Situating machine learning – On the calibration of problems in practice

ABSTRACT In this paper, we employ John Dewey’s notion of the situation as an analytic lens for observing and theorizing machine learning. Based on two ethnographic case studies in art and science, we account for machine learning as practice and examine the dynamics of the situations it gives rise to. Following Dewey, our observations focus on the transformation of situations from an initial state of indeterminacy through to problematizations and their resolution. Rethinking machine learning through the situation, we analyze how cooperating machine learners, both human and non-human, resolve situations and thereby refine their mutual attunement. With Dewey, we first explain how machine learners train through disruption and adaptation as they identify and solve problems. Second, we show that these problems concern issues of latency and addressability in efforts of cooperation between heterogeneous machine learners. Third, we discuss how machine learning practices cultivate situations that feature careful calibrations of problems that allow for their productive transformation. Our empirically grounded approach offers a pragmatist account of machine learning as a continually indeterminate and dynamic situated practice. As a contribution to ongoing discussions in social theory, we reframe existing characterizations of machine learning as issues of latency and addressability in cooperation.