BISWorkingPapersNo1269

ExpectingjobreplacementbyGenAI:Effectsonworkers’

economicoutlookandbehavior

byYusukeAoki,JoonSukPark,YuyaTakadaandKojiTakahashi

MonetaryandEconomicDepartment

May2025

JELclassification:E24,E31,O30.

Keywords:GenerativeArtificialintelligence,labormarket,inflation,productivity.

BISWorkingPapersarewrittenbymembersoftheMonetaryandEconomicDepartmentoftheBankforInternationalSettlements,andfromtimetotimebyothereconomists,andarepublishedbytheBank.Thepapersareonsubjectsoftopicalinterestandaretechnicalincharacter.TheviewsexpressedinthemarethoseoftheirauthorsandnotnecessarilytheviewsoftheBIS.

ThispublicationisavailableontheBISwebsite(

).

?BankforInternationalSettlements2025.Allrightsreserved.Briefexcerptsmaybereproducedortranslatedprovidedthesourceisstated.

ISSN1020-0959(print)

ISSN1682-7678(online)

1

ExpectingJobReplacementbyGenAI:EffectsonWorkers’EconomicOutlookandBehavior*

YusukeAoki?JoonSukPark?YuyaTakada§KojiTakahashi?

Abstract

Thispaperexaminestherelationshipbetweenindividuals’expectationsofjobre–placementbygenerativeAI(GenAI)andtheirmacroeconomicoutlooksandbehaviors.UsingonlinesurveyscombinedwithrandomizedexperimentsconductedintheU.S.andJapan,wederivethefollowingfindingsabouttheeffectsofexpectinggreaterjobreplacementduetoGenAI.First,inboththeU.S.andJapan,respondentsrevisetheirbeliefsafterreceivinginformationaboutGenAI’sjobreplacementratios.Second,inJapan,suchanexpectationleadstoanincreaseininflationexpectationsdrivenbyariseininvestment.Third,itincreasesrespondents’willingnesstouseGenAIinworkplacesinJapan.Fourth,intheU.S.,expectationsofgreaterjobreplacementamplifyconcernsaboutweakershort–termlabordemandandreducedskillrequirements,particularlyamongmoreeducatedrespondents.Inaddition,theserespondentsanticipatelowerinvestment,whilelesseducatedrespondentsexpecthigherinvestment.

Keywords:GenerativeArtificialintelligence,labormarket,inflation,productivity.

JELClassificationcodes:E24,E31,O30

*Previouslycirculatedunderthetitle”FromPerceptiontoExpectation:TheRoleofGenAIJobRe–placementinShapingEconomicOutlooksandBehavioralViews.”TheauthorsthankJonFrost,Leonardo

Gambacorta,ShingoWatanabe,DaisukeIkeda,PawelAdrjan,IakiAldasoro,GuillermoGallacher,Priscilla

KooWilkens,KumarJegarasasingam,YukoUeno,TakujiFueki,KozoUeda,MunechikaKatayamaandparticipantsattheseminarsattheBankforInternationalSettlements(BIS),BankofJapan,andWasedaUni–versity,andatthe26thMacroConferenceand1stCAM–Riskconference.JoonSukParkandKojiTakahashiworkedonthispaperwhileworkingasVisitingEconomistsattheBIS.TheviewsexpressedinthispaperarethoseoftheauthorsanddonotnecessarilyreflecttheofficialviewsofBIS,BankofJapan,BankofKorea,Indeed,ReDataScienceCo.,Ltd.,andIndeedRecruitPartnersCo.,Ltd.ThesurveywasfundedbyIndeedRecruitPartnersCo.Ltd.

?Indeed.E–mail:yaoki@

?BankofKorea.Email:parkjs@bok.or.kr

§ReDataScienceCo.,Ltd.andSpeciallyAppointedResearcher,IndeedRecruitPartnersCo.,

Ltd.Email:yuyatakada@redata.co.jp

?BankofJapan.Email:kouji.takahashi–2@boj.or.jp

2

1Introduction

“Thebestwaytopredictthefutureistoinventit.”

