The Truth about AI 3of3《人工智能的真相（2023）（3）》完整中英文对照剧本.docx

我已有三十年驾龄I'vebeenadriverfor30years,但从未见过像这样的车butveneverbeeninacarlikethisonebefore.这是辆自动驾驶汽车Thisisanautonomousvehicle正自己行驶在牛津的街道上andit'sdrivingitselfthroughOxford.已开启自动驾驶在千禧年前后现实生活可用的无人驾驶汽车Attheturnofthecentury,practicaldriverlesscars似乎还是一场非常非常遥远的梦feltliketheywereavery,verydistantdream.但人工智能和计算机技术的进步ButadvancesinAIandcomputing让无人驾驶汽车指日可待meanthattheycouldbejustaroundthecorner.我们之所以憧憬无人驾驶汽车是因为它们Andthedreamofdriverlesscarsisthattheywillbefarsafer比人类司机更安全更高效andmoreefficientthanhumandriverscouldeverbe.但有人担心Butsomepeopleworry我们可能把太多控制权thatWemightendupgivingtoomuchcontrol交给人工智能toartificialintelligence.会不会有什么隐患或不可预见的后果Couldtherebeanyhiddendangers,anyunforeseenconsequences?人工智能会威胁到我们的存在吗AndcouldAIeventhreatenourexistence?假象真♥相♥人工智能大揭秘2023年英国皇家科学院圣诞讲座第三讲人工智能的未来:美梦还是噩梦欢迎来到第三场也是最后-场Welcometothethirdandfinal麦克伍德里奇教授英国皇家科学院圣诞讲座ChristmaslecturefromtheRoyalInstitution,本节目由CGl集团赞助播出aneventsupportedbyCGI.本次讲座我们将探讨人工智能的前景Inthislecture,we'llbelookingatthepromiseofAI及其在未来生活中的作用anditsroleinyourfuture.我们是该感到兴奋还是担忧Shouldwebeexcitedorshouldwebeconcerned?或者两者兼有之Ormaybealittlebitofboth?都有可能出现什么问题Howmightthingsgowrong我们要怎样做才能确保不出错andwhatcanwedotomakesurethatthatdoesn,thappen?随着人工智能变得越来越聪明AndasAIbecomeseversmarter,它最终会产生意识吗coulditeveractuallybeconscious机器的意识和我们的一样吗andwouldmachineconsciousnessbethesameasours?还是说意识为人类独有Orarehumanbeingsunique?我们先听听DeepMind的联合创始人1.et'sstartbyseeingwhatDemisHassabisl德米斯哈萨比斯对人工智能未来的看法theco-founderofDeepMind,thinksaboutthefutureofAI.我认为能在这个时代从事科技工作Ithinkit'sanincrediblyexcitingtime这是非常激动人心的tobeintechnologyandscience.我将人工智能视作IseeAIasliketheultimatetool帮助各位专家和科学家的终极工具toassistexpertsandscientistswiththeirendeavours.我想给初入职场的人一些忠告那就是AndwhatIwouldsayforanyonestartingoutnowintheircareer你们真的应该深入研究这些系统的工作原理istoreallydeepdiveintohowthesesystemswork学习如何利用这些系统andhowtheycanutilisethesesystems来帮助自己工作tohelpthemintheirownwork.我认为人工智能工具Ithinkit,salmostlimitless,在工作中的应用有着无限的潜力thepotentialforutilisingthesetoolsandinyourownworkflows.但它每时每刻Butit,sgoingtobe,youknow,都在更新换代it,sgoingtobechangingeverymoment.所以我想说的另一件事是学习SoIthinkoneotherthingIwouldsayistolearn,like,迅速学习并适应接下来发生的一切learnreallyfastandbeadaptivetowhatever,scomingnext.因为唯一能确定的是Costheonlythingthat,ssureis,未来将有许多变化there'sgoingtobealotofchange.