搜索
Turing,A.M.(1950).Computingmachineryandintelligence.Mind,59,433-460.COMPUTINGMACHINERYANDINTELLIGENCEByA.M.Turing1.TheImitationGameIproposetoconsiderthequestion,Canmachinesthink?Thisshouldbeginwithdefinitionsofthemeaningofthetermsmachineandthink.Thedefinitionsmightbeframedsoastoreflectsofaraspossiblethenormaluseofthewords,butthisattitudeisdangerous,Ifthemeaningofthewordsmachineandthinkaretobefoundbyexamininghowtheyarecommonlyuseditisdifficulttoescapetheconclusionthatthemeaningandtheanswertothequestion,Canmachinesthink?istobesoughtinastatisticalsurveysuchasaGalluppoll.Butthisisabsurd.InsteadofattemptingsuchadefinitionIshallreplacethequestionbyanother,whichiscloselyrelatedtoitandisexpressedinrelativelyunambiguouswords.Thenewformoftheproblemcanbedescribedintermsofagamewhichwecallthe'imitationgame.Itisplayedwiththreepeople,aman(A),awoman(B),andaninterrogator(C)whomaybeofeithersex.Theinterrogatorstaysinaroomapartfronttheothertwo.Theobjectofthegamefortheinterrogatoristodeterminewhichoftheothertwoisthemanandwhichisthewoman.HeknowsthembylabelsXandY,andattheendofthegamehesayseitherXisAandYisBorXisBandYisA.TheinterrogatorisallowedtoputquestionstoAandBthus:C:WillXpleasetellmethelengthofhisorherhair?NowsupposeXisactuallyA,thenAmustanswer.ItisA'sobjectinthegametotryandcauseCtomakethewrongidentification.Hisanswermightthereforebe:Myhairisshingled,andthelongeststrandsareaboutnineincheslong.Inorderthattonesofvoicemaynothelptheinterrogatortheanswersshouldbewritten,orbetterstill,typewritten.Theidealarrangementistohaveateleprintercommunicatingbetweenthetworooms.Alternativelythequestionandanswerscanberepeatedbyanintermediary.Theobjectofthegameforthethirdplayer(B)istohelptheinterrogator.Thebeststrategyforherisprobablytogivetruthfulanswers.ShecanaddsuchthingsasIamthewoman,don'tlistentohim!toheranswers,butitwillavailnothingasthemancanmakesimilarremarks.Wenowaskthequestion,WhatwillhappenwhenamachinetakesthepartofAinthisgame?Willtheinterrogatordecidewronglyasoftenwhenthegameisplayedlikethisashedoeswhenthegameisplayedbetweenamanandawoman?Thesequestionsreplaceouroriginal,Canmachinesthink?2.CritiqueoftheNewProblemAswellasasking,Whatistheanswertothisnewformofthequestion,onemayask,Isthisnewquestionaworthyonetoinvestigate?Thislatterquestionweinvestigatewithoutfurtherado,therebycuttingshortaninfiniteregress.Thenewproblemhastheadvantageofdrawingafairlysharplinebetweenthephysicalandtheintellectualcapacitiesofaman.Noengineerorchemistclaimstobeabletoproduceamaterialwhichisindistinguishablefromthehumanskin.Itispossiblethatatsometimethismightbedone,butevensupposingthisinventionavailableweshouldfeeltherewaslittlepointintryingtomakeathinkingmachinemorehumanbydressingitupinsuchartificialflesh.Theforminwhichwehavesettheproblemreflectsthisfactintheconditionwhichpreventstheinterrogatorfromseeingortouchingtheothercompetitors,orhearing-theirvoices.Someotheradvantagesoftheproposedcriterionmaybeshownupbyspecimenquestionsandanswers.Thus:Q:PleasewritemeasonnetonthesubjectoftheForthBridge.A:Countmeoutonthisone.Inevercouldwritepoetry.Q:Add34957to70764.A:(Pauseabout30secondsandthengiveasanswer)105621.Q:Doyouplaychess?A:Yes.Q:IhaveKatmyK1,andnootherpieces.YouhaveonlyKatK6andRatR1.Itisyourmove.Whatdoyouplay?A:(Afterapauseof15seconds)R-R8mate.Thequestionandanswermethodseemstobesuitableforintroducingalmostanyoneofthefieldsofhumanendeavourthatwewishtoinclude.Wedonotwishtopenalisethemachineforitsinabilitytoshineinbeautycompetitions,nortopenaliseamanforlosinginaraceagainstanaeroplane.Theconditionsofourgamemakethesedisabilitiesirrelevant.Thewitnessescanbrag,iftheyconsideritadvisable,asmuchastheypleaseabouttheircharms,strengthorheroism,buttheinterrogatorcannotdemandpracticaldemonstrations.Thegamemayperhapsbecriticisedonthegroundthattheoddsareweightedtooheavilyagainstthemachine.Ifthemanweretotryandpretendtobethemachinehewouldclearlymakeaverypoorshowing.Hewouldbegivenawayatoncebyslownessandinaccuracyinarithmetic.Maynotmachinescarryoutsomethingwhichoughttobedescribedasthinkingbutwhichisverydifferentfromwhatamandoes?Thisobjectionisaverystrongone,butatleastwecansaythatif,nevertheless,amachinecanbeconstructedtoplaytheimitationgamesatisfactorily,weneednotbetroubledbythisobjection.Itmightbeurgedthatwhenplayingtheimitationgamethebeststrategyforthemachinemaypossiblybesomethingotherthanimitationofthebehaviourofaman.Thismaybe,butIthinkitisunlikelythatthereisanygreateffectofthiskind.Inanycasethereisnointentiontoinvestigateherethetheoryofthegame,anditwillbeassumedthatthebeststrategyistotrytoprovideanswersthatwouldnaturallybegivenbyaman.3.TheMachinesConcernedintheGameThequestionwhichweputin1willnotbequitedefiniteuntilwehavespecifiedwhatwemeanbythewordmachine.Itisnaturalthatweshouldwishtopermiteverykindofengineeringtechniquetobeusedinourmachines.Wealsowishtoallowthepossibilitythananengineerorteamofengineersmayconstructamachinewhichworks,butwhosemannerofoperationcannotbesatisfactorilydescribedbyitsconstructorsbecausetheyhaveappliedamethodwhichislargelyexperimental.Finally,wewishtoexclu