布魯蓋爾研究所-人工智能和就業(yè)的性別觀點(diǎn)：數(shù)字不平等的惡性循環(huán)

WORKINGPAPER|ISSUE15/2022|30AUGUST2022

AGENDERPERSPECTIVEON

ARTIFICIALINTELLIGENCEAND

JOBS:THEVICIOUSCYCLEOF

DIGITALINEQUALITY

ESTRELLAGOMEZ-HERRERAANDSABINET.KOESZEGI

Theworldwideartificialintelligencemarketisexpectedtoincreaseenormouslyin

thenextfewyears.BecauseofAI’simmensepotential,virtuallyallindustrieswill

beaffectedbytheimplementationofAIsystems,resultinginthedigitalisationand

automationofworkprocesses.Thiswillcausedisruptiveshiftsinlabourmarkets,

intermsofthenumberandprofilesofjobsinindustriesaswellasworkerskill

requirements.

Wetakeagenderperspectiveandanalysehowgenderstereotypesandgenderedwork

segregationontheonehand,anddigitalisationandautomation(asaconsequence

ofAIimplementation)ontheotherhand,areentangledandresultinaviciouscycle

ofdigitalgenderinequality.Weprovideinsightsintothegender-specificimpactofAI

technologies,whichisrelevantforthemitigationofthepotentialriskofthecreation

ofsocialinequalityandexclusion.Weshowthatexistingempiricalevidencealready

indicatesthatAIwillnotincreasegenderequalitybutwillsomewhatfurtherexacerbate

thegenderinequalityinlabourmarkets,rangingfromfurtherhorizontalandvertical

occupationalgendersegregationtoanincreaseinthegenderpaygap.Wesummarise

policyguidanceandmeasurestodecreasegenderinequalityinthefuture.

EstrellaGomez-HerreraisaVisitingFellowatBruegelandaProfessorattheUnversity

ofBalearicIslands

SabineT.KoeszegiisaVisitingFellowatBruegelandaProfessorofLaborScienceand

OrganizationInstituteofManagementScience,TUWien

Recommendedcitation:

Gomez-Herrera,E.andS.Koeszegi(2022)‘Agenderperspectiveonartificialintelligenceand

jobs:theviciouscycleofdigitalinequality’,WorkingPaper15/2022,Bruegel

1

1Whyagenderperspectiveonartificialintelligenceandjobsisneeded

EversinceDeepMind'sAlphaGobeatLeeSedol,theworld'sbestGoplayer,infouroutoffivegamesinMarch2016,thehypearoundartificialintelligence(AI)hasbeenhardtocontain.AlphaGo'sstrategieswerenotpre-programmed;instead,thesystemtaughtitselftoplaybymimickinghumanstrategiesandsubsequentlyusingreinforcementlearningincountlessgamesagainstitself.Inthishistoriccontestbetweenhumankindandalgorithm,humankindwasforthefirsttimeinferiortoamachine,andthenotionthatnothingcanbeathumanintelligence,creativityandintuitionwasshakenprofoundly.Thehuman“raceagainstthemachine”,atermcoinedbyErikBrynjolfssonandAndrewMcAfeeintheirbookontheinteractionofdigitaltechnology,employmentandorganisation(BrynjolfssonandMcAfee,2011),wasshiftedtoanewlevel.

Atthesametime,hopeshavebeenraisedthatAIwillhelpovercomehumanlimitationsandshortcomings.Essentially,AItechnologyissoftwarethatoperatesautonomouslywithoutdirecthumancontrol.Itisinteractiveandcanadapttoitsenvironment.AIsystemsusealgorithmstointerpretstructuredorunstructureddata.Thesealgorithmsresemblearecipe,includingformulatingaproblemandanobjective,aswellasthelogicalsequenceofstepstoorganise,processandanalysethedatasets.AIhasbeenclassifiedasageneral-purposetechnology(Agrawaletal,2019;BrynjolfssonandMcAfee,2017).TheworldwideAImarketisexpectedtogrowto$59billionby2025,representingasignificantincreasefromthe$1.8billionspentin2016(Servoz,2019).