让我们回到之前展示的Solet'sgobacktothatdriverless无人驾驶汽车之旅carjourneythatWeshowedyouearlier,我们将看到人工智能andwe'regoingtoseehowAI如何应对我家乡牛津狭窄拥挤的街道dealtwiththemeanstreetsofmyhometown,Oxford.下面有请牛津大学的保罗纽曼教授Pleasewelcome,fromOxford,ProfessorPaulNewman.欢迎你保罗-谢谢Welcome,Paul.-Thankyou.你已经乘坐过无数次无人驾驶汽车了Now,you'vebeenindriverlesscarsamilliontimes.而我是第一次乘坐Thatwasmyfirsttime真是非常奇怪的经历anditwasareallyweirdexperience.我们在牛津遇到的路况相当复杂WewentthroughsomequitecomplicatedtrafficinOxford汽车的驾驶席却空无一人andtherewasnobodybehindthewheel.那么汽车是如何知道车身周围的情况呢Sohowdidthecarknowwhatwasgoingonaroundit?这辆车装备有各种传感器Well,thevehicle,sequippedwithvarioussensors.有摄像头还有激光探测器Ithascameras,ithaslasers.从这个片段可以看到Andyoucanseethosecamerasthere人工智能正在处理摄像头的画面beinghavingtheirimagesprocessedbyanAI,这是真实旅程吗追踪人们的.Thisisouractualjourney?-trackingpeople.是的这是牛津大街上的真实旅程Thisisyouractualjourney,yes,upthroughOxford.你可以看到它在通过识别图像追踪周围的汽车Andyoucanseeit'strackingcars,identifyingfromimages,在路口还能识别红绿灯trafficlights,wheretherearejunctions,所以它知道周围有什么及如何驾驶soitknowshowtomoveandwhat'saroundit.我们有摄像头能看到Sowehavecamerasthatcansee并捕捉图像中的不同颜色andpickupdifferentcoloursinimages.我们有激光探测器能测量与物体的距离Wehavelasersthatcanmeasurethedistancetothings,有雷达能看到摄像头和激光探测器andwehaveradarsthatcanseeperhapswherelasersandcameras在浓雾中无法看清的东西mayhavetroubleseeingthroughfog.有这许多不同的传感器就能确保它知道Solotsofdifferentsensorsherehavearobustwayofknowing你的具体♥位♥置-好吧whereyouareintheworld.-Ok.所有这些传感器提供了Soallofthosesensors,they'reproviding一连串你周围环境的数据abarrageofdataaboutyourenvironment你们都会用这些信息做什么呢andwhatareyouactuallydoingwiththatinformation?这么做用意何在What,sthepointofit?无人驾驶汽车Therearethreekeyquestions需要解决三个关键问题thatthevehiclesneedtoanswerinthevehicle.第一我在哪里One,whereamI?第二我周围是什么Two,what'saroundme?我周围有什么东西Whatstuffisaroundme?第三那些东西会做出什么行为Three,what'sgoingtohappenwiththatstuff我又该如何应对andwhatshould1dobecauseofthat?对即将发生事情做出预测后我又该怎样反应HowshouldIdrivebecauseofmypredictionsofwhat,shappening?最近的问题是这些车应该共享哪些数据Andmorerecently,then,iswhatdatashouldthosevehiclesshare以便训练其他无人驾驶汽车sotheycantrainothervehicles?好吧以我的经验OK,now,inmyexperience,开车最难的部分莫过广thetoughthingaboutdrivingisotherhumanbeings.道路上的其他人无论他们是驾车It,sotherroadusers,whetherthey,reincars,骑自行车还是步行onbicyclesorjustpedestrians.当你的车遇到突发事件时Sohowdoesyourcardecidewhattodo是如何决定怎么做的whenitencounterssomethingunpredictable?我们有段之前行驶时Ithinkwehaveavideoofsomething发生类似情景的录像thathappenedwhileweweredriving.好OK,yes.