Michael Mandiberg

Michael Mandiberg has collected and analyzed 130 million stock photographs to create Taking Stock, a series of photographs and videos that surface the ideologies haunting these ubiquitous images.

Over the past few years these massive stock image archives have been a key source for training generative AI models. AI tools treat stock photography as factual information, but they are far from neutral. Mandiberg wanted to know what is in these images? Which is to say, what is AI learning about us?

Mandiberg wrote code that systematically downloaded every image of an individual person on every stock photo website they could find, along with all their metadata. They used Machine Learning and AI tools to textually and visually analyze these stock photographs. Sorting and clustering the images by thematic topic, body pose, and hand gesture. These processes led to a series of unique composite image works, where the code merges the 64 images at the statistical center of this cluster of images.