Harold Cohen: The Visionary Artist Who Taught Machines to Create Art
Harold Cohen was a British-born artist, educator, and artificial intelligence pioneer whose career traced an unlikely arc from abstract painting to building one of the first computer programs capable of making original drawings. Best known as the creator of AARON, Cohen spent more than four decades asking a question that still drives debates about generative art: can a machine be creative, and if so, what does that creativity reveal about human minds?
For readers searching for Harold Cohen, the name most often points to the painter and AI-art innovator who died in 2016. His work sits at the crossroads of studio art, cognitive science, and software engineering, making him a foundational figure for anyone interested in how algorithms became collaborators rather than mere tools.
Harold Cohen and the Dawn of AI Art
Cohen emerged from the postwar London art world with a reputation for rigorous abstraction and a restless curiosity about how images are made. By the late 1960s, he had already achieved significant recognition, but he became increasingly dissatisfied with traditional explanations of artistic creation. That dissatisfaction led him away from the easel and toward the computer, a machine he initially viewed with skepticism.
His move to California in 1968 placed him near the emerging computing culture of the University of California, San Diego. There, Cohen learned to program and began to see code not simply as a tool for generating patterns but as a medium for modeling the very act of drawing. This shift transformed him from a painter who used computers into an artist who thought through software.
From London Painter to Algorithmic Pioneer
Born in London in 1928, Harold Cohen studied at the Slade School of Fine Art and quickly became part of the British art scene. His early paintings were abstract, often large and gestural, and in 1966 he represented Britain at the Venice Biennale. That success might have cemented a conventional career, but Cohen’s interests were already moving toward systems, rules, and cognition.
When he joined the faculty at UC San Diego, he encountered students and colleagues who were experimenting with computers in art. Rather than treating the machine as a novelty, Cohen began writing his own programs. He wanted to understand whether the rules behind drawing could be articulated precisely enough for a computer to follow them and, eventually, to extend them.
The Birth of AARON
In the early 1970s, Cohen began developing AARON, a program named with deliberate simplicity. AARON was not designed to copy photographs or execute a human artist’s preplanned sketch. Instead, it was given a set of cognitive and compositional rules that allowed it to make its own decisions about line, form, balance, and closure.
Early AARON drawings were abstract, consisting of closed shapes and wandering lines that nevertheless felt intentional. Cohen acted as programmer, critic, and sometimes colorist, but the program’s output was not fully predictable. This combination of authorship and autonomy made AARON one of the first serious examples of machine creativity in visual art.
How AARON Changed Computer Creativity
AARON’s significance lies in its longevity and ambition. Over decades, Cohen continually rewrote and expanded the program, moving it from simple abstract mark-making to increasingly complex representational imagery. AARON became a lifelong research project, one that blurred the boundaries between art, artificial intelligence, and philosophy of mind.
Cohen also engaged directly with the AI research community, including periods of dialogue with scientists at institutions such as MIT’s Artificial Intelligence Laboratory. He did not treat engineering as separate from aesthetics. For him, building AARON was a way to test theories about perception, representation, and the human drive to make images.
From Abstract Marks to Recognizable Forms
In its early years, AARON produced abstract drawings that resembled improvised compositions. Cohen did not teach it to imitate existing artworks. Instead, he gave it primitives: simple actions and relationships that could be combined in countless ways. The result was a body of work that felt both systematic and surprisingly organic.
By the 1980s and 1990s, AARON had advanced to drawing recognizable subjects, including plants, people, and still lifes. It could arrange objects in space, suggest depth, and generate variations without repeating itself. This evolution showed that machine creativity could move beyond geometry and into the messy, associative territory of visual representation.
Color, Painting Machines, and Public Exhibitions
Color presented a major challenge. Cohen had to translate color theory and compositional judgment into rules that AARON could apply. Eventually, AARON could choose colors and apply them to its own drawings, and Cohen built robotic painting machines that allowed the program to create physical works on paper and canvas.
These paintings and drawings were exhibited internationally, and Cohen’s work entered major public collections, including the Tate and the Victoria and Albert Museum. AARON became a landmark in generative art, influencing later artists who use AI, machine learning, and algorithmic systems. Its legacy is not just a set of images but a long argument about what it means for a machine to make something new.
Harold Cohen’s legacy continues to shape conversations about AI art, authorship, and creativity. His insistence that programming could be a form of artistic thinking opened doors that many contemporary generative artists now walk through. The next time you see an algorithm produce an image, it is worth remembering the painter who spent decades teaching a program to draw, not by copying human hands, but by modeling the rules behind them.