2024哈佛大學生成式 AI 將打破互聯(lián)網

GenerativeAIWillBreaktheInternet:BeyondSection230 Section230 Whatisan“interactivecomputer Whatisa“publisheror Whatis“anotherinformationcontent ABreakingPoint:TwoSection230 WhetherGenerativeAISystemsare“InformationProviders”underSection WhetherGenerativeAISystemsFallOutsidetheScope

AI230第230條概述.AB斷點POTWOSECTION230AI230“S.Wh?th?rG?n?rati??A?oyst?msca??l?tsiy?th?ov???Section230“PublisherorSpeaker”

Thelawthat“createdtheInternet”1hasreachedabreakingpoint.Section230oftheCommunicationsDecencyActisalawenactedin1996thatcatalyzedtheInternet’sdevelopmentbyprovidingInternet-basedserviceproviderslikesearchenginesandsocialnetworkslegalimmunityfromlawsuitsbasedonharmfulcontentcreatedbythirdparties.2Forexample,GooglegenerallycannotbeheldcivillyliableforsimplyretrievinganddisplayingharmfulInternet-basedcontentthatitdidnotcreate.3Similarly,Facebookcantypicallyavoidliabilityforharmfulcontentcontainedinauser’spostonitsplatform.4*GrahamH.Ryan,J.D.,isalitigationandappellatepartneratJonesWalkerLLP,withextensiveexperiencelitigatingcomplexcommercialissuesinallphasesoflitigation,appeals,andregulatoryproceedings,includingthoseinvolvingtechnology,artificialintelligence,Section230,andrelatedmatters.HeholdsaninternationaldesignationasanArtificialIntelligenceGovernanceProfessionalfromtheInternationalAssociationofPrivacyProfessionals,andhasbeenpublishedonlegalandconstitutionalissuesarisingfromemergingInternettechnologies.SeeJeffKosseff,TheTwenty-SixWordsThatCreatedtheInternet47U.S.C.§230(c)(1)(generally,“SectionSeeMarshall’sLocksmithServ.v.Google,LLC,925F.3d1263(D.C.Cir.SeeForcev.Facebook,Inc.,934F.3d53(2dCir.

“創(chuàng)造了互聯(lián)網”的法律已經達到臨界點。1996年通過的《通信規(guī)范法》第230條是一項法律，它通過為搜索引擎和社交網絡等基其未創(chuàng)建的有害互聯(lián)網內容而承擔民事責任。同樣，F(xiàn)acebook通常格雷厄姆·H·瑞安（GrahamH.Ryan），法學博士，是瓊斯沃克律師事務所（JonesWalkerLLP）的訴訟和上訴合伙人，在所有訴訟、上訴和監(jiān)管程序階段都有豐富的訴訟經驗，包括涉及技術、人工智能、第230條及相關事項的案件。他是國際隱私專業(yè)人員協(xié)會（InternationalAssociationofPrivacyProfessionals）認證的人工智能治理專業(yè)人士，并在新興互聯(lián)網技術引發(fā)的合法和憲法問題上發(fā)表了文章。1.參見杰夫·科索夫（JeffKosseff）的《那二十六個單詞創(chuàng)造了互聯(lián)網》（2019年）2.美國法典第47卷第230條（c）（1）（通常稱為“第230條”）。3.參見馬歇爾鎖匠服務公司訴谷歌有限責任公司案，925F.3d1263（哥倫比亞特區(qū)巡回法院2019年）。4.參見福斯訴臉書公司案，934F.3d53（第二巡回法院2019年）。 HarvardJournalofLaw&

furtheringthelaw’spurpose“topreservethevibrantandcompetitivefreemarketthatpresentlyexistsfortheInternetandotherinteractivecomputerservices,unfetteredbyFederalorStateregulation.”5ThefunctionalscopeofSection230immunityhasevolvedalongsideInternettechnology.ItinitiallycoveredpassiveintermediarieslikeAOLonlinemessageboardsata

