The three papers in this book are reprinted from the journal Artificial Intelligence. The purpose of this collection is to reach a wider and more general audience with information on the current status of research work in intelligent tutoring systems. Major changes in approaches to intelligent tutoring systems and the lessons of artificial intelligence research in other areas have resulted in many advances and insights about intelligent tutoring systems.

In their preface, Clancey and Soloway recount the general characteristics of early intelligent tutoring systems research during the 1970s. They identify the major trends of the 1980s and note that the directions of intelligent tutoring systems research are not always the same as those of research in other artificial intelligence areas. Papers were selected for inclusion because of their contribution to intelligent tutoring systems, their articulation of goals and comparisons with other architectures for the same goals, and their ability to support relevant psychological experimentation or their ability to demonstrate generality. The editors list new perspectives for the field and suggest shifts in practice for future development.

The first paper, by Anderson, Boyle, Corbett, and Lewis, focuses on cognitive modeling and tutoring. For these researchers, intelligent tutoring systems are a domain for investigating theories of cognition. ACT* and PUPS are architectures for cognitive modeling that have been developed to embody specific theoretical commitments for the models. Three particular tutors (for LISP, geometry, and algebra) are described in the paper. Each of these presents features that require special handling. The authors introduce the notion of model tracing to capture the conversion of theoretical models to tutoring vehicles. Performance models and learning models are specified. Ideal models are used to guide the learner, and error models are used to detect and correct errors. The authors present results on the use of each of these tutors in various learning environments. They also discuss the modules of the tutor and the issues that result from such implementations.

The second paper, by Johnson, deals with the diagnosis of errors in a novice’s Pascal program. The author is concerned with students learning to program and finding debugging to be a hindrance to learning. The notion of intention-based analysis is introduced in contrast to dataflow analysis. The PROUST system is presented as a way to provide such intention-based analysis. PROUST contains a knowledge base of programming plans and knowledge of common student errors. Johnson contrasts PROUST with other programming advisors, both intention-based and not. The paper contains a description of PROUST’s goal decomposition processes and intelligent tutoring systems handling of plan differences. Empirical results are presented and interpreted. The author mentions that future investigation should include testing PROUST on other programming languages and attempting to incorporate a tutoring component.

The third paper, by White and Frederiksen, investigates the use of causal model progressions in creating an intelligent learning environment. The authors’ approach builds on research in qualitative modeling and the role of qualitative reasoning in novice and expert problem solving. Expertise is believed to be captured in a set of mental models that contain pertinent conceptualizations. Movement from novice to expert is viewed as a series of model transformations. The novice’s simplistic model is changed through problem solving and learning strategies into a more sophisticated model capable of solving more advanced problems. Models are classified by perspective, order, and degree of elaboration. The QUEST system is presented. QUEST deals with qualitative understanding of electrical systems and troubleshooting. The models are designed to facilitate the decomposition of models into simpler models and to allow causality to be transparent. A small sample of students with no background in physics was used to test some of the features of QUEST. The results were encouraging, especially in the area of troubleshooting skills.

The editors achieve their purpose of presenting major works of recent vintage. It is questionable whether this approach will reach the intended wider audience. The papers are of the highest quality and are sufficiently diverse to provide a significant exposure to the themes and techniques of the field. Their quality will also be a source of difficulty for the nonexpert reader, however. The papers are detailed and require a fairly sophisticated background to be appreciated for what they contribute. References are provided for each paper separately. A comprehensive reading list is not included. The book offers few pointers to other current or recent works that are representative of other activity in intelligent tutoring systems. An extensive index to the entire volume is provided.