Before the Prompts: The True Story of the First AI-Generated Image in 1973

When we think of AI-generated images today, our minds instantly jump to modern conveniences: typing a prompt into a browser, waiting a few seconds, and receiving a hyper-realistic, photorealistic masterpiece generated by massive neural networks.

Yet, the lineage of artificial intelligence art does not begin in Silicon Valley tech labs of the 2020s. It stretches back over half a century to 1973, driven not by computer scientists seeking commercial disruption, but by an acclaimed British painter who wanted to answer a radical philosophical question: What is the absolute minimum requirement for an image to be recognized as art?

The Artist Who Wanted to Teach a Machine to Dream

In the late 1960s, Harold Cohen was at the peak of his career as a traditional painter. He had represented Great Britain at prestigious international exhibitions like the Venice Biennale and taught at the University of California, San Diego. But Cohen became deeply fascinated by computing.

Unlike many technologists experimenting with computer graphics at the time—who used machines to plot mathematical geometries or geometric patterns—Cohen was an artist. He did not want to use a computer merely as an electronic paintbrush or a fancy calculator. Instead, he wanted to externalize the cognitive process of drawing.

In 1973, while spending time at the Stanford Artificial Intelligence Laboratory, Cohen began writing the foundational code for a software program that would change art history: AARON.

What Did the First AI Images Look Like?

If you picture an AI image, you expect pixels, color gradients, and complex textures. The earliest images generated by AARON in 1973 looked entirely different:

• The Medium: Simple, stark black-and-white line drawings.
• The Hardware: A room-sized mainframe computer connected by cables to a mechanical drawing machine—often a flatbed plotter or a tethered “turtle” robot that crawled across a massive sheet of paper spread on the floor, holding an ink pen.
• The Style: Abstract, organic shapes, primitive human figures, rocks, and raw foliage.

Crucially, AARON did not copy pre-existing art. Modern generative AI is trained on billions of images scraped from the internet, looking for statistical patterns to synthesize pixels. AARON had no internet, no dataset of millions of photographs, and no neural networks.

Instead, Cohen manually coded a complex rulebook of human cognition and artistic composition. He taught the program how to construct a basic figure, how lines intersect, how to overlap objects so one appeared in front of another, and how to balance visual weight across a canvas.

Because Cohen embedded elements of randomization within these strict logical boundaries, every single image AARON produced was entirely unique. The machine was making autonomous creative choices within the strict framework set by its human creator.

From Crawling Turtles to Global Galleries

As computer hardware evolved over the decades, so did AARON:

• The 1980s and 90s: Cohen eventually taught AARON how to handle color. The software would generate complex color compositions, though Cohen would manually mix and apply the physical paints according to the program’s digital instructions.
• The 2000s: Later iterations evolved to generate lush, intricate botanical and jungle scenes, sometimes projected onto massive gallery walls.

AARON’s work grew so respected that it was exhibited in major cultural institutions around the world, including the Tate Modern in London and the Whitney Museum of American Art in New York.

Why 1973 Matters Today

Harold Cohen spent over forty years—until his death in 2016—refining AARON. He never viewed the computer as a tool meant to replace the artist; rather, he viewed AARON as a separate entity, an electronic collaborator sharing a unique creative partnership. He famously referred to the program as his “doppelgänger”.

The next time you prompt an AI to paint a masterpiece, it is worth remembering that the lineage of machine creativity does not start with modern tech giants. It traces back to a mechanical turtle crawling on a Stanford floor in 1973, driven by a painter who simply wanted to see if code could learn to draw.