Algorithmic Art: From 1960s Plotters to Modern AI Paintings

The first computer artists: Vera Molnár, Frieder Nake (1960s)

In 1965, Frieder Nake exhibited computer-generated drawings in Stuttgart, Germany. The same year, Georg Nees showed similar work in Stuttgart, and the first gallery exhibitions of computer art appeared in New York. These were drawings produced by writing programs that controlled a mechanical plotter, a device that moves a pen across paper following instructions.

Vera Molnár, a Hungarian-French artist working in Paris, was among the most rigorous early practitioners. Before she had regular access to a computer, she ran what she called “machine imaginaire” experiments: manually following procedural rules she invented to understand what kinds of visual patterns rule-based drawing could produce. When she did get access to computing equipment, she was already thinking algorithmically.

Molnár's work explored geometric order with controlled disorder. She would write rules for drawing grids of shapes, then introduce incremental perturbations: rotate this square by a random small amount, shift this line by a random pixel. The result was visual order that felt handmade. Her question was: how much randomness disrupts a rule before it stops being a rule?

Rule-based art and the role of randomness

Early algorithmic art was explicitly rule-based. The artist defined a set of instructions, the computer executed them. The interesting discovery was that inserting a random element into deterministic rules produced work that felt visually alive in ways that pure determinism did not.

Harold Cohen developed a program called AARON starting in 1973. AARON drew figures and abstract shapes by following rules Cohen wrote about how shapes relate to each other and how figures occupy space. The rules were deterministic in structure but included random parameter selection within defined ranges. Each output was different; all outputs were recognizably from the same system.

This became the defining tension of algorithmic art: how much the artist specifies and how much the system decides. Total specification produces repetition. Total randomness produces noise. Interesting work lives in the space between, where the rules create a visual grammar and randomness generates the sentences.

Fractals and mathematical beauty

In the late 1970s and through the 1980s, a different kind of algorithmic art emerged from mathematics: fractals. Benoit Mandelbrot's work on fractal geometry, published in 1975, showed that simple iterative equations produced images of extraordinary visual complexity.

The Mandelbrot set is generated by a two-line iterative formula applied to complex numbers. The images it produces, with their infinitely detailed boundary and self-similar structure at every scale, became one of the visual symbols of the 1980s. The Julia sets, Sierpinski triangle, and other fractal structures followed.

Fractals demonstrated that algorithmic images could be genuinely beautiful in a way that did not depend on simulating human artistic taste. The visual quality came directly from the mathematics. The formula produced the aesthetics.

Generative art in the digital era

In the 1990s and 2000s, artists working in Processing, Max/MSP, and later JavaScript and WebGL expanded what algorithmic art could do. Casey Reas and Ben Fry developed Processing specifically as a tool for visual artists who wanted to program without a computer science background. The result was a generation of artists who built complex generative systems with interactive and time-based dimensions.

The generative art movement produced artists like Manfred Mohr, who explored algorithmic drawing throughout a multi-decade career, and the broader community around platforms like Hicetnunc and Art Blocks, which brought generative art to blockchain distribution in the early 2020s.

Art Blocks in particular demonstrated commercial appetite for algorithmically generated art. Projects by Tyler Hobbs, Dmitri Cherniak, and others sold for hundreds of thousands of dollars. The code that generated the art was stored on the blockchain; the specific output was determined by parameters embedded in the token.

How modern AI differs from earlier algorithmic art

Earlier algorithmic art was rule-explicit: the artist wrote the rules, the computer followed them. Modern AI image generation is rule-implicit: the model learned its rules from data rather than having them specified by a programmer. Nobody wrote a rule that said “Van Gogh style requires short directional strokes.” The model inferred that from examples.

This shifts the nature of artistic control. In traditional generative art, the artist understands the rules because they wrote them. In AI generation, the rules exist as learned parameters that are not human-readable. The artist controls inputs and direction but not the underlying mechanism.

Both are valid modes. The difference is where artistic intention lives: in the explicit rules of the code, or in the prompts, constraints, and curation that guide a learned model.

The name-to-color formula as a contemporary example

STILL Studio's name-to-color system sits closer to the traditional algorithmic art tradition than to pure AI generation. The formula is explicit: A through Z carry values 1 through 26, sum the letters, multiply by 137.508 degrees, take modulo 360 to get a hue on the color wheel. This is a deterministic rule written by a person, not learned from data.

What is AI-generated is the painting itself: the composition, texture, and visual execution in the chosen style. The palette comes from the formula; the painting comes from the model. The combination is both algorithmic in the Molnár sense (explicit rules producing specific outputs) and generative in the modern AI sense (learned visual patterns applied to produce original imagery).

It places contemporary personalized art production directly in the lineage that runs from 1965 Stuttgart to the present. The tools changed; the core question stayed the same: what interesting things happen when you give a rule to a machine?