![]() |
|||
![]()
|
![]() |
![]() |
![]() |
2. STAGES IN THE DEVELOPMENT OF EXPLANATION FACILITIESKnowledge acquisition (KA) represents the process of acquiring knowledge for the purpose of building an expert system. It is the first of three sequential stages -- (1) knowledge acquisition, (2) knowledge validation, and (3) knowledge implementation -- that have to be performed in building an expert system (ES). There exist various methods of knowledge acquisition, some of which are manual and others that are automated. There are also various tests and procedures that can be used to validate knowledge that is acquired prior to its implementation. As well, at the implementation level, there are a variety of knowledge representation formalisms as well as inference procedures available for selection by the knowledge engineer (KE). While it is recognized that the explanation facility is but one critical aspect of the total set of capabilities that constitute an expert system, research into the design question of "how does one go about developing an explanation facility?" has not met with the same level of success as research into the question of "how does one go about developing an expert system?". Part of the reason is that the development process for explanation facilities has received comparatively less attention. Another reason becomes evident if one applies the three stages of developing expert systems to the process of developing an explanation facility, as in Table 1. While significant research has been devoted to all three stages of expert system development -- knowledge acquisition, knowledge validation, and knowledge implementation -- research into the development of explanation facilities has focused only on the Explanation Implementation stage. For example, see the Proceedings of the AAAI Workshop on Explanation (1988) and also Abu-Hakima and Oppacher (1990) for summaries. A review of the literature reveals that issues pertaining to the stages of Explanation Acquisition and Explanation Validation have not been studied to date.
One reason why ES explanations research to date has focused mostly on Explanation Implementation as opposed to the two prior stages of Explanation Acquisition and Explanation Validation has been the belief that all the knowledge required for the purposes of providing explanations, i.e., the explanation base (EB), can be derived from the knowledge base (KB), which is the end-product of the three stages of expert system development, i.e., knowledge acquisition, knowledge validation, and knowledge validation, and represents the knowledge required for the purpose of ES operation. For example, early research into building explanation facilities focused on the manner in which the knowledge base could be represented and implemented to facilitate explanation (Swartout, 1983). Other work focused on the use of inference formalisms that facilitated the provision of explanations and the optimal derivation of an expert system conclusion. To fully understand this belief necessitates a comparison of the relationship between the explanation base (EB) and the knowledge base (KB). The existing belief is that the EB is a subset of the KB, i.e., all the knowledge required for providing explanations can be derived from the total set of knowledge that has been captured and implemented as part of expert systems development. One manifestation of this belief becomes obvious if one considers the explanation facilities that are included in current expert system shells, e.g., VPExpert (Wordtech Systems, 1993). The REPORT command of this shell displays as an explanation the specific production rule in the KB that was last "instantiated" when either of the Why or How explanation options are accessed. These rules, while possibly being of use to knowledge engineers performing systems validation or debugging, as they are designed to facilitate KB functioning, are usually not very useful or comprehensible as explanations to users of expert systems. These explanations that present internal representations of knowledge in an expert system only partially meet the requirements of the How explanation but not those of the Why and Strategic explanations. Considering that these three kinds of explanations, comprising trace, justification, and strategic knowledge, together constitute an acceptable knowledge-based explanation facility, it follows that the KB, while being adequate for the purposes of expert systems problem-solving, is incomplete for the purpose of providing explanations. It is therefore necessary to consider newer models of the relationship between the EB and the KB. This chapter proposes that it is only possible to derive some of the knowledge required for ES explanation from the knowledge base. Significant effort beyond those undertaken to build the knowledge base must also be exerted to complete the explanation base, especially for the justification and strategic knowledge required for providing Why and Strategic explanations. There are other reasons as well that support the argument that the EB should be regarded as a subset of the KB. For example, it is often difficult to predict beforehand the range of users who will ultimately be using an expert system. These users may differ in terms of purpose (problem solving vs. learning), experience (experts vs. novices), etc. It therefore makes sense to decouple the development of the knowledge base from the development of the explanation facility as design of the latter has to include critical user interface characteristics pertaining to whom explanations are to be provided. There are several implications of adopting such models:
|
![]() |
|
Use of this site is subject certain Terms & Conditions. Copyright (c) 1996-1999 EarthWeb, Inc.. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of EarthWeb is prohibited. Please read our privacy policy for details. |