U.S. Departmentof Commerce

National Bureauof Standards

(NASA-CB-169197) A N O V E R V I E W O F EXPESTSYSTEMS (National Bureau of Standards) 73 pHC A04/H.F A01 CSCL 05fl

G3/54

N82-29899

Onclas30266

NBSIR 82-2505

AN OVERVIEW OFEXPERT SYSTEMSMay 1982

Prepared for

National Aeronautics and SpaceAdministration HeadquartersWashington, D.C. 20546

https://ntrs.nasa.gov/search.jsp?R=19820022023 2020-07-08T17:03:32+00:00Z

NBSIR 82-2505

AN OVERVIEW OF EXPERT SYSTEMS

William B. Gevarter

U.S. DEPARTMENT OF COMMERCENational Bureau of StandardsNational Engineering- LaboratoryCenter for Manufacturing EngineeringIndustrial Systems DivisionMetrology Building, Room A127Washington, DC 20234

May 1982

Prepared for:National Aeronautics and Space AdministrationHeadquartersWashington, DC 20546

U.S. DEPARTMENT OF COMMERCE, Malcolm Baldrige. Secretary

NATIONAL BUREAU OF STANDARDS. Ernest Ambler, Director

Preface

Expert systems is probably the "hottest" topic in A r t i f i c i a l

I n t e l l i g e n c e ( A I ) t o d a y . I n t h e p a s t , i n t r y i n g t o f i n d

solutions to problems, AI researchers tended to rely on search

t e c h n i q u e s o r c o m p u t a t i o n a l logic. These t echn iques w e r e

s u c c e s s f u l l y used to solve e l e m e n t a r y or toy p r o b l e m s or v e r y

wel l s t r u c t u r e d p r o b l e m s such a s games . H o w e v e r , r ea l complex

problems are prone to have the character is t ic that their search

s p a c e t e n d s t o e x p a n d e x p o n e n t i a l l y w i t h t h e n u m b e r o f

pa rame te r s involved.. For such problems, these older . techniques

have g e n e r a l l y proved to be i n a d e q u a t e and a new approach was

needed . T h i s new app roach e m p h a s i z e d k n o w l e d g e r a t h e r than

sea rch and has led to the f i e l d of K n o w l e d g e E n g i n e e r i n g and

Expert Systems.

This report provides a cu r ren t overview of Expert Systems —

what it is, techniques used, exis t ing systems, applications, who

is do ing i t , who is f u n d i n g i t , the s ta te -of - the-a r t , r e sea rch

requ i r emen t s and f u t u r e t rends and opportunit ies.

This repor t is in suppo r t of the m o r e genera l NBS/NASA

report, "An Overv iew of Ar t i f ic ia l Intelligence and Robotics."

Acknowledgements

I w i s h to t h a n k those people a t S t a n f o r d U n i v e r s i t y , X E R O X

PARC, MIT and elsewhere who have been ins t rumenta l in developing

the k n o w l e d g e e n g i n e e r i n g f i e ld , and have c o n t r i b u t e d t i m e and

s o u r c e m a t e r i a l t o h e l p m a k e t h i s r e p o r t poss ib le . I

p a r t i c u l a r l y w o u l d l i k e to t h a n k M a r g i e Johnson who has done a

heroic job typing this series and faci l i ta t ing their publication.

ii

TABLE OF CONTENTS

I Introduction 1

II What is an Expert System 2

III The Basic Structure of an Expert System 2

IV The Knowledge Base 6

V The Inference Engine 8

VI.. Uses of Expert ..Systems .. ...,..-• • 10

VII Architecture of Expert Systems 12

A. ' Introduction ..' .'.'... . . ; . .... .... 12

B. Choice of Solution Direction 18

1. Foward Chaining 18

2. Backward Chaining 18

3. Forward and Backward Processing Combined . . 18

4. Event Driven 19

C. Reasoning in the Presence of Uncertainty .... 19

1. Numeric Procedures ...... 19

2. Belief Revision or "Truth Maintenance" ... 20

D. Searching a Small Search Space 20

E. Techniques for Searching a Large Search Space . . 20

1 . Hierarchical Generate and Test 20

2. Dependency-Directed Backtracking ...... 21

3. Multiple Lines of Reasoning 21

F. Methods for Handling a Large Search Space byTransforming the Space 22

1. Breaking the Problem Down into Subproblems . 22

2. Hierarchical Refinement into IncreasinglyElaborate Spaces 23

iii

3. Hierarchical Resolution into ContributingSub-spaces 24

G. Methods for Handling a Large Search Space byDeveloping Alternative or Additional Spaces . . 26

1. Employing Multiple Models 26

2. Meta Reasoning 26

H. Dealing with Time 26

1. Situational Calculus 26

2. Pj.anni.ag with Time Constraints- » . - . . . - .--.-•' 27

VIII Existing Expert Systems 28

IX -Tools for Building Expert Systems .......... 31

X Constructing an Expert Systems 34

XI Knowledge Acquisition and Learning . . 36

A. Knowledge Acquisition 36

B. Self Learning and Discovery 36

XII Who Is Doing It 38

XIII Who is Funding It 39

XIV Summary of the State-of-the-Art 41

XV Current Problems.and Issues 43

XVI Research Required 46

XVII Future Trends 47

References '. < 52

iv

List of Tables

1. Characteristics of Example Expert Systems 13, 54

2. Summary of Characteristics of Systems in Table 1 ... 17

3. Existing Expert Systems by Function 29

4. Programming Tools for Expert Systems 32

List of Figures

1. Basic Structure of An Expert Systems 4

2. Expert Systems Applications Now Under Development . . 48

3. Future Opportunities for Expert Systems 49

ORSGSI . Introduction OF POOR QU*Ui.J.1TiP

In the 70's, it became apparent to the AI community, that

••arch strategies alone, even augmented by heuristic* evaluation

functions, were often inadequate to solve real world problems.

Th« complexity of these problems were usually such that (without

incorporating substantially more problem knowledge than had here-

tofore been brought to bear) either a combinatorial explosion

.occurred that defied reasonable search times, or that the ability

to generate a suitable search space did not exist. In fact, it

became apparent that for. many problems; that expert domain

knowledge was even more important than the search strategy (or

inference procedure). This realization led to the field of

•Knowledge Engineering," which focuses on ways to bring expert

knowledge to bear in problem solving.** The resultant expert

•yatems technology, limited to academic laboratories in the 70's,

ia now becoming cost-effective and is beginning to enter into

CO»mercial applications.

•Heuristics are "rules of thumb," knowledge or other techniquesthat can be used to help guide search.

•*0ne impor tant aspect of the knowledged-based approach is thatthe combinator ia l complexity associated wi th real-world problemsis m i t i g a t e d by the m o r e p o w e r f u l f o c u s s i n g of the sea rch tha tcan be obtained w i t h rule-based heur is t ics usually used in expertsys tems as opposed to the n u m e r i c a l h e u r i s t i c s ( e v a l u a t i o nfunct ions) used in classical search techniques. In other words,the ru le-based sys t em is able to reason about i ts own sea rche f f o r t , in add i t i on to r e a son ing abou t the p r o b l e m d o m a i n . (Ofcourse, this also implies that the search strategy is incomplete.Solutions may be missed, and an ent ire search may fa i l even whenthere is a solut ion " w i t h i n r each" in the p r o b l e m space d e f i n e dby the domain.)

ORK2.'NAL' PASS s.-JOF POOR QUALITY

II. What is an Expert System?

Feigenbaum, a pioneer in expert systems, (1982, p. 1) states:

An "expert system" is an intel l igent computer pro'g^r'amthat uses k n o w l e d g e and i n f e r e n c e p rocedures to solvep r o b l e m s t h a t a r e d i f f i c u l t enough t o r e q u i r e s i g n i f i c a n thuman expertise for their solution. The knowledge necessaryto p e r f o r m at such a level, p lus the i n f e r e n c e procedure '^u s e d , can be t h o u g h t of as a model of the e x p e r t i s e of thebest pract i t ioners of the f ie ld . . ;

The k n o w l e d g e of an expe r t s y s t e m cons is t s of fact'sand h e u r i s t i c s . The " f a c t s " c o n s t i t u t e a b o d y o fin fo rma t ion that is widely shared,, publicly available-, -andg e n e r a l l y a g r e e d u p o n b y e x p e r t s i n a f i e l d . T h e"heuris t ics" are mostly pr ivate , little-discussed rules o'fgood j u d g m e n t ( r u l e s of p l aus ib l e r e a s o n i n g , r u l e s of good

• guessing) that ' character ize ex-pert-level 'decision m a k i n g in.. . the. f i e l d . - . . T h e - p e . r f o r m a n c e - level of an -exper t s y s t e m is

p r i m a r i l y a f u n c t i o n of the s i z e and q u a l i t y of tlieknowledge base that it possesses.

i • . *f i

III. The Basic Structure of an Expert System...... >t- ;•

An expert system consists of:_j ;_; •. 4

1) a knowledge base (or knowledge source) of domain facts

and heuristics associated with the problem;

2) an inference procedure (or control structure) for

utilizing the knowledge base in the solution of the

problem;

3) a working memory - "global data base" - for keepfifg

track of the problem status, the input data for the

particular problem, and the relevant history of what

has thus far been done. - * ;

A human "domain expert" usually collaborates to help develop

the knowledge base. Once the system has been developed, in

addition to solving problems, it can also be used to hel!p

instruct others in developing their own expertise.

Thus, Michie (1980, pp. 3-5) observes:

•..that there ace three different user-modes for an expertsystem in contrast to the single mode (getting answers toproblems) characteristic of the more familiar type ofcomputing:

(1) getting answers to problems — user as client;

(2) improving or increasing the system's knowledge — useras tutor;

(3) harvesting the knowledge base for human use — user aspupil.

Users of an expert system in mode (2) are known as "domainspecial i sts.". It.i.s not possi.ble .to. build an expert sy.s.temw i thou t one...

An expert systems acts as a systematizing repository overtime of" the knowledge accumulated by many specialists- of

• • diverse experience. Hence, it can and does. ul.tim,atelyattain a level of consultant expertise exceeding* that ofa.ny single one of its "tutors."

It is usual to have a natural language interface to

facilitate the use of the system in all three modes. Normally,

an explanation module is also included, allowing the user to

challenge and examine the reasoning process underlying the

system's answers. Figure 1 diagrams a typical (though somewhat

idealized) expert system. When the domain knowledge is stored as

production rules, the knowledge base is often referred to as the

"rule base," and the inference engine the "rule interpreter."

An expert system differs from more convential computer

programs in several important respects. Duda (1981, p. 242)

observes that, in an expert system, "...there is a clear

separation of general knowledge about the problem (the rules

*There are not yet many examples of expert systems whoseperformance consistantly surpasses that of an expert. Andcurrently, there are even fewer examples of expert systems thatuse knowledge from a group of experts and integrate iteffectively. However the promise is there.

a: nq

<C <C LUPQ I— COO <C <C—I .Q-. PQCD

CO

<cCO

CO

LU

cc:LUCO

CJD

Q_

CO

ce:I—<c

CO

CO

Q_

X

ce:n>CD

LUor

cc:CO

CO<£.pa

cr>

— J coCD PQ

CO COa a=3 =Dce: ce:

CD <_>

LU LU__i ce:3: LUO LL.

C_5

ce:ocoLUCD(=)

f o r m i n g a k n o w l e d g e base) f r o m i n f o r m a t i o n a b o u t t h e c u r r e n t

problem (the input data) and methods for applying the general

k n o w l e d g e t o t h e p r o b l e m ( t h e r u l e i n t e r p r e t e r ) . " I n a

conventional computer p rog ram, knowledge per t inent to the problem

and methods for u t i l i z ing this knowledge are all in te rmixed , so

that it is d i f f i c u l t to change the program. In an expert sys tem,

". . . the p r o g r a m i t s e l f i s o n l y a n i n t e r p r e t e r ( o r g e n e r a l

reasoning -mechanism) and .[ideally], t h e . s y s t e m can .be. chang.ed by

simply adding or substract ing rules in the knowledge base."

IV. The Knowledge Base

The most popular approach to representing the domain

knowledge needed for an expert system is by production rules

(also referred to as "SITUATION-ACTION rules" or "IF-THEN

rules"). Thus, often a—knowledge base is made up mostly of rules

which are invoked by pattern matching with features of the task

environment as they currently appear in the global data base.

The rules in a ^knowledge base represent the domain f a.cts .and

heuristics - rules of good judgment of actions to take when

specific situations 'arise. The power of the expert system' lies

in the specific knowledge of the problem domain, with potentially

the most powerful systems being the ones containing the most

knowledge.

Duda (1981, p. 242) states:

Most e x i s t i n g ru le -based s y s t e m s c o n t a i nh u n d r e d s o f ru l e s , u s u a l l y ob t a ined by i n t e r v i e w i n ge x p e r t s fo r w e e k s o r months . . . In any s y s t e m , theru l e s become connec ted to each o ther by assoc ia t ionl inkages to f o r m rule networks. Once assembled, suchn e t w o r k ' s can r e p r e s e n t a s u b s t a n t i a l b o d y o fknowledge. . . . •

An e x p e r t u s u a l l y has m a n y j u d g m e n t a l o r e m p i r i c a l r u l e s ,

f o r w h i c h t h e r e i s i n c o m p l e t e s u p p o r t f r o m t h e a v a i l a b l e

ev idence . In such cases , one approach is to a t t ach n u m e r i c a l

values (cer ta inty fac tors ) to each rule to indicate the degree of

c e r t a i n t y a s s o c i a t e d w i t h t h a t r u l e . ( I n e x p e r t s y s t e m

o p e r a t i o n , these c e r t a i n t y va lues a r e c o m b i n e d w i t h each o the r

and the c e r t a i n t y of the p rob lem data , to a r r i v e at a c e r t a i n t y

value for the f ina l solution.)

Mich ie (1980, p. 6) indicates that the cognitive strategies

of h u m a n exper t s in m o r e complex d o m a i n s are based "...not on

e l a b o r a t e ca l cu l a t i ons , but on the m e n t a l s to rage and use of

large incrementa l catalogs of pattern-based rules." Thus, human

chess m a s t e r s may be able to a c q u i r e , o r g a n i z e and u t i l i z e as

much as 50 ,000 pattern-based rules in achieving their r e m a r k a b l e

per formance . Mich ie (p. 20-21) indicates that such rules are so

p o w e r f u l tha t only some 30 r u l e s a re needed fo r expe r t s y s t e m

p e r f o r m a n c e for a chess subdomain such as King and Knigh t against

K i n g a n d R o o k , w h i c h h a s a p r o b l e m space s i z e o f r o u g h l y

2 , 0 0 0 , 0 0 0 conf igura t ions . .He f u r t h e r observed for chess that the

number, of. rules required, .grows slowly relative .to the .increase in.

d o m a i n c o m p l e x i t y . T h u s , in chess and other complex d o m a i n s

(such as indus t r ia l rou t ing and scheduling) it appears that well-

chosen p a t t e r n se ts m a y m a i n t a i n c o n t r o l ove r o t h e r w i s e

intractable explosions of combinator ia l complexity.

V. The Infe rence Engine

The p r o b l e m - s o l v i n g p a r a d i g m , and i t s m e t h o d s ,o r g a n i z e s and cont ro l s the s teps t a k e n to solve thep r o b l e m . O n e c o m m o n p l a c e b u t p o w e r f u l p a r a d i g minvolves the c h a i n i n g of IF -THEN r u l e s to f o r m a l ineof r e a s o n i n g . If the c h a i n i n g s ta r t s f r o m a set ofc o n d i t i o n s a n d moves t o w a r d some (poss ib ly r e m o t e )conclusion, the method is called f o r w a r d chaining. Ifthe conc lus ion is k n o w n (e.g., it is a goal to bea c h i e v e d ) , but the pa th to tha t conc lus ion is notk n o w n , then w o r k i n g b a c k w a r d s i s called f o r , a n d t h em e t h o d is ^££jiw^£d_ £.h.ainina.• ( H e u r i s t i c P r o g r a m m i n gP r o j e c t , 1980, p. 6)

• — • • • • • • T'he 'problem'' wi th f o r w a ' r d c h a i n i n g , w i t h o u t a p p r o p r i a t e

h e u r i s t i c s f o r p r u n i n g , i s t ha t y o u w o u l d d e r i v e e v e r y t h i n g

possible whether , you needed it. or not.. Backward . cha in ing works

f r o m goals to subgoals (by u s i n g the ac t ion side of r u l e s to

deduce the condition side of the rules) . The problem here, again

wi thou t appropriate heuris t ics for guidance, is the handling of

c o n j u n c t i v e subgoals . In g e n e r a l to a t t a ck a c o n j u n c t i o n , one

mus t f ind a case whe re all in teract ing subgoals are sat isf ied, a

search for which can of ten result in a combinator ia l explosion of

possibili t ies. Thus appropriate domain heur is t ics and suitable

infe rence schemes and archi tectures must be found for each type

of problem to achieve an e f f i c i e n t and e f f ec t ive expert system.

