人工智能對(duì)勞動(dòng)力市場(chǎng)的影響+The+Impact+of+AI+on+the+Labour+Market

NOVEMBER2024

SAMSHARPS

THOMASSMITHJAMESBROWNEOLIVERLARGE

RHEASUBRAMANYAPEICHINTAY

DANAEELLINA

ISABELATKINSONJESSLYTHGOW

RITHIKAMURALIDHARAN

TheImpactofAIontheLabourMarket

TONYBLAIR

INSTITUTEFOR

GLOBALCHANGE

Contents

10

SettingtheScene

16

ThePotentialImpactofAIonLabourMarkets

55

PolicySolutions

78

Conclusion

79

Acknowledgements

THEIMPACTOFAIONTHELABOURMARKET

3

ExecutiveSummary

Technologyhasalonghistoryofprofoundlyreshapingtheworldofwork.

Fromthedawnoftheagriculturalrevolutiontothedigitalage,eachwaveofinnovationhasredefinedthestructure,natureanddynamicsoflabour

markets.

Today,theworldisonthebrinkofanothertechnologicalrevolution,enabledbyartificialintelligence,thathasthepotentialtoreshapetheworldofwork

yetagain.Thiscomingwaveislikelytoarrivemorequicklythanits

predecessors,giventhepaceoftechnologicaladoptionhasbeen

acceleratingandbecauseAIitselfisadvancingsoquickly.Policymakersneedtopreparenowtoensuretheirworkersandlabourmarketsare

equippedtoridethecomingwaveandbenefitfromit.

InthisreportweusenovelanalysistoexplorehowAIcouldaffectthelabourmarket,withspecificreferencetotheUnitedKingdom.Althoughmuch

discussionofAIiscurrentlybasedaroundlargelanguagemodels(LLMs),thisispartofabroaderstoryaboutautomation,whichtakesinphysical

tasksaswellascognitiveones.Ourstudythereforeaimstotakeawider

viewthatincorporatesabroaderdefinitionofAI,includingAI-enabled

hardware,andthatexaminesthreekeywaysthatAIcouldaffectthelabour

market:throughdemand,supplyandbychangingtheworkplaceexperience:

Labourdemand:WeestimatethatfullandeffectiveadoptionofAIbyUKfirmscouldsavealmostaquarterofprivate-sectorworkforcetime–

equivalenttotheannualoutputof6millionworkers.

ThepotentialforAItosavetimeandboostproductivitywillvarysignificantlyattheindividual-workerlevel.Ouranalysissuggeststhatmostofthetime

savingsfromAIarelikelytocomefromtheuseofAI-enabledsoftwarethatperformscognitivetasks,ratherthanmorecostlyAI-enabledhardwarethatfocusesonphysicalwork.Asaresult,occupationsandsectorsthatinvolvecomplexmanualworksuchastheskilledtradesorconstructionarelikelytobelessexposedtoAI.Bycontrast,thoseworkerswhoperformroutine

4

cognitivetasks,particularlyinadministrativeoccupations,andthosewhoworkindata-intensiveindustrieswhereitiseasiertotrainnewAImodels(suchasbankingandfinance),arelikelytobemoreexposed.

Giventhehighdegreeofuncertaintyoverhowquickly,howfully,andhow

effectivelyAIcouldperformexistingtasksandsaveworkers’time,welookat

arangeofscenariostohelpassessitspotentialmacroeconomicimpact.Inallcases,AIisexpectedtogeneratesomejoblosses,butthislabour-

substitutioneffectisonlypartofthestoryofhowAIwillaffectlabour

demand.AIisalsolikelytocreatenewdemandforlabourbyboosting

economicgrowthandspeedingthedevelopmentofnewproductsand

servicesthatcreateentirelynewjobs.Overhistory,technology’simpactonlabourdemandhasbeenataleofthepushandpullbetweenthesetwo

forcesoflaboursubstitutionanddemandcreation,andoverthelongruntheyhavetendedtobalanceout.

