Problems in Indexing Problem Solving Methods

Organizing PSM into a some library for reuse reveals two related problems that have thus far received little explicit attention: What is the scope of reuse of a PSM? How distinctive/similar are PSM? In considering the first question, the working hypothesis in constructing such libraries, and in particular the CommonKADS library, has been that PSM are task specific, and that therefore the major indexing is by task types, and in particular by some taxonomy of tasks; secondary indexing is by other assumptions that are inherent to the PSM, in particular related to domain knowledge features. In [Breuker, 1994b] it was argued that such taxonomy is impossible to construct and that a suite of dependent generic types of problems is a better representation to characterize what a PSM should yield. In this paper some revisions of this suite are proposed. Although this enables a clearer functional indexing of PSM (i.e. by their competence), it is of only tertiary importance where it concerns the indexing of the real, operational PSM that heuristically exploit experience based domain knowledge. The primary indexing proposed here is by classifying the PSM by the way they generate solutions: by classification or by construction, confirming views held by [Clancey, 1985; deVelde, 1988; Puppe, 1993] 1 What is the scope of a PSM? In developing notions about the reuse of problem solving methods (PSM) the emphasis has been on the question of what’s (in) a PSM, rather than on what it can be reused for, i.e. what is the scope of reuse of a PSM. The intuitive answer has been for domains and tasks that look largely similar as for which the PSM was originally conceived, e.g. [McDermott, 1988]. By applying a PSM to a new domain it becomes evident what domain characteristics have to be present to have the PSM being able to perform the task. In other words, the assumptions on which a PSM is based become experimentally explicit. The scope of KNACK is far more limited than that of MOLE (“cover &