ARTIFICIAL INTELLIGENCE-人工智能(英文)

1、ARTIFICIAL INTELLIGENCE人工智能Artificialintelligence(AI)is,intheory,theabilityofanartificialmechanismto demonstratesomeformofintelligentbehaviorequivalenttothebehaviorsobservedin intelligentlivingorganisms.Artificialintelligenceisalsothenameofthefieldof science and technology in which artificial mechan

2、isms that exhibit behavior resemblingintelligencearedevelopedandstudied.ThetermAIitself,andthephenomenaactuallyobserved,invite-indeed demand-philosophicalspeculationaboutwhatinfactconstitutesthemindintelligence.Thesekindsofquestionscanbeconsideredseparately,however,froma descriptionofthevariousendea

3、vorstoconstructincreasinglysophisticated mechanisms that exhibit “intelligence.”ResearchintoallaspectsofAIisvigorous.Someconcernexistsamongworkersinthefield,however,thatboththeprogressandexpectationsofAIhavebeen overstated. AI programsareprimitivewhencomparedtothekindsof intuitive reasoningandinduct

4、ionofwhichthehumanbrainoreventhebrainsofmuchless advancedorganismsarecapable.AIhasindeedshowngreatpromiseintheareaof expertsystems-thatis,knowledge-basedexpertprograms-butwhileprogramsarepowerfulwhenansweringquestionswithinaspecificdomain,theyare neverthelessincapableofanytypeofadaptable,ortrulyinte

5、lligent,reasoning.ExamplesofAIsystemsincludecomputerprogramsthatperformsuchtasksas medicaldiagnosesandmineralprospecting.Computershavealsobeenprogrammed todisplaysomedegreeoflegalreasoning,speechunderstanding,visioninterpretation, natural-languageprocessing,problemsolving,andlearning.Althoughmostoft

6、hese systemshaveprovedvaluableeitherasresearchvehiclesorinspecific,practical applications,mostofthemarealsostillveryfarfrombeingperfected.CHARACTERISTICSOFAI:Nogenerallyacceptedtheorieshaveyetemerged withinthefieldofAI,owinginparttothefactthatAIisaveryyoungscience.Itis assumed,however,thatonthehighe

7、stlevelanAIsystemmustreceiveinputfromits environment,determineanactionorresponse,anddeliveranoutputtoits environment.Amechanismforinterpretingtheinputisneeded.Thisneedhasledto research in speech understanding, vision, and natural language. Theinterpretation mustberepresentedinsomeformthatcanbemanipu

8、latedbythemachine.Inordertoachievethisgoal,techniquesofknowledgerepresentationareinvoked. TheAIinterpretationofthis,togetherwithknowledgeobtainedpreviously,manipulatedmanipulatedwithinthesystemunderstudybymeansofsomemechanismalgorithm.Thesystemthusarrivesataninternalrepresentationoftheresponseor act

9、ion.Thedevelopmentofsuchprocessesrequirestechniquesofexpertreasoning, common-sensereasoning,problemsolving,planning,signalinterpretation,learning. Finally, the system must網(wǎng)7THEFIFTH-GENERATIONATTEMPT:In7THEFIFTH-GENERATIONATTEMPT:Inthe1980s,inanattempttodevelopananexpertsystemonaverylargescale,theJa

10、panesegovernmentbeganbuildingpowerfulpowerfulcomputerswithhardwarethatmadelogicalinferencesinthecomputerlanguagelanguage(Followingtheideaofrepresentingknowledgedeclaratively,thelogiclogicprogrammingPROLOGhadbeendevelopedinEnglandandFrance.PROLOGisactuallyaninferenceenginethatsearchesdeclaredfactsand

11、rulestoconfirmordenydenyahypothesis.AdrawbackofPROLOGisthatitcannotbealteredbytheprogrammer.)TheJapanesereferredtosuchmachinesas“fifthprogrammer.)TheJapanesereferredtosuchmachinesas“fifth-generation”computers.Bytheearly1990s,however,Japanhadforsakenthisplanandevenannounced thattheywerereadytorelease

