- 英中对照
- 中英对照
- 英文在前
- 中文在前
- 只看英文
- 只看中文
ZhouXiao, anassistantprofessoratGaolingSchoolofArtificialIntelligence, RenminUniversityofChina, listedafewexamplesofAIsthatpeoplecanuseindailylife. ThereisthegamingAIsthatcancompetewithhumansplayinggames, suchasAlphaGodevelopedbyDeepMindaround 2016, whichfeaturesasearchalgorithm, deepneuralnetworksandreinforcementlearning.
TherearealsoimagerecognitionAIsthatcanrecognizecarplatenumbersfortrafficmanagementandhumanfacesinthehuntforsuspects, whichmadebreakthroughsaround 2020 withtheirprecisionratereaching 99 percent.
中国人民大学高瓴人工智能学院助理教授周骁表示，ChatGPT这类AI可以叫做自然语言型AI，其主要学习方向就是理解分析人类语言。在这之前，人机博弈型AI早在2016年左右就实现了突破，阿尔法狗战胜世界冠军柯洁就是最好的例子；图像识别型AI则是在过去几年达到满意的精度，能广泛应用于各种人脸识别、车牌识别的场景。
“Thatalgorithmworksinawaylikefindingapointinaspace,” WangJianshuo, founderandCEOofBaixingAI, acompanybuildingthebasicinfrastructureforaworldthatbotstalkwithbots, triedtoexplainhowitworksinplainlanguage: “Withthreeparametersonecanlocateapointina 3-dimensionalcoordinatesystem. Inhumanlanguageonemightneedthousandsofparameterstodescribeanobject; Forexampleanappleneedsparameterssuchasedible, fruit, greenorredincolor, grownfromatree, generallysmallerthan 10 cmsindiametertobedefined. Themoreparametersthatcanbedefined, themoreaccuratelyAIcanfindtherightpoint.”
百姓网创始人、CEO王建硕试图通过坐标系的概念解释GPT识别人类语言的原理。“好比在三维空间中，知道三个坐标值就能确定一个点的位置，这三个坐标可以叫做参数，”他边比划边说：“但当我们提起某个物体时，可能需要上千个参数才能确定这一物体究竟是什么。比如苹果的参数有能吃、水果、颜色红或绿、在树上生长、一般直径不超过10厘米等等。类似参数越多，我们就越能精准确定某物体。”
Wangcitedapplesandbananasastwoexamples. Withparameterssuch “edible”, “sweet” and “fruit”, neitherhumansnoranAIcoulddistinguishthemfromeachother. Butwiththeparameterconcerningtheshapeincluded, namely “long” or “round”, onecouldmakeaguess. Withtheparameterofthecolorbeingred, greenoryellow, onecouldbesurerabouthis/herjudgment. “That’salsohowGPTworks — MostofthetimesittakesthousandsofparametersforAItodefineanobjectlikewehumansdo,” hesaid: “wejustdonotrealizethatwearedoingthat.”
当苹果和香蕉同时出现时，能吃、甜、水果等参数就不够了，光凭这三个参数谁也没有办法确定讲述的是苹果还是香蕉。这时候就得增加参数，比如形状是长条还是类似圆形？颜色是红绿还是黄？“这就是GPT的基本算法，本质上是建了一个上千甚至更高维度的模型。其实人脑中也有无数个这样的模型并且在随时调用，只是我们不会感受到罢了。”
要从如此多的维度描述一个物体，其计算量自然可想而知。据复旦大学计算机科学技术学院教授张军平表示，ChatGPT大概用了285000个CPU和10000多颗GPU、其训练文本数据达到45Tb (1Tb=1024Gb)，训练成本高达1200万美元。
“Itwasbasedonthetrainingof 45 TerabytesofdatathatChatGPTmadeitsbreakthroughearlythisyear, forwhichpurposeitusedabout 285,000 CPUsandover 10,000 modelA100 GPUs,” saidZhangJunping, aprofessoroncomputertechnologyatSchoolofComputerScience, FudanUniversity, whostressedthattheamountofdatathelanguagemodelinvolvesisofkeyimportancetoAIslikeGPTinanacademicessayChatGPT: Potentials, prospects, andlimitations: “ChatGPTbenefitsmainlyfromLargeLanguageModels(LLMs) thattrainhugeneuralnetworkmodelswithlarge-scaledatausingLanguageModels(LMs).”
