人工智能+區(qū)塊鏈可以成為熄燈數(shù)字抵押貸款的一擊兩拳

PAGE

1

AI+Blockchaincanbetheone-twopunchinalights-outdigitalmortgage

by

PrabhakarBhogaraju,LeahPrice,BrianStucky

Abstract

ArtificialIntelligence(AI)andblockchainaretwoofthemostgroundbreakingtechnologiesofourera,withAIreshapingourunderstandingofcognitionandproblem-solving,andblockchainrevolutionizingourperspectivesonmoney,ownership,andvalue.Bothareincreasinglyintertwininginourdailylives,despitebeingintheirearlyphasesofwidespreadadoption

Anemergingareaofinnovationlookstotakeadvantageofthebestofwhatthesetechnologiescanoffer,combiningattributesofbothAI&blockchain.WhileAIbringsautomation,sophisticatedpredictiveanalytics,andgenerativecontentabilities,thereisalackoftransparency-outputscannotbeblindlytrusted.That’swhereblockchaincomesin:offeringtruth,transparencyanddataprovenance.

Inthispaper,weofferahigh-levelviewonwhatthesetwotechnologiesdowellindependently,andwheretheycouldbeevenbettertogether.WhiletheAIandblockchaindevelopercommunitieshavegenerallybeenisolatedfromoneanother,weseetheabilityforthetoolstocomplementeachotherandofferamorecompletesolutionforthemortgageindustry.Wearestillgrapplingwiththeseideasourselves,andourhopeisthatbyofferinganinitialviewonthissubjectwewillsparkevenmorecreativitytodriveourindustryforward.

MortgageTechnology-RecapofKnownFrictions

Withinthemortgageindustry,therearefourwell-knownfrictionsthatcreateopportunity:

DisparateDataSets:Numeroustechnologycomponentsanddatasetsaidthemovementofamortgageloanassetthroughtheproduction&servicingbusinessprocesses–CRMsystems,Point-of-Salesystems(POS),LoanOrigination(LOS),fulfillmentSystemsandmore.

DisparateTransactionSystems:Typically,thesametransactiondataforaloanbeingfundedisentered/managedinmultiplesystemsacrosscounterpartiesinahighlyintermediatedecosystem.

The“checkercheckingthechecker”problem:Dataverificationandduediligenceisavitalcomponentofthemortgageecosystemyetishighlymanualandexpensive.

Paper:Themortgageindustrycontinuestobeapaper-basedindustry,with~90%ofmortgagesoriginatingwithpaperpromissorynotesdespitemorethan20yearsofpushtoadoptelectronicdocuments.

ThesefourfrictionscanbeaddressedbyleveragingAIandblockchaintoolsasaone-twopunch.Together,theyjointlyhavethepotentialtoreducediscrepanciesbetweendisparatedatasetsandtransactionsystems(1and2),automatethevalidationofborrowerinformationanddisseminatetheinformationsecurelytorelevantparties(3),andeliminatetheneedforpaperentirely(4),deliveringthatever-elusivedigitalmortgage

ArtificialIntelligence

GoogledefinesArtificialIntelligenceas“afieldofscienceconcernedwithbuildingcomputersandmachinesthatcanreason,learn,andactinsuchawaythatwouldnormallyrequirehuman

intelligenceorthatinvolvesdatawhosescaleexceedswhathumanscananalyze”.

Artificialintelligencehasbeenmakingsteadyinroadsintothemortgageindustrykeepingwithbroaderfinancialservicesadoption.Inthemid-late1990swesawaproliferationofknowledge-basedexpertsystemsformthefoundationofmanyofourautomatedunderwritingsystems(DesktopUnderwriter,forexample).

Overthepastdecade,breakthroughinnovationsinAI,especiallyinmachinelearninguse-caseslikevision,languageandvoicewereenabledbytheavailabilityofenormousamountsofdigitizeddata,andthecomputingpowertoharnessthatdata.

Whiletherehasbeentremendousprogress,wehavesignificantworkaheadofusintermsofgovernance,compliance,andoversighttogetitrightandscaleitinawaythatdrivesapositiveimpactinourindustry.

Someofthemorecommonuse-casesofAIinourindustryareinthefollowingareas:

Chatbotsforcustomerservice

Docandimagerecognition&processautomation

Task/Workflowautomation

Advanceddata&analyticsandmarketing

Fraudprevention

ContentGenerationformarketing

Blockchain

Blockchainisnewer,relativetoAI,andhasnotyetbeenadoptedatscalebyanyofthelargest

mortgageplayers.Infact,manyinourindustryflippantlyrefertoblockchainas“asolutionlookingforaproblem.”Butthisreductivephraseignoresthefactthatthebasicunderlyingcomponentsofblockchaintecharebeingrapidlyembracedbythemortgageindustry.Simplydefined,Blockchainisasecuredatabasesharedacrossanetworkofparticipants,whereup-to-dateinformationisavailabletoallparticipantsatthesametime(McKinsey).

MITSloanunpacksthispowerfulconceptasfollows:Onablockchain,transactionsarerecordedchronologically,forminganimmutablechain,andcanbeprivateoranonymousdependingonhowthetechnologyisimplemented.Theledgerisdistributedacrossmanyparticipantsinthenetwork—it

doesn’texistinoneplace.Instead,copiesexistandaresimultaneouslyupdatedwitheveryfully

participatingnodeintheecosystem.

Consideringthemortgageindustry,where,onagivenloanthereare10-15individualsmanaging16-20datasets,100+documents,within20-30technologysystems,acrossnumerouscounterpartieswiringfundsbackandforth-themortgagevaluechainisidealforablockchain-likesolution.

