Nicolás Malevé
Exhibiting ImageNet

ImageNet is a gigantic archive of images for the visual training of Artificial Intelligences. It is made up of more than 14 million photos, organized into 21,000 categories. This classification work, which attributes a word to each photo, was carried out using Amazon Mechanical Turk, an online platform that outsources “human intelligence tasks” that machines cannot do. For two years, more than 25,000 workers dedicated them- selves to this mammoth labelling task, at the estimated speed of 2 images per second.

In this work, Nicolas Malevé addresses the conflict between on the one hand, the volume and processing speed of the algorithms and, on the other, the conditions of human work that sustain it. For this installation, he has created a script that displays approximately half of ImageNet –about 7 million photographs– during the 25 days of Getxophoto, at a speed of 180 milliseconds per image. The script randomly stops on some of them to allow slow human intelligence to recognise them. The categories in ImageNet come from WordNet, another database that was later found to classify photos of people with racial and gender biases. The one shown here is the partially edited version.


A. Resano

Nicolas Malevé (Belgium, 1969)
Exhibiting ImageNet

2023 Edition
Indoor installation
Digital curator: Jon Uriarte
Venue: Amezti Hall