Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Alternative Fairness and Accuracy Optimization in Criminal Justice

Shaolong Wu, James Blume, Geshi Yeung

TL;DR

The paper analyzes competing notions of algorithmic fairness in criminal justice, highlighting conflicts among group, individual, and process fairness. It proposes a Modified Alternative Definition of Fairness that minimizes a weighted false-negative loss while bounding differences in false-negative rates across protected groups, introducing a tunable tolerance to improve feasibility and potentially predictive accuracy. It then articulates a Three Pillars framework—Need-based decisions, Transparency and Accountability, and Narrowly Tailored Solutions and Definitions—to guide legitimate deployment of risk assessment tools in public systems. The authors discuss data bias, latent affirmative action, and subgroup complexity as critiques, offering actionable guidance for agencies to align technical design with legitimacy and policy goals in risk assessment and related tools.

Abstract

Algorithmic fairness has grown rapidly as a research area, yet key concepts remain unsettled, especially in criminal justice. We review group, individual, and process fairness and map the conditions under which they conflict. We then develop a simple modification to standard group fairness. Rather than exact parity across protected groups, we minimize a weighted error loss while keeping differences in false negative rates within a small tolerance. This makes solutions easier to find, can raise predictive accuracy, and surfaces the ethical choice of error costs. We situate this proposal within three classes of critique: biased and incomplete data, latent affirmative action, and the explosion of subgroup constraints. Finally, we offer a practical framework for deployment in public decision systems built on three pillars: need-based decisions, Transparency and accountability, and narrowly tailored definitions and solutions. Together, these elements link technical design to legitimacy and provide actionable guidance for agencies that use risk assessment and related tools.

Alternative Fairness and Accuracy Optimization in Criminal Justice

TL;DR

The paper analyzes competing notions of algorithmic fairness in criminal justice, highlighting conflicts among group, individual, and process fairness. It proposes a Modified Alternative Definition of Fairness that minimizes a weighted false-negative loss while bounding differences in false-negative rates across protected groups, introducing a tunable tolerance to improve feasibility and potentially predictive accuracy. It then articulates a Three Pillars framework—Need-based decisions, Transparency and Accountability, and Narrowly Tailored Solutions and Definitions—to guide legitimate deployment of risk assessment tools in public systems. The authors discuss data bias, latent affirmative action, and subgroup complexity as critiques, offering actionable guidance for agencies to align technical design with legitimacy and policy goals in risk assessment and related tools.

Abstract

Algorithmic fairness has grown rapidly as a research area, yet key concepts remain unsettled, especially in criminal justice. We review group, individual, and process fairness and map the conditions under which they conflict. We then develop a simple modification to standard group fairness. Rather than exact parity across protected groups, we minimize a weighted error loss while keeping differences in false negative rates within a small tolerance. This makes solutions easier to find, can raise predictive accuracy, and surfaces the ethical choice of error costs. We situate this proposal within three classes of critique: biased and incomplete data, latent affirmative action, and the explosion of subgroup constraints. Finally, we offer a practical framework for deployment in public decision systems built on three pillars: need-based decisions, Transparency and accountability, and narrowly tailored definitions and solutions. Together, these elements link technical design to legitimacy and provide actionable guidance for agencies that use risk assessment and related tools.

Paper Structure

This paper contains 20 sections, 8 equations.

Table of Contents

  1. Introduction
  2. What is Algorithmic Fairness?
  3. Group Fairness
  4. Definitions
  5. Methods for achieving group fairness
  6. Individual Fairness
  7. Process Fairness
  8. The Canonical Definition of Fairness and Its Critiques
  9. The PAC setup for group fairness
  10. Critiques to the Canonical Definition of Fairness
  11. Modified Alternative Definition of Fairness
  12. Individual fairness vs. group fairness
  13. Viable Framework: Three Pillars of Fairness
  14. Need-based decisions
  15. Transparency and Accountability
  16. ...and 5 more sections