Principal findings and recommendations of a two-year study of machine-aided knowledge acquisition. The report discusses the transfer of expertise from humans to machines, as well as the functions of planning, debugging, knowledge refinement, and autonomous machine learning. The research method emphasizes iterative refinement of knowledge in response to actual experience. A machine's "knowledge" is acquired from a human, who provides concepts, constraints, and problem-solving heuristics to define some minimal level of performance. Semiautomatic methods convert the initial knowledge into a working program whose resulting behaviors can be used to diagnose problems and design refinements. Methods formulated here may reduce or eliminate much of the human involvement currently required in this process. The approach is illustrated by application of the paradigm to the game of hearts. Recommendations suggest increased emphasis on core research problems standing between current technology and the capability of automatic knowledge programming and refinement.