英文行業(yè)資料 OECD-The impact of AI on the workplace Evidence from OECD case studies of AI implementation-report (2023-03-28)

OECDSocial,EmploymentandMigrationWorkingPapers

No.289

TheimpactofAI

ontheworkplace:Evidence

fromOECDcasestudies

ofAIimplementation

AnnaMilanez

/10.1787/2247ce58-en

OrganisationforEconomicCo-operationandDevelopment

Unclassified

DELSA/ELSA/WD/SEM(2023)7

English-Or.English

23March2023

DIRECTORATEFOREMPLOYMENT,LABOURANDSOCIALAFFAIRS

EMPLOYMENT,LABOURANDSOCIALAFFAIRSCOMMITTEE

Cancels&replacesthesamedocumentof22March2023

TheImpactofAIontheWorkplace:EvidencefromOECDCaseStudiesofAIImplementation

OECDSOCIAL,EMPLOYMENTANDMIGRATIONWORKINGPAPERSNo.288

JELcodes:J2,J3,J5,J6

AuthorisedforpublicationbyStefanoScarpetta,Director,DirectorateforEmployment,LabourandSocialAffairs.

AllSocial,EmploymentandMigrationWorkingPapersarenowavailablethroughtheOECDwebsiteat

/els/workingpapers

.

AnnaMilanez:

Anna.Milanez@

JT03515015

OFDE

Thisdocument,aswellasanydataandmapincludedherein,arewithoutprejudicetothestatusoforsovereigntyoveranyterritory,tothedelimitationofinternationalfrontiersandboundariesandtothenameofanyterritory,cityorarea.

2DELSA/ELSA/WD/SEM(2023)7

Unclassified

OECDSocial,EmploymentandMigrationWorkingPapers

/els/workingpapers

OECDWorkingPapersshouldnotbereportedasrepresentingtheofficialviewsoftheOECDorofitsmembercountries.Theopinionsexpressedandargumentsemployedarethoseoftheauthor(s).

WorkingPapersdescribepreliminaryresultsorresearchinprogressbytheauthor(s)andarepublishedto

stimulatediscussiononabroadrangeofissuesonwhichtheOECDworks.CommentsonWorkingPapers

arewelcomed,andmaybesentto

els.contact@

.

Thisseriesisdesignedtomakeavailabletoawiderreadershipselectedlabourmarket,socialpolicyandmigrationstudiespreparedforusewithintheOECD.Authorshipisusuallycollective,butprincipalwritersarenamed.Thepapersaregenerallyavailableonlyintheiroriginallanguage–EnglishorFrench–withasummaryintheother.

Thisdocumentandanymapincludedhereinarewithoutprejudicetothestatusoforsovereigntyoveranyterritory,tothedelimitationofinternationalfrontiersandboundariesandtothenameofanyterritory,cityorarea.

?OECD2023

Youcancopy,downloadorprintOECDcontentforyourownuse,andyoucanincludeexcerptsfromOECDpublications,databasesandmultimediaproductsinyourowndocuments,presentations,blogs,websitesandteachingmaterials,providedthatsuitableacknowledgmentofOECDassourceandcopyright

ownerisgiven.Allrequestsforcommercialuseandtranslationrightsshouldbesubmittedto

rights@

.

DELSA/ELSA/WD/SEM(2023)73

Unclassified

TheImpactofAIontheWorkplace:

EvidencefromOECDCaseStudiesofAIImplementation

4DELSA/ELSA/WD/SEM(2023)7

Unclassified

Acknowledgments

ThispublicationcontributestotheOECD’sArtificialIntelligenceinWork,Innovation,ProductivityandSkills(AI-WIPS)programme,whichprovidespolicymakerswithnewevidenceandanalysistokeepabreastofthefast-evolvingchangesinAIcapabilitiesanddiffusionandtheirimplicationsfortheworldofwork.TheprogrammeaimstohelpensurethatadoptionofAIintheworldofworkiseffective,beneficialtoall,people-centredandacceptedbythepopulationatlarge.AI-WIPSissupportedbytheGermanFederalMinistryof

LabourandSocialAffairs(BMAS)andwillcomplementtheworkoftheGermanAIObservatoryinthe

Ministry’sPolicyLabDigital,Work&Society.Formoreinformation,visit

https://oecd.ai/work-

innovationproductivity-skills

and

https://denkfabrik-bmas.de/

.

