For the past several years, my colleagues and I have been doing work with expert systems and the representation of the knowledge in those systems. We have previously reported on the effectiveness of K-trees, an implementation independent method of knowledge representation, in all of the stages of expert system development. We use K-trees for obtaining and recording heuristics, ordering rules, translating rules into an expert system using a shell, and for documenting the system.

No abstract

data involved (> million points) and system limits, processing time could not be further reduced.Clearly there was no reason to be overambitious or greedy; the results were more than satisfactory.After the BINIT expert system finished processing the data, graphs were generated from the data so that the author could interpret the results and determine the mechanism of the original nuclear reaction. KNOWLEDGE REPRESENTATIONThe Binit expert system used heuristics to represent and apply the knowledge of the nuclear physicist. Heuristics are general rule-of-thumb guidlines for quickly choosing the correct action to take. An important heuristic was that the graphic distance covered by one mass did not vary appreciably from one particle to another. Some additional, more specific heuristics include: 1. Hydrogen has three common masses: 1, 2, 3. 2. Helium has two common masses: 3, 4. 3. Lithium has two common masses: 6, 7. 4. Beryllium has two common masses: 7, 9. 5. The distance between beryllium-7 and beryllium-9 is equal to the distance covered by one mass. This is because beryllium-8 does not show up. ABSTRACTIn this paper I summarize the experiences of four novice knowledge engineers in the development of an expert academic advising system. We discuss the ideas and techniques behind the system and some of the methods developed and used in implementing the system using a frame-based expert systems shell.

tools are being developed, there is more work needed in furthering this effort.This also suggests a need in not only acquiring knowledge to first develop an expert system, but also acquiring knowledge to update the expert system once it is built. Once the expert system is delivered, the life cycle does not end here. There will have to be some persons responsible for maintaining and updating the expert system. For some expert systems, this "maintenance team" might be fairly large. XCON, an expert system developed by DEC for configunng VAX computer systems, requires a maintenance team of about 30 individuals. With better aids for acquiring knowledge in the front-end and back-end of an expert system's life cycle, the cortstruction and maintenance of an expert system could be greatly facilitated. ConclusionEven though expert systems have made great strides over the past ten years, there is plenty more work needed to further increase their "intelligence." This paper discussed some of the major limitations of current expert system technology and applications. In the years to come, these limitations will be addressed and hopefully resolved in order to make expert systems even more used throughout industry, government, and universities. As the field matures, there will be more structured methodologies and standards set for expert system development and implementation [6,7,8]. As this is done, a greater standardization of techniques for building and implementing expert systems will result. Through the setting of standards, the expert systems field will become even more recognized and reputable. As this occurs, the question of "artificial stupidity" in expert systems will become moot!