Intelligent computer-aided instruction: a survey organized around system components

This paper presents a survey of the issues and previous research in Intelligent Computer-Aided Instruction (ICAI). Unlike previous surveys of ICAI, which have been organized around the seminal ICAI systems, this one concentrates on the contribution of each eeort to our understanding of the various components of ICAI systems. Along the way, comparisons are made between ICAI and CAI, although a CAI straw man is used due to the focus of the paper. Various learning scenarios are discussed, including computer coaches, gaming environments , mixed initiative dialog, Socratic tutors, articulate experts, interactive simulation, and discovery learning. Various forms of knowledge representation are discussed along with relevant issues and examples. Several techniques for student modeling and diagnosis are presented, as are their respective advantages and disadvantages. Pedagogical knowledge, its role in ICAI, and several examples are highlighted. The evolution of discourse management techniques for ICAI is outlined. Techniques for the automatic generation of problems from a general base of domain knowledge are presented. Finally, issues in the design of user interfaces for ICAI systems are brieey discussed. The purpose of this paper is to place previous ICAI research in a form which is appropriate for use by future designers of ICAI systems. Although computer-aided instruction (CAI) 1 has been around for some time now, there has been much discontent with the current state of educational software. Traditional \frame-oriented CAI" requires teachers to fully specify presentation text, all questions and their associated answers, and a strict ow of control through the course, allowing at best diierent branches to be taken based on the student's pre-enumerated possible responses. This style of teaching has often been referred to as \electronic page turning." A step in the right direction was made with generative CAI. In these systems, techniques such as textual templates, problem-generation grammars, and random numbers were used to generate problems for the student. Still, the systems lacked the ability to adapt to the student or relate his incorrect answers to the correct ones, and the problems were usually limited to simple drill-and-practice questions 40]. 1 Because this paper concentrates on ICAI, a CAI straw man is used for comparison. In fairness to CAI researchers, it should be noted that the problems with CAI were well known, and many practitioners developed ingenious solutions for them. Nevertheless, our characterization of CAI is adequate for the purposes of this paper.