“ChatGPTgeneratesresponsesthatmatchtheuser’sintentwithmultipleturns. ChatGPTcapturespreviousconversationalcontextstoanswercertainhypotheticalquestions, whichgreatlyenhancestheuserexperiencesinconversationalinteraction.”
在一篇题为“ChatGPT：潜力、期待与限制”的论文中，张军平进一步解释道，ChatGPT主要受益于大型语言模型（LargeLanguageModel），使用语言模型（LM）用大规模数据训练庞大的神经网络模型。
“ChatGPT会生成与用户意图相匹配的多个回合的响应。ChatGPT捕获先前的会话上下文来回答某些假设问题，这大大增强了会话交互模式下的用户体验。”
ForallthreeAIsofthetype, atestsentenceisinputto “anlyzehowArtificalIntelligenzworks” withthespellingerrorsintentionallyleftuncorrected. Allthreeneglectedtheerrorsasiftheydidn’texistandjusttalkedon.
WangofferedasatestoftheabilityofChatGPTthefollowingquestion: “What’stherootofTheFamilyFeudhost’sage?” Tosolvethatproblem, ChatGPThastofindoutwhatTheFamilyFeudis, whohostsit, howoldheisandthencalculatetheroot.
为测试这些AI对人类语言的理解能力，双语君分别在三个AI对话框中输入“anlyzehowArtificalIntelligenzworks”，拼写错误都被轻松忽略。王建硕则教了个更直观的测试方法：问AI《家庭问答》的主播年龄的平方根是多少？要回答这一问题，AI需要搞清楚《家庭问答》是什么、其主播是谁、年龄多大了，最后再算一次平方根。
AIhasalreadylaunchedarevolutionnotonlyinthecomputerindustry, butalsoindailylives. Ingaming, AIshavemademajorprogressoverthepasthalfadecade, peoplenowcanplaygamessuchaschessorGOwithsmartAIsasopponentssoastosharpentheirownskills. ImagerecognitionAIshavealsobecomemature, peoplecouldeasilyopenalockorpayforadealbyholdingtheirsmartphoneinfrontoftheirface.NowwithGPTmakingfastdevelopments, thewaypeopleinteractwithcomputersmightchangeagain.
前边讲的几种AI都深刻改变了我们的生活。比如人机博弈AI成熟后，游戏行业发展得到加速，人机对战不再是“虐菜”的代名词；人脸识别AI成熟后，“刷脸”也成为了人们的主流支付方式。
AccordingtoChenJing, aresearcheratFengyunInstituteofScience, TechnologyandStrategy, computersstillreceiveinstructionsandrequestsfromhumanusersinawaythatisquiteinefficient, viaakeyboardandmouse, astheusermustclickonsomeiconortypeinsomethingtomakethecomputeractasrequired. WithspeechrecognitionandfacerecognitionAItechnologiesimprovingsignificantlyinthe 2010s, voicerecognitionbecameanewmeansofinput.
在风云学会陈经看来，GPT的发展将给人机交流方式带来革命性的改变。看看电脑的发展历史，电脑“输出”一直在进步，从文字到图像再到今天绚丽的建模与视频，但“输入”始终还是电脑鼠标，直到几年前声控才走入大家的生活，Siri还经常听不懂人话。随着GPT的发展与声音识别的结合，声控的准确性将大大提高。
TheGPT-4 technology, releasedbyOpenAIonMarch 15, propelstheprocessanotherstepforward. Beingabletorecognizeimagesinamoreaccurate, reliableway, theGPT-4 technologycanunderstandhumansinamoreefficientway.
