ArtificialIntelligenceandMachineLearning

SurveyAnalysis

Contents

3Introduction

SurveyAnalysis

5TheIndustry’sPerspective

13Theregulatorsperspective

17Pathforward

19References

ArtificialIntelligenceandMachineLearningSurveyAnalysis2

Introduction

AccordingtotheOrganizationforEconomicCo-operationandDevelopment,“Artificial

intelligence(AI)systemsaremachine-basedsystemswithvaryinglevelsofautonomythatcan,foragivensetofhuman-definedobjectives,makepredictions,recommendationsordecisions.AItechniquesareincreasinglyusingmassiveamountsofalternativedatasourcesanddataanalyticsreferredtoas‘bigdata’.Suchdatafeedmachinelearning(ML)modelswhichusesuchdatatolearnandimprovepredictabilityandperformanceautomaticallythroughexperienceanddata,withoutbeingprogrammedtodoso

byhumans.”

Artificialintelligenceandmachinelearning(AI/ML)areonarapidtrajectorytogrowthand

haveprovenindustry-wideacceptanceanduptake.Tomaintainmarketcompetitiveness,regulatoryobjectivesshouldfocusonmaintainingmarketintegrity,economicstability,andconsumerprotectionwhilstenablingthewidespreadconsumptionofemergingtrendslikethedisruptionofAI/ML.Thus,thestreamliningofcomplianceprocesseswithinfinancialinnovationcontextscanimproveexistingproceduresandcatertoefficientworkflows.

Moreover,therearecurrentlynointernationalguidanceandstandardsonAI&ML

specificallyinthecontextoffinancialservices.Thereby,theCentralBankofBahrain

collaboratedwiththeIsraeliSecuritiesAuthorityandtheOntarioSecuritiesCommissiontounderlinethematurenessoftheadoptionandregulationsurroundingAI/MLinitiatives

acrossdifferentjurisdictionsbycirculatingasurveytotheGFIN(GlobalFinancialInnovationNetwork)withalistofcuratedquestionsaimedatunderstandingglobalAI/MLliteracyandsubsequentgovernance.Hence,thesurveyscopedoutthecurrenttrendsofAI/MLinfinancialserviceswithintherespectivejurisdictionsofGFINmembers,exploredthetreatmentofthesetechnologiesbyfinancialregulators,andinducedanoutlookfromaregulator’sperspective.

Thesurveywascirculatedto30GFINmembers,whichincluded:monetaryauthorities,securitiescommissions,centralbanks,insuranceauthorities,commoditycommissions,andpensionregulators.Thesurveyincludedalistofquestionscategorizedintotwomainsections,withonesectionfocusingontheuseofAI/MLtechnologiesbyregulators,

henceexploringthematurenessofsettingadequateandcomprehensivegovernance

protocolsthatadheretoinnovativesolutionofferings,whilsttheothersectionfocusingontheuseAI/MLbyregulatedentities,hencetheadoptionofAI/MLsolutionsbythe

widerindustry.

ArtificialIntelligenceandMachineLearningSurveyAnalysis3

Thefollowingpaperwillprovideinsightintothecirculatedsurveybyanalyzingthe

responsesprovidedbyparticipatingmembers,therebyexploringtheuseofAI/ML

acrossdifferentjurisdictions.This,webelieve,willhelptheGFINnetworkbetter

understandthelevelofadoptiononaglobalscaleandthestanceregulatorstake

whenitcomestoinnovativeofferings.Bydeployingarisk-rewardcontext,regulatorsarecarefultoadoptdisruptorsandensureweighingtheirassociatedriskswiththeirrewardingbenefitswithintheirrespectivejurisdictions.

ArtificialIntelligenceandMachineLearningSurveyAnalysis4

TheIndustry’sPerspective

ArtificialIntelligenceandMachineLearningSurveyAnalysis5

ThefollowingsectionexaminesthesurveyquestionsdirectedtowardsregulatedfinancialentitiesandanalyzestheadoptionofAI/MLtechnologiesbyfinancial

institutions,henceprovidingindustryinsightontheapplicationanduseofAI&MLtechnologies.