Ouranalysistellsasimilarstory.Inallourscenariosweexpect

unemploymenttoriseinitiallyassomefirmschoosetobankthetime

savingsfromAIandreducethesizeoftheirworkforce.Basedonhistoric

ratesoflabourshedding,weestimate1to3millionjobscouldultimatelybedisplacedbyAI.Cruciallyhowever,thesejobdisplacementswillnotoccurallatonce,butinsteadwillrisegraduallywiththepaceofAIadoptionacross

thewidereconomy.Onanannualbasis,thismeansjobdisplacementspeakatbetween60,000and275,000jobsayear,whicharerelativelymodest

figurescomparedwiththeaveragenumberofjoblossesseenoverthepastdecadeintheUK(450,000peryear)andevenmoresorelativetotheoveralllabourforce(33million).

Moreover,theriseinunemploymentislikelytobecappedandultimately

offsetasAIcreatesnewdemandforworkers,whichpullsdisplacedworkers

backintotheworkforce.OurbestguessisthatAI’speakimpacton

unemploymentislikelytobeinthelowhundredsofthousandsandfortheeffecttounwindovertime.Whilethereisagreatdealofuncertaintyoverallthesefigures,acommonlessonisthatAIislikelytoincreasethedynamismofthelabourmarketbypromptingmoreworkerstoleaveexistingjobsandstartnewones.Asaresult,theUK’slabour-marketinfrastructurewillneedanupgradetoadjusttotheAIera.

5

ThesepotentialtimesavingsfromAIandassociatedriseinlabour

productivitywillalsoboosteconomicgrowth.Thescaleofthiseffectis

highlyuncertain,andwilldependbothontheextenttowhichAIisadoptedacrosstheeconomyandhowcost-effectiveitistoimplement.OuranalysissuggeststhatmostkindsofAIhavethepotentialtogeneratesignificant

costsavings,butthisisnotguaranteedforallAItypes.Forexample,

bespokeAIsoftwaretoolsthataretrainedonacompany’suniquedataaswellascomplexAI-enabledhardwarearelikelytoproveprohibitively

expensiveformostsmallfirmstodevelopontheirown.Forsmallfirmsto

accessthesenoveltechnologies,thetechnologywillneedtobedevelopedatscalebyothercompanies,“platformised”(developedintotoolsthatcan

beeasilyimplementedacrossanindustry)andthensoldasaservice(similartohowthemarketforcloudcomputinghasdeveloped).Weexplorethese

differentpossibilitiesinourscenarioanalysisandfindthatAIcouldraiseUKnationalincomebybetween5percentand14percentby2050withour

mostlikelyscenariopointingtoasubstantial11percentboost,equivalenttomorethan￡300billionayearintoday’sterms.

AkeyquestioninassessingAI’simpactontheeconomyisjudgingwhenitsimpactwillbecomesufficientlysignificanttoshowupinmacroeconomic

statistics.Predictingthepointatwhichanytechnologystartstobecome

widelyadoptedisinherentlydifficult,soweexplorearangeoftiming

assumptionsinouranalysis.InallcasesweexpecttheimpactofAIontheeconomytoberelativelymodestinthenear-term–raisingGDPbybetween0.1and1percentoverthenextfiveyearsandthelevelofunemploymentbyupto180,000by2030.By2035,therangeofoutcomesismuchbroader–withthelevelofgrossdomesticproduct(GDP)0.6to6percenthigher

dependingonthescenario.HowthisplaysoutinpracticewilldependnotonlyonthefundamentalnatureofAIordecisionsmadebyprivate-sectorfirmsbutalsopolicydecisionsthathavethecapacitytoaccelerateordelaytheimplementationofAI.

Laboursupply:AIhassignificantpotentialtoimprovethesupplyoflabour,byincreasingthequantity,qualityandeffectiveuseofworkersinthe

economy–allofwhichcanaddfurthertogrowth.

