T02640010220104028T0264-Pert 01-2

8/2/2019 T02640010220104028T0264-Pert 01-2

1/29

8/2/2019 T02640010220104028T0264-Pert 01-2

2/29

Lecture 01Lecture 01 :: IntroductionIntroduction

CourseCourse : T: T02640264 Artificial IntelligenceArtificial Intelligence

YearYear : February 20: February 201111

8/2/2019 T02640010220104028T0264-Pert 01-2

3/29

Learning OutcomesLearning Outcomes

At the end of this session student will be able to:At the end of this session student will be able to:

Describes what AI is and identify the concept ofDescribes what AI is and identify the concept ofintelligent agentintelligent agent

Bina Nusantara University 3

8/2/2019 T02640010220104028T0264-Pert 01-2

4/29

OutlineOutline

What is Artificial Intelligence (AI)What is Artificial Intelligence (AI)

The foundation of AIThe foundation of AI

The History of AIThe History of AI

The State of the ArtThe State of the Art

SummarySummary


8/2/2019 T02640010220104028T0264-Pert 01-2

5/29

The exciting new effort tomake computers thinks

machine with minds, in the fulland literal sense(Haugeland 1985)

The art of creating machinesthat perform functions that

require intelligence whenperformed by people(Kurzweil, 1990)

The study of mental facultiesthrough the use of computational

models(Charniak et al. 1985)

A field of study that seeks toexplain and emulate intelligent

behavior in terms ofcomputational processes(Schalkol, 1990)

WHAT IS AI ?WHAT IS AI ?

B

ina Nusantara University 5

8/2/2019 T02640010220104028T0264-Pert 01-2

6/29

Different Types of Artificial IntelligenceDifferent Types of Artificial Intelligence

Modeling exactly how humans actually thinkModeling exactly how humans actually think cognitive models of human reasoningcognitive models of human reasoning

Modeling exactly how humans actually actModeling exactly how humans actually act models of human behavior (what they do, not how theymodels of human behavior (what they do, not how they

think)think)

Modeling how ideal agents should thinkModeling how ideal agents should think models of rational thought (formal logic)models of rational thought (formal logic)

note: humans are often not rational!note: humans are often not rational!

Modeling how ideal agents should actModeling how ideal agents should act rational actions but not necessarily formal rational reasoningrational actions but not necessarily formal rational reasoning i.e., more of a blacki.e., more of a black--box/engineering approachbox/engineering approach

B

ina Nusantara University 6

8/2/2019 T02640010220104028T0264-Pert 01-2

7/29

ActingActing HHumanly: Turing Testumanly: Turing Test Alan Turing's 1950 articleAlan Turing's 1950 article Computing Machinery andComputing Machinery and

IntelligenceIntelligence discussed conditions for considering adiscussed conditions for considering amachine to be intelligentmachine to be intelligent Can machines think?Can machines think? npnpCan machines behaveCan machines behave

intelligently?intelligently?

The Turing test (The Imitation Game): Operational definition ofThe Turing test (The Imitation Game): Operational definition ofintelligence.intelligence.


8/2/2019 T02640010220104028T0264-Pert 01-2

8/29

8/2/2019 T02640010220104028T0264-Pert 01-2

9/29

ActingActing RRationally:ationally: RRationalational AAgentgent

RationalRational behavior: Doing that was is expected to maximizebehavior: Doing that was is expected to maximize

ones utility function in this world.ones utility function in this world.

AnAn agentagentisis an entity that perceives and acts.an entity that perceives and acts. AA rational agentrational agent acts rationally.acts rationally. This course is about designingThis course is about designing

rational agentsrational agents

Abstractly, an agent is a function from percept histories to actions:Abstractly, an agent is a function from percept histories to actions:

[[ff: P*: P* A]A]

For any given class of environments and tasks, we seek the agentFor any given class of environments and tasks, we seek the agent(or class of agents) with the best performance(or class of agents) with the best performance

Caveat: computational limitations make perfect rationalityCaveat: computational limitations make perfect rationalityunachievableunachievable


8/2/2019 T02640010220104028T0264-Pert 01-2

10/29

Academic DisciplinesAcademic Disciplines IImportant to AImportant to AI

PhilosophyPhilosophy Logic, methods of reasoning, mind as physicalLogic, methods of reasoning, mind as physicalsystem, foundations of learning, language,system, foundations of learning, language,rationality.rationality.