AretheauthorizedfinancialserviceprovidersinyourjurisdictioncurrentlyusingAI/ML?

Availability

YesNo

4

26

Around86%ofrespondentsconfirmedhavingauthorizedfinancialserviceprovidersdeployingAI/MLsolutionswithintheirjurisdictions.Hence,themajorityhavealreadydelvedintothistechnologyanddisperseditwithintheirentities,signifyingamajor

uptakeandappetitefortheuseofadvancedtechnologies.Itisimportanttonotethatthesurveyreferstostrictlyauthorizedfinancialentities.

withinthefinancialecosystemdo

providersinyourjurisdictionAI/MLtechnologies?

ArtificialIntelligenceandMachineLearningSurveyAnalysis6

IfYes,inwhichfieldswithinthefinancialecosystemdotheauthorizedfinancialserviceprovidersinyourjurisdictionuseAI/MLtechnologies?

AI/MLAdoptionschemes

AlgorithmicTradingRoboTrading

OpenFinance

Issuers

AssetManagementCreditIntermediariesOther

19%

15%

7%

4%

18%

18%

19%

ThesurveyresponseselucidatethefactthattheadoptionofAI/MLtechnologies

infinancialservicesmayintroduceaplethoraofbusinessmodelsandindustry-wide

usecases.Giventhatmostoftherespondentsofthesurveyaresecuritiesregulators,themostcommonusecaseshighlightedincludeassetmanagementandalgorithmic

trading.ThisprovesthatAI/MLhasthepotentialtohelpinpredictinginvestment

patterns,forecastingassetperformance,andgaugingcapitalmarketmovements,thusitcanbeautilityforassetmanagerstoremaincompetitive.Acommonexampleistheuseofpredictiveanalysistoenhanceexistingcustomerrelationshipmanagementprocedures,hence,AItechniquescanadviseassetmanagersonthenextbeststeptotakebasedonthecurrentsituation,improvingdecision-making.Likewise,algorithmictradingsharedthesamepopularity,scoringamongthehighestadoptionschemes.

Respondentssharedseveralalternativeusecasesadoptedbytheirauthorizedfinancialserviceproviderswhichgobeyondthelistofusecasesoutlinedinthesurvey.

Thoseincludedthefollowing:

1.Surveillance

2.Frauddetection(AML/CFTmonitoring)

3.InusrTech(Automatingunderwriting)

4.Riskmonitoring(riskmodeling)

5.E-KYC(onboardingautomation)

6.Compliancemanagement

7.Customerengagement(chatbots/virtualassistants)

ArtificialIntelligenceandMachineLearningSurveyAnalysis7

IfYes,tothebestofyourestimation,howmanyauthorizedfinancialserviceprovidersuseAI/MLtechnologies?

12%

Adoption%

4%

Upto33%

Upto50%

15%

Upto75%

Morethan75%

69%

Thediagramaboveisbasedonthe26respondentsthatrespondedYEStohaving

authorizedfinancialserviceprovidersusingAI/ML.Mostoftherespondentsclaim

tohaveupto33%oftheirfinancialserviceprovidersusingAI/ML.Thisisasufficient

percentageofthemarketforarelativelynewtechnology,signifyingthewideacceptanceanddesireforitsdeploymentandintegrationacrosssectorsandtargetingdifferent

corporateworkflows.Likewise,itsincreaseduptakemaintainsanadequategeographicdistributionasthreecountriesfromdifferentcontinents(each)markedamorethan

75%marketshare.Thus,itbecomesapparentthatAI/ML’sincreaseduptakeisnotgeographicallyclusteredbutspansdifferentregionsoftheglobe,furtherdenotingitspopularity.

AretheauthorizedfinancialserviceprovidersinyourjurisdictioncurrentlyusingAI/MLforanyofthefollowingpurposes?

