Introduction to

Artificial Intelligence
Assoc. Prof. Osama fathy

Knowledge Acquisition: is
the bottle neck of building
Intelligent systems
2

Knowledge Acquisition (KA)

is the process of acquiring the knowledge from

human experts or other sources (e.g. books,
manuals) to solve the problem.

Knowledge Sources

Text books,
reports,
databases (data mining),
case studies,
empirical data and
Personal Expertise.

Selection of the Expert

Ask the organization to provide you with the names
of candidate domain experts, that is, those
individuals who are believed to have significant
expertise within the domain in question.

Select a domain expert whose performance is

generally acknowledged to be above and
beyond that of most others performing the same
task.
Select an expert with a successful track record
over a period of time.
5

Selection of the Expert

Select an expert who is both willing and able to

communicate personal knowledge, and who is
relatively articulate in doing so.
Select an expert who is both willing and able to
devote the time necessary to support the
development effort.
If no expert can be identified, or made available,
consider the development of the rule base
through alternative means (as will be
discussed in the sections to follow)
6

Interviewing the Expert

a- There are 4 reasons why K. cant be satisfied
to describe the domain

For some problems there may not be an expert,

Some experts are unable to describe what they
do,
Some experts may not reveal their tricks of the
trade, and
Some experts may actually have poor expertise
7

Interviewing the Expert

b- Initial meeting has some purposes

Relax individual (expert)

Explain Problem
Schedule follow-on meeting
Evaluate the true extent of expertise of our
systems

Interviewing the Expert

c- Documentation
Documentation of the results of the meeting as
soon as possible, it should contains such facts:

Date, time and location of the meeting

Name of Expert (i.e., if more than one expert
is being used)

List and description of the rules identified
during the meeting

Listing of any new objects, attributes and/or
values encountered - and their properties
9

Interviewing the Expert

Identification of any new outside sources and

references
Listing of any new terminology encountered,
and associated definitions
Listing and discussion of any gaps or
discrepancies encountered
Reminders (e.g., of points that need to be
clarified)

10

Techniques used for extracting Knowledge

from a domain expert
a- On-site observation
K.E watches the domain expert (D.E.) solving
real problems on the job.
K.E.

Watches

D.E.

Note: difficulties are that the K.E. extracts the rules of

solving the problems without expansion from D.E.
11

Techniques used for extracting Knowledge

from a domain expert
b- Problem discussion
K.E. explores the kinds of data, K. and procedures
needed to solve the problem through the
discussion.
K.E.

Explores

D.E.

12

Techniques used for extracting Knowledge

from a domain expert
c- Problem description
have the expert describes a prototypical problem
for each category of answer in the domain.

K.E.

Describes
Prototype

D.E.

Note: difficulties are that the D.E. describes a prototype

(not real problem).
13

Techniques used for extracting Knowledge

from a domain expert
d- Problem analysis
K.E. presents the expert with a series of realistic
problems then the D.E solves the problems with
reasoning steps (analyze the results).
K.E.

Describes Presents real problems

solve with reasoning

D.E.

14

Techniques used for extracting Knowledge

from a domain expert
e- System refinement
Have the D.E. gives the K.E. a series of problems
to solve using the rules acquired from the
interviews.
D.E.

Gives the K.E. a series of

problems to solve

K.E.

15

Techniques used for extracting Knowledge

from a domain expert
f- System examination
Have the D.E. examine and critique the prototype
systems rules and control structure.

D.E.

Examine and Critique

K.E.

16

Techniques used for extracting Knowledge

from a domain expert
g- System validation
Present the cases solved by D.E. and prototype
system to other outside domain experts.

D.E. (s)

Examine and Validate

K.E.

17

Example: Identification Problem

Consider aircraft identification problem, draw the
decision tree and find the produced production
rules according to the following order:

Engine type.

Number of engines

Wing shape.

Size and dimensions

Wing position.

Color and markings

Tail shape.

Speed and altitude

Bulges.
18

Assuming that we feel relatively confident with

the filtered set of attributes, we may list these and
their associated values.
Attribute/Type
C130
C141
C5A
B747
Engine type
Propeller
Jet
Jet
Jet
Wing position
High
High
High
Low
Wing shape Conventional Swept-back Swept-back Swept-back
Tail shape
Conventional
T-tail
T-tail
Conventional
Bulges
Under-wing
Aft-wing
None
Aft-cock pit

19

Solution-1
1- Decision Tree
If the attributes are
selected as follows:

Engine type
Wing position
Wing shape
Tail shape
Bulges

20

2- Production Rules
R1: IF X1 is Propeller THEN C130
R2: IF X1 is Jet AND X2 is Low THEN B747
R3: IF X1 is Jet AND X2 is high AND X3 is Conventional
THEN Unknown
R4: IF X1 is Jet AND X2 is high AND X3 is Swept-back AND
X4 is Conventional THEN Unknown
R5: IF X1 is Jet AND X2 is high AND X3 is Swept-back AND
X4 is T-tail AND X5 is Under-wing THEN Unknown
R6: IF X1 is Jet AND X2 is high AND X3 is Swept-back AND
X4 is T-tail AND X5 is None THEN C5A
R7: IF X1 is Jet AND X2 is high AND X3 is Swept-back AND
X4 is T-tail AND X5 is Aft-cockpit THEN Unknown
R8: IF X1 is Jet AND X2 is high AND X3 is Swept-back AND
X4 is T-tail AND X5 is Aft-wing THEN C141

21

Solution-2
1- Decision Tree:
If the Bulges attribute
only is selected

2- Production Rules:
R1: IF Bulges Under-wing THEN C130
R2: IF Bulges is None THEN C5A
R3: IF Bulges is Aft-cockpit THEN B747
R4: IF Bulges is Aft-wing THEN C141
22