In 'Zizi – Queering the Dataset 07' (2019), Jake Elwes addresses the lack of diversity and representation in the training datasets commonly used in facial recognition systems. The video shows the result of disrupting these systems by re-training them with drag and gender-fluid faces the artist sourced online. In doing so, the neural network is broadened out from the normative identities it was originally trained to recognise, moving instead into a realm of queerness. The work offers a glimpse inside the machine learning process, visualising what the network has—and hasn’t—learned. It celebrates ambiguity and difference, encouraging reflection on the pervasive biases in our data-driven society.

This video forms part of 'The Zizi Project' (2019–ongoing), a series of works that explore the intersection of artificial intelligence (AI) and drag performance. Drag, with its challenge to traditional notions of gender and identity, contrasts with AI, often mystified and complicit in perpetuating social biases. Elwes unites these themes through the production of deepfake synthesised drag identities created with machine learning. The project investigates how AI can learn from drag and how drag can, in turn, critique AI. Originally commissioned as a seven-channel video installation by Experiential AI at Edinburgh Futures Institute and Inspace, The University of Edinburgh, 'Zizi – Queering the Dataset' stands as a thought-provoking exploration of AI, identity, and representation. A further iteration of the project, 'The Zizi Show', was commissioned in 2023 by the Victoria & Albert Museum in London, supported by the Manitou Fund. This groundbreaking deepfake drag cabaret, which explores ethical dilemmas within AI, was acquired for the museum’s permanent collection.