Every copy is a new original

In 1514 Albrecht Dürer finished cutting a copper plate that would print an image of a winged figure sitting among instruments, refusing to use any of them. He called it Melencolia I. He then printed it, and printed it again, and the image entered the world already multiple.

It is worth being precise about what that was. The engraving is not a reproduction of a painting. There is no original painting. The plate is the original, and the plate is a machine for making identical objects — which means the question we usually date to photography, or to the scanner, or to the screenshot, was already fully formed in copper four hundred years before anyone worried about it.

What Dürer understood, and what the intervening centuries occasionally forgot, is that a reproduction is not a diminished version of a thing. It is a different thing with its own grain, its own failures, and its own audience.

A reproduction does not fail by losing detail. It fails by keeping the wrong detail and being believed.

Every plate in this exhibition has been put through the same reduction: each point in the image is forced to one of two values. There is no grey. A machine looks at a five-hundred-year-old face and is permitted to say only ink or paper.

Dithering is the set of lies we tell to make that bearable. Ordered dithering compares each point against a fixed grid of thresholds, so that a flat mid-grey becomes a regular weave. Error diffusion — the Atkinson method, written for the first Macintosh — instead carries the arithmetic remainder of each decision forward into its neighbours, so that the error is spread rather than repeated. Bill Atkinson chose to discard a quarter of that error, which is why images from 1984 have blown highlights and a particular nervous brightness.

None of these is more truthful than the others. They are different opinions about which lie is least distracting.

Hang the eleven plates together and the results sort themselves into an order nobody intended.

Dürer survives almost intact. This is not a compliment to the algorithm; it is a fact about engraving. An engraving is already a binary image — the copper is either cut or it is not — and the apparent greys are already produced by spacing lines at varying density. Dürer was dithering. Reducing his work to one bit returns it to something close to its own logic.

The marble does not survive. A Roman portrait bust is almost entirely a record of how light falls across a curved surface, and one bit cannot describe a curve. What comes back is a mask: the eye sockets, the line of the jaw, and an enormous amount of noise where the cheek used to be.

The paintings sit in between, and the most interesting failure is Bronzino's young man, whose face survives while the black doublet he is wearing becomes a void with no information in it whatsoever. Four hundred years of glazing, gone, because a dark thing on a dark ground has nowhere left to go.

It would be easy to present this as a lament, and it is not one. The fresco in Room IV lost half its surface to damp and time long before a camera reached it. The bust of a Roman woman lost her nose to something we will never know about. The Bronzino has been cleaned, relined and re-varnished more than once, and the face you see is partly a decision made by a conservator in the twentieth century.

Reproduction is not the moment these objects began to degrade. It is simply the most recent, most visible, and most reversible step in a very long series — and the only step in that series where you can drag a slider and put it back.

Go and do that