《信息科學類專業(yè)英語》課件第21章

Lesson21IntroductiontoArtificialIntelligence

(第二十一課現(xiàn)代人工智能簡介)

Vocabulary(詞匯)ImportantSentences(重點句)Multiple-choiceQuestions(多選題)Problems(問題)

HumankindhasgivenitselfthescientificnameHomosapiens—manthewise—becauseourmentalcapacitiesaresoimportanttooureverydaylivesandoursenseofself.Thefieldofartificialintelligence,orAI,attemptstounderstandintelligententities.Thus,onereasontostudyitistolearnmoreaboutourselves.Butunlikephilosophyandpsychology,whicharealsoconcernedwithintelligence,AIstrivestobuildintelligententitiesaswellasunderstandthem.AnotherreasontostudyAIisthattheseconstructedintelligententitiesareinterestingandusefulintheirownright.AIhasproducedmanysignificantandimpressiveproductsevenatthisearlystageinitsdevelopment.Althoughnoonecanpredictthefutureindetail,itisclearthatcomputerswithhuman-levelintelligence(orbetter)wouldhaveahugeimpactonoureverydaylivesandonthefuturecourseofcivilization.[1]

AIaddressesoneoftheultimatepuzzles.Howisitpossibleforaslow,tinybrain,whetherbiologicalorelectronic,toperceive,understand,predict,andmanipulateaworldfarlargerandmorecomplicatedthanitself?Howdowegoaboutmakingsomethingwiththoseproperties?Thesearehardquestions,butunlikethesearchforfaster-than-lighttraveloranantigravitydevice,theresearcherinAIhassolidevidencethatthequestispossible.Alltheresearcherhastodoislookinthemirrortoseeanexampleofanintelligentsystem.

AIisoneofthenewestdisciplines.Itwasformallyinitiatedin1956,whenthenamewascoined,althoughatthatpointworkhadbeenunderwayforaboutfiveyears.Alongwithmoderngenetics,itisregularlycitedasthe“fieldIwouldmostliketobein”byscientistsinotherdisciplines.AstudentinphysicsmightreasonablyfeelthatallthegoodideashavealreadybeentakenbyGalileo,Newton,Einstein,andtherest,andthatittakesmanyyearsofstudybeforeonecancontributenewideas.AI,ontheotherhand,stillhasopeningsforafull-timeEinstein.

AIcurrentlyencompassesahugevarietyofsubfields,fromgeneral-purposeareassuchasperceptionandlogicalreasoning,tospecifictaskssuchasplayingchess,provingmathematicaltheorems,writingpoetry,anddiagnosingdiseases.Often,scientistsinotherfieldsmovegraduallyintoartificialintelligence,wheretheyfindthetoolsandvocabularytosystematizeandautomatetheintellectualtasksonwhichtheyhavebeenworkingalltheirlives.[2]Similarly,workersinAIcanchoosetoapplytheirmethodstoanyareaofhumanintellectualendeavor.Inthissense,itistrulyauniversalfield.1WhatisAI?

WehavenowexplainedwhyAIisexciting,butwehavenotsaidwhatitis.Definitionsofartificialintelligenceaccordingtoeightrecenttextbooksareshowninthetablebelow.Thesedefinitionsvaryalongtwomaindimensions.Theonesontopareconcernedwiththoughtprocessesandreasoning,whereastheonesonthebottomaddressbehavior.Also,thedefinitionsontheleftmeasuresuccessintermsofhumanperformance,whereastheonesontherightmeasureagainstanidealconceptofintelligence,whichwewillcallrationality.Asystemisrationalifitdoestherightthing.Table.1AIDefinitionVaryalongtwomaindimensions

Thisgivesusfourpossiblegoalstopursueinartificialintelligence:

