The European Journal on Artificial Intelligence

The following interview is the first of a series on AI and education. In this series four main topics are addressed:

• Theories of learning and cognition

• The relation between such theories and education

• The role of computers in education

• The future of AI and education

About Martial Vivet:

Martial Vivet is professor of informatics at the University of Maine in Le Mans in France. His main focus is the design of learning environments which take the context into account. His background is in mathematics. He began research in Artificial Intelligence in the early seventies with Jacques Pitrat at the University of Paris. Martial Vivet has worked for many years in the area of teachers training in mathematics. In 1977 he was involved in the group which introduced LOGO to France. Surrounding LOGO he now develops microrobots based microworlds that are used to train students in technology (mechanics, electronics, programming). He is involved in several European projects (COMETT and DELTA). In October 1988 he managed the second European meeting on Intelligent Tutoring Systems.

Certain problems associated with knowledge acquisition are identified and examined in this paper. We review a variety of methodologies and tools designed to address these problems and then argue that there is a strong case for a preliminary knowledge analysis or domain phase of KBS development. This phase facilitates subsequent design, development and maintenance phases. The details of our domain characterisation are not expounded upon in this paper. The paper concludes with a suggestion of a re-examination of some of the central metaphore acquisition.

In this paper we develop a conceptual modelling framework for knowledge-level reflection (KLR), i.e., the modelling of tasks that require a self-representation of a knowledge system's own object-level problem solving tasks. This framework builds upon the KADS methodology for knowledge acquisition and design of knowledge systems [Hayward et al., 1987, Wielinga et al., 1991].

We argue for the separation of object and reflective problem solving levels and a self-representation that is distinct from the object-level because it is selective, specialised and knowledge oriented, i.e., it is a knowledge-level model congruent with the KADS conceptual model of the object system. As an example we describe a conceptual model for competence assessment and improvement in Office Plan, a configuration system for office space allocation. A broad comparison with notions of reflection in logic and computational reflection clarifies the distinctiveness of our notion of knowledge level reflection and investigates some of the architectural options that are open for its realisation in knowledge systems.

The aim of this paper is to present a novel approach to the problem of focusing abductive (model-based) diagnosis. The approach we propose is based on the use of compiled knowledge and, specifically, on the possibility of associating with each entity in a model a necessary condition for the presence of the entity itself. Such conditions embed the problem solving strategy and their evaluation on the data characterizing the problem to be solved allows us to prune the search space, yet preserving the completeness of the abductive process (in other words, only useless search is avoided). The use of compiled knowledge thus allows us to mitigate the problems arising from the computational complexity of the model based approach. The final part of the paper is devoted to a comparison of our approach with other ones and to a brief discussion on the role of knowledge compilation in model-based diagnosis.

It will be shown why expert systems should no longer be designed as autonomous problem-solvers but as cooperative systems. A methodology for designing these cooperative expert systems will be presented that combines methods from cognitive engineering with a popular knowledge acquisition approach, namely the the use of task level frameworks. A real world example from the domain of technical troubleshooting will be used to illustrate the four main steps of the methodology.

In the March issue of AI Communications we published an excerpt of the invited talk of Bill Clancey at the DELTA-conference in The Hague (The Netherlands). We invited our readers to react to his ideas. You will find Clancey's reply on p. 108.