A Shadow of a Man in the Mirror presents stills taken from the 1934 experimental film Hands captioned with “subtitles” created using image-to-text software. The captioning program, NeuralTalk, is one of the first machine vision models designed to write sentences that describe images’ contents rather than simply identifying the objects present within them. This book exposes the limits of object recognition technologies; the inaccurate outcomes make explicit the unstable relation between images and what they are deemed to represent. An accidental poetry often arises from the erratic space of the machine. This book explores the transferability of meaning—or, rather, the impossibility thereof—between text and image, and the reproduction of bias-laden classifications in machine vision. The relatively recent ability for computers to segment and operationalize visuals as textual data has marked a major shift in image analysis. But in the cases illustrated here, its anthropocentric mischaracterizations point to the limits of computer vision when confronted with information that falls outside of a specific worldview determined by its training datasets.