智東西·內(nèi)參·《白宮再推人工智能白皮書 技術(shù)過渡期應(yīng)如何應(yīng)對(duì)工種轉(zhuǎn)型》

PREPARINGFORTHEFUTUREOFARTIFICIALINTELLIGENCE

NationalScienceandTechnologyCouncil

ExecutiveOfficeofthePresident

ArtificialIntelligence,Automation,andtheEconomy

CopyrightInformation

ThisisaworkoftheU.S.Governmentandisinthepublicdomain.Itmaybefreelydistributed,copied,andtranslated;acknowledgmentofpublicationbytheExecutiveOfficeofthePresident(EOP)isappreciated.AnytranslationshouldincludeadisclaimerthattheaccuracyofthetranslationistheresponsibilityofthetranslatorandnotEOP.ItisrequestedthatacopyofanytranslationbesenttoEOP.ThisworkisavailableforworldwideuseandreuseandundertheCreativeCommonsCC01.0Universallicense.

EXECUTIVEOFFICEOFTHEPRESIDENTWASHINGTON,D.C.20502

December20,2016

AdvancesinArtificialIntelligence(AI)technologyandrelatedfieldshaveopenedupnewmarketsandnewopportunitiesforprogressincriticalareassuchashealth,education,energy,economicinclusion,socialwelfare,andtheenvironment.Inrecentyears,machineshavesurpassedhumansintheperformanceofcertaintasksrelatedtointelligence,suchasaspectsofimagerecognition.Expertsforecastthatrapidprogressinthefieldofspecializedartificialintelligencewillcontinue.Althoughitisunlikelythatmachineswillexhibitbroadly-applicableintelligencecomparabletoorexceedingthatofhumansinthenext20years,itistobeexpectedthatmachineswillcontinuetoreachandexceedhumanperformanceonmoreandmoretasks.

AI-drivenautomationwillcontinuetocreatewealthandexpandtheAmericaneconomyinthecomingyears,but,whilemanywillbenefit,thatgrowthwillnotbecostlessandwillbeaccompaniedbychangesintheskillsthatworkersneedtosucceedintheeconomy,andstructuralchangesintheeconomy.

AggressivepolicyactionwillbeneededtohelpAmericanswhoaredisadvantagedbythesechangesandtoensurethattheenormousbenefitsofAIandautomationaredevelopedbyandavailabletoall.

FollowingupontheAdministration’spreviousreport,PreparingfortheFutureofArtificialIntelligence,whichwaspublishedinOctober2016,thisreportfurtherinvestigatestheeffectsofAI-drivenautomationontheU.S.jobmarketandeconomy,andoutlinesrecommendedpolicyresponses.

ThisreportwasproducedbyateamfromtheExecutiveOfficeofthePresidentincludingstafffromtheCouncilofEconomicAdvisers,DomesticPolicyCouncil,NationalEconomicCouncil,OfficeofManagementandBudget,andOfficeofScienceandTechnologyPolicy.TheanalysisandrecommendationsincludedhereindrawoninsightslearnedoverthecourseoftheFutureofAIInitiative,whichwasannouncedinMayof2016,andincludedFederalGovernmentcoordinationeffortsandcross-sectorandpublicoutreachonAIandrelatedpolicymatters.

Beyondthisreport,moreworkremains,tofurtherexplorethepolicyimplicationsofAI.Mostnotably,AIcreatesimportantopportunitiesincyberdefense,andcanimprovesystemstodetectfraudulenttransactionsandmessages.

