Second Generation Expert Systems

762 Pages·1993·43.187 MB·English

by  Jean-Marc David, Jean-Paul Krivine, Reid Simmons (auth.), Jean-Marc David, Jean-Paul Krivine, Reid Simmons (eds.)

Second Generation
Expert Systems
Jean-Marc David
Jean-PaulEJivine
Reid Simmons
(Editors)
Second Generation
Expert Systems
With 212 Figures and 12 Tables
Springer-Verlag
Berlin Heidelberg New York
London Paris Tokyo
Hong Kong Barcelona
Budapest
Editors
Jean-Marc David
Renault
Service Sysoomes Experts
860 Quai StaIingrad
92109 Boulogne-Billancourt
France
Jean-Paul Krivine
Electricite de France
Direction des Etudes et Recherches
1, Av. du oem~ral de Gaulle
92141 Clamart Cedex
France
Reid Simmons
Carnegie Mellon University
School of Computer Science
5000 Forbes Avenue
Pittsburgh, PA 15213-3890
USA
ISBN-13: 978-3-642-77929-9 e-ISBN-13: 978-3-642-77927-5
DOl: 10.1007/978-3-642-77927-5
Library of Congress Cataloging-in-Publication Data
Second generation expert systems/J.-M. David, J.-P. Krivine, R. Simmons, eds. p. cm. Includes
bibliographical references and index.
1. Expert systems (Computer science) I. David, J. -M. (Jean Marc) TI. Krivine, J. -P. (Jean Paul) TIL
Simmons, R. (Richard) QA76.76.E95S43 1993 006.3'3-dc20 93-19194 CIP
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcast
ing, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this
publication or parts thereof is permitted only under the provisions of the German Copyright Law of
September 9, 1965, in its current version, and permission for use must always be obtained from
Springer-Verlag. Violations are liable for prosecution under the German Copyright Law.
© Springer-Verlag Berlin Heidelberg 1993
Softcover reprint of the hardcover 1st edition 1993
The use of general descriptive names, registered names, trademarks, etc. in this publication does not
imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
Cover Design: H. Lopka, Ilvesheim
Typesetting: Camera ready by the editors
33/3140 -5 4 3 2 I 0 -Printed on acid-free paper
Foreword
This book results from a series of workshops and conferences that were organized
in France between 1988 and 1991.
A workshop entitled "Second Generation Expert Systems: Combining Heuris
tic and Deep Reasoning" was organized during the IMACS Congress in Paris
in July 1988. About fifty researchers gathered for that workshop, which can
be considered the first meeting devoted explicitly to second generation expert
systems.
The Avignon Conference on Artificial Intelligence, Knowledge-Based Systems
and Natural Language then hosted a separate conference on this topic from 1989
to 1991. More than 200 papers were submitted during this period, and about
60 of them were published in the proceedings. Subsequent to 1991, the second
generation expert system conference was integrated into the main scientific con
ference.
These gatherings played a major role in bringing researchers together, defin
ing the field, and in the exchange of ideas. During those years, they were the
prime forum for presenting research on second generation expert systems. Most
of the contributors to this book were involved in one or more of the proceedings.
As chairman and co-chairs of these workshops and conferences, we would
like to thank those who enabled these events. We are especially indebted to the
authors and to those who acted in the different program committees for their
invaluable contribution to the overall success of these meetings.
