Harold Cohen: The Visionary Who Taught Machines to Draw

In the spare enclave of a university computer lab in the early 1970s, a British painter with tousled hair and an impish smile watched as a rudimentary mechanical arm dragged a pen across paper, sketching abstract forms that seemed to have intention behind them. This was not merely a machine reproducing programmed instructions it was creating something new, guided by algorithmic logic but generating compositions that its human creator couldn't precisely predict. That creator was Harold Cohen, and the machine was AARON, a pioneering artistic intelligence that would become his life's defining work.

Today, as our social media feeds flood with AI-generated imagery and debates rage about machine creativity, we find ourselves standing on ground that Cohen first mapped over fifty years ago. His journey from celebrated abstract painter to computational art pioneer offers a profound lens through which to understand our current moment of algorithmic creation.

The Radical Pivot

Harold Cohen's artistic trajectory represents one of the most fascinating reinventions in modern art history. By the mid-1960s, he had established himself as one of Britain's leading abstract painters. Trained as a painter, Cohen gained renown in the 1950s/60s for his bold abstract canvases, before making a radical mid-career shift into the realm of algorithms and artificial intelligence.

During the 1950s and 60s Cohen built a reputation as an abstract painter, creating bold, expressive canvases often categorized as English Abstract Expressionism. His paintings from this period feature vigorous, biomorphic forms and vibrant colors, reflecting the influence of contemporary abstraction and even the emerging Pop Art palette.

The pinnacle of his conventional success came when he was selected to represent Great Britain at the 23rd Venice Biennale in 1966, alongside his brother (the painter Bernard Cohen) and others. His work hung in the Tate, and according to Tate curator Michael Compton, "from about 1952 until 1968 Harold Cohen built up a reputation as a painter equal to that of any British artist of his generation."

Yet at this apex of recognition, Cohen grew restless. The questions that preoccupied him began to transcend the canvas. He later reflected on fundamental questions: "How do artists process their information in the creation of artworks? What are the minimum conditions under which a set of marks functions as an image? What makes an image evocative?" These questions hinted at a shift from the purely visual and emotional aspects of painting toward a more analytical, even scientific inquiry into art.

Silicon Valley Transformation

The year 1968 marked a turning point. Cohen accepted a one-year visiting professorship at the newly founded University of California, San Diego. Relocating to California placed Cohen in proximity to cutting-edge technology and research in art and computing. His one-year position extended into a permanent relocation and a complete redirection of his artistic focus.

In California, Cohen quickly sought out the computer as a new tool for exploring these ideas. At UC San Diego he met Jef Raskin, a graduate student (who would later become known for his work at Apple). Raskin introduced Cohen to the campus's CDC 3200 mainframe computer and taught him the basics of programming (starting with FORTRAN).

For Cohen, this technological encounter catalyzed an intellectual epiphany: He realized that if art-making is an information process, a computer might model it, since computers "are excellent at processing and do so purely based on defined input".

His early efforts were laborious. He began coding simple programs to generate drawings, initially via punch cards on batch-processing machines ff a slow, frustrating method. But Cohen persisted through these technical limitations, driven by conceptual questions about art's essence that had begun to overshadow his interest in making traditional paintings.

AARON: The First AI Artist

By the early 1970s, Cohen had created a program that would eventually be named AARON one of the most significant and longest-running experiments in computational creativity. Conceived in the late 1960s and named by the early 1970s, "AARON" was not actually an acronym but rather a reference with multiple layers of meaning.

Cohen has noted that he initially imagined developing a series of artmaking programs A, B, C, etc., in parallel with the naming of programming languages like B and C of that era. He chose "Aaron" as the first, and ultimately only, such program to pursue, a name which also alluded to the biblical figure Aaron the brother of Moses who acts as his spokesperson.

What made AARON revolutionary was not merely that it could draw, but that it could generate original compositions following rule systems that mimicked aspects of human artistic decision-making. In its initial incarnations (early-to-mid 1970s), AARON was designed to autonomously generate drawings at first in black-and-white linework. Cohen started simple: he programmed a small set of primitive shapes and rules, and the computer would recursively compose these into abstract compositions.

