How to organize the concepts

  The previous section described what I call the basic level knowledge representation of the ISAAC theory. It is now possible to see how any particular world entity might be represented internally at this basic level. However, this will not be sufficient to allow   creative understanding to occur with respect to novel concepts. In order for novel concepts to be understood, the reasoner will attempt to discover how the novel concepts might fit into the already existing knowledge which they possess. If concepts are simply represented with no higher-level organization, it will be difficult to determine how a new concept should be considered. Instead, a hierarchical organization of the concepts is needed, thereby relating similar concepts.

If a concept can be placed within a larger framework of concepts in a super-concept--concept--sub-concept relationship, then certain functional power is gained. Most importantly, information can be inherited from the higher levels in the hierarchy. If you know, for example, that all mammals have hair and that a dog is a mammal, then you can infer with high probability that a dog will have hair. This idea of hierarchical grouping was seen in the last section with the special :is-a slot in the frame representation. This slot allows the organization of concepts within the knowledge representation system.

The question becomes exactly what ordering to impose. For example, suppose a reasoner wishes to group chairs as a sub-concept under furniture. Should blue chairs, red chairs, and steel chairs be sub-concepts under the concept of chair? From a functional perspective, there is no difference in this approach and in the alternative of having specific chairs under the chair concept which have different attributes attached them. Still, to get the power of hierarchical organization, some grouping must be imposed.

My approach is to impose a top-level organization which mirrors the type and domain breakdown discussed in the first section of this chapter. Nine top level categories exist, formed form the four types and five domains. At the next immediate level are the twenty conceptual categories formed by crossing the types and the domains. This highest-level organization can be seen in Figure 5.

  

Figure 5: Knowledge grid

Beyond this level of description, I impose a loose sub-typing mechanism. The reason is that the preset hierarchical description is largely irrelevant in terms of reasoning. Consider a horse--is a horse an animal or a means of transportation? The answer, of course, is that it is both, although it may not need to always be viewed as both. If a reasoner is trying to discover concepts similar to a horse and is interested in the horse's animal view, then nearby concepts would include zebras and cows. If the reasoner is interested in the transportation view of a horse, nearby concepts should include oxen, cars, and tractors. A preset hierarchy would tend to relate one set at the expense of the other. Therefore, the ontological organization is dynamic, based on the current   functions being considered.