The knowledge of a task domain guides the problem- ',solving steps taken. Somet imes the knowledge is quiteabs t rac t—for example, a symbolic model of "how thingswork" in the domain. In fe rence that proceeds f r o m them o d e l ' s a b s t r a c t i o n s to m o r e de ta i l ed (less a b s t r a c t )s t a t e m e n t s i s called m o d e l - d r i v e n i n f e r e n c e . A l w a y sw h e n o n e i s m o v i n g f r o m m o r e a b s t r a c t s y m b o l i cs t a t e m e n t s to less a b s t r a c t s t a t e m e n t s , one i sg e n e r a t i n g e x p e c t a t i o n s , a n d t h e p rob l em-so lv ingbehav io r i s t e r m e d expec ta t ion d r i ven . O f t e n inproblem solving, however , one is work ing "upwards" f r o mthe details or the specific problem data to the higherlevels of a b s t r a c t i o n (i.e., in the d i r e c t i o n of " w h a ti t al l m e a n s " ) . Steps in this d i r e c t i o n are call da tadr iven. If you choose your next step e i the r on thebas is of some new da ta or on the basis of the last

prob lem-so lv ing step t a k e n , you are r e spond ing toe v e n t s , and the a c t i v i t y is ca l led £v££t: d_£^v e_ 11.(Heurist ic Programming Project 1980, p. 6). ~"

As ind ica ted e a r l i e r , an exper t sy s t em consis ts of t h r e e

m a j o r c o m p o n e n t s , a set of ru les , a global data base and a ru l e

i n t e r p r e t e r . The ru les a re ac tua t ed by pa t t e rns , ( w h i c h m a t c h

the IF s ides of the r u l e s ) in the g loba l d a t a base . The

application of the rule changes the system status and therefore

•the -da ta bas-.e, enabl ing .some rules and d isa-bl i ng...other s. The

rule in terpre ter uses a control strategy for f i n d i n g the enabled

rules and dec id ing wh ich ru le to apply. The basic control

s t r a t eg i e s used may be top down (goal d r i v e n ) , bo t t omup (da ta

dr iven) , or a combinat ion of the two that uses a relaxation-like

convergence process to jo in these opposite l ines of r ea son ing

together at some in termedia te point to yield a problem solution.

VI. Uses of Expert Systems

The uses of expert systems are v i r tua l ly l imit less . They

can be used to:

• diagnose

• monitor . .

• analyze

• interpret

• consul.t _. _ ..„. . . .... .. • .. .-..- • • ...

• plan

• ' ' -design' . . . - • . - • • • •

• • instruct

• explain

• learn .

• conceptualize

Thus they are applicable to:

• Mission planning, monitoring, tracking and control

• Communication

• Signal analysis

• Command and control

• intelligence analysis

• Targeting

• Construction and manufacturing

- design, planning, scheduling, control

• Education

- instruction, testing, diagnosis

• Equipment

- design, monitoring, diagnosis, maintenance, repair,operation, instruction

10

Image Analysis and Interpretat ion

P r o f e s s i o n s ( l a w , m e d i c i n e , e n g i n e e r i n g , a c c o u n t i n g ,

law enforcement)

- Consul t ing, instruction, interpretat ion, analysis

' Sof tware

- S p e c i f i c a t i o n , d e s i g n , v e r i f i c a t i o n , m a i n t e n a n c e ,

instruct ion

' Weapon Systems.

- T a r g e t i d e n t i f i c a t i o n , e lect ronic w a r f a r e , adap t ive

cSontro'l; • • " . • • • - . . • ' . - • • • • . . . . - . • • . . - . .

11

VII. Architecture of Expert Systems

A. Introduction

One way to classify expert systems is by function (e.g.

diagnosis, planning, etc). However, examination of existing

expert systems indicate that there is little commonality in

detailed system architecture that can be detected from this

classification.

...A.more f-rui tfu.l .approach appear s'- 'to' b'e to"look 'at problem

complexity and problem structure and deduce what data and control

structures might be appropriate to handle these factors.

The Knowledge Engineering community has evolved a number of

techniques which can be utilized in devising suitable expert

system architectures. These techniques* are described in the

following portions of this section.

The use of these techniques in existing expert systems is

illustrated in Table 1**. Table 1 describes the basic approach

taken by each of these expert systems and indicates how the

approach translates into key elements of the Knowledge Base,

Global Data Base and Control Structure. A listing of the systems

in Table 1, together with an indication of their basic control

structures, is given in Table 2.

Table 2 represents the expert system control structures in

terms of the search direction, the control techniques utilized

and the search space transformations employed. The approaches

*This chapter is largely derived from information contained inthe excellent tutorial by Stefik et al. (1982).

**Tables 1-1 to 1-4 are shown on the following pages. Table 1-5to 1-17 are at the back of this report.

12

toBeg4Jcaen

5I

W cd

coO

• MCDuu2J8o

m

w2wu

PC,

1p>(X6-cyucPS5cu

Ide/<cfl<t-

C

(

.^£^_

O

Ci^cafc£.[x^

*3o§

60a•He

•H

5U

•aMCO3MOhCO

CO•aSCOM600M

Ucua.

• .?enCO

S

1- -0)60 r-lG 0•H B> .i-l CM Oeg

•a eo4J

1-4 C

0 i-l<4-l Cfl

MCO 4Jeg ca3 0an u

*° • QRSGP-IAL PAGE '3S • • OF POOR QUALITY0)4Jcd • •

egaeg60

G 4Jco en

rH <1)CXi 4J

encu34JCJ3

4JCO CO^JC eg,-j tj v

CO cfl .u .-a .. ..... ... ... . • ., . . . • s. -. •. .- •-.- -. . • -1 1 ^eo 13C GO cfl0 cj

W 1 • . . . .B c f l 1 <C . 60 - - " . ' • • ' ' . • - ' • • • • • - . - . •0 ' M 0 O . ' GM eg 3 u -H s

<4-l Cfl (3M 4J O4J <j> 4-> ss o M

0 ) C f l 6 0 C O I 4 - l -H «-lW -a en TJ •3 M e g - H e g c o4J r-l O 4J 4J Ui ^O cd M-i cfl cfl O. O. . .3 4J -o eo co eoH C eo i-l HMD4 J C U M - O O O O O MCOB 3 C 4J CO M-l Q 3

•H 73 CO 4J M 4JMM cu a eo c co 4J ocfl a) cj eu 1-1 eu o 33 X 2 C 3 M '3 O. 4Jc j e g ( X i - H 4 J 4 J pi co eo

B60 CM a:

rs eu cu4J•r1

a

5KC

^

1=

aT

C^3

X

C

T:cre4J

en

Meg•a,2*^3

ecd

ceg61•Heg

i

eo•H4JCO4-1CUMaMeg4J

M

CO4J(flQ

.31

HH " Z §

•• =3 en o gZ H a: M 2w M O H SH H S O SIen Wl r» Z>" Z 3 3en M < b

•oCU

4J

BOM

«W

CO4JC•HcflM4JCOGOO

eg•H

O•

r-l

nera

teausib

le

CU r-l

O 0.

enegM94-1CJ31 1toeg4Jcd

T?

13

cd

0)

cd

Seu

C3

*CM

1CU

iH

ructu

ra

4Jen

CO

O•H

ese

nta

t

CX

•Q

U<4

enSio.cd600^jUcu(XenenCO

4JU•H•oeu

ex.•

CO

irHo

U•Hacfl60

0

14-4O

eoeuCO

•H

"gcau

COca

Bo

14-1

caeuiH3V

§

ectr

ogr

&en

cO4Jcfl

T3

Si4J1-1

*CDUCO

1 , ' •

•

^f

, •

cd(0•«

CN1

i— i

Ei

£^H

2

SY

STE

M:

CO

• §j iCO

CO

4J

<D0. foX 0H

0) Hi—4 2O. W

cO w. X . • • -iJ

w uM-) >4O &«3

COy•H4J

• . ,co«HJ-lCU4JycO1-1CO

45U

>t4J•H CCO Ol-l flCU 4J> CO

•H ps

S 0fn

•o1-1 4Jo a

M-l 4-1 CDC CO cJCO C C4J CU OCO H] O

oMH •• Z3 en OH .^ MM O HH EC 0co H g

M < fr,

[I

c.cliw•Jgcu

ucrct<jE-

e• H

CCcu•'

wc/«(Uug£5

FJc

fcfl

^

aCO0

Pu

• . ' • GRJGSfOAL PA i ES"§ OF POOR QUALITY

. .

2* . -•u • •

cu ' ' ' 'g '60 ' .

f l ' J • • - • ' - • '

CO COr-l 0)PL, 4j

4J . '. "flj•o•acCOCJ

cu •.•^.W - .. . 5 • . . . . . . . . . . . . . . . . . - . . . - , • : - . . • • •• -. - .

•H a.CO 0) •3 yCO CiH O

1 . 4J 60 . . '• • • • ••' ' ' 4J 'CO • C '

CO '60 0) iHl-l iH (5 M TJ

C > CU CO -H 01 M•H >. c y c 4J co . •

M 0) -H -H • C y . . — '

CO 0) W S cfl r ,iH • 'CU 45 M B 0 0) T3

T3 4-* ' O CU U *^9 *4-*•H tw 45 -H O P

4-1 X o

M Q ) O B l - i O _ t n

C ai n c u a c co w -d< i ) 4 J - H c a i c u i - i c u - H , /C U ( 5 3 6 0 C C U 3 C ) 0 ' O

COCO CU .01 n . .H . 33 4J

4J 44 y •cu y cu doto cu fl) 1-1 (5

i-) CO C -HC 15 CU 0 G•H O 45 y O

O 4J COCO 60 cfl

S o) B c cur-l O -H l-l

*O 43 SJ i »•H tH M-I en IM

co cu oX CO "O pM O 01 01 0)CO CX 4-1 4-1 C4-> Cfl C -HC H-l M . -H i-H01 O 0) COe c co 4J 4Jcu cu at a) ca coIH y eo-a o 450) CO -H B 4J

CL 0145 CO 43 >. <U <U*J H ' 45 3•H Cfl C « 4J W3 • (0 4J M

^ 45 U C cu 34J i y 01 CO O.W O l-i 4J E OCO 01 CO CO 0) O T34J 45 CU 43 i-l 45 CCO 4-1 CO 4J CU U cfl

M-I r-lO CO

O .

>> -Hl-l 4J CO0) CO 4J . .> e a. •o cu cuy 45 yW 4J C .

•H CO O0 S 0

14

§Cfl

S^5sw

. .1 >*^' Td

U&M

. ; -. .. •

'1B

fiM

*

SOSHZ

8

CUCfl

p

2r

c(*<0CHu

KO<;C£

HCJ£1pi

z

'eofl•ri

"a o '<u nt> M

•H 44

r "O O32 CO

Ml ".c •o ••H44CO14S. .

01 • TH 01tJ ^ VW 44M w c: oO CO O 44

44 o caM0). • 44 • 1*4 f3 • •S C cO CUO 0) 1-1 4444 M 44 COCO M M ^3 3 at M0 rj p., >"

P»% * • * • - • • '

rH 604= • C 4400 CO -H X3 44 fi 01o c -H arl <U S C^ e Coo

O 01 44*W CX 44 TJ •O3 01 O) 01 01

O T3 > CO 44CO cj O CO cO4) »J B 43 44•H X 0 CO44 <J <4-l O J«iM > 44 CO 0o> co co aiO.--N 01 C 44 44O O rH 01 43 COV4 0 3 43 3 >>a, <r 06 3 co co

UjvF^SJ^'.^l ? '• [-^ n . ^ ^

'OF PpOjR (^^^ .%

-

-. » • • . - • . . ' - : - - . • • • ' - • . • • - * • * . « . * » ' • . •••• ; . • ^ • • ^ . • • • • r f • • ,

. . - . . • . • . • ' . ' • • ' ' • • • • • • ' • • •

' 44 x4.' 'rH ' ' ' ' " . ' ' " ' '3 C 3 •O O M

•a -H60 C 44 O

•fi o) ca o .•H 44 (-1 OO>v X 3 • 1 'M CU 60 >» 'U . -H tH .CO O <4H 01U 44 C 44

^•^ O crj il-i o 8O CO -H

U-i ^ iH XCO CO O ^>

CO Cfl -H . M rH01 44 44 CX d)rH 43 M (X 443 3 aj •< o

OS CO O. ^ 44

§ H B£ MM O H

H H X UWJ C/l H rZ>• z b 5Cfl M < U,

1cOSOcP-."*4

(dCOO(X05

£

60C•H

O

rHO

cu4344

0

C1-1

CX CO

COS crj0) 44

<H A43 3O COMCXT3

01

CO 010) -OW M

PQ O

COa

X 44< CO> PS

CO0)M M3 4100 44Tt 3

*w CUa eO 00 U

"M01

us to

rn

uCO

So

<4H

•M0113l-iO

G•H

CO

144CO

•H844CJCU(-1

^4O

co•

rH

U-l0

)4

01•0

o

t

/co44

01eoa.SoCJ

.CO440)C

•HncdCJ

•3PLJ,

ooa•H

CO44

01

0CXeo0

443a.•

PJ

1CO

CO4401C

•HnCOU

CO3

•HC3

O44C•H

ca01Xo

43

44

C^4

•01

»COcuX0

43

C•H

CO44COlcoCXsou443CX

•CO44cuG•rl43a)oCO

I•Ha3

carH01a9a443

Cu

*

"*

•

OorH<4-l

ao

§44CO

M*

443O

grjlJ

*10

1

' 4=44

O44

e0

4)>o• s

nj

^60 COC co

•H 44rH 43

•3 S,o01 U

JS <tt44 cuO 01O >

rH• O

VO C/l

.,

5

tVl0)

^4o

01X•rl14-1

0143

c•H

01BO

44X01e•

1

0)

a)4Jco

ca

ss

CUrHO.

I

coo

CU

u<d)-i5

F l

4-1iH

ZMCJS4

*

caIHCU

iHCJ3'

•a^40

>4H

CCO4-1CO

Q)<4HMH•Hi-H4JI-lO

JSM

*• V

Z0

CO•HCOoc60

.cdiHa

MH •• Z

•• 3 V. OZ H CS rH

W M O HH H S CJ0) 00 H Z>< Z 3 3t/> M < fa

oC/3

z1

W2uwhJ'W

s3

K£

c_B.EHC/l

. *

1Zap •U

CdW<ft

E-*. Q

* i—<PC

i—

wCO<pe

tC

g

52t.

K

. g--isFV.P-<

• ^4

WCOOfXiOS3Pu

60

5 .5 3CO rH

JS 3O M

-o <u&J rC

J>

.2 3O Mcd jsca 4J

•acCO •

enX 4JM COo cu4J 4.)ca

•H CJJS iH

4J'4J "COC OCO CiH 604J COco IH

pLt T3

4-1cCU•H4J ecd o

'CX iH4-1

60 CJC CU

•H <4H

I Xi S3C -H•Hi-H O

COCU cd

rH 4-)3 cd

}^60 CUC JS

•H 4-1C <UO 60CO OCO 4JCUIH CO

CUi-H rHcd 34J IHCCU C

T3 CO3 3

CJ4J CV4 0

CU Ua. iX C!(U O

•H4-1 4-1C -HCU T3CO Ccu oIH CJ0.cu ca

p£ cd

UHoCO i-H

•H CdCO -H0 I-lG CU60 4-1CO CJ

iH CO

*

COCUencu4JoexX

CU4J

•H.4-1cacu

~X4JC•HCO4-J

MCU

_CJ

ca4-1Mcua.

cu

cur^4-1

JS

•H

*

"SCO

COco•H4JCJCU

c

•cu

i-H3|

JSCJCOcur(0

UH

CO

oTH4Jed

C0)

S0ocu

•

'

oiH

OiHJ3•H4-1CcdM0

CJ

CO<UCO

cu34-1rn

cflJ3M

CO•Hcncuj:4J

Oa.(H

•inrjcu)HIH3

.CJ

60a

•HC

•H

^3S0CJ

IH0

COCU

r~(3

OS

1cd•a•aCUN•Hca<u

JS4J

0ex^

60^4

CO•acd3CJcd

CiH

COJS

•

cx(0IH<U

JS

•COMO4-1CJCO

UaCO4-1IHCUCJ

CO4JMOcx(X3CO

0)uadi•a-rl

01

cuJ34J

•H

CUCUCO

04J

COCUCO0 •C 4-160 TH

• i

CO '34-1CO4-1CO

4-1ccua4-1co(U

4J

O

CO0)

!_{

3Pi

'•acuucfl<u

COco•H

• co3i-lCJ

oCJ

*cacuCOcuJS4JOa

JS

rHi-HCO

cu4-1CO3

rHCO

CU

^>,i-Hcu

•H4-1

ca3cd.j5

W

CU

_ 3

'O

cdM

^to14H

ca3

4J

OF PCOrt QUALITY '

. -' ' ' ••

*

• . .

60 'CiH

r*1* - ' V

I4H•H4-1 " 'CO3 . '

. . . . .. ' -. • .• . '- •' c n ' : • • • • - • • • • • • • • • •M<U • -

3 JSC 4J . . . . ' • • • . - • • -• • ' . .

• \. • . ; . . - . ' :•• .; • - - • • - • ' -:

'

,

. ' .

