A model asked to write Yoruba strips the diacritics that carry tonal meaning. Prompted for a Yoruba royal palace, it produces a European castle. These structural errors arise from datasets reflecting persistent inequality in whose voices, images, and knowledge have been digitised.

Untangler is a multi-system art and technology project addressing this asymmetry directly. Through parallel systems for African language preservation, cultural knowledge transmission, and visual representation analysis, it works at the root — the data, language, and imagery AI learns from.