Historically,allfourapproacheshavebeenfollowed.Asonemightexpect,atensionexistsbetweenapproachescenteredaroundhumansandapproachescenteredaroundrationality.Ahuman-centeredapproachmustbeanempiricalscience,involvinghypothesisandexperimentalconfirmation.Arationalistapproachinvolvesacombinationofmathematicsandengineering.Peopleineachgroupsometimescastaspersionsonworkdoneintheothergroups,butthetruthisthateachdirectionhasyieldedvaluableinsights.Letuslookateachinmoredetail.2ActingHumanly:theTuringTestApproach

TheTuringTest,proposedbyAlanTuring(Turing,1950),wasdesignedtoprovideasatisfactoryoperationaldefinitionofintelligence.Turingdefinedintelligentbehaviorastheabilitytoachievehuman-levelperformanceinallcognitivetasks,sufficienttofoolaninterrogator.Roughlyspeaking,thetestheproposedisthatthecomputershouldbeinterrogatedbyahumanviaateletype,andpassesthetestiftheinterrogatorcannottellifthereisacomputerorahumanattheotherend.Programmingacomputertopassthetestprovidesplentytoworkon.Thecomputerwouldneedtopossessthefollowingcapabilities:

naturallanguageprocessingtoenableittocommunicatesuccessfullyinEnglish(orsomeotherhumanlanguage);

knowledgerepresentationtostoreinformationprovidedbeforeorduringtheinterrogation;

automatedreasoningtousethestoredinformationtoanswerquestionsandtodrawnewconclusions;

machinelearningtoadapttonewcircumstancesandtodetectandextrapolatepatterns.

Turing’stestdeliberatelyavoideddirectphysicalinteractionbetweentheinterrogatorandthecomputer,becausephysicalsimulationofapersonisunnecessaryforintelligence.[3]However,theso-calledtotalTuringTestincludesavideosignalsothattheinterrogatorcantestthesubject’sperceptualabilities,aswellastheopportunityfortheinterrogatortopassphysicalobjects“throughthehatch.”TopassthetotalTuringTest,thecomputerwillneed

computervisiontoperceiveobjects,and

roboticstomovethemabout.

WithinAI,therehasnotbeenabigefforttotrytopasstheTuringtest.TheissueofactinglikeahumancomesupprimarilywhenAIprogramshavetointeractwithpeople,aswhenanexpertsystemexplainshowitcametoitsdiagnosis,oranaturallanguageprocessingsystemhasadialoguewithauser.Theseprogramsmustbehaveaccordingtocertainnormalconventionsofhumaninteractioninordertomakethemselvesunderstood.Theunderlyingrepresentationandreasoninginsuchasystemmayormaynotbebasedonahumanmodel.3ThinkingHumanly:theCognitiveModellingApproach

Ifwearegoingtosaythatagivenprogramthinkslikeahuman,wemusthavesomewayofdetermininghowhumansthink.Weneedtogetinsidetheactualworkingsofhumanminds.Therearetwowaystodothis:throughintrospection—tryingtocatchourownthoughtsastheygoby—orthroughpsychologicalexperiments.Oncewehaveasufficientlyprecisetheoryofthemind,itbecomespossibletoexpressthetheoryasacomputerprogram.Iftheprogram’sinput/outputandtimingbehaviormatcheshumanbehavior,thatisevidencethatsomeoftheprogram’smechanismsmayalsobeoperatinginhumans.Forexample,NewellandSimon,whodevelopedGPS,the“GeneralProblemSolver”(NewellandSimon,1961),werenotcontenttohavetheirprogramcorrectlysolveproblems.Theyweremoreconcernedwithcomparingthetraceofitsreasoningstepstotracesofhumansubjectssolvingthesameproblems.Thisisincontrasttootherresearchersofthesametime(suchasWang(1960)),whowereconcernedwithgettingtherightanswersregardlessofhowhumansmightdoit.TheinterdisciplinaryfieldofcognitivesciencebringstogethercomputermodelsfromAIandexperimentaltechniquesfrompsychologytotrytoconstructpreciseandtestabletheoriesoftheworkingsofthehumanmind.[4]4Thinkingrationally:Thelawsofthoughtapproach

TheGreekphilosopherAristotlewasoneofthefirsttoattempttocodify“rightthinking,”thatis,irrefutablereasoningprocesses.Hisfamoussyllogismsprovidedpatternsforargumentstructuresthatalwaysgavecorrectconclusionsgivencorrectpremises.Forexample,“Socratesisaman;allmenaremortal;thereforeSocratesismortal.”Theselawsofthoughtweresupposedtogoverntheoperationofthemind,andinitiatedthefieldoflogic.

