Transcript Document
Artificial Intelligence
Chapter 12
Definition:
• Artificial Intelligence (AI):
– “The activity of providing such machines as
computers the ability to display behavior that
would be regarded as intelligent if it were
observed in humans.”
AI
ARTIFICIAL INTELLIGENCE
(AI) SYSTEMS:
(Laudon & Laudon Definition)
AI: COMPUTER-BASED SYSTEMS WITH
ABILITIES TO LEARN LANGUAGE,
ACCOMPLISH TASKS, USE
PERCEPTUAL APPARATUS, EMULATE
HUMAN EXPERTISE & DECISION
MAKING
*
History of AI
1950
• Turing Test
– "Can machines think?"
• Loebner Prize
– $100,000 Grand Prize
– Not yet awarded
1950:
• Alan Turing proposes the “Turing Test”
for computers
• Can a computer pass for a human?
1952:
• UNIVAC correctly predicts Dwight
Eisenhower’s election with only 7% of
votes reported
1956: “Artificial Intelligence”
• John McCarthy coins the term in 1956 as
the theme of a conference held at
Dartmouth College.
“Artificial Intelligence”
• Dartmouth, 1956
• 25-year Prediction (1981):
– Prediction: in 25 years (1981) (would be
before George Orwell’s 1984)
– Intelligent machines would be able to do all
the physical and intellectual work for human
beings.
– Leaving people to devote all their time to
recreational activities.
1958:
• John McCarthy:
• If we work really hard, we’ll have an
intelligent system in from four to four
hundred years.
1958:
• Herbert Simon:
• Said that a program would be chess
champion in ten years (by 1968).
Deep Blue
• 1997 IBM’s computer “Deep
Blue” defeats world chess
champion, Gary Kasparov.
• First time a computer had
defeated a top-ranked chess
player
• Not Undisputed
Major Areas of AI:
•
•
•
•
Expert Systems
Neural Networks
Perceptive systems
Robotics
AI
EXPERT SYSTEMS
KNOWLEDGE - INTENSIVE
CAPTURES HUMAN EXPERTISE
IN LIMITED DOMAINS OF
KNOWLEDGE
*
Development of
Expert Systems
• What is Expertise?
– Skill and knowledge whose input
into a process results in
performance high above the
norm.
First-to-100-game
• Rules:
– 2 Players alternate by adding a number to the
total.
– Numbers must be within 1-10.
– First player to reach 100 wins
Following a Set of Rules /
Pattern Recognition
• The game can easily be won by anyone who
recognizes the pattern…
• You must be the first to 89 in order to be the
first to 100…
Development of
Expert Systems
Components
of Expert
Systems
The
interface or dialog
The
knowledge base
The
interface engine
Development of
Expert Systems
Components of an expert system; numbers indicate the order of the
processes
Expert Systems
• The Benefits
–
–
–
–
Longevity
Cost savings
Availability
Replicable
Contribution of
Expert Systems
• Areas where ESs can help in
business
–
–
–
–
–
–
–
–
–
Planning
Decision making
Monitoring
Diagnosis
Training
Incidental learning
Replication of expertise
Timely response
Consistent solutions
Contribution of Expert Systems
Major reasons for using expert systems
Expert Systems in Action
• Business areas using ESs
–
–
–
–
–
Telephone network maintenance
Credit evaluation
Tax planning
Detection of insider securities trading
Mineral exploration
– Legal Advice/ Medical Advice
– Visa & M/C: 2 purchases + 1
Gas out of town: call for
verification
Knowledge
Representation Methods
• Factors Justifying the
Acquisition of Expert Systems
AI
•
•
•
•
EXPERT SYSTEMS
LIMITATIONS:
Often reduced to problems of classification
Can be large, lengthy, expensive
Maintaining knowledge base critical
Many managers unwilling to trust such
systems
*
Limitations of
Expert Systems
• Three limitations of ESs
– Can handle only narrow domains
– Do not possess common sense
– Have a limited ability to learn
Bobs Cars
• Simple A.I. Application based on weights
+/- w/ each choice you make
• http://www.src-net.com/BobsCars/fbdeal.htm
AI: Neural Networks
Neural Networks
• Biologically inspired flexible
statistical models.
• Function approximations
– Offers not only point estimates but
also converges on the derivatives of
the unknown functions
Neural Networks
• A mathematical model
of the human brain
that simulates the way
that neurons interact to
process data and learn
from experience.
Human Neurons
• Dendrites (input)
• Soma (processor)
• Axons (output)
Biological Neural Network
• Patterns of electrical impulses from cell to cell
form memory.
From Biological to
Artificial Neural Networks
Neural nets simulate the association and inference that take place in a
network of neurons in the human brain. Instead of a network of
neurons, a network of nodes is developed.
Artificial Neuron
• Y is the result of
the weighted
input signals
• Any non-linear
function can be
used, the Sigmoid
function is the
most popular
Multi-Layered Artificial Neural
Network (A.N.N.)
• All possible interactions are considered
• All relationships are considered non-linear
• High inter-correlation is not a problem
Specific Examples of A.N.N.
•
•
•
•
•
•
•
•
•
•
Bankruptcy Prediction
Forecasting Stock Prices
Direct Marketing Mail Prediction
Credit Scoring
Real Estate Appraisal
Finding Gold (testing soil samples)
Thoroughbred Horse Racing: 17 wins in 22 races
Weather Forecasting
Beer Testing
Credit Card Fraud Detection
• Mars-Rock testing
Neural Network Simulator
for Character Recognition
http://diwww.epfl.ch/mantra/tutorial/english/ocr/html/index.html
AI: Ethical and Societal Issues
Ethical and Societal Issues
Too Sophisticated Technology
• Increasing dependence on machine intelligence
raises legal and ethical issues.
– Who is legally responsible for advice provided by a
program?
– Is expert judgment needed to interpret program output?
– Does machine expertise replace or complement the
‘real thing’?
– How do we know if the experts behind expert systems
are expert at all?
Ethical and Societal Issues
Too Sophisticated Technology
• Malfunctions of an ES can be
caused by anyone involved in the
development.
– Experts who contribute knowledge
– Knowledge engineer who builds the
system
– Professional who uses the ES
– The person who is affected by the
decision
British Telecom
• “Soul Catcher”
• Implant a microchip in
the human brain
Questions?
Needed Links:
Bobs Cars
• Simple A.I. Application based on weights
+/- w/ each choice you make
• http://www.src-net.com/BobsCars/fbdeal.htm
Neural Network Simulator
for Character Recognition
http://diwww.epfl.ch/mantra/tutorial/english/ocr/html/index.html