PublicDisclosureAuthorizedPublicDisclosureAuthorized

PolicyResearchWorkingPaper10915

GenerativeAI

CatalystforGrowthorHarbingerofPremature

De-Professionalization?

YanLiu

WORLDBANKGROUP

DigitalDevelopmentGlobalPracticeSeptember2024

PolicyResearchWorkingPaper10915

Abstract

Thispaperpresentsamulti-sectorgrowthmodeltoeluci-datethegeneralequilibriumeffectsofgenerativeartificialintelligenceoneconomicgrowth,structuraltransformation,andinternationalproductionspecialization.Usingparam-etersfromtheliterature,thepaperemployssimulationstoquantifytheimpactsofartificialintelligenceacrossvari-ousscenarios.Thepaperintroducesacrucialdistinctionbetweenhigh-skill,highlydigitalized,tradableservicesandlow-skill,lessdigitalized,less-tradableservices.Themodel’skeypropositionsalignwithempiricalevidence,andthesimulationsyieldnovelandsoberingpredictions.Unlessartificialintelligenceachieveswidespreadcross-sectoradoptionandcatalyzesparadigm-shiftinginnovationsthatfundamentallyreshapeconsumerpreferences,itsgrowth

benefitsmaybelimited.Conversely,itsdisruptiveimpactonlabormarketscouldbeprofound.Thispaperhighlightstheriskof“prematurede-professionalization”,wherearti-ficialintelligencelikelyshrinksthespaceforcountriestogeneratewell-paidjobsinhigh-skillservices.Theanalysisportendsthatdevelopingcountriesfailingtoadoptarti-ficialintelligenceswiftlyriskentrapmentascommodityexporters,potentiallyfacingmassiveyouthunderemploy-ment,diminishingsocialmobility,andstagnatingorevendeclininglivingstandards.Thepaperalsodiscussesartificialintelligence’sbroaderimplicationsoninequality,exploringmultiplechannelsthroughwhichitmayexacerbateormit-igateeconomicdisparities.

ThispaperisaproductoftheDigitalDevelopmentGlobalPractice.ItispartofalargereffortbytheWorldBanktoprovideopenaccesstoitsresearchandmakeacontributiontodevelopmentpolicydiscussionsaroundtheworld.PolicyResearchWorkingPapersarealsopostedontheWebat

/prwp

.Theauthormaybecontactedatyanliu@.

ThePolicyResearchWorkingPaperSeriesdisseminatesthefindingsofworkinprogresstoencouragetheexchangeofideasaboutdevelopmentissues.Anobjectiveoftheseriesistogetthefindingsoutquickly,evenifthepresentationsarelessthanfullypolished.Thepaperscarrythenamesoftheauthorsandshouldbecitedaccordingly.Thefindings,interpretations,andconclusionsexpressedinthispaperareentirelythoseoftheauthors.TheydonotnecessarilyrepresenttheviewsoftheInternationalBankforReconstructionandDevelopment/WorldBankanditsaffiliatedorganizations,orthoseoftheExecutiveDirectorsoftheWorldBankorthegovernmentstheyrepresent.

ProducedbytheResearchSupportTeam

GenerativeAI:CatalystforGrowthorHarbingerofPremature

De-Professionalization?

YanLiu*

WorldBank

Keywords:ArtificialIntelligence,StructuralTransformation,Trade,EconomicGrowth,InequalityJELCodes:O14,F16,F63

*Email:yanliu@.IamgratefulforexcellentcommentsfromStephaneStraub,EstefaniaBelenVergaraCobos,andHeWang.Allremainingerrorsaremyown.

