~j(635)\f5
Early  (1963-1975)  Expert Systems: ~c\f9ub34U
~c\f9b
~c\f9b
MEDICAL CONSULTANTS:~\f9b
DIALOG~l3936\f9
PIP~l3936\f9
GLAUCOMA~l3936\f9
CHEMICAL CONSULTANTS:~\f9b
	 DENDRAL~\f9
	 SECS~\f9
	 Organic Synth. (Gelernter) ~\f9
MATH & BUSINESS CONSULTANTS~\f9b
	 MACSYMA~\f9
	 Management Sci. (Hearn, Burstall, Kuhn)~\f9
EDUCATION CONSULTANTS:
	 BUGGY~\f9b23B
~\f9
OUTLINE: 1981 NCC Talk~c\f9
~c\f9
"Applied Artificial Intelligence:  Expert Systems" ~c\f9
~c\f9
by Randall Davis and Douglas Lenat~c\f9
~j\f5
~j\f5
I. Snappy Introduction  (Davis)~j\f9
			The What & Why of Expert Systems
			Important Themes~j\f8
II. Dull Middle (Lenat)~j\f9
			Case Studies
III. Snappy Conclusion (Davis)~j\f8 16f9 30f8
			Challenges: Where Next?~j\f8 26f9
~j\f5
~j\f5
~j\f5
												-- tentative outline compiled by R. Davis~j\f5
~j\f5
~j\f5
EXPERT SYSTEM  IS CHARACTERIZED BY:~c\f9ub35f0UB
~
~\f9
TASK:   ~z19579\f9
~\f9
REPRESENTATION:   ~\f9 18f0
~
CONTROL STRUCTURE:  ~\f9 20f0
~\f9
~\f9
5  New Kinds of Numbers Proposed by the Previous Rule~c\f9ub53UB
~c\f9
~c\f9
Numbers having no divisors~c\f9
~c\f9
Numbers having only 1 divisor~c\f9
~c\f9
Numbers having only 2 divisors~c\f9
~c\f9
Numbers having only 3 divisors~c\f9
~c\f9
Numbers having only 4 divisors~c\f9
~c\f9
~c
~
~\f9
Lessons from Pre-1975 Work~c\f9ub26UB
~c\f9
~c\f9
Represent the concepts, methods, and in particular the heuristics that experts use.~j\f9 55b10B
~j\f9
Separate domain-specific knowledge from 
domain-independent problem-solving techniques~j\f9
The former (expert knowledge) is the source of power;
the latter (control structure) can be very simple.
~l4576\f5b105f9B
Provide a well-engineered interface.~j\f9
~j\f5
Sources of Uncertainty~c\f9ub22f5UB
~j\f9
DATA are missing or erroneous~j\f9
RULES are missing or erroneous
MODEL is inaccurate~j\f9 30f5 1f9
~j\f9
~j\f5
Corrections for Uncertainty~c\f9ub27f5UB
~j\f9
DATA can be redundant~j\f9
RULES can be redundant
CONTROL can be cautious~j\f9 22f5 1f9
RULES from the expert on detecting 
			 and coping with uncertainties~j\f9
~j\f9
~j\f5
Coping with Incomplete Information~c\f9ub34f5UB
~j\f9
Halt~c\f9
Ask
Infer~c\f9
Default~c\f9
Guess~c\f9
Skip~c\f9
~j\f9
~j\f5
~j\f5
Recent  (1975-81)  Expert Systems:~c\f9ub33f5U1B
~c\f9b
MEDICAL & BIOLOGICAL CONSULTANTS:~\f9b
	CASNET
	INTERNIST~\f9
	MOLGEN~\f9 7f5
	MYCIN/CLOT/PUFF/VM~\f9
	OWL DIGITALIS~\f9
CHEMICAL & PHYSICAL CONSULTANTS:~\f9b
	CONGEN
	CRYSALLIS~\f9
	GAMMA~\f9
	PROSPECTOR~\f9
MATH & BUSINESS CONSULTANTS~\f9b
	AM~\f9b1B
	LDS~\f9
	ODYSSEY~\f9
	OMEGA~\f9 6b
	UTP~\f9b1B3b
~j\f9 1f5
~j\f5
Recent  (1975-81)  Expert Systems (cont'd):~c\f9ub33UB9f5b1B
~c\f9b
~\f9
EDUCATION CONSULTANTS:
	DEBUGGY~\f9b23B
	EXAMINER~\f9
	GUIDON~\f9
COMPUTER CONSULTANTS:~\f9b
	BROWSER
	DART~\f9
	PECOS~\f9
	R1~\f9
KNOWLEDGE ACQUISITION CONSULTANTS:~\f9b
	EURISKO~\f9
	META - DENDRAL~\f9
	PSI~\f9
	TEIRESIAS~\f9

~\f9
Lessons from Recent Expert Systems~c\f9ub34f5UB
~j\f9
~j\f9
Represent knowledge uniformly     ~j\f9
Make the Knowledge Base modifiable	 ---->  Repr. Language~z21167j\f9 36f5b5f9B
Employ a simple control structure   
~j\f9
Accomodate the User
~j\f9 19f5 1f9
Live with Inexactness and Uncertainty~j\f9 37f5
~j\f5
~j\f5
State of the Art in Expert Systems~c\f9ub34f5UB
~
Narrow domain of expertise