JasonFurman JohnP.Holdren

Chair,CouncilofEconomicAdvisers Director,OfficeofScienceandTechnologyPolicy

CeciliaMu?oz MeganSmith

Director,DomesticPolicyCouncil U.S.ChiefTechnologyOfficer

JeffreyZients

Director,NationalEconomicCouncil

ARTIFICIALINTELLIGENCE,AUTOMATION,ANDTHEECONOMY

Contents

TOC\o"1-4"\h\z\u

ExecutiveSummary 1

EconomicsofAI-DrivenAutomation 1

PolicyResponses 3

Conclusion 4

OutreachandDevelopmentofthisReport 5

Introduction 6

EconomicsofAI-DrivenAutomation 8

AIandtheMacroeconomy:TechnologyandProductivityGrowth 8

AIandtheLaborMarket:DiversePotentialEffects 10

HistoricalEffectsofTechnicalChange 11

AIandtheLaborMarket:TheNearTerm 13

WhatkindofjobswillAIcreate? 18

TechnologyisNotDestiny—InstitutionsandPoliciesAreCritical 21

PolicyResponses 26

Strategy#1:InvestInandDevelopAIforitsManyBenefits 27

Strategy#2:EducateandTrainAmericansforJobsoftheFuture 30

Strategy#3:AidWorkersintheTransitionandEmpowerWorkerstoEnsureBroadlySharedGrowth 34

Conclusion 43

References 44

ARTIFICIALINTELLIGENCE,AUTOMATION,ANDTHEECONOMY

PAGE

10

ExecutiveSummary

Acceleratingartificialintelligence(AI)capabilitieswillenableautomationofsometasksthathavelongrequiredhumanlabor.1Thesetransformationswillopenupnewopportunitiesforindividuals,theeconomy,andsociety,buttheyhavethepotentialtodisruptthecurrentlivelihoodsofmillionsofAmericans.WhetherAIleadstounemploymentandincreasesininequalityoverthelong-rundependsnotonlyonthetechnologyitselfbutalsoontheinstitutionsandpoliciesthatareinplace.ThisreportexaminestheexpectedimpactofAI-drivenautomationontheeconomy,anddescribesbroadstrategiesthatcouldincreasethebenefitsofAIandmitigateitscosts.

EconomicsofAI-DrivenAutomation

TechnologicalprogressisthemaindriverofgrowthofGDPpercapita,allowingoutputtoincreasefasterthanlaborandcapital.Oneofthemainwaysthattechnologyincreasesproductivityisbydecreasingthenumberoflaborhoursneededtocreateaunitofoutput.Laborproductivityincreasesgenerallytranslateintoincreasesinaveragewages,givingworkerstheopportunitytocutbackonworkhoursandtoaffordmoregoodsandservices.Livingstandardsandleisurehourscouldbothincrease,althoughtothedegreethatinequalityincreases—asithasinrecentdecades—itoffsetssomeofthosegains.

AIshouldbewelcomedforitspotentialeconomicbenefits.Thoseeconomicbenefits,however,willnotnecessarilybeevenlydistributedacrosssociety.Forexample,the19thcenturywascharacterizedbytechnologicalchangethatraisedtheproductivityoflower-skilledworkersrelativetothatofhigher-skilledworkers.Highly-skilledartisanswhocontrolledandexecutedfullproductionprocessessawtheirlivelihoodsthreatenedbytheriseofmassproductiontechnologies.Ultimately,manyskilledcraftswerereplacedbythecombinationofmachinesandlower-skilledlabor.Outputperhourrosewhileinequalitydeclined,drivingupaveragelivingstandards,butthelaborofsomehigh-skillworkerswasnolongerasvaluableinthemarket.

Incontrast,technologicalchangetendedtoworkinadifferentdirectionthroughoutthelate20thcentury.TheadventofcomputersandtheInternetraisedtherelativeproductivityofhigher-skilledworkers.Routine-intensiveoccupationsthatfocusedonpredictable,easily-programmabletasks—suchasswitchboardoperators,filingclerks,travelagents,andassemblylineworkers—wereparticularlyvulnerabletoreplacementbynewtechnologies.Someoccupationswerevirtuallyeliminatedanddemandforothersreduced.Researchsuggeststhattechnologicalinnovationoverthisperiodincreasedtheproductivityofthoseengagedinabstractthinking,creativetasks,andproblem-solvingandwasthereforeatleastpartiallyresponsibleforthesubstantialgrowthinjobsemployingsuchtraits.Shiftingdemandtowardsmoreskilledlaborraisedtherelativepayofthisgroup,contributingtorisinginequality.Atthesametime,a

1Amoreextensiveintroductorydiscussionofartificialintelligence,machinelearning,andrelatedpolicytopicscanbefoundintheAdministration’sfirstreportonthissubject.SeeTheWhiteHouse,“PreparingfortheFutureofArtificialIntelligence,”O(jiān)ctober2016(https://

/sites/default/files/whitehouse_files/microsites/ostp/NSTC/preparing_for_the_future_o

f_ai.pdf).

slowdownintherateofimprovementineducation,andinstitutionalchangessuchasthereductioninunionizationanddeclineintheminimumwage,alsocontributedtoinequality—underscoringthattechnologicalchangesdonotuniquelydetermineoutcomes.