First,weestimatethatAIcouldassistbothteachersandstudentsinthe

6

classroomandraiseeducationalattainmentbyaround6percenton

averageacrossstudentsovertheiracademiccareer.Emergingevidencesuggeststhatlower-performingstudentsarelikelytoexperiencethe

biggestboostfromAI-enablededucation,suggestingitcouldbeahelpfulsocial-levellingtooltoequaliseaccesstoopportunities.Amoreeducatedworkforcewillbeamoreproductiveone,soasAI-educatedstudents

graduallyentertheworkforcetheywillraiseproductivity.Whilethisisaslow-

burneffect,itcouldbeasubstantialone–addingaround6percentcumulativelytoGDPoverthenext60years.

Second,AIcansupportahealthierpopulationandhenceahealthier

workforce,leadingtofewerlostworkdays,longerandmoreproductive

careers,andlowerwelfarecosts.ThecountryisinthefoothillsofAI

applicationsinhealthservicesbutalreadythereisenormouspotentialforAItospeedupmedicalresearch,enableapreventativeapproachtohealth

care,drivemoreefficientidentification,treatmentanddischargeofpeoplefromthehealthsystem,and,importantly,spurfurtherassistivetechnologiesthatcanhelpdisabledpeopleorthosewithshort-termhealthissuestore-entertheworkforce.

Third,thereispotentialforAItechnologiestosupportbetterjob-matchingandimprovetheeffectiveutilisationoflabour–justastheadventofthe

internetdidattheturnofthecentury.Thisisanareawheredesignand

controlsbecomeincreasinglyimportant,sincebiasarisingfromdata

patternscouldleadtonegativerecruitmentoutcomes.Onekeyelementtoretainisthatofworkers’agency,soratherthanseeingtheroleofAIas

matchingpassiveworkerstoroles,ithelpsequipthemwithstrategiesandinformationthatallowapplicantstopresentthemselvestobesteffect.

Workplaceexperience:People’sday-to-dayexperienceoftheworkplace

couldshiftmateriallyasAItoolsbecomemoreprevalent.AIhasthepotential

toimprovejobqualitybyreducingmundanetasks,improvingaccesstotheworkplacefordifferenttypesofworkers,andhelpingtoimproveworkplacehealthandsafety.AIthushasthepotentialtohelpdeliveramoreengaging,inclusiveandsafeworkingenvironment.However,oursurveyofearly

adoptersofAIinprofessionalsettingsalsohighlightedthatsomehave

concernsthattheworkplaceexperiencemightchangeinotherways,for

7

examplewithextrascrutinyofworkleadingtoamorestressfulworkingday.ThismeansthatcarefulmanagementandcommunicationofhowAIis

appliedintheworkplacewillbeacrucialelementofthetransition.

Overall,thereisemergingevidencethatAIishavingatransformativeimpactattheindividualandcompanylevels.Ourforward-lookinganalysissuggeststheseearlyindicatorsarelikelytobeaforetasteofamuchbigger

transformativechangetotheworldofworkinthedecadesahead.Itwouldbeamistaketoconcentrateallpolicydirectiononlimitingthedisruptionthatthiswillbring.Anypoliciesdesignedtoholdbackthetidewilllikelybe

ineffectiveanddamaginginthelongterm.

Instead,weproposefourkeyareaswhichgovernmentactivityshouldfocusontomaximisethebenefitsofAIintheworkforceforthebroadestpossiblepartofsociety:

A.EncouragethebroadadoptionofAIacrosstheeconomybyreducing

barrierstoAIaccessandbyharnessingitsabilitytoimproveeducationandskillsforall.Bydoingso,governmentscanhelpensureAIdeliversonits

promiseasasocial-levellingtoolthatequalisesaccesstoopportunities.

B.Upgradelabour-marketinfrastructuretocopewiththehigherrateof

churnandmoredynamicpaceofchangethatAIislikelytocreate.This

includesequippingworkerswithknowledgeofthecomingchangesandthesupportfacilities–includingfinancialsafetynets,retrainingopportunities

andjob-matchingservices–tohelpmaximiseemployment.

C.HarnesstheabilityofAItoimprovethequalityofjobs,byshiningalightonhowAIisalreadybeingadoptedacrosstheeconomy–tosharebest

practicetospeeduptakeandidentifyriskareaswherefirmerguardrailsareneeded.

