From the issue dated May 9, 1997
Art Professor Creates a Computer That Draws and Paints
Using artificial intelligence, he"taught" a machine his ideas about the nature of images and composition
By Peter Monaghan
San Diego -- Thirty years ago, a well-established London artist named Harold Cohen asked himself a question that propelled him into a new painting career -- one in which he hardly ever picked up a brush.
The question seemed simple:"What are the minimum conditions under which a set of marks functions as an image?" But it led him on an extraordinary artistic quest, one that he pursued through 29 years as a faculty member on the University of California's campus here, and that he is continuing now as an emeritus professor of art.
The journey began in 1968, when he learned about developments in artificial intelligence during a visit here. He decided almost immediately to see how well he could encode his thoughts on the nature of images into a computer program that would draw and paint.
The result is his unlikely alter ego: a program he calls AARON, which drives a robotic painting machine. A large, flat surface in the machine holds paper, and a painting arm operates on X and Y axes. The arm picks up bottles of fabric dye, manipulates their taps to mix colors, and applies the dye to the paper using sponge brushes. It even cleans the brushes afterward.
The work has attracted great public interest, as well as admiration in the world of computer programming. AARON is"a tour de force of art, artificial intelligence, and robotics," according to Oliver Strimpel, executive director of the Computer Museum in Boston, which has shown Mr. Cohen's work since 1979.
Actually, the paintings aren't altogether his work -- they're collaborations between him and AARON. Although Mr. Cohen wrote the code that operates the machine, he never knows exactly what it will create. He does know, because he wrote the program, that a limited number of humans will appear in an interior, possibly with other elements -- perhaps a table, or plants, or both.
But he can't predict where the people will be in the frame, or where they will be positioned with respect to the objects, or the figures' expressions. AARON"decides" that, according to its code.
Mr. Cohen is a gray-haired man of soft but richly descriptive speech. In his studio here, he talks about his program's achievements by referring to a painting on the wall. It shows two female figures amid overhanging foliage. One woman's eyes and upper head are, surprisingly, obscured by leaves. Her open, large-lipped mouth is oddly expressive. But it is her hands, twisted in front of her, that are most compelling.
The work has undeniable artistic appeal."One of the interesting things about it as a form of imagery -- and which, by the way, identifies it as not being human -- is that if you look at any one part, it's wrong. It's almost all wrong," Mr. Cohen says. But when the parts are put together,"it comes out with a very plausible image."
If, for instance, the program obscures a face behind an object, it does so because it"has no particular reason for preferring one thing or another." AARON"doesn't know very much," Mr. Cohen says. But he has"taught" it -- programmed into it -- lessons about the real world.
Some of the lessons are common-sense:"You're not balancing your pen on the top of your head," Mr. Cohen tells a visitor -- nor would AARON draw a figure doing so. Some involve the forms of things: In figuring out how AARON would generate human faces, for example, Mr. Cohen referred to detailed medical illustrations and wrote code that allows the program to construct faces around internal reference points.
AARON is no Rembrandt. Still, Mr. Cohen points out, no one paints quite like AARON, either:"It does things that I would never have done as a human artist. That's the point where I start to learn things."
In 1968, when he arrived here, Mr. Cohen was at artistic crossroads. A central figure in London's art scene, he had exhibited his paintings in one-man and group shows in major museums and galleries world-wide. In 1966, he had represented the United Kingdom at the Venice Biennale.
He speaks of his achievements without false modesty:"I had a big reputation, and I think deserved it," he says with a glimmer of a smile."I just didn't like the life style that went along with it, so I opted out."
"All my young fellow rebels were settling into being famous artists. I was already on a track where, if I behaved myself, I would get my retrospective at the Tate, and the Arts Council would buy two pictures a year. I'm not into that game."
His own artistic output was the real issue:"I liked the paintings I was doing, but I had an uncomfortable feeling that it hadn't really required me to make them." Most of his early work is in storage in London now."I've not seen it for a number of years," he says."When I do see it, I think, 'You were a pretty damn good painter, just like everybody said.' But I'm not in the habit of looking back."
