Artificial Intelligence - shoib and ruth

8/14/2019 Artificial Intelligence - shoib and ruth

1/57

Artificial IntelligenceCOMP3702/COMP7702

Course Coordinator

Janet Wiles

Lecturers

Shoaib Sehgal

Ruth Schulz

Tutor

Suren Rathnayake


2/57

Aims and Objectives

View: Methods and techniques within the field of

artificial intelligence and solve theoretical and practicalproblems.

Scope: The course provides an understanding of AI and

describes many of the most important algorithms andtechniques which are theoretcally pratically important.

Purpose: The course helps the student to:

gain an appreciation for the scientific context ofartificial intelligence

understand and develop computing algorithms, and toanalyse their properties

find the right techniques for solving specificproblems, and to implement these techniques


3/57

Aims and Objectives (cont)

In general terms, it is expected that the student gains an

understanding of the theories, methods and practices whichform the basis of Artificial Intelligence.

The course aims to introduce the basic concepts andmethods used in the field of artificial intelligence and

provide students with skills in the use of applying thesetechniques.

Specifically the course aims to give students an overview ofthe following topics in artificial intelligence:

Problem solving and optimisation (search algorithms) Reasoning with uncertain knowledge (probability

theory)

Machine Learning (classification, etc)

Probabilistic approaches for information retrieval


4/57


After the course, you should

be familiar with the historical context of artificialintelligence

know several definitions of artificial intelligence

be familiar with an agent-based intelligent system design understand several problem solving and optimisation

techniques based on search (both uninformed andinformed)

be able to implement and apply search techniques

understand general principles of machine learning (bothsupervised and unsupervised)


5/57


In addition, you should

know several machine learning techniques(including decision tree learning and neuralnetworks)

be able to implement, apply and systematicallyevaluate machine learning techniques

understand probability theory and how it can beused for representing and reasoning with uncertain

knowledge be able to apply basic probability theory to

machine learning problems

understand some probabilistic approaches to text

mining and information retrieval


6/57

Resources

Highly recommended reading Russell S. and Norvig P., Artificial Intelligence:

A modern approach, 2nd edition, 2003.

Used extensively (see reading list)

Available for purchase from the University

Bookshop

Several copies are also available at the library


7/57

Resources (cont)

Handouts

At www.itee.uq.edu.au/~comp3702 you will find

Slides used in lectures

Tutorials

Assignments Readings and handouts

Links and resources


8/57

Times and Venues

Sign-up for Tutorials via SI-net


9/57

Mid-semesterexam(optional)

Assignment 1preparation

Mid-semester exam;Discussion of assignment 1,related theory (6pp, 2pp)

1125 August6

10

Adversarialsearch(Comments,hints)

Applications of AIi- Bioinformatics(Shoaib)

ii- TBA - External Speaker

918 August5

8

InformedSearch(Comments,hints)

Chapter 6Adversarial search, gameplaying (6pp, 2pp)

7

11 August4

6

ProblemRepresentatio

n (Comments,hints)

Chapter 4(except

4.4 and4.5)

Informed search and

exploration (6pp, 2pp)

5

4 August3

4

Assignment 1available

The definitionof artificialintelligence

Chapter 3Solving problems bysearching (6pp, 2pp)

328 July2

2No tutorial

Chapters1, 2 and26: pp.

947-949,958-960.

Introduction to artificialintelligence, an agent-based

perspective (6pp, 2pp)

1

21 July1

Shoaib SehgalPart 1

AssessmentTutorialSession

TBA

ReadingRussell andNorvig, 2003

LectureTues 2-4pm

LectureNumber

MondaysDate

Week Number


10/57

Neural networks(Comments)

Assignment 2preparation

Chapter

23

Robotics and discussion ofassignment 2 (6pp,2pp,(corrected

30/10/07) more onapplications of NN)

2313

Oct

ober

12

22

Decision Trees and NaveBayes Classification(Comments, decision tree

solution)

Chapter 20(20.5-onwards)

Learning algorithms for

neural networks (6pp,2pp, (corrected30/10/07) more on NNlearning)

21

6October

11

Mid-semester break (one week)29 Sep.

