Embedded training, and many training exercises played against constructive entities can tend to be repetitive, with similar enemy actions occurring at similar times in the exercise. Evolutionary Algorithms with Genetic Programming (GP) and learning classifier systems (LCS) are seen as a means to provide adaptive enemies that not only process their actions using less rigorous behavior, but also adapt their behavior to the student being trained. This has wide ranging implications from learning a particular student's behavior, Course-Of-Action (COA) planning to After Action Reviews (AARs). Results of experiments in this technology will be presented, as well as indication on how to integrate this architecture with a training management system to provide long-term adaptation for improved student instruction. Additionally, the process of installing a machine learning system into a virtual simulator is discussed as it relates to this effort. New techniques had to be developed, particularly in the mapping of the motion commands to the paradigm of the machine learning technology being used, and especially with regards to the real-time update rate that is expected in the interface with a human participant.