Artificial Intelligence in Medicine 8: 577-580, 1996

In the last decade of artificial intelligence research, qualitative reasoning has proven itself to be a stayer: several large research groups, numerous publications, and an annual international workshop have resulted in a number of comprehensive theories on reasoning about the behaviour of (mostly physical) systems in qualitative terms. One of the most apparent things that the field was lacking, was a proper textbook. 'Qualitative Reasoning: Modeling and Simulation with Incomplete Knowledge' is the first. and yet only, attempt to fill this gap.

Qualitative reasoning is the answer of artificial intelligence to the question of what human common sense reasoning is about. We can reason about (the behaviour of) physical systems like coffee machines, cars, or the human body, without knowing exact formulae or exact quantitative values. Two main reasons for preferring qualitative over quantitative reasoning exist: either no exact formulae or values are known (as applies to reasoning about human physiology), in which case qualitative reasoning is the only alternative, or qualitative reasoning is more informative: if I want to explain someone the working of a refrigerator, the formula to derive the exact pressure in the compressor is not the first consideration. Instead, I may first want to talk about what quantities play a role in the behaviour of a fridge, and how they depend on each other. This second, application area of explanation and tutoring was also one of the main motivations for starting qualitative reasoning research. As the subtitle suggest, Kuipers' book focuses on the first application area of qualitative reasoning with incomplete knowledge. The book, which is entirely set up as a scholars textbook discusses the results of research done at the Qualitative Reasoning Group of the University of Texas in the last decade: the QSIM program and the theories underlying it. In the first five chapters, the ideas behind, as well as the representation and reasoning of QSIM are explained in great detail, supplemented with numerous examples and exercises from both physics and physiology. Without doubt, this part is the best and easiest readable introduction into QSIM written so far. Chapter 6 completes this exposition with three case studies, explaining most ideas and problems with qualitative reasoning (and QSIM in particular).

Chapters 7 through 12 discuss various extensions to the basic QSIM program, dealing with more advanced subjects like predicting the effects of perturbations on the system, reasoning with higher order derivatives, and time-scale abstraction. Most of these extensions aim at reducing the ambiguity that unavoidably results from reasoning with incomplete knowledge, and hence restrict the set of possible behaviours as predicted by QSIM. Chapters 13 and 14 discuss related work, notably the component-connection approach [1] and compositional modelling [2], the latter building on the work on Qualitative Process Theory (QPT) [3]. Finally, the book contains two appendices that may be very valuable for anyone building QSIM models: an overview of all QSIM functions, and a couple of pages on creating and debugging QSIM models.

While reading the book, the question arises who the supposed audience is. For students, the first part of the book provides a clear exposition of QSIM. But QSIM is only one of the main approaches, and the others are dealt with in too less detail to yield a balanced overview of the field of qualitative reasoning as a whole. Given that, as Kuipers also mentions in the preface, few courses are devoted to qualitative reasoning, a 400 page book on QSIM alone might be too much of a good thing. For researchers and engineers who want to apply QSIM, however, the book provides by far the most complete and readable overview of the program, especially due to the numerous case studies and examples provided, and the appendices on QSIM functions, model building and debugging. For medical applications of qualitative reasoning, Kuipers' background in medical reasoning proves useful: a substantial part of the examples are about physiology, and some detailed case studies are presented on equilibrium mechanisms in the kidney. What is particularly missing, however, is a discussion on the use of qualitative reasoning for explanation and tutoring. Especially in medical applications, supporting explanation is a very important feature of qualitative reasoning.

Overall, the book provides a solid, detailed and well-written introduction to qualitative modelling and simulation, albeit from one particular point of view. For researchers who want to learn more about the developments in the field of qualitative reasoning, this book is not the whole story but a major chapter.

Department of SWI
University of Amsterdam
Roetersstraat 15
1018 WB Amsterdam
The Netherlands


[1] J.H. de Kleer and J.S. Brown. A qualitative physics based on confluences. Art. Intell. 24 (1984) 7-83.

[2] B.C. Falkenhainer and K.D. Forbus. Compositional modeling: finding the right model for the job. Art. Intell. 51 (1991) 95-143.

[3] K.D. Forbus. Qualitative process theory. Art. Intell. 24 (1984) 85-168.

[Book home]