The First-Ever AI Image: History and Origin

The Genesis of Synthetic Art: The Story of AARON and the First AI Image

Long before millions of people were typing text prompts into browsers to summon hyper-realistic portraits and surreal landscapes, the world of artificial intelligence art had a very different, humbler origin. There was no cloud-computing cluster, no massive neural network trained on billions of scraped photographs, and certainly no instantaneous generation.

Instead, the first true artificial intelligence image was born in the late 1960s and early 1970s inside the mind of a painter who wanted to crack the code of human creativity itself.

Meet Harold Cohen and AARON

In 1968, British-born artist Harold Cohen was a successful, internationally recognized painter representing Great Britain at prestigious global showcases like the Venice Biennale. Yet, at the height of his traditional career, Cohen grew deeply fascinated by an abstract question: What is the bare minimum required for an image to be recognized as an image?

To answer this, Cohen uprooted his life, moved to the United States, and taught himself computer programming. By the early 1970s—while working at the University of California, San Diego and later during a residency at Stanford University’s Artificial Intelligence Laboratory—he birthed a computer program that would change art history: AARON.

AARON was not an image generator in the modern sense. It didn’t “learn” from a dataset of historical paintings. Rather, Cohen built AARON using Symbolic AI (Good Old-Fashioned AI or GOFAI). He painstakingly, line by line, manually coded the rules of art into the system.

A Note on Early Logic: AARON wasn’t given pictures to copy. Instead, Cohen taught the program how the physical world worked conceptually: how shapes relate to one another, how objects overlap, how to balance visual proportions, and how to govern a composition.

The First Image Takes Shape

By 1973, AARON was capable of producing its first autonomous black-and-white line drawings.

When the system ran, it made independent decisions. Cohen didn’t draw the lines or dictate precisely where a shape would go; he provided the cognitive framework, and the machine executed the code to decide the composition on the fly. Attached to primitive drawing machines—and later, robotic arms—AARON physically held a pen or brush to paper, plotting out striking, organic abstract shapes and primitive figures.

Exhibited at locations like the Los Angeles County Museum of Art and later the Tate Modern and the Whitney Museum of American Art, AARON’s early outputs stunned audiences. This wasn’t just computer-plotted geometric math art (which other programmers were experimenting with at the time); these were fluid, rule-based interpretations of organic forms created by a machine making its own compositional choices.

From Coded Rules to Neural Networks

AARON evolved over decades. In the 1980s and 1990s, Cohen updated the program to understand color theory, and eventually, AARON began painting complex scenes featuring plants and human figures. Cohen passed away in 2016, leaving behind a legacy as the creator of the world’s first sustained AI art collaborator.

Comparing AARON to modern AI tools highlights a massive philosophical shift:

• Then (AARON): Built on human-authored rules of art. The AI acted autonomously based on logic given to it by a single artist.
• Now (Diffusion Models): Built on massive statistical data—absorbing billions of images to predict pixels based on human text prompts.

The Legacy of the First Pixel

The first AI-generated images remind us that artificial intelligence didn’t emerge overnight out of corporate tech labs. It started as a philosophical experiment by an artist who wanted to see if a machine could share the burden of creativity.