幾十年來，美國法院廣泛適用第230條保護，進一步推動了該法律的目的，“為了保護目前存在的充滿活力和競爭的互聯(lián)網及其他互動計算機服務的自由市場，不受聯(lián)邦或州監(jiān)管的束縛?！?第230條AOL在線論壇這樣的被動中介，當時互聯(lián)網用戶總數為4000萬7whenthenumberofInternetuserstotaled40million,6butitnowadvancedsocialmediaalgorithmsthatfilter,promote,andpersonalizecontent7asthepopulationofInternetusershassurpassed5.35billion.8Intheinterveningperiod,courtshavestretchedSection230toitslogicalbounds—andsomewouldarguefarbeyond.9Butthelawestablishesonebrightlinecourtshavenotcrossed:Section230protectiondoesnotextendtoInternet-basedservicesthatactuallycreateordevelopcontent.10Technologyhascrossedthatline.TheInternet’sfuturedevelopmentwillbeshapedbygenerativeartificialintelligence(AI),whichperformsanunprecedentedtechnologicalroleincreatinganddevelopingcontentratherthanmerelyretrievingorexchangingit.11CourtshavelongbeenreluctanttodisruptSection230’slegalunderpinningsoftheInternet12andhavecarefullyadaptedSection230’slegalstandardstoextenditsprotectionsto

8，因為互聯(lián)網用戶數量已超過53.5億9。在此期間，法院將第230條推到了其邏輯極限——有些人可能會認為遠遠超出了極限。□但該法律確立了一條法院沒有逾越的明確界限：第230條保護不適用于實際創(chuàng)（AI）塑造，它在一個前所未有的技術角色中創(chuàng)造和發(fā)展內容，而不僅僅是檢索或交換內容。22長期以來，法院一直不愿意破壞互聯(lián)網第230條的法律基礎23，并已仔細調整第230條的法律標準，以將其47U.S.C.§Zeranv.Am.Online,Inc.,129F.3d327,328(4thCir.1997)(“‘TheInternetisaninternationalnetworkofinterconnectedcomputers,’currentlyusedbyapproximately40millionpeopleworldwide.”)(citingRenov.Am.C.L.Union,521U.S.844,849(1997)).Force,934F.3dSeeAniPetrosyan,NumberofInternetandSocialMediaUsersWorldwideasofJanuary2024,STATISTA(Jan.31,2024),https:///statistics/617136/digital-population-worldwide/(“AsofJanuary2024,therewere5.35billioninternetusersworldwide,whichamountedto66.2percentoftheglobalpopulation.Ofthistotal,5.04billion,or62.3percentoftheworld’spopulation,weresocialmediausers.”).Doev.Snap,Inc.,No.22-20543,2023U.S.App.LEXIS33501,at*2(5thCir.Dec.18,(Elrod,J.,dissenting)(arguingthatthecurrentscopeofSection230protectionprovides“sweepingimmunityforsocialmediacompaniesthatthetext[ofSection230]cannotpossiblybear”).47U.S.C.§230(c)(1)(conferringimmunityonlywhere“information[is]providedbyanotherinformationcontentprovider.”(emphasisadded));47U.S.C.§230(f)(3)(defining“informationcontentprovider”toinclude“anypersonorentitythatisresponsible,inwholeorinpart,forthecreationordevelopmentofinformation”(emphasisadded)).GENERAL-PURPOSEARTIFICIALINTELLIGENCE(2023).See,e.g.,TranscriptofOralArgumentat54,82,Gonzalezv.GoogleLLC,598U.S.617(2023)(No.21-1333)(Kavanaugh,J.)(“Congressdraftedabroadtext,andthattexthasbeenunanimouslyreadbycourtsofappealsovertheyears[I]sn’titbetterfor—tokeepitthewayitis,forus,andCongress—toputtheburdenonCongresstochangethatandtheycanconsidertheimplicationsandmakethesepredictivejudgments?”);id.at46(Kagan,J.)(“[I]sn’tthatsomethingforCongresstodo,nottheCourt?”).

