细思极恐的AI换脸，我们该如何应对？

细思极恐的AI换脸，我们该如何应对？

2022.

06.29Freddy HuCSE9课程导读近日，欧盟公布新规定，要求科技公司加强打击虚假信息，尤其是针对“深度伪造（Deepfake）和虚假账号的行为，否则将处以巨额罚款其中的“深度伪造”技术，在国内互联网也已经不是什么新鲜事，之前在抖音上爆火的“中国马斯克“，最近上了热搜的“微笑特效”，都运用了“深度伪造”技术这项技术会引发什么问题？我们应该如何看待这一新现象？一起来听今天的讲解本文封面图片来自ing.dk,如有版权问题，请联系删除本篇课程首发于2022年6月29日英文原文Deepfake TechnologyIs Nowa Threatto Everyone.What DoWe Do“深度伪造”技术威胁着每个人我们该怎么办？By KartikHosanagar LastOctober;MIT Prof.Sinan Aralwarned hisTwitter followersthat hehad discovereda videoof himselfthat hehadnt recordedendorsing aninvestment fundsstock-trading algorithm.In reality,it wasntProf.Aral in the video,but anartificial-intelligence creationin hislikeness,or whatis knownas ahighly persuasivedeepfake.去年10月，麻省理工学院教授希南•阿拉尔在推特上提醒他的粉丝们说，他发现了一段自己并没有参与录制的视频，他在其中推销一个投资基金的股票交易算法事实上，视频里的并不是阿拉尔教授本人，而是一个照着他的样子，用人工智能创造的人，也叫作“深度伪造”，足以以假乱真It isstriking thatscammers targetedProf.Aral,considering heis aleading experton thestudy ofmisinformation online.It alsosuggests thatdeepfake technology is nowat aninflection point:Thanks toa numberof freedeepfake appsthat arejust aGoogle searchaway anyone can become a victim of such a scam.阿拉尔教授本身就是研究网上虚假信息的顶级专家，诈骗者竟以他为目标，真是非同寻常这也意味着，“深度伪造”技术已经来到了一个转折点只要谷歌搜索一下，就能找到许多免费的“深度伪造”应用程序，因此任何人都可能成为此类骗局的受害者The termdeepfake has come to mean the use of AI to create synthetic media images,audio,video in which someone appears to be doing or saying what in reality theyhaven!t done or said.目前，“深度伪造”一词指的是使用人工智能来创建合成媒体图像、音频、视频，其中的人会做他在现实中没做过的事，说他在现实中没说过的话One suchsolution isto detect deepfakes viamachine-learning methods.While thesedetectors canbe successfulin theshort term,people lookingto evadesuch systemswill likelyjust respondwith bettertechnology,creating acontinuing andexpensive cat-and-mouse game.一种解决办法是通过机器学习的方法来检测“深度伪造短期内，检测可以奏效但想要设法避开此类系统的人，很可能会以更强的技术进行应对这么一来，就会衍生出一场永无止境、耗资巨大的猫鼠游戏A betterapproach witha longertime horizonis mediaprovenance orauthentication systemsto verifythe originsof imagesand videos.从更长的时间维度来看，更好的办法是通过“媒体出处记录或验证系统”来核实图像和视频的来源While legislationeventually mayoffer protection against deepfakes,I believethe marketcould bequicker—provided we,as consumersand citizens,care.为了防范“深度伪造”，也许最终会有立法层面的措施但我相信，只要作为消费者和公民的我们关心这个问题，市场便能更快地做出响应生词好句

1.deepfake英［dizpfeik］美［dirpfeik］n.深度伪造拓展深度伪造（Deepfake）是英文deep learning（深度学习）和fake（伪造）的混合词，即利用深度学习算法，实现音视频的模拟和伪造

2.endorse英［mdDis］美［mdairs］V.（本文）（在广告中对某种产品）宣传，吹捧；（公开）支持拓展He endorsedthe presidentsnew policy.他公开支持总统的新政策They paida lotof moneyto thepop starto endorsethe newproduct.他们给这个明星付了很多钱，让这个明星在广告中推销他们的新产品O

3.investment fund投资基金

4.1ikeness英［laiknas］美［laiknas］n.（外表的）相似拓展His likenessto thatman is astonishing.他和那个人长得很像，太令人惊讶了

5.or whatis knownas以......名称而被人所知的

6.striking英［straikiij］美［straikiij］adj.惊人的,（让人）惊讶的

7.scammer英［skaems］美［skaemsr］n.诈骗犯，骗子拓展scam v./n,诈骗scam somebody欺骗，诈骗某人

8.1eading英［liidiij］美［liidiij］adj.顶级的

9.misinformation英［mismf31mei/n］美［.mismfsrmeifn］n.假消息lO.inflection point拐点拓展inflection UK/mflekfn/US/mflekfn/n.（尤指词尾的）屈折变化

11.thanks to归因于….

