This paper is concerned with knowledge-based knowledge acquisition, with particular reference to data analysis. It discusses the prototype system Student, which was designed to allow expert statisticians working alone to build and test strategies. One successful part of Student is knowledge-based knowledge acquisition; this is defined as the approach to knowledge acquisition that restricts the domain of knowledge that can be acquired and builds a conceptual model of the restricted domain. Student’s conceptual framework was induced from REX, a consultation program in regression analysis that was constructed using standard expert system techniques.
The author discusses the shortcomings of the knowledge-acquisition methods used in REX and considers their contributions to future work. He describes and criticizes the knowledge-acquisition techniques used in Student, and discusses issues related to the use of knowledge-acquisition. Finally, he considers the generality of knowledge-based knowledge acquisition.
This paper lacks scientific method. It is littered with speculative phrases such as “possibly practical,” “based . . . on a feeling,” “will probably want,” “seems like a technique,” “is apparently a useful idea although it has not been tested,” “probably the most efficient,” “possibly pointing out,” and “my impression is.” These phrases give the reader confidence in neither the analysis that has been performed nor the unsubstantiated conclusion that “the method should be usable in domains other than data analysis.”