Artificialintelligence: opportunitiesandimplicationsforthefutureofdecisionmaking

GovernmentOfficeforScience

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

Foreword

Weareincurrentlyinthefoothillsofanewtechnologicalrevolution.Artificialintelligencehasthepotentialtobeastransformativeinourlifetimesasthesteam-poweredeconomyofthe19th

century.

Alreadyit’slettingustotalktooursmartphones,recommendingusmusic,describingphotosforthevisuallyimpairedandflaggingupfirerisksincities.

Inthenearfuturewecouldseeitdeployedineverythingfromdriverlesscars,tointelligentenergygrids,totheeradicationofinfectiousdiseases.

Ingovernmenttoowearelookingatthepotentialapplicationsofthistechnologyinthedeliveryofpublicservices.

OurGovernmentDataProgrammeisincreasingthenumberofprojectsanddatascientistsingovernment,whileplayingaleadingroleinestablishingtheappropriateuseofthesepowerfulnewtools.

Asonetheworld’sleadingdigitalnations,artificialintelligencepresentsahugeopportunityfortheUK.

Getthisright,andwecancreateamoreprosperouseconomywithbetterandmorefulfillingjobs.Wecanprotectourenvironmentbyusingresourcesmoreefficiently.Andwecanmakegovernmentsmarter,usingthepowerofdatatoimproveourpublicservices.

Aswe'veseenalreadyinmanyareas,muchroutinecognitivework-thefiling,siftingandsorting-canincreasinglybeautomated,freeingpeopleuptofocusonthemorehumanaspectsofanyjob.

ThePrimeMinisterhasannouncedanindependentreviewofmodernemploymentpractices,sothatthesupportweprovidebusinessesandworkerskeepspacewithchangesinthelabour

marketandtheeconomy.

Artificialintelligencealsoposesnewquestionsaboutethicsandgovernance,theresponsibleuseofdataandstrongcyberdefences.Torealisethefullpotentialofthisrevolution,againwehavetobereadywithanswers.

IampleasedthattheRoyalSocietyandtheBritishAcademyareconductingareviewthatwillconsiderhowbesttheUKmightmanagetheuseofartificialintelligence.

Thisnotesetsoutwherethescienceisheading,describessomeoftheimplicationsforsocietyandgovernment,andshowshowwecanresponsiblyusethistechnologytoimprovethelivesandlivingstandardsofeveryoneinBritain.

ItisatimelyandimportantpieceofworkfromtheGovernmentChiefScientificAdviser.

MattHancock

MinisterforDigitalandCulture

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

Contents

Foreword 2

Introduction 4

Whatisartificialintelligence? 4

Artificialintelligenceforinnovationandproductivity 8

Theuseofartificialintelligencebygovernment 10

Effectsonlabourmarkets 12

Newchallenges 14

Publicdialogue 17

Conclusion 18

AnnexA:Background 19

AnnexB:Sources 20

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

4

Introduction

Artificialintelligencehasarrived.Intheonlineworlditisalreadyapartofeverydaylife,sittinginvisiblybehindawiderangeofsearchenginesandonlinecommercesites.Itoffershuge

potentialtoenablemoreefficientandeffectivebusinessandgovernmentbuttheuseofartificialintelligencebringswithitimportantquestionsaboutgovernance,accountabilityandethics.

Realisingthefullpotentialofartificialintelligenceandavoidingpossibleadverseconsequencesrequiressocietiestofindsatisfactoryanswerstothesequestions.Thisreportsetsoutsome

possibleapproaches,anddescribessomeofthewaysgovernmentisalreadyengagingwiththeseissues.

Artificialintelligenceisnotadistincttechnology.Itdependsforitspoweronanumberof

prerequisites:computingpower,bandwidth,andlarge-scaledatasets,allofwhichareelementsof‘bigdata’,thepotentialofwhichwillonlyberealisedusingartificialintelligence.Ifdataisthefuel,artificialintelligenceistheengineofthedigitalrevolution.

Muchhasalreadybeenwrittenabouttheuseofartificialintelligenceandbigdata.Thispaper

doesnotattempttosurveythewholefield.ItsoriginslieinaseminarheldattheBritish

AcademyinFebruary2016,chairedbyMarkWalport,GovernmentChiefScientificAdviserandMarkSedwill,PermanentSecretaryattheHomeOffice,thatdiscussedsomeofthelegaland

ethicalissuesaroundtheuseofartificialintelligence.Theissuesdiscussedthereprovidethe

coreofthisreport,withadditionalmaterialdrawnfromtheviewsofa

widerangeofscientific

andlegalexpertsinthefield,

althoughwehavesoughttominimisedetaileddiscussionof

technicalaspectsinordertoconcentrateonthepracticalaspectsofthedebate.Wehopethatitservesasanintroductiontothetopic.