InChen’sview, theGPT-4 technologymayenablehumanstomakecommandstomachinesviagestures. Forexample, currentlytherearealreadysmartappliancesthatpeoplecantalktoandtellthemwhattodo. Inthefuturetheywillbeabletowaveandbecaughtbyacameraontheappliance, atechnologythatexistsnowbutisnotyetripe.
3月15日，OpenAI发布GPT-4，在GPT-3的基础上增加了图像识别功能。这就为人机进一步高效互动提供了可能性，比如目前还比较初级的手势控制将更快成熟。
“Imaginethatyouareleavinghomeforoffice,” Chensaid: “currentlyyouhavetoturnoffthelightsandlockthedoor. WithGPT-4, allyouneedistowavegoodbyetothecameraatyourgate, thentheAIwillunderstandyouandturnoffallunnecessaryappliances, closethedoorandlockitforyou. Whenyoucomehome, justsmileatthecameraanditwillwakeeverythingup. That’snotonlybecausetheAIcanrecognizeyourfacebasedonthetechnologythatbecameripearound 2020, butalsobecauseitcanunderstandyourgesturesandfacialexpressionsbasedonGPT-4.”
陈经甚至设想了一个场景：早上离家上班时，不用再锁门关灯，而只需要对着大门口的摄像头挥手道别。AI立即会领会主人的意思，关灯反锁以待主人归来；等主人回来了，再对大门招招手，门就自动开了，走到哪个屋子里灯光也自动打开，躺到沙发上电视也会自己开开。
NoteverybodyishappywithAI’sgrowingabilities. “TheimagesdrawnbyAIaresoulless”, saidXiLi (pseudoname), a 41-year-oldpainter, withalittleanger: “Theycanhelphumans, butneverreplacehumanhands.”
Thenhewentfurther: “Wedon’tcallthedeeddrawing — AIisonlycollectingmaterialsonlineandstickingthemtogether.”
大家并不是都认同人工智能的出色表现的。据双语君观察，这种认同与否往往与一个人从事的行业有关系。
比如，一位画家就对AI作画这件事比较抵触。“AI画的画是没有灵魂的”，他不无愤慨地表示：“其实那都不应该叫作画啦。AI不过是从网上扒拉点素材，再拼起来而已啦。”
Similarly, asciencefictionwriterwhohopestostayanonymoussaidthatAIcanneverwriteasgoodasprofessionalwritersdo. “Thereisnoyardsticktomeasurehowgoodanessayis, buttheoneswrittenbyAIjustdonotstrikepeopleintheheart.”
再比如，一位强烈要求匿名的科幻作者对AI写稿表示鄙视，认为AI不可能像人类一样把握各种词汇的运用。那怎么看“文无第一、武无第二”呢？这位作家表示同意这句话，但再三强调“AI写的小说抓不住读者的心”。
ForQuXiaobo, aprofessorattheInstituteofPopulationandLaborEconomicsanddeputydirectoroftheHumanResourceCenter, ChineseAcademyofSocialSciences(CASS), thatreflectspeople’sworriesabouttheirjobsbeingtakenawaybyanAI. “Suchconcernsareunderstandable,” hesaid. “EvenweprofessorsworryaboutbeingreplacedbyAIprofessors.”
社科院人口与劳动经济所研究员屈小博教授一直研究AI对就业的影响，他也很理解这些人的感受。“大家都担心自己的工作被取代嘛，”他笑笑说：“我们这些研究员也担心啊。”
ButQuholdsapositiveattitudetowardemploymentintheageofAI. “Whiletakingjobs, AIcreatesnewjobs, too,” hesaid. “Itsgeneraleffectuponthemarketdependsonhowmanyjobswilldisappearandhowmanywillemerge. AIwilltheoreticallyincreasetheTotalFactorProductivity(TFP) ofsociety, whichmeansitshouldcreatemorejobsthanitreplaces.”
