COMPARISON WITH CONVENTIONAL DATA PROCESSING

Anyone who has any experience with a conventional database understands that either organizing a database or writing new applications on top of an existing database is neither easy nor cheap.

Moreover, once the main applications that drove the installation of a database system are installed, there is often a demand to build additional applications for a small group of users. The ability to do this is limited by programmer productivity and ROI of users. Hence, why are database systems easily accepted and knowledge base systems having such a hard time? Some of the differences between the two technologies are summarized in Table 7.3. We add the following observations:

1. The data intended for database systems are relatively easily acquired, having previously existed on paper. Knowledge acquisition, on the other hand, is often the major bottleneck in developing a KBS.
2. The algorithms to manipulate the data are known: payroll, billing, accounts receivable, inventory cash analysis. The operations on a knowledge base are usually known (e.g., simple backward-chaining search through a rule base), but often not codified.
3. The utility of computerizing a database application is easily discerned in advance (ROI can be developed ahead of time). The same is usually true for knowledge-based applications, although the estimate may be more approximate due to lack of experience with this relatively new technology. The ROI on actual deployed knowledge-based systems, however, has been very good.
4. Once the separation of the data in a database from the program is made, it is easy to imagine other reports of value that could be obtained (e.g., tracking fast-selling items, receivables aging, customer profiles), so that developing a corporate database is a worthwhile endeavor. Although knowledge bases are also separated from the inference engines that use them, a knowledge base typically is not reusable. This problem is the subject of much research in the U.S. and in Japan (e.g., Professor Mizoguchi's work at Kyoto University).
5. Initial database applications (e.g., airline reservation systems) are useful to many employees. By contrast, there are usually few users of a KBS. (The same can be said of many database applications after the main ones have been done.) Hence, one sees an emphasis in the database world on query languages so that end-users can write their own applications and generate their own reports. A similar trend has not yet taken hold in the KB applications world, though task-specific shells are an attempt to provide a similar capability.

Table 7.3
Databases Versus Knowledge Bases