JSCJiHr?3

cacuCO

g60Cfl .

•H CO•a cu

3rH rHrH CflCfl t>

O X4-1 4J

co.54J Cfl •C 4JCU Mg o )4J acfl01 JSV4 60W ^1

Jjj

JS •o cu -4-t >tq tfl

. - .; :- ' :' - ;

-. •.

4 ^

16

OF POOR QUAL5TY

0)

34-1U3l-i4-1CO

•^4oVi^Jco

.CJ

CN0)

pQcd**

01 •CJ

aCO

J= <uMCO01

CO

f

c

XuVi

3,CO i

rH

rH

H

cT-l

CO

CO

co,CO

MHO

COo

•H4JCfl

racte

ri:

CO

CJ

COCOHUl

io4-1cocCOM-1

"OM

Coj

coJCJ<0M

a

sains B}3W

uojaniosas TBOfqojBjaTH

3uainauT ia i TB3*£uo«iB*ia3*T?

surajqoji-qng o^Uf ^Bajg

siapoft aidf inw

jo^fpg •^aort^aN

SujuosBa^ 30 sauf if 3TnH

3uatu33TtuDion ^SBai

oUT iOB i 'MOBg ^udAaias

'•Sujssang

rjsax puB a^Baauag

qoaBag aAT^stiHqxg

UaAflQ 3U3A3

pJB/U(3Bg pUB pJBAJOj

paBAMOBg.

pjtB/iao^

tyr

M

jH

gQ

2§Mpj13St*4

SYST

EM

1 — I

X

X

X

^• x ' • • • . - . _ . . . . . . . .

• -x . ' - • • 'x ' x • xv ' " ' • '

X X X

X X

X X

X X ^ X X X . X X

CO4J1-13u

>•> Vi' >•> >> CO0) Vl iH CO ! Vl (U M MC 4J CJ C >* 4J C 4J CU

iH CO 1 - I O O C O 1-1 01 4J COO -H • O O 1-1 O «H 3 4Ji H B c J i H r H B . C ' H B e u O• a c u c u ' a o c u w ' d c u B . oCU *C« rH CU CO »C CO CO «C O OS C J W S O U S S C J C J p i

cO

*rt

4J• eo a •

H iH S Mp. 3 U Cu .vi er o oo n

CO CU U Pn C CO• H 4 - I C O •< 60 i -14- l 00en c i-t c 4J u c CO W CO • • iH Q. O H C iH0 >, H rH C 01 4J 60 C60 CO iH • 5 Vl CJ iH CO iH Cc 0 4 J < o - < o c o d C - u t n c o

iH CO C! ' * C CO O O CO CU rHQ Q - ^ C J ^ h J C J S Q O P -

j^cPS

§ CO

^ Z Q HZ « O 1 ceM O Q < H•J Z W CO H rH - CO

s o w o i 2 s < 3 ! > o o 6 ' < 3

X

K

X X

X X

x: X

X

X

cou

CO iH4J 4JO 0)

JQ ao cuP5 CJ

00

5 cC 00C iH(0 co

rH COPH O

32 cj

1 3Z £

X

"*

X

co4JiH3OViiHU

•acu

P-4u

a00

•HCOcuo

gCO

X

X

'x

X

X

co

cu•oc

j3 5u0)cuCu

CO

MO.MCU4JCM

^ =

00iHco

MM

SH

HER

ARSA

HAR

PY

x

• •

X

X

£CXca00oi-ir-l

CO4JcaViCJ

•

^4Q.ViCU

M

CO4-1COQ

CO

CR

YS

ALI

17

used in the var ious expert systems are d i f f e r e n t implementa t ions

o f two bas ic ideas fo r o v e r c o m i n g the c o m b i n a t o r i a l exp los ion

associated w i t h search in real complex problems. These two ideas

ar e:(1) Find ways to efficiently search a space,

(2) Find ways to transform a large search space into

smaller manageable chunks that can be searched

efficiently.

It^.will be. obser.wed from. Table 2 tha t • ther e •. i s 1 i tt le

architectural commonality based either on function or domain of

expertise. Instead, expert 'system design may best be considered

as an art form, like custom ho'me architecture, in which the

chosen design can be implemented using the collection of

techniques discussed below.

B. Choice of Solution Direction

1. Forward Chaining

When data or basic ideas are a starting point, forward

chaining is a natural direction for problem solving. It has

been used in expert systems for data analysis, design,

diagnosis, and concept formation. .

2. Background Chaining

This approach is applicable when a goal or a

hypotheses is a starting point. Expert system examples

include those used for diagnosis and planning.

3. Forward and Backward Processing Combined

When the search space is large, one approach is to

search both from the initial state and from the goal or

hypothesis state and utilize a relaxation type approach to

18

match the solutions at an in termedia te point. This approach

i s a l so u s e f u l w h e n the s e a r c h s pa ce can be d i v i d e d

hie ra rch ica l ly , so both a bottom up and top down search can

be a p p r o p r i a t e l y c o m b i n e d . Such a c o m b i n e d sea rch is

p a r t i c u l a r l y app l icab le to complex problems incorporat ing

uncerta int ies , such as speech unders tand ing as exempli f ied

in HEARSAY II.

4. Event Dr iven

Th i s p r o b l e m solv ing d i r e c t i o n i s s i m i l a r to f o r w a r d

chaining except that the data or s i tuat ion is evolving over

•••••• t i m e * • In th is -case the next s-tep is" chosen e i t he r on the .

bas is of new d a t a . o r in r esponse to a c h a n g e d s i t u a t i o n

r e s u l t i n g f r o m the last p rob lem solving step t a k e n . This

e v e n t d r i v e n a p p r o a c h i s a p p r o p r i a t e f o r r e a l - t i m e

opera t ions , such as m o n i t o r i n g or cont ro l , and is also

applicable to many planning problems.

C. Reasoning in the Presence o f ; U n c e r t a i n t y

In m a n y cases, we m u s t d e a l ' w i t h u n c e r t a i n t y in data or ini

knowledge. Diagnosis and data analysis are typical examples.

1. Numer ic Procedures :

N u m e r i c p r o c e d u r e s h a v e b e e n d e v i s e d t o h a n d l * e

a p p r o x i m a t i o n s b y c o m b i n i n g ev idence . M Y C I N u t i l i z e s

"certainty factors" (related to probabil i t ies) which use the

r a n g e of 0 to 1 to i nd ica t e the s t r e n g t h of the ev idence .

F u z z y set t heo ry , based on poss ib i l i t i es , can also be

ut i l ized.

19

2. Belief Revision or "Truth Maintenance"

Often, beliefs are formed or lines of reasoning are

developed based on partial or errorful information. When

contradictions occur, the incorrect beliefs or lines of

reasoning causing the contradictions, and all wrong

conclusions resulting from them, must be retracted. To

enable this, a data-base record of beliefs and their

justifications must be maintained. Using this approach,

'truth maintenance techniques 'can exploit redundancies in

experimental data to increase system reliability.

D. Searching ji Small S.earch Space... . .. . .-.-.. . • .. -.

Many straightforward problems in areas such as design,

diagnosis and analysis have small search spaces, either because

1) the problem is small or 2), the problem can be broken up into

small independent subproblems. Often a single line of reasoning

is sufficient and so backtracking is not required. In such

cases, the direct approach of exhaustive search can be

appropriate, as was used in MYCIN and Rl.

E. Techniques for Searching a. Large Search Space

1. Hierarchical Generate and Test

State space search is often formulated as "generate and/>v

test" - reasoning by elimination. In this approach, the

system generates possible solutions and a tester prunes

those solutions that fail to meet appropriate criteria.

Such exhaustive reasoning by elimination can be appropriate

for small search spaces, but for large search spaces more

powerful technique are needed. A "hierarchical generate and

test" approach can be very effective if means are available

20

for evaluat ing candidate solutions that are only partially

specif ied. In these .cases, early pruning of whole branches

( r e p r e s e n t i n g en t i r e classes o f so lu t ions a s soc i a t ed w i t h

t h e s e p a r t i a l s p e c i f i c a t i o n s ) i s p o s s i b l e , m a s s i v e l y

reducing the search required.

"Hie ra rch ica l g e n e r a t e and test" i s a p p r o p r i a t e for

many large data interpretat ion and diagnosis problems, for

w h i c h . .all., solutions .are desired, providing a generator can

be dev i s ed tha t can p a r t i t i o n the s o l u t i o n space in w a y s

. that allow f o r early p r u n i n g . " - ' ' • • • • • . • •

2. Dependency-Directed Backtracking •

In the " g e n e r a t e and test" a p p r o a c h , w h e n a l ine of

r e a s o n i n g fa i l s and m u s t be r e t r a c t e d , one a p p r o a c h i s to

b a c k t r a c k to the mos t r e c e n t cho i ce po in t ( c h r o n o l o g i c a l

backtracking) . However , i t is o f ten much m o r e . e f f i c i e n t to

t race e r rors and i n c o n s i s t e n c i e s back to the i n f e r e n t i a l

steps that created them, using dependency records as is done

in M O L G E N . B a c k t r a c k i n g tha t is based on d e p e n d e n c i e s and

determines what to invalidate is called dependency-directed

(or relevant) backtracking.

3. Mult iple Lines of Reasoning , .

This approach can be used to broaden the coverage of an

incomplete search. In this case, search programs that have

fallible evaluators can decrease the chances of discarding a

good so lu t ion f r o m w e a k ev idence by c a r r y i n g a l i m i t e d

n u m b e r o f s o l u t i o n s i n p a r a l l e l , u n t i l w h i c h o f t h e

solutions is best is clarified.

21

P. Mg^hodg for JLajliLi.lJLS £ La rge Search £pja£e_the Space

1 . Breaking the Problem Down Into Subproblems

a . Non-Interact ing Subproblente

Th i s approach ( y i e l d i n g smal le r sea rch spaces) i s*

applicable for problems in which a number of non-interacting

tasks have to be done to achieve a goal. Unfortunately, few

real world problems of any magnitude fall into this class.

b. Interacting Subproblems. .; '.:. . ..... '.••-•• '•-'•

For most complex problems that can be broken up into

subprobleras, it has been found: that the' subproblems' interact

so that valid solutions cannot be found independently.

However/ to take advantage of the smaller search spaces•*

associated with this approach, a number of techniques have

been devised to deal with these interactions.

(1) £iJl<3 .a ZJL.x.S.d S. SJi6.!! . £.£ §.H^££2^i£S£ £2 ZJl.§J!i .Interactions Occur

S o m e t i m e s i t i s p o s s i b l e to f i n d an o r d e r e d

p a r t i o n i n g so tha t no in t e rac t ions occur . The Rl

sys t em (see Table 1-3) for c o n f i g u r i n g VAX c o m p u t e r s

successfully takes this approach.

(2) Least Commitment

This t e c h n i q u e coo rd ina t e s d e c i s i o n - m a k i n g w i t h

the availability of information and moves the focus of

p r o b 1 e m - so 1 v i ng a c t i v i t y a m o n g the a v a i l a b l e

s u b p r o b l e m s . Dec is ions a re not made a r b i t r a r i l y or

p r e m a t u r e l y , bu t a re postponed un t i l t h e r e i s e n o u g h

informat ion . In planning problems this is exemplif ied

by methods that assign a partial order ing of operators

22

in each subproblem and only complete the ordering when

sufficient information on the interactions of the sub-

problems is developed.

(3) Constraint Proprogation

Another approach (used by MOLGEN) is to represent

the interaction between the subproblems as constraints.

Constraints can be viewed as partial descriptions of

. . . entities, or as relationships (subgoals) that must be

satisfied. Constraint proprbgation is a mechanism for

moving information between subproblems, ..rBy, introducing

constraints instead of choosing particular values, a.

problem solver is able to pursue a least commitment

style of problem solving.

(4) Guessing or Plausible Reasoning

Guessing is an inherent part of heuristic search,

but is particularly important in working with

interacting subproblems. For instance, in the least

commitment approach the solution process must come to a

halt when it has insufficient information for deciding

between competing choices. In 'such cases, heuristic

guessing is needed to carry the solution process along.

If the guesses are wrong, then dependency-directed

backtacking can be used to efficiently recover from

them. EL and MOLGEN take this approach.

2. Hierarchical Refinement into Increasingly Elaborate Spaces

— Top Down Refinement

Often, the most important aspects of a problem can be

23

abstracted and a high level solution developed. This

solution can then be iteratively refined/ successively

including more details. An example is to initially plan a

trip using a reduced scale map to locate the main highways,

and then use more detailed maps to refine the plan. This

technique has many applications as the top level search

space is suitably small. The resulting high level solution

constrains the search to a small portion pf the search space

at the next lower level, so that at each level the solution

can readily be found.. This procedure is an importan.t

.technique f.or preventing combinatorial explosions in

searching for a solution.

3. Hierarchical Resolution into Contributing Sub-Spaces

Certain problems can have their solution space hierarchical-

ly resolved into contributing subspaces in which the

elements of the higher level spaces are composed of elements

from the lower spaces. Thus, in speech understanding, words

would be composed of syllables, phrases of words, and

sentences of phrases. The 'resulting heterogenous subspaces

are fundamentally different from the top level solution

space. However the solution candidates at each level are

useful for restricting the!range of search at the adjacent

levels, again acting as an important restraint on

combinatorial explosion. -Another example of a possible

hierarchical resolution is in electrical equipment design

where subcomponents contribute to the black box level, which

24

in turn contribute to the system level. Similarly/ examples

can be found in architecture, and in spacecraft and aircraft

design.

25

G* H£^]l£5£ JL2JL Handl ing a_ La rge S_eajr_cli Sjaa^c^e b% DAl te rna t ive or Addit ional Spaces

1. Employing Multiple Models

Somet imes the search for a solution u t i l iz ing a single

model is very d i f f i cu l t . The use of a l ternat ive models for

either the whole or part of the probLem may greatly s impl i fy*

the search. The SYN program is a good example of combining

the s t r e n g t h s of m u l t i p l e mode l s by e m p l o y i n g e q u i v a l e n t

f o r m s , o f electrical, circuits.. . . . . . . . . . . • . • • • • • •

2. Meta Reasoning

~. • . It is possible to ad'd: additional layers of spaces to a

sea rch space to help dec ide w h a t to do next . These can be

t h o u g h t of as s t r a t e g y and tac t ica l layers in w h i c h raeta

problem solvers choose among several potential methods for

d e c i d i n g w h a t t o d o ' n e x t a t t h e p r o b l e m leve l . T h e

s t r a t e g y , f o c u s i n g a n d s c h e d u l i n g m e t a r u l e s u s e d i n

C R Y S A L I S and the use of a s t r a t e g y space in M O L G E N fa l l in to

this category.

H. Dealing wi th Time . :

Lit t le has been done in the way of expert -systems that deal

w i t h t i m e exp l i c i t ly . The f o l l o w i n g a re approaches to d e a l i n g

wi th t ime in t e rms of t ime intervals.

1. S i tua t ionaL Calculus

Situational calculus was an early approach by McCar thy

and Hayes (1969) for r e p r e s e n t i n g sequences of ac t ions and

the i r e f f e c t s . I t uses the concept of " s i tua t ions" w h i c h

c h a n g e w h e n s u f f i c i e n t ac t ions have t a k e n p lace , o r w h e n

new data indicates a situational sh i f t is appropriate. Sit-

26

uations determine the context for actions and, through the

use of "frames,"* can indicate what changes and what remains

the same when an action takes place. VM uses the situation

approach for monitoring patient breathing.

2. Planning with Time Constraints

NOAH was an early parallel planner which dealt with

interacting subgoals. The method of least commitment and

backward chaining initially produced a partial ordering of

operators for each plan. When interference between subgoal

plans was observed, the planner adjusted the ordering of the

operators .to re-so-lve the .interference to. produce, a..fina.l

parallel plan with time ordered operators. DEVISER (Vere,

1981) is a recent derivative of NOAH which extends this

parallel planning approach to treat goals with time

constraints and durations. The principal output of DEVISER»

is a partially ordered network of parallel activities for

use in planning a spacecraft's actions during a planetary

fiyby.

*A frame is a data structure for describing a stereotypedsituation.

.2.7.

VIII. Existing Expert Systems

Table 3 is a list, classified by function and domain of use,

of most of the existing major expert systems. It will be observ-

ed that there is a predominance of systems in the Medical and

Chemistry domains following from the pioneering efforts at

Stanford University. From the list, it is also apparent that

Stanford Uhiversirtfy~~dominates in number of systems, followed by

CMU, MIT and SRI, with a dozen scattered efforts elsewhere.

The list indicates that thus far the major areas of expert

systems .development have been in diagnosis, .data analysis and

interpretation, planning and design. "How'ever, the list also

indicates that a few pioneering expert systems already exist in

quite a number of other functional areas. In addition, a

substantial effort is underway to build expert systems as tools

for constructing expert systems.

DENDRAL (Lindsay et al., 1980), which produces molecular

structural representations from mass spectrogram data, has been

the most widely used expert system; It has subsequently been

generalized to CONGEN to produce a set of structural candidates

from whatever constraining data is available.

