The Practices and Politics of Machine Learning: A Fieldguide for Analyzing Artificial Intelligence

This article develops an analytical and methodological field guide for analyzing the practices and politics of machine learning. Drawing on science and technology studies (STS), it seeks to move beyond the opacity/transparency dichotomy that often characterizes discussions of algorithmic systems. In doing this, the article hones in on actors’ mundane practices of constructing machine learners, and develops an analytical stance based on four empirical moments: feature extraction, vectorization, clustering, and data drift. By making visible actors’ work, tinkering, and negotiations involved in these processes, the article aims to demystify the supposed magic of machine learning. This approach allows us to ask: How do actors decide what constitutes a good machine learner? How are things, people, and phenomena translated into the mathematical worlds of machine learning? Through ethnographic attention to these questions, we can begin to grapple with the practices and politics of machine learning in earnest. Thus, the article contributes to ongoing discussions about the societal implications of AI by providing analytical and methodological tools for understanding how the politics of machine learning is assembled in practice.