AIsystemsareencounterednotonlyintheformofalgorithmsorautomateddecisionsystemsbutalsoinembodiedrobots.Aschatbots,voiceassistancesystems,servicerobots,collaborativerobots,autonomousvehiclesortoys,thesemachinescommunicatewithhumans,ofteninnaturallanguage,respondingtohumanbehaviourandadaptingtodifferentsituations.Theyfollowpre-programmedrulesandexpectedbehaviouralnormsandareperceivedassocialactors.Hence,‘AIsystems’referstoabroadrangeofintelligentmachines,embodiedornot,whicharealreadyrelevantinalmosteveryaspectoflifeandwillgainevenmoreimpactinthefuture.BecauseofAI’senormouspotential,virtuallyanyindustrywillbeaffectedbytheimplementationofAIsystems,drivingdigitalisationandautomationofworkprocesses.Hence,AIsystemscanprofoundlytransformjobs.Theirapplicationallowsworkprocessestobeautomatedonshopfloorsandinadministrationandcoremanagementtasks.Specificjobswillbecompletelyautomated,newjobswillemerge,andalmostalljobswillhaveatleastsomeexposuretoAItechnologyandautomation,requiringnew–digital–skills.Thiswillcausedisruptiveshiftsinlabourmarkets,bothinthenumbersandprofilesofjobsinindustriesandalsoinskillsrequirementsforworkers.UnderstandingtheimpactofAItechnologiesonlabourmarketsalsorequiresanunderstanding

S

o：gn〔o山g〔!ou,co山bn〔GL!2g〔!ouguqq!8!〔gl!2g〔!ouo：)op2.lu〔p!2bgbGL,MGgLGbgL〔!cnlgLll!u〔GLG2〔Gq!uMpG〔pGLVlg：：Gc〔2山GuguqMo山Guq!：：GLGu〔ll.MGgugll2GpoM8GuqGL2〔GLGo〔lbG2guq8GuqGLGqMoLk2G8LG8g〔!ou,ou〔pGouGpguq,guqq!8!〔gl!2g〔!ouguqgn〔o山g〔!ou(g2gcou2GdnGucGo：Vl!山blG山Gu〔g〔!ou),ou〔pGo〔pGLpguq,gLGGu〔gu8lGqguqLG2nl〔!ug^!c!on2clclGo：q!8!〔gl!uGdngl!〔l.

Ex!2〔!u88GuqGL!uGdngl!〔!G2Gxp!p!〔g2l2〔G山!cco山bouGu〔〔pg〔qoG2uo〔g：：Gc〔ouG2bGc!：!cg2bGc〔pn〔,ou〔pGcou〔LgLl,!2!u〔L!u2!c〔o〔pGMpolG2oc!gl2〔Lnc〔nLG,〔pn28GuGLg〔!u8!uGdngl!〔!G2g〔^GLlq!：：GLGu〔lG^Gl2.MGbLo^!qGgcoucGb〔ngl：Lg山GMoLk〔pg〔pGlb2〔onuqGL2〔guq〔pGco山blGx!〔lo：〔pGbLoplG山.MG!qGu〔!：l〔pG山g!u：gc〔oL2〔pg〔cou2〔!〔n〔Gg^!c!on2clclGo：q!8!〔gl!uGdngl!〔lguqLG!u：oLcGguq2〔LG22〔pG

Gx!2〔!u8q!2bgL!〔!G2.lp!2gugll2!2!2uGcG22gLl2!ucG山!〔!8g〔!u8bo〔Gu〔!glL!2k2g22oc!g〔GqM!〔pVl〔Gcpuolo8!G2,bgL〔!cnlgLll〔pGL!2ko：cLGg〔!u82oc!gl!uGdngl!〔!G2guqGxcln2!ou,LGdn!LG2!u2!8p〔2!u〔opoM8GuqGL!2!u2cL!pGq!u〔o〔Gcpuolo8!G2guq〔pG!Lgbbl!cg〔!ou2.lpGH!8p-lG^GlExbGL〔eLonbouVL〔!：!c!gl lu〔Gll!8GucG(H「EeVl)

T

cLGg〔Gqpl〔pGEnLobGguCo山山!22!ougqqLG22Gq〔p!2L!2ko：Gxcln2!ouo：bo〔Gu〔!glll^nluGLgplGbGL2ou2guq8Lonb2,2ncpg2Mo山Gu,!u〔pG!LG〔p!c28n!qGl!uG2ou〔Ln2〔MoL〔plVl.「!kGM!2G,〔pGH「EeVlLGdnG2〔2bol!cl山Gg2nLG2〔oGu2nLGGdnglobboL〔nu!〔!G2!ugccG22〔oGqncg〔!ouguq〔Gcpuolo8lg2g2np2〔gu〔!glq!山Gu2!ouo：〔Ln2〔MoL〔plVl2l2〔G山2.