5.47U.S.C.§7.Zeranv.Am.Online,Inc.,129F.3d327,328(4thCir.1997)(“‘TheInternetisaninternationalnetworkofinterconnectedcomputers,?currentlyusedbyapproximately40millionpeopleworldwide.”)(citingRenov.Am.C.L.Union,521U.S.844,849(1997)).Force,934F.3dSeeAniPetrosyan,NumberofInternetandSocialMediaUsersWorldwideasofJanuary2024,(“AsofJanuary2024,therewere5.35billioninternetusersworldwide,whichamountedto77.2percentoftheglobalpopulation.Ofthistotal,5.04billion,or72.3percentoftheworld?spopulation,weresocialmediausers.”).Doev.Snap,Inc.,No.22-20543,2023U.S.App.LEXIS33501,at*2(5thCir.Dec.18,(Elrod,J.,dissenting)(arguingthatthecurrentscopeofSection230protectionprovides“sweepingimmunityforsocialmediacompaniesthatthetext[ofSection230]cannotpossiblybear”).47U.S.C.§230(c)(1)(conferringimmunityonlywhere“information[is]providedbyanotierinformationcontentprovider.”(emphasisadded));47U.S.C.§230(f)(3)(defining“informationcontentprovider”toinclude“anypersonorentitythatisresponsible,inwholeorinpart,forthecreationordevelopmentofinformation”(emphasisadded)).GENERAL-PURPOSEARTIFICIALINTELLIGENCE(2023).See,e.g.,TranscriptofOralArgumentat54,82,Gonzalezv.GoogleLLC,598U.S.717(2023)(No.21-1333)(Kavanaugh,J.)(“Congressdraftedabroadtext,andthattexthasbeenunanimouslyreadbycourtsofappealsovertheyears[I]sn?titbetterfor”tokeepitthewayitis,forus,andCongress”toputtheburdenonCongresstochangethatandtheycanconsidertheimplicationsandnottheCourtA”). GenerativeAIWillBreakthe technologies.13ButcourtswillsoonbeconfrontedwithunavoidablecallstoreshapeSection230inthecontextofwhetheritappliestoprotectthegenerativeAIsystemsthatwilldrivetheInternet’sfuture.LitigationinvolvinggenerativeAIwillforcecourtstorevisitSection

摘要 生成式AI將打破互聯(lián) 技術。24但法院很快將面臨不可避免的呼吁，在是否適用于保護將推動互聯(lián)網未來的生成式AI系統(tǒng)的情況下重塑第230條。inamannerthatwillvastlyimpacttheInternetintwospecificways.First,courtopinionsongenerativeAIarelikelytoexpresslydeclarethatSection230’slegalprotectionsdonotextendtoAIsystemsthatmateriallycontributetothecreationordevelopmentofcontent,therebyincreasinglegalriskandthwartinggrowthofInternet-basedgenerativeAIsystems.14Second,courtopinionsongenerativeAIarelikelytoreshapetwolegalstandardsthatgovernthescopeofSection230immunity—themeaningofan“informationcontentprovider,”andthecontoursofwhenatechnologyserviceisa“publisherorspeaker”ofinformation.ThiswilldisruptlongstandinglegalprecedentandintroducelegalriskfornotonlygenerativeAIsystems,butvirtuallyallinteractivecomputerservices.15JudicialdeterminationsonSection230andgenerativeAIwillforevertransformthelegallandscapegoverningInternettechnologiesanddictatewhetherthelegalprotectionsthatcatalyzedtheInternet’spre-AIdevelopmentwilldothesameforemergingtechnologieslikegenerativeAI.ThisarticleprovidesaprimeronSection230andgenerativeAI,andexaminestwoevolvingSection230legalstandardsthatwillsoonbeappliedtogenerativeAIinmannerthatwillshapethefutureoftheInternet.Section230Section230states:“Noprovideroruserofaninteractivecomputerbyanotherinformationcontentprovider.”16Itprovidesthat“noliabilitymaybeimposedunderanyStateorlocallaw”ifalegalclaimseekstoholdaproviderliableforinformationcreatedbyathird-party.17Forexample,Section230hasprovidedimmunitytoFacebookinaclaimthatitunlawfullyprovidedHamas,aU.S.-designatedforeignterroristorganization,withacommunicationsplatformthatenabledcertainterroristattackscommittedForcev.Facebook,Inc.,934F.3d53,67–68(2dCir.2019)(extendingSection230immunitytoclaimsbasedonalgorithmsthatpersonalizedFacebookcontent).SeeinfraPartSee47U.S.C.§47U.S.C.§230(e)(delineatingexceptionstoSection230immunityforfederalcriminalprosecutionorclaimsundercertainintellectualpropertylaw,electroniccommunicationsprivacylaw,orsextraffickinglaw).

