In this episode we discuss the problem of how to evaluate the ability of a learning machine to make generalizations and construct abstractions given the learning machine is provided a finite limited collection of experiences.