Machine Learning and the Ethics of Induction

This chapter analyzes the inferential structure of machine learning (ML) systems, and shows how these can be value-laden in unexpected ways. ML systems follow an inductive inferential strategy, which is based on two components. First, there is the basic assumption that we are entitled to predict future events on the basis of past occurrences because the world will not drastically change. This assumption is called ‘uniformity of nature’ (UoN). Second, ‘canons of inductive inference’ (CIIs) are required to narrow down the set of possible hypotheses that one can generate from UoN. Debates on the ethics of ML have focused on CIIs. Here I show that UoN plays an important ethical role, in particular in eroding human agency.