—AlanKay

TherapidadvancesingenerativeAI(GenAI)havesparkedsignificantinterestandpromptedwidespreadspeculationaboutitstransformativepotential.However,thehighlevelofun–certaintysurroundingGenAIhasfueledongoingdebateaboutitseconomicimpact.ThisdiscussionbecomesespeciallyheatedwhenitconcernsAI’seffectonthelabormarket.ThisintensedebateshapespublicexpectationsaboutAI’simpactontheeconomy.Asrecentstudiesontheroleofexpectationssuggest,theseperceptionscaninfluenceactualbehavior,whichinturnaffectseconomicoutlooks.Forexample,positiveviewsonAI’simpactcanencourageitsadoption,furtherboostingproductivityandinvestment.Con–versely,negativeperspectivesmayhinderAI’suseandlearning,potentiallyresultinginsmallereffectsontheeconomy.AstheimportanceofexpectationsaboutAI’simpactiswellrecognized,manysurveyshavebeenconductedfocusingonpeople’sviewsregard–ingAI’simpactonthelabormarket.However,tothebestofourknowledge,therearenostudiesexaminingtheroleofexpectationsaboutAIinshapingpeople’seconomicoutlookandbehavior.

Toaddressthisgap,weconductedasurveyonperceptionsofAI’seffectonthelabormarket,alongwithrandomizedexperimentsintheUnitedStatesandJapan.Specifically,wedividedrespondentsintotwotreatmentgroups.ThefirstgroupwasprovidedwithinformationfromanexpertanalysisthatGenAIwouldreplace14%ofcurrentjobs,whilethesecondgroupwasinformedthatAIcouldreplace47%ofjobsbasedontheestimatesby

BriggsandKodnani

(

2023

)and

FreyandOsborne

(

2017

).

1

Wethenaskedparticipantsabouttheirexpectationsregardingthejobreplacementratiobeforethetreatment(referredtoas“priorbeliefs”),followedbytheirupdatedexpectationsafterthetreatment(referredtoas“posteriorbeliefs”).Inaddition,respondentswereaskedtopredicteconomicoutcomes,suchasrealGDPgrowthratesover1–,3–,and5–10–yearhorizons,aswellastheirintentions

1SeeSection3.2formoredetaileddiscussiononthereplacementratio.Wealsocollectresponsesfromathirdgroupthatreceivedunrelatedastronomicalinformation.However,inthispaper,wefocusonthetwotreatmentgroups,asthesetreatmentsreflectrealisticsituationsthatpeoplemayencounter.

3

tolearnanduseAIintheworkplace.UsingtheresponsesaboutAI’simpactonthelabormarketandviewsonmacroeconomicvariables,weidentifythecausaleffectofindividuals’expectationsregardingthelaborreplacementratiobyGenAIontheireconomicoutlooksandbehavior.ItisimportanttonotethatwedonotaimtopredictAI’simpactontheeconomyorthelabormarket.Rather,weseektouncoverhowchangesinviewsonAIaffecteconomicexpectationsandbehavior.

OurstudyshedslightonongoingdebateabouttheimpactofAIonmacroeconomicvariablesbeyondthelabormarket.Infact,

Aldasoroetal.

(

2024

)demonstratethatGenAIcouldexertinflationarypressureontheeconomyinthelongrun,whileintheshortrun,itcouldresultineitherdisinflationaryorinflationaryeffects,dependingonhoweconomicagentsformexpectationsregardingGenAI’simpact.Specifically,ifhouseholdsandfirmsanticipatefutureproductivitygrowthdrivenbyAI,householdsmayincreaseconsumption,leadingtoinflationarypressureevenintheshortrun.Ontheotherhand,ifanincreaseinproductivityduetoAIisunanticipated,consumptiononlyincreasesgradually.Therefore,intheshortrun,ithasadisinflationaryimpactastheproductioncapacityexpands.Ourpaperfocusesontheimpactofpeople’sviewsregardingAIoninflationexpectations,ratherthanassessingwhichpredictionismoreplausible.

