Transcript artificial intelligence

ARTIFICIAL INTELLIGENCE
IS 524
CHANDRA S. AMARAVADI
1 of 41
ARTIFICIAL INTELLIGENCE
IN THIS PRESENTATION
Introduction to AI
Milestones & early work
Definition
Nature and types of knowledge
Knowledge representation
Examples
Neural nets
Expert systems
Business & industry applications
2 of 41
INTRODUCTION TO AI
3 of 41
THE HISTORY OF AI (FYI)
Major milestones
Turing m/c & test for intelligence -Rockefeller & Dartmouth conference --
1950
1956
AI as a field of study
Lisp language
Expert Systems
---
1958
1965
Small Talk, Prolog
Fifth Generation Project
Honda robot
DARPA & driver less vehicle
-----
1972
1981
1995
2004
Stanford driverless car
Driverless vehicles legal in CA
--
2012
Dendral & Mycin
4 of 41
EARLY RESEARCH
Early research on AI focussed on:
Logic
mathematical reasoning
Perceptrons
programs based on “on/off” model
Chess
board game with 8 x 8 squares
Blocks world programs
a world consisting of only blocks
5 of 41
SEARCH STRATEGIES
Search strategies are a result of early research
They are algorithms for finding a solution in a
large solution space. Types include:
Breadth-first
Depth-first
Heuristic
Hill-climbing
?
?
?
6 of 41
DEFINING INTELLIGENCE
7 of 41
DEFINITION
Artificial Intelligence (AI)
AI is concerned with the principles and mechanisms
for achieving intelligent behavior in machines
8 of 41
ACHIEVING INTELLIGENCE
A common approach to achieving intelligence
is to give machines knowledge and reasoning:
Knowledge + Reasoning
= Intelligence
Any other method of achieving intelligence?
9 of 41
THE NATURE OF
KNOWLEDGE
10 of 41
KNOWLEDGE
Information for problem solving
A goal in football is scored when the ball is …
To put in a new battery, unscrew the leads…
Art Deco is a style of architecture…
To become a GSA contractor, an entity must…
The elevators are located in the middle section…
“Lousiana report” refers to AACSB report given..
11 of 41
THE NATURE OF KNOWLEDGE
There are two types of knowledge from a knowledge engineering
perspective:
Declarative – Knowledge about an object (size, shape etc.)
Procedural – Knowledge about how to do something.
(how to install memory)
12 of 41
KNOWLEDGE
REPRESENTATION
13 of 41
KNOWLEDGE REPRESENTATION
Knowledge representation is concerned with
how to encode knowledge
Logic (Predicate logic)
Frames
Scripts
Semantic nets (Snets)
Rules
14 of 41
IDENTIFY THESE AS EXAMPLES
OF LOGIC, FRAMES, SCRIPTS…
EXAMPLE 1
sister_of(X,Y), bird_of_prey(X),
father_of(robin, Y)
father_of(robin,_)
EXAMPLE 2
EXAMPLE 3
If # of users > 300 then,
license fee = $500
If # of users < 300 then,
license fee = $300
is_a
: dbms
software cost
: $3,000
License cost
: check_with_vendor
no of users
: 2000
Max # of tables : 10,000
Supports ODBC : Yes
15 of 41
EXAMPLES OF KNOWLEDGE
REPRESENTATIONS..
EXAMPLE 5
EXAMPLE 4
P PTRANS P to P.O.