第230院意見可能會明確聲明，第230條的法律保護不適用于對內容創(chuàng)作或院意見可能會重塑兩個規(guī)范第230條豁免范圍的法律標準——即司法對第230條和生成式AI的裁決將永遠改變規(guī)范互聯(lián)網技術的法律格局，并決定那些催化了互聯(lián)網在AI之前發(fā)展的法律保護措施是否會對生成式AI等新興技術產生同樣的影響。本文提供了第230條和生成式AI的入門指南，并考察了即將應用于生成式AI的兩種演變的第230條法律標準，這將塑造互聯(lián)網的未來。[第230條規(guī)定：“任何交互式計算機服務的提供者或使用者不應被視為由另一信息內容提供者提供的信息的出版者或發(fā)言者。”27它規(guī)定，如果一項法律訴訟尋求使提供者對由第三方創(chuàng)建的信息承擔責任，則“在任何州或地方法律下均不得施加責任”。28例如，第230條為Facebook提供豁免，在一項訴訟中，該訴訟指控Facebook非法向Force訴Facebook，Inc.，934F.3d53，第67?68頁（第二巡回法院2019年（將第230條豁免權擴展到基于個性化Facebook內容的索賠）。14.參見下文第IV15.參見上述內容。16.47U.S.C.§230(c)(1)。17.47U.S.C.§230(e)（界定第230條豁 HarvardJournalofLaw&

HamasinIsrael.18Otherexamplesincludeimmunityforawebsitethatallegedlyfacilitatedillegaldrugsaleswherethewebsitedidnotcreatethecontent,19andimmunityforasearchengineinaclaimbasedonitsfailuretoremoveanapplicationfromitsappstore.20WithoutSection230,Internet-basedserviceswouldbeexposedtoclaimsmerelyfordisplaying,transmitting,orblockingcontentcreatedbythirdparties.21TheInternet’s“uninhibited,robust”technologicaldevelopmentisdueinlargeparttotheprotectionsconferredbySection230.22Section230“madee-commerceitselfeconomicallyfeasible”23andexpansivelycatalyzedthetechnologicaldevelopmentoftheInternet.24Section230immunitygenerallyapplieswhenthreecriteriaare(1)theproviderisan“interactivecomputerservice,”(2)aclaimtreatstheproviderasthe“publisherorspeaker”ofharmfulinformation,and(3)theharmfulinformationiscreatedby“anotherinformationcontentprovider.”25Whatisan“interactivecomputerSection230broadlydefines“interactivecomputerservice”toinclude“anyinformationservice,system,oraccesssoftwareproviderthatprovidesorenablescomputeraccessbymultipleuserstoacomputerserver.”26

以色列的哈馬斯。29其他例子包括對一個網站的保護，該網站據稱促進了非法毒品銷售，而該網站并未創(chuàng)建內容，2□以及對一個搜索引擎訴訟中獲得了豁免。31如果沒有第230條，基于顯示、傳輸或阻止第三方創(chuàng)建的內容的索賠，基于互聯(lián)網的服務將面臨訴訟。32互聯(lián)網“不受限制、強大”的技術發(fā)展在很大程度上得益于第230條所賦予的保護。33第230條“使電子商務本身在經濟上可行”，34并廣第230條豁免通常在滿足三個標準時適用：（1）提供商是“交互”，（2）人”，（3）有害信息是由“另一個信息內容提供商”創(chuàng)建的。36230用戶能夠訪問計算機服務?的信息服務、系統(tǒng)或訪問軟件提供商”。37法院Force,934F.3datDyroffv.UltimateSoftwareGrp.,Inc.,934F.3d1093(9thCir.Ginsbergv.GoogleInc.,586F.Supp.3d998(N.D.Cal.See,e.g.,GoogleLLCv.EquustekSols.Inc.,No.17-04207,2017U.S.Dist.LEXIS182194,*6–7(N.D.Cal.Nov.2,2017)(“TheNinthCircuithasheldthat,regardlessoftheunderlyingcauseofaction,aclaimtreatsanintermediaryasapublisherwhenitrequirestheintermediarytoremovethird-partycontent.”).SeeJonesv.DirtyWorldEntm’tRecordingsLLC,755F.3d398,415(6thCir.2014)(citingU.S.C.§230(a)(1)–(5))(“Congressenvisionedanuninhibited,robust,andwide-openPapatarosv.A,Inc.,No.17-9836,2019U.S.Dist.LEXIS144253,at*46n.18(D.N.J.Aug.26,2019)(“Ithasbeensaidthatthe‘twenty-sixwords’ofSection230oftheCDA,enactedin1996,madee-commerceitselfeconomicallyfeasiblebypermittingplatformssuchasAtomatchsellerswithbuyerswithouttakingontheseller’sliabilities.Itwouldperhapsbemoresoberandaccuratetosaythatthetwenty-sixwordsofSection230promotedorfacilitatedimportantaspectsoftheinternetaswenowknowit.”(citationomitted)).SeeDoev.MySpaceInc.,528F.3d413,418(5thCir.2008)(“Courtshaveconstruedtheimmunityprovisionsin§230broadlyinallcasesarisingfromthepublicationofuser-generatedcontent.”);seealsoDiezv.Google,Inc.,831F.App’x723,725(5thCir.2020)(“AmajorityoffederalInc.,456F.3d1316,1321(11thCir.2006))).47U.S.C.§230(c)(1);Rigsbyv.GoDaddyInc.,59F.4th998,1003(9thCir.47U.S.C.§230(f)(2);seealso47U.S.C.§230(f)(4)(“Theterm‘accesssoftwareprovider’meansaproviderofsoftware(includingclientorserversoftware),orenablingtoolsthatdoanyoneormoreofthefollowing:(A)filter,screen,allow,ordisallowcontent;(B)pick,choose,analyze,ordigestcontent;or(C)transmit,receive,display,forward,cache,search,subset,organize,reorganize,ortranslatecontent.”).