Feigenbaum (1982, p. 16) states that the most knowledge

intensive system is INTERNIST, a medical diagnosis system which

considers almost 500 diseases and contains over 100,000 pieces o'f

knowledge.

28

,-\r-OF

BOi-lP3•P•H•P09S

60hi •

• 3 99 43

• 09 13EH to P h.• hi P 0

M 4> fl 4n S• MIX B

£ P at3 4i p

OS O CO

S

M

•9

T3£40

<H

Eat

be

rge

rU

.

a3 -0H hi43 0O 4-1 -

CO B*- at• P

EH CO

9

T3hi

M O03 4-tCO E

a)PCO

• •EH EH

M M• •

S 2

43 U60 COhi O3 K

B TJ•H B"0 a)W 03

OOhi

JO• • ^

9 0) EH V• H • B

S flj H 0• H • P

U S CO

>-l

N3hiO

atPBat

CO MK

• CO

a)U

M i-qca a.CO >-9

•9

TJ•O hiB Ofl5 ^4

OH Ca)PCO

•9

•eh<0

4-4Ba)pCO

COCO

BO

UB

•3'

VpB)

CO

EHCO

' b3 j?« p^? ry. &4IX 55 K M Cu KM CO W O 9 <0-. < EH X O. Q

O S5 S

OCO 03M O CO> iJ EH M.0 .<•„•.•'0--.J< 03 rH W <

a < tx caPS S O CO >HW W O OS

• •••so-'r -» S3 W

»-5 CO O OW O W W

COo

I O

'W

EH Qpa-ft.-..

03 CJ(X ••'

co SB

. CO

,. • to. .co a. 03« W M WO < 03 COM O EH M

CO >«-w< 0

CO

OSW . •ss ss% H<! o

a, o• ( scuo

• K-aw

•EH'H•s-

0r-l

43

a

4)

09

CO

ph>

pca•HXH

oo

0)E•HOft S•aai2

s s s

09PH

atEx,

P3

60oH

!>» >> r-l>» h. hi >, atHOP P 60 4»O 09 09 O toH -H -rj H So B e o h,4) 4) 4) 4) O

O0 0

po

09ca op ft3 ato ahi 4)

•H 43O P

atH SajO O

hi r-lP OO 434) 3H SH CO

09SHatB

CO <e4) P•-I at43 4)O hihi 43

CU EH

09 4)O Oft hiB Oat fc43O .*)4) CO

at

at

coEO•HP 60at B

3 S60 3•H at4-i hiBO OO hi

Cu

4) O

CO 4J>» 2CO 3

ohi P4) 3P <3

S1OO

co09 iH

•H W09 4)4) 4343 P

E >,>»CO

CO

P atfl O3 fto Shi 4)•H 43O CJ

0903 CO 60 O

•H >» E fl09 43 i-l P4> >> B 4)43 r-l E Ei ^ QJ G^ £J ^J

BO H C5

O 03

CO P = hi hiO a} T> at

r-l 43 P B Ho) O 4> at 3

E hi O4)

P09•HS4)

S4)

CJ

atrH W rHCU O

S

BO

BO•HPOE3

09OB

at•HO

Pat

M P•H 0)09 fc.

>» fti-H hat VB p< B

MatP "Oat E0 at

09•HM

3)B

C60flca0)

60E

•HBEa]r-lCU

SO

OhO CE U

•H -HB (HU IUat p,0) X

29

0cM

C0•H43

-P•H-PCOCM

P COH

•a oV, h

3 P O+j2 2 = «M C0 0 C O

cd o•pco B

cu•PCO

P

*oJM,

o<M

ccd43CO

•o• V.

0<M

ccd

43CO

G&Gi^AL PA^S IBOF POOR QUALITY

P

•rtViO4-iCcd•PCO

P

TJt,O4Hccd-pCO

COt,n>60•P3K

MKCO

TJCdK

p•av,o

CMCcd4J

CO

HCOM•»* .OCOp

p

•0;ti p02Vi OCcd+3

CO

COCOcd2

<M

O

•P

p

"OV,OV.ccd43CO

•cc0)

CMCOc\

CO

oWoS

6V-pCO

CO

2<:>-« jn x a,< (X ^ CO 2

.CO K P < . > .K <• CO S- ' •< a•w . •w

• • " • • orWE M EH

8 M O CO KO a Q W W COO . Cu l-H . K PL, <<( O D M X WCO CO O. W W. . . . . . ^ ., . -

M .

8 U"\W in MM w <; o coCO O CO H <X,o «< PS'S oK <C W

' • ' • • H • •a

CO SB »a 8W OCO KM O

cd

. Hi-

A

>!

-pCO

oo

ssoo

w_q

PO<:E-"

C•HcdSo0

co•H43

0c3(z,

COo•H•PcdBcuf*

•pcd2

Co•H•PcdSS*o

PL,

4J

ft<UOCoo

(tOc•H•0ccd-pCO£4

CD•o =c

. 3

fl0cucuA

CO

c0•H-pcd-Pcut,ft

cu+»

cM

rHcdc00•HCO

09O•H•PCO3Oo

<JJ

CUC•H_r*Ocd2

0)occdrH•H

CU

h3

CO

Ccdcuoo

c0•H+3

cd£4•HACOcuK

43ccu•H43cdPi

00ciH(jO•P•HCO2

COT-t

CO

HcdC^j

rH

cd(4

343y3(H•PCO

(H001•H

^>T3

^J

<U09P

0cdU00o(4

CU

Vl0)433P^Boo

00C

•H43Oo

J3CO<Ur-f\Q3OIH

EH

O•HC0f-,

+Jo<uH

.M

•a0)•a•H<j

Vicu433ftSoo

0)•H

CO0CbOcd•HorHcdu•HT)CU2

C0•H43

o3^143COCM

. C0

CO iH•rl 43co cd0 43C HbO 3 >»cd co bo•H a oQ O H

o o. H CU

cd H C5U a)•H 0•a -HCU "O2 <l>

2

Co•H43 .•HCO•H

3cr<U

^J

0)to•ocuH

0cbd

co4303SH

43cacooBcu43ca>,

CO

43F*IO*ftXw

00C•H•OCcd•pca

CUsocd

3o

TJc3o

Co5O

COScu•pCO

cucoa)

J3-P -~

rHO CO

+3 ON!-l

CO • —CUU Cc cdcu ct. cdcu A

4-1 O<u 3K m

30

IX. Tools fQJ: Building Expert Systems

To aid in the building of expect systems, special

programming tools have recently begun to be developed. These are

listed in Table 4. The most ambitious is AGE (Attempt to

Generalize). AGE (Nii and Aiello, 1979) has isolated a number of

inference, control and representation techniques from a few

previous expert systems and has reprogrammed them for domain

independence. AGE, itself an expert system, also guides people

in the use of these modules in constructing their own

individualized expert systems. AGE also provides two predefined

configurations o'f- components.'' One called the '""BlackboaTd' •

framework" is for building programs that are based on the

Blackboard model, as was used in HEARSAY II. The Blackboard

model uses the concepts of a globally accessible data structure,

called a blackboard, and independent sources of knowledge which

cooperate in forming hypotheses. The other predefined

configuration, called the "Backchain framework," is for building

programs that use backward chaining production rules like those

used in MYCIN.

31

Table 4

Programming Tools for Bui ld ing Expert Systems

Organ iza t ion N a t u r e

CMU

EMYCIN Standford U,

KAS •SRI-

ROSIE RAND

AGE Stanford U.

HEARSAY III OSC/Informat ionSciencesInst i tu te

UNITS S tanfo rd U .

A p r o g r a m m i n g l a n g u a g eb u i l t o n top of L I S Pd e s i g n e d to f a c i l i t a t et h e u s e o f p r o d u c t i o nrules.

A d o m a i n i n d e p e n d e n tv e r s i o n o f M Y C I N , w h i c haccompanies the backwardcha in ing and explanationapproach w i t h user aids.

S u p e r v i s e s i n t e r a c t i o nw -i-t h • '•• an •" e x p e r' t ••' T hbui lding or a u g m e n t i n g anexper t sys tem k n o w l e d g ebase in a n e t w o r k f o r mimplemented for PROSPEC-TOR.

A general rule-based pro-g r a m m i n g l a n g u a g e tha tcan be used to developl a r g e k n o w l e d g e bases.T r a n s l a t e s n e a r - E n g l i s hinto INTERLISP.

A s o p h i s t i c a t e d e x p e r tsys tem to aid use r s inbuilding expert systems.

A g e n e r a l i z e d d o m a i n -independen t extension ofH E A R S A Y II. I n c l u d e s a"context" mechan i sm, anda n e l a b o r a t e d " b l a c k -board" and scheduler.

A k n o w l e d g e r e p r e s e n t -ation language and inter-ac t ive k n o w l e d g e acqui-s i t i o n s y s t e m . T h el a n g u a g e p r o v i d e s b o t hf o r " f r a m e " s t r u c t u r e sand production rules.

32

Tool

TEIRESIAS

Table 4 (continued)

Programming Tools for Building Expert Systems

Organization Nature

Stanfo rd a. A e x p e r t s y s t e m t h a tf a c i l i t a t e s t h e in t e r -a c t i v e t r a n s f e r o fknowledge f r o m a h u m a ne x p e r t to the s y s t e m viaa ( r e s t r i c t e d ) n a t u r a llanguage dialog.

33

X. Constructing An Expert System

Duda (1981, p. 262) states that to construct a successful

expert system, the following prerequisites must be met:

there must be at least one human expert acknowledged toperform the task well

the primary source of the expert's exceptionalperformance must be special knowledge, judgment, andexperience

the expert must be able to explain the specialknowledge, and experience and the .m.ethpda. .used to. apply,"them to particular problems

the task must have a well-bounded domain of application

. Randy Davis. .(MIT) at IJC.AI.-81*. noted ..that- a good expert

system application:

doesn't require common sense

takes an expert a few minutes to a few hours

has an expert available and willing to be committed.

Hayes-Roth (1981, p. 2) adds that "...the problem should be

nontrivial but tractable, with promising avenues for incremental

expansion."

Having found an appropriate problem and an accessible

expert, it is then necessary to have available an appropriate

system-building tool, such as those described in the last

chapter. Realistic and incremental objectives should then be

set. Major pitfalls to be avoided in developing an expert system

are choosing a poor problem, excessive aspirations, and

inadequate resources.

•The International Joint Conference on Artificial Intelligence,Vancouver, August 1981.

The time for construction of early expert systems was in the

range of 20-50 man-years. Recently, breadboard versions of

simple expert sysems have been reported to have been built in as

little as 3 man-months, but a complex system is still apt to take

as long as 10 man-years to complete. Using present techniques,

the time for development appears to be converging towards 5 man-

years per system. Most systems take 2-5 people to construct, but

not more. (It takes one to two years to develop an engineer or

computer scientist into a knowledge engineer.)

Randy Davis (at IJCAI-81) indicated that the stages of

development = of • an expert system, .can be considered to. .be*: . •• . • ... .

1. System design .

2. System development (conference paper level)

3. Formal evaluation of performance

4. Formal evaluation of acceptance

5. Extended use in prototype environment

6. Development of maintenance plans

7. System release.

*Thus far, no current system has completed all these stages

35

XI. Knowledge Acquisition and Learning

A. Knowledge Acquisition

The key bottleneck in developing an expert system is

building the knowledge base by having a knowledge engineer

interact with the expert(s). Expert systems can be used to

facilitate the process. Some of these expert systems are

indicated in Table 3, with the KAS system being elaborated upon

in Table 1-7.

': The most'ambitious of these systems is TEIRESIAS (Davis and

Lenat, 1982) which supervises interaction with an expert in

building or augmenting a MYCIN .rule set. TEIRESIAS.uses•a model

of MYCIN's knowledge base to tell whether some new piece of

information "fits in" to what is already known, and uses this

information to make suggestions to the expert. An appropriate

expert may not always be continuously available during the

construction of the expert system, and in many cases may not have

all the expertise desired. In these cases other approaches to

acquiring the needed, expertise is desirable.

B. Self-learning and Discovery

Michie (1980, p. 11) observes that "The rule-based s tructure

of e x p e r t s y s t e m s f ac i l i t a t e s a c q u i s i t i o n by the s y s t e m of new

ru les and m o d i f i c a t i o n of ex i s t ing ru les , not on ly by t u to r i a l

interaction with a human domain specialist but also by autonomous

' l e a r n i n g ' . " A t y p i c a l f u n c t i o n a l a p p l i c a t i o n i s

"classification," for which rules are discovered by induction for

la rge co l l ec t ions of s a m p l e s ( Q u i n l i n , 1979) . M i c h i e ( 1 9 8 0 , p .

12) p r o v i d e s a list of e x a m p l e s of va r ious " lea rn ing" expe r t

systems.

36

DENORAL, for obtaining structural representations of organic

molecules, is the most widely used expert system. As the

knowledge acquisition bottleneck is a critical problem, a META-

DENDRAL expert system (outlined in Table 1-8) was written to

attempt to model the processes of theory formation to generate a

set of general fragmentation rules of the form used by DENDRAL.

The method use<3-by—»&'FA—E Elf&R-ftrL~is to generate, test and refine a

set of candidate rules from data of known molecule structure-

spectrum pairs. For META-DENDRAL and several of 'the other

learning expert systems, the generated rules were found to be of

high quality (Feigenbaum,, L98Q . and Michie, 1980),. ,. . . . ...

Another attempt at modeling self-learning and discovery is

the AM Program (Davis and Lenat, 1982) for discovery of

mathematical concepts, beginning with elementary ideas in set

theory. AM (outlined in Table 1-2) also uses a "generate and

test" control structure. The program searches a space of

possible conjectures that can be generated from the elementary

ideas in set theory, chooses the most interesting, and pursues

that line of reasoning. The program was successful in

rediscovering many of the fundamental notions of mathematics, but

eventually began exploring a bigger search space than the

original heuristic knowledge given to it could cope with. A more

recent project - EURISKO - is exploring how a program can devise

new heuristics to associate with new concepts as it discovers

them.

37

XII. Who is Doing It

The following is a list by category of the "principal

players" in expert systems. In each category, the listing

roughly reflects the amount of effort in expert systems at that

institution. Stanford University is the major center of effort

in expert systems.

Universities

StanfordM I T . . . . . . . . . . . . .C M U • ' ' : . ' 'and scattered efforts at perhaps a dozen other universities.

Non-Profit • • • ' • • • . . . . . . - • . • - . . . - . • • • •

SRIRANDJPL

Government

NRL AI Lab, Washington, D.C.NOSC, San Diego, CA

Industrial

FairchildSchlumbergerMachine Intelligence Corp., Sunnyvale, CAXerox PARCTexas InstrumentsTeknowlege, Palo Alto, CADECBell LabsIntelliGenetics, Palo Alto, CATRWBBNIBMHewlett Packard, Palo Alto, CAMartin Marietta, Denver, COHughesAMOCOJAYCOR, Alexandria, VAAIDS, Mt. View, CASystems Control, Inc., Palo Alto, CA

38

I

XIII. Who ia Funding It

To date, the government has been the principal source of

funds of work in expert systems. The funding sources in the

government for expert systems, roughly in decreasing order ofv

expenditure, are:

DARPANIH (National Insitutes of Health)NSFONRNLM (National Library of Medicine)

. - . . - . . ,AFOSR ...... ,.. ......... .:..-. . ,. ,.. .-- ...... :: ... -. ••.---.- •

''" ' USGSNASA

DARPA arid NIH 'have been the pr'im'ary funders of expert systems to

date.

Obtaining precise figures for funding of expert systems (ES)

is virtually impossible because ES is not carried as a separate

funding category. In addition, expert systems are often embedded

in other AI systems such as image understanding systems.

Further, with artificial intelligence becoming heavily knowledge-

oriented, a substantial portion of current AI systems and

activities can be viewed as having expert system components.

Nevertheless, a rough estimate of the current total U.S.

government yearly funding for expert systems research and

development would be in the order of 10 million dollars. Of this•

expendi tu re , app rox ima te ly several mil l ion is spent by D A R P A to

suppor t basic research.

N I H f u n d s t h e A I M ( A r t i f i c i a l I n t e l l i g e n c e i n M e d i c i n e )

ne twork ( N I H , 1980) and its users at a little over three mil l ion

do l l a r s a y e a r . This n a t i o n a l l y sha red c o m p u t i n g r e s o u r c e i s

devoted ent irely to designing AI appl icat ions for the b iomedica l

39

sciences. The community of projects using this resource is

expert systems oriented. Approximately one third of the three

million dollar expenditure in the AIM area can be considered to

be for direct research, the balance being for applications/

experimentation and system support.

NSF, focussed more on basic research, funds approximately

one million dollars per year in the expert systems area. Other

government agencies probably spend another two to three million

dollars per year to support a variety of potential applications.

Finally, government contractors using . IRAD (Independent

Research.and Development) funds (associated with their prime

contracts) probably spend another one to two million dollars a

year in this area.

40

XIV. Summary of the State-of-the-Art

Buchanan (1981, pp. 6-7) indicates that the current state of

the art in expert systems is characterized by:

* Narrow domain of expertise

Because of the difficulty in building and maintaining a

large knowledge base, the typical domain of expertise is

narrow. The principle exception is INTERNIST, for which the

knowledge base covers 500 disease diagnoses. However, this

'broad coverage*is achieved by using ' a'relatively'shallow set

of relationships between diseases and associated symptoms.