MathematicsMathematics Formal representation and proof, algorithms,Formal representation and proof, algorithms,computation, (un)decidability, (in)tractability,computation, (un)decidability, (in)tractability,probability.probability.

EconomicsEconomics utility, decision theory, rational economic agentsutility, decision theory, rational economic agents NeuroscienceNeuroscience neurons as information processing units.neurons as information processing units. Psychology/Psychology/ how do people behave, perceive, process Cognitivehow do people behave, perceive, process Cognitive

ScienceScience information, represent knowledge.information, represent knowledge. ComputerComputer building fast computersbuilding fast computers

engineeringengineering Control theoryControl theory design systems that maximize an objectivedesign systems that maximize an objective

function over timefunction over time LinguisticsLinguistics knowledge representation, grammarknowledge representation, grammar


8/2/2019 T02640010220104028T0264-Pert 01-2

11/29

History of AIHistory of AI 19431943 McCulloch & Pitts: Boolean circuit model of brainMcCulloch & Pitts: Boolean circuit model of brain 19501950 Turing's "Computing Machinery and Intelligence"Turing's "Computing Machinery and Intelligence" 19561956 Dartmouth meeting: "Artificial Intelligence" adoptedDartmouth meeting: "Artificial Intelligence" adopted 1950s1950s Early AI programs, including Samuel's checkersEarly AI programs, including Samuel's checkers

program, Newell & Simon's Logic Theorist,program, Newell & Simon's Logic Theorist,Gelernter's Geometry EngineGelernter's Geometry Engine

19651965 Robinson's complete algorithm for logical reasoningRobinson's complete algorithm for logical reasoning 196619667373 AI discovers computational complexityAI discovers computational complexity

Neural network research almost disappearsNeural network research almost disappears

196919697979 Early development of knowledgeEarly development of knowledge--based systemsbased systems 19801980---- AI becomes an industryAI becomes an industry 19861986---- Neural networks return to popularityNeural networks return to popularity 19871987---- AI becomes a scienceAI becomes a science 19951995---- The emergence of intelligent agentsThe emergence of intelligent agents


8/2/2019 T02640010220104028T0264-Pert 01-2

12/29

State of the artState of the art Deep Blue defeated the reigning world chess champion GarryDeep Blue defeated the reigning world chess champion Garry

Kasparov in 1997Kasparov in 1997 Proved a mathematical conjecture (Robbins conjecture) unsolvedProved a mathematical conjecture (Robbins conjecture) unsolved

for decadesfor decades No hands across America (driving autonomously 98% of the timeNo hands across America (driving autonomously 98% of the time

from Pittsburgh to San Diego)from Pittsburgh to San Diego) During the 1991 Gulf War, US forces deployed an AI logisticsDuring the 1991 Gulf War, US forces deployed an AI logistics

planning and scheduling program that involved up to 50,000planning and scheduling program that involved up to 50,000vehicles, cargo, and peoplevehicles, cargo, and people

NASA's onNASA's on--board autonomous planning program controlled theboard autonomous planning program controlled thescheduling of operations for a spacecraftscheduling of operations for a spacecraft

ProverbProverb solves crossword puzzles better than most humanssolves crossword puzzles better than most humans

Stanford vehicle in Darpa challenge completed autonomously aStanford vehicle in Darpa challenge completed autonomously a132 mile desert track in 6 hours 32 minutes.132 mile desert track in 6 hours 32 minutes.


8/2/2019 T02640010220104028T0264-Pert 01-2

13/29

Consider what might be involved in buildingConsider what might be involved in buildinga intelligent computer.a intelligent computer.

What are the components that might be useful?What are the components that might be useful? Fast hardware?Fast hardware?