Computing gave him a way to move forward when he arrived here in 1968 on a one-year professorship."Once I discovered computing here, it was perfectly obvious there was no way I was going to go back to live in England."
He named his first program, the one he is still building, AARON."When I started doing this kind of stuff, I assumed that I would be writing a whole series of programs, so I thought I'd better start at the beginning of the alphabet. After about six months, I remembered that AARON is my own Hebrew given name. As a devoutly atheist Jew, I'd forgotten that."
He has written AARON in LISP, the early computing language that most artificial-intelligence researchers still prefer. AARON runs on a Silicon Graphics workstation that is networked with a generic 486 personal computer, which controls the painting machine. The machine is programmed in C++, a common computer language, which tells the machine what to sketch, what colors to mix and paint with, what size brush to use.
When Mr. Cohen wants his system to produce a work of art, he switches it on and leaves it to run, usually overnight -- he has better things to do than watch it churn away, he says. In the morning, he decides whether to run the painting machine or simply to view AARON's creation on a computer screen. He commits images to paper only sparingly. This saves storage space in his studio, which is large but crowded and, with machinery all about, has something of the air of a physics lab.
He determined when he began developing the program that its work would have to be compelling as art, not as a computing curiosity. Early on, in his planning, he studied the way children take their first steps in art -- with scribbles and then lines that enclose spaces to create pictures. AARON's early images resembled prehistoric rock drawings. Later he added more-sophisticated concepts. To create a sense of movement, for instance, he allowed AARON to compute a figure's center of gravity so as to place it in front of the visible figure.
Mr. Cohen would like the program"to have the same kind of status as the African sculpture that people like Picasso discovered at the beginning of the century -- that changed the course of European art history," he says. He aspires to create images that, like those sculptures, are alien to Western art yet"clearly done by intelligent entities capable of working thoughtfully, consistently." He does not mind, he says, if AARON neglects some of the standard features of European art of the past 500 years.
He professes to be little interested in arguments about whether AARON can"think" -- about what"artificial intelligence" means. But in an essay in the on-line Stanford Electronic Humanities Review in 1995, he wrote that AARON "constitutes an existence proof of the power of machines to do some of the things we had assumed required thought, and which we still suppose would require thought, and creativity, and self-awareness, of a human being."
What's most important to him is not that AARON possess smarts or a soul, but that it teach him things."I've always used a computer primarily as a way of finding out things," he says. "The problem with being a very skillful artist, whether one is talking about being a colorist or whatever, is that when something comes up and you don't quite know where you are -- you don't know how to handle it -- the pressure to fall back on what you know is enormous."
A measure of Mr. Cohen's success is the response of viewers of AARON's output. It's not just that the paintings have been selling very well -- for as much as $25,000 each, in fact. After his show at the Computer Museum in Boston in 1995, where AARON generated paintings as viewers watched, journalists showed up from around the world."It was unbelievable," he says."I'd been working in the public eye for decades and never seen anything like it."
It is also telling, however, that"not one word was published in the official art literature." He says he has barely minded; it was disdain for"the legitimate art world" that got him and AARON started."The official art world is really rather backward, it's really rather reactionary," he says."Art critics will never let on how stupid they are, and consequently they stay away from me, because the first thing they'd have to admit is they don't know what the hell I'm doing, technologically."
The irony, he notes with relish, is that he quite contentedly considers his work to be"very old-fashioned, by contemporary standards. All the hoo-hah about the World-Wide Web, and I'm still making images on paper!"
Still, it's work that would not exist without his chance encounter with computers.
"Most good things come by accident," Mr. Cohen muses.
And if he hadn't happened to visit San Diego in 1968? He ponders, then chuckles."I'd probably have just gone on being a disgruntled famous English painter."
Copyright © 1997 by The Chronicle of Higher Education
