ABSTRACT
When commenting on advances in artificial intelligence (AI) and its implications for education, the late Herbert Simon (1987) remarked that “a good tutor plays the role of a knowledgeable friend, who only suggests and advises, leaving the control of the learning situation essentially in the hands of the student” (p. 115). When it comes to using AI to create a computer-based tutor, the nature of the information this “knowledgeable friend” draws on is elusive. Surely, it includes a working knowledge of the subject matter (e.g., mathematics, history, or biology). After all, a friend could hardly be considered a helpful tutor if she knew little or nothing about the subject at hand. A good tutor, as Simon implies, knows more than simply the relevant facts or theories of a particular field or discipline. A good tutor, no doubt, has a strong sense of what the student understands. He or she possesses an understanding of what the student knows about the subject and how he reasons using that knowledge. This often includes an assessment of how much the student has learned, as well as an estimate of the amount and rate of learning taking place. The tutor integrates this information, often in real time, with a dynamic, strategic plan that sequences instructional materials and orders and reorders the presentation of those materials to facilitate learning.
