•  
  •  
 

Abstract

This article addresses the judicial duty to explain a decision that is informed by the output of a predictive algorithm. In discharging this duty, courts generally point to evidence about how the algorithm performs: the reason for using an algorithm just is that it makes accurate predictions for all groups within the target population, without causing unjustified disparities between them. As predictive performance can be validated statistically, judges have not sought access to the algorithm itself. I argue for a more robust approach to transparency in certain cases, which involve predictions about individual behaviour. In these cases, there are reasons to want to know which variables count towards a predictive outcome, and how. In the context of a judicial decision, this information falls squarely within the duty to explain, which can ground a right to appeal from any decision that fails to supply it.

Digital Object Identifier (DOI)

10.55496/NDPN4077

Download

Share

COinS