Exploring the Feasibility and Utility of Machine Learning - Assisted Command and Control: Volume 2, Supporting Technical Analysis

Recent high-profile demonstrations of artificial intelligence (AI) systems achieving superhuman performance on increasingly complex games along with successful commercial applications of related technology raise the questions of whether and how the U.S. Air Force can use AI for military planning and command and control (C2). The potential benefits of applying AI to C2 include greater decision speed, increased capacity to deal with the heterogeneity and volume of data, enhanced planning and execution dynamism, improved ability to synchronize multimodal effects, and more efficient use of human capital. Together, the technology push prompted by recent breakthroughs in AI and the market pull arising from emerging C2 needs have prompted the Air Force and the Department of Defense (DoD) to identify AI as a strategic asset. In 2019, the Air Force Research Laboratory, Information Directorate (AFRL/RI) asked RAND Project AIR FORCE (PAF) to examine and recommend opportunities for applying AI to Air Force C2. The research project Exploring the Near-Term Feasibility and Utility of Machine Learning-Assisted Operational Planning was conducted in PAFs Force Modernization program to address this question. A second project was conducted in parallel to examine the separate but related topic of complexity imposition. This report presents the primary result of the study on AI: an analytical framework for understanding the suitability of a particular AI system for a given C2 problem and for evaluating the AI system when applied to the problem. We demonstrate the analytical framework with three technical case studies focused on master air attack planning, sensor management, and personnel recovery (PR).