A Framework for Model-Based Adaptive Training

Training System Foundation - Summary

The foundations upon which the MOBAT specification framework is based are varied. Analysis in this chapter provides the substructure for the MOBAT specification and realisation methods. A number of ideas from the SOAR research [Laird et. al. 1987] have been adapted including the concepts of problem spaces [Newell 1972] and the basic problem solving approach (called the universal weak method in SOAR).

Knowledge acquisition research has identified a variety of problem solving methods [McDermott 1988] which has contributed substantially to the problem solving methods adopted in the MOBAT specification framework (e.g., propose, evaluate and apply methods [van de Brug, Bachant & McDermott 1985].

The QUIC project [Leitch & Gallanti 1992] provides the MOBAT generic task primitives for interpretation, identification, prediction, decision and execution.

The MOBAT specification methods recognise knowledge level and symbol level design [Newell 1981]. MOBAT is encouraged by and identifies with the work on components of expertise [Steels 1990].

The R1-SOAR project [van de Brug, Rosenbloom & Newell 1986] provides experimental results which have been adapted for MOBAT transfer of learning methods (across-trial, within-trial and across-task).


The purpose of this chapter is to introduce the foundation of the MOBAT specification framework and to discuss related research work. There is much work, both theories and systems, that are closely related to the framework presented in this research. The analysis in this chapter is not restricted to Intelligent Tutoring and Training Systems. Much of the work the MOBAT framework is building on is under development in other Artificial Intelligence areas.

The most relevant AI sub-fields are: Intelligent Tutoring Systems, Intelligent System Architectures, Knowledge Based Systems Analysis and Design, Knowledge Acquisition and Machine Learning. Each of these fields draws upon inter-disciplinary work in other areas. For this work, the main inter-disciplinary areas are: educational psychology, cognitive science and software engineering.

The MOBAT specification framework is based on explicit models of cognitive processes and qualitative models of the physical device, product or process on which the training is to be performed. Section 2.3 introduces a model-based approach with the SOAR theory [Laird et. al. 1987] and the concept of problem-spaces [Newell 1972].

As there is a considerable amount of work in other knowledge based system domains (e.g., the QUIC project [Leitch & Gallanti] and KADS Methodology [Tansley & Hayball 1993]), Section 2.4 reviews related research in KBS analysis and design. An important contribution to the MOBAT specification framework has evolved from the work by knowledge acquisition researchers during the 1980’s (e.g., [Marcus 1988, McDermott 1988, Steels 1990].

The specification of problem solving methods [Bachant & Soloway 1989, van de Brug et al. 1985], is introduced in Section 2.5. An intelligent training system needs computer models of learning. Section 2.6 introduces the foundation for the MOBAT specification based on learning techniques from the SOAR project [Laird et. al 1987, van de Brug et al. 1986].

Key concepts introduced in this chapter are revised and extended in the MOBAT specification framework.

Flower Show
© | | Sitemap