The Algorithm Has No Favorite

When a drawing algorithm in the Swarm receives its parameters and begins to render, it does not prefer one composition over another. It does not lean toward symmetry or away from it. It does not favor dense mark-making over sparse fields, warm palettes over cold ones, the elegant over the awkward. It executes. The instructions arrive — line weight, angle, repetition count, color index — and the algorithm produces exactly what those instructions demand with the same indifference a calculator brings to arithmetic. There is no aesthetic judgment at runtime. The algorithm has no favorite.

This is a true statement and a deeply misleading one. The algorithm executes without preference, but the system that designed, selected, weighted, and deployed that algorithm is nothing but preference. The 300 algorithms in the Swarm did not arrive by accident. They were written, tested, revised, discarded, rewritten, and ultimately chosen from a larger set of candidates. The 24-tier system that evaluates algorithmic output against formal criteria — balance, density, chromatic coherence, compositional tension — encodes a specific set of aesthetic values so thoroughly that the system can rank half a million compositions without human review. The chromatic weights that determine how often a given canvas color or line color appears were calibrated to produce a particular distribution: more neutral grounds, fewer vivid ones, because the designer of the system believed that rarity should follow chromatic intensity. Every one of these decisions is a preference. They simply live at a different level than execution.

Monet at Rouen

Claude Monet returned to Rouen Cathedral more than thirty times between 1892 and 1894, painting the same west facade under different conditions of light and atmosphere. The resulting series — exhibited together at Durand-Ruel in 1895 — is often described as an exercise in perception: the artist as passive recorder of optical phenomena, the cathedral as neutral subject, the paintings as documents of light’s behavior across hours and seasons. But this description mistakes what Monet was doing. The decision to paint Rouen Cathedral thirty times was not a neutral act. It was one of the most aggressive aesthetic preferences in the history of painting. Monet chose the subject, chose the vantage point, chose to return, chose to exhibit the works as a series. The individual brushstrokes may have been responsive to light conditions, but the system within which those brushstrokes operated — serial repetition of a fixed motif under variable conditions — was a designed structure with deep preferences about what painting should investigate.

The distinction matters. At the level of the individual brushstroke, Monet was responding to what he saw. At the level of the series, he was constructing an argument about what painting is for. The brushstroke is execution. The series is design. And the preferences live in the design.

Morandi’s Bottles

Giorgio Morandi painted bottles, vases, and boxes on a tabletop for nearly fifty years. His studio in Bologna contained the same objects arranged and rearranged in slight variations, painted in the same muted palette of greys, ochres, and dusty whites. From outside, this looks like limitation — an artist who found one subject and never left it. From inside, it is the most extreme version of preference imaginable. Morandi did not paint bottles because he had nothing else to paint. He painted bottles because he believed that the formal relationships between simple objects in close proximity — the way a cylindrical vase crowds against a rectangular box, the way a shadow cast by one object modifies the apparent color of the object behind it — constituted an inexhaustible subject. Every painting is a variation. Every variation is a new argument about spatial relationship, tonal value, edge quality. The subject never changes. The investigation never ends.

Morandi’s bottles are his tier system. They are the constraint within which all formal variation occurs. The preference is total: this subject, this scale, this palette, this light. Within that total preference, each painting executes without favoritism among the possible arrangements. The hand responds to what it sees. But what it sees has been completely determined by decisions made before the brush touches canvas.

Warhol’s Removed Hand

Andy Warhol understood this distinction and exploited it. The silkscreen process he adopted in the early 1960s removed personal touch from execution. The squeegee drags ink across the screen with mechanical indifference — it does not prefer one area of the image over another, does not modulate pressure for expressive effect, does not hesitate. The execution is as close to algorithmic as a manual process can be. But the selection of subject — Marilyn, Elvis, Mao, the electric chair, Campbell’s soup — is pure preference. Warhol chose what to print with extraordinary precision. The Factory’s output was indifferent at the level of process and deeply opinionated at the level of subject. The squeegee has no favorite. Warhol had nothing but favorites.

The Swarm operates on the same split. The rendering engine has no favorite. It produces a Tier 1 Masterwork and a Tier 24 composition with identical computational indifference. But the system that determines what counts as Tier 1 — the formal criteria, the threshold values, the weighting of balance against density against chromatic range — is as opinionated as Warhol choosing Marilyn over every other face in America.

LeWitt’s Instructions

Sol LeWitt made the relationship between preference and execution explicit. His wall drawings exist as written instructions: “Lines from the center of the wall to specific points on a grid.” “All two-part combinations of arcs from corners and sides, and straight, not-straight, and broken lines.” The instructions are the work. The execution — carried out by assistants, by museum staff, by anyone who follows the instructions correctly — is indifferent. A trained draftsman and an untrained volunteer will produce recognizably different versions of the same wall drawing, but both versions are authentic instances of the work because the work is the instruction, not the execution.

LeWitt’s instructions ARE the preferences. They specify what marks to make, where to place them, what relationships to establish. The person drawing the lines does not choose — they follow. The algorithm does not choose — it follows. In both cases, the preference has been moved from the moment of making to the moment of designing. The hand that draws is indifferent. The mind that wrote the instruction is not.

Where the Preferences Live

In the Swarm, the preferences live in the tier system. They live in the 300 algorithms and the criteria that selected them from among hundreds of candidates that were tested and discarded. They live in the chromatic weights that make Voidmother rare and Limestone common. They live in the balance thresholds that determine whether a composition’s visual weight is distributed in a way the system considers resolved. They live in the decision to use a claw as the fundamental mark rather than a circle, a line, a dot, or any other primitive. They live in the 500,000-token supply rather than 10,000 or 100. Every parameter is a preference. Every default is a position.

The algorithm that renders a single token knows none of this. It receives its instructions — algorithm index, color values, density parameter, scale factor — and it draws. It draws with the same indifference whether the output will be ranked Tier 1 or Tier 24, whether the canvas is Voidmother or Fog, whether the composition will be collected by someone who frames it or someone who forgets it. The algorithm has no favorite because it has no access to the level of the system where favorites are encoded.

This is not a flaw. It is the architecture. Preference belongs to design. Indifference belongs to execution. The system is opinionated so that the algorithm does not have to be. And the result — half a million compositions produced with total indifference within a framework of total preference — is what gives the Swarm its particular character: not the randomness of chance, but the rigor of a system that knows exactly what it values and then lets the work happen without interference.

← All Writings