lu2Gc〔!ouS,MGgugll2G一pg2GqounuqGL2〔guq!u8〔pGGu〔gu8lG山Gu〔o：8GuqGL,〔Gcpuolo8l,guqMoLk一〔pG!山bgc〔o：Vlou)op〔Lgu2：oL山g〔!ou：Lo山g8GuqGLbGL2bGc〔!^G.2Gc〔!ou3q!2cn22G2poM)op2gLG〔Lgu2：oL山Gq〔pLon8pVlguqMpG〔pGLMo山GugLG山oLGg：：Gc〔Gqpl〔pG2G〔Lgu2!〔!ou2.2Gc〔!ou寸on〔l!uG2poM2k!ll2LGdn!LG山Gu〔2cpgu8G〔pLon8pVlguqpoM8GuqGLGqGqncg〔!ou!山bgc〔2Mo山Gu,2gccG22〔o〔Lg!u!u8guq〔Gcpuolo8l,〔pn2uG8g〔!^Gll!山bgc〔!u8〔pG!LcgLGGLob〔!ou2.2Gc〔!ou2gugll2G2G山b!L!cglG^!qGucGouMpG〔pGLVl!ucLGg2G2〔pG8GuqGLbgl8gb.E!uglll,2Gc〔!ouebLobo2G2poMbol!cl2ponlqgqqLG22〔pGLg!2Gq!22nG2.

2Theviciouscycleofdigitalgenderinequality

nuqGLll!u88GuqGL!uGdngl!〔!G2gLGg^!c!on2clclGMp!cp!2pgLq〔opLGgkon〔o：.lonuqGL2〔guq〔pGMpolGbLoplG山,!〔!2uo〔2n：：!c!Gu〔〔ogugll2GlgponL山gLkG〔2oull.eGuqGL2〔GLGo〔lbG2guq!uGdngl!〔!G2!u2oc!G〔!G2gLG〔pGLoo〔cgn2G2o：GgLll2G8LG8g〔!ouo：8GuqGL2!uGqncg〔!ou2l2〔G山2,LG2nl〔!u8!uoullg：GM8!Ll2guqMo山Guob〔!u8：oLGqncg〔!ou!u2c!GucG,〔Gcpuolo8l,Gu8!uGGL!u8guq山g〔pG山g〔!c2(2lEW)guq!u：oL山g〔!ouguqco山山nu!cg〔!ou〔Gcpuolo8!G2(lCl).lpn2,oullLGlg〔!^Gll：GMMo山GuGu〔GL〔Gcp

T2GG

p〔〔b2:\\q!8!〔gl-2〔Lg〔G8l.Gc.GnLobg.Gn\Gu\bol!c!G2\GxbGL〔-8Lonb-g!

.

3

labourmarkets.Disparitiesinwomen'srepresentation,remunerationandpromotionmakeitextremelydifficulttoretainfemaletalentintechnology-relatedfields.Adisproportionatenumberofwomenleaveduringtheirtransitionfromhighereducationandcareercycles.Consequently,homogenous(male)developerteamsdesignAIsystemsandtheirapplications,potentiallyneglectingtheneedsofdiverseusersandperpetuatinggenderstereotypes.UNESCO(2019)pinpointedhowgenderedtechnologiesfurthercreateinequalitiesinaccesstotechnologiesandaggravategenderinequalitiesinsociety.Figure1visualisesthisviciouscycle.Understandinghowgenderisdeeplyinterwovenwithworkandculturalnormsiscrucial.Asustainableandeffectivestrategyagainstgenderdiscriminationmusttakemeasuresateverystageofthiscycle.

Figure1:Theviciouscycleofdigitalinequality

Source:Bruegel.

Inthefollowingsections,wedismantlethisentanglementandprovideempiricalevidencefromtheEuropeanUnion.

