Early AI Pictures: The Evolution of Digital Art

Pixels of the Past: The Fascinating History of Early AI Pictures

Long before type-in prompts could conjure hyper-realistic portraits in seconds, the intersection of computers and art was an experimental frontier. Today, text-to-image generators dominate our feeds, but the story of “early AI pictures” stretches back decades. It is a tale defined by room-sized mainframes, mechanical plotters, and pioneering artists who taught machines how to draw.

The Mechanical Pioneer: AARON and the 1970s

When we think of AI art, we picture software. But in the late 1960s and 1970s, one of the earliest and most significant AI art systems was physical. Developed by British artist and professor Harold Cohen at the University of California, San Diego, AARON was a computer program designed to create original artwork.

AARON didn’t use deep learning or neural networks—those didn’t exist yet in the modern sense. Instead, it used a symbolic rule-based approach. Cohen meticulously coded human rules of drawing, perspective, and composition into the machine. AARON would then output commands to a robotic drawing machine (a mechanical arm holding a pen over paper) to bring its creations to life. Watching a machine autonomously sketch botanical shapes or abstract figures in a gallery felt like stepping into science fiction.

The Psychedelic Era: Google’s DeepDream

Fast-forward to 2015, and the world of computer-generated visuals took a wildly surreal turn with the introduction of DeepDream.

Originally built by Google to help computer scientists understand how artificial neural networks process images, DeepDream used a technique called “inceptionism”. Programmers fed an image into a neural network and asked it to enhance whatever patterns it recognized.

The results were unforgettable. Feed DeepDream a picture of a landscape, and the AI—looking for features it had been trained on—would obsessively amplify dog faces, hypnotic eyes, and bizarre architectural swirls into every cloud and tree. DeepDream didn’t generate pictures from scratch; instead, it acted like an over-caffeinated dreamer hallucinating shapes onto reality. It gave the internet its first massive taste of native machine aesthetics, sparking a viral wave of psychedelic digital art.

The GAN Revolution: Faces That Never Were

The late 2010s introduced a technological leap that brought us closer to modern generative imagery: GANs (Generative Adversarial Networks).

Invented by Ian Goodfellow and his colleagues in 2014, a GAN pits two neural networks against each other in a game of digital cat-and-mouse:

• The Generator tries to create fake images (like human faces or cats).
• The Discriminator acts as a critic, trying to spot the difference between real photos and the generator’s fakes.

Through this constant competition, the generator gets terrifyingly good at its job. This era gave birth to websites like This Person Does Not Exist, which flashed hyper-realistic portraits of humans who were entirely synthetic—pixels calculated by math, representing people who had never lived.

Why Early AI Pictures Matter Today

Looking back at early computer drawings, algorithmic rule-followers, and hallucinogenic neural nets reveals a profound truth: AI art is not an overnight phenomenon.