When we think about artificial intelligence generating images, our minds naturally jump to modern neural networks, massive diffusion models, and instant prompt-to-picture generators. But long before GPUs existed or deep learning became a household term, the blueprint for machine-made art was being sketched out in laboratories.
The true genesis of AI-generated imagery dates back to 1973, born from a unique collision between high-level computer science and traditional fine art.
The Painter Who Turned to Code: Harold Cohen
To understand the first AI-generated image, you have to look at its creator. Harold Cohen was not a computer scientist; he was an acclaimed British painter who had already achieved major international success, representing Britain at the prestigious Venice Biennale.
By the late 1960s and early 1970s, Cohen found himself captivated by a radical question: What is the minimum set of cognitive rules required to make a drawing?
At the time, artificial intelligence researchers were focused on logic puzzles, theorem-proving, and chess. The prevailing assumption inside and outside of computer science was that creative work was inherently human—steeped in emotion, intuition, and a mystical spark that could never be reduced to a formula. Cohen set out to challenge that assumption. He relocated to the United States—working first at Stanford’s Artificial Intelligence Laboratory and later at the University of California, San Diego—to teach a machine how to draw.
Enter AARON: The World’s First Autonomous Artist
In 1973, Cohen’s efforts culminated in a software program that he named AARON.
Unlike modern AI systems that use statistical machine learning to parse billions of scraped internet images, AARON operated entirely on symbolic reasoning and explicit rule-based programming. Cohen meticulously codified his own understanding of physical mark-making into the computer:
- How lines relate to one another.
- What constitutes an enclosed shape or boundary.
- How overlapping objects create a sense of space and composition.
Once Cohen hit “run,” he completely stepped away. AARON generated its compositions autonomously. The computer made its own real-time choices about what lines to draw, where to place them, and when a composition was finished.
Because computer screens of the era were primitive, AARON’s digital decisions were translated into physical reality using a flatbed plotter—a mechanical drawing machine holding a pen that glided across sheets of paper under the software’s direct instruction.
What Did the First AI Images Look Like?
The earliest outputs produced by AARON in 1973 were not hyper-realistic photographs or complex digital paintings. Instead, they were striking, abstract compositions of intricate lines, geometric shapes, and closed forms.
| Feature | Early AARON (1973) | Modern AI Generators (Present) |
| :— | :— | :— |
| Underlying Tech | Rule-based symbolic logic | Deep neural networks & diffusion models |
| Training Method | Manually coded cognitive rules of drawing | Billions of parameters trained on scraped data |
| Output Medium | Physical ink on paper via mechanical plotters | High-resolution digital pixel grids |
| Autonomy | Fully autonomous decision-making | Prompt-driven reactive generation |
While the 1973 drawings were abstract, they possessed a deliberate, organic rhythm that shocked viewers. They did not look like random computer glitches or mathematical equations; they looked like structured, intentional art.
A Forty-Year Partnership
Most technological proofs-of-concept from the 1970s were short-lived, running for a brief demonstration before being archived. But AARON was different. Cohen spent the next four decades continuously updating, expanding, and refining the program.
As computing power evolved, so did AARON. Over the years, the program graduated from abstract black-and-white line drawings to handling complex rules of perspective, shading, and eventually vibrant color. In later decades, AARON even learned to generate figurative elements, producing distinct shapes of people, plants, and rocks.
When AARON’s art was eventually exhibited at major institutions like the Tate Modern and the Whitney Museum of American Art, it forced the traditional art world to confront an uncomfortable question: If a machine can independently conceive and execute a compelling work of art, where does creativity truly reside?
Legacy in the Age of Midjourney and DALL-E
Today, we live in an era where anyone can type a text prompt and conjure a breathtaking image in seconds. Yet, the foundational philosophical and technical debates we have today—about authorship, machine creativity, and the nature of art—were already being mapped out in 1973.
Harold Cohen and his creation, AARON, proved that the boundary between human cognition and machine logic is more porous than we might think, leaving a legacy that echoes through every pixel of modern AI art.
Leave a Reply