Foolproof software?Foolproof software?

ChessChess--playing at grandmaster levelplaying at grandmaster level ??

Speech interaction?Speech interaction?speech synthesisspeech synthesis

speech recognitionspeech recognition

speech understandingspeech understanding

Image recognition and understanding ?Image recognition and understanding ?

Learning?Learning?

Planning and decisionPlanning and decision--making?making?


8/2/2019 T02640010220104028T0264-Pert 01-2

14/29

Can we build hardwareCan we build hardware as complex asas complex asthe brain?the brain? How complicated is our brain?How complicated is our brain?

a neuron, or nerve cell, is the basic information processing unita neuron, or nerve cell, is the basic information processing unit estimated to be on the order of 10estimated to be on the order of 10 1111 neurons in a human brainneurons in a human brain

many more synapses (10many more synapses (101414

) connecting these neurons) connecting these neurons cycle time: 10cycle time: 10 --33 seconds (1 millisecond)seconds (1 millisecond)

How complex can we make computers?How complex can we make computers? 101066 or more transistors per CPUor more transistors per CPU supercomputer: hundreds of CPUs, 10supercomputer: hundreds of CPUs, 10 99 bits of RAMbits of RAM cycle times: order of 10cycle times: order of 10 -- 88 secondsseconds

ConclusionConclusion YESYES: in the near future we can have computers with as many basic: in the near future we can have computers with as many basic

processing elements as our brain, but withprocessing elements as our brain, but withfar fewer interconnections (wires or synapses) than the brainfar fewer interconnections (wires or synapses) than the brainmuch faster updates than the brainmuch faster updates than the brain


8/2/2019 T02640010220104028T0264-Pert 01-2

15/29

Must an Intelligent System be Foolproof?Must an Intelligent System be Foolproof? A foolproof system is one that never makes an error:A foolproof system is one that never makes an error:

Types of possible computer errorsTypes of possible computer errorshardware errors, e.g., memory errorshardware errors, e.g., memory errorssoftware errors, e.g., coding bugssoftware errors, e.g., coding bugs

humanhuman--like errorslike errors Clearly, hardware and software errors are possible in practiceClearly, hardware and software errors are possible in practice what about humanwhat about human--like errors?like errors?

An intelligent system can make errors and still be intelligentAn intelligent system can make errors and still be intelligent humans are not right all of the timehumans are not right all of the time

we learn and adapt from making mistakeswe learn and adapt from making mistakese.g., consider learning to surf or skie.g., consider learning to surf or ski we improve by taking risks and fallingwe improve by taking risks and falling an intelligent system can learn in the same wayan intelligent system can learn in the same way

Conclusion:Conclusion: NONO: intelligent systems will not (and need not) be foolproof: intelligent systems will not (and need not) be foolproof


8/2/2019 T02640010220104028T0264-Pert 01-2

16/29

Can Computers Talk?Can Computers Talk? This is known as speech synthesisThis is known as speech synthesis

translate text to phonetic formtranslate text to phonetic forme.g., fictitiouse.g., fictitious --> fik> fik--tishtish--eses

use pronunciation rules to map phonemes to actual sounduse pronunciation rules to map phonemes to actual sound

e.g., tishe.g., tish --> sequence of basic audio sounds> sequence of basic audio sounds

DifficultiesDifficulties sounds made by this lookup approach sound unnaturalsounds made by this lookup approach sound unnatural sounds are not independentsounds are not independent

e.g., act and actione.g., act and action

modern systems (e.g., at AT&T) can handle this pretty wellmodern systems (e.g., at AT&T) can handle this pretty well a harder problem is emphasis, emotion, etca harder problem is emphasis, emotion, etchumans understand what they are sayinghumans understand what they are sayingmachines dont: so they sound unnaturalmachines dont: so they sound unnatural

Conclusion: NO,Conclusion: NO, for complete sentences, but YES for individual wordsfor complete sentences, but YES for individual words


8/2/2019 T02640010220104028T0264-Pert 01-2

17/29

Can Computers Recognize Speech?Can Computers Recognize Speech?

