Artificialintelligence.docx
- 文档编号:14919175
- 上传时间:2023-06-28
- 格式:DOCX
- 页数:17
- 大小:196.24KB
Artificialintelligence.docx
《Artificialintelligence.docx》由会员分享,可在线阅读,更多相关《Artificialintelligence.docx(17页珍藏版)》请在冰点文库上搜索。
Artificialintelligence
Artificialintelligence (AI)isthe intelligence ofmachinesandthebranchof computerscience thataimstocreateit.
Textbooksdefinethefieldas"thestudyanddesignof intelligentagents," whereanintelligentagentisasystemthatperceivesitsenvironmentandtakesactionsthatmaximizeitschancesofsuccess
∙2 Problems
o2.1 Deduction,reasoning,problemsolving
o2.2 Knowledgerepresentation
o2.3 Planning
o2.4 Learning
o2.5 Naturallanguageprocessing
o2.6 Motionandmanipulation
o2.7 Perception
o2.8 Socialintelligence
o2.9 Creativity
o2.10 Generalintelligence
∙3 Approaches
o3.1 Cyberneticsandbrainsimulation
o3.2 Symbolic
o3.3 Sub-symbolic
o3.4 Statistical
o3.5 Integratingtheapproaches
∙4 Tools
o4.1 Searchandoptimization
o4.2 Logic
o4.3 Probabilisticmethodsforuncertainreasoning
o4.4 Classifiersandstatisticallearningmethods
o4.5 Neuralnetworks
o4.6 Controltheory
o4.7 Languages
∙5 Evaluatingprogress
∙6 Applications
o6.1 Competitionsandprizes
o6.2 Platforms
∙7 Philosophy
∙8 Prediction
Deduction,reasoning,problemsolving
EarlyAIresearchersdevelopedalgorithmsthatimitatedthestep-by-stepreasoningthosehumanswereoftenassumedtousewhentheysolvepuzzlesplayboardgamesormakelogicaldeductions.
Bythelate1980sand'90s,AIresearchhadalsodevelopedhighlysuccessfulmethodsfordealingwith uncertain orincompleteinformation,employingconceptsfrom probability and economics.
Fordifficultproblems,mostofthesealgorithmscanrequireenormouscomputationalresources—mostexperiencea"combinatorialexplosion":
theamountofmemoryorcomputertimerequiredbecomesastronomicalwhentheproblemgoesbeyondacertainsize.
ThesearchformoreefficientproblemsolvingalgorithmsisahighpriorityforAIresearch.
Humanbeingssolvemostoftheirproblemsusingfast,intuitivejudgmentsratherthantheconscious,step-by-stepdeductionthatearlyAIresearchwasabletomodel.
AIhasmadesomeprogressatimitatingthiskindof"sub-symbolic"problemsolving:
Embodiedagent approachesemphasizetheimportanceofsensorimotor skillstohigherreasoning;
Neuralnet researchattemptstosimulatethestructuresinsidehumanandanimalbrainsthatgiverisetothisskill.
Defaultreasoning andthe qualificationproblem
Manyofthethingspeopleknowtaketheformof"workingassumptions."Forexample,ifabirdcomesupinconversation,peopletypicallypictureananimalthatisfistsized,sings,andflies.
Noneofthesethingsaretrueaboutallbirds. JohnMcCarthy identifiedthisproblemin1969 asthequalificationproblem:
foranycommonsenserulethatAIresearcherscaretorepresent,theretendtobeahugenumberofexceptions.Almostnothingissimplytrueorfalseinthewaythatabstractlogicrequires.AIresearchhasexploredanumberofsolutionstothisproblem.[51]
Thebreadthof commonsenseknowledge
Thenumberofatomicfactsthattheaveragepersonknowsisastronomical.
Researchprojectsthatattempttobuildacompleteknowledgebaseof commonsenseknowledge (e.g., Cyc)requireenormousamountsoflaborious ontologicalengineering —theymustbebuilt,byhand,onecomplicatedconceptatatime.[52] Amajorgoalistohavethecomputerunderstandenoughconceptstobeabletolearnbyreadingfromsourcesliketheinternet,andthusbeabletoaddtoitsownontology.
Thesubsymbolicformofsome commonsenseknowledge
Muchofwhatpeopleknowisnotrepresentedas"facts"or"statements"thattheycouldactuallysayoutloud.Forexample,achessmasterwillavoidaparticularchesspositionbecauseit"feelstooexposed" oranartcriticcantakeonelookatastatueandinstantlyrealizethatitisafake.[54] Theseareintuitionsortendenciesthatarerepresentedinthebrainnon-consciouslyandsub-symbolically.[55] Knowledgelikethisinforms,supportsandprovidesacontextforsymbolic,consciousknowledge.Aswiththerelatedproblemofsub-symbolicreasoning,itishopedthat situatedAI or computationalintelligence willprovidewaystorepresentthiskindofknowledge.]
Planning
Mainarticle:
Automatedplanningandscheduling
Learning
Mainarticle:
Machinelearning
Naturallanguageprocessing
ASIMO usessensorsandintelligentalgorithmstoavoidobstaclesandnavigatestairs.
