Why are human epistemic agents not displaced in machine learning scientific inquiries? A practice perspective on ML in science

This chapter considers machine learning (ML) practices used in science. Because ML practices enjoy increasing degrees of automation at various stages of the process, the question whether human epistemic agents are displaced arises. We first point out that shifting focus from the ML outputs to the practice of designing and using ML models allows one to appreciate the role of different actors in this process, from the human designers and modelers to the algorithms themselves. We illustrate this point with a description of ML-based practices in neuroscience. We then go further with problematizing the role of human epistemic agents in ML and argue that they are not displaced.