D.Finally,giventhehighdegreeofuncertaintyaroundthepathofAIanditspotentialimpact,thegovernmentshouldengageindetailedscenario

analysisandpracticalcontingencyplanningtoensureitispreparedforthemoreradicalfuturethatisinprospect.

THEIMPACTOFAIONTHELABOURMARKET

8

Asummaryofdetailedrecommendationsunderthesefourthemesisprovidedinthetablebelow:

THEIMPACTOFAIONTHELABOURMARKET

9

Thefutureisuncertain.Someoftheanalysisinthisreportmayneed

revisitingbeforetoolong.Thiswillcontinuetobeadynamicenvironmentinwhichthetechnologydevelopsrapidlyanditsapplicationsevolveinwaysthatcannotbeforeseen.

Butonlybytakingsensiblestepsnowcanthecountrybestequipitselfforsuccess.ForthisreasonTBIhasaimedtodevelopasetofadozen“no-

regrets”policyrecommendationsthatshouldbringbenefitsinarangeof

scenarios,backedupbyahorizon-scanningfunctioninrecommendation13.

Takentogether,thisanalysisandsetofrecommendationsshouldprovideabigstepforwardintheUK’smissiontomakethemostofAItechnology.AI

canhelpeveryonebecomemoreproductiveandbetteroff.Indoingso,itcansupportmanyoftheotherpolicyoutcomesthegovernmentshouldpursue.

THEIMPACTOFAIONTHELABOURMARKET

10

01

SettingtheScene

Context

Artificialintelligenceish

eretoday.1

Itsimpactisalreadybeingfeltinthe

economy,inpubliclifeandinday-to-dayexperiences.Althoughit’sstillthe

veryearlychaptersofthestory,thegeneral-purposenatureofthetechnologyanditspotentialtobeadoptedquicklymakeAIhugelyconsequential.

Nowhereisthismoreapparentthanintheworldofwork.Thehistoryof

peopleworkingalongsidemachines,andthosemachinestakingonsomeorallofthetasksoriginallycarriedoutbyworkers,isalongone.Thereiseveryreasontobelievethiswillcontinue–andaccelerate.Machineswill

increasinglytakeontasksthatarecurrentlythepreserveofhumanswhile

alsospurringinnovationthatcreatesnewjobsthatdon’tyetexist.Thiswill

leademployerstorestructureentirelythewaysinwhichworkplacesoperate,andsothehumanimpactwillbeapparentandincreasinglyobvious.

Historyalsotellsusthattheroadtotechnologicaladoptionisbumpy.

Althoughtechnologicalprogressisthecornerstoneofrisinglivingstandards,itcanalsoexacerbateinequalitybydisplacingsomejobs,favouringthewell-resourced,andwideninggapsineducationandgeography.AIhasthe

potentialtocausesimilareffects,butitalsopresentsapotentialtoolfor

sociallevelling

.2

Thisneedstobewieldedmindfullyandshapedjudiciouslyinordertorealisea“techno-optimist”futurescenario.

ThequestionishowgovernmentsshouldrespondtotheemergenceofAI.Manygovernmentsmightdecidetowaitandsee,andreacttodisruption

aftertheevent.Otherswillseektoputsafetyfirstandintervenetoestablishasmuchcontrolaspossiblesothatchangecanbemanagedorminimised.

Eachoftheseapproachesismistaken.Thefirstleavespolicymakersatriskofseemingentirelyunprepared;thelatterisacharterfordecline.

Theaimforgovernmentsshouldbetoadoptapro-innovation,pro-

THEIMPACTOFAIONTHELABOURMARKET

11

technologystancewhilealsobeingawareoftherisksandpossibleshort-termconsequences.Thisisvitalnotjustforreasonsofequityandsocialcohesion,butalsotomaintainconfidenceandacceptanceinthe

technology,andtoguardagainstdemandsforeasy-soundinganswersthatwillsimplyundothegains.

ForallthesereasonswefeltitwasimportanttoapplyacriticallenstosomeoftheassumptionsaroundtheeffectofAIonjobs,toexplorearangeof

channelsandtogenerateasetofrecommendationsforaprogressive,ambitiousapproach.

