Co-creating anti-racist datasets in AI workflows utilising films as data

Egbe, Amanda

Abstract

Considering the concern for racial bias within AI algorithms, could creative responses within moving image archival practice and critical film theory foreground possibilities for intersectional approaches. This paper stems from artist practices with moving image archival materials that consider the materiality, historiography and ontology of film, such as Tom Tom the Pipers Son (Jacobs, 1969). Coupled with critical race theory, feminist and intersectional approaches (Anthias and Yuval-Davis, 1992), (Yuval-Davis, 2011), we can consider the broader applications of producing datasets for and with film analysis to resist bias in AI and computer vision workflows. Situated within the context of practice, drawing in broader image archives to produce critical moving image datasets, this approach aims to consider strategies for AI and cinema workflows that are independent of uncritical responses to the film canon and responsive to the notion of the coded gaze (Buolamwini, 2016). The paper reflects on the author’s artistic practice of utilising image archives to supplement the
analysis of the representation of race by creating datasets. These datasets reflect the context of onscreen racial depiction and the situated context of the broader cultural environment of the film’s
making.

• Presentation Conference Type: Presentation / Talk
• Conference Name: AI and Archives: Explorations, Possibilities and Challenges
• Start Date: Apr 27, 2023
• End Date: May 27, 2023
• Deposit Date: May 30, 2023
• Publicly Available Date: Jun 21, 2023
• Public URL: https://uwe-repository.worktribe.com/output/10828711