Teaching a computer to teach

The long-range goal of computer-assisted instruction (CAI) should be to construct computer programs that interact with a student as a human tutor does. In order to make computers act as intelligent tutors, we must consider three aspects of teaching; (1) A tutor makes use of many different knowledge sources. (2) A tutorial session is a kind of problem-solving situation in which the tutor is trying to solve the problem of imparting understanding to the student. (3) There is an analogy between the process of communication and those of teaching in that both require the participants to make a number of inferences about each other’s states and responses. A tutor designs an “instructional act,” anticipating how the student will interpret and learn from it, and the student interprets and learns from the instruction in a manner that depends on his hypotheses about the intent of the tutor. The paper reviews the relevant literature and discusses how we will incorporate such ideas into CAI systems.