Mindthe

AIDivide

ShapingaGlobalPerspective

ontheFutureofWork

MindtheAIDivide:ShapingaGlobalPerspective

ontheFutureofWork

Copyright?2024UnitedNations

Allrightsreservedworldwide.

Nopartofthispublicationmay,forcommercialpurposes,bereproducedortransmitted

inanyformorbyanymeans,electronicormechanical,includingphotocopy,recordingor

anyinformationstorageandretrievalsystemnowknownortobeinvented,withoutwritten

permissionbythepublisher.

RequeststoreproduceexcerptsortophotocopyshouldbeaddressedtotheCopyright

ClearanceCenterat.

Allotherqueriesonrightsandlicenses,includingsubsidiaryrights,shouldbeaddressedto:

UnitedNationsPublications,405East42ndStreet,S-11FW001,NewYork,NY10017,United

StatesofAmerica.Email:permissions@.Website:.

Thedesignationsemployedandthepresentationofthematerialinthispublicationdonot

implytheexpressionofanyopinionwhatsoeveronthepartoftheSecretariatoftheUnited

Nationsconcerningthelegalstatusofanycountry.

PDFISBN:9789211066524

Foreword

TheunevenadoptionofArti?cialIntelligence(AI)isacriticalissuethatgoesbeyondeconomic

growth.Itimpactsglobalequity,fairnessandthesocialcontractthatisattheheartofsocialjustice.

Disparitiesinaccesstorobustinfrastructure,advancedtechnology,qualityeducationandtraining

aredeepeningexistinginequalities.AstheglobaleconomyincreasinglyshiftstowardsAI-driven

productionandinnovation,lessdevelopedcountriesriskbeingleftfurtherbehind,exacerbating

economicandsocialdivides.Withouttargetedandconcertedeffortstobridgethisdigitaldivide,

AI’spotentialtofostersustainabledevelopmentandalleviatepovertywillremainunrealized,leaving

signi?cantportionsoftheglobalpopulationdisadvantagedintherapidlyevolvingtechnological

landscape.

DuringtheconsultationsheldbytheSecretary-General’sHigh-levelAdvisoryBodyonArti?cial

Intelligence,ithasbecomeclearthattheworldofworkisattheheartoftheadoptionofAI.Itis

thuscriticaltounderstandthepotentialforAItoaffectlabourdemandandtransformoccupations.

Itisattheworkplacewherethepotentialforproductivitygainsandimprovedworkingconditions

forthebene?tofworkers,theirfamilies,andsocietiesatlarge,canberealized.Butsuchbene?ts

willnothappenontheirown;theywillonlybeachievediftherightconditionsareinplace,including

theavailabilityofdigitalinfrastructureandskills,butalsoacultureofsocialdialoguethatfostersa

positiveintegrationoftechnology.

PromotinginclusivegrowthrequiresproactivestrategiestosupportAIdevelopmentincountrieson

thewrongsideoftheAIdivide.Thisinvolvesenhancingdigitalinfrastructure,promotingtechnology

transfer,buildingAIskills,andensuringthatalljobsalongtheAIvaluechainareofgoodqualityand

improvethelivesofworkingpeople.ByprioritizinginternationalcollaborationinAIcapacitybuilding,

wecancreateamoreequitableandresilientAIecosystem,unlockingopportunitiesforshared

prosperityandhumanadvancementworldwide.

WelookforwardtocontinuingourcollaborativeeffortstoshapetheglobalgovernanceofAI,uphold

humandignityandlaborstandards,andexpandeconomicopportunityforall.