Today,itmaybechallengingtopredictexactlywhichjobswillbemostimmediatelyaffectedbyAI-drivenautomation.BecauseAIisnotasingletechnology,butratheracollectionoftechnologiesthatareappliedtospecifictasks,theeffectsofAIwillbefeltunevenlythroughtheeconomy.Sometaskswillbemoreeasilyautomatedthanothers,andsomejobswillbeaffectedmorethanothers—bothnegativelyandpositively.Somejobsmaybeautomatedaway,whileforothers,AI-drivenautomationwillmakemanyworkersmoreproductiveandincreasedemandforcertainskills.Finally,newjobsarelikelytobedirectlycreatedinareassuchasthedevelopmentandsupervisionofAIaswellasindirectlycreatedinarangeofareasthroughouttheeconomyashigherincomesleadtoexpandeddemand.

RecentresearchsuggeststhattheeffectsofAIonthelabormarketintheneartermwillcontinuethetrendthatcomputerizationandcommunicationinnovationshavedriveninrecentdecades.

Researchers’estimatesonthescaleofthreatenedjobsoverthenextdecadeortworangefrom9to47percent.Forcontext,every3monthsabout6percentofjobsintheeconomyaredestroyedbyshrinkingorclosingbusinesses,whileaslightlylargerpercentageofjobsareadded—resultinginrisingemploymentandaroughlyconstantunemploymentrate.Theeconomyhasrepeatedlyprovenitselfcapableofhandlingthisscaleofchange,althoughitwoulddependonhowrapidlythechangeshappenandhowconcentratedthelossesareinspecificoccupationsthatarehardtoshiftfrom.

Researchconsistentlyfindsthatthejobsthatarethreatenedbyautomationarehighlyconcentratedamonglower-paid,lower-skilled,andless-educatedworkers.Thismeansthatautomationwillcontinuetoputdownwardpressureondemandforthisgroup,puttingdownwardpressureonwagesandupwardpressureoninequality.Inthelonger-run,theremaybedifferentorlargereffects.Onepossibilityissuperstar-biasedtechnologicalchange,wherethebenefitsoftechnologyaccruetoanevensmallerportionofsocietythanjusthighly-skilledworkers.Thewinner-take-mostnatureofinformationtechnologymarketsmeansthatonlyafewmaycometodominatemarkets.Iflaborproductivityincreasesdonottranslateintowageincreases,thenthelargeeconomicgainsbroughtaboutbyAIcouldaccruetoaselectfew.Insteadofbroadlysharedprosperityforworkersandconsumers,thismightpushtowardsreducedcompetitionandincreasedwealthinequality.

Historicallyandacrosscountries,however,therehasbeenastrongrelationshipbetweenproductivityandwages—andwithmoreAIthemostplausibleoutcomewillbeacombinationofhigherwagesandmoreopportunitiesforleisureforawiderangeofworkers.ButthedegreethatthismaterializesdependsnotjustonthenatureoftechnologicalchangebutimportantlyonthepolicyandinstitutionalchoicesthataremadeabouthowtoprepareworkersforAIandtohandleitsimpactsonthelabormarket.

