Reference Human is a generative collage of people from ImageNet. The collage assembles figures from a carefully-curated selection of ImageNet classes in order to materialize dynamic relationships beyond those captured in the ImageNet taxonomy. These configurations — demagogue, tyrant, and protester; matriarch and birthing coach; press photographer and artist’s model — suggest new narratives, plausible chains of cause-and-effect. The individual images are stitched together using a GAN into a unified (if chaotic) whole. When the work is shown, the camera pans slowly over the synthesized landscape, eternally revealing new figures and new relationships.
By blurring the boundaries between items in the dataset (both visually using a GAN, and conceptually by suggesting new categories), Reference Human attempts to restore the messy richness of the world we live in. Reference Human celebrates the ways in which our world stubbornly resists quantification and categorization and, in doing so, challenges the hegemony of the dataset.
The title Reference Human refers to the people depicted in the dataset (and shown in the collage), who become archetypes that machine learning models use to classify others. The title also nods to Alex Krizhevsky, a computer scientist who set the standard for human accuracy on ImageNet classification by manually classifying images for 1,000 classes in ImageNet. This sense of the title acknowledges the human labor used to construct the dataset, and the way that the task of object recognition is woven into human experience. Finally, the title Reference Human, identifies the artist as the reference for a new set of meanings brought to the images through the act of curation and collage.