Speech Recognition:Speech Recognition: mapping sounds from a microphone into a list of words.mapping sounds from a microphone into a list of words. Hard problem: noise, more than one person talking,Hard problem: noise, more than one person talking,

occlusion, speech variability,..occlusion, speech variability,.. Even if we recognize each word, we may not understand itsEven if we recognize each word, we may not understand its

meaning.meaning.

Recognizing single words from a small vocabularyRecognizing single words from a small vocabularysystems can do this with high accuracy (order of 99%)systems can do this with high accuracy (order of 99%)

e.g., directory inquiriese.g., directory inquiries limited vocabulary (area codes, city names)limited vocabulary (area codes, city names) computer tries to recognize you first, if unsuccessfulcomputer tries to recognize you first, if unsuccessful

hands you over to a human operatorhands you over to a human operator saves millions of dollars a year for the phone companiessaves millions of dollars a year for the phone companies


8/2/2019 T02640010220104028T0264-Pert 01-2

18/29

Recognizing normal speech is much more difficultRecognizing normal speech is much more difficult speech is continuous: where are the boundaries betweenspeech is continuous: where are the boundaries between

words?words?e.g., Johns car has a flat tiree.g., Johns car has a flat tire

large vocabularieslarge vocabulariescan be many tens of thousands of possible wordscan be many tens of thousands of possible wordswe can usewe can use contextcontextto help figure out what someone saidto help figure out what someone said

try telling a waiter in a restaurant:try telling a waiter in a restaurant:I would like some dream and sugar in my coffeeI would like some dream and sugar in my coffee

background noise, other speakers, accents, colds, etcbackground noise, other speakers, accents, colds, etc

on normal speech, modern systems are only about 60%on normal speech, modern systems are only about 60%accurateaccurate

Conclusion: NO,Conclusion: NO, normal speech is too complex to accuratelynormal speech is too complex to accuratelyrecognize, butrecognize, butYESYES for restricted problemsfor restricted problems

Can Computers Recognize Speech?Can Computers Recognize Speech?


8/2/2019 T02640010220104028T0264-Pert 01-2

19/29

Can Computers Understand speech?Can Computers Understand speech? Understanding is different to recognition:Understanding is different to recognition:

Time flies like an arrowTime flies like an arrow

assume the computer can recognize all the wordsassume the computer can recognize all the words

but how could it understand it?but how could it understand it?

1. time passes quickly like an arrow?1. time passes quickly like an arrow?

2. command: time the flies the way an arrow times the flies2. command: time the flies the way an arrow times the flies

3. command: only time those flies which are like an arrow3. command: only time those flies which are like an arrow

4. time4. time--flies are fond of arrowsflies are fond of arrows

only 1. makes any sense, but how could a computer figureonly 1. makes any sense, but how could a computer figure

this out?this out? clearly humans use a lot of implicit commonsense knowledgeclearly humans use a lot of implicit commonsense knowledge

in communicationin communication

Conclusion: NO,Conclusion: NO, much of what we say is beyond themuch of what we say is beyond thecapabilities of a computer to understand at presentcapabilities of a computer to understand at present


8/2/2019 T02640010220104028T0264-Pert 01-2

20/29

Can Computers Learn and Adapt ?Can Computers Learn and Adapt ?

Learning and AdaptationLearning and Adaptation consider a computer learning to drive on the freewayconsider a computer learning to drive on the freeway we could code lots of rules about what to dowe could code lots of rules about what to do and/or we could have it learn from experienceand/or we could have it learn from experience

machine learningmachine learning allows computers to learn to do thingsallows computers to learn to do thingswithout explicit programmingwithout explicit programming

Conclusion: YES,Conclusion: YES, computers can learn and adapt, whencomputers can learn and adapt, whenpresented with information in the appropriate waypresented with information in the appropriate way

Darpas Grand Challenge. Stanfords Stanley droveDarpas Grand Challenge. Stanfords Stanley drove150 without supervision in the Majove dessert150 without supervision in the Majove dessert


8/2/2019 T02640010220104028T0264-Pert 01-2

21/29

Recognition v. Understanding (like Speech)Recognition v. Understanding (like Speech) Recognition and Understanding of Objects in a sceneRecognition and Understanding of Objects in a scene

look around this roomlook around this roomyou can effortlessly recognize objectsyou can effortlessly recognize objects

human brain can map 2d visual image to 3d maphuman brain can map 2d visual image to 3d map

Why is visual recognition a hard problem?Why is visual recognition a hard problem?