1

1Introduction

ArtificialIntelligence(AI),particularlyrecentadvancesingenerativeAI,hassparkedintensedebateaboutitspotentialimpactoneconomicgrowth,labormarkets,andglobaltradepatterns.EstimatesandprojectionsofAI’seconomiceffectsvarywidely.GoldmanSachs(2023)predictsthatgenerativeAIcouldincreaseglobalGDPby7%,equivalenttoUS$7trillionoveradecade.Acemoglu2024providesamoreconservativeestimate,predictingaGDPincreaseofonly0.9%to1.1%overthenextdecade.D.Autor2024arguesthatgenerativeAIoffersauniqueopportunitytoexpandthemiddleclassbyenablingmiddle-skillworkerstoperformhigher-stakedecision-makingtaskscurrentlyreservedforeliteexperts.Incontrast,FreyandOsborne2024suggeststhatbyloweringbarrierstoentryincognitiveoccupations,generativeAIwillincreasecompetition,eventuallydrivingdownwagesandcausingsignificantlabormarketdisruptions.IfgenerativeAItoolscommodifyexpertiseandreducethereturnstospecializedskills,individualsmaybelessincentivizedtoacquireadvancedexpertise,leadingtofurtherdownsidestoproductivityandwages(Capraroetal.2024).

WhilediscourseonAI’seconomicimpactpredominantlyfocusesonadvancedeconomies,itsinfluenceondevelopingcountriesandglobalproductionpatternsremainsuncertainandprofound.KorinekandStiglitz2021warnthatAIcouldworsenthetermsoftradefordevelopingcountriesbyerodingtheirlabor-costadvantage,potentiallyleadingtofurtherimpoverishment.Conversely,AImightreducegeographicalandlanguagebarriers,shrinkhumancapitalgaps,andincreaseoutsourcingofcognitivetaskstothesenations,potentiallydiminishingincomedisparitieswithwealthiercountries.However,thisoptimisticoutcomehingesonrapidAIadoptionindevelopingcountries—aprospecthamperedbylowlaborcostsandsignificantbarrierssuchaspoorinfrastructure,education,andregulatoryframeworks.

GivenAI’svariedimpactsacrossindustriesandcountries,anuancedanalysisisrequired.ThispaperexaminesgenerativeAI’spotentialeffectsoneconomicgrowth,labormarkets,andinequalitythroughthelensofstructuraltransformationandglobalproductionspecialization,aimingtoofferinsightsintohowAImayreshapeeconomiesandsocietiesworldwide.

Structuraltransformation—thereallocationofeconomicactivitiesacrosssectors—isintegraltoeco-nomicgrowth.Decadesago,economistsnotedadeclineintheshareofagricultureinbothemploymentandoutput,atemporaryriseinmanufacturing,andalong-termshifttowardsservices(Kuznets1957). Multi-sectorgrowthmodelsofferuniqueadvantagesforstudyingAI’simpact.Thedominantgrowthmodelsineconomicshavehistoricallybeensingle-sectorconstructs(Solow1956;Romer1986;LucasJr1988;AghionandHowitt1992;GrossmanandHelpman1991).Thesemodelshaveprovedpowerful

2

andappealing,thoughtheyarenotwithoutlimitations.Suchmodelsemphasizedtheimportanceofcapital,labor,education,andinnovationindrivinggrowth,butoverlookthedifferencesandintricateinter-linkagesbetweensectors.Consequently,single-sectormodelsfallshortinelucidatinghowanecon-omy’scompositionalstructurerelatestoitsgrowthtrajectory.Theeconomiccomplexitytheory(HidalgoandHausmann2009)showsthatacountry’sindustrialcompositionandspecializationpatternsstronglyinfluenceitsgrowthprospects.Understandingthecomplexinterplaybetweengrowthandstructuraltransformationisessentialforformulatingpoliciesthatenablesustainabledevelopmentatscale(GollinandKaboski2023).