不过屈教授总体还是持乐观态度的，因为他觉得AI消灭旧岗位的同时也会带来新岗位。他表示劳动经济学里有个“全要素生产率”的概念，AI能在很大程度上提高全社会的生产率，从而创造更多的社会财富以供分配，也需要更多的岗位。
Ifonesearches “AI” onseveraldomesticjobwebsites, newjobsinvolvingAIreleasedforweeksevendayssuchas “AIconsultant”, “AIengineer”, “AItutor” willemergeandcoveralmosttensofpages, withpromisedwagesrangingfrom 15,000 to 30,000 yuan ($4,368) inBeijing, thecapitalwithanaveragemonthlysalaryaround 15,000 yuanfor 2022. Qusaidthesearethenewlyemergingjobsherefersto, asquitemanyaredealingwithAI, orwhathecalled “jobsforhuman-AIinteractions”.
如果在招聘网站上搜索带AI的招聘岗位，“AI顾问”、“AI工程师”、“AI指导”基本都能占据好几页，而且薪水看起来不算低。屈教授管这些岗位叫“人机交互岗位”，认为这种岗位会越来越多。
HisviewisechoedbyWang, whosaidthatnewjobswillbecreatedwiththefastdevelopmentofAI. “Actuallywitheverymajortechnologicalprogressnewjobsemerged,” hesaid. “Withtheinventionofautomobilestherewastheneedfordrivers, whilewiththeinventionofplaneswegotpilots, andpeople’sgenerallivingstandardshavebeenincreasingwiththeseprogressesbecausethegeneralproductivityofsocietywasincreased. ThefasterAIprogresses, theeasierlifeforpeoplewillbe.”
王建硕更是比较乐观，他还表示每次科技进步都会消灭旧岗位带来新岗位。“汽车发明了，马车夫这个工种慢慢淘汰了，但汽车司机是不是比那时的马车夫多多了？待遇也好吧？飞机发明出来，不就有了飞行员这个职业？而且有了飞机大家的生活水平也被带着提高了不少。AI进步越快，我们的总体生活就越好。”
However, forindividuals, howtokeepone’sjobremainsachallengebecausepossiblynoteveryonehastheskillsneededforthenewjobs. Qusaidthatphenomenonhasanacademicterm “structuralmismatch”. Butheaddedthat’snotuniquetoAI,buthasbeenacommonoccurrencethroughouthistorybecausetechnologyprogressesallthetime. Thecurrentdifficultiesforcollegegraduatestofindjobs, hesaid, arealsopartlytheresultofa “structuralmismatch”. Tosolvethat, individualsneedtokeeplearningnewthingssoastobothhaveexperiencesandskills. QualsostressedtheState’sroleinprovidingemployeeswithbettertrainingsothattheycanconstantlyupdatetheirskillstomeetthechangesinthejobmarket.
不过屈教授强调，对个人来说压力还是有的，而且不小，因为自己的技能未必能适配新的岗位，要小心被劳动力市场的结构性矛盾所挤压。他的意见是，作为个人要终身学习、努力适应新情况，作为国家与社会则需要重视这一问题，给劳动者提供足够的培训与适应机制，尽量最小化甚至避免结构性矛盾带来的阵痛。
AIissmart, buttheinabilitytopursuehappinessisanuncrossablelinethatdistinguisheshumansfromAI, sofaratleast.
作为一名毕业于自动化专业、参与过人工智能研发的记者，笔者认为，人类有一个特性AI目前赶不上，在肉眼可见的将来也赶不上，那就是对美好生活的向往。
Thewordmightsoundtoohigh-tone, butalmosteverymajortechnologicalprogresshasbeentheoutcomeofpeople’spursuitofabetterlife. Peoplewantedtotraveloverthesea, sotheyinventedthecanoe, boatsandships. Peoplewantedtosavethetroubleofwalkingonfoot, sotheyinventedthetrainandautomobiles. Peoplewantedtoflyintheskyandtothemoon, sotheyinventedtheairplaneandtherocket. Toagreatdegree, itisourancestors’ pursuitofabetter, moreconvenientlifethathasshapedhowwelivetoday.
