A Framework for Model-Based Adaptive Training

RIME: Imposing Organisation on a Knowledge Base

An explicit problem solving method has been deployed in the current version of R1 using the RIME methodology.The primary problem solving method used by RIME has been derived from work done on R1-SOAR [van de Brug, Rosenbloom & Newell 1986].

R1-SOAR is an experiment in knowledge-intensive programming using the SOAR general problem-solving architecture [Laird et al. 1987].

The major difference between R1-SOAR and RIME is that for RIME the SOAR general problem solving method has been tailored for tasks that can be solved by a strongly recognition-driven problem-solver.

Also, RIME has several auxiliary methods (not described in this section) that are useful in a variety of special circumstances. Problem solving in RIME (as in R1- SOAR) is done in problem-spaces. A problem-space consists of a set of operators and pieces of knowledge that indicate the conditions under which these operators might appropriately be applied. A problem-space is the arena within which part of a complex operator from a parent problem-space is implemented. RIME's problem spaces serve much the same function that subtasks do for R1; each problem space corresponds to a part of the configuration task that expert configurers have named. The difference between R1 and RIME is that RIME's problem solving method imposes an additional level of organisation on its knowledge. RIME's primary problem solving method is defined in terms of the knowledge roles described earlier in this chapter.

Just as R1 always selects the next piece of knowledge to apply from among those pieces of knowledge associated with the currently active subtask, so RIME selects the next piece of knowledge to apply from among the knowledge associated with the currently active problem space. But whereas there was little more to say about how R1 selects knowledge, RIME's problem solving method is substantially better specified. R1 and RIME use essentially the same knowledge to perform their tasks, but because RIME's method makes the roles that knowledge can play explicit, it is easier to talk (and think) about how RIME uses its knowledge and about what knowledge it needs to perform its tasks. Whenever it performs a subtask, RIME always sequences one or more times through a series of steps. Within each step, what action to perform next is never at issue because the method was designed to eliminate all control issues inside steps. Therefore, although RIME ordinarily has many rules that are satisfied at any given time, it never matters in what order the satisfied rules are executed. Within a step, any of the rules that are satisfied can be executed in any order. When no rules are satisfied, control moves to the next step.

Making Adding Knowledge Easier

The problem solving methods used by RIME provides more direction to someone adding knowledge than does R1’s method. This is in part because RIME’s method integrates explicitly defined knowledge roles and in part because the person adding the knowledge can specify the conditions under which one piece of knowledge should be applied in preference to another. Knowledge must be added in such a way that the new knowledge interacts well with existing knowledge.

A limited number of clearly identified knowledge roles make adding knowledge easier because there is less uncertainty in the mind of the person adding the knowledge about how and when knowledge gets used. Each rule is responsible for some aspect of the system’s behaviour. Rules that have the same role of knowledge interact minimally; rules with different roles have well-defined interactions.

This means there is less danger that adding a new piece of knowledge will result in some unexpected interaction with existing pieces of knowledge.

Flower Show
© | | Sitemap