. ,,. (INTERNIST is now. being., .replace by. QADUCEUS, .wh.ich can

diagnose simultaneous .unrelated diseases).

* Limited knowledge representation languages for facts andrelations

* Relatively inflexible and styli zed input-output languages

* Stylized and limited explanations by the systems

* Laborious construction

At present, it requires a knowledge engineer to work

with a human expert to laboriously extract and structure the

information to build the knowledge base. However, once the

basic system has been built, in a few cases it has been

possible to write knowledge acquisition systems to help

extend the knowledge base by direct interaction with a human

expert, without the aid of a knowledge engineer.

* Single expert as a_ "knowledge czar."

We are currently limited in our ability to maintain

consistency among overlapping items in the knowledge base.

Therefore, though it is desirable for several experts to

41

contribute, one expert must maintain control to insure the

quality of the data base.

In addition, most systems exhibit fragile behavior at the

boundaries of their capabilities, so that occasionally even some

of the best systems come up with wrong answers. Another

limitation is that for most current systems only their builders

or other knowledge engineers can successfully operate them.

.,. .... Nevertheless,.- Randy. Davis (at IJCAI-81) observed that>"there

have been notable successes. A methodology has been developed

'for explicating informal' knowledge. ' Representing and using

empirical associations, four systems have been routinely solving

difficult problems - DENDRAL, MACSYMA, MOLGEN and PUFF - and are

in regular use. The first three all have serious users who are

only loosely coupled to the system designers. DENDRAL, which

analyzes chemical instrument data to determine the underlying

molecular structure, has been the most widely used program (see

Lindsay et al., 1980). Rl, which is used to configure VAX

computer systems, has been reported to be saving DEC several

millions of dollars per year, and is now being followed up with

XCON.

In addition, as indicated in Table 3, several dozen systems

have been built and are being experimented with.

42

XV C u r r e n t Problems and Issues

B u c h a n a n (1981, p. 11) s t a t e s , "Because of the i n c r e a s e d

e m p h a s i s o n l a r g e k n o w l e d g e b a s e s , t h e t h r e e i s s u e s o f

e x p l a n a t i o n , a c q u i s i t i o n and v a l i d a t i o n a r e b e c o m i n g c r i t i c a l

issues for expert systems."

Explanat ion

E x p l a n a t i o n is needed because use r s c a n n o t be expec ted to

know or under s t and the whole p rogram.

Knowledge 'Acquisition ' • • ' • • • • ' . • • • ; . - . ' • • • • • . - • • > • • , . - . , . • . . • . • . . . . . , _ . . - . , . . . . ..... ;

F e i g e n b a u m (1982, p. 13) s ta tes , " . . . knowledge a c q u i s i t i o n

is., the. criti.cial bottleneck problem in A r t i f i c i a l Intelligence."

Knowledge acquis i t ion is d i f f i c u l t and t ime consuming. The most

d i f f i c u l t par t is he lp ing ' the expert to ini t ial ly s t ruc tu re the

d o m a i n . The k n o w l e d g e e n g i n e e r t akes an ac t ive role in the

knowledge acquis i t ion process - in te rp re t ing and in tegra t ing the

experts answer s to quest ions, d r a w i n g analogies, posing counter-

examples , and ra is ing conceptual d i f f i cu l t i e s .

D u d a (1981, p . 2 6 4 ) o b s e r v e s , " P a s t e f f o r t s t o speed

knowledge acquisition have been along three lines: (1) to develop

smar t editors that assist in en ter ing and m o d i f y i n g rules , (2) to

deve lop an i n t e l l i g e n t i n t e r f a c e tha t c an i n t e r v i e w the e x p e r t

and f o r m u l a t e the r u l e s , and (3) to deve lop a l e a r n i n g s y s t e m

that can induce rules f r o m examples, or by reading textbooks and

papers." Duda also notes "...that it is d i f f i c u l t for experts to

desc r i be exact ly how they do w h a t they do , espec ia l ly w i t h

respec t to t h e i r use of j u d g m e n t , e x p e r i e n c e , and in tu i t ion . . .

We need to deve lop m o r e e x p r e s s i v e l a n g u a g e s tha t a l low the

e x p e r t to a r t i c u l a t e m o r e of the n u a n c e s and deta i l s of t h o u g h t

43

p r o c e s s e s . " D i v e r s e s o u r c e s o f k n o w l e d g e a r e a l so o f t e n

r e q u i r e d , but t h e r e i s c u r r e n t l y no good way to i n t e g r a t e these

sources in reaching, a solution.

A few k n o w l e d g e - a c q u i s i t i o n s y s t e m s do e x i s t , such as

TEIRESIAS, that are interact ive and semi-automatical ly steer the

expe r t to the needed piece of k n o w l e d g e to i n t r o d u c e in to the

e x p e r t s y s t e m u n d e r , deve lopment . H o w e v e r , these ex i s t ing

knowledge acquisition system's- have 'on ly been used to expand and

improve a knowledge base a f t e r a vocabulary and knowledge repre-

s e n t a t i o n had a l r eady been chosen and upon w h i c h the bas ic

knowledge base had already been built. The knowledge-acquisi t ion

problem remains ext remely d i f f i c u l t and a major imped iment .

Val idat ion

Al l complex c o m p u t e r p r o g r a m s tend to have e r r o r s and are

t h e r e f o r e d i f f i c u l t to cer t i fy . At the moment , empir ica l studies

( s u c h a s has b e e n used to v a l i d a t e M Y C I N a s a s u p e r i o r

d i a g n o s t i c i a n and t h e r a p i s t ) may be the best we can hope f o r .

H o w e v e r , the c r e d i b i l i t y of the sys t em can be inc reased if the

system is made intelligible and unders tandable , so that the user

can be made r e spons ib le for the sys t em and be able to m o d i f y i t

t o h i s o r h e r o w n s a t i s f a c t i o n . M o r e f u n d a m e n t a l l y , a

methodology of validation needs to be developed.

Other problems are:

Lack of Adequate and Appropriate Ha rdware

F e i g e n b a u m (1980, p. 10) s ta ted tha t "...applied AI is

m a c h i n e l i m i t e d . " Th i s i s still t r u e , t h o u g h special LISP

44

machines and more general large, fast computing machines are

beginning to become available.

Inadequate Special Knowledge Engineering Tools

Though software packages such as EMYCIN and OPS-5 are

beginning to become available, there is much room for improvement

and extension to capture more of the existing expert systems

approaches and architectures and make them readily available to

the new expert systems builders. Further, the concepts and' • • " • ' " " *' • • • \ * * ' ! * • * ' » • • * • • : , • •"• . . • . . . , .,

techniques thus far developed need to be sy'stematicalTy" drawn

together and synthesized into higher-order patterns, so that a

firm base- for future systems can. be built, and reinventing the

proverbial wheel can be avoided.

Orderly Development and Transfer

To capture the interest of domain experts and develop a

major expert system requires continuous funding over several

years, which has not always been available. Further, there is as

yet no orderly system in the research funding agencies for

effectively taking a successful research project and moving it

on to appropriate applications.

Shortage of Knowledge Engineers

The field is relatively new and few knowledge engineers are

currently being trained by the universities. Because of the huge

number of potential applications, shortages of knowledge

engineers currently exist and probably will continue to exist

for some time.

45

XVI Research Required

Buchanan (1981, pp. 8-14) indicates that research is re-

quired to develop:

Improved knowledge acquisition systems

Learning by example

Better explanation systems and friendlier user interfaces

More adequate knowledge engineering tools

Better expert system architectures and inference

procedures

More efficient and workable.techniques for .working with.

multiple experts and. knowledge sources '• ' ""

More adequate methods for dealing with time

The ability to make appropriate assumptions and

expectations about the world

The ability to exploit causal physical and biological

models and couple them with other knowledge

General methods for planning

Analogical reasoning

Methods for coupling formal deduction into expert systems

Parallel processing approaches

Better knowledge representation methods

46

XVII Future Trends

Figure 2 lists some of the expert system applications

currently under development. It will be observed that there

appear to be few domain or functional limitations in the

ultimate use of expert systems.

Figure 3 (based largely on Hayes-Roth IJCAI-81 Expert System

tutorial and on Feigenbaum, 1982) indicates some of the future

opportunities for expert systems. Again no obvious limitation is

It thus appears that expert systems will eventually find use

in most endeavors which require symbolic reasoning with detailed

professional knowledge - indeed most of the world's work. In the

process, there will be exposure and refinement of the previously

private knowledge in the various fields of application.

Feigenbaura (1980, p. 10) states that, "The gain to human

knowledge by making explicit the heuristic rules of a discipline

will perhaps be the most important contribution of the knowledge-

based systems approach."

On a more near-term scale, in the next few years we can

expect to see expert systems with thousands of rules. In

addition to the increasing number of rule-based systems we can

also expect to see an increasing number of non-rule based systems

as not all problems are homogeneous enough to be readily cast in

the production system framework. We can also expect much

improved explanation systems that can explain why an expert

system did what it did and what things are of importance.

By the late 80's, we can expect to see intelligent, friendly

and robust human interfaces. Much better system building tools

47

Figure 2

Expert System Applications Now Under Development

Medical diagnosis and prescription

Medical knowledge automation

' Chemical data interpretation

' Chemical and biological synthesis

' Mineral-and-oil exploration " "•"• '

Planning/scheduling

*'Signal interpretation

' Military threat assessment

Tactical targeting

Space defence

Air traffic control

Circuit diagnosis

' VLSI design

Equipment fault diagnosis

Computer configuration selection

Speech understanding

48

Figure 3

Future Opportunities for Expert Systems

Building and Construction

Design, planning, scheduling, control

Equipment

Design, monitoring, control, diagnosis, maintenancerepair, instruction.

Command and Control

..•• Intelligence-'analysis, planning, targeting, .communication

Weapon Systems * • • • • . • . • • • • • • • .

Target identification, adaptive control, electronicwarfare

Professions

(Medicine, law, accounting, management, real estate,financial, engineering)

Consulting, instruction, analysis

Education

Instruction, testing, diagnosis, concept formationand new knowledge development from experience.

Imagery

Photo interpretation, mapping, geographic problem-solving.

Software

Instruction, specification, design, production, verification,maintenance

49

Figure 3 (continued)

Home Entertainment and Advice-giving

Intelligent games, investment and finances,purchasing, shopping, intelligent informationretrieval

Intelligent Agents

To assist in the use of computer-based systems

Office Automation • • • • • • .

Intelligent systems

Process Control

Factory and plant automation

Exploration

Space, prospecting, etc.

50

should also be available. By 1990, we can anticipate knowledge

acquisition systems which, after being given a basic domain

context, can rapidly guide a human expert in forming the needed

expert system knowledge base. Somewhere around the year 2000,

we can also expect to see the beginnings of systems which semi-

autonomously develop knowledge bases from text. The result of

these developments may very well herald a maturing information

society where expert systems put experts at everyone's disposal.

In the process, production and information costs should greatly

diminish, opening up major new opportunities for societal better-

ment; • • • • • • • • » ' • - . ••• -. ' ,, • . :••', ... ... ..... ,'• _ .. ....... .. .......

51

• .• : Refe rences

1. Mich ie , Donald, "Knowledge-based Systems," Univers i ty of ILat Urbana-Champaign , Report 80-1001, Jan. 1980.

2. Fe igenbaura , E. A., "Knowledge Eng inee r ing : The Applied Sideo f A r t i f i c i a l In t e l l i gence , " C o m p u t e r Science Dept . , M e m oHPP-80-21, S t an fo rd Unive r s i ty , July, 1980.

3. F e i g e n b a u m , E. A. , " K n o w l e d g e E n g i n e e r i n g for the 1980's,"Computer Science Dept . , S tanford Un ive r s i ty , 1982.

4. N i i , H. P. , and Aie l lo , N. , "AGE ( A t t e m p t to G e n e r a l i z e ) : AKnowledge -Based P r o g r a m f o r B u i l d i n g K n o w l e d g e - B a s e dPrograms ," Proceedings of the Sixth In te rna t iona l Conf . on

. . A j^t if i£ia. 1^ I_njt &1 1 jL gjsjiae^ ( LJ-C/A I - .7. 9 ) , . . T o k-y o ,• . • A u g . , . 2 0- 2 3 ,' 1979, pp. 645-655 .

5 . B u c h a n a n , B. G. , "Resea rch on E x p e r t S y s t e m s , " S t a n f o r dUnive r s i ty Computer Science Depar tmen t , Report No. STAN-CS-81-837, 1981. . .....

6. Hayes-Roth, F., "AI The New Wave - A Technical Tutor ia l forR&D M a n a g e m e n t , " (AIAA-81-08 27) , San ta M o n i c a , CA: R a n dCorp. , 1981.

7. D u d a , R. 0. " K n o w l e d g e - B a s e d E x p e r t S y s t e m s Come of Age , "Byte , Vol. 6, No. 9, Sept. 81, pp. 238-281.

8. Heur i s t i c P r o g r a m m i n g Project 1980, Computer Science Dept.,S tandford Univers i ty , 1980.

9. S t e f i k , M., et al., "The O r g a n i z a t i o n of Expe r t S y s t e m s : APrescr ipt ive Tutorial", XEROX, Palo Alto Research Centers ,VLSI-82-1, J anuary 1982.

10. Quin l in , J. R., "Discovering Rules by Induction f r o m LargeCol lec t ions of E x a m p l e s , " in E x E e r _ t S^s_^e_mg _ijv ,tji,eMicro-Electronic Age, D. Mich ie (Ed.), Ed inburgh : Ed inbu rghUn ive r s i t y Press, 1979, pp. 168-201.

11. D a v i s , R, and L e n a t , D. B. K_££ wA r t i f i c i a l Intel l igence, New Y o r k : McGraw-Hi l l , 1982.

12. L i n d s a y , R. K., et al.,In t e l l igence^ f.o_r_ Organic C h e m i s t r y ; Tjie_ D E N D R A L P£oj.ec j:,New Y o r k : McGraw-Hil l , 1980.

13. McCar thy , J., and Hayes, P. J., "Some Philosophical Problemsf r o m the Standpoint of Ar t i f i c ia l Intelligence," in MachineIntelligence ±, Mel tze r , B., and Michie , D. (Eds.)., HalstedPress, 1969, pp. 463-502.

14. V e r e , S., Planning in £.i.me_:_ W i n d o w s and D u r a t i o n s £Activities and Goals, Pasadena, CA: JPL, Nov. 1981.

52

15. The Seeds of A r t i f i c i a l Intel l igence; S u m e x - A I M , NIH Publ.No . 30 -2071 , B e t h e s d a , MD: NIH Oiv . o f R e s e a r c h R e s o u r c e s ,March 1980.

53

W

enEcu4-1CO>

CO

cua.

IM-lO

COCJ

0

en&2wISijw

SS*

w«!HCJgpiHcoijOftEH2OCJ

wCO

«<;H

O

j303O|_J

O

DOC/3<3jPQ

W

8u1o§

eoccoCOcflcup

•OpCO3

£ipa

^

j-1

o.acfl

ca4JCJCOfa

CcuCOcupacu1-14J

J4-1

COCUr-l

5

60c

1-1

43w•H

CJp

U

M

•H3CJ

*r4U

r-lCOU

1-1p4JCJCU

<U

r-lCOP01ccu

•acu

cuccCU

"§COcucaCOcu3

•ocuc<ucuCucu*""

»rj

<D

>»•HCO

«.

CO<Ui-lCU•HCJC

•HP

60

COP

iOCO

43

^JeCO>

• 5 3rH

•fi

<l»-CO

C coOJ ^

C CO

cfl

cjo cu4J 1-1P 4^cu toCO CJcd O

caCO COcO. CO.

60C

•HV.

.

0U-l

COcu3

•COcu34-1Uca1-)ao

•acujjaCU•H

^401

4J•HjjO•HM

PL.

cup3

U4)1-1C

U

0

>*

04JCO

*H»

•

r-lO

rjOCJ

"Ocu0)CO

43

cu3CU

a-

caC

•3CO3

•H

COU

•acCO

•O01CO3

COJ=*CO)c1-1T3i-lUcu•0

o(4-1

cacur-i

cr»;A2. p^2 &OF POOR QUALITY

*^

^

.

•3

424J

C co - - - • ~~

CU 4J3 a4J cad . - . . - . . . • . - • • • • • •cu g . • •C iHCU 43a o

^3 • CW • / •. . - ' ' .'