你们来到一个环岛一辆出租车超车Soyoucomeuptoaroundaboutandacarpasses与你们擦身而过andthattaxiwasveryclose.当时车载人工智能所做的Sowhatthevehiclewasdoingatthatpoint,就是对可能发生的事做出许多假设theAI,iswe'remakinglotsOfhypothesesofwhatcouldhappen.那辆出租是会停车还是靠得太近Thevehiclewasgoingtopullin,gettooclose.还是会让你先走Itwasgoingtoletyougoinfront.许多假设1.otsofthings.好吧你们从视频中看不出来OK,whatyoudon'tgetfromthatvideo当时我们正在过一个环岛isweweregoingroundaroundabout,那辆车就从夹转间穿过thiscarreallydroveinthemiddle.实际上假如当时是我在开车And,actually,ifIhadbeenahumandriverbehindawheel,那将是非常恐怖的经历thatwouldhavebeenaveryscaryexperience.但这辆车只是缓缓地踩下了刹车Butthecarjustseamlesslyputitsbrakeonandyouhardly,你几乎都注意不到真是太不可思议了youhardlynoticed,itwasreallyincredible.保罗我们听说在其他国家Now,wehearinothercountries,Paul,有更多无人驾驶汽车therearemanymoredriverlesscars,比如在美国等国家的街头forexample,onthestreetsoftheUnitedStatesandsoon.我们也听说发生了些事故Andwehearaboutaccidents,还有人员受伤甚至死亡Wehearaboutinjuriesandfatalitiesandsoon.自动驾驶汽车在街道上行驶真的安全吗Isitreallysafertohaveautonomousvehiclesonourstreets?我想会很安全Ithinkitwillbe.我认为有充分的理由相信And1thinkthere'sareallyverystrongreason这项技术whythistechnology将对交通运输产生巨大影响isgoingtomakeanextraordinarydifferencetotransport.作为人类当你学开车时Soasyouasahuman,whenyoulearntodrive,你只有自己的经验youonlyhaveyourownexperience.但这些自动驾驶汽车就不同了Butthat'sgoingtobesodifferentforautonomousvehicles,它们可以向所有其他有行驶经验的车学习thatwilllearntodrivefromeveryothervehiclethathasdriven,基于人工智能我们能让这些自动驾驶汽车costheseAIs,wecanhavethemshareeverymile分享行驶的每一英里thatwasdrivenbyanautonomousvehicleever.这真的非常了不起Andthatisextraordinary.这跟我们人类恰恰相反Andit,stheexactoppositewaytoweashumansdoit,因为我们的经验仅限于自己不会大规模分享coswe'rejustourselves,notsharingatscale.基于这个了不起的原因Andthatisanextraordinaryreason我相信这项技术whyIreallybelievethistechnology将会当无人驾驶汽车出错时呢isgoingto.-So,whenadriverlesscarmakesamistake?该信息将在整个汽车群内共享Thatwouldthenbesharedacrossthewholefleet每辆车都会知道andeveryvehiclecangetanupdateintheway而我们人类司机则不会thatweasahumansetofdriverscan'tgetanupdate.如果在遥远的哥本哈根发生了什么事Youcan'tgetanupdate,animprovement,你就无法立即得知并加以改进becauseofsomethingthatjusthappenedinCopenhagen.好吧保罗感谢分享你的无人驾驶之旅OK,PaU1.thanksforthejourneyinyourcar谢谢你来参加今天的讲座andthanksforjoiningustoday.非常感谢Thankyouverymuch.现在大家可能还不觉得无人驾驶车是机器人Now,youmightnotthinkofadriverlesscarasarobot,但它确实就是butitreallyisarobot它用到了许多先进的科技anditusesmanyofthesametechniques在不久的将来我们在家里和工作场所也会见到thataregoingtobeusedbytherobotsthatwe'regoingtoencounter使用相同技术的机器人inourhomesandourworkplacesverysoon.