Usingtherandomizedexperiment,weobtainthefollowingfindings.First,people’sviewsonGenAI’slabormarketimpactcanbeupdatedbyexpertopinion.InboththeU.S.andJapan,respondentsrevisetheirbeliefsafterreceivinginformationaboutGenAI’sjobreplacementratios.Thisupdatingbehavioralignswiththepreviousliteratureonpeople’sexpectationaboutmacroeconomicvariablessuchasrecessionprobabilitiesandisconsistentwithaBayesianprocess.

Second,inJapan,higherposteriorbeliefsaboutthejobreplacementratioleadtohigherinflationexpectations.Inaddition,ahighreplacementratioisassociatedwithpositiverealGDPgrowth,particularlyamonghigh–incomeindividuals,thoughthisestimateofthemarginaleffectisnotstatisticallysignificantonaverage.Furthermore,ahighreplace–mentexpectationleadstohigherprivateinvestmentgrowthamongworkersincreativeoccupations.Wecaninferthatthisresultispartlyduetotheirhighrelianceonmanylabor–intensivetaskssuchaswritingtext,whicharelikelytobereplacedbyGenAI.These

4

resultssuggestthattheinvestmentdemandassociatedwithGenAIcouldcontributetorisinginflationrates.Moreover,respondentsinJapanshowanincreasedintentiontouseGenAIintheirworkplacewhentheyadjusttheirexpectationsregardingthereplacementratiotohigherlevels.

Incontrast,intheU.S.,wedonotfindanysignificantaverageeffectofahigherreplace–mentratiooninflationexpectations.However,theexpectationofahigherreplacementratioleadstoanexpectationofweakerlabordemandintheshorttermandadeclineintheskillsrequiredfortherespondent’scurrentjobs.Theimpactontheoutlookforlaborde–mandismorepronouncedamongindividualswithhighereducationlevels.ThissuggeststhatpeoplewithhighereducationanticipatemorenegativeeffectsofGenAIonthelabormarket,aligningwithpreviousliteratureontheheterogeneousimpactsofAIonlabor.Finally,respondentsintheU.S.donotincreasetheirwillingnesstouseorlearnGenAIintheirworkplace,eventhoughtheyhavechangedtheirviewsonitsimpactontheirjobs.

Therestofthepaperisorganizedasfollows.Section2summarizestheliteraturerelatedtothispaper.Section3explainsthedataandthesettingoftherandomizedcontroltrial.Section4introduceseconometricmodelsbyillustratingtheidentificationstrategyandthenreportsthesurveyandexperimentresults.Finally,Section5concludes.

2Literaturereview

Thisstudyisrelatedtofourstrandsofliterature.First,webuildonthepreviousstudiesontheimpactofAIsonlabormarket,especiallywhenweimplementtherandomizedexperiment.Amongmany,

FreyandOsborne

(

2017

)isaseminalpaperandanalyzestheimpactofautomationonjobsusingdetailedjobdescriptions.Theyfindthatalargeshareofcurrentsjobsareexposedtotheautomationbycomputers.

BriggsandKodnani

(

2023

)discussthepotentialofGenAIandtheireffectsoneconomy,byshowingdifferentscenariosofAIdevelopmentsandthesubsequentimpactonthelabormarket.Wealsousetheirestimatesintherandomizedexperiments.

Webb

(

2019

)usespatentdatatoidentifywhichtaskwouldbemostaffectedbyautomationwithAI.AgrowingnumberofstudiesanalyzetherelationshipbetweenlabormarketandAIincluding

Manyikaetal.

(

2017

),

5

Huietal.

(

2023

),

Feltenetal.

(

2021

)and

AcemogluandRestrepo

(

2020

).

2

Babinaetal.

(

2023

)findthatcompanieswithalargerinitialproportionofmoreeducatedworkerswithexpertiseinscience,technology,engineering,andmathematicstendtoinvestmoreinAI.Inaddition,theyreportthatAIinvestmentsareassociatedwithaflatteningoforganizationalhierarchies,withariseinjunior–levelemployeesandadeclineinmiddle–managementandseniorpositions.

Cazzanigaetal.

(

2024

)demonstratethattherearecleartrendsinAIexposure:womenandcollege–educatedindividualsfacehigherexposurebutarealsobetterpositionedtobenefitfromAIadvancements,whileolderworkersmaystrugglemoretoadapttothenewtechnology.

