Artificial Intelligence (AI)

1
Artificial Intelligence (AI)

• Dr. Merle P. Martin
• MIS Department
• CSU Sacramento

2
Acknowledgements

• Dr. Russell Ching (MIS Dept) Source
  Materiel / Graphics
• Edie Schmidt (UMS) - Graphic Design
• Prentice Hall Publishing (Permissions)
• Martin, Analysis and Design of Business
  Information Systems, 1995

3
Agenda

• Gate Assignment Problem
• Artificial Intelligence
• Expert Systems (ES)
• ES Examples

4
In the Airline Industry

• United Airlines' GADS (Gate Assignment Display
  System)
• Trans World Airlines' GATES (Gate Assignment and
  Tracking Expert System)

5

• Boeing 747, 387-427 capacity

Lockheed L-1011, 252 capacity
6
Boeing 767, 170-227 capacity
Boeing 727, 115-134 capacity
7
McDonnell Douglas DC-9, MD-80 73-132 capacity
8
Gate Assignment Problem
9
Gate Assignment Problem

• Constraints
• Matching size of aircraft to gate 8 different
  types with United 6 with TWA
• Minimizing distances between connecting flights
• Foreign vs. domestic flight

10
GATES Constraints

• Constraints without exceptions
• Gate size
• Constraints with exceptions
• International versus domestic flights
• Constraints with changing tolerances
• Turn-around times

11
GATES Constraints

• Guidelines
• Taxiway congestion
• Convenience constraints
• Time between flights
• Distance between connecting flights

12
Gate Assignment

• ES benefits
• Task of scheduling gate assignments for a
  month reduced from 15 hours
  to 30 seconds.
• ES can be transferred to other airport
  operations, reducing training / operating
  costs.

13
Gate Assignment

• Benefits (Cont.)
• Decrease susceptibility of schedule to
  moods and whims of schedulers.
• Gate assignments can be done on demand
  with little interference to current
  operations.

14
Gate Assignment

• Benefits (Cont.)
• Managers can review impact of
  changes, implement changes (i.e., what-if
  analysis).
• ES integrated into airlines' major
  operations / scheduling systems through
  direct electronic interfaces, thus expediting
  scheduling.

15
Artificial Intelligence (AI)

• Effort to develop
  computer-based systems
• that behave like humans
• learn languages
• accomplish physical tasks
• use a perceptual apparatus
• emulate human thinking

16
AI Branches

• Natural Language
• Robotics
• Perceptive Systems
• Expert Systems
• Intelligent Machines

17
Human Processing Capabilities

• Induction
• act on inconsistently formatted
  data
• fill in the gaps
• CN U RD THS
• Wheel of Fortune
• Adaptiveness

18
Human Processing Capabilities

• Insight
• creativity
• create alternatives
• chess game
• perspicuous grouping

19
Perspicuous Grouping

• Recognize that we can handle
  only a few alternatives
• Short Term Memory (STM)
• Millers 7 /- 2 Rule
• Zero in on a few viable alternatives
• Enumerate / select best
• Satisficing, rather than optimizing
• Herbert Simons 1958 Chess prediction

20
Computer Processing Capabilities

• Handle large volume of data
• quickly
• Detect signals
  where humans sense noise
• Tireless

21
Computer Capabilities

• Consistent
• Objective
• no selective perception
• Not distracted
• Minimal down-time

22
Issue

• A Stanford Research Institute (SRI)
  scientist once said, You
  neednt fear intelligent machines. Maybe
  theyll keep us as pets.
• Will intelligent machines replace us?
• Why or why not?
• WHAT DO YOU THINK?

23
What is an ES?

• Feigenbaum, 1983
• intelligent computer program
• using knowledge / inference procedures
• to solve problems difficult enough
• to require significant human expertise
• a model of the expertise of
• the best practitioners

24
Components of an Expert System
Knowledge Base
Knowledge
Acquisition
Facts and Rules
Facility
Recom-mended
Inference Engine
Action
Explanation
User Interface
Facility
User
25
Rule Induction
Rules Induced From Example Cases
Case Classified Through Deduction
Individual Cases Applied to the Rules
Induction(Inductive Logic)
Deduction(Deductive Logic)
26
Check Overdraft Cases
Decision Attributes
Decision
Overdraft for Single or Multiple Checks
Pay or Reject
Type of Account
Credit Rating
Pay
Regular
Good
Multiple
Pay
Student
Unknown
Single
Reject
Student
Poor
Single
Reject
Student
Good
Multiple
Pay
Student
Good
Single
27
Check Overdraft Cases (Cont.)
Decision Attributes
Decision
Overdraft for Single or Multiple Checks
Pay or Reject
Type of Account
Credit Rating
Pay
Regular
Unknown
Multiple
Pay
Regular
Good
Single
Reject
Regular
Poor
Single
Reject
Student
Unknown
Multiple
Reject
Regular
Unknown
Multiple
28
Pay or Reject?
Overdraft for Single or Multiple Checks
Pay or Reject
Type of Account
Credit Rating
?
Regular
Unknown
Single
29
Bank Overdraft Application