为了更好的了解机器人如何工作Andtoseehowtheserobotsaregoingtowork,下面有请来自牛津机器人研究所的fromOxfordRoboticsInstitute,莫里斯法♥伦♥和机器狗点点pleasewelcomeMauriceFallonandSpot,therobotdog.欢迎莫里斯欢迎点点Welcome,Maurice,andwelcome,Spot.今天稍早我们放点点在阶梯教室随处走Now,earlierintheday,weletSpotlooseinthelecturetheatre在旁边这位操作员米凯尔的协助下withthehelpofitsoperator,Mikhail,standingoverthere,测绘阶梯教室tomapthetheatre.大家可以在影片中看到点点的行动AndontheprojectoryoucanseeSpotdoingthat.莫里斯请告诉我们点点是如何对周边环境So,Maurice,canyoutellus,howexactlyisSpotmapping展开测绘的thespacearoundit?点点身上装了大量传感器So,Spotisblessedwithavarietyofsensorsonboard.顶部有激光探测器就是保罗刚刚It,sgotthislaserontop,thesamekindofsensorthatPaul提到过的那种wastalkingaboutearlier.它装有许多摄像头It'sgotavarietyofcamerasthat提供360度的视角所以它能看到地上的障碍物formaringarounditsoitcanseeobstaclesthatareintheground其腿上装有传感器所以能行走-好andithasasensorinitslegssoitcanwalk.-Ok.现在听起来有点像刚才提到的无人驾驶车Now,thatsoundsabitlikethedriverlesscarsearlier.理念相同吗非常类似Isthatthesameidea?-Verysimilar.好那它在做的事就是给周围环境OK.Andsowhatit'sdoinginthisisbuildingupamap画出一张地图吗ofitsenvironment?对它在行走过程中使用传感器Yeah,it,susingthesensorsasitwalksaroundtobuildup对周围环境建立起数据库adatabaseofwhat'saroundit.这里可以看到它建立起的数据库Andwehavehereaviewofitsdatabase.可以看到机器狗所搭载的摄像头Youcanseethecamerafeedfromtheonboardsensor,传回的画面theonboardcameraonboardtherobot.从这里能看到一个俯瞰的视角Andhereyoucanseeakindofviewfromabove灰色是之前的地图thatshowsthepriormapingrey彩色斑点则是那些andthecolorisedpointsarethepoints激光探测器正在捕捉的地方thatarebeingcapturedrightnowfromthelaser.这些直线则代表Thesestraightlinesindicate机器人可以选择的thekindofhighwaythattherobothas,可通行路径whereitcanchoosetogoto.点点解读了所有数据Ok.NOw,Spot'sinterpretedallthedata,就像我们刚刚看到的无人驾驶车一样justlikethedriverlesscarsthatwesawearlieron.人工智能则协助机器狗AndtheAIhashelpedittobuildamap为空阶梯教室画出地图oftheemptylecturetheatre.现在我们认为点点已经可以Now,wethinkatthispoint,Spotcanautonomously在这个环境中为自己导航navigatethisenvironment.今天稍早我们给点点看了这只球So,earliertoday,weshowedSpotthisballhere.这次我居然接住了Wow,Icaughtitthistime!我们让点点看到了这颗球的样子AndweshowedSpotwhatthisballlookedlike.现在我们要让点点完全自动的Andwhatwe,regoingtoaskSpottodoistonavigate去找到这只球tothisballcompletelyautonomously.过程中不会有任何人员操作So,notunderanybody'scontrol.好Ok.但我们给点点制♥造♥了些难题Butwe'vemadeSpot'slifealittlebitmoredifficult增加了一些障碍物We'veintroducedsomeobstacles.点点需要弄清楚Spot'sgoingtohavetofigureouthowtoget如何绕过障碍及如何通过aroundtheseobstaclesandalsohowtotraversethosesleepers地上的轨枕onthefloorthere.