Browse Prior Art Database

Some Approaches to Knowledge Acquisition

IP.com Disclosure Number: IPCOM000128291D
Original Publication Date: 1985-Dec-31
Included in the Prior Art Database: 2005-Sep-15
Document File: 4 page(s) / 21K

Publishing Venue

Software Patent Institute

Related People

Bruce G. Buchanan: AUTHOR [+3]

Abstract

Knowledge acquisition is not a single, monolithic problem for Al. There are many ways to approach the topic in order to understand issues and design useful tools for constructing knowledge-based systems. Several of those approaches are being explored in the Knowledge Systems Laboratory (KSL) at Stanford.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 25% of the total text.

Page 1 of 4

THIS DOCUMENT IS AN APPROXIMATE REPRESENTATION OF THE ORIGINAL.

Some Approaches to Knowledge Acquisition

Bruce G. Buchanan Department of Computer Science' . Stanford University Stanford, CA 94305

Knowledge Systems Laboratory October 1985 Report No. KSL-85-38

ABSTRACT

Knowledge acquisition is not a single, monolithic problem for Al. There are many ways to approach the topic in order to understand issues and design useful tools for constructing knowledge-based systems. Several of those approaches are being explored in the Knowledge Systems Laboratory (KSL) at Stanford.

BACKGROUND

In 1969, while working on DENDRAL, we recognized the "bottleneck" problem of acquiring knowledge from experts for use by a knowledge-based system [1]. From that initial recognition, born out of our first efforts at systematizing the process now known as "knowledge engineering", developed a line of research that is still active at Stanford. In the context of DENDRAL, we first initiated research on interactive editors and automatic rule induction [2]. Then, in the context of MYCIN, we were again confronted with very practical problems of knowledge engineering, and further worked on interactive debugging tools and languages for expressing new knowledge [3]. To date, knowledge engineering has been the only means of building a complex knowledge base, but this remains a tedious process. Thus, we are seeking to develop tools that aid knowledge engineers. Studies in progress include exploring methods by which programs can acquire knowledge by induction from examples, by analogy, by watching, by SOAR's process of chunking, by discovery, and by understanding written text. In this brief overview, we summarize research recently completed or in progress in the KSL. Although we are developing programs in the context of particular, suitable domains, we are seeking methods that are domain independent. Thus some of our research has resulted in papers that analyze and discuss general problems [4, 5, 6, 7].

GENERAL MODEL

Knowledge acquisition cannot be thought of as a single problem; there are several dimensions to the transfer and transformation of problem-solving expertise from a human expert or other knowledge source into a program. In our research, we have identified three different stages of knowledge acquisition and are examining different kinds of learning appropriate to each stage. We have given the chess labels of the "opening," "middle game," and "end game" to these stages (see Chapter 5 of [8 ). All three can be seen as different perspectives on the general model for learning systems in [4]. In the opening, an expert must lay out the terminology and the problem-solving framework. All subsequent knowledge-acquisition work depends on making this conceptual foundation correct. The middle game builds on the framework that was established initially. In a rule-based system, a specialist provides a large block of rules to cover many cases.

Stanford University Page 1 Dec 3...