生成式人工智能的经济潜力（英）

iiTheeconomicpotentialofgenerativeAI:Thenextproductivityfrontier Contents Keyinsights 3 Chapter1:GenerativeAIasatechnologycatalyst4 Glossary 6 Chapter2:GenerativeAIusecasesacrossfunctionsandindustries 8 Spotlight:Retailandconsumerpackagedgoods27 Spotlight:Banking 28 Spotlight:Pharmaceuticalsandmedicalproducts 30 Chapter3:ThegenerativeAIfutureofwork:Impactsonworkactivities,economicgrowth,andproductivity 32 Chapter4:Considerationsforbusinessesandsociety48 Appendix 53 TheeconomicpotentialofgenerativeAI:Thenextproductivityfrontier1 2TheeconomicpotentialofgenerativeAI:Thenextproductivityfrontier Keyinsights e1.neGrativeAI’simpactonproductivitycouldaddtrillionsofdollarsinvaluetotheglobaleconomy.Ourlatestresearch estimatesthatgenerativeAIcouldaddtheequivalentof$2.6trillionto$4.4trillionannuallyacrossthe63usecasesweanalyzed—bycomparison,theUnitedKingdom’sentireGDPin2021was$3.1trillion.Thiswouldincreasetheimpactofallartificialintelligenceby15to 40percent.ThisestimatewouldroughlydoubleifweincludetheimpactofembeddinggenerativeAIintosoftwarethatiscurrentlyusedforothertasksbeyondthoseusecases. 2.About75percentofthevaluethatgenerativeAIusecasescoulddeliverfallsacrossfourareas:Customeroperations,marketingandsales,softwareengineering,andR&D.Across16businessfunctions,weexamined63usecasesinwhichthetechnology canaddressspecificbusinesschallengesinwaysthatproduceoneormoremeasurableoutcomes.ExamplesincludegenerativeAI’sabilitytosupportinteractions withcustomers,generatecreativecontentformarketingandsales,anddraftcomputercodebasedonnatural-languageprompts,amongmanyothertasks. 3.GenerativeAIwillhaveasignificantimpactacrossallindustrysectors. Banking,hightech,andlifesciencesareamongtheindustriesthatcouldseethebiggestimpactasapercentageoftheirrevenuesfromgenerativeAI.Acrossthebankingindustry,forexample,thetechnologycoulddelivervalue equaltoanadditional$200billionto$340billionannuallyiftheusecaseswerefullyimplemented.Inretailandconsumerpackagedgoods,thepotentialimpactisalsosignificantat$400billionto$660billionayear. 4.GenerativeAIhasthepotentialtochangetheanatomyofwork,augmentingthecapabilitiesofindividualworkersbyautomatingsomeoftheirindividualactivities. CurrentgenerativeAIandothertechnologieshavethepotentialtoautomateworkactivitiesthatabsorb60to70percentofemployees’timetoday.Incontrast,wepreviouslyestimatedthattechnologyhasthepotentialtoautomatehalfofthetimeemployeesspendworking.1 TheaccelerationinthepotentialfortechnicalautomationislargelyduetogenerativeAI’sincreasedabilitytounderstandnaturallanguage,whichisrequiredforworkactivitiesthataccountfor25percentoftotalworktime.Thus,generativeAIhasmoreimpactonknowledgeworkassociatedwithoccupationsthathavehigherwagesandeducationalrequirementsthanonothertypesofwork. 5.Thepaceofworkforcetransformationislikelytoaccelerate,givenincreasesinthepotentialfortechnicalautomation.Ourupdatedadoptionscenarios,includingtechnologydevelopment,economicfeasibility,anddiffusiontimelines,leadtoestimatesthathalfoftoday’sworkactivitiescouldbeautomatedbetween2030and2060,withamidpointin2045,orroughlyadecadeearlierthaninourpreviousestimates. 6.GenerativeAIcansubstantiallyincreaselaborproductivityacrosstheeconomy,butthatwillrequireinvestmentstosupportworkersastheyshiftworkactivitiesorchangejobs.GenerativeAIcouldenablelaborproductivitygrowthof0.1to0.6percentannuallythrough2040,dependingontherateoftechnologyadoptionandredeploymentofworkertime intootheractivities.CombininggenerativeAIwithallothertechnologies,workautomationcouldadd0.2to3.3percentagepointsannuallytoproductivitygrowth.However,workerswillneedsupportinlearningnewskills,andsomewillchangeoccupations.Ifworkertransitionsandotherriskscanbemanaged,generativeAIcouldcontributesubstantivelytoeconomicgrowthandsupportamoresustainable,inclusiveworld. 7.TheeraofgenerativeAIisjustbeginning.Excitementoverthistechnologyispalpable,andearlypilotsarecompelling.Butafullrealizationofthetechnology’sbenefitswilltaketime,andleadersinbusinessandsocietystill haveconsiderablechallengestoaddress.TheseincludemanagingtherisksinherentingenerativeAI,determiningwhatnewskillsandcapabilitiestheworkforcewill need,andrethinkingcorebusinessprocessessuchasretraininganddevelopingnewskills. TheeconomicpotentialofgenerativeAI:Thenextproductivityfrontier3 1 GenerativeAIasatechnologycatalyst TograspwhatliesaheadrequiresanunderstandingofthebreakthroughsthathaveenabledtheriseofgenerativeAI,whichweredecadesinthemaking.ChatGPT,GitHubCopilot,StableDiffusion,andothergenerativeAItoolsthathavecapturedcurrentpublicattentionaretheresultofsignificantlevelsofinvestmentinrecentyearsthathavehelpedadvancemachinelearninganddeeplearning.ThisinvestmentundergirdstheAIapplicationsembeddedinmanyoftheproductsandservicesweuseeveryday. ButbecauseAIhaspermeatedourlivesincrementally—througheverythingfromthetechpoweringoursmartphonestoautonomous-drivingfeaturesoncarstothetoolsretailersusetosurpriseanddelightconsumers—itsprogresswasalmostimperceptible.Clearmilestones,suchaswhenAlphaGo,anAI-basedprogramdevelopedbyDeepMind,