1 . . • • . • • • • - • •c f l • . • • -

4JaoCJ

c PCU 3 • .f. CJ3 oo4JCU CO60 CP O.0 i-l

(4-1 4JCJ

0 i-l

IQPS

P-4

HM

cc{CC

C/

b<

cc£r

^f

4f |W

• .*J Zw o

W

•rt

COXr-lCO

PLI

<^

WM ' •• 52H •• T. °

• • i r« r~ pLI» r t f*s ~* p^

S M <~ =- I3r ^ « ^ ^ rt .r—4 fcH ^ O

fc_, tgF ^— t — iCO H-l <• i.

o 34-1 CU

CCUca cuCO 4-1

,a co •cu cu

CO P CO4J O COCO A

T3 ^2CJ CO

CU 1-1 4-1

4J 3 TJ

C co cuiH CU JS

rH 4JCO 3C ^ CO -rl

•H CU4J J3 COP -u aCU OCO P i-lCO CU 4J

60 CUcu 1-1 coCO P CO

1cu•a

cu o4J M-l 1-1CO O T>4-1 1CO CO P

•H 0>. CO 4J

TJ X COCO i-i -H0) CO CO4-1 C CUen cfl P

CO|4O4-)CO

•HCOC3COp4-1

PO

<4J

COcu4J

4JCO

60 '0•H^J

CO

CU COcu cuO T3

0CU iH

3CO T3CO C< cfl

1 1P P

•H CU COCJ 4-1 CU

CU 60P -O COO 4-14-1 O r-lCO 4-1 O

•H >ca coC 4-1 0)CO i-l CP 3 i-l4j cj a

•CO4-1eCU

P3CJ

T3

(H

COpCU

cuecd

cflacu

42u

Mo

COcuiH

Cfl.•H

^JCO>

HIa3

*O

24J

c

rr4J

C1-1 4J

o caCU 4J•a coO CO- pC 3 cu

CO 4Jcu r-i cu0 E

CO p4J CJ Cflcd *t"^ Cu

pcu cj cu60 CU 4Jcfl r-l 34J CU CU

O CU O> CO CJ

ap x •O T3 r-l CO

14-1 C rH CUCO CO -O•cu o ou 1-1 c

•1-1 r-l

60 3 42 cu• P S . ^

CU -H >, 4J

caCUiH3

Id

cu4=4-1fa

3

occcu»J

CO(U •P cu3 !"H

4J 43O COCU CJTTHC i-lO CUo cu

cfl

^ ftjcQ M

1 cn o0 CUCJ 3

u

cu aca cu•H -d> C!cu cuM Oi

cuTJ "O

ca caci

ca oo co•H 3

CJ CJ•H a•o 5CO CJ

4_l 13c 5O COCJ

COcu cu

CU 4JCO CJ

43 uO in

. '

B -cu4-1

54

Cd

I

CO

cu4-1CO

CO

§COHSZW

3.,jCd

Cd&M

(ig

C.

5HCO

1£c.CJ

c/

pc<JE-•<

Q. ^

•pCoJ

cs

CdW• 3pQ

wBwI

cCU

I-l•o4JBcu

CO• 3

4J

01T3Os

. •. -4J• B

CU•u34-1CO

01. 60

•o01r-\

. . soB

^

£a

0

aCJ

^*

T3B

• . ._

icd<*-i

OlCOcd

•HCO0B00

0160B

5.CJ

Ol4-1B

•H

- .4 -1 .pj0)

3CJ '

COcuiH3M

i-lcdo•H4-1

•ocd0 MO 0<N M-l

CO34Jcd4-1CO

jd

u1-1[J

000

1-1cd•H•a01

•H•a•H

Ol

'

0?aOF

o

i-ioCO•H

4J.0

CO

01

. CU

4-1

g

•a4JCO

"o- ••::-. --i!-i

r.LlJ

FCCR QUALITY

CU00B

43CJl-iCU4-1B•H

i-HCdCJ

•HT)

CUeooB•H4-1

resen

Cu

1

CO.CU1-100CU4-1COM4-1CO

u•H4-1COO

oo•HT3

«

a0)..

T334-1CO

cd00Bi-l4JO3M4J i-lCO 0)g,-go g1

a)4-1

o4J

V .

'

•co . . . , . . . • • . . ;: - . , . . •B • . . . . . . .O

•H4J 'cd • ••H4J•H . ' -

OO-rH

•H CO '4J

B B0 01CUTSOB 3CU 4J

1CO

4J•HCOM01

t-lC

•aMO

U-lCcd4J09

o

UCJ

c/J5 BU M O HH H — OC/> CO r-i 2>• z s r>CO M < fo

cd cd*^cu -o

COCd i-H

43 Cdo

CU 1-1oo -a•a cucu giH3 00O B

53* sb cu

Mu oO M-l

e caM 01

01 TJCO T3

Mo

14-1

cacu

co e4J CUCJ •-!Cd 43 i-l

VM o 00 00M B CU

4= CUrH 4Jy > cdCO T> r-l VJ0) B O 4JH cd co (o

00

00-HB 4J

•H cdco cuO MB 4-100

i-l eu4-1 4J•H CJ00 COB 43

B cao n

•H (04J >,CJ r-l

1 - I B C U B M 3 3 B

55

w

i0)

CO

4-1MOIO.

14-1

O

MOI4-1OcdMCO

CJ

c

CO

2MPiCO

0^QOI

Z0

o1-1•HCO

3crCJ

0160•a0)

^H30

E- •• 253 CO Oz: H pfi M

UJ .M O EHH E- 23 CJ.CO CO H Z>• 2: 3 3CO rH < b

Cu0

COHZ

S

SW

5ws

£ECJcr_

ff

^pt

%oCJ

Wcr

««!E-<

a

Pc0

wo<PCwocbito^^

_CO

M 3 CCU 0 -H•H 1-1U-l M CO

M ' i-l Cd CJ •o *o > M co4J O CO ^•H SO) M•o oi -a ocu . •> 4.) C 3co oi cd w

! .* oi i-i aiM 4-i cu en eO CO T3 CU3 ai -a oi4J M M O X!01 O O' C 4-1 .Z 1

CO

1 *

cd oi-o oico a.U go

•H CJ

45- CD- - '•j_.

CO O

' CU 160 60•a w -a0) ^ 3 •

3 0 <l~> 60o 3 ao-oC 4J C CD

,*i 01 -H rH.C co 3

M CO O0 0) 01 C4-1 CJ M &o no.01 01 X rHQ. • M OI cdCO OI 01 4-1O CO U-l M CM cd C O OIPu .0 M M-l S

CU C C. C O - H O

CO CO i-l CXi en 4-i 1-1. 1 OI CJ cfl •a> g j_, co g M60 gs o, M „ 0•a P- x oi oi co jj

B IM • • O-*1

oi 1-1 o -cu -..60 60 Off

C CO M 0) CO • t-'f0) 3 cu co 3> 60 CO 3 60

•H C 3 CM CO M cd•a IH w .cu i-i

1 .O. O,

cO C O M B

f ee M a. cdB £u 3

CO g CU gM o o x: o

CJ CJ 4-1 4J CJ

* * .

60 • UM C -CO ... -HO i-l CU 0)

U-l C W r-l cd CU CCO C 3 en XI X: -H

CO 0) O M ~& 4J efl

M 4J 01 0 C 60 0o cu 1-1 x: 3 1-1 c -a3 XI CO 4J 4-1 4-1 O4J 4-1 O 01 C 6 01oi a. c c cu co x;C 60 O -H CO 4J

CM CJ CU 01O i-l O. t3 i-l M 60 C

•H Cfl CU S O. *O -H4J CO 01 >. O 01 01 •c cu x; o c M r-i enCfl M 4-1 rH O 38

§ a. a. x M o MX **-i B • CO O C CU

CO 01. O 0) H >4H AJ 4J

Xo34JCU

1 "^ & iVt ' '. •"' jp' *."• "/"Iv2ii .***:&• 8 ~v«i

.

; p ".oi-j 'j«-3

POOR QUALITY

• ... .

60

CO i-l3 -a coO 01 3

•H >, MM o o -aCfl r-l >4-l C

g pdUH 01 O 600 H C

co cj 1-1 .ai B W 4J <u60 CO P* C 60•a -H co oi -a01 C O CO CU

rH CO PS 01 iH3 J2 PL, M 30 cj o-oC 01 C 0) C« g -H M J<!

. . • • ' . . .

COB •. . • . • .CO•Hc2 . . rCJcusCJc01

CO•HCOc0

CJ

asCJ

ccdPUPu

• <i

Ci3COoPip-

,_]30c

OAogHCJ

PuCOo£

60c1-1CO

0)

CU

COH

CO

CO•rl>MCUQ.

c•H

COs

1-1ccdX!(J

§3B

'o0)60

CU

"c

*A

"ccd

0

4-1CJcdM014-1

M0

U-l

J2MQ

34J01c

T3

CO

T3

OHCJ

P-!

CO

4-1M01O.X

flCO

J24-1

CiH

)O

<4H

CO0)iH3M

•Oa

0)60

"SrH

0

^

0)c

M0

60C

T3r-l•H3

f*

H

•4JMCUexOI

c•HcdE

T3

C

§3

cdX!

•H3

cd00

•H4-1CcuB60

*;0)4=4-1

M

01*o•HCOCOCJ

4-1

CO•H

0)

•Hc

0)4-1

e01

CO

CO

4JMOIG.

*T301

0)CO0a.3p.

rHCOM01

0160

CO

CO

Bcu4J

•HU-l•HCJ01 ,.a. . .CO .

0)^34j

U-lO

0)60

T3

iH

O

5

COI

X-HCO

OICOCOXi

60

OIrH30

60

1 .^ CUM rHo a M oi4J M U-l HOI CJc -a w

• 01 CUcd B « co

M C O

COiH 3i-l CO O>.MT<U-I SCW

56

OO1

3-i

COeCU

CO

en

t-icuCUX

cu

Iw

COCJ

1-1cu4-1uCOr<

O

a. Z

4-1•HCOMCU

|

(-1O

BCO

r*CO9-V5

H10

cc4:c.

W3&.CXe-t

cuo

MCU

i00

rn

Zor-<

SZ3U.

112HW

•jOoic8

u

•<CQ

3o

•3

§3.

••••

CdC/3

M

r_J

S

h_25o2

COM2 .B *eu woo co

cuB"CflTJrH BCU CO

4-1

cu4-1B0CJ

T)cu•Hi-HCUCU-

CO

r-l

<UCOp-. . _

.•

^40

14-1

COcurH

A

cfl4-1

*.

4J'3O.BI-*

rH•Cflrl4-1OCUCUCO

ooBiH4-1CUrl0.l-lCU4-1

Bi-l

1

ooB

•HN

v4}^Cflg§3CO

•oBCO

CO

CO•a

i£4

34JU34-1

-CQ

5oB

1

•CO4JrH3COCU

cu4JCO

iHco -aB B

•H cflcO UCU

CU8 rH.3 .0

4-1 CO0 3CU COCU rHCO CU

. . 1

'cu '4-1 COCO CU

•O rH•H. 3•O »-iBcd (30 0

•H00 4-1B CO

•H 4J4J BCO CU

cu i)B cfl

60<4H

ORlOF

acu3M

BO•H4Jat4-1.B ..(11

i00CO

g^

•

0)uBCU•a•HKCU •

cuJC4J

B0M

POOR 'QUALITY

rlO

00B

•HN

THrHCflt-lCUBCU00

1•H•aBcflU

00B•HN

•HrH

COiHCJ(U(XCO

• s .

•

.

. . . . . . . . . . . . . . . '

|4 " " * '"" •" • ' • ' - ' • • • ' • • : • • • • > • .•CU4J4J •CU- CU

JS CJB

o cu -*J T3

•HCO ' ^cu cu

rH3 0)U X

4-1cuw 4JCfl *H

*O iw

EC

1«CU

*

til

PUR

POS1

<u14-1 rH0 3

U4J CUCU rHCO O

BCO

a) 3 •B o to•H B MM-l ^6 *H0) COrl B CU

O•a ^ 6CO rl

CO 4J4J CU CJCO rH CUcu 3 cu4J )-l CO

« CU CUCU 4-1 Vl4J Cfl 30) -O 4JM -H Ucu -o 3B B rlCU Cfl 4-1O 'CJ CO

rHeg ca B cu

M 0 JSCU CU iH 4J4J B 4J(0 CU CO <4-lM 00 4J OCU BB 14-1 CU «CU 0 S COoo oo cu

4J CO rHO CU U 3H CO <4H lJ

Bcu B

>> > 9JZ -H 0

00 Bcu -CO iJ 14-1S3 °s a caM Z 4JO td <U14-1 a co

• •

COM•Hcfl

1 Cu(U

s I4J rlCJ 4-13 CJi-i cu4J p.co ca

57

wJw

enEcu4-1CO>,

CO

0)Q.

01iHcx

03O

o2ca

o)60Cd

CO

S HSJ Mr- HCO CO>< zGO M

CCO60edF*

CO02Oas

CO4-1cd

I4.)

i-((001

oo

zO

Ci.g

c*—5a.

•pc0!E"gCCJ

Wfa CO0 <!

pa

1 aW *

S . £

§ 3^ Sfw oW ,Js* o

Wc/:^PO

HC

S^S

^J

^

aPuQJH-

• ^

C=3WOP-iOS

P-i

S OR!O

** OF420l-t -Ocd oi

• Ol IH •C CO 01 CO01 • . 60 B> 01 01 60 Oi-T > 4J i-l -HU i-l CO M 4J

' T3 4-1 4J 4J CdCO CO 4J

4J 3 CU CJC CO 4= 4J 01o) 43 o ca a.> X CO 4J Xu u oi co oi

*->• >>i • o) e i-i i-iCO 4J 4J O 1 O 1 O-Old CO -H fi <4H rJ 4J601 4 J 4 J ' 3 3 co •

•H CO ed CD CJ -H 60U 4J 3 T3 4J 42 C

• H c d *J4J B 1-1 s 1-1•aa. C-H co B i-t 4J MO 01 CO tH CO•HIM M co r-i en ai 4JMO V- l M C 4J. iH-H01 3 O O O I B O I 4J CO, CO CJ TH r-l CU 4-1 CdO

4J . 4J . 4 J 4 3 B c d - D . B ->4-l C 4 J X C O C d O l 4 JO Ol BO) 4 J 4 - I M C O 4 J 6 0

B OI4J C J C X 3 B Bco a) -H C <u cu co 4-i oii-i4J»- i 4 j o n - c j c d c C J M013 nj CJ X O 01 <U 013

c o c a a. ^ w co B VH • rt -o

CO01

•~J^13

ca MOl

_^ r^r— 1 (2 •3 O UM fl CO 01

4J Ol i-(B cd i-l 3O N 3 Mi-l -H V4-1 i-l >%•H cd co aco i-l 3 cda 4J 4_i ucd -H cd oiH C 4J AH M CO EH

B 0)01 -H 4J4-> SI I • Cdcd C cd o >% 4J4-1 01 01 -H P. CO <4Hca 4J B • ca co o

0) 01 ^ M (IH4J T3 0) 4J 4 = 0 1 i H 4 J -C i-H CO p. 43 01 0101 O 4 3 4 J 4J Ol CO >

•H 4J Cd CO 01 01 -H4-1 4J > 6 CO 3 J-lC O C O O - 4 J i - l l - l C U MO. 01 0) co M O B c d

-— s iH CJ 43 01 01 4JI-HB 3 O 4 J T 3 C O o i c oc d o M c d 3 c a B u•rt -rl ClH O O I - H B4-1 4J B3O 4- IV l i-l 4-1 Ol•H cd O >»H O 3 BB 3 -H-O CO<4H M 4 3 0 I•H4J 4 J C C O 0 1 C J J j

•H C d C 0 4 J r-l CO B c d 3C tn N iH 30) o o i t aCd - H C O B IH4J iH Cd

IH i H B - H cfl 4 J T S O I4 3 Q C O O i - H C04- I - H O J S4-1 in 1-1 3 ca co eo*TH U -W iJ -U ^J £« C2 O3 X - H c d B cdi-H C O c O - H

01 B 4 - ) O I 4J CO M 4 3 - O4 J 4 J - H O B C O C J 4J CJ OMB 01 01 iH i-lcdo 0) p. rJ B60 B C O M4- i c j W X 3 3O 3 CO 01CO^ S Ol CO pi rH Cd 43 O.

• • • • •i-H CN CO «<T

0) 1 rH43 60<4H ed4-1 iH O Cd O

CO iH 60 • •4J 01 B 60 B MH 60O l i - H C J O O - H O BM C d S l H r H l H 4J-Hc x u c a > 4 - i o o B B ^ 31-l iHCJ ^H 4-1 (U 0) 4JO I B i H c d C O i H 4 J - H C d4 J i H > t H 4 J > > B C 0 4 J O IB r H . i H C 0 4 3 O > , C d MM o C T 3 a . E c n o . 4 3

GiKAL PA3E jgrPOOR QUALITY

• .

•as .. .. : . ..4-1

CO4Jcai-HcdCJ•H60o - • . •i-Ho

•HCO>,43PH • • • •

f

014-1cd4-1CO

s0)B

43CJ

S - .•M0

<4H

enT3BCO

CN

4Jcd0)Cu01

06 • •» • . . . .

."">

58

co><

CO

AJW<uo.

orHI

Wt

ocdMcd

CJ

oU3

5=

5sw(JW

SM

S

WcdHuOSHCA

CO

NTR

OL

[V]

cq

-^SQ

_

"Jsoo

w«5" *CQ

HOQW

gQ

S

CU4J

cdM01a0160

rHCOy

iera

rchi

X

•cd4J

T3

4JC

. 01

• • ^4

3CO

Ol

' COCU

i-H3l-i

CO•H4Jy0)

i-ioytflcda

CO60

•H CC CO3 4J •

4-1 p CO CflCO (X TH OlOl CO JS4-> x c y

i-l- O cfl•a ia y oC cd C Mcd W -H -p.

