CFDI世代人工智能如何改变全球南方的IT服务业（英）

HowGenerativeAIIsChangingtheGlobalSouth’sITServicesSector ByJulianJacobs|June10,2024 Theemergenceoflargelanguagemodels(LLMs)haskindleddiscussionsaboutthepotentialforartificialintelligence(AI)toincreaseproductivityandboosteconomicgrowth.AlthoughtheeconomiceffectsofAIhavebeenatopicofdebateforthelastdecade,thegrowingadoptionofLLMscreatesanewurgencyforunderstandingthetechnology’simpact. Todate,mostresearchontheeconomiceffectsofAIhascenteredalmostentirelyontheGlobalNorth,yetthediffusionofLLMsaroundtheworldeconomymayhavesignificantimplications—bothauspiciousandforeboding—fortheGlobalSouth.1Ononehand,theproductivity-enhancingbenefitsofLLMscouldgreatlybenefitindustriesintheGlobalSouth.Ontheotherhand,unevenadoptionofLLMscouldexacerbatethedigitaldivideandleadtoanunevendistributionofthetechnology’sbenefits. PreliminaryevidencesuggeststhatLLMsareimprovingproductivityacrossavarietyofwhite-collartasks,includingwriting,coding,andcustomerservice.SuchgainshavebeenparticularlyconcentratedintheITsector.2ITserviceshaveplayedagrowingroleinthedevelopingeconomiesoftheGlobalSouth,manyofwhichexportkeydigitalservicestothewealthierGlobalNorth.Thisincludestelecommunicationservices,copywriting,gigwork,andotherformsofcontentgeneration.TheefficiencygainsusheredinbyLLMsmaybedisruptivetothegrowthoftheseITsectors,reshapingthebalanceandflowofinternationalITexports. ThisreportinvestigatesthepotentialimpactofLLMsontheGlobalSouth’sITsector.ItbeginsbycontextualizingthegrowingroleofITinmajorGlobalSoutheconomies,andthenhighlightsthekindsofIToccupationsthataremostrepresentedinGlobalSouthcountries.ItconcludeswithadiscussionregardingtheimpactofLLMsontheglobalflowofITservicesandanassessmentofcurrentpolicyresponsestoAIbyGlobalSouthcountries. ThisreportfindsthatgrowthintheITsector’sshareofGlobalSouthemploymentandexportsislikelytobedisruptedduetotheinfusionofLLMsacrosstheworldeconomy.AlthoughGlobalSouthcountriesexperiencecomparativelylowlevelsofLLMexposureduetosmallersharesofITservices,theITservicesthataremostrepresentedinGlobalSouthexports,growth,andemploymenttendtobeoneswithhighlevelsofautomationpotential.Inotherwords,theITservicesintheGlobalSouthappearmorelikelytoexperiencethedisplacing,asopposedtocomplementing,effectsofAI.GiventhepotentialforcountriestoreshoreandautomatepreviouslyoutsourcedIToccupations,theGlobalSouth’sITservicesappearvulnerabletoLLMadoption.ManycountriesareawareoftheserisksandaremovingrapidlytopromotereskillinganddevelopdiversifiedandmoreadvancedITsectors.Yet,existingpolicyresponsesmaybeinsufficienttoaddresstheserisksintheGlobalSouth. Toaddressthesechallenges,policymakersshouldtakethreekeysteps: 1.PolicymakersintheGlobalSouthshouldsupportworkforcedevelopmentpoliciesthatprovideworkerswiththenewdigitalskillstheywillneedfortheAIeconomy.AIwillreducedemandforsomedigitalservicesbutincreasedemandforothers,andeconomiesintheGlobalSouthwillneedtoadjust.3 2.PolicymakersinthesecountriesshouldpursuewidespreadadoptionofAItoboostproductivityandcompetitivenessacrosstheireconomiesanddevelopdomesticAIimplementationskillsandcapabilities. 3.Policymakersshouldcontinuetopursuepoliciesthatfacilitatedigitalfreetrade,suchasbyopposingrestrictionsoncross-borderdataflows,toensurethattheircompaniesandworkershaveaccesstobest-in-classdigitalservices. WHYSTUDYLARGELANGUAGEMODELS? TherearetwokeyreasonstostudyLLM’seconomiceffects.ThefirstisthatLLMshaveapplicationsacrossthelabormarket,augmentingorautomatingskillsetsacrossadiversityofoccupations.ThismakesLLMsusefulforcomparativeinternationalstudieswhereindustryconcentrationsareheterogeneous.Second,LLMsappeartobecreatingobservableincreasesinlaborproductivitywithintheglobalITsector.PreviousattemptstostudyAI’seconomicimpactshaveeitherbeenforward-lookingorindustry-specific(e.g.,robotexposureinmanufacturing). RecentstudiesprovideestimatesofLLMexposure,automationpotential,andcomplementaritybasedonreal-worlddata,asopposedtospeculation.OnepapershowsthatChatGPTreducedtheaveragetimetakentocompletewritingtasksby40percentwhileincreasingoutputby18percent.4Similarproductivityeffectshavebeenobservedinotherformsof contentgeneration,includingcodingandadvertising.5Inaddition,LLMssupportimprovementsforcallcenteroperators,copywriters,andcontentgeneratorsmorebroadly.Forexample,onepaperfindsthatLLMsimprovedtheworkerproductivityofcustomerserviceagentsby14percent.6ManyoftheseIToccupationsareover-representedintheGlobalSouthandcompriseasignificantshareofthosecountries’ITexports. Thisreport’sfindings—abouttheGlobalSouth’sITsectorLLMexposure—relyonleveragingpublicITdatafromtheInternationalMonetaryFund(IMF),theInternationalLaborOrganization(ILO),andtheOrganizationforEconomicCooperationandDevelopment(OECD).WelookattherepresentationofITsectoroccupationcategorieswithintheGlobalSouththatarehighlysusceptibletoLLM-enabledautomation,definedastheweightedshareoftaskswithinoccupationalcategoriesthataresusceptibletoLLMsubstitution. OnelimitationofthisapproachistherelianceonGlobalNorthevidenceforLLMoccupationalexposure.Itispossible,forexample,thatcertaino