Sentencing and the Conflict between Algorithmic Accuracy and Transparency

Abstract

Abstract Predictive accuracy and transparency are generally recognized as ethically desirable features of algorithms at sentencing. However, it is often explicitly or implicitly assumed that there may be a conflict between transparency and accuracy. If an algorithmic tool is made more transparent, this will result in a loss of predictive accuracy and vice versa. The purpose of the present chapter is to discuss the nature of this conflict. More precisely, it is first argued that even if there is a conflict between transparency and accuracy, this does not demonstrate the conflict to be of genuine ethical significance. Second, even when there is a genuine ethical conflict between transparency and accuracy, this may sometimes be resolved in ways other than by engaging in trade-offs. Finally, the chapter discusses the theoretical and practical implications of these conclusions.