Machine learning (ML) is all around us. From virtual assistants to automated testing, it provides capabilities that we now depend upon. Yet, while ML enables significant advantages in organizing and differentiating complex data in many domains, it has not yet made a significant impact on US Department of Defense (DoD) training systems or training methods. The question is not should ML be integrated into DoD training, but what techniques are efficient, effective, and feasible? The answer to these questions, and many others, are critical to developing the next generation of trainers and live, virtual, and constructive (LVC) training capabilities. The potential impact is vast, but well researched and intentional integration is crucial, as the costs will be significant. This paper describes ML and discusses emerging/innovative technological ideas on integrating ML into two categories of training systems. First are multiperson training simulators, such as convoy trainers, which - with the injection of ML - could realize decreases in training time and increases in proficiency. Second, the analysis expands these insights into the context of LVC training simulations. For LVC, it summarizes precursor semi-automated systems, highlights current ML applications, discusses the roles ML could play in future LVC environments, and describes how these systems could be wrapped in advanced training delivery approaches. This paper concludes with thoughts and considerations regarding ML topics that are critical in simulation-based training (uncertainty, metrics, DoD/commercial interaction, and data) and then recommends possible next steps.