•CO4Jccd4J

CO MH CO4-» O CCO O

•O 4J yOl

T3 CO Ma> o4J -O 4J

• . -- tj - ' • »- o) - (0 *•' •<U > laI-l -H CUri VJ CO CU 01CJ Q 60

• • 60 ' W '.- •••C 4Ji-4 CC -H

•tH Ua co •B o l-i coV4 W Q.)Ol 4J CO r-l4J 01 C 301 CO O M

rH £3

O 1H O •*

T4 Cfl CO

Ol iH 01 01

3 C CU 01« Cfl 60 0

OR1GJ&AL ?£££ ---5OF POOR QUALITY

*

.

I-lcfl

CO rHC 3o o•H 014J rH . - -

3 0rH e •O COCO 01 CU

4-1 MrH Cfl 3W : T3 4J •• . -. ', .-,. . - . . • . . , - . . ^ - • - . .•H tH y . • • • . . . . . . . . . .4-1 tJ 3 •

Cfl Cfl 4-1a, cj ca

coj " '" ; • " " • "• • • '•• • •' • • • » • . .•H 4J M

cfl co60 *a eo i-+C -H c 3

c c . 4J at3 CO CO r-lM U CU OCX 4J B •

4-1 CDi-i c M oi aiO . cfl • o 4J la

CO Ol "O 4J

. S "w co S ^ 33 O i-l 3 cfl 4J

rH

O

•P

•ala0

14-1gcd4J

w

•»14(4C

6

C0•H4-1cd01Mala014JC

: Hti cd1 4JJ Cfl1 O

H •• Z•• S 00 OS H =i MU3 M O HH H — CJCA CA -E-> Z

CO M < fi

goena.^*

1t^

^w

ca4JCfl•a01A4J

Ole3•H14-1CU

OS

r4

1

S0ucdMOl

t4_t

CM

lacfl

3ycu

rHOeOl4JCUrHCX

Cflla01S0)60

UH0

Jj

CUcarHCO•rl4-1•HeiH

(0

cuc•rl£jaCU4J01o

CM

SoM

tta

cu34Jy3MUca

•CO4-1cd•o

1M

(4H

CO4-1a

•iHCOla4JCOcoy

UaO

4JC01 la

01 rHM 33 yCO OlCd rHOl Oe 6

ou

ca4JciHcdM4JCOeQy

^Scdcd4Jcd•a01ciH

e0

c^i

4JrH9COOlla

CO01y01iHex

•

CO01la

4Jy34JcaOl4J

cd•H•accfly

Ol4J

CO

0)C0160

1•H13

eola

U-i

ooC

•tH

60C

V43•aCOOlnycaoexex

' cfl

Ol4-1cfl

T3CcdyOl

3la

cm

*

cca>.

.

O CU•H S

CO NOl C60 CU

1C0yc

CUlacfl

4JCO

4J

COCO01yOexc0

•H4-1<0la01C0160

V4O

<4H

COCU

r_|3

rHCOlaOl

Ol60

A4J•H34JCcfl4Jca•HCO

•COCUla34-1y34-1CO

laCO

3ycui-H

B

01

^j

>4HiH

CUCUCO

o4J

COOl4-1 •cd cfl

•a 4J•H Cd•a T}G

y MHCO

4-1 iHCO 4-1Ol cdH cfl

in.

59

0)8cu4409>,to

(I)CUXw

"l-lrH <U

I <-l-H Cu

H eg

l-l<uuocflMefl

CJ

§C/lH52

•Sw£S

b.ee

Pc-N-

OSHcn

V-

oEHZO

' CJ

wCfl<pa•<H<Q

<ff

O

UCfl<;«u0ou£oZM

•acu44ocuVJ•rl•arH

COoo

rHefloo

CO4-1ccuBa00•HcaCOcfl

>,4-1•tHrHcaCJ•H44•HVlCJ

1

oo5c•HcaCJ

•aidca?.5CJCO

43N_X

44O

43Ol-l

C•rl

CO44CCU§l-lcu

(4-1O

At

•

X— NrHCUJ>CU

•

CUd•HU-l0)h

60

•HCO

COacdrH

str

act

42cdrHcdo

r-l CUfH

42CJCflCU

44Cfl

H-lO

cu^Jcfl4JCO

rH

4344•HaM

•c

TH

g|O

•O

00

|

pjcflrHP.

1

>%44•HrHcdCJ•H4j^H(j.y^^

§44CO>,CO

§O

•O

CuOH

^^

§§

THrlcdB

44CJ

CU4344

<4-lO

Co

•H44cdt-l

00TH

>4Hdo

CJ

f*>

>4HO

44Ccu

•B

•CO80O)H

4=O«MwMCUT-l

4=

•

Mcd

*4-4

CO3

4=44

Q)

cflrHPL,

•CO0)CJ0)o.CO

4=orlCOcuCO

,ol-lCu

>H•Hcu

4344

'O

CO

CO44ocu1-)

43o

QRSGIMAL PASS '3OF POOR QUAUTY

*

' * ....

*

>S. 4J•Hr-(cdCJ

•H4J • *•fj•Mj^O -•uc - . . • •CU t-Hl-l CUrl >

3 CUCJ rH

- . 6 0 ' CO . . • •C ' C CU•H iH rH .CO 4J 3B d • rt

O CU rH•O B <U 43.

cu > y • ^UJ M CU rl rH .

42 O i-H cfl CU44 CU CU >

•a >» co cuC 44 rHiH rl iH O

O rH iH 43 'CO <4H CB 44 O •cu u ca co•H ca iH . -H eu44 CU 44^ h .

IH rH 1-i 3 rlCU 3 M CU Ocu cd u n M-I

CO

1 ' e . -2-- '

si

H•• 3 '-Ox: H ^W I-H OH H Scn cfl E-tX 3 2Cfl M <

ZO

HCJ

U.

.<004PM

Si<

PJCOoSSPJ

1 1cu cu a oo 8

T3 43 3 CT S44 (0 -H i-H

44 M 43co a a cu oM o 5J -a v

•H 43 TH CU<4H -a 44 CO

cu c cu>, CO • O 43

43 cd B CJ 4443 CU

00 rH >, >4HC C 43 43 O

•H cd oC rH l-l 44 COC O. Cu TH 44ca o

rH rH CU 00 CUCu CU 43 C Cu

> 44 TH COi-H CU C CdC8 i-H 14-1 iHCJ O H-l rH

i-l Cu <U Cfl43 O CO I-i CJO 4-1 44 iHi-l CJ >. 44CO CO CU rH lHM O. CU heu oo co > cj cu

TH C CO -rl Cu43 iH CO CO -H

CO >» CO CO CJO -H 01 01 CU CUQ > M CJ rH fii

oCO 44 COC ca eCO 44 44 O

rH O CJ OO.43 CU M

O t-iCfl l-l 43 dcu o cuca co cu•H CU 3> rl > 44CU O O <U

O•H443i-l

SMO

I4H

COrHCU>CUrH

Co•H44Ucflp44 •CO ^

43 COCfl d

COX rH

•H Cub •**•

rH

.-

0 ^13 a•H

CUUH0 TH44 CJ

(UCO CU•a co(UCU 44cj cao orl B

CuO

e 4-1o

•H 4444 O3 cfl

rH UO 44CO CO

43B a5rH 4443 COO OM 8

PLi -

•

CM

•o •SCO

rH •cu » - •> 0CU rH

i-H CU43

CUC rHO CU

44 CUCfl rH

d 440 XiH CU44 d

rH OO 44CO

fll

cu £4J gcu srH B

O. C8 50 43

CJ 44

•

cn

__ • '

Q M-I 6 43

60

I

CO

01U0)>>co

feocoHztdwE3*JW

JnM

3

183OSHCO

rH0OSEH530U

Cden<pa<E-<

.O

*-

at»o

&2(2

Cs

PJ-

5CpP6

.

Cg24J•H

§

OCJ

4-1COcd01rJ .

COrHCd000

3CO

•COrl0t 1

cdl-l01a.O

00c£

1

cd •4-u

•acd3

CJcd

- , _ , - •"4-1 Cd0 0

00oo .aC 3 .

•H COl-lCU G

T3 TH)-lO CO

VlrH 0Cd 4J •

•H Cd CO' 4J >-l C

i-i cu cdCd CUrHP-, o a.

00c ciH 0)N CU

TH 3

00 010 43o0) 01M CJ

r«t-i cu0 rl

>4H OlMl ,

CO h CO.01 CU GI-H w td3 (3 rH

«S -H (X

OF'* POOR QUALJTY•.

*

*.

•

. . • '

c01

.01 . . •34J01

,jQ

Oloc . •l-l01

V4 CO .

CU C4J CdC rHM 0.

00C

•H

rHO •CO CO01 CU)-i U

cl-l CU0 M

CO h01 OlrH 4-1

*TH .

X<0z

MosCO

•H4-1O•ai-iCUocd

00

zo

0c

•r1

CcCC

pi

-

HH "H •• Z

•• 3 CO OS E-< Ci MW M O HH H S UCO CO H &

CO M <C fa

3;*•-

cPS

P-^

pjCOoPuQM|~1

PM

COCcd

rHCL.

rHcd

•a•H

TH•d•!-(

rH01

rHrHcdcda.

^^•oscdO.

w

ooc

THcccd

i-HPL,

4JO

43

<§

,rHrHcdiH4J•He

•H

4-13

,jQ

ACOrHcdO00

43

CO

00G•H4J

cd

cu4JciH

Mo

M-l

CQ

Goo

TH

U)COcd

E014JCO

CO

o4J

00G

•Hl-l01-ao101e•H4J

rHcd•H4-1l-lcda.cdi

rHGO

a00

THCOCOcd

bU

•HI-i0)•aoi01E

•H4J

cd

0)CJGCU}4

0)14-1VlCU4JGiH

G

13

0.04JCO

.COrlOUcdrlCUO.0

•H

COl-lo4-1cd0)a.oo4J

COiH

COC

•tdi-HexrHcdO00

433CO

rHcd

•H

l-lcdex01

4J

COl01

4J0)

*l,

ccd

i-Hcxcd

14-1O

00c•rll-lCU•ol-loCUf.4->

4-1CO3

cd»rjCcdM

•a0)

MCUCO

0

CU4=

01

rH0COCU

O4J

•a0)

T3Ol01cCOcdCOvioUcdIH0)a.oCU

4=4J

•

CUUcCUIHCU

I4H

0)UC

•H

•

61

cnrHI

1

COEcu4-1COX

CO

Vicua.

exIM-lotoCJ

CO•HVIcuuo

ca43O

j2WOnJ

^s

3

MO

fj

. cd4JCO

•K

•HM-lCU4JCA

0

Coo

•HCOCUo

H •• Z•• 3 CO OZ H P- rHW M O HH H K 0Cfl C/3 H 5

C/5 M < CJL.

G

faO

en55pj§Pjw

^4W*yj

•

4Jc0)

•a

woA3E-iCJ

ft£-03

»J

§E-

cCJ

Wr/B^t-«(

a

^(Xc{;

Ci

^jpcEx

t:t-t-ij***

orHCU

CU•oVi4)•o§

TJ

CO

by F

riedl

AP

PR

OA

CH

GO•HCOVlcu^

g00rH0

Vicu4j r*io coc o< CL,* 6!S

px.

O

Ucd00oexoMex4JG

TSM4JCOCOo

COco

•HU3

rH .0CO

rHCfl•H4-1Vlcd

Cu

VirH CUcu a> 0curH OC

1 Ccd i-l4J >CU rHS 0ca4-1 1•H ey cu

•H rHrH 43ex oX Mw ex

i433CO

gcu4-1CU43 •

CO

era

ctions

onstr

ain

t

4J Oa

•H COcd

4JG COcu SCO CUCU rHVi 43ex ocu vi06 °"

00 (_,G cdiH rHC 300 a

"H fliCO ,ujcu oo e

•4JCcuE4J4-1 ••rl

o

CJ

4-1COCflCUrJ

CO, •CU COCO 4JCO CJcu cu3 M 100 M-l

0)M-lO Vl

>, cuVl 430 4J

• 4JCO -O

•H GX cfl

434-1•H5

G0COcfl •CU COVl 4-1

c50 i-l4-1 CO

MCO 4-1Vi co0 G4J Ocd o

cu^o4J

CO

cdo00

COcfl

str

ain

ts

GOCJ

cu4-1CflrH3

gO

fa

CO4-1Ccu

u S•H -HU \j

cu cuC CX<u Xoo cu

,

00c

tHCOCOCD300

o•H4JCO•H

• Vl

<UP3

•

••«cu

0CO

. •00

•H*rftCJcd

- Vl4J

iCJcd

4J13cdcurHCU

OS

•CO4JeCO

4JCOCoCJ

COCUrH3 00Vl C

•Hoo caG CO

•H CU> 3iH 00Oco vt1 0

rH CO43 CUO rHVl 3CU K

rHCOCU

^ *CU COVl S

<u0 rH

4-> 43O

git•H 434J 3

t pro

poga

be

twe

en s

c•H COcd GM 04-1 1-tCO 4JG CJo cdU M

<UCU 4JCO G3 -H

O4J

COCU 1

3 OVi CJ1cd 43W U

CU -H .S 3 co

4JM-4 C GO O i-l

CO cflCU cfl ViCO CU 4J3 Vi CO

03G G 1

•H CU Cvi cu ocu 3 cj^ WO CU cflCJ 43 -HCO >

•H CO•a e coVl *H CUO 4J rH

M-l U 43. cfl O

CO Vl Mcu cu arH 4J 43s a s

«5 -H CO

•a. "

^s 1-flvi vi co03 CJ 4-1CO T3 CCO -HCU CO CflU i-l ViCU 4Jc e co

0 GCfl -H OCfl 4J 0

cd00 8 M-l

em

-solv

inent

info

nrc

hange

o

rH -H 0)43 CJ 4J0 i-l GVl M-l iHCXM-I

3 CU"O CO 43G 4JCU rH(3.1-1 6CO 4J O3 C VI

CO 3 M-l

1cug

•H

cu(_l

<rHCOy-r44=y(jcfl «Vl 4Jcu e

•H CU33 S

.

GO

4JCflVi

'00oex0Viex4-1

•HcflVi4-1CO .

1

^*cuy

•H4JCO•HC

4JVloexex0

A

4J

CUe4-14J-Hc§0 •O ^

. e•u oCO -HCO COCU G

rH Cfl

B

eo. !

y3

"O•CUVl

• * cutj

cucu

•HP

COcu eou G•H iHO rH

43 rHy CU

ex^ o

CO U

e g4-1

cu

guessin

go

the

rwis

e

y co•H iH4-1CO CU•H ViVi CU3 43CU 4J

43

cu cu3 3

(3RiG""'^''-L p. 'i? i1^1"tt^'*^-*B « i*"""'**.-!. f.^-.-^

• OF POOR QUALITY

QCO

OT)

O4J

CoCOcflCUVl

•

•.

ecdccu3

VlCOCOCOcuocu13

COCO

COcuCOCOcu00

4-1yCOVi4J

(*

'

1

.

• '

.

,

•oCUVlCU4JC

OCJ

g

CO•H

§rH430ViexcurH43Cfl

rHOCOCUVic

62

cua.Xw

Pn0

COH[jl

2-i

Cd

S

g

1§H

,J0

E2;oU

Ed

<3pc

^JpH

Q•

»

W

fc

^***iff

bpb512fe<

00c •

1-1C!0COcdcuM-

• 73

cd5

o

•4-1•H

CJ

1-1o

CO5

<0CJ

•HI-l4-1CJcu

i-Hu

14J

§73CO CUCU 73M CUCu CUcu e

COoo tdC

•H COoo aC 0Cd vt

J3 4-1cj cd

cuMcu

73•HCOPJ

oCJ

ooC!

•Hcu

JO

0)o•HiHCO

oo•H00d

50

uo

caca cucu u

r~4 *H3 rH

oa ca

ca•rlcoX

r-4COecdCU2atH•Ueouo4-1

73cuCJ3•acu

73

U-lO

4JCU

CO

00c

•H4-1cdcuVJo

0<4-l

car«4

«

QRJG^AL FAQS 13 -OF POOR QUALITY

i

.

• .CO(U

iHCO>

COcu • *u

1-4f-l . •

CO

cucd

cuocuCucd

1-101cu

CUQ

CO

H

g

cu"cucu

CO

cTcdscaCO

CO

z

c00•HcaCU

O iM . .

™

H •• Z•• 3 CO OS H £<: MW M O HH H S CJcn co 6-1 ~z

CO

gCD-PU0^<

'-'.

c/:OCuOi3ft.

