What Was the First AI-Generated Image Ever Made?

The Pixel Before Time: What Was the First AI-Generated Image?

Long before typing a few words into a prompt box could conjure up photorealistic landscapes or surreal oil paintings in seconds, the concept of machine-made art sounded like science fiction. Today, we take text-to-image generators for granted. But the lineage of artificial intelligence art didn’t begin with neural networks or massive server clusters scraping the internet.

Instead, it started in a laboratory in the early 1970s with a British painter, a primitive computer, and a program named AARON.

The Birth of AARON (1973)

When art historians and computer scientists look for the definitive “first” AI-generated image, they point to the work of Harold Cohen.

In the late 1960s, Cohen was a successful, internationally acclaimed painter who represented Great Britain at the prestigious Venice Biennale. Yet, he grew deeply fascinated by the cognitive mechanics of art—specifically, what makes a human recognize a collection of lines as a figure or an object. This curiosity led him away from his canvases and into the world of computing, eventually landing him at the Stanford Artificial Intelligence Laboratory in 1973.

There, Cohen wrote a symbolic, rule-based computer program that would change art history: AARON.

How the First “AI” Made Art

Unlike modern generative models, AARON did not use machine learning, neural networks, or billions of scraped reference images. It possessed no concept of a “dataset.”

Instead, AARON was built entirely on human-coded logic and cognitive rules. Cohen meticulously translated his own understanding of the physical world into code:

• How objects overlap.
• How to distinguish a foreground from a background.
• The geometry of closed shapes and abstract lines.

AARON would take these programmed rules and make autonomous decisions on composition. It was connected to a physical drawing machine—a robotic arm holding a pen over paper, known as a plotter—which physically executed the program’s real-time decisions. The resulting minimalist, abstract line drawings were entirely original; no human hand explicitly mapped out where those specific lines would fall.

Evolution Beyond the First Stroke

AARON represents the dawn of algorithmic art, but the definition of an “AI-generated image” evolved dramatically over the decades through three major technological milestones:

1. The Symbolic Era (1970s–1980s): Programs like AARON operated strictly on hard-coded human rules, gradually expanding from basic black-and-white lines to complex geometric shapes, plants, and human figures over decades of updates.
2. The Deep Learning Era & DeepDream (2015): Google introduced DeepDream, using convolutional neural networks to find and amplify patterns in existing photos. This produced psychedelic, surreal, dreamlike imagery that proved neural networks could “see” creatively.
3. The GAN & Diffusion Revolutions (2018–Present): Generative Adversarial Networks (GANs) brought about milestones like Portrait of Edmond de Belamy in 2018, quickly followed by modern diffusion models (like DALL-E, Midjourney, and Stable Diffusion) that translate raw text prompts into hyper-realistic masterpieces.

Why the First AI Image Still Matters

Looking back at AARON’s primitive line drawings next to today’s hyper-detailed digital creations highlights a profound shift. AARON was essentially a digital apprentice executing its master’s philosophy. Modern AI, by contrast, acts as a statistical mirror of humanity’s collective visual culture.

Yet, Harold Cohen’s experiment established a vital philosophical precedent: that a machine could participate in the act of creation. When AARON’s robotic arm put pen to paper in 1973, it crossed a threshold, proving that art does not exclusively require a heartbeat—it requires a system capable of choice.