Automation and Fairness: Assessing the Automation of Fairness in Cases of Reasonable Pluralism and Considering the Blackbox of Human Judgment
12 Pages Posted: 16 Nov 2020
Date Written: September 24, 2020
In this article we probe whether the automation of fairness is in itself inherently unfair i.e. that the very attempt to automate a particular process that involves issues of justice, will necessarily be unfair. We do this by first explicating what we mean by fairness, and then discuss the notion of discernment i.e. deciding what kind of fairness to choose between in cases of reasonable pluralism. Following this we outline the difference between automating implementation and automating fairness. We discuss the latter in terms of discernment i.e. the making of a fairness judgment, and the legitimacy of the judgment made. An argument is then presented regarding the benchmarking of performance of automated systems and critiqued in terms of an erosion of the judicial process. We close with a speculative discussion concerning the comparative legitimacy of the fairness judgments of human reasoning and blackbox systems, and argue that notwithstanding the opacity of human psychology, the processes that appeal to human reasoning sit within a cluster of accountable and public/democratic institutions (something that the developers and deployers of automated systems lack). We conclude with a note on the challenge the question of fairness and automation poses in a world of increasing automated decision making.
Keywords: Automation, Bias, Fairness, AI Ethics, Judicial Reasoning, Public Reasoning
Suggested Citation: Suggested Citation