Lesson 9

Knowledge-based Systems:
Rule-based Knowledge Systems

Knowledge-based Systems
(Expert Systems)
 Rule-based systems:
 Overview:
 Deductive versus Reactive rule-systems
 “Zookeeper”: a deductive rule-system
 “Bagger”: a reactive rule-system
 Procedures for inference in Rule-systems
 Explanation facilities

CSC 215: Introduction to A.I 2

 Deductive Rule-Systems:
Consist of rules of the form:

If Condition_1 Then Conclusion_1

and Condition_2 and Conclusion_2
… …
and Condition_n and Conclusion_m

Winston’s notation:

Know Deduced
facts facts

AND gate
CSC 215: Introduction to A.I 3
Basics for Deduction Systems:
true true
assertions facts
false
The working memory = a collection of assertions

Can be either facts or hypothesis:

In FORWARD chaining, we usually start from facts and add
assertions that are new facts:

Know Deduced
facts facts

In BACKWARD chaining, we usually start from hypothesis and

add assertions that are new hypothesis.

New
Hypothesis
hypothesis
CSC 215: Introduction to A.I 4
 ZOOKEEPER:
Winston’s simplification of MYCIN.
Robby robot attempts to identify animals in the ZOO.

Z1 If ?x has hair
Then ?x is a mammal

Z2 If ?x gives milk
Then ?x is a mammal

Z3 If ?x has feathers
Then ?x is a bird

Z4 If ?x flies and ?x lays eggs

Then ?x is a bird

Z5 If ?x is a mammal and ?x eats meat

Then ?x is a carnivore
CSC 215: Introduction to A.I 5
 ZOOKEEPER (cont.):
Z6 If ?x is a mammal and ?x has pointed teeth
and ?x has claws and ?x has forward-pointed eyes
Then ?x is a carnivore

Z7 If ?x is a mammal and ?x has hoofs

Then ?x is an ungulate

Z8 If ?x is a mammal and ?x chews cud

Then ?x is an ungulate

Z9 If ?x is a carnivore and ?x has tawny color

and ?x has dark spots
Then ?x is a cheetah

Z10 If ?x is a carnivore and ?x has tawny color

and ?x has black stripes
Then ?x is a tiger
CSC 215: Introduction to A.I 6
 ZOOKEEPER (cont.2):
Z11 If ?x is an ungulate and ?x has long legs
and ?x has long neck and ?x has dark spots
Then ?x is a giraffe

Z12 If ?x is an ungulate and ?x has white color

and ?x has black stripes
Then ?x is a zebra

Z13 If ?x is a bird and ?x does not fly

and ?x has long legs and ?x has long neck
Then ?x is an ostrich

Z14 If ?x is a bird and ?x does not fly

and ?x swims and ?x is black and white
Then ?x is a penguin
CSC 215: Introduction to A.I 7
Forward chaining in Zookeeper:
Working Stretch has hair. Stretch has a long neck.
memory Stretch chews cud. Stretch has dark spots.
Stretch has long legs.

First rule fires

Has Is a mammal
Z1
hair
Second rule fires
Chews cud Is an ungulate
Z8

Has long legs Third rule fires

Has a long neck Is a giraffe
Z11
Has dark spots
CSC 215: Introduction to A.I 8
 Backward chaining in Zookeeper
Knowledge Swifty has forward-pointing eyes. Swifty has claws.
base Swifty has pointed teeth. Swifty has dark spots.
Swifty has tawny color. Swifty has hair.

Working memory Swifty is a cheetah.

Has forward-pointing eyes

Third rule
Has claws Is a carnivore
Z6
Has pointed teeth

Fourth rule
Has hair Is a mammal
Z1
Second rule tried
Eats meat Is a carnivore
Z5
First rule used
Has a tawny color
Z9
Has dark spots Is a cheetah
CSC 215: Introduction to A.I 9
 Old backward/forward
choice:
 Mostly: the branching factor argument.
 Winston terminology: “fan out”, “fan in”

 Also:
 Are you able to gather more facts?
 If not: just forward chain to see what you can
get as conclusions
 If you are only interested in a specific question:
 Is it a carnivore?
– Backward better

CSC 215: Introduction to A.I 10

 Example: Mycin
- Diagnoses Bacterial Infections of the blood.
How to choose antibiotic treatment, without knowing
exactly which organism?

- Preferably: don’t prescribe a broad-spectrum drug

- Narrow down to more specific, using information

concerning:
patient symptoms, general condition,
quick lab-tests, ...

- Questions and advises physician.

CSC 215: Introduction to A.I 11

Excerpt of sample session:
> Male or female?
Male.
> Age?
55.
> Have you obtained positive cultures indicating general type?
Yes.
> What type of infection is it?
Primary bacteremia.
> When did symptoms first appear?
May 5.
> From where was the positive culture taken?
From the blood.
>…
CSC 215: Introduction to A.I 12
 Mycin (cont.)
 About 500 If-Then rules .
 Can recognize about 100 causes of infections.

 Example rule:
If ?x ’s type is primary bacteremia
the suspected entry point of ?x is the gastrointestinal tract
the site of the culture of ?x is one of the sterile sites
Then there is evidence that ?x is bacteroides

 Uses: backward chaining.

… and certainty factors.

CSC 215: Introduction to A.I 13

 Query-the-user facility:
 Where do the questions come from? > Age?
55.
 The query-the-user rule:
 Any system includes a rule of the type
If query-the-user( ?x )
Then ?x

 Here ?x could be any predicate

 If we want to establish ?x and there is no
information on ?x in the rule-base, ask the user
about ?x
If age_patient(?x) No “age_patient(..)” query-the-user(
?x > 50
age_patient(?x))
Then ... in rule-base

CSC 215: Introduction to A.I 14

 Reactive Rule-Systems:
Consist of rules of the form:

If Condition_1 Then Action_1

and Condition_2 and Action_2
… …
and Condition_n and Action_m

No more relation to logical statements !

