Using problem-solving methods to impose structure on knowledge

The author outlines a taxonomy of methods that identifies some of the discriminating characteristics of the kinds of methods expert systems use and that suggests how methods can be mapped onto tasks. It is suggested that half-weak methods are an important class of methods because they can have a quite broad scope of applicability, but provide substantially more help than the weak methods do in specifying what task-implementation knowledge needs to be collected to perform a particular task and how that knowledge is appropriately represented. It is claimed that it should be possible to devise a set of inference engines (half-weak methods), each of which defines the roles that the task-implementation knowledge it requires must play and the form in which that knowledge is to be encoded. Two half-weak methods, MOLE and SALT, are considered in support of these points.<<ETX>>