Thereportconsidersthefollowingquestions:

.Whatisartificialintelligenceandhowisitbeingemployed?

.Whatbenefitsisitlikelytobringforproductivity?

.Howdowebestmanageanyethicalandlegalrisksarisingfromitsuse?

SirMarkWalport

GovernmentChiefScientificAdviser

MarkSedwill

PermanentSecretary,HomeOffice

5

Artificialintelligenceisparticularlyusefulforsortingdata,findingpatternsand

makingpredictions.Currentexamplesineverydaylifearewidespread:theyincludetranslationandspeechrecognition

servicesthatlearnfromlanguageonline,searchenginesthatrankwebsitesontheirrelevancetotheuser,andfiltersforemailspamthatrecognisejunkmailbasedonpreviousexamples(seeboxformore

uses).Thislistofapplicationsisgrowingrapidly:artificialintelligenceisenablinganewwaveofinnovationacrossevery

sectoroftheUKeconomy.

Artificialintelligenceisabroadterm(seebox).Moregenerallyitreferstothe

analysisofdatatomodelsomeaspectoftheworld.Inferencesfromthesemodelsarethenusedtopredictandanticipatepossiblefutureevents.

Whatisartificialintelligence?

Artificialintelligenceismorethanthesimpleautomationofexistingprocesses:itinvolves,togreaterorlesserdegrees,settinganoutcomeandlettingacomputerprogramfinditsownwaythere.Itisthiscreativecapacitythatgivesartificialintelligenceitspower.Butitalsochallengessomeofourassumptionsabouttheroleofcomputersandourrelationshiptothem.

Someusesofartificialintelligence

ProductrecommendationsfromservicessuchasNetflixandAmazonthatevolvethroughusers’webexperiencesarepoweredbymachinelearning.

TheUK’s‘smartmotorways’usefeedbackonroad

conditionsfromembeddedsensorsandneuralnetworksystemstoanticipateandmanagetrafficflow.

Infinancialmarkets,‘high-frequencytrading’algorithms

usepre-determineddecisioncriteriatorespondtomarketconditionsmanytimesfasterthanhumantradersareableto.Similaralgorithmsarebeingusedbysomefinancial

advisorstoautomaticallyspotinvestmentopportunitiesforclients.

CornellUniversityworkedwithmachinelearning

specialiststoidentifythecallsofrightwhalesmore

accurately,makingiteasiertotrackindividualwhales.

Digitalimagesfrommillionsofsatelliteobservationscanbeanalysedforenvironmentalorsocio-economictrendsusingmachinelearningtoidentifypatternsofchangeanddevelopment.

Statisticalmodelsarecreatedusingseriesofalgorithms,orstep-by-stepinstructionsthat

computerscanfollowtoperformaparticulartask.Computeralgorithmsarepowerfultoolsforautomatingmanyaspectsoflifetoday,takingthestep-by-steproutinesthatunderpinthe

administrativeandoperationaltasksoforganisationsanddigitisingthem,makingthemfasterandmoreconsistent.Oneapproachtoautomationistochooseaseriesofrulestoapplyto

inputs,leadingaparticularoutput.Mostcurrentmedicalself-diagnosissystems,bothinbooksandonline,usethislogic.Certaincombinationsofanswerstoquestionsaredeterministicallylinkedtocertainindividualoutputs.Ifyouprovidethesameanswersagain,thealgorithmwillshowthesameresult.

6

Mostmachinelearningapproachesarenotrestrictedtoproducingasingle

predictionfromgiveninputs.

Manyalgorithmsproduceprobabilistic

outputs,offeringarangeoflikely

predictionswithassociatedestimatesofuncertainty.Thealgorithmsproducing

theseprobabilisticoutputsarecapableofbeingunderstoodbyhumans.However,inthecaseofmorecomplexmachine

learningsystems(suchasdeeplearning:seebox),therearemanylayersof

statisticaloperationsbetweentheinputandoutputdata.Theseoperationshavebeendefinedbyanalgorithm,ratherthanaperson.Becauseofthis,notonlyistheoutputprobabilistic,aswithsimpler

algorithms,buttheprocessthatledtoitcannotbedisplayedinhuman-

understandableterms.Thismakes

machinelearningfundamentallydifferenttothekindsofalgorithmsusedto

automatestandardorganisationalroutines.