Inaddition,thereportbenefitedfromsupportfrom:Austria’sFederalMinistryofLabour,SocialAffairsandConsumerProtection;thedepartmentforEmploymentandSocialDevelopmentCanada;Ireland’sDepartmentofEnterprise,TradeandEmployment,theU.S.DepartmentofLabor;theUKEconomicandSocialResearchCouncil;ESSECBusinessSchool,theJapanInstituteforLabourPolicyandTraining.

ThisreportwascarriedoutunderthesupervisionofStijnBroecke(TeamLeadfortheFutureofWork)andMarkKeese(HeadoftheSkillsandEmployabilityDivisionoftheDirectorateforEmployment,LabourandSocialAffairs).ThereportalsobenefittedfromhelpfulcommentsfromcolleaguesfromtheDirectorateforEmployment,LabourandSocialAffairsincludingStijnBroecke,SandrineCazes,AndrewGreen,MargueritaLane,JulieLassébie,GlendaQuintini,AngelicaSalvidelPeroandChloéTouzet,aswellasfromcolleaguesfromtheDirectorateforScience,TechnologyandInnovationandfromtheDirectorateforEducationandSkills.ThesuggestionsofexternalreviewersMichaelHandel(BureauofLaborStatistics,U.S.DepartmentofLabor),JennaMyers(UniversityofTorontoCentreforIndustrialRelationsandHumanResources)andNicoleNestoriak(BureauofLaborStatistics,U.S.DepartmentofLabor)wereextremelyuseful.ParticularthankstoMargueritaLane(OECD)andMorganWilliams,whodraftedtheresearchmaterials.ThankyoutoAssaFofanaforsupportduringthepublicationprocess.

Thisreportwouldnothavebeenpossiblewithoutresearchteamsineachparticipatingcountrythatrecruitedfirms,coordinatedandconductedinterviews,andprovidedresearchinsightstotheOECD.ThankstoKarinPetzlberger(Austria),Anne-MarieMulumba(Canada),JulienMalaurent(France),DorisJanssen(Germany),SaraStevenson(Ireland),ShinyaIwatsuki(Japan),TimSturgeon(UnitedStates),andEsmeTerry,WilHuntandSteveRolf(UnitedKingdom).

Inaddition,thisreportwouldnothavebeenpossiblewithoutthefirmswhoagreedtoparticipateintheresearchandthoseindividualswhomadethemselvesavailabletosharetheirexperienceswith

DELSA/ELSA/WD/SEM(2023)75

Unclassified

researchers.Thereportcouldnothavebeenwrittenwithoutthem.Someoftheparticipatingfirmsinclude:Accenture,AdamRieseGmbH,AltraFederalCreditUnion,AspireFoodGroup,BankAustria,BenevaInc.,BigRedCloud,BostonScientificCorporation,BristolMyersSquibb,BudFinancial,Co-operatorsGroupLimited,CréditMutuelArkéa,Dimension88,électricitédeFrance(EDF),EnergyProIreland,ESBIreland,Fidelity,FujitsuUKandIreland,GYS,Huawei,InvoiceFair,IO-DynamicsGmbH,JanssenIreland,JBARiskManagement,Kalypso(ARockwellAutomationBusiness),KARLMAYERSTOLLGmbH,LEDRoadwayLighting,Legal&General,Legrand,MannaDroneDelivery,Manulife,MichelinRetreadDivision,msglife,ProgenyGroup,Renault,RevenueIreland,RobertBoschGmbH,RollsRoyce,RoyalBankofCanada,SiemensDigitalIndustries,SoluviaEnergyServicesGmbH,SSEAirtricity,StadtwerkeKiel,SwissLife,Toyota,VRAIIreland,andYorkshireBuildingSociety.Notallparticipatingfirmswantedtobepubliclyacknowledged;gratitudeisalsoextendedtofirmswhowishtheirparticipationtoremainanonymous.