Marketing&promotion

15

Scope

22

21

21

4

Fraud&operationalriskComplianceandAMLsurveillance Internaloperations&processesNotusingAI/ML

0510152025Numberofrespondents

ArtificialIntelligenceandMachineLearningSurveyAnalysis8

Asdepictedbythegraphonthepreviouspage,thesurveyresponsesillustrate

substantialadoptionofAI/MLbyauthorizedfinancialservicesprovidersforinternal

proceduresfocusingonenhancingexistingworkflows.ThisincludestheuseofAI/

MLfordifferentbusinessfunctionssuchasautomateddecision-making,predictive

analysis,portfoliomanagement/investmentresearch,aswellasspecificcorporate

valuessuchasESGandCSR,inwhichdifferentAI/MLtechnologiescanbeused

tonotonlyimplementtheagendasbuttoensurethatitdoessoinareliableand

sustainablemanner.Thetwomostcommonusecasesfallundertheriskmanagementandsupervisoryparadigmofanorganization,hencefraudcontrol,compliance

management,andAMLsurveillancescoredamongthehighestpurposes,andallserveacommonunderlyingprinciple,mitigatingorganizationalrisk.AI/MLplaysasignificantroleinriskmanagementinwhichintelligenttechniquessuchasanomalydetection

andthreatcorrelationareimplementedwithinexistingriskmanagementframeworks,makingtheexecutionofthisstrategywithinanorganizationmoreefficientwith

minimizedhumanerrorsassociatedwithwrongfulthreatdetections.

Likewise,marketingandpromotionscoredthelowest,amongtheadoptionuse

cases,denotingagreaterprioritytosecureanadequatelyfunctioningandprofuselyprotectedorganizationoverassociatingwithexternalaffairsandplacingdominanceoverincreasedbrandrecognitionandhighersales.Thisanalogyindicatesthatenterprisestodayplacepivotalvalueonmaintainingsecureandadequateinternalworkflows

andsecuritygovernanceprotocolsandarewillingtouseAI/MLtoimproveexistingprocedurestofurtherenhancetheireffortsandmaximizereturnbenefits.Hence,organizationsplacegreaterimportanceonhavingareliableinternalsystembeforerevertingtomarketingandpromotionalagendas,ultimatelyhelpingbuildamore

ethicalgroundforfutureandexistingcorporations.

ArtificialIntelligenceandMachineLearningSurveyAnalysis9

Inyouropinion,whatarethemainbarriersinyourjurisdictionforauthorizedfinancialserviceproviderstouseAI/ML?

Nobarriers3

Outsourcing1

Accesstoserviceproviders3

Entrybarriers

Privacyconcerns4

Dataavailability6

Resourceconstraint8

Resistancetochange8

Lackoflegislation9

Highcosts17

Staffing/Skillshortage17

0510152025Numberofrespondents

ThispartofthesurveyaimstodiscoverthecommonimpedimentsfacingfinancialentitieswithintheirrespectivejurisdictionsintermsofadoptingAI/MLsolutions.

Asillustratedbythegraphabove,threecountriesreportedhavingnobarriers.Theremainingrespondentsoutlinedawiderangeofinterdependentbarriersclassifiedandillustratedinthegraphabove,withthemostcommonlyreportedbarrierbeingassociatedwithalackofadequatefundingduetothehighcostsofimplementation.Astaggering17respondentsindicatedhavingthisbarrierdecipheringamajorprogressionobstaclehurdlingtheadoptioncapacityworldwide.Thestudyfoundthatlackoffundingisacommonissuethatcorrelateswithsomeoftheremainingbarrierssuchasresourceconstraint,ofwhich8respondentsconfirmedhaving,wherealackofcapitallimitsa

firm’sabilityinfundinginfrastructurerequirementstoimplementAI/MLsolutions.