Iconstructasimplemulti-sectorgrowthmodeltoconceptualizethedistinctchannelsthroughwhichAIinfluencesgrowthandstructuraltransformation.Themodelbuildsonrecentstructuraltransformationframeworks(Rodrik2016;Comin,Lashkari,andMestieri2021;Matsuyama2019)andincorporatesnon-homotheticpreferences,whereincomeelasticitiesofdemandvaryacrosssectors.Theeconomyhasfoursectors:agriculture(A),manufacturing(M),high-skillservices(Sh),andlow-skillservices(Sl).Laboristheonlyfactorofproduction.Themodelinitiallyassumesaclosedeconomy,laterextendingtoanopeneconomycontext.Italsoexploresvariousmodelextensions,transitioningfromexogenoustoendogenousgrowthandincorporatinginter-sectoralproductivitylinkages.Themodel’spropositionsarestronglysupportedbyempiricalevidence.

High-skillservicesincludeinformationandcommunicationtechnology(ICT)services,financeandinsurance,andprofessional,scientific,andtechnicalservices.Theterm”high-skill”servesasashorthandforsectorscharacterizedbyhighlevelsofskill,income,tradability,anddigitalization.Theseindustriessignificantlydifferfromotherservicesectorsacrossallfourdimensions.BasedondatafromtheU.S.andChina,thesethreeindustriesexhibitthehighestaverageearnings,ICTserviceinputintensity,tradeintensity,andthelargestshareofemployeeswithadvanceddegrees(master’sorhigher).Otherservicesareclassifiedaslow-skillservices.Althougheducationandhealthcarealsohaveahighproportionofemployeeswithadvanceddegrees,theyareconsideredlow-skillduetotheirlowtradabilityanddigitalization.

AIaffectsgrowthandstructuraltransformationthroughthreedistinctchannels:

1.Demand:AIcreatesradicallynewproductsandshiftsconsumerpreferences.Itreshapesutilityfunctionsbyalteringincomeelasticitiesforcertainsectorsandchangingtheelasticityofsubstitutionacrosssectors.

2.Supply:AIaffectsrelativepricesacrosssectors.Whileprevioustechnologiesprimarilyaffectedthegoods-producingsectorandroutinetasks,generativeAItargetshigh-skillservices,particularly

3

cognitiveandcreativetasks(Eloundouetal.2023;Gmyrek,Berg,andBescond2023).

3.Internationalproductionspecialization:AIchangescomparativeadvantagesamongcountries.Na-tionsthatadoptgenerativeAIearlycanenhancetheircomparativeadvantageinhigh-skillser-vices.Additionally,generativeAIhasbeenshowntoimprovetheproductivityofless-skilled,less-experiencedworkers(Brynjolfsson,Li,andRaymond2023;NoyandW.Zhang2023;Pengetal.2023;Dell’Acquaetal.2023),potentiallyenablingdevelopingcountriestocompeteinhigh-skillservicesandreplacemoreexpensivelaborindevelopednations.

Usingmodelparameterssourcedfromtheliterature,thepaperquantifiesAI’spotentialimpactthroughsimulations.IexplorethreescenariosregardingAI’sinfluence:

1.Short-term/pessimistic:AIexclusivelyincreaseslaborproductivitygrowthinhigh-skillservices.

2.Medium-term/neutral:AIenhanceslaborproductivitygrowthacrossallfoursectors,withthemostsubstantialincreaseobservedinhigh-skillservices.

3.Long-term/optimistic:AInotonlyboostslaborproductivityinallsectorsbutalsocatalyzesthecreationofrevolutionaryproducts,fundamentallyalteringsocietalpreferences.

Thesimulationsrevealseveralnovelandsoberingfindings:

1.Therecentriseinhigh-skillservicesemploymentshareinmanycountrieswilleventuallystagnateordecline,followingahump-shapedcurvesimilartomanufacturing.Althoughhigherincomesincreasedemandforhigh-skillservices,advancesinAIwillreducetheneedforwhitecollarworkers,shiftingjobconcentrationtolow-skillservices.

2.AIwillfurtherlimitthepotentialforcreatingqualityjobsinthehigh-skillservicessector,partic-ularlyindevelopingcountries.Similartoprematurede-industrialization,AIwillcause”prematurede-professionalization”,wherehigh-skillservicesemploymentpeakearlierandatlowerincomelev-els.