Thedevelopmentofformallogicinthelatenineteenthandearlytwentiethcenturies,providedaprecisenotationforstatementsaboutallkindsofthingsintheworldandtherelationsbetweenthem.(Contrastthiswithordinaryarithmeticnotation,whichprovidesmainlyforequalityandinequalitystatementsaboutnumbers.)By1965,programsexistedthatcould,givenenoughtimeandmemory,takeadescriptionofaprobleminlogicalnotationandfindthesolutiontotheproblem,ifoneexists.(Ifthereisnosolution,theprogrammightneverstoplookingforit.)Theso-calledlogicisttraditionwithinartificialintelligencehopestobuildonsuchprogramstocreateintelligentsystems.

Therearetwomainobstaclestothisapproach.First,itisnoteasytotakeinformalknowledgeandstateitintheformaltermsrequiredbylogicalnotation,particularlywhentheknowledgeislessthan100%certain.Second,thereisabigdifferencebetweenbeingabletosolveaproblem“inprinciple”anddoingsoinpractice.Evenproblemswithjustafewdozenfactscanexhaustthecomputationalresourcesofanycomputerunlessithassomeguidanceastowhichreasoningstepstotryfirst.[5]Althoughbothoftheseobstaclesapplytoanyattempttobuildcomputationalreasoningsystems,theyappearedfirstinthelogicisttraditionbecausethepoweroftherepresentationandreasoningsystemsarewell-definedandfairlywellunderstood.5ActingRationally:theRationalAgentApproach

Actingrationallymeansactingsoastoachieveone’sgoals,givenone’sbeliefs.Anagentisjustsomethingthatperceivesandacts.(Thismaybeanunusualuseoftheword,butyouwillgetusedtoit.)Inthisapproach,AIisviewedasthestudyandconstructionofrationalagents.

Inthe“l(fā)awsofthought”approachtoAI,thewholeemphasiswasoncorrectinferences.Makingcorrectinferencesissometimespartofbeingarationalagent,becauseonewaytoactrationallyistoreasonlogicallytotheconclusionthatagivenactionwillachieveone’sgoals,andthentoactonthatconclusion.Ontheotherhand,correctinferenceisnotallofrationality,becausethereareoftensituationswherethereisnoprovablycorrectthingtodo,yetsomethingmuststillbedone.Therearealsowaysofactingrationallythatcannotbereasonablysaidtoinvolveinference.Forexample,pullingone’shandoffofahotstoveisareflexactionthatismoresuccessfulthanasloweractiontakenaftercarefuldeliberation.

Allthe“cognitiveskills”neededfortheTuringTestaretheretoallowrationalactions.Thus,weneedtheabilitytorepresentknowledgeandreasonwithitbecausethisenablesustoreachgooddecisionsinawidevarietyofsituations.Weneedtobeabletogeneratecomprehensiblesentencesinnaturallanguagebecausesayingthosesentenceshelpsusgetbyinacomplexsociety.Weneedlearningnotjustforerudition,butbecausehavingabetterideaofhowtheworldworksenablesustogeneratemoreeffectivestrategiesfordealingwithit.Weneedvisualperceptionnotjustbecauseseeingisfun,butinordertogetabetterideaofwhatanactionmightachieve—forexample,beingabletoseeatastymorselhelpsonetomovetowardit.