Inrecentyears,however,availabledataandcomputingpowerhavereachedthepointwhereit

hasbecomepracticaltodevelopmachinelearning:algorithmsthatchangeinresponsetotheir

ownoutput,or“computerprogramsthatautomaticallyimprovewithexperience”

1

.

Machinelearningsystemshaveoftenbeenshowntopickupdifficult-to-spotrelationshipsindatathatmayotherwisehavebeenmissed.

Terminology

Therangeofdifferentstatisticaltechniquesreferredtoherewiththegeneralterm‘a(chǎn)rtificialintelligence’haveemergedoveralongtimefrommanydifferentresearchfieldswithinstatistics,computerscienceandcognitivepsychology.

Consequently,authorsfromdifferentdisciplinestendtomakedifferentdistinctionsbetweentermslike‘machinelearning’and‘machineintelligence’,usingthemtorefertorelatedbutdistinctideas.

Asthesetechniqueshavebeenappliedindifferent

businessareasthey’vebecomerelevanttoothertasks–sothey’relikelytofeaturealsoindiscussionsabout‘datamining’and‘predictiveanalytics’.

Whilethispaperlooksaheadtoatimewhenmachine

learningismorewidespreadthanatpresent,manyoftheopportunitiesandchallengesitdiscussesariseinothercontextstoo.Soratherthanbedistractedwithan

academicdiscussionaboutterminology,we’vechosentousetheumbrellatermartificialintelligence.

Therearemanydifferentkindsofalgorithmusedinmachinelearning.Thekeydistinctionbetweenthemiswhethertheirlearningis‘unsupervised’or‘supervised’.

Unsupervisedlearningpresentsalearningalgorithmwithanunlabelledsetofdata–thatis,withno‘right’or‘wrong’answers–andasksitfindstructureinthedata,perhapsbyclustering

elementstogether–forexample,examiningabatchofphotographsoffacesandlearninghowtosayhowmanydifferentpeoplethereare.Google’sNewsservice

2

usesthistechniqueto

groupsimilarnewsstoriestogether,asdoresearchersingenomicslookingfordifferencesinthedegreetowhichagenemightbeexpressedinagivenpopulation,ormarketerssegmentingatargetaudience.

Supervisedlearninginvolvesusingalabelleddatasettotrainamodel,whichcanthenbeusedtoclassifyorsortanew,unseensetofdata(forexample,learninghowtospotaparticular

personinabatchofphotographs).Thisisusefulforidentifyingelementsindata(perhapskeyphrasesorphysicalattributes),predictinglikelyoutcomes,orspottinganomaliesandoutliers.Essentiallythisapproachpresentsthecomputerwithasetof‘rightanswers’andasksittofindmoreofthesame.

1Mitchell,T.(1997),MachineLearning

2

/

7

currentinterestinmachinelearningis

focusedon‘deepIearning,,asupervised

learningtechniquecombininglayersof

neuralnetworkstoautomaticallyidentify

thefeaturesofadatasetthatarerelevant

todecision-making.Deeplearningisa

powerfuladditiontothemachinelearning

repertoire:however,itrequiresverylarge

amountsofdatatobeeffective.

TheLondon-basedfirmDeepMind(owned

byGoogle)isaworldleaderinthis

technique.

centraltotheinterestaroundmachine

learningisthepotentialitoffersfor

autonomousdecision-making.

Manyalgorithmicprocessescanbeused

tomakedecisionswithouthumaninput

(suchaswhenmortgageproviders

automaticallyapprovelendingto

individualsbasedontheircreditscore).

Butrealautonomycomeswhenasystem

isabletolearncontinuouslyandmake

deductionsabouttheworldwithouthumaninput.Forexample,self-drivingcarsareabletomakereal-timedecisionsaboutspeedanddirectionwithoutinputfromahumandriver,usingmanyinterlinkedmachinelearningsystemstomakesenseofinformationabouttheir

surroundings.Theyarenotfollowingpre-programmeddecisionsbutrespondingtochangesaroundthem.

usingautonomousdecision-makinginotherareasofsocietywouldbeasignificantchangein

thewayweusedataandmakechoices,bringingwithitimportantquestionsaround

accountabilityandtrust.Thisisparticularlytrueforitsusebygovernment,inlightofthe

compactthatexistsbetweenitandcitizens,anditsresponsibilityfortheirwellbeingand

security.Atthepresenttimethereisgenerallyagreementamongstexpertsandpolicy-makersthatimportantdecisionsmustinvolvea‘humanintheloop,一buttheexactnatureoftheirrole,orthedegreetowhichtheyinfluencetheoutcome,issomethingthatislikelytoevolveasthetechnologydevelopsovertime.

