IMF-人工智能的经济影响与调控——学术文献与政策行动综述（英）-2024.3_市场营销策划_202.docx

《IMF-人工智能的经济影响与调控——学术文献与政策行动综述（英）-2024.3_市场营销策划_202.docx》由会员分享，可在线阅读，更多相关《IMF-人工智能的经济影响与调控——学术文献与政策行动综述（英）-2024.3_市场营销策划_202.docx（55页珍藏版）》请在优知文库上搜索。

1、1 IntroductionThisreviewpaperinvestigateshowArtificialIntelligence(AI)affectstheeconomyandhowthetechnologyhasbeenregulated,relyingonacademicandpolicysourcesthroughearly2024.Wecoverinsightsonemploymentandwageeffects,productivity,andeconomicgrowthfromtheeconomicliterature.Inthepolicyrealm,wesummarizet

2、heregulatoryactionsundertakenindifferentregions,detailingtheirrationales,approachesandareasofcoverage.GiventherapidevolutionofAItechnologiesandtherelatedliterature,thepaperaimstoprovideastructuretoorganizethelatestcontributionsfortheuseofpolicymakers,economists,researchers,andindustrystakeholders.Be

3、foredelvingdeeperintothecontentofourpaper,weshallclarifyitsscopewithsomekeydefinitions.ProfessorJohnMcCarthy,oneoftheorganizersofthe1956DartmouthresearchprojectthatstartedAIasafield,definedAIasthescienceandengineeringofmakingintelligentmachines(McCarthy2007).TheoriginaldocumentcoiningthetermAIwasthe

4、1955proposalfortheDarthmouthResearchProjectcoauthoredbyMcCarthy,Minsky,RochesterandShannon,seeInthisreview,weconcernourselveswiththeeconomicimpactandregulationofrecentadvancesinAI,suchasmachinelearning(M1.),anditssub-fieldsofdeeplearning,generative-Al(gen-AI)and1.arge1.anguageModels(1.1.Ms).M1.studi

5、eshowmputeragentscanimprovetheirperception,knowledge,thinking,oractionsbasedonexperienceordata.1.1.MsarebroadlydefinedasneuralnetworksthatIeamsntextandmeaningbytrackingrelationshipsinsequentialdata.SeeBroadly,wecoverempiricalstudiesfocusingoneitherM1.applicationsexcludinggen-AI(44pre-gen-A,untillate

6、2022)oronthelatestgen-AIand1.1.Ms(post-gen-AI,”mostlysince2023).Thisdistinctionismotivatedbythefactthat1.1.Msandgen-AIhavecometotheforeoftheeconomicandpolicydebateafterthereleaseofDall-E2andChatGPTbyOpenAIinlate2022,andrelateddata,applicationsandresearchremainrelativelyscarce.Accordingly,werefertoM1

7、.asmachinelearningordeeplearningforprediction,imageandpatternrecognition,textanalysisanddataanalysis.WhenusingthetermM1.,weexcludegen-AI,whichistreatedseparatelyduetothewideavailabilityoftext-andimage-generating1.1.Mtools.WewillclarifythetypeofAIcoveredbyeachpaperwhenthecontextdoesnotmakeitclear.Con

8、sistentwiththisdiscussion,wewilluse“M1.todenoteresearchthatappliestomachinelearningandexcludesgen-AI,usuallyforlackofavailabledate.Weinsteaddenotestudiesas“gen-AI“iftheyfocusexclusivelyongen-AI.WhenweusethetermAI,werefertocontextsencompassingbothM1.andgen-AItechnologies.Thisdistinctionismostlyreleva

9、nttoempiricalorexperimentalpapers,sincetheoreticalstudiesoftendonotdistinguishgen-AIfromM1.Inthecaseofregulation,thereisnoglobalconsensusonadefinitionof“AIorthetechnologiescoveredbyAIregulation.SeeOShaughnessy(2022)andalsoAppendixA.lWestartbydescribingIheimpactofAIonlabormarkets,wheretheliteratureha

10、sfocusedonemploymentandwagesofvariousoccupationalgroups.MostofthestudieswereviewadoptimplicitlyorexplicitlythetaskframeworksofZeira(1998),AcemogluandAutor(2011),andAcemogluandRestrepo(2018)thathavebeenappliedtostudyhuman-robotsubstitutionintheautomationliterature.Aswebrieflydiscuss,theseframeworksmo

11、deloutputasabundleoftaskscarriedoutbyeitherworkersorcapital,andobtainthatemploymentandwagesofdifferentoccupationalordemographicgroupsaredirectlyrelatedtothequantityoftasksassignedtoeachgroup.Accordingly,theearlierpre-gen-AIliteratureandmuchofthepost-gen-AIliteraturefocusedoncomputing“taskexposures,-

12、theshareoftasksthatcanpotentiallybereplacedbyAI-togiveestimatesofthepotentialimpactofAIongroupsofworkers.Followingtheearlierautomationliterature,theseworksgenerallyassociateahighertaskexposurewithlargerpotentialdisplacementforaffectedgroupsofworkers.Moststudiesagreethatwhite-collar,higher-skilledocc

13、upationshavehighertaskexposuresandthereforefacestrongeremploymentriskfromAIadoption(e.g.,ongen-AI,Eloundouetal.(2023).Otherresearchersinsteadseparatedtaskexposureandemploymentrisk,highlightingtheaugmentationpotentialofAItechnologiesorother“shieldingfactors”(e.g.,Cazzanigaetal.(2024).Empiricalstudies

14、onM1.showthatemploymenteffectsmightoverallbenilorpositive(Acemoglu,Autor,Hazell,etal.2022;Albanesietal.2023),whileexperimentalpapersongen-AIhighlightproductivitygainsforlow-skilledworkersthatcouldharborinwagecompression(NoyandW.Zhang2023;Brynjolfsson,1.i,andRaymond2023).Ultimately,theorysuggeststhat

15、thefinalverdictonlabormarketimpactsofAIwilldependontheracebetweenjobdisplacementandproductivityincreases,resultingfromdirectworkercomplementationoreconomy-widegainsfromAI.Thefindingsthatwepresentprovidesomeelementstoevaluatethistrade-off,butleaveustarfromadefinitiveunderstanding.Inparticular,allstud

16、iesseemtoagreethatexposureispervasive,butremaininconclusiveonhowsuchexposuremaytranslateinsubstitutionofcomplementationofworkers.WecloseourdiscussionoflabormarketimpactsofAIwithadiscussionofpotentialpolicytotacklethepotentiallabordisplacementbroughtaboutbythistechnology.Next,weproceedtosurveythemorelimitedstudiesconcerningproductivityandgrowtheffectsofAI,whichencompasstheory,firm-levelstudiesonM1.,andtheexperimentalevidenceongenAIcitedabove.Theworkinth