The Editors
Table of Contents
Part I: Introduction ............................................................... 1
1. Second Generation Expert Systems:
A Step Forward in Knowledge Engineering .............................. 3
JM. David, JP. Krivine & R. Simmons
Part II: Combining Multiple Models & Reasoning Techniques ........... 25
2. The Roles of Knowledge and Representation in Problem Solving .... 21
R. Simmons & R. Davis
3. Combining Heuristic Reasoning with Causal Reasoning
in Diagnostic Problem Solving ........................................... 46
L. Console, L. Portinale, D. Theseider Dupre & P. Torasso
4. Combining Causal Models and Case-Based Reasoning ............... 69
P. Koton
5. Generate, Test and Debug: A Paradigm for Combining
Associational and Causal Reasoning .................................... 79
R. Simmons
6. The Business Analyzer:
A Second Generation Approach to Financial Decision Support. ....... 93
W. Hamscher
7. QUAWDS: Diagnosis Using Different Models
for Different Subtasks ..................................................... 110
T. Bylander, M. Weintraub & S. Simon
8. Integrating Functional Models and Structural Domain Models
for Diagnostic Applications ............................................... 131
J. Hunt & Ch. Price
9. Multiple Models for Emergency Planning ............................. 161
O. Paillet
10. Knowledge-Based Design Using the Multi-Modeling Approach ..... 174
G. Guida & M. Zanella
VIII
m:
Part Knowledge Level Approaches .......................................2 09
11. Issues in Knowledge Level Modelling ..................................2 11
W. Van de Velde
12. Generic Tasks and Task Structures:
History, Critique and New Directions ................................... 232
B. Chandrasekaran & T. Johnson
13. The Componential Framework and its Role in Reusability ..........2 73
L. Steels
14. Towards a Unification of Knowledge Modelling Approaches ........2 99
B. Wielinga, W. Van de Velde, G. Schreiber & H. Akkermans
15. On the Relationship between Knowledge-based Systems Theory
and Application Programs:
Leveraging Task Specific Approaches ..................................3 36
J. Sticklen & E. Wallingford
16. Generic Models and their Support in Modeling
Problem Solving Behavior ................................................3 5O
P. Rademakers & J. Vanwelkenhuysen
17. Building and Maintaining a Large Knowledge-Based System
from a 'Knowledge Level' Perspective: the DIVA Experiment ........3 76
JM. David, JP. Krivine & B. Ricard
Part N: Knowledge Acquisition ..............................................4 03
18. An Overview of Knowledge Acquisition ................................4 05
M. Musen
19. Knowledge Acquisition Process Support
Through Generalised Directive Models ................................. 428
P. Terpstra, G. van Heijst, N. Shad bolt & B. Wielinga
20. Using the System-Model-Operator Metaphor
for Knowledge Acquisition ...............................................4 56
W. Clancey & M. Barbanson
21. Explicit and operational models as a basis
for second generation knowledge acquisition tools .................... 465
M. Linster
IX
22. ACTE: A Causal Model-Based Knowledge Acquisition Tool ......... 495
J. Charlet
23. Acquisition and Validation of Expert Knowledge
by Using Causal Models ...................................................5 17
Ch. Reynaud
Part V: Explanation ............................................................5 41
24. Explanation in Second Generation Expert Systems ...................5 43
W. Swartout & J. Moore
25. Explanation Using Task Structure
and Domain Functional Models .........................................5 86
M. Tanner, A. Keuneke & B. Chandrasekaran
26. Second Generation Expert System Explanation ........................6 14
M. Wick
Part VI: .Architectures ..........................................................6 41
27. Architectural Foundations for Real-Time Performance
in Intelligent Agents ......................................................6 43
B. Hayes-Roth
28. An Investigation of the Roles of Problem-Solving Methods
in Diagnosis ................................................................ 673
W. Punch & B. Chandrasekaran
29. Knowledge Architectures for Real Time Decision Support ...........6 99
J. Cuena
30. MODEL-K for prototyping and strategic reasoning
at the knowledge level .....................................................7 21
W. Karbach & A. VofJ
31. A Framework for Integrating Heterogeneous Learning Agents ..... 746
B. Silver, J. Vittal, W. Frawley, G. fba, T. Fawcett,
S. Dusseault & J. Doleac
Part I
Introduction
Second Generation Expert Systems:
A Step Forward in Knowledge Engineering
Jean-Marc David1 and Jean-Paul Krivine2 and Reid Simmons3
1 Service Systemes Experts, Renault, Boulogne Billancour.t, France
2 Direction des Etudes et Recherches, Electricite de France, Clam art , France
3 School of Computer Science, Carnegie Mellon University, Pittsburgh, PA
Abstract. Second generation expert systems are characterized by two
approaches: combining multiple models and reasoning techniques, and
using knowledge-level approaches for designing systems. These approaches
are complementary ways to overcome drawbacks of first generation ex
pert systems. This paper describes these approaches and reviews how
they benefit knowledge-based systems with respect to knowledge acqui
sition, explanation, robustness and efficiency, and reuse of models, knowl
edge bases and problem-solving code. Many of the points are illustrated
with examples taken from papers in this volume.