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

DeepIearning

Deeplearningisasubsetofmachinelearningthatdependsonusinglayersofnon-linearalgorithmic

processestofindpatternsorclassifydata.Thereare

manydifferenttechniqueswithinthisgeneralapproach一butthekeyfeatureisthattheyeachusealayeredor

stageddesign,inwhichoutputsfromthepreviouslayerareusedasinputsforthenext.

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

8

Artificialintelligenceforinnovationandproductivity

Artificialintelligenceholdsgreatpotentialforincreasingproductivity,mostobviouslybyhelpingfirmsandpeopleuseresourcesmoreefficiently,andbystreamliningthewayweinteractwithlargesetsofdata.Forexample,firmslikeOcadoandAmazonaremakinguseofartificial

intelligencetooptimisetheirstorageanddistributionnetworks,planningthemostefficient

routesfordeliveryandmakingbestuseoftheirwarehousingcapacity.Artificialintelligencecanhelpfirmsdofamiliartasksinmoreefficientways.Importantly,itcanalsoenableentirelynewbusinessmodelsandnewapproachestooldproblems.Forexample,inhealthcare,datafromsmartphonesandfitnesstrackersthatisanalysedusingnewmachinelearningtechniquescanimprovemanagementofchronicconditionsaswellaspredictingandpreventingacuteepisodesofillness.

Artificialintelligencecanhelpbothcompaniesandindividualemployeestobemoreproductive.Routineadministrativeandoperationaljobscanbelearnedbysoftwareagents(‘bots’),whichcanthenprioritisetasks,manageroutineinteractionswithcolleagues(orotherbots),andplanschedules.EmailsoftwarelikeGoogle’sSmartReplycandraftmessagestorespondentsbasedonpreviousresponsestosimilarmessages.Newsroomsareincreasinglyusingmachine

learningtowritesportsreportsandtodraftarticles:intheoffice,similartechnologycanproducefinancialreportsandexecutivebriefings.

Artificialintelligencecanreducetheburdenofsearchinglargesetsofdata.Inthelegalsector,groupslikeROSS,LexMachinaandCaseTextareusingartificialintelligencetosiftcourt

documentsandlegalrecordsforcase-relevantinformation.Otherfirmsareusingsimilar

techniquesaspartofduediligence.Artificialintelligencecanalsoofferawayofinteractingwiththesedatasets,withplatformssuchasIBM’sWatsonabletosupportexpertsystemsthatcananswerfactualnaturallanguagequestions.Forcybersecurityfirms,artificialintelligenceoffersawayofrecognisingunusualpatternsofbehaviourinanetwork.

Theseexamplesfocusonusingsoftwaretodothesamethingashumansbut,inmanycases,analysingdataofvolumeorcomplexitythatisbeyondtheanalyticalcapabilityofindividual

humans.Indeed,artificialintelligenceisnotareplacement,orsubstituteforhumanintelligence.Itisanentirelydifferentwayofreachingconclusions.Artificialintelligencecancomplementorexceedourownabilities:itcanworkalongsideus,andeventeachus,asshownbyLeeSedol’sunbrokenstringofvictoriessinceplayingAlphaGo

3

.Thisoffersnewopportunitiesforcreativityandinnovation.Perhapstherealproductivitygainfromartificialintelligencewillbeinshowingusnewwaystothink.

TheUKisaworldleaderinthescienceunderpinningthistechnology,witharichecosystemofinvestors,employers,developersandclients,andanetworkofsupportingbodiessuchastheAlanTuringInstitute.InnovationsdevelopedfirstinuniversitiessuchasCambridge,Oxford,

ImperialandUniversityCollegeLondonhavefoundtheirwayintotoolsusedbymillionsaround

3“LeeSedolhaswoneverysinglegamehehasplayedsincethe#AlphaGomatchinc.usingsomenewAG-likestrategies-trulyinspiringtosee!”DemisHassabis,CEOofDeepMind,4thMay2016

(/demishassabis/status/728020177992945664,

/player/Lee+Sedol)

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

9

theglobe,withincreasingnumbersofUKstartupselectingtoremainintheUK,furtherstrengtheningexpertiseandcapabilityinthiscountry.

