Can a machine be trained to recognize shapes in clouds? The Cloud Doesn’t Contain Sheep adopts the conventions of machine vision—bounding boxes, labeled datasets, and iterative accumulation—to test the assumption that the world can be discretized, categorized, and detected. Applied to clouds, whose shifting forms depend as much on projection and memory as on visual pattern, the logic of object detection models begins to unravel.

Over 800 cloud formations are hand-labeled by the artist and printed in real time as a growing stack of indexed pages. What first appears to be a training dataset gradually becomes a record of perception, where the act of  labeling shifts from neutral annotation to a reflection of bias, desire, and persistent belief that something is there to be found in the clouds.