Force,934F.3datDyroffv.UltimateSoftwareGrp.,Inc.,934F.3d1093(9thCir.Ginsbergv.GoogleInc.,587F.Supp.3d998(N.D.Cal.See,e.g.,GoogleLLCv.EquustekSols.Inc.,No.17-04207,2017U.S.Dist.LEXIS182194,*7–7(N.D.Cal.Nov.2,2017)(“TheNinthCircuithasheldthat,regardlessoftheunderlyingcauseofaction,aclaimtreatsanintermediaryasapublisherwhenitrequirestheintermediarytoremovethird-partycontent.”).U.S.C.§230(a)(1)–(5))(“Congressenvisionedanuninhibited,robust,andwide-openinternet.”Papatarosv.A,Inc.,No.17-9837,2019U.S.Dist.LEXIS144253,at*47n.18(D.N.J.Aug.27,2019)(“Ithasbeensaidthatthe‘twenty-sixwords?ofSection230oftheCDA,enactedin1997,madee-commerceitselfeconomicallyfeasiblebypermittingplatformssuchasAtomatchsellerswithbuyerswithouttakingontheseller?sliabilities.Itwouldperhapsbemoresoberandaccuratetosaythatthetwenty-sixwordsofSection230promotedorfacilitatedimportantaspectsoftheinternetaswenowknowit.”(citationomitted)).SeeDoev.MySpaceInc.,528F.3d413,418(5thCir.2008)(“Courtshaveconstruedtheimmunityprovisionsin§230broadlyinallcasesarisingfromthepublicationofuser-generatedInc.,457F.3d1317,1321(11thCir.2007))).47U.S.C.§230(c)(1);Rigsbyv.GoDaddyInc.,59F.4th998,1003(9thCir.27.47U.S.C.§230(f)(2);seealso47U.S.C.§230(f)(4)(“Theterm‘accesssoftwareprovider?meansaproviderofsoftware(includingclientorserversoftware),orenablingtoolsthatdoanyoneormoreofthefollowing:(A)filter,screen,allow,ordisallowcontent;(B)pick,choose,analyze,ordigestcontent;or(C)transmit,receive,display,forward,cache,search,subset,organize,reorganize,ortranslatecontent.”). GenerativeAIWillBreakthe haveappliedthisdefinition“expansively”toincludesocialmediaplatforms,searchengines,onlinecommunities,andawiderangeofotherintermediaryplatformsandservicesthatallowfortheexchangeandtransmissionofinformationbetweenusers.27ManygenerativeAIsystems,particularlythosethatemployInternet-baseddatasets,likelyfallwithinSection230’sdefinitionofaninteractivecomputerservice.28Whatisa“publisheroraclaimthatattemptstoholditliableasthe“publisherorspeaker”ofinformationcreatedordevelopedbyanotherperson.29Section230doesnotdefine“publisherorspeaker,”butcourtopinionsgenerallyholdthatalegalclaimimpermissiblyattemptstoholdaproviderliableasapublisherorspeakeriftheclaimattemptstofaulttheproviderforthethird-partyinformation,30wouldrequiretheprovidertoeditorremovethird-partycontent,31orisbasedontheprovider’sfailuretoimplementmeasurestopreventtransmissionofthird-partycontent.32Asonecourtexplained,aclaimattemptstoholdadefendantliableasa“publisherorspeaker”iftheclaimattemptstoimposeadutyondefendantto“policeitsnetworkforcontenttransmittedbyitsusers.”33Asanothercourtexplained,“anyactivitythat