2.1GenderStereotypesandinequalitiesinsociety

Theviciouscycleofdigitalgenderinequalityisdeeplyrootedintheentanglementofworkandgender:“Inshort,doinggenderanddoingwork,whileanalyticallyseparable,appeartobeempiricallyintertwined”(Fenstermakeretal,2002,p.34).Tounderstandthegendereddivisionoflabour,itis

4

essentialtoscrutinisewhatisdoneathomeandwhatisdoneatwork,whodoeswhatandwhoispaidforwhat.Insteadofcastinggenderasanattributeofindividuals,WestandZimmerman(1987)reframedtheconceptofgenderas“anaccomplishment”.Theyconceivedgenderas“anemergentpropertyofsocialsituations:bothanoutcomeofandarationalefor…justifyingoneofthemostfundamentaldivisionsinsociety”(p.9).Thefundamentalshiftinourtheoreticalunderstandingofgenderfromanattributetoanaccomplishmentshiftstheattentionto‘doing’andthustogenderedprocessesmanifestingenderedstructures(Acker,1990).Genderhastobeaccomplishedthroughsocialinteractionsandisalsoaccomplishedbyhowwedistributeanddowork.

Oneofthemainareasofdebateoverwomen’spositioninthelabourmarketistherelationshipbetweenwomen’sroleinpaidandunpaidwork(Bagilhole,2002).TheEuropeanInstituteforGenderEquality(EIGE)summarisedincriticalfindingsongenderinequalitiesintheEUthatthereisadirectlinkbetweentheunequaldivisionofunpaidcareinhouseholdsandgenderinequalitiesinthelabourmarket(EIGE,2020a).Womendothebulkofunpaidworkinhouseholds:theydo2.6timestheamountofunpaidcareanddomesticworkthatmendo(UNWomen,2018).Consequently,theyhavelesstimetoaccessemploymentandgrowtheircareers.Furthermore,withinthelabourmarket,womenarefoundpredominantlyincertainoccupations,whichreflectthetraditionaldivisionofrolesinthedomesticsphere.Traditionalfeminineoccupations,includingchildcare,careforolderpeople,nursingandeducation,aresignificantlylesswell-paidthanconventionalmaleoccupations,suchasconstruction,manufacturingortechindustries.Theestablishedsocialsystemleadsgirlsandwomentoabsorbunconsciouslytheideathattheyaresupposedtohavecertainqualitiesassociatedwiththerolesofdomesticworkandcare.Atthesametime,boysandmenaresocialisedintorolesof‘breadwinners’.

2.2Inequalityandgendersegregationineducation

ThedifferentrolesocialisationsofboysandgirlsleadtoanoverallreductionoftalentinSTEMsubjects,underminingEuropeanindustry'scompetitiveness.However,thegendergapintheeducationalchoicesofgirlsandboysonlybecomesvisibleinsecondaryandtertiaryeducation.WhilegirlsandboysarenearlyequalintheirinterestinSTEMsubjectsinprimaryschool(UNESCO,2019),theirinterestdevelopsinsignificantlydifferentdirectionsatlaterstagesofeducation.Attheageof15,whenchoosingthefieldofspecialisationineducation,only0.5percentofgirls,while5percentofboys,wishtobecomeICTprofessionalsacrossOrganisationforEconomicCo-operationandDevelopmentcountries(OECD,2017).Thesegender-specificexpectationsexistindependentlyoftheactualperformanceofgirlsandboysinrelatedsubjectsatschool.Correll(2001,2004)showedthatculturalbeliefsaboutgenderskewtheperceptionsofgirls’competenciesandconstraintheircareeraspirations.Compellingevidenceofthe

5

impactofgenderstereotypesisfoundintheself-efficacygapinmathematicsanddigitalcompetencies,ordifferencesbetweengirls’andboys’confidenceandbeliefintheirabilities.Althoughgirlsandboysdosimilarlywellinmathsanddigitalliteracytestsuntilpuberty,girlsratetheirskillssignificantlylower(UNESCO,2019).

Figure2:ICTabilitiesofgirlsandtheirperceptionofabilities

Source:UNESCO(2019).Figure2showsthatdespitestrongperformanceincomputerandinformationliteracy,girlsdonothaveconfidenceintheirICTabilities.Unfortunately,thelackofself-confidence,structuralbarriersandprevailinggenderstereotypesleadtoselectiveeducationalchoices.Girlspredominantlychoosehumanitiesandsocialsciencemajors,whileboyschoosecomputerscienceandengineeringstudies.Theglobalproportionoffemaleenrolmentsineducationis70percent,inhealthandwelfare69percent,inartandhumanities61percent,innaturalsciences(includingbiology)56percent,inSTEM(average)36percentandICT29percent(EqualsResearchGroupcitedinUNESCO,2019,p.23).InEurope,the

numbersareevenlower:Only34percentofSTEMgraduatesand17percentofICTgraduatesarefemale(EuropeanCommission,2021).Only2.4percentoffemaletertiarygraduatesearnICTdegreesversus9.2percentofmaletertiarygraduates.

