Can a machine be truly creative? Let’s ask AARON

Harold Cohen didn’t fit the mold of an AI pioneer. He wasn’t a computer scientist. He wasn’t a programmer. He was an artist – brilliant, world-renowned, and celebrated for his explosive use of color and bold compositions. In 1966, he represented Britain at the Venice Biennale, one of the most prestigious art events in the world. Cohen had arrived. He was at the top of his field.

But within two years, something shifted. In 1968, Cohen found himself on the other side of the world, at the University of California, San Diego – the very campus where, decades later, I sat in the audience, enthralled by his passion. It was there, almost by accident, that he stumbled into computer programming.

By 1971, at Stanford University’s Artificial Intelligence Laboratory, Cohen began building AARON, a machine that could make art. But not just any art – art that could stand on its own, that could hang in galleries. This project wasn’t just about machines; it was about Cohen himself. In an audacious act of self-exploration, he set out to crack open his own mind and decode the mystery of his creative process.

Capturing the depth of Harold Cohen’s creativity in lines of code was no simple task. It became a decades-long obsession. AARON’s earliest works in the 1970s were humble – simple, monochrome line drawings that felt like a child’s first tentative sketches. But by the 1980s, AARON had matured. It began drawing intricate forms – human figures, plants, everyday objects – capturing the natural world with surprising sensitivity.

For years, AARON acted as Cohen’s silent collaborator. The machine laid down the structure; Cohen brought it to life with his vibrant colors. Color was the soul of his work, and he doubted AARON would ever master it. “It’s taken me 20 years to teach AARON to draw. How can I possibly teach it to color before I die?” he mused.

But by the mid-1990s, Cohen had done the seemingly impossible. Using a robotic arm and a sophisticated color-mixing system, the machine could now create vibrant, full-color images without any human touch.

This breakthrough took center stage in 1995 at the Computer Museum in Boston. The exhibit, The Robotic Artist: AARON In Living Color, showcased AARON’s newfound abilities. Viewers watched as the machine, brush in hand, brought vivid canvases to life. It wasn’t just a technical milestone – it was proof that a creative machine could step out from under its creator’s shadow and reach much greater creative independence.

Despite Cohen’s groundbreaking achievements, many remained skeptical, questioning whether AARON’s creations could truly be called art. Computer scientist Larry Cuba famously dismissed AARON as merely a “Harold Cohen simulator”.

But Cohen didn’t shy away from the critique. Instead, he acknowledged its validity. AARON was, after all, an extension of himself – a system designed to echo his creative process.

While some sought to minimize Cohen’s accomplishments, others were awestruck. In a presidential address at the American Association for Artificial Intelligence, Buchanan claimed AARON “is a much more talented and creative artist than most of us would claim to be.” But Cohen wasn’t swayed by the praise. He knew exactly where AARON’s boundaries lay… Cohen refused to exaggerate AARON's abilities. In a world driven by self-promotion and hype, his honesty still stands out. He built AARON during the age of 'Expert Systems' – machines that followed explicit, hand-coded instructions. There was no independent thought, no learning. AARON could only do what Cohen told it to.

But things have changed. Today's AI systems don't rely on hand-coded rules. They learn from vast data sets, consuming more information than any human ever could. Modern creative machines routinely produce work that surpasses the capabilities of their makers, forcing us to grapple with deeper challenges around the creative capabilities of machines.

And yet, even as generative AI tools far outpace the pioneering efforts of expert systems, there's still much to learn from Cohen. His unwavering commitment to integrity over acclaim offers a lesson today's AI developers would do well to remember.

Excerpted with permission from the publisher, Wiley, from Creative Machines: AI, Art & Us by Maya Ackerman. Copyright © 2025 by John Wiley & Sons, Inc. All rights reserved.