3.SmallopeneconomiesfaceacriticaljuncturewithAIadoption.FailureordelayinembracingAItechnologiesriskserodingexistingcomparativeadvantagesinhigh-skillservicesandmanufacturing,orimpedingthedevelopmentofsuchadvantages.Thesecountriesmayconsequentlyfindthemselvestrappedascommodityexporters,withemploymentheavilyconcentratedinagricultureandlow-skillservices.Incontrast,successfulandtimelyAIadoptioncouldcatalyzethedevelopmentofcomparativeadvantagesinmanufacturingorhigh-skillservicesector.

4

4.UnlessAIiswidelyadoptedacrosssectorsanddrivestransformativeinnovationsthatpermanentlyshiftconsumerpreferences,itsgrowthbenefitswilllikelybeunderwhelming.Evenwitha50%increaseinlaborproductivitygrowthinhigh-skillservices,realincomeseesonlyamodestinitialboostof0.046percentagepoints,whichgraduallydiminishesovertime.After100years,realincomeisjust2.4%higherthanwithoutAI.IfAIenhancesproductivityacrossallfoursectors,realincomeatyear100is19%higher.Inthemostoptimisticscenario,realincomeatyear100is28%higherthanthebaseline.

ThepaperthenaddressesthelimitationsofthemodelandconsidershowgenerativeAImightinfluencegrowthandinequalitymorebroadly.Themodel’sexclusionofcapitaloverlooksakeyaspectofAI’simpact:automation,jobdisplacement,andthedeclininglaborincomeshare.Additionally,byomittingintermediateinputs,themodelmayunderestimatethedemandforhigh-skillservices.Theassumptionofhomogeneous,freelymobilelaboracrosssectors,whichleadstoequalizedincomes,divergesfromrealityandprovideslittleinsightintoincomeinequality.GenerativeAIcouldlowerentrybarriersforcertainhigh-skillservicejobs,potentiallydrivingdownaveragewagesandreducingincomeinequalityacrossindustriesandoccupations.However,itmayalsoinfluenceincentivesforeducationandskillacquisition.IfAIshrinkstheshareofhigh-skill,well-paidjobsanddepresseswagesinthesefields,individualsmaybecomelessmotivatedtoinvestinhighereducation,whichcouldhinderlong-termgrowthandlimitsocialmobility.

ThepapercontributestothegrowingbodyofliteratureontheeconomicimpactsofAIinseveralways.

First,tothebestofmyknowledge,itisthefirststudytoanalyzetheeffectsofgenerativeAIthroughthelensofstructuraltransformationandinternationalproductionspecialization.Unlikemanyexistingstudiesthatfocusonoccupationalexposurebasedoncurrenttasks,thismodelincorporatesdemand-sidefactors,rootedinnon-homotheticpreferencesacrosssectors.ThisbroaderapproachcapturesAI’spotentialtoreshapetaskcontentandcreatenewoccupations,offeringamoreholisticviewatthemacroe-conomiclevel.

Second,thispaperadvancestraditionalstructuraltransformationmodelsbyintroducingacriticaldistinctionbetweenhigh-skill,highlydigitalized,tradableservicesandlow-skill,lessdigitalized,non-tradableservices.ThisdifferentiationyieldsdeeperinsightsintotheunevenimpactofAIacrossservicesectorsandoccupations,aswellasitsimplicationsforeconomicgrowthandinequality.

Third,alongsidenon-homotheticpreferences,thisstudyexploresanewdemand-sidechannelthroughwhichAImaydrivegrowth:shiftingconsumerpreferences.OnescenariomodelsAIasacatalystfor

5

game-changingnewproductsandindustries,permanentlyalteringincomeelasticityincertainsectorsandtheelasticityofsubstitutionacrosssectors.Combinedwiththechannelofinternationalproductionspecialization,thisapproachoffersrichinsightsintoAI’svariedeconomicimpactsacrossdifferenttypesofcountries.