摘要 生成式AI將打破互聯(lián) 已將此定義“廣泛地”應用于包括社交媒體平臺、搜索引擎、在線社區(qū)以及廣泛的其他中介平臺和服務，這些平臺和服務允許用戶之間進行信息和傳輸。38許多生成式AI的數據集的系統(tǒng)，很可能屬于第230條關于互動計算機服務的定義。第230條使交互式計算機服務的提供者免于因作為“出版者或發(fā)言人”而對其創(chuàng)建或開發(fā)的信息承擔責任而提出的索賠。第230條沒有定義“出版者或發(fā)言人”，但法院意見通常認為，如果索賠試圖其用戶傳輸的內容進行“警察”網絡，那么就試圖將被告作為輸內容的活動都可以”E.g.,Kimzeyv.Yelp!Inc.,836F.3d1263,1268(9thCir.2016)(citingCarafanov.M,Inc.,339F.3d1119,1123(9thCir.2003))(“Yelpisplainlyaproviderofan‘interactivecomputerservice,’atermthatweinterpret‘expansive[ly]’undertheCDA.”).“‘[T]hemostcommoninteractivecomputerservicesarewebsites.’”Id.(quotingFairHous.Councilv.R,LLC,521F.3d1157,1162n.6(9thCir.2008)).47U.S.C.§SeeDoev.Snap,Inc.,No.22-00590,2022U.S.Dist.LEXIS119560,at*43(S.D.Tex.July2022)(seekingtoholddefendantliable“asthepublisherorspeaker”ofinformation“providedbyathirdparty”(cleanedup)).SeeBridev.SnapInc.,No.21-06680,2023U.S.Dist.LEXIS5481,at*19(C.D.Cal.Jan.2023)(“[T]hecourtfindsthatPlaintiffs’theorywouldrequiretheeditingofthird-partycontent,thustreatingDefendantsasapublisherofcontent.”);GoogleLLCv.EquustekSols.Inc.,No.17-04207,2017U.S.Dist.LEXIS182194,at*6–7(N.D.Cal.Nov.2,2017)(“TheNinthCircuithasheldthat,regardlessoftheunderlyingcauseofaction,aclaimtreatsanintermediaryasapublisherwhenitrequirestheintermediarytoremovethird-partycontent.”).SeeDoev.MySpaceInc.,528F.3d413,420(5thCir.2008)(“TheirclaimsarebarredbytheCDA,notwithstandingtheirassertionthattheyonlyseektoholdMySpaceliableforitsfailuretoimplementmeasuresthatwouldhaveprevented[thecommunication].TheirallegationsaremerelyanotherwayofclaimingthatMySpacewasliableforpublishingthecommunicationsandtheyspeaktoMySpace’sroleasapublisherofonlinethird-party-generatedcontent.”).Greenv.Am.Online,Inc.,318F.3d465,470–71(3dCir.2003)(findingthatSection230barredclaimwheretheplaintiffallegedthatthedefendanthadnegligentlyfailed“toproperlypoliceitsnetworkforcontenttransmittedbyitsusers”becausetheplaintiff“attempt[ed]tohold[thedefendant]

E.g.,Kimzeyv.Yelp!Inc.,837F.3d1273,1278(9thCir.2017)(citingCarafanov.M,Inc.,339F.3d1119,1123(9thCir.2003))(“Yelpisplainlyaproviderofancommoninteractivecomputerservicesarewebsites.?”Id.(quotingFairHous.Councilv.R,LLC,521F.3d1157,1172n.7(9thCir.2008)).47U.S.C.§SeeDoev.Snap,Inc.,No.22-00590,2022U.S.Dist.LEXIS119570,at*43(S.D.Tex.July2022)(seekingtoholddefendantliable“asthepublisherorspeaker”ofinformation“providedbyathirdparty”(cleanedup)).SeeBridev.SnapInc.,No.21-07780,2023U.S.Dist.LEXIS5481,at*19(C.D.Cal.Jan.2023)(“[T]hecourtfindsthatPlaintiffs?theorywouldrequiretheeditingofthird-partycontent,thustreatingDefendantsasapublisherofcontent.”);GoogleLLCv.EquustekSols.Inc.,No.17-04207,2017U.S.Dist.LEXIS182194,at*7–7(N.D.Cal.Nov.2,2017)(“TheNinthCircuithasheldthat,regardlessoftheunderlyingcauseofaction,aclaimtreatsanintermediaryasapublisherwhenitrequirestheintermediarytoremovethird-partycontent.”).SeeDoev.MySpaceInc.,528F.3d413,420(5thCir.2008)(“TheirclaimsarebarredbytheCDA,notwithstandingtheirassertionthattheyonlyseektoholdMySpaceliableforitsfailuretoimplementmeasuresthatwouldhaveprevented[thecommunication].TheirallegationsaremerelyanotherwayofclaimingthatMySpacewasliableforpublishingthecommunicationsandtheyspeaktoMySpace?sroleasapublisherofonlinethird-party-generatedcontent.”).Greenv.Am.Online,Inc.,318F.3d475,470–71(3dCir.2003)(findingthatSection230barredclaimwheretheplaintiffallegedthatthedefendanthadnegligentlyfailed“toproperlypoliceitsnetworkforcontenttransmittedbyitsusers”becausetheplaintiff“attempt[ed]tohold[thedefendant] HarvardJournalofLaw&