Yang

(

2022

)studiestheimpactofAIinTaiwan’selectronicsindustryforthe2002–2018periodandfindsthatAItechnologyispositivelyassociatedwithproductivityandemployment.

Hering

(

2023

)usestheonlinejobpostingdataandfindsthat20%ofjobsfacesthehighestlevelofpotentialexposure.

Second,weextendtheliteratureonexpectationformationsofeconomicvariables.

3

Inparticular,wefollowtheexperimentalsettingof

RothandWohlfart

(

2020

)thatstudytherelationshipsbetweenmacroeconomicexpectationsandindividualbehavior.Agrowingbodyofresearchstudiespeople’sexpectations.Forexample,

Dasetal.

(

2020

)examinetheheterogeneityinexpectationformationacrosspeoplewithdifferentsocioeconomicstatuses.Theyfindthatindividualswithhigherincomeoreducationlevelstendtobemoreoptimisticaboutfuturemacroeconomicdevelopments.

KuchlerandZafar

(

2019

)explorestherelationshipbetweenpersonalexperiencesandviewsonthemacroeconomy,findingthatindividualswhoexperienceunemploymentpersonallybecomemorepes–simisticaboutfuturenationwideunemployment.Theextentofthisextrapolationismorepronouncedamonglesssophisticatedindividuals.

MalmendierandNagel

(

2011

)demon–stratesthatpastindividualexperiencesplayasignificantroleinexplainingrisk–takingbehavior.Regardingdifferencesacrossagegroups,

MalmendierandNagel

(

2016

)findsthatinresponsetoinflationsurprises,youngerpeopleupdatetheirexpectationsmorestronglythanolderindividuals,asrecentexperiencesweighmoreheavilyintheiraccu–mulatedlifetimehistory.However,householdexpectationsabouttheeconomymaybe

2

BIS

(

2024

)comprehensivelydiscussestheimpactofAIontheeconomyincludinglabormarket.

3

Manski

(

2018

)providesacomprehensiveliteraturereviewonmacroeconomicexpectations.

6

biased.Infact,

Mianetal.

(

2023

)findthatanindividual’sexpectationsforfutureeconomicgrowtharebiaseddependingonwhetherherfavorablepoliticalpartycontrolstheWhiteHouse,andthisbiasisnotnecessarilylinkedtoactualbehaviorssuchasconsumption.Thisresultsuggeststhatfurtherintensivestudiesonhouseholdexpectationsareneeded.Theimportanceofindividualbeliefsinfinancialdecision–makingishighlightedby

Bailey

etal.

(

2019

).WecontributetotheliteraturebyfocusingontherelationshipbetweenviewsonAI’sroleandmacroeconomicconditions.

Third,ourstudyextendsexistingresearchontheimpactofAIonthelabormarketthroughtheuseofsurveys.Amongothers,

Laneetal.

(

2023

)provideacomprehensiveviewofAI’simpactinworkplaces,enablinginternationalcomparisons.

McElheranetal.

(

2024

)useasurveyonbusinessestostudytrendsinAIadoption,findingthatdynamicyoungfirmswithmoreeducated,experienced,andyoungerownershavethehighestratesofAIuse.Weextendthissurveyapproachusingarandomizedcontrolledtrial.

Finally,ourpapershedslightonthedifferencesinviewsregardingtheimpactofAIacrosscountries.Intermsoftheimpactofautomationandroboticsontheeconomy,adifferentlandscapeappearsbetweenJapanandtheU.S.ManystudiesontheU.S.labormarket,including

AcemogluandRestrepo

(

2020

),demonstratethatrobotsreplacehumanlabor.Ontheotherhand,

Adachietal.

(

2024

)showthattheincreaseinrobotusageincreasesemploymentbyraisingtheproductivityandproductionscaleofrobot–adoptingindustries.Althoughthereasonsforthedifferingimpactsofroboticsbetweenthetwocountriesremaindebatable,suchexperiencesmayinfluencetheirexpectationsabouttheimpactofAI,leadingtodifferentresults.WecontributetothisliteraturebyconductingthesamesurveyinbothJapanandtheU.S.,uncoveringdifferencesinthecausaleffectsofviewsonAIinshapingmacroeconomicoutlooksandindividualbehaviorsbetweenthetwocountries.

