In November 2022, the Museum of Modern Art in New York filled its lobby with a 24-foot-square screen of shapes that never stopped moving. Refik Anadol’s Unsupervised was built from the metadata of 138,151 records in MoMA’s own collection (titles, makers, dates) that the museum had posted publicly online in 2016. A machine learning model mapped that collection in 1,024 dimensions and then wandered through it, imagining modern art that might have existed. It was extended four times. Visitors spent an average of 38 minutes in front of it, which in museum terms is close to a miracle! In October 2023 MoMA acquired it, its first generative artwork for the permanent collection.
That’s the headline version of data in the studio. But artists have been working with numbers for a long time, and the more interesting question is what they do with them. Some let the data paint. Some let the data vote. And some count things by hand.
When the data paints
Anadol is the best-known name, so let’s be honest about him. His studio, founded with Efsun Erkılıç in 2014, makes big, liquid, colorful projections out of huge datasets. For the LA Philharmonic’s centennial in 2018, WDCH Dreams poured the orchestra’s archive (587,763 images, 1,880 videos and 17,773 audio files) across the outside of Frank Gehry’s Walt Disney Concert Hall using 42 projectors. Gehry’s team handed over the building’s 3D files so the light could fit every curve.
His critics say the results can look alike from one dataset to the next, swirling pigment that could be anything. We think that’s partly fair. The most moving piece of his, to us, is the smallest in concept. Melting Memories (2018) came after his uncle was diagnosed with Alzheimer’s, and it used EEG data from a neuroscience lab at UC San Francisco to make the physical traces of memory visible. When the data has a person behind it, the swirl means something.
Anadol’s work isn’t new in spirit, either. Harold Cohen began writing AARON, a program that made its own drawings, at UC San Diego in 1968, and kept refining it until he died in 2016. AARON showed at the Tate. Generative art has a 50-plus-year history. The datasets just got bigger.
When the data votes
From 1994 to 1997, the Russian-born duo Komar and Melamid did something sly. They hired professional polling companies in 11 countries, including the United States, Russia, China, France and Kenya, and asked people what they wanted to see in a painting. Then they painted each country’s “most wanted” and “most unwanted” pictures, and published the statistics in a 1997 book, Painting by Numbers.
It was a joke, mostly. But it was a joke with teeth. Komar and Melamid were testing what happens when you let audience preferences decide the art, which is exactly what engagement metrics now do to so much of what we see online. A painting made by poll looked, unsurprisingly, like a compromise. Thirty years later, a lot of feeds look like that too. We’d put this project on any reading list about analytics and creativity, ahead of anything about dashboards.
When the data is drawn by hand
Then there’s the slow, human end. In 2014 the information designers Giorgia Lupi and Stefanie Posavec, living on opposite sides of the Atlantic, started mailing each other one hand-drawn postcard a week for a year. Each card charted something small from their lives: how often they complained, the doors they walked through, the times they said thank you. It became the book Dear Data in 2016, and in 2017 MoMA added the original postcards to its collection. Lupi calls her approach “data humanism.” The cards are wonderful because they’re imprecise! You can see the pen wobble.
The climate scientist Ed Hawkins of the University of Reading went the other direction, stripping data down until only color was left. On May 22, 2018, he posted his warming stripes: a row of vertical bands of blue and red, ordered by year, cooler years blue and warmer years red, using the most saturated shades from the ColorBrewer palettes. No axes, no numbers. It became one of the most reproduced images of the decade, on the cover of The Economist in September 2019, on book jackets, on scarves. In 2023 he had to add a darker red because the year ran so hot. We’ve written about a similar idea worn as clothing in climate data you can wrap around your neck.
What separates the good work
So how are artists actually using analytics? The cynical answer is audience data: post at the right time, chase the color palette that performs. That’s real, and it’s mostly dull. The work that lasts treats data as material, not as a manager.
Our rule of thumb from these examples: the size of the dataset doesn’t matter much. Lupi and Posavec used a few dozen data points a week. Hawkins used one temperature per year. Anadol used hundreds of millions of images. What they share is a clear decision about what the viewer should feel. For more data turned into form, see how artists use 3D printing to turn natural data into physical sculptures, or the live data visualization built for the Tokyo Olympics.