Fourth,thepaperusessimulationstoilluminatethegeneralequilibriumeffectsofAIoneconomicgrowthandlabormarketsundervariousscenarios,quantifyingthescaleoftheseeffects.ConsistentwithAcemoglu2024,theresultssuggestthatAI’sshort-tomedium-termgrowthimpactwilllikelybemodest.

Notably,thispaperisthefirsttopredictAI’spotentialtoreduceopportunitiesforwell-paidjobsinhigh-skillservices.ThiscouldbeparticularlyharmfulforlateAIadoptersanddevelopingcountries,aphenomenonIdubas”prematurede-professionalization”.Itunderscoresthediminishingprospectsfordevelopingeconomiestofosterhigh-skillserviceemployment,cautioningthatthoseslowtoadoptAIriskbeingrelegatedtocommodityexporters,facinglargescaleyouthunderemployment,reducedsocialmobility,andpotentialdeclinesinlivingstandards.

Therestofthepaperisorganizedasfollows:Section2presentsstylizedfactsaboutstructuraltransformation,thedivergencebetweenhigh-andlow-skillservices,andAI’sdisproportionateeffectsonhigh-skillservices.Section3introducesthefour-sectorgrowthmodelandderivesseveralempiricallysupportedpropositions.Section4simulatesthebaselinemodelusingparametersfromtheliteratureandquantifiesAI’spotentialimpactthroughthreechannelsacrossthreescenarios.Section5discussesmodellimitationsandthebroaderimplicationsofAIforgrowthandinequality.Section6concludes.

2StylizedFacts

2.1Structuraltransformation

Demandsidedriver:Differentincomeelasticitiesacrosssectors

Differencesinincomeelasticitiesofdemandacrosssectorsareakeydriverofstructuraltransformation.Economistshavelongstudiedsystematicvariationsinthesectoralcompositionofdemandasincomechanges,aconceptexploredthroughEngelcurves.Typically,servicesexhibithigherincomeelasticitiesthanagriculturalandmanufacturedgoods(1).Asaresult,non-homotheticpreferencesacrosssectorscanexplaintheshiftinconsumerspendingfromfoodandmanufacturingtoservices,observableinbothrealandnominalterms(Kongsamut,Rebelo,andXie2001;FoellmiandZweim¨uller2008;Boppart2014;Comin,Lashkari,andMestieri2021).

6

Figure1:Estimatedincomeelasticitybygoodstype

Note:BasedonUSdatafromAguiarandBils2015

Supplysidedriver:Differentproductivitygrowthacrosssectors

Changingrelativepricesacrosssectorsisakeysupply-sideforceaffectingstructuraltransformation.Changesinrelativepricesareprimarilydrivenbytheheterogeneityinsectoralproductivitygrowthrates.Empiricalestimatesshowsignificantvariationsinproductivitygrowthratesacrosssectors(MartinandMitra2001;DuarteandRestuccia2010;LawrenceandEdwards2013).Ifsectoraldemandisinelastic,rapidproductivitygrowthandsubsequentpricedeclinesinasectorcanleadtoreducedshareofspendingandemploymentinthatsector.Historically,productivitygrowthinagricultureandmanufacturinghasexceededthatintheservicesector,partlyexplainingthedecliningshareofmanufacturinginrichcountrieswithhighmanufacturingproductivity(NgaiandPissarides2007).Thismechanism,oftenreferredtoastheBaumol’sdisease,describeshoweconomicactivityshiftsfromindustrieswithfasterproductivitygrowthtothosewithslowergrowth(Baumol1967).Additionally,AcemogluandGuerrieri2008highlightcross-sectoraldifferencesinfactorintensityandcapitaldeepeningascausesofrelativepricechanges.