DefinitionandScope

“Artificialintelligence”mustbeclosetothemostused–andmostflexiblyused–phraseincurrentpoliticaldiscourse.Yetdespitetheexplosionof

interestinAI,thereisnoconsensusonwhatthetermmeans.Forthe

purposesofthisproject,weusetheOrganisationforEconomicCo-

operationandDevelopment(OECD)definitionof

AI3

asourfoundation:“An

AIsystemisamachine-basedsystemthat,forexplicitorimplicitobjectives,infers,fromtheinputitreceives,howtogenerateoutputssuchas

predictions,content,recommendations,ordecisionsthatcaninfluencephysicalorvirtualenvironments.DifferentAIsystemsvaryintheirlevelsofautonomyandadaptivenessafterdeployment.”

ThisdefinitionfocusesontheoutputsthatanAIsystemgenerates–for

example:content,predictions,recommendations,decisions–andthe

objectivesthattheseoutputsaremeanttoachieve.Itiswidelyrecognised,andaversionofithasbeenadoptedbyboththeEuropean

Union4

andtheUnitedStat

es5

intheirAIacts.

AIisnotnew.ThefirstwaveofAIbeganinthemid-20thcenturyandwas

calledsymbolicAI,flourishingfromthe1950suntilthe1990s.Itreliedon

quantifiabledataandstrictrule-basedprocedureswithsomeallowancesfor

uncertainty.MajorsymbolicAIachievementsincludedoneofthefirst

chatbots,ELIZA,andthechess-playingexpertsystemDeepBlue,whichin1997beatGarryKasparov,theworldchampionatthetime.

12

Thesecond(andcurrent)waveofdata-drivenAIkickedoffinthe2000sandacceleratedinthe2010swithrapiduptakeoftheinternet,whichunlocked

oneofthekeybarriersstoppingAIfromscaling:data.Thiswaveclosely

mimickedhowhumanbrainsprocessinformationthroughdeeplearningandmachinelearning,andmodelleduncertainty,randomnessandcomplexity.ItisbestreflectedinrecentAIinnovationssuchasChatGPTorautonomous

drones.

Boththesewavesconstitute“weak”or“narrow”AIinthattheyarespecificandboundbytask-basedparameters.We’veexcludedsymbolicAIfromouranalysisundertheassumptionthatanylabour-marketimpactsfromthis

wavehavebeenwitnessedandadjustedforalready,giventhetechnology’sageandprevalence.

AIcapabilities,however,arefastevolving.Thenextwaveisexpectedtobe“strong”or“general”AI(AGI).AGIapplicationswillexhibitmoreautonomy,beabletoadapttodifferentcontextsandnotbelimitedtospecifictasks–similartohumancapability.Beyondthis,AIsystemssmarterthanany

individualhuman(“artificialsuperintelligence”)andsystemssmarterthanallhumans(“thesingularity”)arepossiblefurtherinthefuture.

Forthepurposesofthisproject,weexcludethesemoreadvancedversionsofAIbecausetheirdisruptivepotentialissogreatandtheyarestillthoughttobesomedecadesaway.Forexample,theGoodJudgementProject’s

“superforecasters”,whoareparticularlyskilledatassessingprobabilitiesoffutureoutcomes,estimatethereisonlya12percentchanceofAGIbeing

developedby2043,40percentby2070and60percentby2100

.6

AI

expertsareonlyslightlymoreoptimistic,withhalfofarecentsurveyofmorethan350AIprofessionalsestimatingthereisa50-50chanceAGIwillbe

developedby2060

.7

OurStartingPoint

TBIanalysedarangeofliteraturefrompublicandindustrysourcesand

complementedthiswithstakeholderengagementintheUnitedStatesandEurope

.8

13

Fromthisreview,wecanseethatexpertshavecometoabroadconsensusonseveralfronts:

?AIisalreadydrivingdouble-digitproductivitygainsforearlyadoptersonindividualtasks

.9

Todate,AIhasnotbeenimplementedwidelyenoughtomakeanoticeabledifferencetoaggregateproductivitystatistics,butthisisunsurprising–newinnovationsoftentaketimetobeintegratedinto

productionprocessesandfullyutilised.TheproductivityeffectsofAIarelikelytofollowastandardJ-curve

,10

whereinitialinvestmentleadsto

aggregateproductivitygainswithalag.

