Advancing Equitable Decisionmaking for the Department of Defense Through Fairness in Machine Learning (Summary Report)
Abstract
Machine learning (ML) algorithms are increasingly used as an aid to human decision-making. However, there is a growing recognition that the use of ML algorithms may reinforce or exacerbate human biases, thereby perpetuating inequities. This situation is commonly referred to as algorithmic bias. The U.S. Department of Defense (DoD) is investing heavily in the development of ML algorithms to assist in many decision-making processes. At the same time, DoD has a strong stated interest in promoting diversity, equity, and inclusion (DE and I) at all levels of the organization. The goal of this report is to provide policymakers and developers of ML algorithms with a framework and tools to produce algorithms that are consistent with DoDs equity priorities. This report represents part of a larger effort to advance equity in DoD. Although predictive ML algorithms are deployed in some sectors within DoD including intelligence and surveillance ML algorithms are in the preliminary stages of development and are not at this time deployed in decision making processes in the personnel space, where DoD has expressed equity goals. Despite this, we observe a growing interest in using ML algorithms as part of personnel decisions, as evidenced by the prototype tools developed in this space. Therefore, the utility of this report is primarily to preempt the possibility of algorithmic bias in eventual personnel decision-making applications within DoD rather than to address existing instances of algorithmic bias.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2023
- Accession Number
- AD1203429
Entities
People
- Inez Khan
- Irineo Cabreros
- Joshua Snoke
- Marc N. Elliott
- Osonde A. Osoba
Organizations
- RAND Corporation