Algorithmic Judicial Ethics

Abstract

Judges have a brand new bag—an algorithmic accessory in criminal adjudication. It scores criminal defendants, aiming to inform judges which defendants are likely reoffenders or flight risks and which ones are not. The downsides, however, include that the algorithms score defendants primarily on the basis of other defendants’ (mis)conduct and that certain races effectively score lower than other races. This article explores these algorithmic developments in criminal courts across the country and makes four contributions: (1) a survey and preliminary application of judicial ethics to this development; (2) a preliminary moral argument, informed by related judicial ethics and legal standards, suggesting that judges should use these algorithmic tools only to help, not hurt, individual defendants; (3) an approach to judicial decision-making in the shadow of structural injustice that promises to deal less algorithmic damage to defendants and their family members; and (4) a technical constraint on algorithmic design that ensures equal (indeed, better than equal) protection on the basis of race.

Keywords: Algorithms, Artificial Intelligence, Judicial Ethics, Risk Assessment, Bias, Discrimination, Statistical Methods

JEL Classification: K1, K4, K23