Introduction-To-AI(L-1) for civil engineers

AI for civil engineers.

1. INTRODUCTION TOARTIFICIAL
INTELLIGENCE

2. Objectivesof thisCourse
•Thisclassisabroadintroductiontoartificial
intelligence(AI)
oAIisaverybroadfieldwithmanysubareas
•Wewillcovermanyoftheprimaryconcepts/ideas
•Butin15weekswecan’tcovereverything

3. Today’sLecture
•What is intelligence? What is artificialintelligence?
•AverybriefhistoryofAI
oModern successes: Stanley the drivingrobot
•An AIscorecard
oHowmuchprogresshasbeenmadeindifferentaspectsofAI
•AI inpractice
oSuccessfulapplications

4. AIandSoftComputing:ADifferentPerspective
AI: predicate logic and symbol manipulation
techniques
User
Interface
Inference
Engine
Explanation
Facility
Knowledge
Acquisition
KB:
•Fact
•rules
Global
Database
Knowledge
Engineer
Human
Expert
Question
Response
ExpertSystems
User

5. AI and SoftComputing
ANN
Learningand
adaptation
Fuzzy Set Theory
Knowledge representation
Via
Fuzzy if-thenRULE
Genetic Algorithms
Systematic
RandomSearch
AI
Symbolic
Manipulation

6. AI and SoftComputing
cat
cut
knowledge
Animal? cat
Neural character
recognition

7. 5
What is Hard Computing?
•Hard computing, i.e., conventional
computing, requires a precisely stated
analytical model and often a lot of
Computational Time.
•Many analytical models are valid forideal
cases.
•Real world problems exist in a non-ideal
environment.

8. 6
Premises and guiding principles of
HardComputing
Precision, Certainty, andRigor.
•Many contemporary problems do not lend
themselves to precise solutions such as:
Recognition problems (handwriting,
speech,objects, images,texts)
Mobile robot coordination,forecasting,
combinatorial problemsetc.
Reasoning on naturallanguages

9. 7
Recognition problems (handwriting, speech,
objects, images,texts)

10. 8
Mobile robotCoordination
weatherforecasting
Natural LanguageProcessing

11. What isArtificial
Intelligence?

12. Some Definitions(I)
The exciting new effort tomake
computers think…
machines with minds,
in the full literalsense.
Haugeland,1985

13. Some Definitions(II)
The study of mental faculties through theuse
of computationalmodels.
Charniak and McDermott,1985
A field of study that seeks to explainand
emulate intelligent behavior in terms of
computationalprocesses.
Schalkoff, 1990

14. Some Definitions(III)
The study of how to make
computersdothingsatwhich,at
the moment, peopleare better.
Rich & Knight,1991

15. Outline of theCourse
•Knowledgerepresentation:
opropositionallogicandfirst-orderlogic
oinference in Expert Systems
oFuzzylogic
oRoughset
oMachinelearning: classification trees
oNeuralnetworks
oOhers?

16. What isntelligence?
•Intelligence:
o―thecapacitytolearnandsolveproblems‖(Webstersdictionary)
oinparticular,
•theabilitytosolvenovelproblems
•theabilitytoactrationally
•theabilitytoactlikehumans
•ArtificialIntelligence
obuildandunderstandintelligententitiesoragents
o2 main approaches: ―engineering‖ versus ―cognitivemodeling‖

17. WhatisArtificial Intelligence?
(John McCarthy, StanfordUniversity)
•What isartificialintelligence?
Itisthescienceandengineeringofmakingintelligentmachines,especiallyintelligent
computer programs. It is related to the similar task of using computers to understand
human intelligence, but AI does not have to confine itself to methods that are
biologicallyobservable.
•Isn't there a solid definition of intelligence that doesn't depend on
relating it to human intelligence?
Not yet. The problem isthat we cannot yet characterize in general what kinds of
computationalprocedureswewanttocallintelligent.Weunderstandsomeofthe
mechanismsofintelligenceandnotothers.
•More in:http://www-formal.stanford.edu/jmc/whatisai/node1.html

18. What’s involvedin
Intelligence?
•Ability to interact with the real world
oto perceive, understand, andact
oe.g.,speechrecognitionandunderstandingandsynthesis
oe.g., imageunderstanding
oe.g.,abilitytotakeactions,haveaneffect
•Reasoning andPlanning
omodelingtheexternalworld,giveninput
osolvingnewproblems,planning,andmakingdecisions
oabilitytodealwithunexpectedproblems,uncertainties
•Learning andAdaptation
owearecontinuouslylearningandadapting
oourinternalmodelsarealwaysbeing“updated”
•e.g.,ababylearningtocategorizeandrecognizeanimals

