The Fascinating History of the First AI-Generated Image

The Ghost in the Machine: The Fascinating True Story of the First AI-Generated Image

Long before neural networks could conjure hyper-realistic portraits from a simple text prompt, a quiet revolution took place in a university computer lab. Today, we live in an era where anyone can type a few descriptive words and watch an artificial intelligence render breathtaking art in seconds. But the lineage of machine-made art didn’t begin with billions of scraped parameters or massive server clusters.

It began in 1973 with a British painter, a primitive computer program, and a profound question: Can a machine learn to see, think, and create like an artist?

Meet Harold Cohen and AARON

In the late 1960s, Harold Cohen was a celebrated, establishment artist. He had represented Great Britain at the prestigious Venice Biennale and enjoyed a thriving career in the traditional art world. Yet, Cohen found himself consumed by an intellectual itch. He became fascinated by the emerging world of computing and wanted to understand the cognitive mechanics behind how humans make art.

Rather than treating the computer as a passive tool—like a digital paintbrush—Cohen wanted to see if he could codify the actual creative process.

Moving to the United States (working first at Stanford’s Artificial Intelligence Laboratory and later at the University of California, San Diego), Cohen embarked on an audacious project. He began writing a software program that would eventually be known as AARON.

How the First AI Image Was Made

If you compare AARON to modern generative tools, the differences are staggering. Modern AI models rely heavily on deep learning, consuming massive datasets of billions of human-made images to recognize statistical patterns.

AARON had none of that. There was no internet, no massive image database, and no neural network. Instead, Cohen approached AARON like a human student:

• Rule-Based Logic: Cohen manually wrote symbolic rules of art into the code. He taught the program geometry, the principles of perspective, how lines relate to one another, and how shapes overlap.
• Autonomous Agency: Crucially, Cohen did not dictate every stroke. Once the program ran, AARON made its own independent compositional choices. The software decided where lines would go, how shapes would close, and how elements would balance on the page in real-time.
• Physical Embodiment: Because computer monitors of the era could not display rich, high-resolution artistic strokes, AARON was connected to a specialized drawing machine (a plotter)—a mechanical arm holding a pen that physically moved across paper based on the computer’s live instructions.

When AARON produced its first coherent, abstract black-and-white compositions in 1973, it achieved something unprecedented: a machine was generating original visual art entirely through autonomous decision-making.

From Abstract Lines to Living Worlds

The earliest images created by AARON were geometric and abstract—focused on structural mark-making rather than recognizable objects. But Cohen’s project didn’t stop in 1973.

Over the next four decades, Cohen continuously updated AARON, teaching it new tricks. As computing power evolved, so did the software:

• By the 1980s and 1990s, AARON learned to handle color, filling in its own shapes.
• Later iterations transitioned from pure abstraction into figurative work, generating complex human figures, stones, and rich, jungle-like foliage.

AARON’s work eventually transcended the computer lab, earning a permanent place in world-class institutions like the Tate Modern and the Whitney Museum of American Art.

Why the First AI Image Still Matters

Looking back at AARON’s primitive 1970s line drawings today, they might appear modest compared to the photo-realistic digital paintings produced by modern algorithms. Yet, their historical weight is immeasurable.