Thepotentialisdrivingarapidtake-upofartificialintelligenceacrossarangeofsectors

4

.InthewordsofthetechnologypunditKevinKelly:“thebusinessplansofthenext10,000startupsareeasytoforecast:TakeXandaddAI”

5

.Estimatingthesizeofthisgrowthischallenginggiventhedifferentdefinitionsof‘a(chǎn)rtificialintelligence’,‘machinelearning’andrelatedterms:itisalsohardtospecifywheresectorslike‘bigdata’endand‘machinelearning’begins.Buta2015US

report

6

suggeststhattheglobalmarketin‘robotsandartificialintelligence-basedsystemswillgrowfrom$58bnin2014to$153bnin2020.

Lookingatbigdatamorewidely,areportearlierthisyearonthevalueofbigdataandthe

internetofthingsestimated￡240billionincumulativebenefitstotheUKbetween2015-20;

manufacturingshouldderivethegreatestbenefits,withthegreatestgainsacrossallsectorssettocomefromefficiencysavings

7

.Anotherreportfrom2015predictsmajoroperationalsavingsforEuropeangovernmentsbyusingbigdata,inadditiontoopportunitiestoincreasetax

revenuesandreducefraudanderror

8

.A2014studyof500UKbusinesses,meanwhile,

concludedthatthosewhomakebetteruseofcustomerandconsumerdataarebetween8and13percentmoreproductivethanfirmswhodon’t

9

.Morebroadly-definedforecastsforthe

impactofsystemscombiningrobotics,dataandartificialintelligence–sometimeslabelled“Industry4.0”

10

–alsopromisesubstantialgains.

AccordingtoMcKinsey,companiesthemselvesanticipateIndustry4.0toincreaserevenuesby23percentandproductivityby26percent

11

.Artificialintelligencehasacentralroleinenablingallofthisgrowth.

4AnoverviewofcurrentworkisofferedbyShivonZillisofBloombergBETAat:

/machineintelligence.

5

/2014/10/future-of-artificial-intelligence/

6

/fastft/2015/11/05/robotics-ai-become-153bn-market-20-bofa/,

/content/dam/boamlimages/documents/PDFs/robotics_andaicondensed_primer.pdf

7TheValueofBigDataandtheInternetofThingstotheUKEconomy,CEBRforSAS,February2016.8Bigdata:Thenextfrontierforinnovation,competition,andproductivity,McKinseyGlobalInstitute,2011.9Bakhshi,Bravo-BioscaandMateos-Garcia(2014)–fromDataDrivenInnovation(OECD).

10BundesministeriumfürBildungundForschung(2013),

Zukunftsbild“Industrie4.0”,

PwC

,Industry4.0:Building

thedigitalenterprise.

11McKinseyIndustry4.0GlobalExpertSurvey2015.

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

10

Theuseofartificialintelligencebygovernment

Governmentisalreadyusingdatasciencetechniquessuchasmachinelearning,andthroughtheworkoftheGovernmentDataProgrammetheiruseisgrowing

12

.Thesetechniquesareprovidinginsightsintoarangeofdata,fromfeedbackondigitalservicedeliverytoagriculturallandusethroughtheanalysissatelliteimages.Astheirsophisticationimprovesmorebenefitsmayberealised.Forexample,wemight:

.Makeexistingservices–suchashealth,socialcare,emergencyservices–moreefficientbyanticipatingdemandandtailoringservicesmoreexactly,enablingresourcestobe

deployedtogreatesteffect.

.Makeiteasierforofficialstousemoredatatoinformdecisions(throughquicklyaccessingrelevantinformation)andtoreducefraudanderror.

.Makedecisionsmoretransparent(perhapsthroughcapturingdigitalrecordsoftheprocessbehindthem,orbyvisualisingthedatathatunderpinsadecision).

.Helpdepartmentsbetterunderstandthegroupstheyserve,inordertobesurethattherightsupportandopportunityisofferedtoeveryone.

Otherapplicationswillbecomeapparentastheuseofdataandartificialintelligencebecomesmoremainstream.

Governmentisaspecialbody,withuniqueobligationsthatdonotfallonprivateorganisations.Itmustbetransparentaboutthewayitacts,followdueprocess,andbeaccountabletoits

citizens.Thismeanstherearespecialresponsibilitiesforgovernment,beyondthegeneralpointsoutlinedabove,whichfollowfromitsuseofartificialintelligenceandbigdata.