Globally,averageannuallaborproductivitygrowthduring1950-2010wasthehighestinagriculture(3%),followedbymanufacturing(2.8%),high-skillservices(1.5%),andthelowestinlow-skillservices(1.1%)(Figure2).Productivitygrowthvariessignificantlyacrosscountryincomegroups.Amonghigh-incomecountries,laborproductivitygrowthinagricultureandmanufacturingisabove4%,morethan

7

twicethegrowthrateinhigh-skillandlow-skillservices(around1.5%).Inuppermiddle-incomecountries,agricultureandmanufacturingreportthehighestlaborproductivitygrowthofaround3%,followedby2%inhigh-skillservices,and1.2%inlow-skillservices.Inlowermiddle-incomeandlow-incomecountries,agriculturalproductivitygrowthisonly1.5%,followedbyhigh-skillservicesat1.2%.Productivitygrowthinmanufacturingisonly0.7%,andamere0.4%inlow-skillservices.

Figure2:Laborproductivitygrowthacrosssectors

Source:GGDC10-sectorDatabase.Laborproductivityiscalculatedasvalue-addedat2005fixedprices

ineachsectordividedbysectoralemployment.SeeAppendixAformoredetails.

Internationalproductionspecializationandprematurede-industrialization

Technology-enabledglobalizationhasimpactedthedynamicsofemploymentandoutputsharesthroughpatternsofinternationalspecialization.Advancesintransportationandcommunicationtechnologiesinthelate20thcenturyenabledproductionunbundling,outsourcing,andoffshoringonamassivescale,leadingtounprecedentedgrowthininternationaltradeandinvestment.Manystudieshavedocumentedtheeffectsoftradeandmultinationalcorporations’offshoringonde-industrializationintheUS(D.H.Au-tor,Dorn,andHanson2013;Acemoglu,D.Autor,etal.2016;Boehm,Flaaen,andPandalai-Nayar2020).Suchforcesmaybeevenstrongerforsmallerdevelopingcountries,whichtypicallyactaspricetakersinworldmarkets.Theorysuggeststhatinternationaltradecancreatepatternsofstructuralchangedis-tinctfromclosedeconomymodels(Matsuyama2009;Matsuyama2019).Globalizationamplifies,ratherthanreduces,thepowerofendogenousdomesticdemandcompositiondifferencesasadriverofstructuraltransformation(Matsuyama2019).Arecentlineofworkhasdocumentedpronouncedtrendstowardearlierde-industrializationindevelopingcountries(DasguptaandSingh2007;Rodrik2016).

Mosthigh-incomecountrieshaveexperiencedacontinuousdeclineinagriculturalemploymentshare,whichhadfallenbelow3%by2010.Themanufacturingsectorinitiallysawanincreaseinemployment

8

share,peakingataround30%beforebeginningtodecline(Figure3).Concurrently,thehigh-skillservicessectorhasgraduallyincreaseditsemploymentshare,withsomecountriesalreadyreachingapeakorstagnationpointofapproximately20%.By2010,low-skillserviceshadcometodominateemployment,accountingfor70%-85%oftotaljobs.

Therapidproductivitygrowthinagriculture,alongwithindustrializationandurbanization,beganmorethan150yearsagoinWesternEuropeandtheUnitedStates.Duringthisperiod,domesticsupplyanddemandforcesplayedamuchmoresignificantrolethanglobaltrade.Whenglobalizationacceler-atedinthe1990s,mosthigh-incomecountrieshadalreadytransitionedtoservice-dominatedeconomies,enjoyingacomparativeadvantageinproducingandexportinghigh-skilltradableservices.

Figure3:Sectoralemploymentshare-Highincomecountries

Upper-middle-incomecountries,ontheotherhand,exhibitpatternsofprematurede-industrializationintheirstructuraltransformation.Agriculturalemploymentsharehasdeclinedovertimeandhasbeensurpassedbyemploymentinlow-skillservices.Manufacturingemploymentinthesecountrieshaspeakedataround20%,asignificantlylowerlevelthanthepeakreachedinhigh-incomecountries,andatalowerrealincomelevel.Theemploymentshareinhigh-skillserviceshasbeguntoincreaseinsomeofthesecountriesbutremainswellbelowthelevelsseeninhigh-incomenations,asshowninFigure4.