终F这句台词我已经期待了好几周了Now,thisis,I'vebeenwaitingweekstosaythisline.开始Herewego.出发点点开始搜寻Go,Spot,fetch!好的Ok.莫里斯现在它在做什么Maurice,tellus,what'sitdoing?显而易见它爬上了楼梯Soit'sobviouslywalkedupthestairs并且这次的搜寻已经结束andgottotheendpointinitssearch.大家看这里的图片Andyoucanseehereapicture,是向左的相机拍到的图片thepicturefromthecamerafacingtotheleft.它已经找到球了Theballhasbeenspotted.球外圈了一个框Aboxisaroundtheball.我们甚至能看到一个百分数Andwehaveevenhereapercentage显示这个物体有多大可能whichcorrespondstohowconfident就是那只球wearethatitisaball.不过这次搜寻真是很严谨Butthisisreallyseriousresearch.像点点这样的机器人都会应用在哪里呢SohowmightrobotslikeSpotactuallybeused?比如检查所核设施Forexample,inspectinganuclearfacility,这类的工作thatkindofwork.就是说我们会让点点这样的机器人Sotheideaiswe'llhaverobotslikeSpot去做一些危险工作thatdodangerousthings因为那些地方对人类可能会有伤害inplaceswhereitwouldbeunsafetosendhumanbeings.没错-好的Exactly.-Ok.莫里斯感谢到访也感谢你点点Maurice,thankyousomuchforcomingon,andthankyou,Spot!像点点这样能定位能走动的人工智能ArtificialintelligencethatcannavigatetheworldlikeSpot也可以装备武器mightalsobeequippedwithweapons变成机器士兵toturnitintoarobotsoldier.现在我要说明一下点点的制♥造♥者们Now,weshouldsaySpot'smanufacturershaveaverystrict对F机器人的武器化有着十分严格的政策policyagainstWeaponisingtheirmachinesinthatway.但其他机器人公♥司♥则无类似的规定Butotherroboticscompanyhavenosuchreservations,能自动作战andautonomousfightingmachines自主决定攻击目标的战争机器thatareempoweredtodecidewhotoattack将从根本上改变战争的性质wouldfundamentallychangethenatureofwarfare.包括我在内的诸多人工智能研究者都认为AndmanyAIresearchers,myselfincluded,thinkthisissomething我们永远不该这么做thatweshouldjustneverdo.使用致命的自动武器是否合理Whetherit'sacceptabletouselethalautonomousweapons应该是近年人工智能成功所引发的isprobablythesinglemostimportantandprominentissue最重要最尖锐的问题that'sbeenraisedbytherecentsuccessofAI.今天我们请来了一位政♥治♥理论家Sojoiningustodiscussthepossibilityofthesemachines来探讨智能机器的可能性及其带来的问题andtheissuesthattheyraiseispoliticaltheorist,有请伦敦玛丽女王大学的埃尔克施瓦茨博士fromQueenMaryUniversityof1.ondon,DrElkeSchwarz.谢谢埃尔克来到现场-感谢邀请Elke,thanksforjoiningus.-Thanksforhavingme.我觉得很多听众都认为机器士兵Now,Ithinkalotoftheaudienceimaginethatrobotsoldiers,自动致命武器等要么只是科幻设定lethalautonomousweaponsiseithersomethinginsciencefiction要么是未来才会发生的事orit,ssometimeinthefuture.是这样吗Isthatthecase?实际上不是自动致命武器Well,no,actually,autonomousweaponsystems已经存在了几十年至少在七十年代就有havebeenaroundformanydecades,atleastsincethe'70s,非常原始和基础的雏形存在inamuchmorerudimentary,basicform.但装备人工智能的系统是新鲜事物ButthosesystemsthatareequippedwithAl,theyarenew,但也是现实存在的东西我们认为其已经buttheyarealreadyhereandwethinkthatthey'rebeingused被应用F一些冲突中insomeconflictsalready.