6

2.3Gendersegregationinprofessions

Althoughwomencan,inmostplacesinprinciple,enteranyprofessiontheywant,thesegregationofmenandwomenisreinforcedinoccupations.Professionalaspirationsareincorporatedintheindividualself-imageschildrendevelopduringsocialisationfromearlychildhoodthroughadolescence.Whencomparingone’sself-imagewiththeimageofaprofession,genderplaysacrucialroleincareerchoices.Whenthegenderimageofaprofessiondoesnotmatchtheself-concept,theattractivenessofoccupationsandindividualinterestsmaybeoverruledbyfeelingsofinadequacy(Haasetal,2016).Andevenifwomenaspiretomasculinejobs,social,culturalandstructuralbarriersforwomencomplicatecareersintheseoccupations.Asskillsandcompetenciesneededtodoaspecificjobareinherentlytiedupwithmasculinityandfemininity,genderisusedasadiscriminantcriterionforhiringinsteadoflookingattheactualcompetenciesofpotentialemployees(Lorber,1994,p.199).

Unfortunately,onceinthefield,womenareagainexposedtoculturalbiaseswithinprofessionsthatcontributetopatternsofretentionandattrition(Carrolletal,2016).Professionalsocialisationentailsnotonlythemasteryofskillsandspecialisedknowledgeoftheprofessionbutalsorequiresamatchbetweentheprofession’svaluesandpersonalvaluesandself-conceptions.Carrolletal(2016)showedhowsocialisationprocessesintheengineeringprofessionleadwomentodeveloplessconfidencethattheyfitintotheengineeringculture.Initiationritualsincoursework,informalinteractionswithpeersandeverydaysexisminteamworkandinternshipsareparticularlysalientbuildingblocksofgendersegregation.EmilyChang(2018),aSiliconValleyinsider,usedtheterm"brotopia"forthisphenomenon.Shedescribedhowwomenfacetoxicworkplaceswithdiscriminationandsexualharassment.Shearguedthatthe“aggressive,misogynistic,work-at-allcostsbro-culture”excludeswomenfromtechnologydevelopmentandaccess.

Apartfromhorizontalsegregationintospecificoccupations,womenalsoexperienceaglassceilingeffectresultinginverticalgendersegregation.WomeninSTEMfieldsandthedigitalsectorarelesslikelytoholdhigh-levelpositions.AccordingtoUNESCO(2019),onlyoneineveryfourleadershippositionsintechindustries(includingnon-technicalpositionsinmarketing,human-resourcemanagementandthelike)isoccupiedbyawoman.Chang(2018)citedculturalandstructuralbarriersforwomenastheleadingcausesofthisphenomenon.

ICToccupationsarealsoheldprimarilybymen(EIGE,2020b).IntheEuropeanUnion,only17percentofICTspecialistsarewomen.Althoughthesefiguresvaryacrosscountries,withabalancedpictureinRomaniaandLatvia,significanteffortsarerequiredinallEUcountriestoreducesegregation.The

7

situationisnotbetterforAI.Inthisfield,menrepresent84percentoftheEUworkforce,withonlya16percentshareofAI-skilledwomen.Overthecareerpath,thedifferencegetsevenmoreexpansive.InpositionsthatrequiremorethantenyearsofAIexperience,only12percentareoccupiedbywomen(EIGE,2021).Notably,ICTandAIjobsrepresentasignificantproportionofEUemployment,andthedemandforthesespecialistsisonlyexpectedtogrow.ThelowentrynumbersofwomenintothefieldandthedifficultiesinprogressingwiththeircareersonceinthearearesultinratherahomogeneousAIdeveloperteams,whichtendtoreproduceandperpetuategenderbiasesinthetechnologiestheydevelop.

2.4Inequality-reproducingtechnologiesandtechnologyaccess

AItechnologiesareneitherobjectivenorgender-neutral(Tufekci,2015).ThequalityofAIapplicationsdependsmainlyonthequalityofthetrainingdata(indata-drivensystems),themodelling(whatwerefertoasthealgorithm),design(voice,shapeandothercharacteristicsofembodiment)andtheactualimplementationofthesysteminthespecificcontext.