AmandeepSinghGillGilbertF.Houngbo

UnitedNationsSecretary-General’s

EnvoyonTechnology

Director-GeneraloftheInternational

LabourOrganization

MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork|3

Contents

Foreword

3

Section1.Introduction

5

Section2.Unevenground:UnderstandingAI’sroleinreshapinglabourmarkets

6

Ensuringjobqualityunderaugmentation

10

Section3.TheAIvaluechainandthedemandforskills

11

AdaptingskillsfortheAIlandscape

14

Section4.Movingforward:Strengtheninginternationalcooperation,building

nationalcapacity,andaddressingAIintheworldofwork

17

StrengthenedinternationalcooperationonAI

17

BuildingnationalAIcapacity

18

TowardsapositiveintegrationofAIintheworldofwork

18

Acknowledgments

20

References

21

4|MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork

Section1

Introduction

TherapidadvancementofArti?cialIntelligence

(AI)promiseswidespreadtransformations

foroursocieties,oureconomiesandthe

worldofwork.Whilesuchadvancesoffer

tremendousopportunitiesforinnovationand

productivity,theunevenratesofinvestment,

adoptionanduseamongcountriesrisks

exacerbatingthealreadywidedisparities

inincomeandqualityoflife.Thereisa

pronounced“AIdivide”emerging,wherehigh

incomenationsdisproportionatelybene?tfrom

AIadvancements,whilelow-andmedium-

incomecountries,particularlyinAfrica,lag

behind.Worse,thisdividewillgrowunless

concertedactionistakentofosterinternational

cooperationinsupportofdevelopingcountries.

Theabsenceofsuchpolicieswillnotonly

widenglobalinequalities,butriskssquandering

thepotentialofAItoserveasacatalystfor

widespreadsocialandeconomicprogress.

WhileAIwillpotentiallyaffectmanyaspects

ofourdailylives,itsimpactislikelytobemost

acuteintheworkplace.Wealthiercountries

aremoreexposedtothepotentialautomating

effectsofAIintheworldofwork,buttheyare

alsobetterpositionedtorealizetheproductivity

gainsitoffers.Developingcountries,onthe

otherhand,maybetemporarilybuffered

becauseofalackofdigitalinfrastructure,but

thisbufferrisksturningintoabottleneckfor

productivitygrowth,andmoreimportantly,for

thefutureprosperityoftheirpopulations.

Ensuringinclusivegrowthinthefuture

requiresproactivemeasurestoempowerAI

developmentincountriesatthedisadvantaged

receivingendofthedigitaldivide,fostering

digitalinfrastructureaswellasAIskills,and

promotingtechnologytransferandabsorption.

Suchdigitalskillscanalsoenableamore

positiveintegrationofAIintheworkplace,

particularlywhencombinedwithsocial

dialogue.Socialdialogueonthedesign,

implementationanduseoftechnologyatthe

workplace,aswellasinthedevelopmentof

regulationsessentialforensuringrespect

ofworkers’fundamentalrights,isneeded.

Indeed,whethertheintegrationoftechnology

intoworkprocessesspursproductivitygrowth

orimprovesworkingconditionsinsupport

ofsocialjusticedependsinlargepartonthe

strengthofsuchcollaborationanddialogue.

Sovereigneffortsplayacrucialroleinshaping

AIcapacitybuildingascountriesassert

theirautonomyindevelopingstrategies

andpoliciestailoredtotheiruniquesocio-

economiccontexts.Localprocesses,driven

byculturalvalues,politicaleconomies,and

societalneeds,cansigni?cantlyimpactthe

effectivenessandsustainabilityofAIinitiatives.

However,disparitiesinresourcesandexpertise

remainandcanhinderAIdevelopmentinthe

GlobalSouth.Inresponse,thereisagrowing

recognitionoftheresponsibilityofdeveloped

countriestosupportcapacitybuildingefforts

inresourcescarcecountries.Asoutlined

intherecentInterimReportoftheUnited

NationsSecretary-General’sHigh-levelAdvisory

BodyonAI1,thisrecognitionextendsbeyond

?nancialaidtoincludeknowledgesharing,

skillsdevelopment,technologytransfer,aswell

ascollaborativeresearchanddevelopment

partnerships.Byleveragingtheiradvanced

AIecosystems,theGlobalNorthcanhelp

bridgethegapandempowercountriesinthe

GlobalSouthtoharnessAIforsustainable

development,whilerespectingtheirsovereignty

andpromotinglocalinnovationecosystems.By

prioritizingglobalcollaborationforAIcapacity

building,theinternationalcommunitycan

nurtureamoreequitableandresilientglobalAI

ecosystem,unlockingopportunitiesforshared

prosperityandhuman?ourishingacrossthe

world.