PolicyResponses

Technologyisnotdestiny;economicincentivesandpublicpolicycanplayasignificantroleinshapingthedirectionandeffectsoftechnologicalchange.Givenappropriateattentionandtherightpolicyandinstitutionalresponses,advancedautomationcanbecompatiblewithproductivity,highlevelsofemployment,andmorebroadlysharedprosperity.Inthepast,the

U.S.economyhasadaptedtonewproductionpatternsandmaintainedhighlevelsofemploymentalongsiderisingproductivityasmoreproductiveworkershavehadmoreincentivetoworkandmorehighlypaidworkershavespentmore,supportingthiswork.But,someshockshaveleftagrowingshareofworkersoutofthelaborforce.Thisreportadvocatesstrategiestoeducateandpreparenewworkerstoentertheworkforce,cushionworkerswholosejobs,keepthemattachedtothelaborforce,andcombatinequality.MostofthesestrategieswouldbeimportantregardlessofAI-drivenautomation,butalltakeonevengreaterimportancetothedegreethatAIismakingmajorchangestotheeconomy.

Strategy#1:InvestinanddevelopAIforitsmanybenefits.Ifcareistakentoresponsiblymaximizeitsdevelopment,AIwillmakeimportant,positivecontributionstoaggregateproductivitygrowth,andadvancesinAItechnologyholdincrediblepotentialtohelptheUnitedStatesstayonthecuttingedgeofinnovation.GovernmenthasanimportantroletoplayinadvancingtheAIfieldbyinvestinginresearchanddevelopment.AmongtheareasforadvancementinAIarecyberdefenseandthedetectionoffraudulenttransactionsandmessages.Inaddition,therapidgrowthofAIhasalsodramaticallyincreasedtheneedforpeoplewithrelevantskillsfromallbackgroundstosupportandadvancethefield.PrioritizingdiversityandinclusioninSTEMfieldsandintheAIcommunityspecifically,inadditiontootherpossiblepolicyresponses,isakeypartinaddressingpotentialbarriersstemmingfromalgorithmicbias.Competitionfromnewandexistingfirms,andthedevelopmentofsoundpro-competitionpolicies,willincreasinglyplayanimportantroleinthecreationandadoptionofnewtechnologiesandinnovationsrelatedtoAI.

Strategy#2:EducateandtrainAmericansforjobsofthefuture.AsAIchangesthenatureofworkandtheskillsdemandedbythelabormarket,Americanworkerswillneedtobepreparedwiththeeducationandtrainingthatcanhelpthemcontinuetosucceed.Deliveringthiseducationandtrainingwillrequiresignificantinvestments.Thisstartswithprovidingallchildrenwithaccesstohigh-qualityearlyeducationsothatallfamiliescanpreparetheirstudentsforcontinuededucation,aswellasinvestingingraduatingallstudentsfromhighschoolcollege-andcareer-ready,andensuringthatallAmericanshaveaccesstoaffordablepost-secondaryeducation.

AssistingU.S.workersinsuccessfullynavigatingjobtransitionswillalsobecomeincreasinglyimportant;thisincludesexpandingtheavailabilityofjob-driventrainingandopportunitiesforlifelonglearning,aswellasprovidingworkerswithimprovedguidancetonavigatejobtransitions.

Strategy#3:Aidworkersinthetransitionandempowerworkerstoensurebroadlysharedgrowth.Policymakersshouldensurethatworkersandjobseekersarebothabletopursuethejobopportunitiesforwhichtheyarebestqualifiedandbestpositionedtoensuretheyreceiveanappropriatereturnfortheirworkintheformofrisingwages.Thisincludesstepstomodernizethesocialsafetynet,includingexploringstrengtheningcriticalsupportssuchasunemployment

insurance,Medicaid,SupplementalNutritionAssistanceProgram(SNAP),andTemporaryAssistanceforNeedyFamilies(TANF),andputtinginplacenewprogramssuchaswageinsuranceandemergencyaidforfamiliesincrisis.Workerempowermentalsoincludesbolsteringcriticalsafeguardsforworkersandfamiliesinneed,buildinga21stcenturyretirementsystem,andexpandinghealthcareaccess.Increasingwages,competition,andworkerbargainingpower,aswellasmodernizingtaxpolicyandpursuingstrategiestoaddressdifferentialgeographicimpact,willbeimportantaspectsofsupportingworkersandaddressingconcernsrelatedtodisplacementamidshiftsinthelabormarket.

