2019, multi-channel digital video, 135 minute loop
‘Zizi - Queering the Dataset’ aims to tackle the lack of representation and diversity in
the training datasets often used by facial recognition systems. The video was made by
disrupting these systems* and re-training them with the
addition of drag and gender fluid faces found online. This causes the weights inside the
neural network to shift away from the normative identities it was originally trained on
and into a space of queerness. ‘Zizi - Queering The Dataset’ lets us peek inside the
machine learning system and visualise what the neural network has (and hasn’t) learnt.
The work is a celebration of difference and ambiguity, which invites us to reflect on
bias in our data driven society.
*A Style-Based Generator Architecture for Generative Adversarial Networks
The Zizi Project (2019 - ongoing) is a collection of works by Jake Elwes exploring
the intersection of Artificial Intelligence (A.I.) and drag performance. Drag
challenges gender and explores otherness, while A.I. is often mystified as a concept
and tool, and is complicit in reproducing social bias. Zizi combines these themes
through a deepfake, synthesised drag identity created using machine learning. The
project explores what AI can teach us about drag, and what drag can teach us about
Instagram @zizidrag - machine learning
generated captions trained on drag profiles.
Zizi was originally commissioned as a seven channel video installation by Experiential
AI at Edinburgh
Futures Institute and Inspace, The University of Edinburgh. Presented
site specific video installation with between 3 and 10 projected video channels.
Zizi: Queering the Dataset 2019, 30 second extract of single channel
7-channel installation at Inspace City Screen, Edinburgh Fringe Festival 2019
Zizi: Queering the Dataset 2019, custom LED screen