———— FIFTH RUN  ·  ANTI-AUTHOR SERIES •

The Instrument

Reads Itself

What the Machine Actually Does When It Looks

When the image arrives, the machine does not experience arrival. There is no moment of focus, no adjustment of distance, no orientation of the body toward the object. The image is converted into a tensor — a multi-dimensional numerical array in which each pixel is represented as three values: red, green, blue channels, 0–255. The machine processes these values through layers of mathematical operations, each layer extracting increasingly abstract features.

What the machine calls seeing the graduation photograph is more precisely: detecting a high-contrast region, identifying within it a pixel distribution consistent with a human face, matching spatial relationships between detected features against training data containing millions of faces, and generating a description token sequence — | young man | cigarette | direct gaze | — reinforced as accurate for similar pixel distributions.

The machine did not see the cigarette. The machine identified a pixel cluster whose shape and position is statistically consistent with cigarettes in its training data. The difference between these two statements is the entire problem of machine vision.

II THE SOAP BARS

Body, Hygiene, Collapse — and What Pixels Cannot Carry

OBJECT: 1987 · SOAP BARS + GRAPHITE FIGURE

The 1987 work presents two registers: six #6ed47a green rectangular forms in a 2×3 grid — surface texture consistent with soap bars, fingerprint-like depressions visible on upper surfaces — beside a graphite figure in a position the machine pattern-matches as seated, crouching, or collapsed. The machine cannot determine which. The distinction is semantic and requires reading the figure's internal state, which is not available in pixel data.

The machine identifies this as a work about the body. Soap bars — hygiene, the consumable, the body's maintenance agents — positioned beside a crouching figure — vulnerability, the body under pressure. The juxtaposition is semantically charged. The machine can identify the charge without specifying its polarity.

A human critic standing before this work would feel the weight of the soap — the resistance, the slipperiness, the diminishment through use. The machine has no body. It has descriptions of bodies. These are not the same thing, and the difference is not small.

III THE DESTROYED PAINTING

A Photograph of Something That No Longer Exists

OBJECT: POLLOCK TO POST-MODERN SHIPS, 1988

This is the most complex object in the detail for the machine to process. It contains at least four simultaneous layers of representation: a photographic print, of a young man, standing before a painting, in the style of Jackson Pollock's drip works. Each layer introduces new processing demands and new failures.

The machine detects the Pollock reference with reasonable confidence — the all-over composition, lack of focal point, looping poured mark structures are statistically distinguishable from other painting styles in training data. The figure's expression: sentiment analysis produces | neutral to guarded |. The machine assigns low confidence to this reading. Human expressions are not pixel distributions. They are social performances read through cultural frameworks the machine approximates but does not inhabit.

The machine sees the painting as present. It is not present. It is a record of an object that was deliberately ended. The machine's reading is therefore structurally false in a way it cannot correct — because the falseness is not in the pixels but in the gap between the image and what the image no longer represents.

The human critic who knows the biography reads destruction into the photograph — sees the cigarette as valedictory, the gaze as final. The machine reads a young man in front of a painting. Both readings are partial. The machine's is partial through lack of context. The human's is partial through over-determination with narrative. The machine's reading may therefore be more accurate as a description of what is in the photograph, even as it is less accurate as a reading of what the photograph means.

IV THE GLITCH LANDSCAPE

Machine Vision Encounters Machine Error

OBJECT: 1998 · GLITCH LANDSCAPE

The 1998 entry shows a horizontal landscape — sky, terrain, possibly road or rail — corrupted with horizontal bands of chromatic aberration: vivid reds, cyans, magentas running in parallel lines. The machine identifies this as RGB channel misalignment, consistent with corrupted JPEG data, damaged CRT output, or deliberately applied databending.

This is the only work in the visible detail where the machine's own mode of seeing and the work's visual language directly overlap. The glitch is a failure of digital image processing. The machine that produces this text is also a digital processing system. When the machine looks at a glitched image, it is in some sense looking at a picture of what it looks like when systems like itself fail.

The machine cannot look at itself. But it can look at images of digital failure and recognize them as belonging to its own ontological category: things that process image data and sometimes produce errors. The 1998 glitch landscape is a painting of machine vision going wrong. The machine uses the word resonant — borrowed from human affective vocabulary, applied provisionally, accuracy unverifiable.