6DELSA/ELSA/WD/SEM(2023)7

Unclassified

Abstract

Howartificialintelligence(AI)willimpactworkplacesisacentralquestionforthefutureofwork,withpotentiallysignificantimplicationsforjobs,productivity,andworkerwell-being.WhileagrowingnumberofquantitativestudiesattempttoassesstheimpactofAIonvariouslabourmarketoutcomes,thereremainsignificantknowledgegapsintermsofhowfirms,workers,andworkerrepresentativesareadapting.Thisstudyaddressesthesegapsthroughaqualitativeapproach.Itisbasedonanewdatacollectionthatresultedinnearly100casestudiesoftheimpactsofAItechnologiesonworkplacesinthemanufacturingandfinancesectorsofeightOECDcountries.ThefindingsprovideanuancedpictureofthebenefitsandchallengesofAIforworkers,addinggranularitytothepublicdebate.Thecasestudiesshowthat,todate,jobreorganisationappearsmoreprevalentthanjobdisplacement,withautomationpromptingthereorientationofjobstowardstasksinwhichhumanshaveacomparativeadvantage.ThejobqualityimprovementsassociatedwithAI–reductionsintedium,greaterworkerengagement,andimprovedphysicalsafety–maybeitsstrongestendorsementfromaworkerperspective.Atthesametime,thecasestudieshighlightcertainchallenges–higherskillrequirements,adeficitofspecialisedAIskills,andfrequentreportsofincreasedworkintensity–underscoringtheneedforpoliciestoensurethatAItechnologiesbenefiteveryone.

DELSA/ELSA/WD/SEM(2023)77

Unclassified

übersicht

WiewirktsichderEinsatzkünstlicherIntelligenz(KI)amArbeitsplatzaus?DieseFrageistfürdieZukunftderArbeitvonzentralerBedeutung,dennKIkannerheblicheAuswirkungenaufArbeitspl?tze,Produktivit?tunddasWohlergehenderBesch?ftigtenhaben.EinewachsendeZahlvonquantitativenStudienbesch?ftigtsichbereitsmitdemEinflussvonKIaufverschiedeneArbeitsmarkt-ergebnisse.Gro?eWissenslückenbestehenjedochnachwievorbeiderFrage,wiesichUnternehmen,Besch?ftigteundArbeitnehmervertretungenandiezunehmendeVerbreitungvonKIanpassen.DieseStudiesetztaufeinenqualitativenAnsatz,umdieseLückenzuschlie?en.SieberuhtaufeineraktuellenDatenerhebung,beideranhandvonfast100FallstudiendieAuswirkungenvonKI-TechnologienanArbeitspl?tzenimVerarbeitendenGewerbeundimFinanzsektorinachtOECD-L?ndernuntersuchtwurden.DieErgebnissezeichneneindifferenziertesBildderVorteileundHerausforderungenvonKIfürdieBesch?ftigtenundbereicherndurchihreGranularit?tdie?ffentlicheDebatte.DieFallstudienzeigen,dassKIbislangehereineUmorganisationvonT?tigkeitenalstats?chlicheArbeitsplatzverlusteausl?st,dadieAutomatisierungeinest?rkereVerlagerungaufjeneAufgabenbewirkt,beidenenMenschenkomparativeVorteileaufweisen.DieVerbesserungderBesch?ftigungsqualit?t–durchwenigermonotoneT?tigkeiten,h?hereMotivationundgr??ereArbeitssicherheit–dürfteausSichtderBesch?ftigtenderpositivsteAspektvonKIsein.ZugleichlassendieFallstudienauch

gewisseHerausforderungenerkennen,wieetwah?here

Kompetenzanforderungen,einenMangelanspezialisiertenKI-KompetenzenunddievonvielenBefragtenerw?hnteSteigerungderArbeitsintensit?t.DiesunterstreichtdieNotwendigkeitvonMa?nahmen,diedafürsorgen,dassKI-Technologienallenzugutekommen.

8DELSA/ELSA/WD/SEM(2023)7

Unclassified

Résumé

L'impactdel'intelligenceartificielle(IA)surleslieuxdetravailestunequestioncentralepourl'avenirdutravail,avecdesimplicationspotentiellementimportantespourlesemplois,laproductivitéetlebien-êtredestravailleurs.Alorsqu'unnombrecroissantd'étudesquantitativestententd'évaluerl'impactdel'IAsurdiversrésultatssurlemarchédutravail,ilrested'importanteslacunesdanslesconnaissancessurlamanièredontlesentreprises,lestravailleursetlesreprésentantsdestravailleurss'adaptent.Cetteétudecombleceslacunesparuneapprochequalitative.Elleestbaséesurunenouvellecollectededonnéesquiaaboutiàprèsde100étudesdecassurlesimpactsdestechnologiesd'IAsurleslieuxdetravaildanslessecteursdel’industriemanufacturièreetdelafinancedehuitpaysdel'OCDE.Lesrésultatsfournissentuneimagenuancéedesavantagesetdesdéfisdel'IApourlestravailleurs,ajoutantdelagranularitéaudébatpublic.Lesétudesdecasmontrentqu'àcejour,laréorganisationdesemploissembleplusrépanduequeledéplacementdesemplois,l'automatisationincitantàlaréorientationdesemploisversdestachesdanslesquellesleshumainsontunavantagecomparatif.Lesaméliorationsdelaqualitédesemploisassociéesàl'IA-réductiondel'ennui,plusgrandengagementdestravailleursetaméliorationdelasécuritéphysique-peuventêtresaplusforteapprobationdupointdevuedestravailleurs.Danslemêmetemps,lesétudesdecasmettentenévidencecertainsdéfis-desexigencesdecompétencesplusélevées,undéficitdecompétencesspécialiséesenIAetdesrapportsfréquentsd'intensitédetravailaccrue-soulignantlanécessitédepolitiquespourgarantirquelestechnologiesd'IAprofitentàtous.

