Criminal Vision 1

This is one of the earlier renderings of KATSU’s AI criminal portraiture series. Using machine learning and thousands of criminal images an AI learns to draw non existent figurative portraits. KATSU's portraiture series consists of a generative adversarial network built using Google's TensorFlow machine learning library and trained using vintage black and white mug shots scraped from the web, to output unique portraits. The pieces explore themes of inherent bias in algorithms, Wired Magazine notes that "there's more to AI portraits in KATSU's view, than just a comment on the growing power of technology—it's a warning about humanity's lack of control over what these tools can build".