Mainarticle:
Naturallanguageprocessing
Naturallanguageprocessing[64] givesmachinestheabilitytoreadandunderstandthelanguagesthathumansspeak.Manyresearchershopethatasufficientlypowerfulnaturallanguageprocessingsystemwouldbeabletoacquireknowledgeonitsown,byreadingtheexistingtextavailableovertheinternet.Somestraightforwardapplicationsofnaturallanguageprocessinginclude informationretrieval (or textmining)and machinetranslation.[65]
Motionandmanipulation
Mainarticle:
Robotics
Thefieldof robotics[66] iscloselyrelatedtoAI.Intelligenceisrequiredforrobotstobeabletohandlesuchtasksasobjectmanipulation[67] and navigation,withsub-problemsof localization (knowingwhereyouare), mapping (learningwhatisaroundyou)and motionplanning (figuringouthowtogetthere).[68]
Perception
Mainarticles:
Machineperception, Computervision,and Speechrecognition
Machineperception[69] istheabilitytouseinputfromsensors(suchascameras,microphones,sonarandothersmoreexotic)todeduceaspectsoftheworld.Computervision[70] istheabilitytoanalyzevisualinput.Afewselectedsubproblemsare speechrecognition,[71] facialrecognition and objectrecognition.[72]
Socialintelligence
Mainarticle:
Affectivecomputing
Kismet,arobotwithrudimentarysocialskills
Emotionandsocialskills[73] playtworolesforanintelligentagent.First,itmustbeabletopredicttheactionsofothers,byunderstandingtheirmotivesandemotionalstates.(Thisinvolveselementsof gametheory, decisiontheory,aswellastheabilitytomodelhumanemotionsandtheperceptualskillstodetectemotions.)Also,forgood human-computerinteraction,anintelligentmachinealsoneedsto display emotions.Attheveryleastitmustappearpoliteandsensitivetothehumansitinteractswith.Atbest,itshouldhavenormalemotionsitself.
Creativity
Mainarticle:
Computationalcreativity
TOPIO,arobotthatcanplay tabletennis,developedby TOSY.
Asub-fieldofAIaddresses creativity boththeoretically(fromaphilosophicalandpsychologicalperspective)andpractically(viaspecificimplementationsofsystemsthatgenerateoutputsthatcanbeconsideredcreative).Arelatedareaofcomputationalresearchis ArtificialIntuition and ArtificialImagination.
Generalintelligence
Mainarticles:
StrongAI and AI-complete
Mostresearchershopethattheirworkwilleventuallybeincorporatedintoamachinewith general intelligence(knownas strongAI),combiningalltheskillsaboveandexceedinghumanabilitiesatmostorallofthem.[12] Afewbelievethat anthropomorphicfeatureslike artificialconsciousness oran artificialbrain mayberequiredforsuchaproject.[74]
Manyoftheproblemsaboveareconsidered AI-complete:
tosolveoneproblem,youmustsolvethemall.Forexample,evenastraightforward,specifictasklike machinetranslation requiresthatthemachinefollowtheauthor'sargument(reason),knowwhatisbeingtalkedabout(knowledge),andfaithfullyreproducetheauthor'sintention(socialintelligence). Machinetranslation,therefore,isbelievedtobeAI-complete:
itmayrequire strongAI tobedoneaswellashumanscandoit.[75]
Approaches
Thereisnoestablishedunifyingtheoryor paradigm thatguidesAIresearch.Researchersdisagreeaboutmanyissues.[76] Afewofthemostlongstandingquestionsthathaveremainedunansweredarethese:
shouldartificialintelligencesimulatenaturalintelligence,bystudying psychology or neurology?
OrishumanbiologyasirrelevanttoAIresearchasbirdbiologyisto aeronauticalengineering?
[77] Canintelligentbehaviorbedescribedusingsimple,elegantprinciples(suchas logic oroptimization)?
Ordoesitnecessarilyrequiresolvingalargenumberofcompletelyunrelatedproblems?
[78] Canintelligencebereproducedusinghigh-levelsymbols,similartowordsandideas?
Ordoesitrequire"sub-symbolic"processing?
[79]
Cyberneticsandbrainsimulation
Mainarticles:
Cybernetics and Computationalneuroscience
Thereisnoconsensusonhowcloselythebrainshouldbe simulated.
Inthe1940sand1950s,anumberofresearchersexploredtheconnectionbetween neurology, informationtheory,and cybernetics.Someofthembuiltmachinesthatusedelectronicnetworkstoexhibitrudimentaryintelligence,suchas W.GreyWalter's turtles andthe JohnsHopkinsBeast.ManyoftheseresearchersgatheredformeetingsoftheTeleologicalSocietyat PrincetonUniversity andthe RatioClub inEngland.[24] By1960,thisapproachwaslargelyabandoned,althoughelementsofitwouldberevivedinthe1980s.
Symbolic
Mainarticle:
Goodoldfashionedartificialintelligence
Whenaccesstodigitalcomputersbecamepossibleinthemiddle1950s,AIresearchbegantoexplorethepossibilitythathumanintelligencecouldbereducedtosymbolmanipulation.Theresearchwascenteredinthreeinstitutions:
CMU, Stanford and MIT,andeachonedevelopeditsownstyleofresearch. JohnHaugeland namedtheseapproachestoAI"goodoldfashionedAI"or"GOFAI".[80]
Logicbased
Unlike Newell and Simon, JohnMcCarthy feltthatmachinesdidnotneedtosimulatehumanthought,butshouldinsteadtrytofindtheessenceofabstractreasoningandproblemsolving,regardlessofwhetherpeopleusedthesamealgorithms.[77] Hislaboratoryat Stanford (SAIL)focusedonusingformal logic tosolveawidevarietyofproblems,including knowledgere
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- Artificialintelligence