?WhatsinglesoutAIfromprevioustechnologicalwavesisthescaleandspeedofitspotentialimpact.Morethan50per

cent11

oftheUS

populationusedChatGPTwithin10monthsofitslaunch–theinternettook17yearstoreachthesamelevel.Around40percentofglobal

employmentisexpectedtobeaffectedinonewayoranotherby

generativeAIaccordingtotheInternationalMonetaryFund's(IMF)lateststudy

.12

AIshouldthereforebeviewedasoneofthefewtechnologieswithnear-universalapplication,similarinscopetoelectricityorthe

internet.Butunlikeelectricity,whichtookdecadestointegrateinto

productionprocesses,muchofthebasicinfrastructureforAI–

computersandtheinternet–alreadyexists,meaningAI’seffectsarelikelytomanifestwithinyears,notdecades.

?Itseemshighlylikelythenthatdemandforlabourwillsoonbedisruptedonasubstantialscale.Developedeconomies,white-collarworkersand

theservicessectoraremostlikelytofeeltheeffectsofthischangeintheshortterm,reflectingthealignmentoftheirworktogenerative-AI

technologies.However,thissamegroupisalsolikelytobebetterabletonavigateandleveragethedisruptiveimpactofAI,givenhigherskilllevelsandtechnologicalreadiness.

?SomeofthecommentaryaroundautomationviewsAIasarisktohumanwork.Itisreasonabletoconsiderthispointofview–atthisstageitis

possibletomakeprojectionsbasedonlyonanalysisanditisnotagiventhatAIwillresultinnetjobcreation.Whatwecansayisthatthepast200yearssuggestthatnewtechnologiestendtocreatemorejobsinthelongrunthantheydestroy.Itistruethatthiseffectisnotautomaticandhas

notbeenasvisiblesincethe1980s,aperiodcharacterisedaccelerating

14

au

tomation,13

butwithmorelimitedproductivitygainsandalowerrateofnewtaskcreation–aprocessknownas“so-soautomation”.Giventhe

largepotentialproductivitygainsfromAI,itseemsunlikelyitwillfollowthatpattern–andtheoutcomeoftheseprocessesaretractable,inthatpolicycanaffecttheoutcome.

?AIhasthepotentialtoincreaselong-termeconomicgrowthbyboostingthequalityofthelabourforcebothbyhelpingworkersattainmoreskillsandbylivinghealthierlives.AIcouldimproveskillsbyincreasing

educationalattainmentandprovidingtailoredon-the-jobtraining.Itcouldimprovehealthbyenablingearlyandpredictivediagnoses,expanding

health-systemcapacityandincreasinglabour-marketaccessforpeoplewithdifferentabilities.However,theseeffectsarenotguaranteedandwillrequireshrewdgovernmentalpolicytocultivatethepotentialgains.

?AIisalreadybeingusedtoimprovelabour-marketefficiencybybetter

matchingworkerswithemployers–similartohowonlinejobboards

improvedhiringrates

.14

However,thereisariskthatAIcouldalsoamplifyexistingbiasesinrecruitmentdecisions.Policyguardrailswillbeimportant

–includingaddinghumancheckpointsandevaluationsofAIrecruitmenttoensuretheyworkasintended.

?TheimpactofAIonindustrialrelationswillbecomplex.WorkersalreadyrecognisethepotentialbenefitsofAIintermsofimprovingjobquality–byautomatingroutinetasks,improvingphysicalsafetyatworkand

freeingupmoretimeformoreengagingtasks.However,workersalsoarewaryoftherisks–fearingjoblosses,lossofwage-bargainingpower,

increasedintensityandstress,andadditionalmanagementscrutiny.Thissuggestsasignificantroleforpolicyinmanagingthesedisparateeffects.