1/ai-advisory-body

MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork|5

Section2

Unevenground

UnderstandingAI’sroleinreshapinglabourmarkets

Researchonthepossibleeffectsofgenerative

AIonemploymentacrosstheworldsuggests

thatwhiletherearelikelytobeimportant

transformativeeffectsonsomeoccupations,

impactsintermsofjoblossesaremuchless

thanheadline?guresappearinginthemedia,

andcertainlydonotpointtoajoblessfuture.

Accordingtoananalysisundertakenbythe

InternationalLabourOrganizationonthe

potentialexposureoftaskstogenerativeAI

technology,clericalsupportworkersarethe

mostexposedoccupationalgroupwith24

percentofthetasksinthesejobsassociated

withhighlevelofexposuretoautomation

andanother58percentwithmedium-level

exposure(seeFigure1).2Otheroccupational

groupsarelessexposed,withonly1to4

percentoftasksconsideredashavinghigh

automationpotential,andmedium-exposed

tasksnotexceeding25percent.Thismeans

that,whilecertaintasksintheseoccupations

couldpotentiallybeautomated,mosttasks

stillrequirehumanintervention.Suchpartial

automationcouldenableef?ciencygains,by

allowinghumanstospendmoretimeonother

areasofwork.

Importantly,taskautomationdoesnot

necessarilyimplyredundancies,asthe

technologycanalsocomplementoraugment

humanlabourwhenonlycertaintasksare

automated.Whethertheadoptionofthe

technologyleadstoautomation(jobloss)or

augmentation(jobcomplementarity)depends

onthecentralityoftheautomatedtasktothe

occupation,howthetechnologyisintegrated

Figure1:Taskswithmediumandhigh-levelexposuretogenerativeAItechnologybymajor

occupationalgroup(ISCO1-digit)

Source:Gmyreketal.,2023.

2Thestudyanalysesthepotentialforautomationwiththe436internationallystandardizedISCO-08occupationsand

thenclassi?estheoccupationbasedonthemeanandstandarddeviationofthescore.Formoredetailssee[1].

6|MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork

intoworkprocessesandmanagement’s

desiretoretainhumanstoperformoroversee

someofthetasks,despitethepotentialof

automation.

TheILOanalysisusesoccupationalexposure

scores(themeanexposureofeachofthe

taskswithinanoccupation)andappliesthese

scorestoemploymentdatafromlabourforce

surveysofmorethan140countriestoassess

potentialemploymentimpactattheglobal

andregionallevel.Withrespecttoautomation,

theshareofemploymentthatisexposed

ishighestinEuropeandNorthernAmerica,

re?ectingthegreatereconomicandlabour

marketdiversi?cationoftheseregions.In

LatinAmerica,AsiaandAfrica,theshareof

employmentpotentialexposedtoautomation

ismuchsmaller,duetothegreatershareof

workersemployedinoccupationsthatwould

notbeexposedtogenerativeAItechnology

suchasinagriculture,transportorfood

vending.

Nevertheless,women’spotentialexposure

totheautomatingeffectsofgenerative

AItechnologyismuchhigher,duetotheir

over-representationinclericaloccupations

(see?gure2).Inmostregions,thepotential

exposureofwomenismorethandoublethatof

men’sexposure.Someofthisemploymentisin

businessprocessoutsourcing,suchascontact

orcallcenterwork,whichisanimportantpart

oftheeconomyofseveraldevelopingcountries,

includingIndiaandthePhilippines.Theindustry

isanimportantsourceofformalandrelatively

well-paidemployment,particularlyforwomen.

