desired results.
So how does the AI program create art as well as new art styles? A generator network produces images by taking on one of the neural net roles. This network creates images. A discriminator network, meanwhile, plays the other neural net role. The discriminator network is what distinguishes images that we would consider to be art from images that we wouldn't consider to be art. This network can also distinguish between different art styles.
The researchers made it so that the generator network can create images that the discriminator would recognize as art, but wouldn't be able to classify in the different art styles. This way, the AI was able to create new images in new art styles, as opposed to AI that can recreate images in the styles of famous artists.
Of course, what is art without an audience? The researchers showed the images that the AI created alongside images that humans created. Audiences viewed the artwork online and answered a survey on the works they saw. They answered questions on how complex they found the images to be, how inspiring, and how the images affected their mood. The researchers found that the AI-generated works received slightly better scores than works that humans produced.
Is this the future of art? Maybe. The researchers note that AI has already produced works for galleries, and has produced images for an animated film.
An important question, of course, is how do human artists fit in here? Will AI replace our artists as well? This kind of AI-generated art still relies quite heavily on humans, so we won't be out of the picture any time soon. Perhaps artists may even learn a thing or two from the new art styles that the AI program has generated.
Get weekly science updates in your inbox!