ThestudyofAIasrationalagentdesignthereforehastwoadvantages.First,itismoregeneralthanthe“l(fā)awsofthought”approach,becausecorrectinferenceisonlyausefulmechanismforachievingrationality,andnotanecessaryone.Second,itismoreamenabletoscientificdevelopmentthanapproachesbasedonhumanbehaviororhumanthought,becausethestandardofrationalityisclearlydefinedandcompletelygeneral.Humanbehavior,ontheotherhand,iswell-adaptedforonespecificenvironmentandistheproduct,inpart,ofacomplicatedandlargelyunknownevolutionaryprocessthatstillmaybefarfromachievingperfection.6TheStateoftheArt

InternationalgrandmasterArnoldDenkerstudiesthepiecesontheboardinfrontofhim.Herealizesthereisnohope;hemustresignthegame.Hisopponent,Hitech,becomesthefirstcomputerprogramtodefeatagrandmasterinagameofchess.

“IwanttogofromBostontoSanFrancisco,”thetravellersaysintothemicrophone.“Whatdatewillyoubetravellingon?”isthereply.ThetravellerexplainsshewantstogoOctober20th,nonstop,onthecheapestavailablefare,returningonSunday.AspeechunderstandingprogramnamedPegasushandlesthewholetransaction,whichresultsinaconfirmedreservationthatsavesthetraveller$894overtheregularcoachfare.Eventhoughthespeechrecognizergetsoneoutoftenwordswrong,itisabletorecoverfromtheseerrorsbecauseofitsunderstandingofhowdialogsareputtogether.

AnanalystintheMissionOperationsroomoftheJetPropulsionLaboratorysuddenlystartspayingattention.Aredmessagehasflashedontothescreenindicatingan“anomaly”withtheVoyagerspacecraft,whichissomewhereinthevicinityofNeptune.Fortunately,theanalystisabletocorrecttheproblemfromtheground.OperationspersonnelbelievetheproblemmighthavebeenoverlookedhaditnotbeenforMarvel,areal-timeexpertsystemthatmonitorsthemassivestreamofdatatransmittedbythespacecraft,handlingroutinetasksandalertingtheanalyststomoreseriousproblems.

CruisingthehighwayoutsideofPittsburghatacomfortable55mph,themaninthedriver’sseatseemsrelaxed.Heshouldbe—forthepast90miles,hehasnothadtotouchthesteeringwheel.Therealdriverisaroboticsystemthatgathersinputfromvideocameras,sonar,andlaserrangefindersattachedtothevan.Itcombinestheseinputswithexperiencelearnedfromtrainingrunsandsuccessfullycomputeshowtosteerthevehicle.

Aleadingexpertonlymph-nodepathologydescribesafiendishlydifficultcasetotheexpertsystem,andexaminesthesystem’sdiagnosis.Hescoffsatthesystem’sresponse.Onlyslightlyworried,thecreatorsofthesystemsuggestheaskthecomputerforanexplanationofthediagnosis.Themachinepointsoutthemajorfactorsinfluencingitsdecision,andexplainsthesubtleinteractionofseveralofthesymptomsinthiscase.Theexpertadmitshiserror,eventually.

Fromacameraperchedonastreetlightabovethecrossroads,thetrafficmonitorwatchesthescene.Ifanyhumanswereawaketoreadthemainscreen,theywouldsee“Citroen2CVturningfromPlacedelaConcordeintoChampsElysees,”“LargetruckofunknownmakestoppedonPlacedelaConcorde,”andsoonintothenight.Andoccasionally,“MajorincidentonPlacedelaConcorde,speedingvancollidedwithmotorcyclist,”andanautomaticcalltotheemergencyservices.

