Paper short abstract:

Based on an ethnographic enquiry at the BBC, this paper contributes to the demystification of AI through a situated and non-deterministic account of the epistemic techniques employed to know and collaborate around an AI system in the making and the epistemological politics that shape this process.

Paper long abstract:

AI systems are generally portrayed as powerful yet abstract and inscrutable entities. In reaction to these dramatising portrayals, STS scholars have provided alternative, ‘grounded’ narratives of AI systems focussing on the practical and localised efforts required to make AI systems work. Building on these efforts to demystify AI, this paper provides a situated account of AI systems in the making. It ethnographically traces the everyday work and decisions of data scientists, engineers, product managers and editors within the BBC as they collaborate to develop recommender systems that can better distribute their vast collections of content. Whereas the existing literature has helped to ground the study of AI in the materiality of hardware and infrastructure as well as the socio-material labour of producing datasets for AI systems, this paper highlights a different socio-material practice. When making new AI systems, localised epistemic techniques are employed to enable different actors to ‘know’ and collaborate around emerging AI systems. Particularly, the paper highlights the role of visualisations as techniques of knowing, as different visualisation tools are often at the centre of the collaborative practices. By analysing observational and interview data from an ethnographic enquiry at the BBC conducted from September 2023 to February 2024, the paper shows how the visualisation tools shape the development of AI systems by enabling the actors to see certain 'particularities' of the system, while also abstracting away other ways of knowing the system. Thereby, the paper sheds light on the epistemological politics that shape the making of AI systems.