12、itssoftware.Althoughtheydidnotdetailreasonsfor theirabandonmentofthefifth-generationprogram,U.Sscientistsfaultedtheirefforts atAIasbeingtoomuchinthedirectionofcomputer-typelogicandtoolittleinthe directionofhumanthinkingprocesses.ThechoiceofPROLOGwasalsocriticized.Other nations were by then not devel

13、oping software in that computer language and wereshowinglittlefurtherenthusiasmforit.Furthermore,theJapanesewerenot makingmuchprogressinparallelprocessing,akindofcomputerinvolvingmanyindependentprocessorsworkingtogetherinparallelaincreasinglyimportantinthefieldofcomputerscience.TheJapanesehavenow de

14、fined a “sixth -generation” goal instead, called the Real World Computing Project, thatveersawayfromtheexpert-systemsapproachthatworksonlybybuilt-inlogical rules.THEFUTUREOFAIRESEARCH:Oneimpedimenttobuildingevenmoreusefulexpertsystemshasbeen,fromthestart,theproblemofinput-inparticular,the feedingofr

15、awdataintoanAIsystem.Tothisend,muchefforthasbeendevotedto speechrecognition,characterrecognition,machinevision,andnatural-language processing.Asecondproblemisinobtainingknowledge.Ithasprovedarduousextractextractknowledgefromanexpertandthencodeitforusebythemachine,soagreat dealofeffortisalsobeingdevo

16、tedtolearningandknowledgeacquisition.1010OneofthemostusefulideasthathasemergedfromAIresearch,however,isthatfactsandrules(declarativeknowledge)canberepresentedseparatelyfromdecision-makingdecision-makingalgorithms(proceduralknowledge).Thisrealizationhashadaprofoundprofoundeffectbothonthewaythatscient

17、istsapproachproblemsandontheengineeringengineering techniquesusedtoproduceAIsystems.Byadoptingaparticularproceduralelement,calledaninferenceengine,developmentofanAIsystemisreducedreducedtoobtainingandcodifyingsufficientrulesandfactsfromtheproblemdomain.domain.Thiscodificationprocessiscalledknowledge

18、engineering.Reducingsystemsystemdevelopmenttoknowledgeengineeringhasopenedthedoortonon-AIpractitioners. In apractitioners. In a11Inparticular,alargenumberoftheseproblemsintheAIfield11Inparticular,alargenumberoftheseproblemsintheAIfieldhavebeenassociatedassociatedwithrobotics.Thereare,firstofall,them

19、echanicalproblemsofgettingamachinemachinetomakeverypreciseordelicatemovements.Beyondthatarethemuchmoredifficultproblemsofprogrammingsequencesofmovementsthatwillenablearobottotointeracteffectivelywithanaturalenvironment,ratherthansomecarefullydesignedlaboratorylaboratorysetting.Muchworkinthisareainvo

20、lvesproblemsolvingandplanning.Aradicalapproachtosuchproblemshasbeentoabandontheaimof developing “reasoning” AI systems and to produce, instead, robots that function “reflexively”. A leading figure in this field has been Rodney Brooks of the MassachusettsInstituteofTechnology.TheseAIresearchersfeltth

21、atprecedingefforts inroboticsweredoomedtofailurebecausethesystemsproducedcouldnotfunctionintherealworld.Ratherthantryingtoconstructintegratednetworksthatoperate underacentralizingcontrolandmaintainalogicallyconsistentmodeloftheworld, theyarepursuingabehavior-basedapproachnamedsubsumptionarchitecture.Subsumption architecture employs a design technique called “l(fā)ayering,”-a formofparallelprocessinginwhicheachlayerisaseparatebehavior-producing networkthatfunc