Calculating Fairness. Enacting statistical justice in the case of risk recidivism

• In Event: Digital Technologies in Policing and Security - Session 3

Abstract

For some time now predictive algorithmic systems have been utilized within juridical practice, most prominently discussed in the case of Northpointe’s COMPAS. In the last years, this case sparked discussions on racial bias, predictive accuracy, and algorithmic fairness. Interestingly, in the discussion different mathematically formalized notions of fairness were used to justify or contest the algorithmic system. In our contribution we will focus on these conflicts and their particular subtext: a) the trade-off between classification markers that are potentially biased and the plain, predictive power of models; b) the evaluation of error rates in different representations of the same statistical model; c) different modes of subjectivation based on applied categories assessing the changeability of variables in terms of social intervention and individual accountability. This raises questions of whose knowledges, visions, and interests are dominating the technological restructuring of punitive jurisdiction institutions and practices. As a democratic legal state institution, judiciary institutions may operate under a formalistic-juridical tension, however, the question of justice and justification cannot be decided by judiciary intuitions alone, but should be embedded in a broader democratic discussion. Hence, we argue that the technological shift happening at certain courts challenges the very democratic questions of participation and representation. Which social identities are affected by the used statistical variables? Which ideas of the citizen or social ecology are used and reproduced by the discourse hegemonic voices? Who can and does evaluate the usage of SDSS at courts?

Authors

• Nikolaus Pöchhacker, University of Graz

• Peter Müller, Technical University Munich