Conclusion: mostly NO:Conclusion: mostly NO: computers can only see certaincomputers can only see certaintypes of objects under limited circumstances: buttypes of objects under limited circumstances: butYESYES for certainfor certainconstrained problems (e.g., face recognition)constrained problems (e.g., face recognition)

Can Computers see?Can Computers see?


8/2/2019 T02640010220104028T0264-Pert 01-2

22/29

Can Computers plan and make decisions?Can Computers plan and make decisions? IntelligenceIntelligence

involves solving problems and making decisions and plansinvolves solving problems and making decisions and plans e.g., you want to visit your cousin in Bostone.g., you want to visit your cousin in Boston

you need to decide on dates, flightsyou need to decide on dates, flightsyou need to get to the airport, etcyou need to get to the airport, etc

involves a sequence of decisions, plans, and actionsinvolves a sequence of decisions, plans, and actions

What makes planning hard?What makes planning hard? the world is not predictable:the world is not predictable:

your flight is canceled or theres a backup on the 405your flight is canceled or theres a backup on the 405 there is a potentially huge number of detailsthere is a potentially huge number of details

do you consider all flights? all dates?do you consider all flights? all dates?

no: commonsense constrains your solutionsno: commonsense constrains your solutions AI systems are only successful in constrained planning problemsAI systems are only successful in constrained planning problems

Conclusion: NO,Conclusion: NO, realreal--world planning and decisionworld planning and decision--making is still beyond themaking is still beyond thecapabilities of modern computerscapabilities of modern computers exception: very wellexception: very well--defined, constrained problems: missiondefined, constrained problems: mission

planning for sateplanning for satellites.ites.Bina Nusantara University 22

8/2/2019 T02640010220104028T0264-Pert 01-2

23/29

Intelligent Systems in Your Everyday LifeIntelligent Systems in Your Everyday Life Post OfficePost Office

automatic address recognition and sorting of mailautomatic address recognition and sorting of mail

BanksBanks automatic check readers, signature verification systemsautomatic check readers, signature verification systems

automated loan application classificationautomated loan application classification Telephone CompaniesTelephone Companies

automatic voice recognition for directory inquiriesautomatic voice recognition for directory inquiries

Credit Card CompaniesCredit Card Companies automated fraud detectionautomated fraud detection

Computer CompaniesComputer Companies automated diagnosis for helpautomated diagnosis for help--desk applicationsdesk applications

Netflix:Netflix: movie recommendationmovie recommendation

Google:Google: Search TechnologySearch Technology


8/2/2019 T02640010220104028T0264-Pert 01-2

24/29

AI Applications: Consumer MarketingAI Applications: Consumer Marketing Have you ever used any kind of credit/ATM/store card while shopping?Have you ever used any kind of credit/ATM/store card while shopping?

if so, you have very likely been input to an AI algorithmif so, you have very likely been input to an AI algorithm

All of this information is recorded digitallyAll of this information is recorded digitally Companies like Nielsen gather this information weekly and search forCompanies like Nielsen gather this information weekly and search for

patternspatterns general changes in consumer behaviorgeneral changes in consumer behavior tracking responses to new productstracking responses to new products identifying customer segments: targeted marketing, e.g., they find outidentifying customer segments: targeted marketing, e.g., they find out

that consumers with sports cars who buy textbooks respond well to offersthat consumers with sports cars who buy textbooks respond well to offersof new credit cards.of new credit cards.

