As part of their AI for Good Global Summit, the UN has explored the role of artificial intelligence in the creation of art, putting forward a call for “AI-powered” artwork from creators across the globe.

Really, almost anyone with enough patience, basic IT capacity and the desire to learn could scribble off some machine learning-based art. WIRED's Tom Simonite did it with open source tools and the same machine learning software used by researchers at Facebook and IBM.

You can try it too, with premium and free programs, like Runaway ML, GANBreeder, Magenta and Processing. Want professional training? Visit ml5.js or the more costly UC San Diego’s first-ever Machine Learning for the Arts.

Machines in Art Class

In truth, it's the machine learning programs, such as the popular generative adversarial network (GAN), that do the art, not you. You feed the algorithm images of an original picture. You, then, train it to choose data that appears in the base picture and to reject data outside of it. The training is repeated over and again, until the program reshapes the base picture into something new with its computer-processed art.

Sometimes, programs surprise their artists with unexpected results. For example, a GAN fed with portraits spat out a series of surprising deformities that Christie's actually exhibited for sale.

Given the growing extent to which AI art invades galleries and museums, there´s small wonder ¨regular¨ artists are concerned they´ll lose their jobs.

AI masterpieces can make a killing, with prices like $16,000 USD for a work called Future Blink that people say they have a hard time realizing wasn't made by human hand, and a pre-sale-estimate-smashing $432,500 USD for a blurred face titled "Portrait of Edmond Belamy."

"The Portrait of Edmond Belamy" produced by Paris-based arts-collective Obvious was collated from images of 15,000 portraits from various periods. Christie's sold it for well over double the combined price of an Andy Warhol print and a bronze work by Roy Lichtenstein displayed in the same room.

Is it art?

Philosophers have long squabbled over how you define art. Take the urinal of Marcel Duchamp’s controversial sculpture Fountain, for example—industry opinion (somewhat controversially) tags that art.

Says Mike Rugnetta of PBS Idea Channel,:

"Art needs an audience and tastes vary. Computer programmers haven't actually made art. They have made works—objects, media, performance, software—works. What makes art is an audience. An artist makes a work and audiences turn that work into art through appreciation.¨

So you may not understand, or like, AI-generated art—but that doesn't mean it's not art.

Graphic artists use software programs like PaintShopPro to create art. That sector has become a recognized art—graphic art. Likewise, cameras produce photos. Yes, it's humans snapping those photographs, but it's humans who manipulate computers too. (Read also: Can AI Have Biases?)

AI Art Lacks Emotion

Pulitzer Prize–winning art critic Jerry Saltz contrasts AI-generated art with Japanese and Indian pornographic statues. Displaying a picture of AI-created "The Butcher’s Son" (which incidentally won the Lumen Prize for best-technology created art,) he told viewers,

¨In India every god is having sex with a thousand other people, animals, each other. and then this! You see how it has been tamed and held in. Open your sphincter, man, and let go!¨

AI-produced art, the argument goes, lacks emotion. It is dull, generic, and boring,

Humans Live in a Social Context that Machines Lack

Back on Idea Channel Mike Rugnetta says,

¨A good work of art is supposed to stimulate. It's traditionally beautiful, hard to make and should tell an affecting story. Human art is chaotic. Machine learning, on the other hand, is reflective, boring, [dispassionate]. It is humanity scrubbed away.¨

The Argument for AI Art: Novelty

Ahmad Elgammel, director of the Art and Artificial Intelligence Laboratory at Rutgers University, says he and his associates follow the theory of psychologists like Colin Martindale who found novelty is the most important stimuli in works of art:

¨Artists seem to intuitively understand that they’re more likely to arouse viewers and capture their attention by doing something new.¨

And AI is nothing if not stimulating. Check out this immersive art exhibit that reimagines New York City! The artist, Refik Anadol, used over 300 million images to render the city's architectural past, present and future.

¨Artificial intelligence,¨ Elgammel explains, ¨creates something new that builds off what already exists.¨ (Read also: 5 Nature-Imitating Technologies.)

Or not?

Philosopher Sean Dorrance Kelly of MIT disagrees that AI art contains anything new. The computer did not create the art. It piggy-backed on the work of humans. Was there an innovation? Something new? Certainly not. Therefore you can´t call machine-learning generated content art.

¨I want the robot,” Saltz exclaims, ¨to tap into its inner robot. Briefly I would give the computer a work and I would say: ´Çomputer make a KulumSchtadt. And then put it in hell. And then put it in the 46th century BC.´ Give us something passionate and new!¨

Further, the algorithms aren't autonomous. They're the decisions of humans who design those algorithms. So, in essence, it is the humans who created the art and humans who trained the machine and the machine has simply spat out the original according to its programming.

Rugnetta poses the question:

¨We can easily find novelty in machine art but will humans always be the gold standard for meaningful art-making? Will we ever see machine-made works as made independently of human control?¨

Only then can we call AI art “novel.” (Read also:11 Quotes About AI That Will Make You Think.)


Art - ex nihilo

It seems to me the most persuasive argument against AI as art was put forth by Rugnetta who, citing Magenta researcher Doug Eck, told PBS viewers:

¨Art is something from nothing. You look at someone or something and draw it or conceptualize some idea.¨

Yes, machines can make art - but they only copy your art. For example, Google-made Magenta, with its recurrent neural networks, generates art and music from pre-existing works.

¨All a computer does,¨ Rugnetta added, ¨is learn what art is by analyzing art and making its own art based on another piece of art… How can you make art truly generative? Start from something to nothing.¨

In short, artistic works are supposed to be significant, innovative and original—not a repeat of one or more human-made artifacts.

Final Thoughts

Art or not, the paintings are lovely. If I had the time and interest, I'd love to experiment with some AI-teaching programs myself. After all, the money would come in handy and, who knows, I may even win some competitions...

Resources