Likewise,almost30%ofrespondentsreportedresistancetoorganizationalchange

duetohavingadequatetraditionalsystems.Hence,firmsmayfinditdifficultinderivingthevalueofAI/MLandlikelyjustifyingthenecessaryfundsrequiredtooperateit.ThisresistancetochangeisalsoassociatedwithalackoftrustinAI/ML,hencefirmsfind

thatdependingonAItechniquessuchaspredictiveanalysisandautomated-decisionmakingisveryrisky,andthecostsassociatedwithitsfailureareextremelyhigh.

ArtificialIntelligenceandMachineLearningSurveyAnalysis10

ThisriskispartiallydepictedwithinthePrivacyconcernsbarrierpillarinwhich4respondentsreporteddataprotectionchallengesassociatedwithitspotentialmisuse,especiallythesecurityofpersonallyidentifiabledatacanprovetobearprofoundcoststoan

organization.ThisisespeciallybecausetheavailabilityofqualitydataisanintegralpartofAI/MLimplementation.Beingadata-driventechnology,itiscrucialtohaveasufficientflowofdatatooperatethosetechnologies,anothercommonbarrier

reportedbyroughly22%ofrespondents.

Similarly,membersalsoindicatedthehesitancyofsomefirmsinadoptingnewtools

thathavenotdemonstratedmarketefficiencyforaconsiderableamountoftime.

ThisoutlinesagenerallackofunderstandingofAI/ML,whichcanbeseenwithinthe

Staffing/SkillandLackofregulationbarrier.TheStaffing/Skillbarrierscoredamong

thehighest,withalmost62%ofmembersconfirmingitspresence,indicatingamajor

lackofappropriatetalent(datascientists/dataengineers)necessarytooperatethosetechnologies.Despiteitsminimizedhumaninputparadigm,AI/MLstillrequiresskilledlaborwithinitsproductlifecyclefromdevelopmenttoobsolescence.Likewise,this

lackofunderstandingissimilarlydepictedwithinthelackofregulationbarrier,inwhichalmost33%ofmembersreportedashortageinappropriatelegislationandregulationsgoverningthedevelopmentandadoptionofAI/MLsolutionswithintheirrespective

jurisdictions.AvitalcomponenttothesuccessfuldispersionofAI/MLwithinanyregion.

Moreover,despitehavingallthebarrierpillarscorrelatedandinterdependent,

thefoundationthatindirectlystimulatesthosebarriersisthelackofcompetentunderstandingofAI/MLtechnologies,henceAI/MLliteracyenablesafirmor

jurisdictiontodecipheritsvalue,justifyitscosts,supportitsgovernance,andeffectivelydeployandreapitsbenefits.

Fromyoursupervisoryexperience,whichofthefollowingbusinessareaswouldbenefitthemostfromAI/MLforyourauthorisedfinancialserviceproviders?

Businessareas

Other

7

2

10

11

Customerfacingservices Compliance/AMLsurveillanceFrauddetection/riskmanagement

0510152025Numberofrespondents

ArtificialIntelligenceandMachineLearningSurveyAnalysis11

Thispartofthesurveyfurtheraddstothepreviousquestionconcerningadoption

usecases/purposeswhichdenotedagreatersignificanceplaceduponsurveillance

implementationschemesfallingundertheriskmanagementparadigm.Hence,this

questiondelvesdeeperintothepreciseimplementationmeasuresfallingunderthe

supervisoryscope.Hence,compliance/AMLsurveillanceandfrauddetection/risk

managementscoredamongthehighest,averaging36%and33%respectively.This

denotesthekeyroleAI/MLtechnologiesplayintheoverarchinginternalandexternal

enterprisegovernanceprotocol.Automateddecision-makingprogrammedtoexecutewhenaspecifiedcriterionismetisacommontechniqueutilizedinAMLandanti-fraudsystemsdependentonAI/ML.Further,roughly6%ofrespondentsclaimedtoutilizeitforcustomer-facingserviceswhichcanincludechatbotsandvirtualassistantsutilizedtomaintainanorganization’scustomerexcellencyandqualityassurance.Hence,a

virtualassistantcanhelpacustomerfreezetheircardimmediatelywhendetectingsuspiciousbehavior,contributingtosafeguardingtheorganizationanditsclientsfromfraudulentactivity.

