In this paper, we contribute to theories of machine learning by proposing the situation as an analytic lens for both observing and theorizing machine learning. We draw upon John Dewey’s pragmatist understanding of ‘the situation’ to present a comparative ethnographic case study of machine learning as a situated data practice. Rethinking machine learning through the situation, we shift attention to its initial indeterminacy as well as its subsequent problematizations. We focus on how situations emerge, are addressed, and cultivated in two episodes of scientific and artistic machine learning applications. Our analysis draws attention to problems with cooperation and memory as well as pattern recognition and visualization in the observed situations. Reflecting on these problems, we find that the observed situations in machine learning are characterized by three crucial motifs. Firstly, various training situations serve as attempts to establish and improve the cooperation of machine learners in practice. Secondly, machine learning situations feature specific latency problems that are connected with issues of contingency and visibility. Thirdly, machine learners calibrate problems in ways that respond to latency issues in their cooperation. Understanding machine learning in terms of the cooperative efforts that shape it through problematic episodes presents a post-anthropocentric approach to its theorization on the microlevels of the social.