Finally,ifasignificantproportionofAmericansareaffectedintheshort-andmedium-termbyAI-drivenjobdisplacements,policymakerswillneedtoconsidermorerobustinterventions,suchasfurtherstrengtheningtheunemploymentinsurancesystemandcountervailingjobcreationstrategies,tosmooththetransition.

Conclusion

RespondingtotheeconomiceffectsofAI-drivenautomationwillbeasignificantpolicychallengeforthenextAdministrationanditssuccessors.AIhasalreadybeguntotransformtheAmericanworkplace,changethetypesofjobsavailable,andreshapetheskillsthatworkersneedinordertothrive.AllAmericansshouldhavetheopportunitytoparticipateinaddressingthesechallenges,whetherasstudents,workers,managers,technicalleaders,orsimplyascitizenswithavoiceinthepolicydebate.

AIraisesmanynewpolicyquestions,whichshouldbecontinuedtopicsfordiscussionandconsiderationbyfutureAdministrations,Congress,theprivatesector,academia,andthepublic.Continuedengagementamonggovernment,industry,technicalandpolicyexperts,andthepublicshouldplayanimportantroleinmovingtheNationtowardpoliciesthatcreatebroadlysharedprosperity,unlockthecreativepotentialofAmericancompaniesandworkers,andensureAmerica’scontinuedleadershipinthecreationanduseofAI.

OutreachandDevelopmentofthisReport

ThisreportwasdevelopedbyateamintheExecutiveOfficeofthePresidentincludingstafffromtheWhiteHouseCouncilofEconomicAdvisers(CEA),DomesticPolicyCouncil(DPC),NationalEconomicCouncil(NEC),OfficeofManagementandBudget(OMB),andOfficeofScienceandTechnologyPolicy(OSTP).ThisreportfollowsapreviousreportpublishedinOctober2016titledPreparingfortheFutureofArtificialIntelligenceandtheaccompanyingNationalArtificialIntelligenceResearchandDevelopmentStrategicPlan,developedbytheNationalScienceandTechnologyCouncil’s(NSTC)SubcommitteeonMachineLearningandArtificialIntelligence.ThissubcommitteewascharteredinMay2016byOSTPtofosterinteragencycoordinationandprovidetechnicalandpolicyadviceontopicsrelatedtoAI,andtomonitorthedevelopmentofAItechnologiesacrossindustry,theresearchcommunity,andtheFederalGovernment.Thisreportalsofollowsaseriesofpublic-outreachactivitiesasapartoftheWhiteHouseFutureofArtificialIntelligenceInitiative,designedtoallowgovernmentofficialstolearnfromexpertsandfromthepublic,whichincludedfiveco-hostedpublicworkshops,andapublicRequestforInformation(RFI).2

ThisreportmoredeeplyexaminestheimpactofAI-drivenautomationontheeconomyandpolicyresponsestoit.Itconsiderstheeconomicevidencetobetterunderstandthelessonsfrompastwavesofautomation,theimpactalreadycausedbythecurrentwaveofAI-drivenautomationanditsprospectsforthenearfuture,andhowAI-drivenautomationmayaffectworkersinthefuture.Thereportalsoconsiderspolicystepsthatareneededtoaddresstheeconomicdislocationcausedbythearrivalofthesetechnologiesandtoprepareforlonger-termtrendsintheeconomycausedbyAI,automation,andotherfactorsthataresystemicallydisadvantagingcertainworkers.Thereportlaysoutthreebroadstrategiesforpolicymakerstoconsider.

2EdFeltenandTerahLyons,“PublicInputandNextStepsontheFutureofArtificialIntelligence,”Medium,September62016

(

/@USCTO/public-input-and-next-steps-on-the-future-of-

artificialintelligence-458b82059fc3).FurtherdetailsonthepublicworkshopsandtheRFIcanbefoundintheOctober2016report,PreparingfortheFutureofArtificialIntelligence.