19. Academic Disciplines important
toAI.
•Mathematics Formal representation andproof,
algorithms,
computation, (un)decidability,(in)tractability,
•Economics
probability.
utility, decision theory, rationaleconomic
neurons as information processingunits.
agents
•Neuroscience
•Psychology/ howdopeoplebehave,perceive,process
CognitiveScience
•Computer
information,representknowledge.
building fastcomputers
engineering
•Controltheory design systems that maximize anobjective
•Linguistics
function overtime
knowledge representation,grammar

20. History ofAI
•1943: earlybeginnings
oMcCulloch&Pitts:Booleancircuitmodelofbrain
•1950:Turing
oTuring's "Computing Machinery andIntelligence―
•1956: birth ofAI
oDartmouthmeeting:"ArtificialIntelligence―
nameadopted
•1950s: initial promise
oEarly AI programs,including
oSamuel's checkersprogram
oNewell & Simon's LogicTheorist

21. History ofAI
•1966—73: Realitydawns
oRealizationthatmanyAIproblemsareintractable
oLimitations of existing neural network methodsidentified
•Neural network research almostdisappears
•1969—85: Adding domainknowledge
oDevelopment of knowledge-basedsystems
oSuccess of rule-based expertsystems,
•E.g., DENDRAL,MYCIN
•Butwerebrittleanddidnotscalewellinpractice
•1986--Rise of machinelearning
oNeural networks return topopularity
oMajoradvancesinmachinelearningalgorithmsandapplications
•1990--Role ofuncertainty
oBayesian networks as aknowledge representation framework
•1995--AI asScience
oIntegration of learning, reasoning, knowledgerepresentation
oAImethodsusedinvision,language,datamining,etc

22. Different Types of Artificial
Intelligence
1.Modelingexactlyhowhumansactuallythink
2.Modelingexactlyhowhumansactuallyact
3.Modelinghowidealagents―shouldthink‖
4.Modelinghowidealagents―shouldact‖
•ModernAIfocusesonthelastdefinition
owewillalsofocusonthis―engineering‖approach
osuccessisjudgedbyhowwelltheagentperforms

23. The Origins ofAI
•1950AlanTuring’spaper,ComputingMachineryand
Intelligence,describedwhatisnowcalled―TheTuringTest‖.
•Turingpredictedthatinaboutfiftyyears"anaverage
interrogatorwillnothavemorethana70percentchanceof
makingthe rightidentificationafterfiveminutesof
questioning".
•1957NewellandSimonpredictedthat"Withintenyearsa
computerwillbetheworld'schesschampion."

25. The ChineseRoom
Set ofrules,in
English, for
transforming
phrases
Chinese
Writing is
given tothe
person
Correct
Responses
Shedoesnot
know
Chinese

26. The ChineseRoom
•Soimagineanindividualislockedinaroomandgivena
batchof Chinese writing.
•ThepersonlockedintheroomdoesnotunderstandChinese.
NextheisgivenmoreChinesewritingandasetofrules(in
Englishwhichheunderstands)onhowtocollatethefirstsetof
ChinesecharacterswiththesecondsetofChinesecharacters.
•SupposethepersongetssogoodatmanipulatingtheChinese
symbolsandtherulesaresogood,thattothoseoutsidethe
roomitappearsthatthepersonunderstandsChinese.
•Searle's point is that, he doesn't really understand Chinese,it
reallyonlyfollowingasetofrules.
•Followingthisargument,acomputercouldneverbetruly
intelligent,itisonlymanipulatingsymbolsthatitreallydoesn't
understandthesemanticcontext.

27. Can these Questions are Answerable?
CanComputersplayHumansatChess?
Can ComputersTalk?
Can Computers RecognizeSpeech?
CanComputersLearnandAdapt?
Can Computers“see”?
CanComputersplanandmakedecisions?

28. CanComputersplayHumansatChess?
•Chess Playing is aclassic AI problem
owell-definedproblem
overycomplex:difficultforhumanstoplaywell
•Conclusion:YES:today’scomputerscanbeateven
the besthuman
3000
2800
2600
2400
2200
2000
1800
1600
1400
1200
1966197119761981198619911997
Ratings
Garry Kasparov (current WorldChampion) DeepBlue
DeepThought
Points
Ratings

29. Can ComputersTalk?
•This is known as ―speechsynthesis‖
otranslate text to phoneticform
•e.g., ―fictitious‖ ->fik-tish-es
ousepronunciationrulestomapphonemestoactualsound
•e.g., ―tish‖ -> sequence of basic audiosounds
•Difficulties
osoundsmadebythis―lookup‖approachsoundunnatural
osounds are notindependent
•e.g., ―act‖ and―action‖
•modernsystems(e.g.,atAT&T)canhandlethisprettywell
oaharderproblemisemphasis,emotion,etc
•humansunderstandwhattheyaresaying
•machinesdon’t:sotheysoundunnatural
•Conclusion:
oNO, for completesentences
oYES, for individualwords

