The Ghost in the Machine: The Untold Story of the World’s First AI-Generated Image

If you type a prompt into an image generator today, a hyper-realistic, breathtaking piece of art materializes on your screen in a matter of seconds. It feels like modern magic—a marvel born from the era of deep learning, vast datasets, and neural networks.

However, the lineage of artificial intelligence art goes back much further than you might think. Long before the public ever heard of diffusion models, Generative Adversarial Networks (GANs), or Midjourney, a lone artist and programmer was quietly building a digital entity that would create the world’s first AI-generated images.

Meet AARON, the pioneering software that kicked off a revolution in the 1970s.

Meet Harold Cohen and His Digital “Doppelgänger”

The story of the first AI-generated image begins with Harold Cohen, a successful British-born artist who represented Great Britain at the prestigious Venice Biennale in 1966. At the height of his traditional career, Cohen experienced a radical career pivot. Fascinated by the burgeoning field of computing, he moved to the United States and began asking a radical question: What makes a drawing a drawing?

Cohen didn’t want to use a computer merely as a digital paintbrush. Instead, he wanted to see if a machine could understand the cognitive processes behind art-making.

In the late 1960s and early 1970s at the University of California, San Diego, Cohen began writing a computer program that would eventually be named AARON. By 1973 (and publicly showcased in 1974), AARON successfully generated its earliest autonomous compositions, cementing its place in history as the world’s first AI art system.

How Did AARON Create Art Without Neural Networks?

Today’s generative AI learns by analyzing billions of images scraped from the internet. AARON, by contrast, had never seen a photograph or a painting in its life.

Instead, AARON was a rule-based expert system. Because deep learning did not exist yet, Cohen had to manually translate his own understanding of human cognition and visual art into code.

• Hand-Coded Rules: Cohen didn’t teach AARON what a specific object looked like by showing it examples; he taught it how to think about space, form, and composition. He programmed rules for how objects occupy three-dimensional space, how lines overlap, and how to balance proportions.
• Autonomous Decision-Making: Once the parameters were set, AARON made independent choices. It decided how compositions would be arranged, where figures would stand, and how lines would intersect, meaning no two outputs were identical.
• The Physical Output: Because computer screens of the era were primitive, AARON didn’t just live in a digital vacuum. It was connected to a mechanical drawing machine—a pen plotter—that physically moved across paper with ink pens to render its creations line by line.

The Evolution of the First AI Artist

AARON did not stay static; it grew alongside computer science over several decades:

• Phase 1: Abstract Line Drawings (1970s): Early versions of AARON produced minimalist, abstract black-and-white geometric shapes and lines.
• Phase 2: Figures and Flora (1980s–2000s): As computer processing power grew, Cohen updated AARON to understand human anatomy, rocks, and eventually complex plant life. Later iterations, like Aaron KCAT (2001) and Gijon (2007), could paint vibrant, jungle-like scenes.
• Phase 3: Painting in Color: Cohen eventually taught AARON how to mix colors, though Cohen himself would sometimes physically paint the shapes generated by the machine based on its color-coding instructions.

“AARON represents one of the longest-running contemporary art projects in history,” operating as Cohen’s artistic partner and intellectual doppelgänger until Cohen’s passing in 2016.

Why AARON Still Matters Today

It is easy to look at AARON’s primitive 1970s line drawings and compare them to the hyper-realistic digital paintings generated by modern models. However, AARON’s historical importance cannot be overstated.

Modern AI often acts as a black box—scraping data, weighting probabilities, and churning out outputs based on statistical correlations. AARON was fundamentally different: it was an intellectual exercise in explicit logic. It forced humanity to confront the mechanics of creativity long before algorithms dominated our daily lives.

When you look at the raw, mechanical ink lines plotted by Harold Cohen’s computer program half a century ago, you are looking at the true dawn of machine creativity—the moment a machine first picked up a digital pen and began to draw.