Introduction

RecentprogressinArtificialIntelligence(AI)hasbroughtrenewedattentiontoquestionsaboutautomationdrivenbytheseadvancesandtheirimpactontheeconomy.ThecurrentwaveofprogressandenthusiasmforAIbeganaround2010,drivenbythreemutuallyreinforcingfactors:theavailabilityofbigdatafromsourcesincludinge-commerce,businesses,socialmedia,science,andgovernment;3whichprovidedrawmaterialfordramaticallyimprovedmachinelearningapproachesandalgorithms;whichinturnreliedonthecapabilitiesofmorepowerfulcomputers.4Duringthisperiod,thepaceofimprovementsurprisedAIexperts.Forexample,onapopularimagerecognitionchallengethathasa5percenthumanerrorrateaccordingtooneerrormeasure,5thebestAIresultimprovedfroma26percenterrorratein2011to3.5percentin2015.Thisprogressmayenablearangeofworkplacetasksthatrequireimageunderstandingtobeautomated,andwillalsoenablenewtypesofworkandjobs.ProgressonotherAIchallengeswilldrivesimilareconomicchanges.

TechnicalinnovationhasbeenexpandingtheAmericaneconomysincethecountry’sfounding.AmericaningenuityhasalwaysbeenoneoftheNation’sgreatestresources,akeydriverofeconomicgrowth,andasourceofstrategicadvantagefortheUnitedStates.Remarkablehomegrowninnovationshaveimprovedqualityoflife,createdjobs,broadenedunderstandingoftheworld,andhelpedAmericansapproachtheirfullpotential.Atthesametime,theyhaveforcedAmericanstoadapttochangesintheworkplaceandthejobmarket.Thesetransformationshavenotalwaysbeencomfortable,butinthelongrun—andsupportedbygoodpublicpolicy—theyhaveprovidedgreatbenefits.

ThecurrentwaveofAI-drivenautomationmaynotbesodifferent.Forexample,robotshavemadetheeconomymoreefficient.A2015studyofrobotsin17countriesfoundthattheyaddedanestimated0.4percentagepointonaveragetothosecountries’annualGDPgrowthbetween1993and2007,accountingforjustoverone-tenthofthosecountries’overallGDPgrowthduringthattime.6SomeofthatgrowthhasbeenachievedbyU.S.manufacturersadoptingrobots,allowingmoregoodstobeproducedwhileemployingfewerworkersatsomefacilities.AIinitsmanymanifestationsalsoholdspromisetotransformthebasisofeconomicgrowthforcountriesacrosstheworld;arecentanalysisof12developedeconomies(includingtheUnitedStates)

3“BigData:SeizingOpportunities,PreservingValues,”ExecutiveOfficeofthePresident,May2014,https://

/sites/default/files/docs/big_data_privacy_report_may_1_2014.pdf.

4

For

moreinformationaboutAIanditspolicyimplications,see:TheWhiteHouse,“PreparingfortheFutureofArtificialIntelligence,”O(jiān)ctober2016.(https://

/sites/default/files/whitehouse_files/microsites/ostp/NSTC/preparing_for_the_fut

ure_of_ai.pdf).

5TheImageNetLargeScaleVisualRecognitionChallengeprovidesasetofphotographicimagesandasksforanaccuratedescriptionofwhatisdepictedineachimage.Statisticsinthetextrefertothe“classificationerror”metricinthe“classification+localizationwithprovidedtrainingdata”task.See

/challenges/LSVRC/.

6GeorgGraetzandGuyMichaels,“RobotsatWork,”CEPRDiscussionPaperNo.DP10477,March2015(

/sol3/papers.cfm?abstract_id=2575781).

foundthatAIhasthepotentialtodoubleannualeconomicgrowthratesinthecountriesanalyzedby2035.7

SomeexpertshavecharacterizedtheriseofAI-drivenautomationasoneofthemostimportanteconomicandsocialdevelopmentsinhistory.TheWorldEconomicForumhascharacterizeditasthelynchpinofaFourthIndustrialRevolution.8Furthermore,theeconomistAndrewMcAfeewrote,“DigitaltechnologiesaredoingforhumanbrainpowerwhatthesteamengineandrelatedtechnologiesdidforhumanmusclepowerduringtheIndustrialRevolution.They’reallowingustoovercomemanylimitationsrapidlyandtoopenupnewfrontierswithunprecedentedspeed.It’saverybigdeal.Buthowexactlyitwillplayoutisuncertain.”9