Recognisingthis,thegovernmenthaspublishedaguidetotheethicaluseofdatasciencetoolswithingovernmentforgovernmentanalysts

13

.Thisfirstiterationofacodeofpracticewas

developedwithextensiveexternalinputanditeratedwithdatascientistsinsidegovernmenttomakeitaspracticalandusefulaspossible.

Twouseswithparticularrelevanceforgovernmentarehighlightedhere:theuseofartificialintelligencetoadvise,andpossiblelegalimplicationsoftheuseofartificialintelligence.

Advice

Artificialintelligencehasaclearadvisoryroletoplaybeyondtraditional‘decision-support

systems’,supportingdecision-makingthroughassemblingrelevantdata,identifyingpertinentquestionsandtopicsfortheattentionofpolicy-makers,andinhelpingtogeneratewritten

advice.Already,governmentisbeginningtofindvalueinusingartificialintelligencetoassist

publicservantsinthedeliveryofdigitalservices.Itislikelythatmanytypesofgovernment

decisionswillbedeemedunsuitabletobehandedoverentirelytoartificialintelligencesystems.Therewillalwaysbea‘humanintheloop’.Thisperson'srole,however,isnotstraightforward.Iftheyneverquestiontheadviceofthemachine,thedecisionhasdefactobecomeautomaticandtheyoffernooversight.Iftheyquestiontheadvicetheyreceive,however,theymaybe

thoughtreckless,moresoifeventsshowtheirdecisiontobepoor.

12

.uk/category/data-science/

13

.uk/government/uploads/system/uploads/attachment_data/file/524298/Data_science_ethics

_framework_v1.0_for_publication1.pdf

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

11

Aswithanyadviser,theinfluenceofthesesystemsondecision-makerswillbequestioned,anddepartmentswillneedtobetransparentabouttheroleplayedbyartificialintelligenceintheir

decisions

Legalconstraints

Therearecurrentlyspecificlegalframeworks,inadditiontogenerallegislationsuchastheUKDataProtectionAct(1998)andtheEUGeneralDataProtectionRegulation(2016),thatgoverntheuseofcitizens’databygovernmentanalysts,protectingrightstoprivacy,ensuringequal

treatmentforall,andsafeguardingpersonalidentity.Theseareanessentialingredientin

maintainingpublictrustingovernment’sabilitytomanagedatasafely.Teamsmakinguseof

artificiallearningapproachesneedtounderstandhowtheseexistingframeworksapplyinthis

context.Forexample,ifdeeplearningisusedtoinferpersonaldetailsthatwerenotintentionallyshared,itmaynotbeclearwhetherconsenthasbeenobtained.

Thesecurrentprotectionsareeffectiveandwell-established.However,understandingthe

opportunitiesandrisksassociatedwithmoreadvancedartificialintelligencewillonlybepossiblethroughtrialsandexperimentation.Forgovernmentanalyststobeabletoexplorecuttingedgetechniquesitmaybedesirabletoestablishsandboxareaswherethepotentialofthistechnologycanbeinvestigatedinasafeandcontrolledenvironment.

Inadditiontothesethreeareas,theproductiveuseofartificialintelligenceingovernment

dependsonresolvingwiderdatascienceissues:skills,privacy,dataqualityandsoon.

TheworkoftheDataSciencePartnership,ledbytheGovernmentDigitalService(GDS),is

raisingtheawarenessofthepotentialofdatascienceacrossgovernment.Italsoprovidesafocalpointforsharingexperiencesandlessonslearnttopromoteinnovationandthespreadofbestpracticebetweendepartmentsandagencies.

Artificialintelligence:opportunitiesandimplicationsforthefutureofdecisionmaking

12

Effectsonlabourmarkets

Theemergenceofmachinelearning,aswellasrobotics,bigdataandautonomoussystems,islikelytohavesignificantimplicationsfortheeconomyandlabourmarkets.Thesetechnologiestogethercanbeseenaspartofanewwaveof‘generalpurpose’digitaltechnologies

14

,

comparabletothesteamengine,andthemovingassemblyline,withthepotentialtodrive

significantsocio-economicchange.Thereisevidencetosuggestthatthesetechnologiescould

driveproductivitygrowthandsoboosteconomicgrowth,butthereismuchuncertaintyaboutthescaleandthespeedofthesechanges.Theywilldependonboththepaceoftechnological

developmentandthespeedofitsdeploymentbyfirmsacrosstheeconomy.

Inparticular,thesetechnologiesmayhaveaparticularimpactonrolesintheservicesector,

which