The control of understanding

   The four tasks of reasoning act together to produce an understanding of all concepts entering the reasoner. These four tasks must be synchronized in order to work smoothly together. One possibility is simple linear control. That is, memory retrieval is called; if it fails, then analogical mapping is tried; upon its failure, base-constructive analogy is attempted; finally, problem reformulation is used if the other three tasks have failed to provide an adequate understanding; and the cycle repeats. However, this single path through the four tasks is not the only one possible. For example, a reader may fail at memory retrieval but receive a cue that problem reformulation is needed, skipping the analogy steps completely. An analysis of possibilities reveals the following:

 
• Memory retrieval is the entry point into the understanding process. If the memory retrieval is successful, then understanding is successful. On the other hand, if memory retrieval fails, two possibilities could be the result. First, an analogue (or set of analogues) may have been discovered; at this point, analogical mapping can be executed. Second, evidence may have been returned that enables the   reasoning supertask to determine that problem reformulation is needed.

   

• The task of analogical mapping is followed by a memory retrieval request if it is successful. If the mapping fails, the   reasoning supertask can elect to attempt to dynamically create a new base with base-constructive analogy. Or, it might be the case the failed analogy indicates that a problem reformulation is required. Finally, the reasoning supertask can decide to attempt another memory retrieval upon a mapping failure.

   

• If base-constructive analogy fails, it can be followed by a new memory retrieval attempt, thereby starting a new cycle. Or, the failure can indicate that a problem reformulation needs to occur.

   

• Problem reformulation, if unsuccessful, can be followed by memory retrieval.

      This view of the understanding process interprets it as a search process (e.g., [#!plan:newell!#]) with the operators being the four tasks. The possible search space can be compactly represented due to the redundant nature of possibilities, as seen in Figure 29.

  

Figure 29: Search space for understanding

Given the overall cyclic nature of the algorithm, some method should constrain the amount of understanding; otherwise, it is possible that   bizarre (i.e., non-useful) understandings could result. One possibility is that the researcher can enforce bounds on the number of cycles through understanding which is allowed, but this seems to restrict the outcomes in an unreasonable fashion. There is no way to know a priori how much reasoning should be ``permitted'' in every reasoning episode. Alternatively, arbitrary choices could be made. However, both of these options have serious problems (see, for example, [#!acm:birnbaum!#]). To avoid both the arbitrariness and unreasonably tight control, I suggest that three factors combine to provide this necessary limit: the satisfaction of the reasoner, the interest level of the reader, and the knowledge ontology.