The image that becomes the fact

There is usually plenty to photograph, and almost none of it is the reason anyone buys. A drug delivery device can be shot from any angle, while the plume geometry, the particle size distribution and where the dose lands in a lung cannot. A satellite ground station has an antenna standing in a field and a coverage argument that exists only as a diagram. A research group has a lab, and a finding that exists as a chart or does not exist at all. So the object is visible and the reason to choose it is not, and the picture that carries the argument has to be made by somebody.

A photographer chooses the lens, the light, the moment, what stays outside the frame. All of it shapes the picture, and all of it answers to something that exists. The other case is when there is nothing for the camera to answer to, and every property of the image has to be decided from the start. That is the territory of scientific illustration.

Almost every picture of the coronavirus you have seen descends from one drawing. A grey sphere, textured like stone, studded with red spikes. It was made by two medical illustrators at the US Centers for Disease Control (CDC), Alissa Eckert and Dan Higgins, and released on 31 January 2020. Eckert has said she thought it might turn up on a few news programmes.

The CDC opened its emergency operations centre for Covid-19 on 20 January, and the pair were asked the following day. Eckert has described the job as making a mugshot, something that represents what this enemy is…to most creatives this would sound like a branding brief.

The virus is too small to be seen in visible light at all. It has no colour, no texture and no shadow. All of it had to be decided. The difficulty, as Dan Higgins put it, was presenting something with no visual reference.

The structures came from the Protein Data Bank and were assembled in Chimera, ZBrush and 3DS Max.

The grey sphere is the lipid envelope, given a stony surface rather than a smooth or glassy one, on the reasoning that relating a thing to something that already exists makes it easier to believe.

The S proteins, the spikes, are red. The M proteins across the membrane are orange and the small E proteins are yellow.

The palette came from the CDC’s own Covid response colours. Blues and greens were tried and set aside; reds and oranges carried the register the brief had asked for. Higgins has described holding the saturation down so the result did not read as playful.

A real virion carries far more M proteins than spikes, so a drawing that gave every protein its true share would bury the part that mattered most, which is also the part that does the infecting. It would be far less frightening: bumps read as texture, spikes read as a threat and the brief had asked for an alert, not a diagram.

In editing what the eye reaches first, Eckert made significance the priority rather than abundance, and produced a picture the world has read ever since as a likeness.

What followed was a wave of simplified versions: the same silhouette flattened, redrawn, reduced to an icon on a poster or a slide and for sale on image stock libraries. Almost none of them went back to the structural data. They copied the drawing, which by then had become the reference.

The black hole at the centre of the galaxy Messier 87 sits 55 million light years away. It was observed on four nights in April 2017 by seven radio telescopes at five sites, on volcanoes in Hawaii and Mexico, mountains in Arizona and Spain, and in the Atacama Desert. Each produced around 350 terabytes of data per day. The hard drives from the telescopes were flown to supercomputers in Bonn and Massachusetts. The final image of the black hole was published two years later.

The telescopes do not record a picture. Each pair of them measures a single component of an image, and the array captures a few thousand of these where a complete picture would need billions. The collaboration is explicit about what follows: the problem is ill-posed, reconstructed images are not unique, and they always require assumptions beyond the measurements themselves. Many different images fit the data equally well. Something other than the data has to choose between them.

They chose smoothness, and a preference for the simplest image that fits. They tested the pipeline by feeding it fake data made from pictures they had drawn themselves, to see whether it gave those pictures back. Four teams worked in isolation for seven weeks, forbidden from discussing not just their results but even which telescopes might be unreliable, to see whether they would independently produce the same ring. They did, though its thickness and brightness varied between them. The published image is an average of three separate reconstructions, each blurred to a common resolution. The orange is a colour map applied to brightness temperature, chosen because it reads evenly to the eye.

They would not call this design, and they built the whole process to keep judgement out of it. It is still a judgement, because the measurements alone could not choose between the images that fit them.

These two images were made before it became possible to generate forty versions of anything in an afternoon, which changes the economics but none of the substance: the decisions are still there; what has gone is the account of them. Eckert can tell you why the M proteins were pulled back, and the collaboration published fifty pages on which assumptions it chose and why. An AI-generated image carries the same decisions, but nobody can say what was chosen, what was rejected, or why.

Once an image is drawn it circulates: into the press, the partner deck, the conference stand, the teaching slide, the competitor’s comparison.

The timing is the difficult part: Eckert’s account of why the image landed is that there was nothing else, so it mattered to do it early and to race to get it done. That is very close to the standard condition. A canonical image is almost always made early, quickly and cheaply, before anyone knows that it is going to be load-bearing.

You cannot identify it in advance and you cannot withdraw it either, because a later and better version does not retrieve the copies already in circulation, it joins them.

Rendered, filled in, fixed. One could be photographed but not in a way people could read, the other could not be photographed at all. Both needed somebody to decide what they looked like.

So the question is never whether an image is good enough for now, it is whether it is good enough for the ten years in which nobody revisits it.

Sources: Alissa Eckert interviewed by Alice Rawsthorn for Wallpaper; Dan Higgins interviewed by Dezeen; Event Horizon Telescope.