真正的问题是是否有人Butthequestionreallyis,isthereahumansomewhere在对这种事进行监管督查overlookingthatlooporoverseeingthatloop?人们对全自动武器的应用讳莫如深Peopleareverysecretiveaboutfullautonomoususe.这是敏感话题因为现在说不清It,sadelicatesubject,becauseit'snotclearatall此事是法律范畴还是伦理范畴whetherthat'slegalorethical.我知道你对人工智能在战争中的使用Now,Iknowyouhavelotsofconcernsabouttheuse非常担忧ofAIinwarfare.可否与我们分享下你的想法Canyoutalkusthroughsomeofthose?我们必须知道战争难以预料Well,wehavetoremember,warfareisnotpredictable.战争存在诸多偶然因素Warfarehasmanyelementsofchance.敌方会欺骗会使诈Theenemytriestodeceive,tobetricky.众所周知战局既迷局There'stheproverbialfogofwar.我们无法得知战争中会发生什么Wedon'tknowwhathappensinwarfare.这就表示不可能有足够数据训练人工智能ThatmeansanAIisnotlikelytohaveenoughdatatobetrained使其可靠或者能被可靠使用reliablyortobeabletobeusedreliably,也就是说会犯错误whichmeansmistakeshappen.这将是大问题That'sabigissue.好好的Ok,ok.在你所有的顾虑中Now,ofallofyourconcernsthatyouhave,对人工智能用于战争的最大担忧是什么what'sthesinglebiggestconcernaboutAIinwarfare?能想到的盥梦场景有很多但其中之一TherearemanynightmarescenariosIcanthinkof,butoneofthe令我们非常担心的就是所谓的闪电战thingsthatwe'reconcernedaboutareso-calledflashwars.闪电战Flashwars?闪电战是指一个人工智能系统SoaflashwarwouldbewhenanAI甄别出潜在威胁systemidentifiesapotentialthreat然后在几秒之内andthen,withinseconds,向另一人工智能系统发出行动信♥号♥♥pingsoffanactionontoanotherAIsystem使其启动武器平台whichactivatesperhapsaweaponsplatform.随即由下一个人工智能进行反应AndthenanotherAIreactstothat.所有这些都瞬间发生几秒之内Andallofthathappenswithinaflash,withinseconds,速度之快没人能及时监管somuchsothatnohumanwillhaveoversight了解发生了什么overwhatisactuallyhappening,发生原因或者如何停止whyitishappening,orhowtostopit.即人工智能将事态升级到战争却没人Sothat'sAIjustescalatinguntilwereachawarwithouthumans有机会进行干预-没错havingtheopportunitytointervene.-That,sright.好请大家看大屏幕Well,OK,thisisonthescreeninfrontofus.这是斯坦尼斯拉夫彼得罗夫空军中校Thisis1.ieutenantColonelStanislavPetrov.我想大部分观众对他一无所知Now,Ithinkmostofyouwon'thaveheardofhim,但要不是彼得罗夫我们今天都不会存在butifitweren'tforPetrov,wemightnotbeheretoday.1983年冷战正酣时彼得罗夫当值In1983,attheheightoftheColdWar,Petrovwasonduty苏联的预警系统显示whentheSovietUnion'searlywarningsystemsindicated美国的导弹可能即将来袭apotentialincomingmissilestrikefromtheUnitedStates.彼得罗夫的职责是立刻上报该攻击AndPetrov'sjobwastoimmediatelyreporttheattack.如果他当时上报了Andifhehaddone,苏联就会反击发射原♥子♥弹♥aretaliatorynuclearstrikecouldhavebeenlaunched.但他没有上报Buthedidn't.他介入了决策过程认为这是误报Heintervenedanddecideditwasafalsealarm并上报了系统的错误andreportedafaultinthesystem.他是对的Andhewas