Whilepotentialexposuredoesnotnecessarily

translatetodisplacement,itisclearthatthe

advancesintechnologymayputsomeofthese

jobsatrisk.3

Another?ndingisthatasigni?cantlylarger

shareoftotalemploymentisinoccupations

withhighaugmentationpotential,andthis

holdsacrossregions,from10.2percent

inSub-SaharanAfricato16.1percentin

SoutheasternAsiaandthePaci?c(See?gure

3).Thus,thepotentialforoccupationsto

bene?tfromtheproductivity-enhancingeffects

ofthetechnologyisrelativelysimilaracross

countries.Inpractice,however,itislesslikely

Figure2:Potentialexposuretoautomationbyglobalsub-region

3‘AICouldKilloffMostCallCentres,SaysTataConsultancyServicesHead’,April25,2024.

MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork|7

Figure3:Potentialexposuretoaugmentationbyglobalsub-region

toberealizedduetoconstraintsinphysical

infrastructure(electricityaccess,broadband)

aswellasdigitalskills.Indeed,subsequent

researchthatincorporatesdataoncomputer

useatwork[2]revealsthatmanyofthe

occupationswithpotentialforaugmentation

haverelativelylowusageofcomputeratwork,

suggestingthattheconditionsarenotinplace

forrealizingthepotentialproductivitygains.

AscanbeseeninFigure4,theshareof

workerswithoutaccesstoacomputeratwork

(“nocomputer”)exceedsthosewhousea

computerin9ofthe16countrieslisted.As

Figure4:Potentialexposuretoaugmentationandcomputeruseatwork

Source:Gmyrek,WinklerandGarganta,2024.

8|MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork

such,thelikelihoodtorealizeproductivitygains

fromgenerativeAItechnologywillbelimited.

Figure5givesinformationonthe

characteristicsofthosewhomightbeaffected

byautomationfromgenerativeAItechnologyin

LatinAmerica.Asthedatashow,itiseducated

womenlivinginurbanareasandbelonging

tothetop?fthoftheincomedistributionthat

aremostexposed.ForLatinAmerica,these

occupationsareoverwhelminglyinsalaried,

formalemploymentandinthesectorsof

?nance,professionalservicesandpublic

administration.Inshort,theyaregoodjobs,

whoselosswouldhavenegativemultiplier

effectsbotheconomicallyandsocially.

Theanalysisdoesnotaddressthepotentialfor

newjobcreation.Thus,whilemiddle-income

countriessuchasIndiaandthePhilippines,

aremoreexposedtotheautomatingeffects

ofgenerativeAItechnologyintheircallcentre

work,theirdigitalinfrastructureandskilled

workforcecanalsobeanassetforspawning

thegrowthofcomplementaryindustries.

Harnessingsuchpotentialisparamount.

Indeed,withtherightconditionsinplace,a

newwaveoftechnologycouldfuelgrowth

opportunities.Inthepast,technological

advancementshavespurrednewand

successfulindustriesinmanydeveloping

countries.OnesuchexampleistheM-Pesa

moneyservice,whichreliedonthediffusion

ofmobiletelephonesinKenya.Theservice,

inturn,increased?nancialinclusionwhich

helpedtopropelthegrowthofSMEsandled

tocreationofanetworkof110,000agents,

40timesthenumberofbankATMsinKenya

[3];[4].Similarly,astudyofthediffusionof3G

coverageinRwandabetween2002and2019

foundthatincreasedmobileinternetcoverage

Figure5:Characteristicsofpersonsholdingoccupationsmostexposedtoautomation,

LatinAmerica

Source:Gmyrek,WinklerandGarganta,2024(forthcoming).

MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork|9

waspositivelyassociatedwithemployment

growth,increasingbothskilledandunskilled

occupations[5].Scholars[6]also?ndpositive

employmenteffects,fromthearrivalofinternet

in12Africancountries,albeitwithaslight

biastowardsskilledoccupations.Thesegains

areattributedtoincreasesinproductivityand

growthofmarketsthatfollowedincreased

connectivity,underliningtheneedforsuch

investments,givenimportantmultipliereffects

ontheeconomyandlabourmarkets.

