The Ancestors of Modern Pixels: A Look Back at Early AI Pictures

Today, we take text-to-image wonders for granted. With a quick typed sentence, anyone can conjure hyper-realistic photographs, whimsical oil paintings, or surrealist landscapes in mere seconds. But long before modern diffusion models and neural networks dominated headlines, a quiet, experimental revolution was taking place.

The story of early artificial intelligence pictures is not one of instant digital gratification. Instead, it is a fascinating tale of mainframe computers, plotting machines, robotic arms, and decades of visionary artists who tried to teach machines what it means to draw.

1. The Pioneer in the Lab: Harold Cohen and AARON

When we look for the true genesis of AI art, the trail leads us back to the late 1960s and early 1970s. Enter Harold Cohen, a renowned British painter who represented his country at the Venice Biennale before packing up his bags to explore computing at the University of California, San Diego.

Cohen asked a radical question: What if, instead of using a computer as a passive tool, we could program the computer to actually make autonomous creative decisions?

From that question emerged AARON, widely recognized as the world’s first AI art system.

• How it worked: AARON was not a neural network in today’s sense; it was a symbol-based, rule-driven program rooted in Good Old-Fashioned AI (GOFAI). Cohen didn’t feed it thousands of photos to copy. Instead, he codified his own deep understanding of human anatomy, perspective, and drawing mechanics into thousands of lines of code.
• The Execution: AARON would output code that drove physical, custom-built turtle-like drawing machines or robotic arms holding pens and brushes. Watching AARON work in galleries—such as the Los Angeles County Museum of Art in 1972—was mesmerizing. Audiences watched a machine autonomously map out botanical shapes and human figures on paper, stroke by calculated stroke, proving that a computer could execute acts of artistic creation.

2. The Era of Artificial Evolution and Genetic Code

As computing power accelerated through the 1980s and 1990s, artists and researchers began experimenting with evolutionary algorithms—borrowing mechanisms from natural selection to “breed” images.

• Karl Sims and Artificial Life: In the early 1990s, computer graphics artist Karl Sims created installations like Galápagos. Viewers could walk up to screens displaying various abstract, mutating 3D forms. By voting on their favorites, humans acted as the environment, selectively “breeding” generations of digital organisms. The computer mutated the underlying mathematical code of the winners, resulting in breathtaking, otherworldly animations and visual patterns that evolved in real-time.
• Scott Draves and Electric Sheep: Launched in 1999, this decentralized screensaver project utilized distributed computing to evolve “fractal flames”. Millions of participating computers pooled their processing power to render and evolve psychedelic, endlessly morphing mathematical art, guided by user voting and automated genetic mutation.

3. DeepDream: When Machines Start to “See”

If AARON represents the thoughtful, deliberate childhood of AI art, DeepDream (released by Google in 2015) was its chaotic, psychedelic teenage rebellion.

DeepDream introduced the public to the capabilities of Convolutional Neural Networks (CNNs). Originally trained to classify objects in photographs (like spotting a dog, a car, or a coffee mug), researchers decided to run the network in reverse.

[Input Image] ---> [Neural Network Classifier] ---> (Amplified Feature Recognition) ---> [Surreal, Dreamlike Output]

When fed a random photo and told to “enhance whatever you see here,” the network hyper-fixated on familiar patterns. A cloudy sky or a patch of grass would suddenly bloom into terrifyingly intricate collages of dog eyes, swirling fur, and bizarre multi-headed animals. While impractical for traditional portraiture, DeepDream viralized the concept that artificial intelligence didn’t just calculate pixels—it possessed a bizarre, algorithmic “imagination” of its own.

4. The Bridge to Modern Marvels: Style Transfer and GANs

Right on DeepDream’s heels came the mid-2010s breakthroughs that paved the way for today’s landscape:

• Neural Style Transfer (2015): Algorithms suddenly learned how to mathematically separate the content of one image from the style of another. You could feed a photograph of your cat and a digital copy of Van Gogh’s Starry Night, and the AI would meticulously redraw your cat as if Van Gogh himself had painted it.
• Generative Adversarial Networks (GANs, 2014–2018): Pioneered by Ian Goodfellow, GANs pitted two neural networks against each other: a generator trying to fake realistic images, and a discriminator acting as an art critic trying to spot the fakes. Through this high-stakes digital game of cat-and-mouse, GANs rapidly learned how to generate hyper-realistic human faces of people who never existed, fashion items, and abstract art pieces.

Looking Back to Move Forward

The early days of AI pictures remind us that technology doesn’t emerge in a vacuum. Long before the era of multi-billion-parameter text prompts, artists, programmers, and mathematicians were already wrestling with the philosophical and aesthetic boundaries of machine creativity.

Those clunky mainframe printouts, robotic ink-pen drawings, and glitchy neural hallucinations laid the crucial groundwork for the visual world we navigate today. They prove that our fascination with co-creating art alongside machines is not a modern fad—it is a half-century-old human dream finally coming into focus.

What are your thoughts on the evolution of machine art? Did you ever cross paths with early algorithmic screensavers or digital art experiments back in the day?