20

Current best learning and

decision trees(Comments)

Chapter 20(20.5-

onwards)

Neural networks (6pp, 2pp(corrected 31/10/07),

more material on NNs,NetTalk audio (8MB))

19

22 Sep.10

18

Machine learningbasics (Comments)

Chapter 20(20.1-20.2)

Statistical machinelearning (6pp, 2pp)

1715 Sep.9

16

Assignment 2available

Probabilisticreasoning (Comments)

Chapters18and

19

Principles of machinelearning, decisiontrees (6pp, 2pp,

examples withdecision tree learning)

15

8 Sep.8

14

Assignment 1

deadline(Monday 1stSep., 5pm)

Chapter 13+ 7.1-7.2

Probabilistic reasoning(6pp, 2pp, example)

13

1 Sep.7

Ruth SchulzPart 2


11/57

10 NovemberExam Week 2

Final Exam

3 NovemberExam Week 1

Revision Period

27 October


12/57

Assessment

Assignments (30%)

2 assignments Assignment 1 (10%): Search game playing

Assignment 2 (20%): Machine learning pattern recognition

Tutorials (10%)

Active participation mark (1% per tutorial) Final Examination (60%)

During final examination period

Covers lecture material

Details 2 hours

Closed-book

Primarily short answer / short essay


13/57

Important Assessment Information

All assessment is due at 5pm of the due date

Assignments need to be submitted on-line at

http://submit.itee.uq.edu.au

Late submission not accepted except for medical or

strong personal reasons (documentation required)

The programming language will be Java

Tutorials for C/C++ programmers are available at

the course website.

Why Java ?


14/57

Why Java http://shootout.alioth.debian.org/

Intel AMD


15/57

Assignments

Two assignments Two problems/applications which

require intelligence (artificial or natural)

Problem-solving/optimisation

Approach: clever search algorithms

optimising outcomes on basis of a well-defined current state

exposes computational complexity issues

Pattern recognition Approach: learning-by-example/machine learning

exposes difficulties of representation by rules

illustrates the use of probabilistic methods and neural networks


16/57

Tutorials

Aims

to provide examples of potential examination

questions

to enable and encourage peer-tutoring

to provide an opportunity for questions

to explore the theoretical concepts

to apply the theoretical concepts


17/57


18/57

Lectures

Aims

to outline theories, methods and applications of

the field of AI

to explain difficult concepts from the

recommended text and other sources

to illustrate concepts in AI with diagrams and

examples

to provide a forum for general discussion andquestions about the subject matter


19/57

Some tips

Dont be shy, participate in lectures, askquestions.

Buy the book, read chapters as noted. It iswell-written, up-to-date, and an excellent

reference for later. AI is actually quite fun and useful, but you

need to work hard. Assignments and tutorials

will help you to work.


20/57

Examination

Closed book, non-programmable calculator allowed

Knowledge questions - theory may have been in tutorials

explicit in the recommended text book

assessed on correctness of answer

Knowledge questions - practical similar to those in tutorials

method described in the recommended text and lectures

assessed on correctness of method application

Discussion questions may have been addressed in lectures

not necessarily explicit in the readings

assessed on insight / justifications


21/57

Artificial intelligence:

An introduction

after which you should

know a couple of definitions if intelligence and AI

know how the Turing Test works

know the Physical Symbol System Hypothesis

understand the point made in Searles Chinese Room

Argument

have some familiarity with the history of AI

understand the concept of agents in the context of AI

be familiar with some intelligent agent designs

and feel prepared to discuss actual ways of makingmachines clever and making them learn bythemselves


22/57

What is Intelligence?

For each of the following, give three reasons

why:

(a) A dog is more intelligent than a worm.

(b) A human is more intelligent than a dog.

(c) An organisation is more intelligent than an

individual.


23/57

What is Intelligence?(adapted from J. McCarthy Stanford University)

What is intelligence?

One Answer

Intelligence is the computational part of the ability to

achieve goals in the world. Varying kinds and degrees

of intelligence occur in people, many animals and

some machines.

Do you agree with this definition?


24/57

What is AI?(adapted from J. McCarthy Stanford University)

What is artificial intelligence?

One possible answer

It is the science and engineering of making intelligent

machines, especially intelligent computer programs. It

is related to the similar task of using computers to

understand human intelligence AI does not have to confine itself to methods that are

biologically observable.


25/57

What is AI?(adapted from J. McCarthy Stanford University)

Is there a solid definition of intelligence that doesn't

depend on relating it to human intelligence?

Answer.

Not yet. The problem is that we cannot yet

characterise in general what kinds of computational

procedures we want to call intelligent. We understandsome of the mechanisms of intelligence and not

others.


