Existing comment:
Adversarially Evolved Hallucinations
Adversarially Evolved Hallucinations may best be understood as "impossible images." Here, Paglen developed his own taxonomies from literature, psychoanalysis, political economy, and poetry. An artificial intelligence (AI) was trained to recognize different types of portents, or omens. Images whose metadata were tagged with labels such as rainbow, black cat, or eclipse then became the training set. Once it had ingested a sufficient number of training images to recognize portents, a second AI was used to "paint" that subject. "Together," Paglen said, "the AIs have evolved an image that is entirely synthetic and has no referent in reality, but that the pair of AIs believe are examples of things they've been trained to see."
Paglen describes these images as hallucinations because they have no relation to light, vision, memory, narrative, or any of the traditional components of human image making. As in "Goldfish," the titles explain how each was made. In these impossible images, plant, animal, and human figures appear to emerge and mutate within a primordial data ooze, the dream world of the AI. |