Thesearejustafewexamplesofartificialintelligencesystemsthatexisttoday.Notmagicorsciencefiction—butratherscience,engineering,andmathematics.1.?Homosapiensn.智人(現(xiàn)代人的學名)

2.?antigrarityn.反重力,反引力。

3.?endeavorn.努力,盡力vi.盡力,努力。

4.?dimensionn.尺寸,尺度,維(數(shù)),度(數(shù)),元。

5.?rationalityn.合理性,唯理性。

6.?hypothesisn.假設(shè)。Vocabulary

7.?aspersionn.灑水,誹謗,中傷。

8.?interrogatorn.訊問者,質(zhì)問者。

9.?extrapolatev.推斷,[數(shù)]外推。

10.?cognitiveadj.認知的,認識的,有感知的。

11.?syllogismn.[邏]三段論法,推論法,演繹。

12.?mortaln.凡人,人類adj.必死的,致命的,人類的,臨終的。

13.?agentn.代理。

14.?inferencen.推論。

15.?stateoftheartn.技術(shù)發(fā)展水平。16.?Neptunen.[天]天王星。

17.?lymphn.淋巴腺，淋巴。

18.?pathologyn.病理學。

19.?fiendishlyadv.惡魔似地,極壞地。

20.?eruditionn.博學。

[1]Althoughnoonecanpredictthefutureindetail,itisclearthatcomputerswithhuman-levelintelligence(orbetter)wouldhaveahugeimpactonoureverydaylivesandonthefuturecourseofcivilization.

雖然沒有人可以詳細地預測未來，但是很顯然，具有人類智力水平(或更高水平)的電腦將會對我們的日常生活以及未來的文明進程產(chǎn)生巨大的影響。主句中it為形式主語，真正的主語是that引導的定語從句。ImportantSentences

[2]Often,scientistsinotherfieldsmovegraduallyintoartificialintelligence,wheretheyfindthetoolsandvocabularytosystematizeandautomatetheintellectualtasksonwhichtheyhavebeenworkingalltheirlives.

通常，其他領(lǐng)域的科學家逐步進入到了人工智能領(lǐng)域，他們在那里發(fā)現(xiàn)了能夠?qū)⑺麄円恢彼鶑氖碌墓ぷ飨到y(tǒng)化和自動化的工具和詞匯。where引導定語從句，修飾“artificialintelligence”。

[3]Turing’stestdeliberatelyavoideddirectphysicalinteractionbetweentheinterrogatorandthecomputer,becausephysicalsimulationofapersonisunnecessaryforintelligence.

圖靈測試刻意回避詢問者和計算機之間直接的物理交互，因為人的物理模擬對智能來說是不必要的。

[4]TheinterdisciplinaryfieldofcognitivesciencebringstogethercomputermodelsfromAIandexperimentaltechniquesfrompsychologytotrytoconstructpreciseandtestabletheoriesoftheworkingsofthehumanmind.

認知科學這個跨學科領(lǐng)域匯集了人工智能學的計算機模型以及心理學的實驗技巧，試圖構(gòu)建人類頭腦運轉(zhuǎn)的準確的、可檢驗的理論。本句為一簡單句，結(jié)構(gòu)為Theinterdisciplinaryfield…brings…to….。

[5]Evenproblemswithjustafewdozenfactscanexhaustthecomputationalresourcesofanycomputerunlessithassomeguidanceastowhichreasoningstepstotryfirst.

除非有應該首先執(zhí)行哪個推理步驟的提示，否則即使只有幾十個論據(jù)的問題也能耗盡任何一臺計算機的計算資源。

(1)?OnereasontostudyAIistolearnmoreaboutourselves,itisbecausethat().

A.?AIattemptstounderstandintelligententities

B.?AIattemptstobuildintelligententities

C.?AIisanintelligententities

D.?weareintelligententities

Multiple-choiceQuestions

(2)?Inthethirdparagraph,“AI,ontheotherhand,stillhasopeningsforafull-timeEinstein.”,whatisthemeaning?()

A.?InAI,there’remanynewideasforonetocontributeandmoreeasilytostudy.

B.?AIisnotoneofthenewestdisciplines.

C.?AllthegoodideashavealreadybeentakenbyGalileo,Newton,Einstein,andtherest.

D.?AIwasinitiatedformanyyears.

(3)?WhichistheTuringTest?

A.?Thecomputerandahumanshouldinterrogateeachother,andthec