26/57

Definitions of Artificial Intelligence

Computational Intelligence is the study of

the design of intelligent agents [Poole et al:

1998]

AI is concerned with intelligent

behaviour in artefacts [Nilsson: 1998]

The act of creating machines that

perform functions that require intelligence

when performed by people [Kurzweil:

1990]

The study of how to make computers do

things at which, at the moment, people are

better [Rich and Knight, 1991]

Systems that act rationallySystems that act like humans

The study of mental faculties through theuse of computational models [Charniak &

McDermott: 1985]

The study of the computations that make it

possible to perceive, reason, and act

[Winston: 1992]

The exciting new effort to makecomputers think machines with minds,

in the full and literal sense [Haugeland:

1985]

[The automation of] activities that we

associate with human thinking, activitiessuch as decision-making, problem solving,

learning [Bellman, 1978]

Systems that think rationallySystems that think like humans


27/57

Science of AI


28/57

The Turing Test Turing proposed a definition of Artificial

Intelligence independent of the innermechanisms of the machine

http://www.abelard.org/turpap/turpap.htm

Operational test of intelligent behaviour

Given an interrogator, and

(A) a computer and (B) a person in adifferent room

The interrogator, through asking questions,

must determine whether respondent A orrespondent B is the computer. If thecomputer fools the interrogator then we

can call it intelligent.


29/57

The Turing Test

What is

needed?

Knowledge

Reasoning

Language

Learning

http://chatbotgame.com/

Chatbot: Hello there human.

You: are you bot

Chatbot: Of course. I'm a collection of software rules.You: how are you

Chatbot: Life is tolerable. How are things in your world?

You: excellent

Chatbot: Can we change the subject please? Do you like

swimming?You: nop

Chatbot: "nop" ... What does that mean? It's not in my

dictionary.

You: nop = no

Chatbot: That's too bad.You: why

Chatbot: Because of an intense metaphysical connection in

the properties of submolecular particles

You: what?Chatbot: Do you understand me?

You: noChatbot: Well I don't understand humans.. Lol


30/57

Attributes of the Turing Test

Positive

standard test

bypasses the true nature of intelligence

removes bias

Negative

focus on symbolic tasks

compares machine with human intelligence far too restrictive

difficult to do in practice


31/57

Lets play: Spot the Intelligence!

Order the following from most to least

intelligent

Cockroach

Computer

Elevator

Hammer

Human

Model-T ford motor car

Sunflower


32/57

Answers?

Simple yes / no answers

cannot be made from

current definitions.

Your answer reflects

your beliefs, and is not

necessarily correct or

incorrect.

My Ranking Human

Cockroach

Sunflower

Computer

Elevator

Model-T Ford

Motor Car

Hammer


33/57

Examples


34/57

Average Rank (from 2000)

Average Ranking

0

1

2

34

5

6

7

8

Cockroac

h

Compute

r

Elevato

r

Hamme

r

Human

Model-

T

Sunflowe

r

Average


35/57

Results of Intelligence Ranking

Most Popular Response

0

1

2

34

5

6

7

8

Human

Cock

roach

Sunflowe

r

Compute

r

Elevato

r

Model-

T

Hamme

r

Ra

nk


36/57

Physical Symbol System Hypothesis(Simon and Newell, 1976)

A physical symbol system has the necessary and sufficient means for intelligentaction.

A system:

Consists of a set of entities, called symbols,

Symbols can occur as components of another type of

entity called an expression (or symbol structure) A symbol structure is composed of a number of instances

(or tokens) of symbols related in some physical way (suchas one token being next to another).

Contains a collection of these symbol structures

Contains a collection of processes that operate on expressionsto produce other expressions: processes of creation,modification, reproduction and destruction

Set of finite symbols that can be composed to form a potential

infinite set of expressions


37/57

Physical Symbol System Hypothesis(Simon and Newell, 1976)

A physical symbol system has the necessary and sufficient means forintelligent action.

Intelligent human thought: the symbols are encodedin our brains. The expressions are thoughts. Theprocesses are the mental operations of thinking.

A running artificial intelligence program: Thesymbols are data. The expressions are more data. The

processes are programs that manipulate the data.


38/57

Searles Chinese Room Thought

Experiment

You are: Monolingual English speaker locked in a room

You are given a large batch of Chinese writing

a second batch of Chinese script as an output

a set of rules in English for correlating the second batch with thefirst batch

The rules correlate one set of formal symbols with another set offormal symbols

Your responses are indistinguishable from those of Chinese

speakers

Just by looking at your answers, nobody can tell you "don'tspeak a word of Chinese."


39/57

Searles Chinese Room

The Point: No matter

how intelligent a computer seems to behaves and

no matter what programming makes it behave that

way

since the symbols it processes are meaningless

(lack semantics) to it

its not really intelligent


40/57

Searles Chinese Room

Demonstrates that a system can be merely following

rules, but not understanding anything at all.