Ensuringjobqualityunder

augmentation

Anotherareaofconcernisabouttheimpact

ofAItechnologyonworkingconditionsand

jobqualitywhenthetechnologyisintegrated

intotheworkplace.Whilesuchintegration

intoworktaskscanpotentiallypromotemore

engagingworkifroutinetasksareautomated,

itcanalsobeimplementedinwaysthat

limitsworkers’agencyoraccelerateswork

intensity.ConcernsoverAI’sintegrationat

theworkplacehasfocusedonthegrowthof

algorithmicmanagement,essentiallywork

settingsinwhich“humanjobsareassigned,

optimized,andevaluatedthroughalgorithms

andtrackeddata”[7].Algorithmicmanagement

isade?ningfeatureofdigitallabourplatforms,

butitisalsopervasiveinof?ineindustries

suchasthewarehousingandlogisticssectors.

Inwarehousesanautomated,“voice-picking”

systemdirectswarehousestafftopickcertain

productsinthewarehouse,whileusingdata

collectiontomonitorworkersandsetthe

paceofwork[8].Besideslackingautonomyto

organizetheirworkorsetitspace,workersalso

havelittleabilitytoprovidefeedbackordiscuss

withmanagementabouttheorganizationof

work[9].TheintegrationofgenerativeAIinto

other?eldssuchasbanking,insurance,social

services,andcustomerservicemorebroadly

mayhaveasimilareffect.

Technologicaladvancementsareoftenfelt

moreimmediatelyattheworkplaceleveland

areusuallybestaddressedattheworkplace.

Asaresult,whethertheeffectoftechnology

onworkingconditionsispositiveornegative

dependsinlargepartonthevoicethatworkers

haveinthedesign,implementationanduseof

technology.Havingsuchagencyreliesinturn

ontheopportunitiesforworkerparticipation

anddialogue.Thiscantakeplaceeither

throughformalizedsettings,suchasworks

councilsorguidanceprovidedincollective

bargainingagreements,orlessformally,in

workplaceswherethereisahighdegreeof

employeeengagement.StudiesinEurope

haveshownthatitiscountrieswithstronger

andmorecooperativeformsofworkplace

consultation,essentiallytheNordiccountries

andGermany,whereworkersaremoreopento

technologicaladoptionattheworkplace[10].

10|MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork

Section3

TheAIvaluechainand

thedemandforskills

Liketheproductionofmanygoodsand

servicesintheglobaleconomy,AIhasitsown

valuechain.AsdepictedinFigure6,thereare

differentstagesoftheAIvaluechain,eachwith

speci?chumanandsocialinfrastructureneeds.

Asistypicalinmostglobalvaluechains,stages

differintheamountofvaluereceivedforthe

contributionmade,withlower-valueadded

activitiespredominantinmiddleandlow-

incomecountriesanddesignanddeployment

associatedwithhigher-incomecountries.

Dataisfundamentaltothedevelopmentand

operationofAIsystems.Human-prepared

dataisfedintoAIsystemstohelpthemlearn

thenecessaryconnectionsandpatternsfor

functionality.Thesourcesofthisdataare

diverse,dependingonthesystem’spurpose.

Publiclyavailabledata,suchasUnitedNations

documentsusedfortrainingtranslation

programs,contributedtoadvancesinnatural

languageprocessing.Proprietarydataisalso

crucial,particularlyinworkplaceapplications,

likecallcenterrecordingsusedtotrain

chatbotsforcustomerservice.Withglobal

connectivity,datacollectioncontinuesto

providetheessentialrawmaterialforfutureAI

applications.

