Acquiring Maintainable AI-Enabled Systems

Abstract

The Army and other services are quickly entering into an age where many, if not all, acquisitions programs will need to contend with acquiring Artificial Intelligence (AI)-enabled systems. While there has been research on how to acquire the data or model for an AI-enabled systems, sustainment considerations have been overlooked. Given the importance of sustainment for any acquisition program of record – both in terms of cost and in terms of program effectiveness – it is imperative that the Army, and the rest of DOD, plan for AI-enabled system maintenance. To address this gap, this paper proposes a framework and practices that draw on best practices from industry, program maintenance, and Machine Learning Operations (MLOps) to integrate AI maintenance into a product support strategy and Life Cycle Sustainment Plan. The framework outlines necessary components for sustainable AI and considers varying levels of maintenance to reduce operation and sustainment costs.