1 Introduction
This book contains a collection of articles about knowledge-based systems that
are collectively known as "second generation expert systems." This term is some
what fuzzy: there is not a clear distinction between "first" and "second" gen
eration systems; it is more of an evolution of ideas, styles, and techniques for
constructing knowledge-based systems. The ideas, as evidenced by the diversity
of papers in this volume, in many cases arose independently as researchers saw
both the potential in knowledge-based systems and the limitations of the then
state of the art in technology.
While we cannot give a precise definition of "second generation expert sys
tems," they have certain common characteristics. First and foremost is the ac
knowledgement that knowledge is central in problem solving and that the explicit
modeling of knowledge is important for creating understandable and maintain
able systems. Different models and problem-solving methods are needed for dif
ferent aspects of the process, and the choice of models and methods to be used
has a large impact on the efficiency and competence of the knowledge-based sys
tem. Another common characteristic is in distinguishing between what knowl
edge is used and how it is implemented (the distinction between the "knowledge
level" and the "symbol level" [48]). In particular, second generation expert sys
tems demonstrate the importance of using the appropriate knowledge for given
problems, and representing that knowledge in appropriate ways. Such systems
often combine multiple representations, problem solving strategies, and learning
methods within a single system.
4
In short, the field is characterized by an increased understanding of what
knowledge it takes to solve problems and how best to encode that knowledge so
that a computer can make use of it.
Research in these "second generation" topics is still progressing and new
systems continue to be developed. It is one of the main motivations of this volume
to contribute to the dissemination and cross-fertilization of these ideas and to
accelerate advancement of the field. The current state of the art, as reflected by
this volume, is just a snapshot of an ongoing process. Yet it is valuable, at this
point in time, to show where the field is and where it has yet to go to achieve
truly "expert" systems.
Before discussing the promises and problems of second generation expert sys
tems, we briefly review the development of the field of knowledge-based systems.
Solving complex problems has always been the hallmark of Artificial Intelli
gence. Early work in AI concentrated on developing general techniques for solv
ing problems [32, 49]. In the late 1970's and early 1980's, however, researchers
came to the realization that knowledge about tasks and problems, rather than
clever search techniques, was a main source of power in problem solving. Even
more importantly, they realized that this knowledge could be acquired from ex
perts and represented in a form that computers could use. Expert systems were
developed for quite complex domains, including medical diagnosis [56], geology
[28, 31], computer layout [45], etc.
The knowledge in these early expert systems was often obtained by inquiring
from experts how they solved particular problems. The "knowledge engineer"
then encoded the experts' advice in the form of associational (also referred to
as heuristic or empiricaQ rules-of-thumb that mapped from observable features
of the problem to conclusions. For example, MYCIN has rules of the form "if
given symptoms are observed, then there is evidence for a particular infection."
These systems solved problems by chaining the rules together, either forward
(from premises to conclusions) or backwards (from goal to initial conditions).
Thus, early expert systems typically had a simple control structure and uni
form representation of knowledge, in the form of associational production rules.
The knowledge was typically represented at a single level of abstraction and im
plicitly combined knowledge about "how" to perform a task, "what" is in the
domain, and "why" things work. While it was initially thought that this would
make systems fairly easy to develop, in fact it leads to several problems related
to knowledge acquisition, explanation, brittleness, and maintainability [12, 26].
Knowledge acquisition can be difficult because experts often are unable to
adequately explain how they solve problems, or they provide rationalizations
that in practice do not lead to effective problem-solving behavior. While experts
may be quite adept at describing why something works, they often have much
more difficulty explaining how to diagnose malfunctions, such as indicating what
information is relevant or how to match symptoms and causes.
Creating readily understandable explanations was often difficult for these
early systems, in part because the absence of explicit "how", "what", or "why"
knowledge limited their explanations to traces of how the problem was solved.