Limited language for expressing facts and relations
A "knowledge engineer" must be there to help
		At least to design the knowledge base
		Which can then be easily extended by the expert
Limited I/O languages
Limited explanations of lines of reasoning
Limited self-knowledge (scope, limitations)
One single expert as  "knowledge czar"~\f9 124f5b90f9B147f0
~
~
How do we represent...~c\f9ub22f0UB
~
Causal models
Strategies and Plans
Expectations and defaults
Temporal and spatial continuity
Abstractions and Approximations
Analogies
Modalities and Beliefs
Conflicts and Contradictions~\f9
Multiple Sources of Expertise~\f9
Parallel Processing
Learning from Experience~\f9
Focus of Attention
Mass nouns~\f9
~
VALIDATION:    The MYCIN Survey~c\f9ub31f0UB
~
~
Ratings by 8 Experts on 10 Meningitis Cases~c\f9
~c\f9
					(Perfect score = 80)~\f9
~\f9
~\f9
						Acceptable to Evaluator			Missed Treatable Pathogen~z23495\f9 6f6u23f9U1f6 2u25f9U
MYCIN					52									0
FACULTY-1			50									1
FACULTY-2			48									1
Infec.Dis.Fellow	48									1
FACULTY-3			46									0
ActualTherapy		46									0
FACULTY-4			44									0
RESIDENT			36									1
FACULTY-5			34									0
STUDENT				24									3
~\f9 254f0
PUFF~c\f9ub4f0UB
~
~\f9
TASK:  Diagnose pulmonary function disorders.~\f9
~\f9
REPRESENTATION:   55  If/Then   rules~\f9 30f5 2f9 5f0
~
CONTROL STRUCTURE:  Exhaustive back-chaining~\f9 44f0
~
~
BROWSER~c\f9ub7f0UB
~\f9
~\f9
TASK:  Open-ended sifting through a database:~\f9
			  define concepts, make and test conjectures.~z19579\f9
~\f9
REPRESENTATION:   If/Then   rules   and  Frames~\f9 26f5 2f9 5f0 1f9 13f0
~
CONTROL STRUCTURE:  Agenda of plausible topics~\f9 46f0
~c\1f9b
LDS~c\f9ub3f0UB
~
~\f9
TASK:  Legal decision making --
		  	  in the area of product liability law.~\f9
~\f9
REPRESENTATION:   If/Then  rules  and  Assertions~z21167\f9 26f5 1f9 22f0
~
CONTROL STRUCTURE:    Forward-chaining~z19579\f9 38f0
~
~
~
R1~c\f9ub2f0UB
~
~\f9
TASK:  Configure DEC VAX -11/780 systems.~\f9
~\f9
REPRESENTATION:   772  If/Then  rules  and  420 Frames~z21167\f9 31f5 1f9 22f0
~
CONTROL STRUCTURE:    Forward-chaining~z19579\f9 38f0
~
~
R1:  Typical Rule~c\f9ub17f0UB
~\f9
IF:
	1) Currently assigning devices to the unibus modules,
	2) and there is an unassigned dual port disk drive,~z20743\f9 111f0
	3) and the type of controller it requires is known,~z20743\f9
	4) and there are two such controllers still unassigned,
~z20743\f9
THEN:
	Assign the disk drive to each of the controllers,  
	and note that each controller supports one device.~\f9 56f0 1f9 53f0
~
~c
~c\1f9b
PUFF:  Typical Rule~c\f9ub19f0UB
~\f9
IF:
	1) The severity of OAD is at least Mild,
	2) and the degree of diffusion defect is at least Mild,~z20743\f9 102f0
	3) and the TLC observed/predicted ratio is at least 110,~z20743\f9
	4) and the RV/TLC observed-predicted difference > 10,
~z20743\f9
THEN:
	There is strongly suggestive evidence (.9) that
	the subtype of OAD is Emphysema.~\f9 88f0
~
~
LDS:  Typical Rule~c\f9ub18f0UB
~\f9
IF:
	1) The plaintiff is injured, or
	1b) the plaintiff's property is damaged,
	2) and the incidental-sale defense is inapplicable,~z20743\f9 131f0
	3) and the product is sold or leased by the defendant,~z20743\f9
			.~z20743\f9
			.~z20743\f9
			.