The system he describes will pass the Turing Test

Behaviour cannot determine extent of understanding

Is following rules sufficient for intelligence?


41/57

Strong AI vsWeak AI

Strong AI

duplication of intelligence

aims to understand intelligence

Weak AI

simulation of intelligence

aims to make computers more useful


42/57


43/57

Modern History

1956

John McCarthy coined the term artificial intelligence as

the topic of the Dartmouth Conference.

Demonstration of AI program - the Logic Theorist (LT) -

written by Newell, Shaw and Simon (CMU).

1952-62

Samuel (IBM) wrote the first game-playing program that

learns.

1962 First industrial robot company, Unimation, is founded.


44/57

Modern History

1965

Weizenbaum (MIT) built ELIZA, an interactive programthat carries on a dialogue in English on any topic. Itbecame a popular toy at AI centres.

See http://www-ai.ijs.si/eliza-cgi-bin/eliza_script

Or http://www.manifestation.com/neurotoys/eliza.php3

1967 Dendral program (Feigenbaum, Lederberg, Buchanan and

Sutherland) is the first successful knowledge-basedprogram for scientific reasoning.

1968 Minsky and Papert publish Perceptrons, demonstrating

the limits of simple neural nets.


45/57

Modern History

1969

SRI robot, Shakey, demonstrates combining

locomotion, perception and problem solving.

1974

The first expert system - MYCIN (Stanford) -

demonstrates the power of rule-based systems for

knowledge representation and inference in the

domain of medical diagnosis and therapy.


46/57

Modern History

1969-1979

Knowledge-based (expert) systems

1980-1988

Expert systems industry booms

1988-1993

Expert systems industry busts AI winter1966 US Military report on machine translation: The spirit is willing but the

flesh is weak was re-translated (English-Russian-English) intothe vodka is good but the meat is rotten. Prospects seemed poor and

funding was cut.

2008 version (Babel fish): Spirit is willingly ready but flesh it is weak


47/57

Modern History

Mid-1980s-present The return of Neural Networks

1985 Brooks (MIT) develops the concept of behaviour based robots

1988-present Resurgence of probability: general increase in technical depth

Computational Intelligence / soft computing (Evolutionarycomputing, Swarm Intelligence, Fuzzy systems, NNs)

1995-present Agents, agents everywhere


48/57

Agents

Generally computers are obedient, literal,

unimaginative servants

acceptable for most applications

However, increasingly we require systems

that can decide for themselves

Such computer systems are known as agents


49/57


50/57

I, Robot


51/57

Agents and Environments

Environment Agent?

SensorsPercepts

ActionsThe agent takessensory input from the

environment and

produces as outputsactions that affect it.

Agents include humans,robots, softbots,thermostats, etc

Effectors

Rationality depends on

performance, degree of success,perceptual history,knowledge of environment, andactions available for deployment

Lets look at these aspects in terms

of different agent designs

An agent consists of

an architecture, anda program.


52/57

Agent Types

Simple reflex agents

Model-based reflex agents Goal-based agents

Utility-based agents

Example: Automated taxi driver

Cameras,

speedometer,

GPS, sonar,

microphone

Steer,

accelerate,

brake, talk to

passenger

Roads, other

traffic,

pedestrians,

customers

Safe, fast, legal,

comfortable trip,

maximise profits

Taxi Driver

SensorsActuatorsEnvironmentPerformance

MeasureAgent Type


53/57

Simple reflex agents

En

vironment

Agent

Condition-action rulesWhat action I

should do now

What the worldis like now

Actuators

Sensors


54/57

Model-based reflex agents

En

vironment

Agent

Condition-action rulesWhat action I

Should do now

What the world islike now

Actuators

Sensors

What my actions do

How the world evolves

State


55/57

Goal-based agents

En

vironment

Agent

GoalsWhat action I

should do now


Actuators

Sensors

What my actions do


State

What it will be likeif I do action A


56/57

Utility-based agents

En

vironment

Agent

UtilityHow happy I will be

in such a state

Actuators

Sensors

What my actions do


State

What it will be likeif I do action A

What action Ishould do now



57/57

Questions ?

Office HoursTuesday 10-12 AM

Room 308, Axon building ITEE

Acknowledgements:http://domus.usherbrooke.ca

www.ironbot.com

for some of the images used in making of these slides.

Artificial Intelligence - shoib and ruth

Documents

Transcript of Artificial Intelligence - shoib and ruth