9

Figure4:Sectoralemploymentshare-Uppermiddleincomecountries

Lower-middle-incomeandlow-incomecountriesmaintainextremelylowandstagnantemploymentsharesinmanufacturingandhigh-skillservices(Figure5).Inmostofthesenations,agriculturestillaccountedformorethanhalfoftotalemploymentby2010.Unlikethehigh-incomecountrytrajectoryofmovingfromfarmstofactoriesandthentooffices,laborfreedfromagricultureintheselower-incomecountriestendstomovedirectlyintolow-skillservices.Theselow-skillservicejobsarepredominantlyconcentratedinretail,restaurantsandhotels,andpersonalservicessectors,presentingsignificantchal-lengesforlong-termeconomicdevelopmentandindividualprosperity.Suchjobstypicallyofferlimitedopportunitiesforlearning,skillsupgrading,andtechnologicaladvancement,whicharecrucialfordrivingproductivitygrowthandinnovation.Moreover,thesesectorsgenerallyhavelowexportpotential,restrict-ingacountry’sabilitytogenerateforeignexchangeandparticipateeffectivelyintheglobaleconomy.Unlikemanufacturingorhigh-skillservices,whichoftenbenefitfromeconomiesofscaleandtechnologicalprogress,manylow-skillservicejobsarelessamenabletosuchefficiencygains.Thiscanleadtostagnantwagesandlivingstandardsintheseeconomies,orevendeterioratingtermsoftrade.

10

Figure5:Sectoralemploymentshare-Lowermiddleandlowincomecountries

Thesediversepatternsofstructuraltransformationacrossdifferentincomegroupshighlightthecom-plexinterplayofdomesticandglobalfactorsinshapingeconomicdevelopmenttrajectories.

2.2Divergencewithintheservicesector

Distinctcharacteristicsofhigh-skillandlow-skillservices

Thediversenatureofactivitiesandjobswithintheservicssectornecessitatesamoregranularlevelofanalysis.Theservicesectorisanextensivecategoryencompassingawiderangeofactivitiesnotclassifiedunderagricultureormanufacturing.Thesubsectorsandoccupationsinservicesarehighlyvaried,rangingfromlow-skilljobssuchascashiersandcleanerstohighlyspecializedroleslikesoftwareengineers,fundmanagers,lawyers,andsurgeons.Globally,servicesnowcontributetohalfofGDPandaccountforover60%ofemployment.Inhigh-incomecountries,theservicesectorrepresentsthree-quartersofbothemploymentandGDP.Eveninlow-incomecountries,servicesaccountedforaroundone-thirdofGDPandemploymentin2022.However,theservicesectorisoftentreatedasaresidualcategoryafteragricultureandmanufacturing,anafterthoughtineconomicmeasurement.Manycountriesstillusetheoutdatedthree-sectorframeworkineconomicaccounting,andstructuraltransformationmodelsoftenfocusonthesebroadsectors,offeringlimitedinsightintotheincreasinglyservice-basedglobaleconomy.

11

Risingincomeinequalitywithinindustries,especiallyamongdifferentserviceindustries,furtherun-derscorestheneedtodistinguishvarioustypesofservicesineconomicresearch.Card,Rothstein,andYi2024havedocumentedsignificantheterogeneityinindustrywagepremiumsintheUS,showingthatworkersinhigher-payingindustriespossesshigherobservedandunobservedskills,thuswideningbetween-industrywageinequality.Haltiwanger,Hyatt,andSpletzern.d.foundthatmostoftheriseinearningsinequalityintheUSisduetoincreasingbetween-industryinequality,withthedisparityparticularlypronouncedintheservicesector.

High-skill,highlydigitalized,tradableserviceindustries,suchasinformationandcommunicationtechnology(ICT)services,financialservices,andprofessionalandbusinessservices,arepullingfurtheraheadofotherserviceindustries.In2022,theaverageannualearningsintheinformation,financeandinsurance,andprofessionalservicesectorsexceededUS$120,000intheUS,placingthemamongthetopfivehighest-earningindustries.InChina,ICTservice