Explanation-Based Theory Revision: An Approach to the Problems of Incomplete and Incorrect Theories
Abstract
Knowledge intensive Artificial Intelligence systems rely on a model of the domain, called a domain theory, to fulfill their tasks. A domain theory consists of a encoding of the knowledge required by the system to draw inferences about situation of interest. Systems that rely on domain theory face two difficult problems. 1) Their performance is directly related to he amount of knowledge in the domain theory. In order to insure a satisfactory level of performance, the expect who constructs the domain theory has the tedious chore of anticipating the wide variety of examples on which the system may be run. For most complex real-world domains it is impossible to anticipate handcode all the required knowledge. The expert is forced to make approximations assumptions. This results in brittle systems that tend to draw erroneous inferences and fail frequently. 2) systems that rely on a domain theory are limited to reasoning within the deductive closure of the knowledge in the domain theory. Since the knowledge content of the domain theory remains constant, these systems are incapable of modelling dynamic or under-specified domains in which new knowledge is being constantly acquired. Furthermore, large amounts of additional knowledge must be provided to the system if it is to process new examples. Consequently, such systems tend to be inflexible and inextensible. This thesis describes a method called explanation-based theory revision for augmenting and correcting an inadequate domain theory. In brief, the method consists of detecting failures due to the inadequacies of the domain theory.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 1988
- Accession Number
- ADA205337
Entities
People
- Shankar A. Rajamoney
Organizations
- University of Illinois Urbana–Champaign