Likewise,around23%ofmembersreportedhavingotherusecasesthatwouldbenefitthemostfromAI/MLwhichincludethefollowing;

1.Marketing/promotion

2.Productdesignandinnovation

3.Robo-advisoryandinvestmenttrading

ArtificialIntelligenceandMachineLearningSurveyAnalysis12

TheRegulator’sPerspective

ArtificialIntelligenceandMachineLearningSurveyAnalysis13

Mostoftheregulators(60%)don’tuseAI&MLintheircontinuoussupervision.However,aconsiderablenumberofregulators(40%)useAI&MLandthisrateisexpectedtoincreaseinthefuture.

TheuseofAI&MLintheregulators’continuoussupervision

No

Yes

1218

Amongthesecuritiesauthorities,mostofthem(55%)useAI&MLintheircontinuoussupervision.Inthiscontext,fromthesurveyresultsitcanbesaidthatthesecurities

industryisthemostdominantuserofAI&ML(outofallrespondentsfromthevariousfields;seesection3below).

In-houseofthirdparties’developmentofAI&MLsystemsorapplications

Outofthe12regulatorsthatuseAI&ML,6respondedthattheydevelopedtheAI&MLsystemsorapplicationsin-house,while4regulatorsrespondedtheyusedthirdparty

serviceproviders.2outofthe12regulatorsthatuseAI&MLrespondedthattheyuse

bothin-houseandthird-partyserviceprovidersforthedevelopmentofAI&MLsystemsorapplications.Overall,onlya1/3ofthesurveyrespondentsconfirmtheiradoptionofAI/MLtechnologieswithintheirrespectivejurisdictions,asmallandinconclusivefigurethatmaynotsufficeinrepresentingthewideradoptionscaleandtheappropriateanalysis.Thus,surveyanalysisindicatesthatmostjurisdictionsareatanearlystagewhenitcomes

todelvingintothedispersionofAI/MLtechnologiesandareyettogainacompetentunderstandingthatwillenablethemtosafelydeployandregulateit.

ArtificialIntelligenceandMachineLearningSurveyAnalysis14

AI&MLsystemsorapplicationsdevelopmentbytheregulatororthroughathird-party

Bytheregulator

2

throughathirdpartyboth-byaregulator

6

andthroughathirdparty

4

89%oftheregulatorsthatdeclaredtheydonotuseAI&ML(16outof18)answered

thattheuseofAI&MLtechnologiesisapartoftheirstrategicplanforthenearfuture(coming3years).Therest11%(2outof18)statedthattheuseofAI&MLtechnologiesisn’tapartoftheirstrategicplanforthenearfuture(coming3years),butpartoftheir

long-termstrategicplan(5-10years).

Regulation

12outof30(40%)respondentsansweredthatlaws/regulations/rulesregardingAI&MLtechnologiesarealreadyinplaceintheirjurisdictions.9outofthese12regulatorsrespondedthattheyissuedguidelineswithrespecttoAI&ML.

Yes

No

1812

ItispossiblethatinlightofthegrowingtrendintheuseofAI&MLinthefinancialfields,ledtothepublicationofsuchguidelines.AnotherconclusionmightbethattheadoptionofAI&MLregulationsbythelegislatormakesiteasierandmoreconvenient(evenfroma

practicalorpoliticalstandpoint)fortheregulatorstopublishguidelinesinthisregard.

ArtificialIntelligenceandMachineLearningSurveyAnalysis15

Yet,itshouldbenotedthatthesurveydidnotexaminewhetherthefinancialregulatorsinjurisdictionsthatdidnotissuelaws/regulationsregardingAI&MLtechnologieshadissuedguidelinesinthesematters,anissuethatmaycontradictthisview.

