The practices and politics of machine learning: a field guide for analyzing artificial intelligence

Artikel i vetenskaplig tidskrift, 2025

This article develops an analytical and methodological field guide for studying the mundane practices that constitute machine learning systems. Drawing on science and technology studies (STS), I move beyond the opacity/transparency dichotomy that has dominated critical algorithm studies to examine how machine learning is assembled through everyday work. Rather than treating algorithms as black boxes or magical entities, I focus on four empirical moments of translation—feature extraction, vectorization, clustering, and data drift—where technical work becomes political choice. By ethnographically attending to practitioners' tinkering, negotiations, and valuation practices in these moments, we can trace how classification systems are constructed and stabilized. This approach allows us to ask: How are particular features of the world selected as relevant for prediction? Through what practices are people and phenomena translated into mathematical vector spaces? How are temporal assumptions encoded in data? By studying these mundane processes of construction, we can understand how machine learning systems enact particular ways of seeing, classifying, and predicting the world. This field guide thus contributes methodological tools for analyzing how the politics of machine learning is assembled in practice, opening analytical space for critical engagement beyond calls for transparency or fairness.