While nothing in the foreseeable future appears capable of replacing the requirement of actual aircraft flight in developing Air Combat Maneuvering (ACM) proficiency, the technology is now available to augment actual flight hours with meaningful training from computer-based simulations. At the Aircrew Training Research Division of the Armstrong Laboratory we have been developing several Artificial Intelligence (AI) based approaches to augmenting pilot training in simulation-based ACM. Because air-to-air combat is such a fast moving, complex task, automating (through AI) such tasks as performance measurement and assessment can provide a very important enhancement to a simulation. In addition, using AI techniques to create "smart bogeys" can provide a real boost in the training capabilities of a simulation. In this paper we describe our efforts to create three AI systems (two based upon rule-based production system technology and one based upon artificial neural systems technology) and detail their strengths and weaknesses in providing pilot training in ACM. Particular emphasis is given to the lessons we have learned in applying AI in this rich, fast moving, and complex task environment