The physical manifestation of AARON's virtual creations evolved over time. The output was then realized physically using a plotter or a motorized turtle robot dragging a pen on paper. These early drawings had a distinctive aesthetic: AARON's early drawings often resembled the free, unsteady lines of children's drawings, an intentional choice influenced by Cohen's studies of child art and ancient petroglyphs as examples of "minimal" image-making.

Beyond Abstraction: Teaching AARON to See the World

As the decades passed, Cohen continuously refined AARON, pushing its capabilities into ever more sophisticated territory. As time went on, Cohen continually expanded AARON's capabilities, effectively training the program to handle more complex imagery. He viewed this as a decades-long artistic collaboration.

In the 1980s, he began incorporating figurative and representational elements into AARON's repertoire, moving beyond the purely abstract forms. Cohen incrementally added code for AARON to draw recognizable objects first rocks and plants, and eventually human figures by the mid-80s.

By the late 1980s, AARON could draw a domestic scene with potted plants, furniture, and people a limited world of symbols that Cohen felt represented a significant chunk of an artist's subject matter.

Color presented the next frontier. Cohen quipped in 1989, "It's taken me 20 years to teach AARON to draw. How can I possibly teach it to color before I die?" But he succeeded in this challenge too, first by hand-coloring AARON's drawings himself, and later by teaching the program to apply its own chromatic palette.

Up to the early 1990s, AARON's artworks were either colored by Cohen manually or were displayed as digital images on-screen. But Cohen wanted AARON to autonomously paint in full color.

This required significant technical advances: he transitioned AARON's software core from the C language to LISP, the AI community's favored language, because C was "too inflexible, too inexpressive, to deal with something as conceptually complex as color".

The culmination of these efforts debuted in 1995, when Cohen unveiled a new version of AARON a robotic painting system nearly 8 feet long that could not only draw but also mix and apply paint on a physical canvas or paper. This remarkable machine used a vacuum table to hold paper, a robotic arm with pens and brushes, and even a paint-mixing palette.

The Question of Creativity

Throughout his exploration of algorithmic art, Cohen grappled with profound questions about the nature of creativity and authorship. He concluded that AARON was not creative in the human sense, since it didn't originate its goals or improve itself autonomously, but he asserted that the process AARON used was analogous to an artist's process.

He likened his relationship with AARON to that of a master painter with a very skilled apprentice or assistant – he set the direction, but AARON executed much of the work, and together they produced the art.

This collaborative model anticipated today's debates about AI image generators like DALL-E, Midjourney, and Stable Diffusion. While these contemporary systems use fundamentally different technical approaches (neural networks rather than rule-based programming), they raise the same essential questions that Cohen explored for decades: What is the relationship between human and machine creativity? Who is the author of algorithmically-generated art? Can machines truly create?

Legacy and Relevance

Harold Cohen died in April 2016, having devoted almost fifty years to developing and refining AARON. Notably, AARON did not outlive its creator: it was a labor of love, written in bespoke code that only Cohen fully understood, so when he passed, AARON fell silent.

Yet his work remains profoundly relevant today. Historically, Cohen's AARON provides essential context for today's boom in AI-generated imagery. As curator Christiane Paul noted, AI-driven art is "not a new development, but has been around for more than 50 years." Cohen's exhibits and writings through the years continuously invited viewers and scholars to consider what it means for a machine to draw or paint.

Recent institutional recognition confirms Cohen's historical importance. In 2024, the Whitney Museum of American Art in New York presented "Harold Cohen: AARON," a major museum exhibition tracing the evolution of AARON and placing Cohen's work in dialogue with contemporary AI image generators. The Whitney also acquired AARON's digital software artworks for its collection, a significant institutional recognition of Cohen's art-historical importance.

In our current era of generative AI, when algorithms create images, text, and music with unprecedented sophistication, Cohen's pioneering work offers valuable perspective. He demonstrated that the most interesting questions aren't simply about what machines can make, but about the nature of creativity itself. His five-decade journey with AARON reminds us that technology is not merely a tool but can be a collaborator in the creative process if we approach it with the right questions and human intentionality.

Harold Cohen crossed the divide between traditional art and computational creation not as a technologist seeking to make art, but as an artist seeking to understand creativity through technology. In that crucial distinction lies the enduring power of his legacy, and a model for how we might navigate our own algorithmic future with both critical insight and creative possibility.