Hence,AIsystemdesignersdeterminewhichdataandparametersarerelevantforthetrainingofthesystem,theydecideontheoperationalisationofperformanceindicatorsandgoals,andtheyalsodecideontheAIsystem’sappearanceinitsembodiments,theirnames,voicesandcharacters,andontherolesandtaskstheyshouldtakeon.

Consequently,theinitialsocialjudgmentofsystemdesignersismathematicallyspecifiedinalgorithms,strategicgoalsandindicatorsformeasuringsystems'success.Thus,algorithmsorAIsystemsrefer,infacttoanundefinednetworkofsocio-technicalarrangementsinwhichtheinvolvementofhumansremainshiddenateverystepoftheprocess.Theterms‘a(chǎn)lgorithm’and‘AIsystem’obscurethatcultural,societalandpoliticalvalues–andwiththem,potentialdiscriminationandbias–areinherentinAIsystems.O'Neil(2016,p.53)expressedthispointedly:"Analgorithmisnothingmorethananopinionformulatedinaprogramminglanguage."Hence,stereotypicalnotionsofwomenandmenandtheirtasksandrolesinsocietyarereflectedbythemachinesdesignedinengineeringlabs.

TheBerkeleyHaasCenterforEquity,GenderandLeadership,trackedavailableinstancesofbiasinAIsystems(Smithetal,2021).Theiranalysisof133biasedsystemsacrossindustriesfrom1988to2021foundthatalmosteverysecondanalysedsystemdemonstratedgenderbias,andeveryfourthsystemexhibitedbothgenderandracialdiscrimination.TrainingdataforAIsystemsplaysacrucialroleinthis.Accordingtotheprinciplegarbage-in/garbage-out,AIsystemslearnwhatisintrainingdataandoncelearnedbyanalgorithm,willneverforget.Smithetal(2021)identifiedseveralcriticalimpactsofgender-

8

biasedAI.First,genderbiasresultsinlowqualityforwomenandnon-binaryindividuals.Forexample,voicerecognitionsystems,increasinglyusedinmanyproductsandservices,fromautonomouscarstohealthcareproducts,oftenworklesswellforwomen’svoices.Suchproductsorservicescouldeventhreatenwomen'sphysicalandmentalwellbeing.ExamplesareAIsystemssupportinghealthdiagnosesbasedonhealthdatathatdoesnotrepresentgenderappropriately.Second,genderbiasinAIsystemsleadstounfairallocationofresources,informationandopportunitiesforwomen.Forinstance,systemsusedinrecruitingandhiringdeprioritisewomen’sapplications(seealsoBogenetal,2018).Third,genderbiasalsoperpetuatesexisting,harmfulstereotypesandprejudicesandleadstoa“derogatoryandoffensivetreatmentorerasureofalreadymarginalisedgenderidentities”(Smithetal,2021).

Smithetal(2021)referredheretoexamplesoftranslationsoftwareortheuseofthegenderbinaryingenderclassification,whichbuildsaninaccurate,simplisticviewofgenderintoolssuchasfacial-analysissystems.Hence,genderedAIsystemsimpactindividualsandalsocontributetosetbacksingenderequalityandwomen’sempowermentinsocieties.SaniyeGülserCorat,UNESCO'sDirectorforGenderEquality,wrote,"obedientandcompliantmachinesposingaswomenenterourhomes,carsandoffices.Theirhard-wiredsubservienceaffectshowpeoplespeaktofemalevoicesandhowwomenrespondandexpressthemselvesinresponsetorequests.Tochangecourse,weneedtopaymuchmoreattentiontohow,when,andifAItechnologiesaregenderedandwhoisgenderingthem"(UNESCO,2019).

Giventhis,itisunsurprisingthataccesstotechnologiesisalsodistributedunevenlyacrossgender(UNESCO,2019).Notably,theaccessgapcannolongerbeattributedtotechnologyprices.Therehasbeenasignificantdeclineinthepriceofconnectivityandhardware,whichhasnotbeentranslatedintoareducedgendergap(UNESCO,2019).Thisproblemisevenmoresevereindevelopingcountries.Availablestatisticsshowthatwomenareabout50percentlesslikelythanmentobeconnected,controllingforage,educationandincome(WorldWideWeb,2015).Althoughthisproblemismoresevereinless-developedcountries,therearealsonotabledifferencesinEuropeacrossregions.WhilewesternEuropehasnarrowedthedigitalusagegendergap,centralandeasternEuropeancountriespresentanaveragegapof3percent,whichisevenhigherforGreece.Thiscomesasaconsequenceofskillsandeducationbeingthemostcriticaldeterminantsoftechnologicalaccess.Inparallel,experiencewiththeuseoftechnologycontributestoabetterunderstandingofitsbenefitsandhencetoagreaterinterestinaccessingtherequiredskills.