Currently a very hot area in marketingCurrently a very hot area in marketing

How do they do this?How do they do this? Algorithms (data mining) search data for patternsAlgorithms (data mining) search data for patterns based on mathematical theories of learningbased on mathematical theories of learning completely impractical to do manuallycompletely impractical to do manually


8/2/2019 T02640010220104028T0264-Pert 01-2

25/29

AI Applications: Identification TechnologiesAI Applications: Identification Technologies

ID cardsID cards e.g., ATM cardse.g., ATM cards can be a nuisance and security risk:can be a nuisance and security risk:

cards can be lost, stolen, passwords forgotten, etccards can be lost, stolen, passwords forgotten, etc

Biometric IdentificationBiometric Identification walk up to a locked doorwalk up to a locked door

cameracamerafingerprint devicefingerprint devicemicrophonemicrophoneiris scaniris scan

computer uses your biometric signature for identificationcomputer uses your biometric signature for identificationface, eyes, fingerprints, voice pattern, iris patternface, eyes, fingerprints, voice pattern, iris pattern


8/2/2019 T02640010220104028T0264-Pert 01-2

26/29

AI Applications: Predicting the Stock MarketAI Applications: Predicting the Stock Market

The Prediction ProblemThe Prediction Problem

given the past, predict the futuregiven the past, predict the future very difficult problem!very difficult problem!

we can use learning algorithms to learn a predictive model fromwe can use learning algorithms to learn a predictive model fromhistorical datahistorical data

prob(increase at day t+1 | values at day t, tprob(increase at day t+1 | values at day t, t--1,t1,t--

2....,t2....,t--k)k)

?

?

time in days

Value ofthe Stock


8/2/2019 T02640010220104028T0264-Pert 01-2

27/29

AIAI--Applications: Machine TranslationApplications: Machine Translation Language problems in international businessLanguage problems in international business

e.g., at a meeting of Japanese, Korean, Vietnamese and Swedishe.g., at a meeting of Japanese, Korean, Vietnamese and Swedishinvestors, no common languageinvestors, no common language

or: you are shipping your software manuals to 127 countriesor: you are shipping your software manuals to 127 countries solution; hire translators to translatesolution; hire translators to translate

would be much cheaper if a machine could do this!would be much cheaper if a machine could do this!

How hard is automated translationHow hard is automated translation very difficult!very difficult! e.g., English to Russiane.g., English to Russian

The spirit is willing but the flesh is weak (English)The spirit is willing but the flesh is weak (English) the vodka is good but the meat is rotten (Russian)the vodka is good but the meat is rotten (Russian)

not only must the words be translated, but their meaning also!not only must the words be translated, but their meaning also!

Nonetheless....Nonetheless.... commercial systems can do alot of the work very wellcommercial systems can do alot of the work very well

(e.g.,restricted vocabularies in software documentation)(e.g.,restricted vocabularies in software documentation) algorithms which combine dictionaries, grammar models, etc.algorithms which combine dictionaries, grammar models, etc. see for example babelfish.altavista.comsee for example babelfish.altavista.com


8/2/2019 T02640010220104028T0264-Pert 01-2

28/29

Summary of Todays LectureSummary of Todays Lecture Artificial Intelligence involves the study of:Artificial Intelligence involves the study of:

automated recognition and understanding of speech, images, etcautomated recognition and understanding of speech, images, etc learning and adaptationlearning and adaptation reasoning, planning, and decisionreasoning, planning, and decision--makingmaking

AI has made substantial progress inAI has made substantial progress in recognition and learningrecognition and learning some planning and reasoning problemssome planning and reasoning problems

AI ApplicationsAI Applications

improvements in hardware and algorithms => AI applications inimprovements in hardware and algorithms => AI applications inindustry, finance, medicine, and science.industry, finance, medicine, and science.

AI ResearchAI Research many problems still unsolved: AI is a fun research area!many problems still unsolved: AI is a fun research area!


8/2/2019 T02640010220104028T0264-Pert 01-2

29/29

End of Session 01End of Session 01

Good Luck

Good Luck


T02640010220104028T0264-Pert 01-2

Documents

Transcript of T02640010220104028T0264-Pert 01-2