The process of seeing paintings, or seeing anything else, is less spontaneous and natural than we tend to believe. A large part of seeing depends upon habit and convention.
– from Ways of Seeing (1972) by John Berger

In 2018, the London-based digital media researcher Gabriel Pereira teamed up with the Brazilian artist Bruno Moreschi to investigate how algorithms would interpret contemporary art. For the project, the duo ran 654 works from the Van Abbemuseum in the Dutch city of Eindhoven through a homemade image-recognition program composed of six different commercially available AIs. In Recoding Art, Moreschi and Pereira guide us through the museum’s collection as viewed by AIs, revealing their mostly strange, often comedic readings that both simplify and expand upon the artworks’ meanings. To contrast AI and human perspectives, Moreschi and Pereira also asked Amazon Mechanical Turk workers – digital labourers who take on tasks that computers can’t yet do – to add their own reactions to the artworks, adding another rich layer to the project. The result is an entertaining and thought-provoking exploration of subjectivity, capitalism and machine learning that shows how AIs reflect the imperfect humans who train them.