Whendataiscollected,itisusually

unstructured.Highlyskilleddataengineers

willpre-processthedataintoausableformat,

but‘datalabelers’areneededtolabeland

classifydatasothatitisusable.Labelled

andannotateddatasetsarecriticalforthe

developmentandeffectivenessofmachine

learningmodels.Workersinvolvedindata

enrichmentcarryoutanarrayoftasksthat

includemarkingradiologyscanstoaidin

creatingAIsystemscapableofdetecting

cancer;categorizingtoxicandunsuitable

onlinecontenttoimprovecontentmoderation

algorithmsordiminishthenegativityinlarge

languagemodelresponses;annotating

videofootagefromdrivingsessionstotrain

autonomousvehicles;editinglargelanguage

modeloutputstoboosttheirfunctionality;and

more.4

Contentmoderationistheprocessof

monitoringand?lteringuser-generated

contentondigitalplatforms,suchassocial

media,forums,andwebsites,toensurethat

itcomplieswiththeplatform’sguidelinesand

policies.Thegoalofcontentmoderationis

tomaintainasafe,respectful,andpositive

environmentforallusersbyremovingor

Figure6:ValuechainofAI

1234567

Note:Orangerepresentstheactivitiesthathavelowervalue-added.

Source:Authors’elaboration.

4ValuingDataEnrichmentWorkers:TheCaseforaHuman-CentricApproachtoAIDevelopment|UnitedNations

MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork|11

?aggingcontentthatisinappropriate,offensive,

harmful,orillegal.Contentmoderationcanbe

performedmanuallybyhumanmoderators

orautomaticallybyusingalgorithmsand

machinelearningtools.Thetypesofcontent

thatmaybesubjecttomoderationcanvary

widely,includingbutnotlimitedtohate

speech,harassment,violence,nudity,andfalse

information.Evenwiththeuseofalgorithms

andmachinelearningtoolsforcontent

moderation,thereistypicallyalwaysahuman

involvedintheprocess.Thesetechnologies

canhelpautomateandscalethemoderation

process,buttheyarenotperfectandcan

sometimesmakemistakesormissnuances

thatahumanmoderatorwouldbeabletopick

upon.

Inmanycases,algorithmsareusedto?ag

orprioritizecontentforreviewbyhuman

moderators,whothenmakethe?naldecision

onwhetherthecontentshouldberemovedor

allowedtoremainontheplatform.Additionally,

humanmoderatorsmayalsobeinvolvedin

trainingandimprovingthealgorithms,by

providingfeedbackandlabellingdatathatcan

beusedtore?nethesystem’saccuracyand

effectiveness.Individualstaskedwithcontent

moderationdutiesinsocialmediaplatforms

oftensufferfromanxiety,depression,andpost-

traumaticstressdisorder,adirectconsequence

oftheircontinuousexposuretodistressing

materialssuchasmurder,suicide,sexual

assault,orchildabusevideos.

Theseexamplesdemonstratehowhumansare

integraltotheprovisionofservicesmarketed

ordescribedas“arti?cialintelligence”.Indeed,

JeffBezosdescribedAmazon’sMechanical

Turk(AMT)platformas“arti?cial-arti?cial-

intelligence”asitwashumanintelligence

thatwasprovidingthelabour-intensivework

neededforarti?cialintelligencesystemsto

operate.AsdescribedontheAMTsite,the

platformprovides“anon-demand,scalable,

humanworkforcetocompletejobsthat

humanscandobetterthancomputers,for

example,recognizingobjectsinphotos”.5

Workersontheplatformareaccessiblethrough

anapplicationprogramminginterface(API),

allowingprogrammerstocallonworkerswith

afewsimplelinesofcodewhenworkingonan

algorithm[11].

InadditiontoplatformssuchasAMTand

Appen,datalabelerssometimesworkthrough

third-partycompanieshiredbyleading

tech?rms,inasubcontractingrelationship.

