Judge: A Case-based Reasoning System

Abstract

SUMMARY OF RESEARCH People tend to improve their abilities to reason about situations by amassing experiences in reasoning. The more situations which a person knows about. the more able he is to account for feature differences between a new input and old knowledge. A computer program which can improve its ability to reason must also have access to situations which it has previously analyzed or reasoned about. Previous experiences thus require some mechanism for orderly storage and retrieval. The inability to save accounts of previous experiences for future application and modification represents a serious shortcoming of most. if not all. rule-based expert systems. A reasoner's goals and beliefs can influence how he (or it) processes input situations in several ways. Chief among these are the abilities to store experiences differentially and to retrieve particular kinds of episodes. Goals can help to impart a systematicity to the task of organizing memory by virtue of the fact that they can be prioritized with respect to one another. A goal which is hierarchically more important than others will have an earlier and more substantive role in comparing and indexing episodes. For example. the fact that a person is wearing a particularly attractive blue shirt pales in significance to most people if the person is also using a gun to hold them up. The greater relevance of the latter feature can be demonstrated by the experiencer recalling the situation faster with a query such as. "Remember the person who held you up last month?" than with. "Remember the blue shirt that you liked that you saw someone wearing last month?" or by his not remembering the experience at all with the second query. The JUDGE system has been developed to demonstrate the utility of using the goals of a program to reason about input situations by indicating differences (and their effects) between the input and other stored situations. and then storing the input along with some notation of the reasoning which was applied to it. This is called "case-based" reasoning. In addition to storing episodes in and retrieving them from its memory. the system is also able to construct. use and modify its own rules for T. M. Mitchell et al., Machine Learning