3Dataandsettingofrandomizedexperiment

Thissectiondescribesthesurveymethodology,samplingstrategy,datacleaningprocesses,anddetailsoftherandomizedtreatmentgroups.

7

3.1Surveymethodology

Wecollected4,144responsesfromboththeU.S.andJapanthroughlarge–scaleinternetsurveypanelsadministeredbyMacromill,Inc.,Japan’sleadingsurveycompany.Onlyfull–timeemployeesbetweentheagesof20and59wereinvitedtoparticipateinthesurveyandrespondentswhoappropriatelycompleteditcouldearncashoragiftvoucher.Thesurveywasimplementedtoensurethatthenumberofrespondentswasequalizedacrossgender,age,andjobcategory.Morespecifically,respondentsarecategorizedinto16distinctdemographicgroupsbasedongender,age(20to39yearsoldand40to59yearsold),andjobcategory(SalesandAdministrative,Engineering,PlanningandSpecialist,andCreative).Weshouldnotethatthesefourjobcategoriesarenotcomprehensive.However,wefocusonthempartlytoensureasufficientnumberofresponseswithineachcategory,andpartlybecauseworkersinthesecategoriesareexpectedtobehighlyexposedtogenerativeAI,accordingtoexistingstudiessuchas

Chuietal.

(

2023

).ThedetailsofthejobcategoriesaresummarizedinTable

A.1

inAppendixA.ThesurveywasconductedfromMay20toJune3,2024,fortheU.S.participantsandfromMay20toMay29,2024,forJapaneseparticipants.

3.2Settingofrandomizedexperiment

WerandomlyassignedrespondentsintothreegroupstoevaluatetheimpactofGenAIonjobreplacement.Thefirstgroupwaspresentedwiththefollowinginformation:“Awell–knownstudyonAIestimatesthat14%ofcurrentjobscouldbereplacedbyGenAIinthefuture.”Thesecondgroupwasinformedthat“Awell–knownstudyonAIestimatesthat47%ofcurrentjobscouldbereplacedbyGenAIinthefuture.”

Thereplacementratiosof14%and47%arebasedonestimatesprovidedby

Briggs

andKodnani

(

2023

).Weshouldnotethat

BriggsandKodnani

(

2023

)donotreportthereplacementratioofthelaborforcebyAIperse.Instead,theyestimatetheexposureofthelaborforcetoAI–drivenautomationacrossvariousscenarios,assumingvaryinglevelsofAIdevelopment.

4

The14%replacementratioreflectsthelowestestimatein

4

BriggsandKodnani

(

2023

)refertothisestimateasthe”shareoffull–timeequivalentUSemploymentexposedtoautomationbyAI.”

8

theirscenarios,whilethe47%figureisbasedonestimatesfrom

FreyandOsborne

(

2017

),whichalsosuggeststhat47%ofjobsareexposedtoautomation.Weoptednottousetheterm“exposed”asintheoriginalpapersbecauseitcouldbeambiguousandinterpreteddifferentlybyrespondents.However,weacknowledgecertainlimitationsinourapproach.Specifically,theterm”replaced”mayevokeanegativeperceptionofAI,eventhoughitsimpactcouldbepositiveinmanyareas,suchasincreasedproductivity.Moreover,respondentsfamiliarwiththeoriginalresearchmightnoticethedifferenceinterminologyandbecomelessengagedinansweringsubsequentquestions.Despitethesecaveats,ourapproachreducesambiguityandprovidesafinerunderstandingoftheroleofpeople’sviewsonAI.Inaddition,wedidnotmentionthesourcesoftheseestimatesinthesurvey.Thisdecisionwasmadetoavoidintroducingbias,asbeliefinthecredibilityofdatasourcescanvarydependingonindividualcharacteristics,potentiallyintroducingnoiseintotheirresponses.Furthermore,wedidnotproviderespondentswithinformationonaforecasthorizonfortheprofessionalestimates,giventhattheestimatesdonotspecifyatimehorizonforAI’slabormarketimpact.Thisapproachallowsustoobserverespondents’ownbeliefsaboutwhenGenAIeffectsmaymaterialize.

