Where have you seen that sculpture? Perhaps at an art fair, in one of hundreds of dizzying booths, or maybe featured in the glossy pages of a design magazine? The answer is neither, because none of the works pictured above are real. If the sculptures seem uncannily familiar, though, it’s because they are.
For his project “This Sculpture Doesn’t Exist,” artist Matteo Rattini trained a neural network — a set of algorithms designed to recognize patterns — to create images of contemporary sculptures based on Instagram’s suggestions. The resulting totemic structures, biomorphic forms, and sleek stacked shapes against drab gray backgrounds have a strange, mass-produced patina, as vaguely similar as Ikea furniture.
“The algorithm just accommodates users’ tastes, creating an artificial environment in which diversity is replaced by repetition and standardization,” Rattini told Hyperallergic. “While seeing images of a show you haven’t been to yet could spoil the experience, the algorithms and strategies employed by social media in content managing are working on a deeper level, changing and reshaping the perception of art itself.”
Rattini, who is currently studying multimedia arts at Università Iuav di Venezia, wanted to understand how Instagram’s algorithm was impacting his experience of art as well as his own practice.
“The way to discover it was to have the neural network exposed to the same quantity of images and information and see what it would produce,” he said.
The artist opened a new Instagram account, followed a few contemporary art profiles, and created a script that would automatically engage with the suggested posts — what he calls “biting into the algorithm recommendations baits.”
In the beginning, Instagram’s suggestions were wide and diverse. “But with time everything started to look the same,” Rattini said. “From pictures of paintings, installation, performances, it slowly reduced to pictures of minimal modernist sculptures showcased inside an aseptic white cube.”
The artist then fed about 4,000 photos of artworks into a Generative Adversarial Network (GAN), a type of neural network made of two competing algorithms — a discriminator, which assesses the images in a dataset, and a generator, which produces random images. Based on the discriminator’s feedback, the generator slowly improves, yielding increasingly more realistic visuals.
“In a way, this mechanism reflects how our brain works, in a very simple way: you feed it a lot of information on trees, for example, and when it comes the time to draw a tree, not only is your idea of a tree informed by all the trees you’ve seen, but while you are drawing it, your brain is constantly judging it through your eyes and giving real-time feedback on how to compensate and adjust,” Rattini explained.
Indeed, the computer-generated sculptures became more and more convincing, evolving from blurry, amorphous blotches to clearly defined, structured, and realistic imitations of art. Any apparent diversity in form, texture, and color is flattened by a more pernicious sameness, an underlying monotony that is harder to pinpoint or describe.
The project helped Rattini visualize the otherwise imperceptible processes implemented by platforms like Instagram to tailor their users’ experiences, with the inevitable result of homogenizing the content they see — showing them artworks they are guaranteed to like, or, more dangerously, political views they already agree with.
“Every post you like, every image you share, every topic you show interest in, becomes data used to train the recommendation algorithms to better understand you, manipulate you, and predict what you’ll like and what will keep you online,” Rattini says.
Still, Rattini says he is not against social media; he just wants to get a better sense of how it operates, and more insight into how it might affect us.
“For me, it comes down to always remembering that the same tool that made me discover artists such as Rothko or Bacon could be altering my perception of art and my artistic production if not used consciously,” he said.