Monday, January 22, 2018

Can Artificial Intelligence Make Art?


In February's Scientific American, technology columnist David Pogue wonders if human artists and composers should start worrying about a new development in artificial intelligence (AI):  the automated composition of music and production of paintings.  Computer scientist Ahmed Elgammal's Art and Artificial Intelligence Lab at Rutgers University is developing algorithms that start with well-known famous works of real art and abstract elements of style and composition from them.  Then the machine can either be set to do imitations in the same style, or else he turns on a "style ambiguity signal" that forces the digital Rembrandt to deviate from the style it's learned. 

I have viewed some of these products on the lab's website, and while I make no claims to be an art critic, my impression is that Rembrandt doesn't have much to worry about.  In fairness to Elgammal, he doesn't claim that what his system is doing is just as good as human-created art.  Rather, he sees himself as exploring theories of art creation using AI to see what happens if style rules are either slavishly followed or intentionally broken.

When he mixed the products of his AI "artist" with works by actual humans in a couple of different sets—abstract expressionism and contemporary art from a recent European art show—he found that people who viewed the artworks without knowing which was by a human and which was by a computer, often picked the computer-generated ones as more "intentional, visually structured, communicative, and inspiring" than those made by unaided humans.  He was surprised by this outcome, but he shouldn't have been.

Most people will agree that much visual art that is bought and sold for millions of dollars today doesn't look much like the artworks that were painted, say, a hundred and fifty years ago. Elgammal has happened to come along with his AI artist at a time when the so-called standards for what constitutes art are all but nonexistent.  Last year The New Yorker carried a story about a young man named Jean-Michel Basquiat who mainly wanted to be famous.  He tried music as a path to fame first, but was discouraged by the fact that it takes years of practice to become even an adequate musician.  So he switched to art.  Starting with graffiti, he attracted the attention of the art world, rocketed to international fame, and died of a heroin overdose at the age of 27.  The magazine's art critic Peter Schjeldahl thinks that his art is worth looking at, but probably not worth paying $110 million for, as a Japanese business man did last May for a Basquiat work from 1982.  Schjeldahl himself described it as a "medium-sized, slapdash-looking painting of a grimacing skull."  Judging by the photograph in the article, that's a pretty accurate description.

My point is that what passes for art these days is a departure from what has passed for art in the past, well, several thousand years.  Up until the nineteenth century, artistic works represented both recognizable objects, and also the higher operations of the human mind and spirit, operations that distinguish human beings from the lower animals.  G. K. Chesterton regards the production of art as one of the primary distinctions between people and other animals, and points to the cave paintings such as those in Lascaux, France, as being evidence that those who painted them were humans like us. 

One chronic concern that arises as AI advances into more areas of endeavor formerly regarded as exclusively human, is that when AI starts to do a certain kind of thing better and cheaper than people, what will happen to the people who earn their living doing it now?  So far, humanity has survived the replacement of telephone operators by automatic dialing, elevator operators by pushbutton elevators (everywhere except at the United Nations building, I'm told!), and more recently, the advance of AI into the professions of engineering, medicine, and even law.  Right now, the unemployment rate in the U. S. is at a historic low, but that is due mainly to an economy that is close to overheating, and doesn't take into account the millions of people who neither look for work or are particularly troubled that they're not working.  And here is where we find the real matter to be concerned about.

The issue isn't whether AI will send some artists to the unemployment line.  The real issue is how we regard art and how we regard humanity.

When Chesterton wrote in 1925 that "Art is the signature of man," he didn't mean just any random scrawl.  He had a particular thing in mind, namely, that the portrayal of nature as interpreted by the human spirit is unique to man.  Certainly no other animal produces anything that is generally regarded as a work of art.  I am aware of the bowerbirds of Australia and New Guinea which construct large elaborate arches of sticks and decorate them with blue objects and sometimes even paint the walls.  But this is simply instinctive behavior directed at attracting a mate.  No one has seen bowerbirds exchanging worms for a particularly fine bower and signing bills of sale. 

If people today can't seem to tell the difference between computer-generated art and human-generated art, the reason isn't that the computer is now as artistic as a human artist.  The problem is that artists have degraded their craft to the level of a machine-made product, and taught the general public that yes, that is indeed art even if I tied brushes to two turtles and let them crawl across the canvas.  When Marcel Duchamp tried to exhibit an ordinary urinal as art in a 1917 New York art show, the show's committee rejected it, but photographer Alfred Stieglitz allowed him to put it up in his studio.  In 2004, 500 "renowned artists and historians" reportedly selected this work, called simply "Fountain," as the most influential artwork of the twentieth century.  And it was made by a machine.

Sources:  David Pogue's column "The Robotic Artist Problem" appeared on p. 23 of the February 2018 issue of Scientific American.  Some creations of Prof. Elgammal's AI artist can be viewed at https://sites.google.com/site/digihumanlab/home.  (The ones with "style ambiguity" turned off, at https://medium.com/@ahmed_elgammal/generating-art-by-learning-about-styles-and-deviating-from-style-norms-8037a13ae027, are truly creepy.) A brief introduction to Jean-Michel Basquiat can be found on the New Yorker website at https://www.newyorker.com/culture/cultural-comment/lets-pause-to-appreciate-basquiats-hundred-million-dollar-painting.  Chesterton's comments about cave paintings are from pp. 30-34 The Everlasting Man, reprinted in 2008 by Ignatius Press (originally published 1925).  And I also referred to Wikipedia articles on Marcel Duchamp and Fountain. 


No comments:

Post a Comment