Feeding the machine: Practitioner experiences of efforts to overcome AI's data dilemma

This paper examines the human implications of AI's ‘data dilemma' in three different and contrasting sectors: pharmaceuticals, higher education, and the arts. The ‘data dilemma' refers to the challenge of obtaining sufficient and suitable data to effectively train AI algorithms. The research, conducted in the UK, involved interviews, focus groups, and observations with 65 practitioners employed across these three sectors. The findings reveal that addressing the data dilemma often involves practitioners being pressured to generate data for AI, either passively in the context of data extractivism or actively by engaging in new forms of data production. We explore how this pressure to ‘feed the machine’ manifests differently in each sector, and how modes of resistance to these emergent data practices vary across sectors. We observe that the push to resolve the data dilemma is fundamentally driven by capitalist and technological solutionist values; values that often conflict with those of practitioners who are expected to adapt their practices in the service of AI-driven capitalism. We conclude with a call for exploring different approaches to AI development that align with alternative value systems.