Advancing Equitable Decisionmaking for the Department of Defense Through Fairness in Machine Learning
Abstract
There is growing concern that machine learning (ML) algorithms can reinforce or exacerbate racial biases in the many sectors in which these algorithms are applied. The U.S. Department of Defense (DoD) is investing in the development of ML methods to assist a wide array of decisions. If the possibility for algorithmic bias is not anticipated and addressed, discriminatory practices analogous to those observed in other sectors may be repeated in DoD. In this report, we aim to provide developers of ML algorithms for DoD with a framework and tools to develop equitable algorithms. We propose a process for developing algorithms that are consistent with DoD's equity priorities. We also introduce the RAND Algorithmic Equity Tool, which allows algorithm developers to enforce equity constraints on predictive algorithms while assessing the inherent trade-offs to doing so. The research reported here was conducted as part of a RAND Project AIR FORCE (PAF) initiative to support diversity, equity, and inclusion (DEI) within the Department of the Air Force. Oversight of the initiative was provided by Dr. Ray Conley.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 14, 2023
- Accession Number
- AD1203428
Entities
People
- Inez Khan
- Irineo Cabreros
- Joshua Snoke
- Marc N. Elliott
- Osonde A. Osoba
Organizations
- RAND Corporation