18respondentssaidtheydon’thaveanyregulationregardingAI&MLtechnologies.However,6ofthemsaidtheyandtheauthorizedfinancialserviceprovidersintheirjurisdictionsuseAI&MLtechnologies,andasso8ofthemsaidonlytheauthorizedfinancialserviceprovidersintheirjurisdictionsuseAI&MLtechnologies.

83%oftheregulatorsthatdon’tregulateAI&MLtechnologiesintheirjurisdictions(15outof18),believethattheuseofAI&MLshouldberegulated,mostlybyexistinglawsor\andgeneralonesandbetechnologyneutral.

Toconclude,mostregulatorsdon’tuseAI&MLintheircontinuoussupervisionbut

intendtouseAI&MLtechnologiesaspartoftheirstrategicplanforthenearfuture.

Incontrast,theuseofAIbyfinancialregulatedentitiesismuchbroader,andour

expectationisthatthenumberwillcontinuetorise.Giventhissituation,itisnot

surprisingthatthemajorityoftheregulatorsbelievethatAI&MLshouldberegulated.

ArtificialIntelligenceandMachineLearningSurveyAnalysis16

Pathforward

TheuseofAI/MLgloballyhasaccelerated2,andAI/MLsolutionshavethepotentialtosignificantlychangethefinancialservicesindustryinthenearfuture.Despitetherecentpublicattentionrelatedtochatbots,AI/MLisseenasanopportunitytofreeupresourcestofocusonmorecognitiveaspects.3WhiletherearerisksassociatedwithAI,therearealsobenefitsthatcanimproveconsumerprotectionandfacilitatecompetitivefinancialservices.

OursurveyresultsprovidedausefuloverviewofglobalapproachestotheuseofAI/

MLsolutionsbyfinancialserviceprovidersandregulators.Thesurveyfoundthatthe

vastmajorityofrespondentsthatarefinancialserviceprovidersalreadyuseAI/ML

solutionswithintheirrespectivejurisdictions.Basedonthesurvey,themostcommonAI/MLusecasesformarketparticipantsarerelatedtoriskmanagement,compliance

andAMLsurveillance,andinternaloperationsandprocesses.However,commonbarriersassociatedwithadoptingAI/MLsolutionsincludethelackofadequatefundingand

technologistscapableofimplementingthesesolutions.Similarly,themajorityof

regulatorsfromthediversemembershipofGFINthatrespondedtothesurveydonotcurrentlyuseAI/MLsolutionsforcontinuoussupervision,butmanyrespondentsexpecttheiruseofAI/MLtoincreaseovertime.

MostrespondentsdonotcurrentlyhaveregulationsaddressingtheuseofAI/ML

technologies.However,itisanticipatedthatmoreregulatorswillpublisheither

regulationsorguidelinesinthefuturewhiletryingtofindtherightbalancebetweenprinciples,guidelines,and/orregulationthatisspecificenoughtobehelpfulwhile

remainingbroadenoughtobeadaptabletotechnologicaladvancement.Atthesametime,regulatorsmayalsoneedtoconsiderimplementingnewframeworksandtoolstoaddresstherisksandbenefitsofAI/ML.4

TosupportGFINmembersaddressingthis,weareplanningtwodistinctworkinggroupstreams:

Stream1:FocusonuseofAI/MLbyregulators

Stream2:FocusonuseAI/MLbyregulatedentities

1Chui,M.etal.(2022)ThestateofAIin2022-andahalfdecadeinReview,McKinsey&Company,Accessed:

February9,2023.

2IOSCO(2021)Theuseofartificialintelligenceandmachinelearningbymarketintermediariesandassetmanagers,TheInternationalOrganizationofSecuritiesCommissions,IOSCOFR06/2021

3(2022)Adopt,Innovate,Regulate:EmergingSolutionsfortheUseofAIinFinancialServices,SchwartzReismanInstituteforTechnologyandSociety,Accessed:February9,2023.

ArtificialI