| 1992 Publications |
|
| Ripple Down Rules: Turning Knowledge Acquisition into Knowledge Maintenance |
|
| P. Compton, G. Edwards, A. Srinivasin, R. Malor, P. Preson, B. Kang and L. Lazarus
|
| Artificial Intelligence in Medicine |
|
| The most successful applications of medical expert systems seem to be in the interpretation of laboratory data. However even in this domain, knowledge acquisition and maintenance are major problems. We have developed a knowledge acquisition technique ('ripple down rules') based on using knowledge only in the context in which it is acquired. The method also guides the expert to enter rules that are valid. This method trivialises knowledge acquisition so that building a pathology expert system becomes the minor daily task for the expert of correcting wrong interpretations and tuning the knowledge base to current expertise. A major expert system based on this technique, PEIRS (Pathology Expert Interpretative Reporting System), is now in use. The current limitations of the technique are that the underlying tree structure of the knowledge base may require the expert to re-enter some knowledge and that multiple diseases are handled as composite diseases. |
|
|
|
| Towards a Process Memory |
|
| B. Kang and P. Compton
|
| AAAI Spring Symposium: Cognitive aspects of knowledge acquisition |
|
| Process memory implies that knowledge is not recalled but is created anew and what appears to be recalling is actually rethinking. Memory is concerned with how to do this rethinking. An important feature of this would concern how we have changed out thinking, i.e. how we have learned to come to correct conclusions in a domain, how we have evolved from being an error prone novice to an expert etc. We contend that this type of process memory, enabling us to consider why we would change our minds in other contexts and how and why we have corrected errors in the past helps us to be wise. Humans are wise compared to expert systems in that they know when their conclusions are likely to be unreliable. We have developed a way for ripple down rule expert systems to use past experience in other contexts to warn that their current conclusions may be unreliable. This method is very primitive but in a preliminary evaluation was able to provide correct warnings for 24% of the test cases. We conclude from this study that information as to how and why its knowledge has evolved should be an integral part of a knowledge base. If this extra information is used expert systems may perform in a more wise fashion. |
|
|
|
| Knowledge Acquisition in Context : the Multiple Classification Problem |
|
| B. Kang and P. Compton
|
| The 2nd Pacific Rim International Conference on Artificial Intelligence |
|
| "Ripple down rules" (RDRs) is a knowledge acquisition and representation strategy which restricts the use of knowledge to the context in which it is acquired. This is a very successful approach which allows large classification expert systems to be built by experts using only their domain expertise. It also blurs the distinction between acquisition and maintenance and provides for easy long term incremental modification of the knowledge base. The disadvantage of the approach, along with decision trees and ordered rules, is that only a single classification can be provided per case. An extension is proposed to allow the method to be used to provide multiple independent classifications. The results here demonstrate that this extension does not significantly increase the knowledge acquisition task over single classification RDRs. Importantly, the method also results in less brittle expert systems. This sort of expert system can make comparisons about how knowledge has evolved in various contexts and thus make more prudent conclusions. |
|
|
