Can a computer learn to fly a helicopter simply by observing the actions of a pilot? Should we use neural networks, fuzzy controllers or some other approach? What are the best lessons to teach the computer? Are humans really good teachers or do they use queues, information and experiences that aren't available to the observing learning system? How does this all compare to simply hand-tuning conventional control systems and is the hand-tuning really as hard as people say? These are the questions that this work explores and should interest anybody in the area of machine learning, robotics, or computer science.