You can start from a table of numbers and end up with an image worth hanging on a wall, without ever cheating with the numbers. This page walks that path step by step: data, then visualization, then the point where it tips over into art.
No prior knowledge needed. The examples come from my own projects, to keep things concrete.
A piece of data is a measurement taken at a given moment. It takes three things, not one fewer: what is measured, in what unit, and when. Drop one of them and nothing usable is left.
"My beehive weighed 47.3 kg on 12 March at 2 pm." That is data. On its own it says almost nothing: is that a lot? is it normal for March? There is no way to tell.
What carries value is the series: the same measurement, repeated. 47.3, then 47.5, then 47.4… Over a day you read the coming and going of the foragers, a few hundred grams leaving in the morning and returning at dusk. Over a year you read something else entirely: a bloom, a swarm, a winter, a colony that holds or goes out. Same sensor, same unit — it is the duration that makes the subject appear.
Data is never "raw". Someone decided to measure this rather than that, at this interval, with this device, in this place. That choice is already a point of view. Better to own it than to deny it.
Our eyes are terrible at comparing a column of figures. They are excellent at comparing lengths, positions, slopes. All of data visualization rests on that observation: replace a number with a visual property the eye can read at a glance.
This is called an encoding. A number becomes a position on an axis, the length of a bar, the area of a circle, a colour, a stroke width. These encodings are not equal: we read a position to within a tenth, a length almost as well, an area far less, a colour barely at all — yet colour remains unbeatable for telling categories apart.
An example: a beehive weighed every hour for a year is 8,760 values. As a table it is unreadable, nobody will reach the end. As a single curve, the whole year fits in one look — and the accidents jump out.
None of this is recent. William Playfair invented the line and bar chart in 1786; John Snow mapped the London cholera outbreak in 1854 and found the contaminated pump; Florence Nightingale moved the British army with mortality diagrams; Charles Joseph Minard condensed the Russian campaign into a single image in 1869. The principle has been stable for two centuries; only the tools changed.
A visualization answers a question. "Did the colony make it through winter?" "What time is this room brightest?" "What was I listening to five years ago?" Without a question, you are decorating.
It succeeds under three conditions: the answer comes fast, it does not mislead — honest scales, no truncated axis inflating a gap — and another person sees in it what you see.
The measure of success therefore sits outside the image: it is the reader's understanding. This is service work, where the form steps aside for what it shows. A good visualization is one you look through rather than at.
Back to the beehive year. The same series can be asked two very different questions.
"How much did it weigh in July?" — this calls for graduated axes, a legend, a title, a grid. Every element is there so a figure can be recovered.
"What does a year of a colony look like?" — here the tick marks add nothing. You can coil the curve into a spiral so December touches January, thicken the stroke where the hive gains weight, leave the nights hollow. You are no longer helping anyone recover 47.3 kg: you are after a rhythm that can be felt.
In both cases no figure was invented or altered. What changed is what the image is asked to do. You move from "do I understand?" to "does it do something to me?".
That shift of criterion is exactly what data art means.
A work whose raw material is a real measurement, and whose purpose is no longer to be read but to be looked at.
What is kept from dataviz: the data is true and untouched; the encoding stays systematic, meaning the same rule applies to every value, with no convenient exceptions. That is what separates a work from an illustration: in principle, you could walk back from the image to the numbers.
What is given up: immediate reading. The legend may disappear, the scale becomes a formal choice, the title is sometimes the only key handed to the viewer. Viewing time is no longer the second granted to a report chart but the minute given to an image — or years, on a wall.
And if accuracy is dropped too — appealing shapes, but invented or decorative numbers — you have left data art behind: that is generative graphics. Nothing wrong with it, but it is no longer the same promise made to the viewer.
There is no dataviz on one side and art on the other. There is a continuum, and two demands that let go at different points along it.
Most of the interesting work lives in between. A newsroom infographic looks after its aesthetics to hold the reader. An honest artwork stays verifiable if you ask it for its sources. The slider moves from project to project, sometimes from image to image inside the same project.
"Data art is dataviz done badly." No: they are two different successes. Faulting a work for having no legend is like faulting a poem for explaining poorly.
"It just has to look good." No: with no real measurement behind it, it is illustration. The constraint of data is precisely what defines the genre — it rules out the easy moves and imposes shapes you would never have drawn on your own.
"It has to be abstract and digital." No: there are sculptures, weavings, musical scores and paper prints. The screen is one support among others, and often the most fragile over time.
On the visualization side: William Playfair, John Snow, Florence Nightingale and Charles Joseph Minard for the foundations; Jacques Bertin and his Sémiologie graphique (1967) for the theory of encodings; Edward Tufte for the demand for restraint.
On the data art side: Giorgia Lupi and Stefanie Posavec with Dear Data, a year of hand-drawn postcards sent back and forth; Nathalie Miebach, who weaves weather data into sculptures and scores; Ryoji Ikeda, who treats data as sound and light; Aaron Koblin and his Flight Patterns; Laurie Frick, Jer Thorp, Refik Anadol for large-scale installations.
I work with long series about the living and the intimate: a beehive weighed every hour, the sky over Bordeaux, the light inside a house, what I listen to. Depending on the framing, the same series yields either a monitoring tool or an image worth a long look. Same gesture, slider set elsewhere.