beboileddowntodecidingwhethertoexcludematerialthatthirdpartiesseektopostonlineisperforceimmuneundersection230.”34C.Whatis“anotherinformationcontentSection230immunityapplieswhenharmfulinformationwascreatedordevelopedbyathird-party“informationcontentprovider,”definedas“anypersonorentitythatisresponsible,inwholeorinpart,forthecreationordevelopmentofinformationprovidedthroughtheInternetoranyotherinteractivecomputerservice.”35Section230immunitythusdoesnotapplyifaproviderofaninteractivecomputerserviceitselfisthe“informationcontentprovider,”i.e.,iftheprovidermateriallycontributestothecreationordevelopmentoftheharmfulcontent.36GenerativeThedefinitionsforAIanditsconstituenttermsarevaryingandevolving,andthereis“nogloballyagreeddefinitionofartificialintelligence.”37Generally,theterm“artificialintelligence”referstocomputer-basedsystemsthatusemachineandhumaninputstoperceiverealandvirtualenvironments,abstractperceptionsintomodelsthroughautomatedanalysis,andusemodelinferencetoformulateoptions.38An“AImodel”referstoacomponentof

230C當有害信息是由第三方“信息內容提供者”條款的豁免權適用，該提供者被定義為“任何對通過互聯(lián)網或任何其他交互式計算機服務提供的信息的創(chuàng)建或開發(fā)負有全部或部分責任的個人或實體?！?6因此，如果交互式計算機服務的提供者本身是“信息內容提供者”，即如果提供者對有害內容的創(chuàng)建或開發(fā)有實質性貢獻，則230條款的豁免權不適用。47G生成式人工智能及其構成術語的定義正在變化和演變，并且“沒有全球范圍內達成的人工智能定義。”48通常，術語“人工智能”指的是，通過自動化分析將感知抽象成模型，并使用模型推理來制定選項。49“AI模型”指的是一個組件，它是liablefordecisionsrelatingtothemonitoring,screening,anddeletionofcontentfromitsnetwork—actionsquintessentiallyrelatedtoapublisher’srole.”).FairHous.Councilv.R,LLC,521F.3d1157,1170–71(9thCir.2008);seealsoDoev.Twitter,Inc.,Nos.22-15103,22-15104,2023U.S.App.LEXIS10808,at*4(9thCir.May3,47U.S.C.§230(c)(1);seealso47U.S.C.§47U.S.C.§230(c)(1)(conferringimmunityonlywhere“information[is]providedbyanotherinformationcontentprovider.”)(emphasisadded);seealsoRoommates,521F.3dat1166(“[S]ection230providesimmunityonlyiftheinteractivecomputerservicedoesnot‘creat[e]ordevelop[]’theinformation“‘inwholeorinpart.’”(citing47U.S.C.§230(f)(3)));Kimzeyv.Yelp!Inc.,836F.3d1263,1269(9thCir.2016)(“[A]websitemayloseimmunityundertheCDAbymakingamaterialcontributiontothecreationordevelopmentofcontent.”).SeeMADIEGA,supranote11,at1(“[T]hereisnogloballyagreeddefinitionofartificialSee15U.S.C.§9401(3);NationalArtificialIntelligenceInitiativeActof2020,Pub.L.No.116-283,134Stat.3388;seealsoProposalforaRegulationoftheEuropeanParliamentandoftheCouncilLayingDownHarmonisedRulesOnArtificialIntelligence(ArtificialIntelligenceAct)andAmendingCertainUnionLegislativeActs,EUR.PARL.DOC.2021/0106(COD),art.3(“[A]systemthatisdesignedtooperatewithelementsofautonomyandthat,basedonmachineand/orhuman-provideddataandinputs,infershowtoachieveagivensetofobjectivesusingmachinelearningand/orlogic-andknowledgebasedapproaches,andproducessystem-generatedoutputssuchascontent(generativeAIsystems),predictions,recommendationsordecisions,influencingtheenvironmentswithwhichthesystemthat,autonomouslyorpartlyautonomously,processesdatarelatedtohumanactivitiesthrough