• 340 Cases of check
  overdrafts
• Classification Variable
• Check unpaid(0) or paid (1)

30
ID3 DECISION TREE
CR DIFFlt6.5
176
Yes
No
130
CRDIFFlt5.5
DIFFlt10.5
60
116
5
125
Yes
No
Yes
No
CR DIFFlt.035
DIFFlt20.5
DIFFlt9.4
DIFFlt40.3
59
1
15
101
57
68
4
1
ACTDIFF
COVDIFFlt1.5
DIFFlt42.2
DIFFlt1.65
lt19.6
0
14
1
69
50
9
1
32
53
2
2
0
1
56
15
1
Pay
Reject
Pay
Reject
DIFFlt5.55
ACTDIFFlt.175
48
4
1
0
5
2
32
0
0
0
0
15
56
1
0
1
Reject
Reject
Reject
Pay
Reject
Pay
0
2
3
ACTDIFFlt3
2
1
0
54
2
Overall Classification Rate 97.7
Pay
Reject
Pay
2
0
1
1
Reject
Pay
31
Reasons For Using ES

• Consistent
• Never gets bored / overwhelmed
• Replace absent, scarce experts
• Quick response time

32
ES Reasons

• Reduced down-time
• Cheaper than experts
• Integration of multi-expert opinions
• Eliminate routine / unsatisfactory jobs
  for people

33
ES Limitations

• High development cost
• Limited to relatively simple problems
• operational mgmt level
• Can be difficult to use
• Can be difficult to maintain

34
When to Use ES

• High potential payoff
• OR
• Reduced risk
• Need to replace experts
• Campbells Soup

35
When to Use ES

• Need more consistency
  than humans
• Expertise needed
  at various locations
  at same time
• Hostile environment dangerous
  to human health

36
ES Versus DSS

• Problem Structure
• ES structured problems
• clear
• consistent
• unambiguous
• DSS semi-structured problems

37
ES Versus DSS

• Quantification
• DSS quantitative
• ES non-mathematical reasoning
• IF A BUT NOT B, THEN Z
• Purpose
• DSS aid manager
• ES replace manager

38
Issue

• Does your company use Expert
  Systems (ES)?
• How do they?
• How might they?
• WHAT ARE YOUR EXPERIENCES?

39
MYACIN

• Diagnose patient symptoms
  (triage)
• free doctors for
  high-level tasks
• Panel of doctors
• diagnose sets of symptoms
• determine causes
• 62 accuracy

40
MYACIN

• Built ES with rules
  based on panel consensus
• 68 accuracy
• Why better than doctors?
• Heuristics

41
Stock Market ES

• Reported by Chandler, 1988
• Expert in stock market analysis
• 15 years experience
• published newsletter
• Asked him to identify data
  used to make recommendations

42
Stock Market ES

• 50 data elements identified
• Reduced to 30
• redundancy
• not really used
• undependable
• Predicted for 6 months of data whether
  stock value would increase, decrease, or stay the
  same

43
Stock Market ES

• Rule-based ES built
• Discovered that only
  15 data elements came into play
• Refined the ES model
• Results were better than expert
• WHY?

44
USA Expert Systems
Manufacturing Planning
HICLASS - Hughes (process plans, manufacturing
instructions)
CUTTECH - METCUT (plans for machining operations)
XPSE-E - CAM-I (plans for part fabrication)
45
USA Expert Systems
Manufacturing Control
IMACS - DEC (plans for computer hardware
fabrication and assembly)
IFES - Hughes (models dynamic flow of factory
information)
46
USA Expert Systems
Factory Automation
Move - Industrial Technology
Institute (material handling)
Dispatcher - Carnegie Group, Inc. (materials
handling system)
GMR - GM Corp. (flexible automation assembly
system)
FMS/CML - Westinghouse (simulation for FMS
design, planning, control)
47
Issue

• Expert systems are dangerous.
  People are likely to be
  dependent on them rather than
  think for themselves.
• WHAT DO YOU THINK?

48
Points to Remember

• What is AI?
• What is an ES?
• When to use an ES
• Differences between
  ES and DSS
• ES examples