RespondentsansweredthequestionsintroducedinSection

3.3

bothbeforeandafterre–ceivingthetreatmentinformation.Thisdesignallowedustomeasurechangesinresponsesresultingfromtheprovidedinformation.Wefocusonresultsfromthetwotreatmentgroupswith14%and47%jobreplacementratiobyGenAI,followingthemethodologyof

RothandWohlfart

(

2020

).

3.3Surveyquestionsanddefinitionofvariables

Weoutlinethequestiondefiningthetreatmentvariableforthecausaleffectanalysis,followedbythequestionsdefiningtheoutcomevariables.Respondentsansweredeachofthesequestionsbothpriortoandfollowingthetreatment.

Toconstructthetreatmentvariable,weaskrespondentsabouttheshareofjobsbeingreplacedbyGenAIin1,5,and10years.Therespondentsentervaluesinpercentagesforeachhorizon.DetailsofthisvariablecanbefoundinthefirstlineofTable

1

.Astotheoutcomevariablesintherandomizedexperiment,weaskquestionsregardingexpectations

9

forkeymacroeconomicvariablesoverdifferenttimehorizons(1year,3years,andanaverageof5–10years).RespondentsansweredquestionsonConsumerPriceIndex(CPI),privateinvestmentgrowth,andrealGDPgrowthbothbeforeandaftertreatment,withanswerchoicesin0.5%increments,asshowninrows2to4ofTable

1

.

Table1:Detailsofthetreatmentvariableandtheoutcomevariables

Variablename

Question

Choices

Processingmethod

ShareofjobsreplacedbygenerativeAI

WhatpercentageofcurrentjobsdoyouthinkwillbereplacedbygenerativeAIinthefuture?Pleaseansweraboutsocietyingeneral,notaboutyourownworkspecifically.

values:%

Noprocessing

CPI

Howdoyouthinkthefollowingindicatorsinyourcountrywillchangeinthefuture?

ConsumerPriceIndex(YoY)

Choices

0.5increments

+0.5%,

+1.0%,

+1.5%,...etc

Scaledbetween

aminimumof0

andamaximumof1

Private

investment

Howdoyouthinkthefollowingindicatorsinyourcountrywillchangeinthefuture?

Privateinvestment(YoY)

Choices

0.5increments

+0.5%,

+1.0%,

+1.5%,...etc

Scaledbetween

aminimumof0

andamaximumof1

Real

GDPGrowth

Howdoyouthinkthefollowingindicatorsinyourcountrywillchangeinthefuture?

RealGDP(YoY)

Choices

0.5increments

+0.5%,

+1.0%,

+1.5%,...etc

Scaledbetween

aminimumof0

andamaximumof1

Notes:WhenaskingaboutmacroeconomicvariableslikeCPI,privateinvestment,realGDP,weprovidethepastvaluesfor2022and2023asreferencepoints.Thechoicesforthequestionsofmacroeconomicvariablesareinincrementsof0.5%,suchas+0.5%,+1.0%,+1.5%,etc.,withamaximumof+5.0%.Iftherespondentwantstochooseavaluegreaterthanthat,theyshouldselectthechoice”Increaseofmorethan5.0%”.Similarly,theminimumis–5.0,andiftherespondentwantstochooseavaluelessthanthat,theyshouldselectthechoice”Decreaseofmorethan5.0%”.

Asadditionaloutcomevariables,wealsosurveyrespondentsabouttheirperspec–tivesontheirownjobs,specificallyregardingwagegrowth,skills,labordemand,andproductivity.

5

DetailsofthesevariablescanbefoundinTable

2

.

WealsopreparedvariablesfortheintentionforlearningandusingGenAIintheworkplace.DetailsofthesevariablescanbefoundinTable

3

.

5Theyareaskedabouttheprojectionsofthosevariablesover1,3,5,and10–yearhorizons.The1–,5–,and10–yearprojectionsareusedinthispaper.