Althoughtherearestillmanydatalabelers

workingintheUnitedStatesinEurope,muchof

theworkisbeingdoneindevelopingcountries,

giventhelowremunerationassociatedwiththe

work.Whileprecise?guresonthenumbersof

personsworkingasdatalabelersdonotexist,

estimatesrangeinthetensofmillions,and

demandforsuchworkislikelytocontinueas

AIdatasetsandtrainingneedsgrow[12].The

sizeofthemarketisestimatedatbetweenUS

$1-$3billionandlikelytoexperiencedouble-

digitgrowthoverthenext5years[13].

Datalabelingworkdoesnotrequiremany

quali?cations,besidesliteracy,digitalskills

andaccesstocomputer(ormobiledevice)and

internet.Studiesofearningsofonlineplatform

workersintheUSthatperformthiswork,

regularlyreportmedianearningsofroughly$2

-$3perhour,orwellbelowthefederalminimum

wageofUS$7.25[14];[11].Giventhelowlevel

ofpay,itisunsurprisingthatmuchofthiswork

hasmovedtodevelopingcountries.

Butevenfromadevelopingcountry

perspective,theearningsarelow,particularly

consideringtheskillleveloftheworkforce,

withmanyworkersholdinguniversityand

post-graduatedegrees[11].Fortheworkers

whoworkthroughdigitallabourplatforms–

andnotbusinessprocessoutsourcing?rms

–thereistheaddedconcernthattheyare

hiredasindependentcontractorsandarethus

notcoveredbytheprotectionsandbene?ts

emanatingfromastandardemployment

relationship.Moreover,analysesofearnings

differentialsbetweenworkersinIndiadoing

similartypesofdataannotationworkrevealed

thatplatformworkersearnedtwo-thirds

lessthancomparable,non-platformworker

employees,evenbeforeaccountingforother

bene?tssuchassocialinsurancecontributions

[15].

5SeeAmazonMechanicalTurkAPIReference-AmazonMechanicalTurk.Accessedon9June2024.

12|MindtheAIDivide:ShapingaGlobalPerspectiveontheFutureofWork

Butevenamongbusinessprocessoutsourcing

?rms,thereareconcernsabouttheworking

conditionsoftheseworkers,withonecase

studyofadataannotationenterprisewith

of?cesinKenyarevealinglowpay,insecure

workandgender-basedworkplaceviolence

[16].Furthermore,thestudyarguedthatthe

dataannotationskillsusedinthislineofwork

werenotessentiallytransferable,questioning

thecareer-enhancingimpactofthislineof

work.

Movingalongthevaluechain,thesubsequent

parts–modeldesign,modeltrainingand

tuning,deploymentandmaintenance–

representacontrastingpicturewiththe

skillsneedsandworkingconditionsof

dataannotationwork.Theyalsoinvolve

muchgreaterrequirementsforphysical

infrastructure,particularlycomputepower

necessaryformodeltrainingandtuning.These

stagesrequiretheskillsofhighlyquali?ed

computerscientistsorgraduatesfrom

otherSTEM6?eldsinadditiontosigni?cant

investmentsinresearchanddevelopment.

ApartfromChinaandIndia,emergingmarkets

havegarneredonlyasmallportionofglobal

investmentinadvancedtechnologies.From

2008to2017,totalventurecapital?owsto

emergingmarkets,excludingChinaandIndia,

amountedtojust$24billion,whiletheUnited

Statesaloneattracted$694billionduringthe

sameperiod.7

Annually,morethan$300billionisspent

globallyontechnologytoenhancecomputing

capacity.However,theseinvestmentsare

unevenlyspread,makingthedisparityinaccess

tocomputinginfrastructurebothwithinand

amongvariousregionsincreasinglyevident.A

limitednumberofcountriesareleadingtheway

indevelopingcomputecapacity,whilemany

othersarebeginningfromalowbase.TheUS

holdsasigni?cantadvantageindata-centre

construction,farsurpassinginvestmentsmade

byanyothernation.AlthoughChina,Singapore,

theNetherlands,andafewothershave

developedsubstantialcapacity,mostcountries

havefewerthan20top-tierdatacentres.