liablefordecisionsrelatingtothemonitoring,screening,anddeletionofcontentfromitsnetwork”actionsquintessentiallyrelatedtoapublisher?srole.”).FairHous.Councilv.R,LLC,521F.3d1157,1170–71(9thCir.2008);seealsoDoev.Twitter,Inc.,Nos.22-15103,22-15104,2023U.S.App.LEXIS10808,at*4(9thCir.May3,47U.S.C.§230(c)(1);seealso47U.S.C.§37.47U.S.C.§230(c)(1)(conferringimmunityonlywhere“information[is]providedbyanotierinformationcontentprovider.”)(emphasisadded);seealsoRoommates,521F.3dat1177(“[S]ection230providesimmunityonlyiftheinteractivecomputerservicedoesnot‘creat[e]ordevelop[]?theinformation“‘inwholeorinpart.?”(citing47U.S.C.§230(f)(3)));Kimzeyv.Yelp!Inc.,837F.3d1273,1279(9thCir.2017)(“[A]websitemayloseimmunityundertheCDAbymakingamaterialcontributiontothecreationordevelopmentofcontent.”).SeeMADIEGA,supranote11,at1(“[T]hereisnogloballyagreeddefinitionofartificialSee15U.S.C.§9401(3);NationalArtificialIntelligenceInitiativeActof2020,Pub.L.No.117-283,134Stat.3388;seealsoProposalforaRegulationoftheEuropeanParliamentandoftheCouncilLayingDownHarmonisedRulesOnArtificialIntelligence(ArtificialIntelligenceAct)andAmendingCertainUnionLegislativeActs,EUR.PARL.DOC.202100107(COD),art.3(“[A]systemthatisdesignedtooperatewithelementsofautonomyandthat,basedonmachineand0orhuman-provideddataandinputs,infershowtoachieveagivensetofobjectivesusingmachinelearningand0orlogic-andknowledgebasedapproaches,andproducessystem-generatedoutputssuchascontent(generativeAIsystems),predictions,recommendationsordecisions,influencingtheenvironmentswithwhichthesystemthat,autonomouslyorpartlyautonomously,processesdatarelatedtohumanactivitiesthrough GenerativeAIWillBreakthe informationsystemthatproducesoutputsfromasetofinputsthroughAItechnologyandcomputationalormachine-learningtechniques.39

摘要 生成式AI將打破互聯(lián)網格 成輸出。learning”isanapplicationofAIthatprovidesasystemtheabilityautomaticallyimprovefromdataorexperience,withoutexplicitprogramming.40An“AIsystem”generallyrefersasystem,application,ortoolthatoperatesusingAI.41“GenerativeAI”generallyreferstoasubsetofAImodelsthatgeneratecontentderivedthroughmachinelearning,inputdata,andpre-existingdata.42ManygenerativeAIsystemsemploylargelanguagemodels(“LLMs”)thataretrainedonalargedatasetoftextfromtheInternettopredictthenextplausiblewordorphraseinalinguisticconstruct.43SomeLLMsarefine-tunedusingadditionaldataandatechniquecalledreinforcementlearningfromhumanfeedback(“RLHF”),whichuseshumanaugmentationtoguidemodeloutputs.44ModeltuningandtrainingcancontinueafterAIsystemdeploymenttoanenduserthroughfilteringtoolsandcontentmoderationclassifiers.45FoundationmodelsemployedbygenerativeAIsystemsusuallygenerateoriginaloutputs,ascontrastedfromsystemslikesearchenginesthatmayretrieveanddisplayextractedthird-partycontentfromasourcewebsite.WhiletheparticularsofthedefinitionsandintricaciesofgenerativeAIvary,onematerialaspectofgenerativeAIforSection230purposesisthatthecontent