P ATTEND eyes to counter
P MBUILD line position
P PTRANS P to line
P PTRANS M to X
X PTRANS Stamps to P
bird
bird-of-prey
Is-a
Is-a
eagle
has-a
wings
1.5 m
max
wingspan
16 of 41
NOTES ON FRAMES
Frames are a representation scheme based on the way
we process situations:
“slots” and “fillers”
fillers can have:
values
is-a links
procedures (procedural attachments)
Most useful in classification problems
17 of 41
NOTES ON PREDICATE LOGIC
Predicate logic is a representation scheme based on logic.
predicates express relationships between symbols
symbols assigned meaning
father(X, Y) -- father of X is Y
subsidiary(A, B) – subsidiary of A is B
‘assertions’ or ‘propositions’ are made with predicates
sister(A, B) :- parents(A, P, Q), parents(B, P, Q)
i.e. sister of A is B if both have the same parents.
versatile technique useful in general reasoning
18 of 41
NOTES ON RULES
Rules are a representation scheme based on the way
we interact with everyday situations (S-R):
Thought to be used by experts
Have this format:
IF conditionTHEN action/conclusion
condition is expressed in terms of variables
If tax_bracket is high
if interest_rate > 5%
Useful if knowledge is conditional
Most useful in specialized domains
“shallow” reasoning
Note: S-R: stimulus/response
19 of 41
NOTES ON SCRIPTS
Scripts are a representation scheme based on the way
we react to complex situations (similar to frames):
Description (conceptual representation) of
actions in a pre-defined situation
Originated from film industry
Includes events, actions, actors/props
Used in understanding stories/narratives
20 of 41
NOTES ON SEMANTIC NETS
Semantic nets are a representation scheme based on
associative memory:
“node” + “link” formalism
nodes represent concepts or values (atomic)
links are of two types
structural – represent structure
descriptive – describe object
Useful for modeling relationships
21 of 41
DISCUSSION
1. Represent the following as semantic nets:
a) Onyx, Topaz and diamonds are precious stones
b) Diamond is a metamorphic rock!
c) Topaz is blue or yellow in color
d) Ann sent the memo to Mary and Jack.
e) John gave the red rose to his favorite cousin
2. Represent the following as rules:
a) For customers paying with credit card, discount is 15%
b) If a loan is greater than $10,000 classify as risky
c) Poly analyst (a sw program) will not work unless
you enter a registration code
3. Develop a frame suitable for mortgages (house or car)
4. Write using predicate logic:
a) manager of A is B
b) two people are office mates if they sit next to each other
22 of 41
DISCUSSION
What representation (if) any is suitable for each of the
following:
 According to the AACSB guidelines, MBA students
should be capable of providing leadership in complex organizational
situations.
 A conference paper is accepted if it is written with
clarity, the objectives are well stated, methodology is
sound, and the objectives are fulfilled.
 Insurance coverage is the obligation to compensate an insured for loss
suffered in a mishap or catastrophe
23 of 41
FOR DISCUSION
Identify knowledge items and represent
them using any method.
DP World has a portfolio of more than 65 marine terminals
across six continents, including new developments underway
in India, Africa, Europe, South America and the Middle East.
Container handling is the company’s core business and
generates more than three quarters of its revenue. In 2013, DP
World handled 55 million TEU (twenty-foot equivalent
container units). With its committed pipeline of developments
and expansions, capacity is expected to rise to more than 100
million TEU
http://www.tejari.com/cms/uae/resource-centres/case-studies/case_study_dp_world
24 of 41
EXAMPLE OF PREDICATE LOGIC
facts:
has_qualification(brad,3.2,620).
has_qualification(jill,4.0,540).
has_qualification(ted,3.5,320).
has_qualification(matt,3.8, 600).
clauses:
select(X) :- has_qualification(X,GPA,GMAT),
GPA>3.2, GMAT>550;
goals:
select(brad)? jill? matt? andrew?
25 of 41
APPLICATION OF FRAMES
Rocks
Sedimentary
Is_a: sedimentary rock
Name: limestone
Hardness: soft
Marks y/n: yes
Texture: coarse
Structure: amorphous
Metamorphic
Feldspar
Is_a:
Name:
Hardness:
Marks y/n:
Texture:
Structure:
Igneous
Obsidian
Quartz
26 of 41
THE TEST FOR MACHINE INTELLIGENCE
The Turing test: If a person interacting with an entity from a remote
location is unable to judge whether he/she is dealing with a computer or
a human, and the entity a machine, it is said to possess intelligence.