10

Table2:Additionaloutcomevariables:Wagegrowth,Skills,andLabordemand

Variablename

Question

Choices

Processingmethod

Wagegrowth

HowdoyouthinkthespreadofgenerativeAIwillimpactwagesforyourcurrentjobinthefuture?

A1:20%orgreaterincrease

A2:10–19%increase

A3:5–9%increase

A4:1–4%increase

A5:Nochange

A6:1–4%decrease

A7:5–9%decrease

A8:10–19%decrease

A9:20%orgreaterdecrease

A10:Other

Scaledbetweenaminimumof0

andamaximumof1

RespondentswithA10areexcludedfromthesamplefortheestimation.

Skills

HowdoyouthinkthespreadofgenerativeAIwillchangetheskillsrequiredforyourcurrentjobinthefuture?

A1:Nochange

A2:Lessskillswillberequired

A3:Moreskillswillberequired

A4:Idon’tknow

Standardizingresponsesthatindicateadecreasetoaminimumvalueof0

andthosethatindicateanincreaseto

amaximumvalueof1

Labordemand

HowdoyouthinkthespreadofgenerativeAIwillchangethedemandforyourcurrentjobinthefuture?

A1:Nochange

A2:Decreaseddemand

A3:Increaseddemand

A4:Idon’tknow

Standardizingresponsesthatindicateadecreasetoaminimumvalueof0

andthosethatindicateanincreaseto

amaximumvalueof1

Productivity

HowdoyouthinkthespreadofgenerativeAIwillchangeproductivityinyourcurrentjobinthefuture?

A1:Nochange

A2:Decreasedproductivity

A3:Increasedproductivity

A4:Idon’tknow

Standardizingresponsesthatindicateadecreasetoaminimumvalueof0

andthosethatindicateanincreaseto

amaximumvalueof1

3.4Representativenessanddatacleaning

Toensurebalanceddemographicrepresentation,oursurveyaimedforequalnumbersofrespondentsacrossgender,age,andjobcategory.Eachgrouphasnearlythesamenumberofrespondentsacrossgenders.Forage,respondentswereaskedtospecifytheiragein5–yearincrements.Weexcludedrespondentswhowereunder20yearsoldandthosewhowere60yearsoldorolder.Thesampleincludesanequalnumberofrespondentsaged20–39and40–59,ensuringbalancedrepresentationacrosstheseagegroups.Respondentswereaskedtochooseoneof60jobcategories,andweaimedtocollectanequalnumberofvalidresponsesacrossfourconsolidatedgroups:SalesandAdministrative,Engineering,PlanningandSpecialist,andCreative.However,weobtainedfewervalidresponsesfromindividualsintheCreativejobcategory,resultinginalowerproportionofcollectedsamplesforthisgroup.

Weimplementedadatacleaningprocessfocusedonthetreatmentvariable,ortheshareofjobsreplacedbyGenAI.Weexcludeextremeresponsesof0%or100%fromrespondents’

11

Table3:Additionaloutcomevariables:GenAIlearning/useintention

Variablename

Question

Choices

Processingmethod

GenerativeAIlearning

intention

DoyouhavetheopportunitytolearnaboutgenerativeAI?

Pleaseselectallthatapply.

A1:Idonothavetheopportunity

tolearnitatpresentanddonotplantolearninthefuture

A2:Idonothavetheopportunity

tolearnitatpresent,butIwouldliketolearninthefuture

A3:Iamlearningitonmyownduringworkhours

A4:Iamlearningit

throughonlinecourses,seminars,etc.duringworkhours

A5:Iamlearningit

atagraduateschool,etc.

A6:Iamlearningitonmyownduringmyprivatetime

A7:Iamlearningitthroughonlinecourses,seminars,etc.duringmyprivatetime

A8:Other

A1andA2areused

Convertingintobinary

A1=0

A2=1

A3toA7arenotusedSeeNotes

GenerativeAI

useintention

DoyouwanttousegenerativeAIinyourworkinthefuture?

Selectoneonly.

A1:Idonotwanttouseit

A2:Iwanttoactivelyexplore

the