•

=14-40

cacdcu

73•H

CO

1-14J

cd00oCUoa.CUCOp

ca•H

4J CO•H <U3 x:CJ 4Jh C•H ^cj ca

oCOc0

•H4J<d4-1aCUCOcuMacuM

4Jacu•-I<d•H3crcu04-1

ACJ4-1•rl3

CO

1Mcu>0

O4-1

730)73CUCUeecu

4=3

COC0•H4JMOa4J1-13o^11-1CJ

COcuc•HsMCU4-1CU

731

«-4

oCO•H4JCOooOaoIMacux4J

C•H

COCU00

^&CJoiHJQ

cuEoCJ

V4

o

CO<u3

1-1cd>

CU•H4-1r-4.

g

1scacuu•HrHCO

=

CUCOcu

JSH

*CO4-1e•H

CO

4-1COcou

ca4-1acucQCu

0CJ

cuM

CU73•H>0I-lCu

1caa0fl4JI-loCu

4J1-13CJ

1-1CJ

M-loca3cu•H>

cdCJ

•H^44-1CJCU

<-(cuc•H

cu>cdI-l4J

o4-1

ao1-14JcdGA

oM-lc1-1

I-lo

M-l

car^4-1

cflCU

4-1ccd

73c3

•CO4J•H3o

1-4o

1iH

g

<4-l

oCOAu00cCUV44-1cacu£4-1

cuc

-HXI6oCJ

caCUCJ•H!-!.

CO

•s**i

CO1-1cu

73oecu

•H -Cw1-14J

63

OF POOR QUALJTf,

s*^5sH

^js2PS

g58

•C/3

CO

Hj

^

^1CJ

[i

PC

(£C

§

2

1a a.03 Mu B

B . 01.IM 0 S 4SO <-> . U 4J 4J

*J -H *J 13BO- U • -H i-lO 43 CO CO 60 H 3•H B i-l B B 1w 73 i-l B i-t O 0)

S B CO 3 i-l U i-4 •0) CO W 3 43 4=

i-l 01 M 73 U eg U43 B O O 01 CO i-i fclB 3 O ChjS eg M egO O u tx.cj 01 eg oi

CJ 73 D. O CO »4 > 03

•-I ' <M•01 r-l 0) O> eg s>01 rl 01 B•H 4J r-l lH •

B 01 Ui- oi B > eg

13 0) 01 IM6 -H 4> f-lO J» 3 0

01 4J X • 309 13 a) u m js01 h IM 43 -H Ol Ua eg o i-l caat o <J IH ai • co

43 43 13 hi 43 43 UU j£. U O -H U rHo a o a. 73 o 3a. a u o. ai a. coXi-i oi 3 MX egX 43 Ki CO CJ 43 CU

• R01 CO4J 0101 CO

co B 0)00 DO B I-l U -73 B Ol eg - o cooi 1-1 § a. to a, .ui-l u So u x B3 eg oi r-i B 43 oio oi to eg oi BB H B B Ol 60

M D 73 60 01 iH 01Ol i-l tl 43 CO

01 H r-l CO 3 CO60 O 01 CO r-l 0eg IM 43 M eg rH o .3 eg o at -X hi60 •-< IM B co IMB <n3 % * ~

^

iTJB •01 COa. B

oIM rlO 'U• egni >*•*3 U

ISeg coB01 00oo a^

ohi

IM

CO01a01..

u •O 00D. 0)Xr-l

43 43cd

2 >»3 co

*>

%rf« " —

•

.

•

•* "• • >

0. 6 73 • •3 oi aH S W *"O U 0) 01U 01 b COU 43 -H CD •O 0) 43 .43 CO O. U

01 Oeo 1-1 oi a. •> U CO X

B eg js§ 0) U

U 43 Uo co a B7-H r oi

CO 13 Ba s M GoO O O 01H O 3 oo •

3 . S -3 B Sx 3 §S JIS 'ff x i ,» i gSi i

J S O O O J M T < 4 S 0 0 0 1 * 2 S < uO I l M f f l i H w U i H 1 0 0 . . . S S S

8 ' 5 °.S $£ S o -"S TJ.S » o 4 ? 5 S .§ g . S ^ ^ a . B i 3 § § o i £ T & I 4 '3 » X O 6 0 B O B C L « i J . e 5 ' i

ITS x 25 8 3 3 . S " S o T 'S^ 2 3« S^i;i 3 II J3I! s ss . 8 - 8 s 5s ss§1 1 U H3l"~m || |.| . 52 ill• • • • c o 5 9 i ' i

- 3 > I A ^ O r * . « 1 | « 1 1

co

5.8.i-l U 01a) eg u §y M oj a

•H M hiX to ca «so 43 oi a.u co oeg B uM CO 01 Bai 3_ — 41

O B BJS 01 41 01

B to hia. oi 33 6043 CO

O 4J egB M rl 01

CO 13CO C01 Su.U'O Xb Ol

O J3 4J

7 0 ega -oa o -H

O M T5•u a Ba. egT3 CO O

cd B i-iO eg

D.-H -H

r-i a.a* oi .> 01 00

*O a4J CO

•H

"w8

fi-SaB eg o0 1-1ca 73 ueg oi eg01 co uhi co oi

3 hIM u O.O O hi

.a.VI

oo0160

X(d

BO

CO01CD0)

O

.5

>

CD

cctu

•s•a01

W01

J3

Uoo.o.

B01

U W I U i-l IB "M -H -H •

IM eo o 43 w co0 B O 00.1-1 O.-0

: oo B -n "S 73 -S60 C O ^ Jj Q) ^B eg IH o S a)•H JS 60 3 60 >rH o B 0 73 -H -rl3 -H 01 01 00 4J

T3 M -O H C O y01 O B

43 IM 01 .u a u00 CO 01 B CO 01 M

ot 73 IH eg 1-1u u eg 3 oo

•H hi JS 4J CO 13 BrH 4J43 010 43

aCO

01 r-lJ3 CJ•U >

U 003 W

1W01

73

S M

co eg B41 B O0 60 Ueg 1-1 Ma. co oca 43

43 *B 01o eg 43

eg a.01 O *Jco ^ so

S |U i-lu 01

eg

ega41

43 S•U B OO i-l O

430073

0 Beg

o co01 O

rH B OCO -H Uhi -H01 60> rH B4> 0) •*CO i-l 41

rH 43

X " ahi hi SM eg o5 CL hi

^ IM

uBrl

UCJ01

hiouOl

43

•

01 <HU 01eg >hi 01eg r-lD.0) 43co u

egeg 01

01 MCO O3 <M

-

*: i01 O> * u01 cor-l 0) CO

COf» 01 O9

eg 44 13O O) hi

01 (3« ^ COco X eg o3 43 as 43

w

1CO

us

iv w — -n• CO 13

73 73 01 r-l 01B hi X 01 hieg a) u > u

O 0)Ol 43 73 r-l B

eg u OB 1-1oi at u oi coH r-l 01 01 CO0 43 hi 3 eg

r-l CM «n

co•H

B3

OO.a

_o

0)COa

•^ W MO hi 01to eg uOl Ol Bhi co 3•-I IM IMg O OO 43 CO• H U BU 13 Oeg eg -H4J Q) 4J

3 M -H0.43 13

43IM OO CO

11MU coeg 6085B egrl hi

^

0.

IMO

CO011-1u•H

O

01 COa. uw co

64

Sa

COEcu4-1CO>>en

a)

I

8-QX

Ed

CD:0

(II4JU

2CO

6

C

^4^2*

^

*"-»

O

01l-iVi0135

o

o•H44cfl

01

a01

c•H

i-HCOC01

•I-len

H •• 23 C/2 O

2 H ci MW W O HEH £-* £ CJen en H Sen M < t,

•SE™uc. eenH-

oK

2ou

wtn a0 <

paenz 2Cx3 5g P

*J T

U <P

J3j P

wu<caw^gc§S

Ic04CX,P-i

UenOSP&,

M

a §CO W01 CUpq C

3C UCU J2> u

•HI-l =•0 4S

uCO l-i44 cflCO CUQ en

ju0101exen4-1

ft.

1 CM

Q

S01PJ

eo•H4J•HCOC

H

l-iCU O CO4J C 44 014 - 1 0 .1-13 i-l co M

4J 01 CO01 O i-( . T3r-l 3 43 . C43 -O CO 3•H 0 i-l 0CO I-l i— 1 43CO O. >,O CO TJQ, >, ^

43 iH o"H Cfl 5

CO i-l 60 4-1

14-1 0 cnO O co

T3 CU 0)

01 CO cfl 3CO - i-H VI

>i CUcu CM l-i

43 S 014-J ^3 4^2 Vi

33 O 34J 1-1 4-1C C 42 U01 -H S Cco . 301 co cfl cn i™)l-i 0) cu •«Q, CJ iH u -O0) C 3 o -a •Cti CO U 5 <

1Vl01•ac •3 00

C42 i-tU T)01 C01 cfl(X 4J

cn co

ORJGJKAL FAQS W . . . .OF POOR QUALITY

0

• \ .•

.

•

*

• • . • . .' • • . . • • • - • • . . . . . • - . - .• .••V .

O .•H CO '

« (fl 43 -O 4Jx 3 a c • • •eg o cu cu4J 4J C SC C -H C S)>M-» cfl euCO Ji CO

cu 1-1 o01 00 O 4J U-l

4S -O S O4-1 01 44 0)

iH a) -a o) .01 3 C O 0CO C C•H C c eu43 ^ 0 4-1 3S i-t u erO 01 4J cfl O)O M -H 4-1 CO .

3 0) coO 4J C Cfl CU44 U CO CO U

C M CD CM 3 44 g 4-1 CUcu T-I o e 44

rH CU U CU C•H T3 tlO<4-l CO 0)a. c l-i cu coS cfl cfl 43 ViO i-H 44 Qu CUU iH CO 01 S

CO 01 13, Vi OCO U rH 00

i-4 604S 0)01 X C CJ T> WCO CU i-l CO O OP i-l CO CU C U-l

65

0)

01uCOXca

rH 01I i-H

rH D.

(d CO

CO (d•<

3

a3

£j

p

gCO

a

1cCJ

t/3

A

c

iw

»*CT2

Ct

Ci

5§

e01>Vl•0

wB01

Id

aiH

•§

J

TJ

ack

boar

rHca=

,

u01CO

OIrH

2000)4J09ViUen

to01u *

TJ ' 3 •B U

". .4. .0) eg MN U OI

CO JS 04 0001 O U OIJS Vl 0. UU 4) U 01 J* .O - Vl 01 Vl COQ. 03 iJ u . OS 01X iH C CO U &

ST. 3B.TH 1 1 1

. -H ISca vi o •01 3 V) TJ rH

JS O UH B 01x u ca TJ

rH O 01 CO .O1-1 a. oi ws IH js sca x n u eo TJ a.U JS 3 01 CU CO rH•H u vi jr B? vi aJ S T J U 0 W 0 00 0O CU 3 UH O iH TH

•• Vi N VI O . 4 J C BC O C O l H U 4 J X V I O C U01 Vi C CO V i j C l O U j S

TJ CU iH O O. OI CJ3 iH to OJ Ck U i-H

rH js GO u c x e o i o i o iO iH ta 3 01 JS J* JSB c a O T J CO Vl W CO U

•H 1 1

uiHVI

.000

B rHo <o

4J iHaj . u aoco jj ca B

01 N . iHoi to IH a BrH B ji 03 CU O 4J iH

3 OI Vl iHX VI i-l 0 U

CO 01 50 14co en M rH a)H M en « pu

: *

"

1•uoo.X

01

4J

UHO

CU •U 0CO iH4J 0cn cu

• OF POOR QUAL-TY

'•

d « o > . • * * ' •3 « js a TJ .

'W M W " flv» oi u caO rH UH V . .a oi o h ca •a o p.O U u • *H X ' '

to B CO O S: a) -H . .TJ vi a B « xrH o a. a) co ja •g UH o £ * e

JS rH U rH CO O01 CO 01 O 01 iHO 01 £* O» ^ bt\U '«J> IH oi x O B ca

£ 4J TJ «B V CQ CJ1 ^ _ Q

W O rH

'

,

^«_J

COMF-3

cnoiU

toVlOI

BJ3

TJViO

Beg4Jcn

ViVlaiH*Q

OIrlO€

rH61B[d

O

B0

4Jto4JOIVloVi01uBM

ca4Jto0

H •• Z•• 3 C/l OS H fA t-tiS M o HH H S Uco to H 5>• z » 5co M <: u.

w . pVl U Uo 1-1

cu cdX £co uC O )

COI 01 01

CO 4J JS,e B iJ Vi

u a) o . CX

o01 iH •Vt JS 4J

2* 8"o S C

e

w a uS O 01

UH coCU£ O 01

s.

CU 01

•-I C•rH Vl

M Via.01 «

JS U^ T-»

60 COrH Ba u

niz

ation

lectr

on

0 01uOI 01rH JC01 4J

O 4JVi« «

01 Bto oS 0

*rH

s0^H4Jto

U 4J•H 01

W

BOICOOIViQ.OIVl

eo4JOIrH

JSCO

1OI

•HrH

CO

B•H01 C

U-rl

VlCOQ.

J4JiH

corH

CO •

o

chem

icele

ments

±JB 01

M -H eo o60 JS J3

CO

ca01COB.

CN

COD.

:

CO 0vi ca00 J3

C TJ0 B4J CO

rH01 OI

SSI

hypoth

es

idence 1

UHO C4J 0

01 TJrH 013" UVl 10

iHOI (Jto o3 co.

m

wCO

aJ3

to

§Ua]

Vl01o.3CO

TJBCO

CO

ouca

Ml —U

O.UHca o

01CS TJ

° irH rHoi caa u

CO iH

T J §01 J=B U

3 sVi Oca B

»

X•§4JtoVi01

•s3

Hf

2ca

rH10oiHJSOIHCOIH01•HJS

OICOS.

"*

* «H

01i-H3Vi

01

toiHVla.0

a.a.to

ucurH

•

CO01o01

p

hypoth

Q

^^OI

T)

O4J

01a•HCJOI0!

wW

COTJ

s to

ehol

•HcoV

uoa£01CO3.

rH

h-iH VlS a>U TJ01 iHu to01 BT, 0

0u o

4JCO01 UrH 0)

2 "01

XrH60 301 U

ca jtfVi COjj cato u

••• •«U 4Jto caB iH•H rHca60 J->CD B

01a >01 01

rH3 CUVl JS

CO UHto ou

CJ 3

to o£ u

^CN

A-

OICO

J MOJ

O COu oQ

10 Ou

B COCU OI> -101 3

Viai

o X

w w4J >CO 01UCO 01

to uiHCO UHCU O

4J BO Oa. -rix w

JS VIo« a.

u s

•5 Sa •§£ =

a

«n

oi uJS COU TH

rH

ncr

em

ent

e event

"rl .BU

Ou o

01 TJrH TJs ca

cn s

•o u aB iHca •

Vl 01

N S'SO VI

O 00 60OI 01

CU JS UrH O COCJ 4J Vtx tg *Ju 8 coX Vi O

rH 01 4J

11 iJ rH

* % 2a) 3 )rfH) m jjjo* tscu o o

OS S 0

^

U IH U O X§ 0. O Vl 4J

Vl Vl U -rlo oi a. o a4J U 01 B3 B UH rH 01<'i-l O 01 TJ

66

NBS-114A (REV. 2-aoiU.S. OEPT. OF COMM.

BIBLIOGRAPHIC DATASHEET (See instructions)

1. PUBLICATION ORREPORT NO.

NBSIR 82-2505

2. Performing Organ. Report No, 3. Publication Data

May 19824. TITLE AND SUBTITLE

AN OVERVIEW OF EXPERT SYSTESM

5. AUTHOR(S)

William B. Gevarter

6. PERFORMING ORGANIZATION (If joint or other than NSS. see instructions)

NATIONAL BUREAU OF STANDARDSDEPARTMEMT ffF^TOMMERCF~WASHINGTON, D.C. 20234

7. Contract/Grant No.

8. Type of Report & Period Covered

9. SPONSORING ORGANIZATION NAME AND COMPLETE .ADDRESS (Street. City. State. ZIP)

10. SUPPLEMENTARY NOTES '

Document describes a computer program; SF-185, PIPS Software Summary, Is attached.

11. ABSTRACT (A 200-word or less (actual summary of most significant information. If document Includes a significantbibliography or literature survey, mention it here) \

This report provides an overview of Expert Systems - currently the hottesttopic in the field of Artificial Intelligence. Topics covered include what itis, techniques used, existing systems, applications, who is doing it, who isfunding it, the state-of-the-art, research requirements, and future trends andopportunities.

12. KEY WORDS (Six to twelve entries; alphabetical order; capitalize only proper names; and separate key words by semicolons)Applications; Artificial Intelligence; Expert systems; Forecast; Intelligentcomputer programs; Knowledge engineering; Machine Intelligence; Overview; Research;State-of-the-Art; Funding sources. •

13. AVAILABILITY

|X I Unlimited . . .^_| | For Official distribution. Do Not Release to NTIS

| | Order From Superintendent of Documents, U.S. Government Printing Office, Washington, D.C.20402.

[X] Order From National Technical Information Service (NTIS), Springfield. VA. 22161

14. NO. OFPRINTED PAGES

73

IS. Price

$9.00

USCOMM-DC SO43-PSO