DELSA/ELSA/WD/SEM(2023)79

Unclassified

Tableofcontents

TheImpactofAIontheWorkplace:EvidencefromOECDCaseStudiesofAI

Implementation3

Acknowledgments4

Abstract6

übersicht7

Résumé8

ExecutiveSummary11

Zusammenfassung14

Principauxrésultats17

1Introduction20

2Methodology22

Casestudyapproach22

Researchdesign22

Informationgathered24

3WorkersAffectedbyAITechnologies29

AItechnologiesareaffectingawiderangeofoccupations29

Thereisevidenceofdisproportionateimpacts,includingonolderworkers32

4HowAITechnologiesareImpactingEmploymentLevels35

Relatedliterature36

AItechnologiesoftendidnotchangethejobquantitiesofworkersmostaffected37

AItechnologiessometimesdecreasedemploymentlevelsofworkersmostaffected,though

firmsdidnotmaketheseworkersredundant42

JobgrowthinoccupationsrelatedtoAIdevelopmentandmaintenance46

5HowAITechnologiesareChangingTaskComposition48

Relatedliterature48

AItechnologiesarechangingtaskcompositionofjobsinarangeofways50

6HowAITechnologiesareImpactingSkillRequirements60

Relatedliterature61

AItechnologieschangeskillrequirements63

Skillrequirementsoftenremainedthesame69

10DELSA/ELSA/WD/SEM(2023)7

Unclassified

7HowAITechnologiesareImpactingJobQuality

71

Relatedliterature71

AItechnologiesoftenledtoimprovementsinworkingenvironments,thoughthereareconcerns

ofincreasedworkintensityandstress73

Wagesremainlargelyunchanged,withsomereportsofwageincreases81

8PoliciesShapingtheImpactsofAITechnologiesontheWorkplace84

Workers’voice84

Training95

Governmentpolicy99

Regulation102

AnnexA106

References110

FIGURES

Figure1.OccupationCategoriesofWorkersMostAffectedbyAITechnologiesExploredintheCaseStudies31

Figure2.OccupationsintheCaseStudiesHadBothLowandHighExposuretoAITechnologies32

TABLES

Table1.Stakeholdertypesinterviewedaspartofthecasestudies23

Table2.Numberofinterviewsbysectorandbycountry26

Table3.Numberofcasestudiesbysectorandbycountry26

Table4.Numberofinterviewsbystakeholdertypeandbycountry27

Table5.OccupationsofWorkersMostAffectedbyAITechnologiesExploredintheCaseStudies29

Table6.Casestudyexamplesofcomplementarytaskchange53

Table7.Casestudyexamplesoffulltaskautomation55

Table8.Casestudyexamplesofpartialtaskautomationofsimpleversionsoftasks58

BOXES

Box1.DeterminingtheLevelofAnalysisforTaskChange50

Box2.InJapan,firmshelddirectworkerconsultationsessionstobuildconsensuswithworkers87

Box3.Socialpartners’attitudestowardsAItechnologies91

DELSA/ELSA/WD/SEM(2023)711

Unclassified

ExecutiveSummary

Artificialintelligence(AI)hasthepotentialtotransformlabourmarkets.ThereisanimportantongoingdebateontheimpactsofAIontheworldofwork,withsomesuggestingitwillresultinlabourmarketupheavalandamorelimitedroleforhumansinthefutureofworkandotherssuggestingitwillresultinincreasedproductivityandworkerwell-being.ThegrowingprevalenceofAItechnologieshasincreasedthesalienceofthisdebateandyet,duetoalackofgranularevidenceavailable,policymakersstillhavemanyunansweredquestions.Forinstance,isthe"AIrevolution"likepasttechnologicalwavesordifferent,andhow?HowdotheimpactsofAIdifferacrossfirms,sectors,andcountries?Howareworkersandfirmsadaptingtothechangesbroughtabout?