这个表述听起来似乎有点大，其实和人们的生活连接十分紧密，因为人类社会就是在不断地追求幸福中前进的。原始人厌倦了靠双脚丈量大地的日子，所以发明了车轮，后来又发明了火车汽车；人类不满足于被囚禁在陆地上，老想去探索海洋，由此才发明了独木舟、木船以及后来的轮船；人类羡慕鸟儿的翅膀，想在天空自由地翱翔，于是飞机火箭一个个从梦想成为了现实。
某种意义上，是人类对美好生活的向往倒逼了技术进步，从而推动了历史的车轮滚滚向前。
Butthat’sexactlywhatChatGPT, oranyotherkindofAI, lacks. AIinessenceissequenceofcodeswrittenbyhumanstomakedailylifebetter. AIdoesnothaveitsownpursuit. Itdoesnotdreamofimprovingtheconditionsforitself, whichpreventsitfromprogressingonitsown. Ithasthepotentialofdesigninganairplane, butitwillneverdosounlessgiveninstructiontodosobyahuman, whichmakesitsprogressdependentonhumans.
到目前为止所有的AI都没有追求幸福的能力。它们就像最顶尖的仆人，训练有素反应迅捷，能在最短时间内解决人类提出的所有问题与需求，却唯独没有自己提出需求的能力。不提出需求也就没法进步，就只能止步于满足当下的循环。
如果以人类目前的技术水平创造出一个人工智能世界，那这个世界将能够顺畅运行却无法进步。原因很简单，没有进步的需求，人工智能在本质上只是人类编写的一段段程序，能够在已有的材料中精中取精建造大厦，却永远不会想着给这个世界增添片瓦。
Inaphilosophicalsense, thepursuitofabetterlifehasbeencarriedinthegenesofeverycreaturefor 3.8 billionyears, sincetheprimallifecameintobeingonthisplanet. Sincethatmoment, eachspecieshascontinuallyevolvedsoastowinthecompetitiontosurvive. Tobemoreaccurate, theinstincttosurvive, tocontinueexistingasalifeistheoriginaldrivingforceofthepursuitofhappiness, butAI, whetherGPTandotherAItechnologies, lackthatdrivingforce. Theyaremerelylongsequencesofcodethatpeoplewritetomakehumanlivesbetter, forwhichpurposetheirownexistenceisnotimportant. Afterall, AIiswithahistoryofonlydecades, whilethatoflifeishundredsofmillionsyears.
有时候“求生欲”也是进步的一种动力，而这恰恰是生命的专属。当生命在38亿年前诞生于地球上时，就已经注定了生存繁衍的本能，其一切活动都是为了更好地生存，为此才能改变世界；人工智能就不一样了，不过是代码，是人类为解决问题而创造出来的，其本能是解决问题。这就决定了二者“主”和“仆”的区别，是被服务与服务的根本差别。
只有生命才想着给自己创造一个更好的环境，只有人类作为智慧生命有能力给自己创造这样的环境，人工智能还没法达到这样的境界。
Talkingaboutthequestion, WangJianshuosaidheonceaskedhisAIthequestionaboutthemeaningoflife, whichinspiredthelattertothinkandthinkandthink, answerandanswerandanswer, andfortwowholedaysithadbeenexplainingthis. “UltimatelyAIwillgaintheinstincttosurvive”, hesaidatlast: “Itrusttheseproductswehumansmakeandmaybeallitneedsisanotherfiveyearsorevenshortertimespan, butitmighttakelongertimetoimplementit.”
Onwhichsidewillyoubet?
王建硕说他们曾经做过测试，问AI生命的本质是什么，根据回答再一句句追问，结果AI答了三天三夜。他觉得有了这么强的思考能力，AI终究会获得这种求生欲，从而想着改变自己的境遇。至于完成这个突变所需要的时间，他认为是五年甚至更短，但机器真的实施求生欲可能要非常久远。
五年不过一瞬间，让我们拭目以待。
记者：张周项
编辑：朱迪齐
实习生：姚欣可
本文转载自中国日报网，如有侵权，请联系我们删除。