Decision Making in Asylum Law and Machine Learning

Abstract

This article avails an autoethnography of the authors’ attempt to construct a post hoc intervention machine learning (ml) system responsive to the problem of discrimination in asylum law decisions. In the article we revisit the conjunction of law as a slow hermeneutic, against the fast-paced pull of ai and commercial imperatives to ask whether a ml-driven post hoc intervention system such as the one set up in the research project reduces the overall risk of discrimination emerging from human discretion in legal decision making on asylum. We conclude that a ml-driven ‘anti-discrimination machine’ will displace rather than reduce that overall risk. We warn that similar attempts at using ml as part of legal decision making, decision support, and post hoc interventions, in international law and beyond, may need to take seriously the risks of human discretion embedded in ml design and data selection.