Adjudicating with Inscrutable Decision Tools

in MACHINES WE TRUST: GETTING ALONG WITH ARTIFICIAL INTELLIGENCE, Marcello Pelillo and Teresa Scantamburlo (Eds.) (MIT Press, 2020 Forthcoming)

NYU Law and Economics Research Paper No. 20-36

NYU School of Law, Public Law Research Paper No. 20-52

29 Pages Posted: 24 Jul 2020 Last revised: 6 Nov 2020

Date Written: December 20, 2019

Abstract

Machine learning models are increasingly used in making decisions important to peoples’ lives. Often, however, humans have difficulty understanding how these automated decision tools arrive at their assessment. This inscrutability has drawn the attention of data scientists, legal scholars and others, but the discussion so far has focused on explanations aimed at decision subjects. This chapter highlights the previously neglected importance of explanations to human adjudicators, who generally retain ultimate responsibility for significant decisions. It approaches this issue by comparing inscrutable automated decision tools to the rule-like decision criteria that adjudicators have traditionally implemented, often in combination with more standard-like criteria. The chapter analyzes the novel difficulties that inscrutable automated decision tools create for adjudicators and concludes with some suggestions about how to address them.

Keywords: machine learning, algorithmic decisionmaking, adjudication, explainability, interpretability

Suggested Citation

Strandburg, Katherine J., Adjudicating with Inscrutable Decision Tools (December 20, 2019). in MACHINES WE TRUST: GETTING ALONG WITH ARTIFICIAL INTELLIGENCE, Marcello Pelillo and Teresa Scantamburlo (Eds.) (MIT Press, 2020 Forthcoming), NYU Law and Economics Research Paper No. 20-36, NYU School of Law, Public Law Research Paper No. 20-52, Available at SSRN: https://ssrn.com/abstract=3546380

Katherine J. Strandburg (Contact Author)

New York University School of Law ( email )

40 Washington Square South
New York, NY 10012-1099
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Downloads
97
Abstract Views
510
rank
332,099
PlumX Metrics