30. Can Computers RecognizeSpeech?
•SpeechRecognition:
omappingsoundsfromamicrophoneintoalistofwords
oclassic problemin AI, very difficult
•Recognizingsinglewordsfromasmallvocabulary
•systemscandothiswithhighaccuracy(order of99%)
•e.g., directoryinquiries
olimitedvocabulary(areacodes,citynames)
ocomputertriestorecognizeyoufirst,ifunsuccessfulhandsyou
overtoahumanoperator
osavesmillionsofdollarsayearforthephonecompanies

31. Recognizinghumanspeech
(ctd.)
•Recognizingnormalspeechismuchmoredifficult
ospeechiscontinuous:wherearetheboundariesbetweenwords?
•e.g.,―John’scarhasaflattire‖
olargevocabularies
•canbemanythousandsofpossiblewords
•wecanusecontexttohelpfigureoutwhatsomeonesaid
oe.g., hypothesize andtest
otrytellingawaiterinarestaurant:
―Iwouldlikesomesugarinmycoffee‖
obackgroundnoise,otherspeakers,accents,colds,etc
oonnormalspeech,modernsystemsareonlyabout60-70%accurate
•Conclusion:
oNO,normalspeechistoocomplextoaccuratelyrecognize
oYES, for restricted problems (small vocabulary, singlespeaker)

32. CanComputersLearnandAdapt?
•Learning andAdaptation
oconsideracomputerlearningtodriveonthefreeway
owecouldcodelotsofrulesaboutwhattodo
oand/orwecouldhaveitlearnfromexperience
omachinelearningallowscomputerstolearntodothingswithout
explicitprogramming
•Conclusion: YES, computers can learn and
adapt,whenpresentedwithinformationinthe
appropriateway

34. Can Computers“see”?
•Recognition v. Understanding (likeSpeech)
oRecognitionandUnderstandingofObjectsinascene
•look around thisroom
•you caneffortlessly recognize objects
•humanbraincanmap2dvisualimageto3d―map‖
•Why is visual recognition ahard problem?
•Conclusion:
omostlyNO:computerscanonly―see‖certaintypesofobjectsunder
limitedcircumstances
oYESforcertainconstrainedproblems(e.g.,facerecognition)

35. CanComputersplanandmake
decisions?
•Intelligence
oinvolvessolvingproblemsandmakingdecisionsandplans
oe.g.,youwanttovisityourcousininBoston
•youneedtodecideondates,flights
•youneedtogettotheairport,etc
•involvesasequenceofdecisions,plans,andactions
•What makes planninghard?
othe world is notpredictable:
•yourflightiscanceledorthere’sabackuponthe405
othereisapotentiallyhugenumberofdetails
•doyouconsiderallflights?alldates?
o no: commonsense constrains yoursolutions
oAIsystemsareonlysuccessfulinconstrainedplanningproblems
•Conclusion:NO,real-worldplanninganddecision-
making is still beyond the capabilities of modern
computers
oexception:verywell-defined,constrainedproblems:missionplanningfor
satelites.

36. SummaryofStateofAISystems
inPractice
•Speech synthesis, recognition andunderstanding
overyusefulforlimitedvocabularyapplications
ounconstrainedspeechunderstandingis stilltoohard
•Computervision
oworksforconstrainedproblems(hand-writtenzip-codes)
ounderstandingreal-world,naturalscenesis stilltoohard
•Learning
oadaptivesystemsareusedinmanyapplications:havetheirlimits
•Planning andReasoning
oonlyworksforconstrainedproblems:e.g.,chess
oreal-worldis toocomplexforgeneralsystems
•Overall:
omanycomponentsofintelligentsystemsare―doable‖
otherearemanyinterestingresearchproblemsremaining

37. Intelligent Systems in Your
EverydayLife
•PostOffice
oautomaticaddressrecognitionandsortingofmail
•Banks
oautomatic checkreaders, signature verification systems
oautomated loan applicationclassification
•TelephoneCompanies
oautomatic voicerecognition for directory inquiries
•Credit CardCompanies
oautomated frauddetection
•ComputerCompanies
oautomated diagnosis for help-deskapplications
•Netflix:
omovierecommendation
•Google:
oSearchTechnology

38. AI Applications: Identification
Technologies
•IDcards
oe.g., ATMcards
ocanbeanuisanceandsecurityrisk:
•cardscanbelost,stolen,passwordsforgotten,etc
•BiometricIdentification
owalkuptoalockeddoor
•camera
•fingerprintdevice
•microphone
•irisscan
ocomputer uses your biometric signature foridentification
•face, eyes, fingerprints, voicepattern, iris pattern

39. The agendaof AI class:
1.Fuzzy logic
2.Prepositional logic –prolog –expert systemswith
inferencealgorithms
3.Rough settheory
4.Decision trees, kNN, NaiveBayes
5.Neuralnetwork