A Framework for Model-Based Adaptive Training

Methodology Review - Summary

This chapter reviews the initial methodology for the MOBAT framework with:

  • a training system specification and modelling methodology,
  • an adaptive intelligent system realisation methodology and,
  • a flexible life-cycle methodology.

The ITSIE project [ITSIE 1992, D7] and MOBIT project [MOBIT 1994] provide an initial training specification methodology with three basic modes of instruction (rote, inductive and deductive), three domain model types (procedural knowledge, associative knowledge and principled knowledge) and seven modelling dimensions (scope, ontology, generality, perspicuity, precision, accuracy and uncertainty).

This methodology includes a classification of trainee performance, based on work by [Rasmussen 1986], using three goal expertise levels (skill-based expertise, rule-based expertise and model-based expertise).

The SOAR project provides an adaptive realisation methodology for model selection and model switching using a preference mechanism [Laird, Newell & Rosenbloom 1987].

Recent object oriented software engineering techniques provide a flexible life-cycle methodology called concurrent engineering [Nichols 1992, Odell 1992]. This technique can be used for MOBAT application because the training system architecture [Slater, Brown, van de Brug & Brown 1984] is object-oriented and the use of training agents allows a different knowledge base to be added a step at a time.


Surprisingly few methodological approaches to building Intelligent Training Systems exist. Most ITS’s have been developed within research organisations as experiments. The MOBAT framework is extending the ITSIE methodology which is reviewed in this chapter.

Section 3.3 reviews the main elements of the ITSIE training methodology for specification of levels of performance, types of learning and modelling dimensions. A mechanism to select and switch models was not addressed in ITSIE.

In Section 3.4 a model switching approach is proposed based on the SOAR mechanism with preferences.

When stripping away the terminology, tools and techniques that make an AI application unique, the general work area is software engineering. Successful software has to be built in a disciplined way. Lifecycle methods can provide a framework for ensuring software quality. Section 3.5 considers general Software Life Cycle Methods.

Finally, to build on the ITSIE and MOBIT methodology, Section 3.6 summarises a set of questions within the proposed MOBAT framework.

Flower Show
© | | Sitemap