![]() |
|||
![]()
|
![]() |
![]() |
![]() |
2. EXPERT SYSTEMS: THE BACKGROUNDThe first expert systems were developed by researchers in both academic and industrial milieus. Although there were not many expert systems, the best of them clearly demonstrated the potential of the new technology. MYCIN, Prospector, and Xcon are well-known examples. The accomplishments of these expert systems attracted widespread interest and resulted in hundreds of companies applying expert system techniques to a large variety of problems. There were many early successes but, with hindsight, it became apparent that most of the successful expert systems were solving relatively simple problems (Durkin, 1996). By the mid-1980s, expert systems were being built to solve new kinds of problems, many of which were extremely complex. Many of these systems failed to meet expectations, and there was a period of disillusionment. There were several reasons for failure: designers, impressed by the promises of "thinking machines" attempted to solve problems that were beyond the reach even of human experts; other designers took on projects so large that they could not be completed within a reasonable time frame; and yet other designers developed impressive systems that were never used because they failed to meet the requirements of the client (Durkin, 1996). During the late 1980s, confidence returned and the popularity of expert systems began to rise again. The resurgence is probably due to two factors: first, there was a return to straightforward, well-defined applications and consequently projects were completed successfully; second, techniques from disciplines outside the mainstream of artificial intelligence -- software engineering in particular -- enabled the development of larger and more complex expert systems. Another important development that occurred during the 1980s was the emergence of software tools for expert system development. Rather than offering complete expert systems, vendors began to provide expert system shells. A shell is a set of tools that simplify the development of an expert system. A company in need of an expert system no longer had to go through the time-consuming and expensive process of procuring a custom-built expert system; instead, they could acquire a shell and use it to create as many expert systems as they needed. Sales of software tools for expert system development increased from $60 million in 1988 to $140 million in 1993 (Durkin, 1996). Meyer and Curley (1991) provide a two-dimensional classification scheme for knowledge-based systems; see Figure 1. The vertical dimension is the level of knowledge complexity embodied in the system; this includes the depth and specialization of the encoded knowledge, the scope of the decisions made by the system, and the level of expertise needed to solve problems in the domain of the system. The horizontal dimension is the level of technological complexity used to build the system; this includes the nature of the hardware and software used in the construction of the system, the complexity of the user interface, and the use of database, network, user interface, and other specialized technologies.
Bradley and Hauser (1995) identify four categories of expert system within this classification. In Figure 1, expert systems in category 1 are the result of "personal productivity": they embody a relatively small amount of knowledge and little technology. Expert systems in category 2 incorporate extensive knowledge provided by one or more experts. Category 3 systems are like category 1 systems in that they capture only a modest body of knowledge, but they are enriched by advanced computing technology. Finally, expert systems in category 4 combine extensive knowledge and advanced technology. Although a particular expert system does not necessarily fall precisely into one of these categories, the categories provide reference points for many kinds of expert system. Expert systems with high technology content -- categories 3 and 4 in Figure 1 -- are likely to require software engineering techniques in their development. It is expert systems of this kind that are the concern of this chapter.
|
![]() |
|
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. |