9

Further,thelackofskillsfostersgenderstereotypesandgenderedworksegregationofpaidandunpaidwork.Consequently,thecycleclosesinevitablyagain.Differentcomponentsintheviciouscycleinfluenceeachother,contributingoveralltothepersistenceofinequalitiesovertime.

ToensureasmoothtechnologicaltransitionandtheequaldistributionofbenefitsderivedfromAI,properdiagnosisandunderstandingoftheelementsinthecyclethatrequirefurtheractionisneeded.Inthefollowingsection,weprovideaspecificanalysisofthetransformationofjobs,skillsandjobsegregationandthegenderpaygap.

3AIandthetransformationofjobs

OneofthemostsignificantimpactsofAIrelatestothelabourmarket.First,AIisexpectedtoimpactnetjobcreationsignificantly.Theso-called‘riskofautomation’ispresentinalldiscussionsaboutthefutureofwork.ThemorepessimistichavepredictedarateofdestructionofjobsbecauseofAIandautomationrangingfrom30percent(PwC,2018)to47percent(FreyandOsborne,2013).Inaddition,morethan60percentofcompanieshaveacceleratedtheirautomationandAIcapabilitybuildingasareactiontotheCOVID-19pandemicMcKinsey(2021).ThiscombinationofautomationandAIhasbeencalled‘intelligentautomation’andimpliesthereplacementofasignificantproportionoftasks.Inparallel,arangeofnew,previouslynot-existing,jobsarebeingcreated.Theserelatetotheneedtoimplementanddevelopnewtechnologies.

AsecondimportantimplicationofAIisthetransformationofalmostalloccupations,atleasttosomedegree.Asageneral-purposetechnology,theimpactofAIhasatransversalnature,affectingallsectorsoftheeconomy.Anessentialchallengeforpolicymakingisthusfacilitatingtheproperintegrationbetweenmachineandhumancapacityinthelabourmarket,i.e.toendowworkerswiththerequiredabilitiesandskillstoproperlyusethenewtechnologies.Theproportionoftimedevotedtoworkbyhumansandmachineswillbearound50percenteachin2022(WEF,2020).Figure3showsthischangefordifferenttasks.

10

Figure3:Ratioofhumanvsmachineworkinghours,2018vs2022

Source:BruegelbasedonWEF(2020).

Avoidingasymmetriesorexclusionsisessentialwhenhelpingdifferentgroupsofworkersadapttothesenewsetsoftasks.Thiscanbeachievedbyreskillingandupskillingworkersaffectedbythechangewhilereducingthebarrierstoentryintonewjobs.Evenifmostofthepredictionsaboutthefutureofworkareoptimistic,thetransitionwillbechallenging,andthereisastrongneedforpolicyguidance.Pastepisodesofeconomicandsocialintegrationofnewtechnologiestookseveralyearsorevendecades.Thesuddendisappearanceofapercentageofjobsandthecreationofanewsetofjobsimpliessignificanteconomicdisruption.Itcouldaffectdifferentgroupsofpeopleunevenly,withamoresubstantialimpactongroupsatriskofexclusion.Theadjustmentisexpectedtobeslow,thusgeneratingamismatchofskillsandtechnologiesintheshortandmediumruns.Therefore,fromapolicyperspective,itisessentialtolooknotonlyattheoverallemploymentfiguresbutalsoatthefiguresatadisaggregatedlevel(regions,sectors,agegroups,etc).SucceedingintheAIintegrationprocessrequiresthesocietalandmarkettransformationtobeadequatelyaddressedataprofoundlevel.ThesystematicnatureofthegenderinequalityproblemreflectedinallstagesoftheviciouscycleisreinforcedbytherapidgrowthofAI.ThegendersegregationwomenfaceintheeducationphaselatercreatesadisadvantagefortheminaccesstoAIjobs,thuscontributingtogendersegregationwithinandbetweenprofessions.

TodeterminetheextenttowhichwomenaremorenegativelyaffectedbytheAItransformationthanmen