Machine Learning Fairness in Justice Systems: Base Rates, False Positives, and False Negatives

TL;DR

This paper explores the challenges of implementing fairness in machine learning models within justice systems, focusing on error tradeoffs like false positives and negatives across racial groups and emphasizing stakeholder engagement.

Contribution

It analyzes the implications of error disparities in justice ML applications and discusses practical, stakeholder-inclusive approaches to fairness standards.

Findings

Higher false positive rates for one group and false negatives for another can lead to fairness issues.

Tradeoffs in error rates are complex and often require stakeholder dialogue.

Computational methods alone cannot fully resolve fairness tradeoffs.

Abstract

Machine learning best practice statements have proliferated, but there is a lack of consensus on what the standards should be. For fairness standards in particular, there is little guidance on how fairness might be achieved in practice. Specifically, fairness in errors (both false negatives and false positives) can pose a problem of how to set weights, how to make unavoidable tradeoffs, and how to judge models that present different kinds of errors across racial groups. This paper considers the consequences of having higher rates of false positives for one racial group and higher rates of false negatives for another racial group. The paper examines how different errors in justice settings can present problems for machine learning applications, the limits of computation for resolving tradeoffs, and how solutions might have to be crafted through courageous conversations with leadership, line workers, stakeholders, and impacted communities.