~z20743\f9
THEN:
	Assert:  The defendant is liable 
				   under the theory of strict-liability.~\f9 85f0
~
~c
Specific Programs  --->  Representation Languages~c\f9u17U2f5b4f9B2u24f0U
~c
~c
					 CASNET  --->  EXPERT~\f9 14f5b4f9B
		  EURISKO&AM  --->  RLL~\f9 16f5b4f9B
						MYCIN  --->  EMYCIN~\f9 13f5b4f9B
				   MOLGEN  --->  UNITS~\f9 15f5b4f9B
						   RITA  --->  ROSIE~\f9 15f5b4f9B7f0
~
EXPERT~c\f9ub6f0UB
~
~\f9
TASK:  Consultation.~\f9
~\f9
REPRESENTATION:   Assertions and decision rules.~\f9
	Distinguish  "findings"  from  "hypotheses"~\f9
	So rules can be  F->F,  F->H, or  H->H~\f9 39f0
~
CONTROL STRUCTURE:  Complicated but fixed.~\f9 42f0
~
COMPLICATIONS:  Causal net may disagree with Rules.~z19579\f9 51f0
~
Typical EXPERT Rules~c\f9ub20UB
~c\f9
FINDING-->FINDING~\f9ub
	 F(M,T) --> F(PREGP,F)~\f9
~\f9
FINDING-->HYPOTHESIS~\f9ub
	IF rapid palpitations and finger tremor~\f9
	THEN hypothesize hyperthyroidism (confidence .5)~\f9 49b
~\f9b
HYPOTHESIS-->HYPOTHESIS~\f9ub
	IF there are hypothesized eye and thyroid 
		 dysfunctions, with confidence at least .5,~z20743\f9
	THEN hypothesize Graves' disease (confidence .9)~\f9 49f0
~
~
MYCIN~c\f9ub5f0UB
~
~\f9
TASK:  Diagnose blood infections and meningitis,
			  and recommend drug treatment.~\f9
~\f9
REPRESENTATION:   400  If/Then   rules~\f9 31f5 2f9 5f0
~
CONTROL STRUCTURE:  Exhaustive back-chaining~\f9 44f0
~
COMPLICATIONS:  Each rule has a certainty-factor,
	thereby spawning multiple plausible lines-of-reasoning~z19579\f9 105f0
~
~
TEIRESIAS~c\f9ub9f0UB
~
~\f9
TASK:  Within a MYCIN consultation, interact with
 			  the physician to debug or acquire rules.~\f9
~\f9
REPRESENTATION:   If/Then   rules~\f9 26f5 2f9 5f0
~
CONTROL STRUCTURE:  Exhaustive back-chaining~\f9 44f0
~
COMPLICATIONS:  Must explicitly -- and dynamically --model the transfer of expertise from human to program.~z19579\f9 107f0
~
~
TEIRESIAS:   Typical Rule~c\f9ub25f0UB
~\f9
IF:
	1) The patient is a compromised host, and
	2) there are rules whose "IF" mentions pseudomonas,~z19579\f9 99f0
	3) there are rules whose "IF" mentions klebsiellas,
~\f9
THEN:
	There is suggestive evidence (.4) that the former 		rules should be done before the latter ones.~\f9 103f0
~
~
AM~c\f9ub2f0UB
~
~\f9
TASK:  Open-ended research in elementary math:~\f9
			  define concepts, gather data, conjecture.~z19579\f9
~\f9
REPRESENTATION:   If/Then   rules   and  Frames~\f9 26f5 2f9 5f0 1f9 13f0
~
CONTROL STRUCTURE:  Agenda of plausible topics~\f9 46f0
~
COMPLICATIONS:   Integrate rules and frames.~z19579\f9
									 Manage an open-ended task.~z19579\f9 36f0
~
~
EXPERT SYSTEM  IS CHARACTERIZED BY:~c\f9ub35f0UB
~
~\f9
TASK:   ~z19579\f9
~\f9
REPRESENTATION:   ~\f9 18f0
~
CONTROL STRUCTURE:  ~\f9 20f0
~
COMPLICATIONS:   How the knowledge was acquired.~z19579\f9
								  	 How the system was evaluated.~z19579\f9 41f0
~
~
~
AM:   Typical Rule~c\f9ub18f0UB
~\f9
IF:
	1) f  : A -> B, and
	2) there is some already-known extremal subset E of B,~z21167\f9 8bi1BI6f8 2f9 63f0
~\f9
THEN:
	It is plausible to define and study  f -1 (E).~\f9 42f1 1f9 1bi1f1o4I1f5 1f0 1f1o0B1f9 4f0