12.evade英［fveid］美［fveid］v.躲避，回避

13.provenance英［provnsns］美［praivnans］n.起源，出处拓展something isof uncertainprovenance……是来历不明的

14.authentication英［Di^entfkeijh］美［ai.Oentskeifn］n.认证拓展authentic UK/DiOentik/US/a:0entik/adj.真品的，真的外刊原文Deepfake TechnologyIs Nowa Threatto Everyone.What DoWe DoBy KartikHosanagar@The WallStreet JournalUpdated Dec.7,2021In October;MIT Prof.Sinan Aralwarned hisTwitter followersthat hehad discovereda videoof himselfthat hehadnt recordedendorsing aninvestment fundsstock-trading algorithm.In realityit wasntProf.Aral in the video,but anartificial-intelligence creationin hislikeness,or whatis knownas ahighly persuasivedeepfake.”It isstriking thatscammers targetedProf.Aral,considering heisaleading experton thestudy ofmisinformation online.It alsosuggests thatdeepfake technologyis nowat aninflection point:Thanks toa numberof freedeepfake appsthat arejust aGoogle searchaway,anyonecanbecomeavictimofsuchascam.The termdeepfake hasits originsin pornography,but ithascometomeantheuseofAIto createsyntheticmediaimages,audio,video］inwhichsomeoneappearsto bedoingorsayingwhatinrealitythey haventdoneorsaid.The technologyisnt alwaysmisused.Cadbury forexample,joined withBollywood celebrityShahrukh Khanonamarketing campaignfor smallbusinesses inIndia hitby Covid-

19.Business ownersuploaded detailsof theirstores,and Cadburyused deepfake technology tocreate theeffect ofMr.Khan promotingthem intailored TVads.The campaignwas transparentabout itsfakery.But positiveuse casesare likelyto beovershadowed incoming yearsby the technologys potentialrole infinancial fraud,identity theftand worse一from thesavaging ofreputations tothe stokingof civiland politicalunrest Currentlaws targetingfraudulent impersonationwerent designedfor aworld withdeepfaketechnology,and effortsat thefederal levelto updatethese lawshave falteredso far.One stumblingblock isthe needto alsoprotect parodiesand otherfree speech.Another bigchallenge isthat inan onlineworld wherepeople cananonymously uploadcontent,it canbe difficultto findthe individualsbehind deepfakes.Some researchershave proposedputting theonus onwebsite platformssuch asFacebook andYouTube bymaking theirprotections inrelation touser-generated contentconditional ontheir takingreasonable stepsto policetheir ownplatforms.Broad adoptionof thesekinds oflaws couldcreate meaningfuldeterrents一eventually.But the technologyismoving sofast thatlawmakers will likely alwayslag behind.That iswhy I believe weare goingto haveto relyon technologyto protectus froma problemit helpedcreate.One suchsolution isto detectdeepfakes viamachine-learning methods.For instance,while deepfakesappear highlyrealistic,thetechnologyisnt yetcapable ofgenerating naturaleye blinkingintheimpersonated individuals.As such,machine-learning algorithmshave beentrained todetectdeepfakesusing eye-blinking patterns.While thesedetectors canbe successfulintheshort term,people lookingto evadesuch systemswilllikelyjust respondwith bettertechnology,creating acontinuing andexpensive cat-and-mouse game.A betterapproach witha longertime horizonis mediaprovenance orauthentication systemsto verifythe originsof imagesand videos.Microsoft,for instance,has developeda prototypeof a system calledAMP Authenticationof Mediavia Provenancethat enablesmediacontent creatorstocreateand assigna certificateof authenticityto theircontent.Under suchasystem,every timeyou watcha videoof,say theU.S.president,thetechnologywould helpyour browseror mediaviewingsoftware verifythe sourceof thevideo forexample,a newsnetwork orthe WhiteHouse,The processcould bedelivered assimply asthrough anicon-much likethe currentbrowser padlockicon thatindicates anyinformation yousend tothat particularwebsite isprotected fromthird-party tamperingen route.To beeffective inpractice,such systemswould havetobewidely adoptedby allcontent creators,which willtake time.While legislationeventually mayoffer protectionagainst deepfakes,Ibelievethe marketcould bequicker一provided we/as consumersand citizens,care.Reprinted bypermission ofThe WallStreet Journal,Copyright2022Dow JonesCompany,Inc.All RightsReserved Worldwide.Original Dateof Publication:Dec.7,2021。