Cross-examination as a model of knowledge elicitation in the design of expert systems.
Item
-
Title
-
Cross-examination as a model of knowledge elicitation in the design of expert systems.
-
Identifier
-
AAI9108104
-
identifier
-
9108104
-
Creator
-
Fulda, Joseph Simcha.
-
Contributor
-
Adviser: Michael Anshel
-
Date
-
1990
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Computer Science | Law | Artificial Intelligence
-
Abstract
-
Statement of the problems. (1) The methodologies for eliciting knowledge from domain specialists are "soft" and unreliable--i.e., the variations on the interview scheme, even those which make heavy use of modern technology, do not result in a reproducible knowledge base. Of course, some methodologies are superior to others and advances have been made, but most existing methodologies do not rise to the level of genuine knowledge engineering. (2) If there are n items of evidence, each of which can take on m values, and based on the evidence D decisions are possible (in general, there is no reason to assume that the decisions are mutually exclusive or that the evidence must be binary), then the knowledge engineer must ask the domain specialist D(m{dollar}\sp{lcub}\rm n{rcub}{dollar}) questions, a combinatorially impossible task for even moderate n, before he can build his rule base.;Thesis. Use of the most highly structured question-and-answer paradigm, namely that of the legal process, can result in both a reproducible knowledge base and in one built without combinatorial explosion, thus resulting in a tractable problem. The most important technique of the legal question-and-answer process and the one requiring the most creativity on the part of the attorney is adversary examination, particularly cross-examination. By using ten principal heuristics, we introduce an algorithm for cross-examination which makes it possible to focus on the hard cases--those where inconsistency is likely to occur--and to effectively cover the remainder of the D(m{dollar}\sp{lcub}\rm n{rcub}{dollar}) cases. This is not always possible. But in cases where the evidence is more or less cumulative and independent, it is not only possible, it has been done for this dissertation using the medical domain. The heuristics for cross-examination are used on the hard cases until all inconsistencies are removed from the knowledge base and it proves futile to try to produce additional hard cases or additional inconsistencies.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