Questions
?
Responses
27 of 41
BRANCHES OF AI
Artificial
intelligence
Expert
Systems
NLP
Robotics
Vision
Systems
Machine
Learning
These are traditional branches of AI
28 of 41
TRADITIONAL BRANCHES OF AI ..
NLP: Natural language processing, concerned with understanding text and
speech as well as with language translation, handwriting recognition etc.
Expert Systems: A computer system that emulates the decision-making
ability of a human expert. Typical tasks include portfolio allocation,
locomotive repair etc.
Vision Systems: Computer based systems where software performs tasks
assimilable to "seeing", usually aimed at industrial quality assurance, part
selection, defect detection etc.
Robotics: The branch of AI that deals with mechanical or virtual intelligent
agents that can perform tasks automatically or with guidance, typically by
remote control e.g. painting, welding etc.
Machines Learning: Machine learning is the science of getting computers to
act without being explicitly programmed.
29 of 41
OTHER BRANCHES OF AI (FYI)
AI – OTHER
BRANCHES
Neural
Nets
Fuzzy
Logic
Intelligent
agents
Genetic
Algorithms
These are recent extensions of AI
30 of 41
EXPERT SYSTEMS
31 of 41
EXPERT SYSTEMS
Expert systems incorporate knowledge
of domain experts (SME)
predominantly in the form of thumb
rules so as to function
like an expert in a
specialized area.
KA
Subsystem
User
interface
Inference
engine
Knowledge
base
Note: SME– Subject Matter Expert
KA – Knowledge acquisition
32 of 41
EXAMPLE KNOWLEDGE BASE (FYI)
(defrule compare-object
=>
(printout t "What do you want to compare (cartridge-case or
bullet)?")
(assert (object-to-compare (read))))
(defrule comparable-ejector-mark
(object-to-compare cartridge-case)
=>
(printout t "Are the ejector marks comparable (yes or no)?")
(assert (ejector-mark-comparable (read))))
(defrule similar-ejector-mark
(and (object-to-compare cartridge-case)
(ejector-mark-comparable yes))
=>
(printout t "What is the similarity ratio of the ejector marks
(high or low)?")
(assert (ejector-mark-similarity (read))))
33 of 41
NEURAL NETWORKS
34 of 41
NEURAL NETS
Mathematical models to simulate neural models of the brain,
often used in applications requiring pattern recognition e.g.
crime, fraud, intrusion detection etc.
Neurons
Dendrites
nose
hair color
eyes
gait
Neural Net
(a math model)
The brain
35 of 41
MORE ON NEURAL NETS (FYI)
X
Y
1
1
OUTPUT
AF
1
-2
Activation Function
X+Y–2 >0
1) X = 1, Y = 1 => ?
2) X = 2, Y = 1 => ?
36 of 41
A NEURAL-NETWORK MODEL
Age
Loyal
Region
Hopper
Call
Rate
Lost
Length
Cust
Service
37 of 41
AI APPLICATIONS
38 of 41
BUSINESS APPLICATIONS OF AI
Following is a sampling of AI applications in businesses
Marketing
data/text mining
Automated voice response
Production applications
machine design
robotics
paper thickness
Scheduling of cranes
Accounting applications
detect irregularities
Financial applications
portfolio selection, credit approval
39 of 41
INDUSTRIAL APPLICATIONS OF AI
driverless vehicles
facial recognition
crime prevention
pothole recognition
locomotive fault diagnosis
drones
40 of 41
CHALLENGES IN AI
hardware for real time AI
understanding natural language/speech
recognizing objects
representing knowledge
recognizing user context
41 of 41
IS THE SINGULARITY NEAR?
Discuss the TECHNICAL feasibility and time frame for
each of the following technologies:
Domestic robots .
Neural uploads/downloads of information.
Replacing an organ involved in cognitive processing (eyes etc.)
Flexible manufacturing (fully automated)
What are the impacts of AI on business?
Society?
42 of 41
The End!
43 of 41