CSC 215: Introduction to A.I 15

 BAGGER:
Winston’s simplification of XCON.
Robby robot attempts to help bagging in a supermarket.

It considers 4 steps in the procedure:

1. ”check order” :see if something is missing
2. ”bag the large items” and put big bottles in first
3. “bag medium items” and put frozen things in
freezer bags
4. “bag small items” : put wherever there is room

Use a
database
of objects

CSC 215: Introduction to A.I 16

 Bagger’s Working Memory:

 Current step is “Check order”

 Current bag is Bag_1(empty)

 Not bagged are bread, glop, granola (2),

ice cream, patato chips
+ the database of objects

CSC 215: Introduction to A.I 17

 Bagger rules:
B1 If step is “check order”
and patato chips are in not bagged
and Pepsi is NOT in not bagged
Then suggest to add Pepsi to not bagged

Assume suggestion is accepted and Pepsi added in not bagged.

B2 If step is “check order” Delete

Then discontinue step “check order”
And start step “bag large items”
Add

Note: whenever B1 is triggered, B2 is triggered as well !

Need: CONFLICT RESOLUTION
CSC 215: Introduction to A.I 18
 Conflict Resolution:
A rule in a rule-based system is triggered
if the precondition of the rule is satisfied
A rule is said to fire
if its then-part is executed

In a deductive system, every triggered rule fires.

In a reactive system, we need a “conflict resolution

strategy” to determine which rule may fire.

Assume for now that we use textual order here.

CSC 215: Introduction to A.I 19

 Bagger (cont.):
 Current step is “bag the large items” Working memory
 Current bag is Bag_1(empty)
 Not bagged are bread, glop, granola (2),
ice cream, patato chips, Pepsi

B3 If step is “bag the large items”

and a large item is in not bagged
and a large bottle is in not bagged
add
and current bag has < 6 large items
Then put large bottle in current bag delete

B4 If step is “bag the large items”

and a large item is in not bagged
add
and current bag has < 6 large items
Then put large item in current bag delete

B5 If step is “bag the large items”

and a large item is in not bagged add
Then start a fresh
CSC 215:bag
Introduction to A.I
delete
20
 Possible conflict resolution
strategies:
Specificity ordering:
highest priority rule has superset of conditions
Textual rule ordering
(see Bagger)
Data ordering
give a priority number to each possible aspect
Size ordering
highest priority rule has largest list of conditions
Context limiting
put rules into groups: activate/deactivate the
groups

CSC 215: Introduction to A.I 21

 Example: XCON
- System for Configuration of Computer Systems.
How to set up all components of a big main frame
computer system.

- Includes processors, memory, terminals, discs,

peripheral controllers, etc.
- All components must be arranged sensibly with
appropriate input/output buses.
- Must deal with hundreds of possible options: how
to configure?

CSC 215: Introduction to A.I 22

 XCON (cont.)
 About 10000 If-Then rules .
 Can deal with about 200 components of VAX systems.
 Example rules:
If context is doing layout and assigning power supply
an sbi module has been put in a cabinet
the position the sbi module occupies is known
there is space available for a power supply
there is no available power supply
the voltage and frequency of components is known
Then add an appropriate power supply

Similar to the Pepsi rule in bagger

If context is doing layout and assigning power supply

an sbi module has been put in a cabinet
the position the sbi module occupies is known
there is space available for a power supply
there is an available power supply
Then put the power supply in the cabinet in the available space
CSC 215: Introduction to A.I 23
 Explicit procedures for
Back/Forward reasoning:
 Depth-first search Even for forward reasoning!

 Database operation (Join - Projection - Selection)

for Forward Reasoning
 Tuple-at-a-time database operations:
the Rete algorithm
Ref: Winston for details.

CSC 215: Introduction to A.I 24

 Mixing Forward and
Backward Reasoning:
Many commercial systems promote their ability to provide
both forward and backward reasoning (together) on the
same set of rules.

Is this useful? Or even reasonable??

- Forward reasoning starts from facts and produces new

fact.
Why would one want to start backward reasoning:
from hypothesis building new hypothesis from a fact??

- Backward reasoning starts from hypothesis and builds

new hypothesis.
These are not facts and cannot be used to derive new
fact from them in a forward reasoning !!

CSC 215: Introduction to A.I 25

 One way to combine
forward and backward:
 Assume that by forward reasoning we derived:
 the facts a, b and d.

 If there is a rule:
If a and b and c and d
Then e

then it may be a good idea to switch to backward

reasoning to try to prove c (and thus conclude e)

 BUT: VERY hard to control this and completeness

becomes VERY obscure.
CSC 215: Introduction to A.I 26
 Explanation facilities:
Expert Systems are supposed to provide explanations to
convince the user that their proposed conclusions are
reasonable.

 Typical user questions:

 How did you arrive to the conclusion that … .
 Why did you show that … .
 Technique:
 Keep a trace of all inferences done:

If a and b and c and d

Then e
And record e <- a, b, c, d

CSC 215: Introduction to A.I 27

 All these systems require FULL
knowledge of the world !
 What if information is Missing?
 Information is Unclear?
 Information is Vague?
 Information is Contradicting?
 Information Only probabilistically available?

 Requires a study in Uncertainty, Fuzziness, ….

CSC 215: Introduction to A.I 28