RuleWorks

A Framework for Model-Based Adaptive Training

Expertise Classification

Expertise in training is considered the ability to exhibit the correct behaviour in a given situation [MOBIT 1994, W8]. There are two main aspects to expertise classification. First the scope of the situation and second the level of required behaviour. For Proto-MOBAT, the scope of a situation is defined in problem spaces [Newell 1981] and the level of behaviour is based on work by Jens Rasmussen [Rasmussen 1986]. The behaviour classification is in 3 levels: skill-based expertise, rule-based expertise and model-based expertise. This section starts with the Proto-MOBAT problem space structure and then relates the problem spaces to levels of desired trainee behaviour.

The scope of a situation in Proto-MOBAT, as in SOAR [Laird et al. 1987], is bounded by defining the behaviour as taking place within a problem space. The problem space provides a useful means to limit the arena of a training problem. A problem space has a goal (and subgoals) and a context within which problem solving takes place. A problem space goal and its context of problem solving within the subject domain can be derived from a training objective. The breakdown of training objectives is shown in Section 3.4. A problem space map is a structure (possibly a hierarchy with sub-problem-spaces) based on a trainee’s training objectives. The training objectives are reflected in a problem space structure as shown in Figure 4-5.

Proto-MOBAT Problem Space Map

Figure 4-5 Proto-MOBAT Problem Space Map

The problem space structure is created to define the situations within which a trainee is expected to achieve a certain level of performance. It should be noted that a problem space map is not a rigid structure. As shown with RIME (see Section 2.5.5), the knowledge structures for implementing tasks in problem spaces are small and re-usable. For example, without changing the knowledge structures, the conceptual problem space map for R1-SOAR was easily re-configured several times (i.e., with major changes in viewpoints [van de Brug, Rosenbloom & Newell 1986] ). The actual problem space map during run time is dynamic as tasks can cause automatic subgoaling to any other problem space. The realisation of domain models in problem spaces for Proto-MOBAT is further described in Section 4.9.

For expertise classification, the desired level of trainee performance is defined within a problem space. Jens Rasmussen created a model of human performance in problem solving situations which describes the different classes of operation required under different situations within a complex dynamic environment. He defines three human behaviour classes as:

  1. skill-based behaviour;
  2. rule-based behaviour; and,
  3. model-based behaviour.

Clearly, the expected trainee performance level depends on the target audience. In Proto-MOBAT all sub-problem-spaces for plant operation are targeted to train plant operators to become skilled in plant operating procedures. The goal expertise level for these operators is skill-based expertise. The sub-problem-spaces for circuit board debug are aimed at technicians to become competent problem solvers in finding faults and identification of problem causes. The goal expertise for these technicians is rule-based expertise.

*
Flower Show
  *  
© RuleWorks.co.uk | | Sitemap