AIandtheUK

TheAIrevolutionisunlikelytoleaveanypartoftheworldunaffected.ButinthisreportwefocusontheUK.Thecountryhasalargeeconomy,iswell

integratedwithglobalmarkets,inpossessionofaskilledworkforce,withstrongcapitalmarketsandacompetitivenativetechscene–itshouldbewellpositionedtotakeadvantageoftechnologicalinnovations.

TheUKalsoneedsnewmeansofmakingprogress:thenationhasendured

15

alongperiodofslowproductivitygrowth,feedingintodisappointingoveralleconomicperformance.Successiveeffortstoaddresssomeofthese

shortcomings,basedaroundinfrastructureorskills,havemadeonlylimitedimpact.Somethinghasbeenmissing.Atthesametime,othercountriesare

poisedtosurgeaheadviatechnologicaltransformation,leavinganyunreformedeconomieslessandlesscompetitive.

Inotherwords,theupsideofAIfortheUKissubstantial,butthedownsideofamissedormismanagedadoptionofnewtechnologyisalsosignificant.

Theremainderofthereportisstructuredintwochapters.In“ThePotentialImpactofAIonLabourMarkets”weanalysethewaysinwhichAIcould

affecttheworldofwork–exploringitspotentialimpactonlabourdemand,laboursupplyandtheworkplaceexperience.Thenin“PolicySolutions”weprovideasetofrecommendationsfortheUKtobesttakeadvantageoftheopportunitiesofAIandmanageanyresultingdisruption.

THEIMPACTOFAIONTHELABOURMARKET

16

02

ThePotentialImpactofAIonLabourMarkets

TherearethreemajorwaysinwhichAImighthaveanimpactonthelabourmarket.

?Demandforlabour:MuchofthedebateaboutAIandthefutureofworkhascentredontheriskthatAIwillreplacejobsthroughautomation.

SomejobswillcertainlybereplacedbyAI,butitisonlyoneofthewaysthatAIwillaffectlabourdemand.AIcouldalsoincreasethedemandforlabourandcreatenewjobsthroughtwochannels.First,AIcould

complementworkersbymakingthemmoreproductive,whichwould

increaseeconomicgrowthandhencedemandforworkers.Second,AIcouldfollowthepatternofprevioustechnologicalwavesandcreatenewproducts,marketsandsectorsoftheeconomythatrequireworkerstoperformnewtasksandjobs–againincreasingthedemandforworkers.Therelativestrengthofthesethreeforces,ratherthanjustthefirst,willdeterminethenetimpactofAIonlabourdemand.

?Supplyoflabour:AIhasthepotentialtoincreasethequantity,qualityandeffectivedeploymentoflabour.Forexample,AI-enablededucationcouldraiseeducationalattainmentandproductivitygrowth,whileAI-assisted

healthcarecouldleadtoalargerandmoreeconomicallyactivelabour

force.AIalsohasthepotentialtoimprovethefunctioningofthelabour

marketbymoreefficientlymatchingemployerswithemployees,reducingfrictionalunemploymentandhelpingtoreducetheunder-utilisationof

labour.

?Workplaceexperience:AIhasthepotentialtoreshapetheenvironmentinwhichpeopleworkbyautomatingrepetitivetasks,increasingaccesstotheworkplaceincludingfordifferentlyabledgroups,andenablingbettermonitoringtosafeguardagainsthealth-and-safetyrisksandtohelp

boostproductivity.Thisisnotwithoutrisks,asanynewtechnologycanbeusedbothforgoodandill.ButcarefuladoptionofAIclearlyhasthepotentialtomakeworkmoreengaging,inclusiveandsafer,andtomakeworkersmoreproductive.

THEIMPACTOFAIONTHELABOURMARKET

17

Inthissection,weanalyseeachoftheseeffectsinturn–drawingonnoveltechniquestoanalysethefirsttwoforcesandanewAI-usersurveyto

explorethepotentialimpactofthelatter.

FIGURE1

ChannelsthroughwhichAImightimpactthelabourmarket

18

LabourDemand: