Icons {100% recognizable}

An icon is a sign that resembles its signified. It can also be a type of utilitarian image, stylized and designed to be immediately, unambiguously, and universally recognizable. An AI for image classification was trained on millions of photos tagged by people. This AI represents a consensus of recognition. Furthermore, it does not simply label an image as it sees it, but rather quantifies the likelihood that the image is similar to a certain object. This can be used to construct an image that maximizes the recognizability of a given object. This image is, in a sense, the maximum stimulus for the AI viewing it, and by extension for the average human perception that this AI is a surrogate of. This stimulus however, is not a photographic likeness, nor is it a stylized image that captures the "essence" or "archetype" of the object. Instead it is an assemblage of features and parts of an object. For example, the most ostrich-like image would be something like a fragment of an ostrich head, a flock of ostriches, or an ostrich body viewed simultaneously in profile and from the front from multiple directions and at multiple scales. In this series, we start with a grid of triangles and look for the configuration that will maximize AI's recognition of the image as a particular object. In many of these images, we also impose additional stylistic constraints to ensure that the AI only sees symmetrical images. The images with the highest recognition are selected and laser engraved onto a copper plate. Many of these images read very differently at different distances, and the metallic sheen helps visually isolate and highlight individual features as the viewer's position relative to the light changes. This replicates AI's fragmented perception and a sort of partial feature assembly rather than the concrete objects it sees.