Instagram’s algorithmic feed performed the machine-driven curation. In recent months, Instagram has publicly announced that the content displayed in users’ Home feed will increasingly be decided by a “black box” algorithm, rather than what friends or family have recently posted.

This means that we do not know exactly what Instagram chooses to prioritise, or why. Through its prioritised selections, the algorithm reveals its own ways of seeing, providing the audience with an intimate lens into its perceptual mechanisms.

We captured Instagram’s curatorial decisions by uploading 800 images from the Metropolitan Museum of Art’s Open Access collection to a new Instagram account: @thealgorithmicpedestal. At the time, this account was followed only by one other account (@following_algorithmicpedestal). Visiting the Home feed of the following account, we saw which images Instagram chose to display from @thealgorithmicpedestal, and in which order. No captions, metadata, or social information was provided, such that the algorithmic curation was determined only based on the image itself.

Artist Fabienne Hess elected to select images that correspond to a concept: loss. Hess believes that loss is a uniquely—and universally—human experience, inescapable in human lives.

The images displayed here are part of Hess’ “Dataset of Loss,” which she has created over the course of three years as a resistance to the dominant algorithmic ways of seeing, which are shaped by the commercial interests of Big Tech. Her curatorial process is driven by the human experiences of time, curiosity, and patience; she has spent years physically exploring collections in an embodied fashion, learning about each object’s stories and photographing them during site visits. In this way, Hess’ curation represents both a very human process and a very human selection criteria.

Hess’ curation contributes to the field of machine vision from an artistic angle, rather than a commercial or technological one. Algorithms are trained on labelled datasets; Hess views her curation as a dataset as well. However, the organizing principle of Hess’ dataset—the theme of loss—is not a label stating what the images depict, which is how current algorithmic datasets are labelled for the purposes of computer vision. Instead, the images in Hess’ dataset ask questions about origin and content while resisting being categorized and labelled. She is working on a forthcoming book of a broader “Dataset of Loss,” including her own photographs and images from other collections.

Hess graduated from the Royal College of Art in 2012 and is based in London. Her work can be seen on her website and Instagram.