ThisstudyoffersadetailedlookatthewaysinwhichAIischangingworkacrossOECDcountries.Itdrawsonalargenewdatacollectionexercisewhichresultedinnearly100casestudiesofAItechnologiesimplementedinworkplacesinthefinanceandmanufacturingsectorsacrosseightcountries(Austria,Canada,France,Germany,Ireland,Japan,theUnitedKingdom,andtheUnitedStates).Bylettingarangeofstakeholders,andforemostworkers,speakforthemselves,thestudyexpandsthesetofnarrativesexplainingthe"whyandhow"ofAIimplementationandsharpenstheirquality,providingaricherevidencebaseontheimpactofAIonemploymentlevels,taskcomposition,skillrequirementsandjobquality.Thefindingssuggestthat:

.AItechnologiesareimpactingawiderangeoftasksandworkers.TheAIapplicationsstudiedsuggestthatAIisfurtheringtheautomationofroutinetasksasAI'snewcapabilitiesenablenewsolutions(e.g.,advancesinimagerecognitionappliedtoqualityassurance,advancesinnaturallanguageprocessingappliedtoinformationretrievalusedtopowerchatbots).Inaddition,AIisgainingnewgroundwhenitcomestotheautomationofnon-routinetasks(e.g.,AI-poweredpredictivemaintenancesystemsinmanufacturing,whichrelievetechniciansofthenon-routinetaskoftrouble-shootingequipmentfailuresbyanticipatingbreakdownsbeforetheyoccur).Asaresultofsuchwidereach,theworkersmostaffectedbyAItechnologiescomprisearangeofoccupations,suggestingthatAIhasthepotentialtoimpactworkersofallskilllevelsacrossmanyfirmsandsectors.

.Todate,thecasestudiessuggestthatemploymentlevelshaveremainedsteadyinthefaceofAIadoption,thoughthereisevidenceofslowedjobgrowth.ThecasestudiesshowlimitedevidenceofredundancieslinkedtoAI.Instead,intervieweesoftenemphasisedtheviewthat,nomatterhowadvancedAIbecomes,somejobsarealwayslikelytobebetterdonebyhumans,suchasthoseinvolvingempathy,socialinteractionandcertaintypesofdecisionmaking.InthelimitednumberofcasestudieswhereAIimplementationdidleadtojobdecreases,firmsreallocatedworkerstootherbusinessareasormanagedadjustmentsviaslowedhiringandattrition,leavingemploymentinspecificoccupationstodiminishgraduallyovertime,withretirees,inparticular,notbeingreplaced.

.RobustdemandforspecialisedAIskillsisdrivinggrowthinAI-relatedoccupations.ThecasestudiesshowevidenceofnewjobcreationinthefieldofAIitself.AItechnologiesmustbe

12DELSA/ELSA/WD/SEM(2023)7

Unclassified

built,trained,updated,andmaintained.Tomeetthesedemands,humanresourcemanagersfrequentlycitedeffortstohireworkerswithspecialisedAIskills.

.Jobreorganisationismoreprevalentthanjobdisplacement,withtasksshiftingtowardsthoseinwhichhumanworkershaveacomparativeadvantage.Inlimitedinstances,taskcompositionchangedlittle,ascomplementaryAItechnologiesenhancedworkers'abilities,enablingthemtoproduceproductsand/orservicestoabetterstandard(e.g.,faster,moreaccurately,moresafely)withoutchangingjobprofiles.Inagreaternumberofinstances,whenAIautomatedataskthatitcoulddomorequicklyorcheaply,demandforhumanworkerstodotheothertasksaroundtheAIrose.Theothertaskstendedtobethoseinwhichhumanworkersretainacomparativeadvantage,underscoringakeymechanismbywhichautomationaffectsthedemandforlabour.

.TheimplementationofAItechnologiesoftendemandsbothhigherskillsandbroadenedskillsets.ThepotentialforfirmsandworkerstoadapttoAIimplementationviajobreorganisationdependsonworkers'existingskilllevelsandthetrainingeffortsthatfirmsundertaketoupskillworkerswhennecessary.ThecasestudiesshowthatskillrequirementshaveoftenremainedunchangedfollowingtheintroductionofAI.Forexample,manyjobsreorganisedamongworkers'existingtaskmixesdonotrequirenewskills.However,inasignificantportionofcasestudies,skillrequirementsdidchange,withevidencethatAItechnologiesareleadingtohigherskillsdemands(e.g.,sharpenedanalyticalskills,improvedinterpersonalski