Critical Data & AI Lecture Series #8. Abstracting to ‘see’ the particularities: Visualisation tools as techniques of knowing AI systems in the making

Lecture by Anna Schjøtt Hansen, University of Amsterdam.

Abstract

Artificial Intelligence (AI) systems are increasingly moving out of the research labs and into different industrial domains of application, such as the media, health, or public sector. This industrialisation of AI produces new questions about the localised knowledge practices that unfold within organisations and requires us to study how these systems are constituted beyond the lab. Some important work has already addressed how algorithms are constituted via the processes of ‘ground truthing’ (Jaton, 2021) or illustrated the particularities of how recommender systems are developed and deployed in the music sector (Seaver, 2023). This talk builds on this work but places the attention, particularly on how AI systems in the making are made ‘knowable’ within the public service media BBC, where editors, curators, data scientists, engineers and product managers must collaborate to ensure the system delivers on both commercial and public service goals. The talk will centre on the notion of ‘techniques of knowing’, to describe the “standardised yet plastic approaches or methods” (Rieder, 2020) that help to make AI knowable to non-experts and shape the development of AI systems. This conceptual proposal will be grounded in recent fieldwork conducted with DataLabs, the BBC team responsible for developing recommender systems. Following the daily work of making new Machine Learning (ML) powered recommendation systems for both BBC Sounds and iPlayer, two central techniques of knowing were employed; 1) visualisations of the system’s performance and 2) A/B testing.

This talk will predominately focus on the role of visualisations and how they both abstract and particularise certain ways of knowing the system in the making – as well as how these techniques are used strategically by the actors involved. Thereby, the talk will address questions of how to engage with the opacity and explainability around AI systems, but particularly foreground how these techniques of knowing participate in both epistemological and ontological politics (Mol, 1999; Schwandt, 2007) by privileging some knowledge and knowers in the process and by enabling some conditions of possibility over others.

Bio

Anna Schjøtt Hansen is a technological anthropologist and PhD Candidate in the Media Studies Department at the University of Amsterdam. In her PhD research, she ethnographically explores different epistemic spaces where AI systems are discussed, presented, developed, and evaluated to critically examine the politics of AI design processes and their implications. She is the co-organiser of the Critical AI Seminar series based in the Faculty of Humanities at the University of Amsterdam and one of the co-editors of the topical collection on the ‘Politics of Machine Learning Evaluation’ to be published in Digital Society.

About the lecture series

The Critical Data and AI Lecture series is organized by Louis Ravn and Nanna Bonde Thylstrup as a joint venture between the research projects AI REUSE (DFF) and Data Loss: The Politics of Disappearance, Destruction and Dispossession in Digital Societies (DALOSS, ERC Stg).