This paper is a transcription from memory of a short talk that used overhead projector slides, with musical examples played on an Apple Macintosh computer and a Yamaha CX5 synthesizer. The slides appear in the text as reduced "icons" at the point where they would have been used in the talk. The paper concerns ways in which artificial intelligence techniques could be applied to enable computers to make some kind of useful input into teaching musical performance skills. The overall plan of the paper falls into three main parts, namely: "What should be taught?""How could it be taught?" and "Could it and should it be done?" A particular psychological characterization of musical performance is discussed as a means of answering the first question: what should teachers be trying to teach? The question, how could teachers teach it, then becomes a problem of finding a knowledge representation which both models these cognitive processes and is suitable for the teaching strategies desired. The question as to whether the proposal could be done is a matter of having the needed human and technological resources--and of stretching them to the limit in anticipation of future advances. As to whether it should be done, what is needed is a reasonable assurance that the research will make some useful contribution both to music education and to general design principles in compouter aided instruction. Design considerations for the technology for creating the appropriate environment to develop music performance skills are examined and practical implications of the human and technological resources used for computer design and for music education are explored. An appendix contains questions, suggestions, and answers that were raised in the original discussion, together with a short list of works referred to in the paper. (DB)