Atthesametime,AI-drivenautomationhasyettohaveaquantitativelymajorimpactonproductivitygrowth.Infact,measuredproductivitygrowthoverthelastdecadehasslowedinalmosteveryadvancedeconomy.Itisplausible,however,thatthepaceofmeasuredproductivitygrowthwillpickupinthecomingyears.TothedegreethatAI-drivenautomationrealizesitspotentialtodrivetremendouspositiveadvancementindiversefields,itwillmakeAmericansbetteroffonaverage.But,thereisnoguaranteethateveryonewillbenefit.AI-drivenchangesinthejobmarketintheUnitedStateswillcausesomeworkerstolosetheirjobs,evenwhilecreatingnewjobselsewhere.Theeconomicpainthiscauseswillfallmoreheavilyuponsomethanonothers.Policymakersmustconsiderwhatcanbedonetohelpthosefamiliesandcommunitiesgetbackontheirfeetandassemblethetoolstheyneedtothriveinthetransformedeconomyandshareinitsbenefits.

7PaulDaughertyandMarkPurdy,“WhyAIistheFutureofGrowth,”2016(https://

/t20161031T154852

w/us-en/_acnmedia/PDF-33/Accenture-Why-AI-is-the-Future-of-Growth.PDF#zoom=50).

8KlausSchwab,“TheFourthIndustrialRevolution:whatitmeans,howtorespond,”WorldEconomicForum,January2016(https://

/agenda/2016/01/the-fourth-industrial-revolution-what-it-means-and-how-

to-respond/).(Thefirstthreeindustrialrevolutionsarelistedasthosedrivenbysteampower,electricity,andelectronics.)

9AmyBernsteinandAnandRaman,“TheGreatDecoupling:AnInterviewwithErikBrynjolfssonandAndrewMcAfee,”HarvardBusinessReview,June2015(/2015/06/the-great-decoupling).

EconomicsofAI-DrivenAutomation

AcceleratingAIcapabilitieswillenableautomationofsometasksthathavelongrequiredhumanlabor.Ratherthanrelyingonclosely-tailoredrulesexplicitlycraftedbyprogrammers,modernAIprogramscanlearnfrompatternsinwhateverdatatheyencounteranddeveloptheirownrulesforhowtointerpretnewinformation.ThismeansthatAIcansolveproblemsandlearnwithverylittlehumaninput.Inaddition,advancesinroboticsareexpandingmachines’abilitiestointeractwithandshapethephysicalworld.Combined,AIandroboticswillgiverisetosmartermachinesthatcanperformmoresophisticatedfunctionsthaneverbeforeanderodesomeoftheadvantagesthathumanshaveexercised.Thiswillpermitautomationofmanytasksnowperformedbyhumanworkersandcouldchangetheshapeofthelabormarketandhumanactivity.

Thesetransformationsmayopenupnewopportunitiesforindividuals,theeconomy,andsociety,buttheymayalsoforecloseopportunitiesthatarecurrentlyessentialtothelivelihoodsofmanyAmericans.ThischapterexplorestheimportantrolethatAI-drivenautomationislikelytohaveingrowingtheeconomyandpotentialeffectsonlabormarketsandcommunities.Itdrawsoneconomictheoryandempiricalstudiesofpasttechnologicaltransformationsandappliestheselessonstothecurrentcontext.WhiletherearemanyreasonstothinkthatchangesinthelabormarketpromptedbyAI-drivenautomationwillbesimilartowhathasbeenobservedinthepast,thischapterwillalsodiscussargumentsforhowthecurrentperiodcouldbedifferentfromprevioustechnologicalrevolutions.

Critically,technologyalonewillnotdeterminetheeconomicoutcomesintermsofgrowth,inequalityoremployment.Theadvancedeconomiesallhavehadaccesstosimilarlevelsoftechnologybuthavehadverydifferentoutcomesalongallofthesedimensionsbecausetheyhavehaddifferentinstitutionsandpolicies.Butunderstan