Butunlikeartificialintelligence,blockchaintechnologyhasnotyetbecomeembeddedinthemortgagelifecycle(orourdailylives).Broadly,themostvisibleusecaseiscryptocurrency,and,withrecenthigh-profilecryptocompanymeltdownsandnegativecommentarybyregulators,theunderlyingtechnologyitselfseemstohavebeenunfairlyperceivedas“guiltybyassociation”.

Onceyougetpastthehype,thefunctionofablockchainissimple:ittracksownershipandvalueinrealtime,inawaythatissecure,inadisintermediatedfashion.Andsmartcontractcodecreateswhatisessentially“programmablemoney.”Infinancialservices,whereeveryparticipantinatransactioncollectsafee,theeliminationofintermediariesandreal-timesettlementofablockchainsolutionhasthepotentialtoreducethecosttotransactdramatically.Earlyadoptersinspacearealreadyfindingwaystoleverageblockchaintodojustthat.

Usecasesforthemortgageindustry:

Simpletrackingofownershipofatokenizedassetsuchasamortgage

Real-timesettlementinanexchangeofvaluebetweentwoormoreparties

Permissionedaccesstoreal-timeservicingdata

Holisticdigitalassetswheretheentireloandocuments,dataandownershipareaccessibletoallpartieswithinterestintheloan

WherecanAI+blockchainbestdeliver?

Two-prongedapproachtoborrowerdatavalidation

WhileAIcanidentifyworkflowexceptionsinnear-real-time,blockchaincanhelpvalidateandfetch/storeunderlyingdatatoresolveexceptionsinstantly.Withincomecalculationsforexample,AIcanhelpspeed-uptheprocessofreading,harvestingandcomputingincomefrompaystubs,taxreturnsandmore.Thenblockchaintoolscanquicklyconfirmtheunderlyingsourcesasbeingauthentic–deliveringaone-twoefficiencypunchforourindustry.Thisvalidatedpackageofdatacanthenbeassociatedwithatokenizedelectronicmortgagerecordwhichcanbepassed(onablockchain)topermissionedpartiessuchasawarehouselenderoraninvestorintheasset,inaninstantaneousfashion.Thiswouldenableatrulyend-to-endfullydigitalprocess(zeropaper)andwouldreducethetimeforaloanpackagetomakeitswaythroughthesecondarymarket,fromweeksdowntosecondsinrealtime.

WhereAIgeneratesDeepFakes,Blockchainvalidatesauthenticity

AgrowingconcernwithrespecttoAIistheproliferationofdeepfakes,andIPtheft.AI-detectionsoftwareandwatermarkingexisttodayastools,buttheyareconsideredunreliable,anditisexceedinglydifficulttofigureoutwhetherornotmaterialwasgeneratedbyartificialintelligenceorbyahuman.Blockchaintechnology,withitsrootsindataprovenanceandproofsisagreatsolution-setforAIpractitionersworriedaboutdeepfakesandIPtheft.Asitbecomeseveneasierto“fake”documentsanddatawithnewerAItools,weexpectdataprovenancetobecomeevenmoreimportant.

Themortgageindustryalreadyisfamiliarwiththisusecase,using“tamperevidentseals”ondocumentslikeeNotes.Weexpectotherusecases,forexample,appraisalimagesorpropertyinspectionimagesusedinAVMscouldbestampedwithacryptographicsealandstoredasanassetonablockchain,whichcouldbesecurelyaccessedbyanapprovedparticipantinthenetwork.

KYC/AMLUse-Cases

Forregulatedfinancialservicesentities,ariskofusingdecentralizedfinance(DeFi)systemsistheanonymityofactors:howdoesoneconductKYCchecksinapublic,permissionless

blockchainsystem,forexample,whereanyonecanjoin?AIexcelsatfindingabnormalpatternsinwhatseemstobenormaldata.AIcancomplementDeFisystemsinhelpingflagfinancialmalfeasanceexploitingblockchainanonymity(sourcesoffundsforexample).AIcanalsohelpbridgeandtraceblockchainuseractivitywithotherdatasourcesintriangulatingpotentialidentityissues.

MitigateAIdataandmodeltransparencyissues

AImodelsingeneral,andDeeplearningmodelsinparticularrequiresubstantialamountsoftrainingandtestingdata.Thesedatasourcescanbehardtomanage,duetoinherentvolatility,poortaggingandlabelingandconsequentlypotentiallossoftraceability.BlockchaincanhelptrackMLdatasets&modelsmoreefficientlybycreatinguniquedigitalfingerprintsforeachiterationofthedatasetandthealgorithmandenablingimmutableandauditreadymodelgovernanceartifacts.Similarly,modelcontributorsthatpublishmultiplemodelversionsmayexperiencethesameissueswithtagging,modelversionmanagementetc.,whichamplifytransparencyissueswithmachinelearningmodels.Blockchaincanhelpwiththenecessarytracking&tracingofdatasetsusedtobuildthemodels,aswellastrackthemodelsthemselvesonwhatversion/iterationofamodelwasbuiltonexactsetsofdata–greatlyimprovingreplay/rebuildreliabilitynecessaryforvalidation

AccelerateResponsibleAIModelDevelopment

DevelopinghighqualityAImodelsrequireshighqualitydata.Blockchain’sstrengthswithmanagingdata(andbyextension,models)canhelpcreatebetterAImodeldevelopmentplatformsthatcanbesharedacrosscollaborators.Certifieddatasetsbyagenciesorregulatorscanbepublishedonablockchain–andaccelerateresponsibleAIinnovation.

Blockchainefficiency&optimizationuse-cases

AI(likeneuralnetwork-basedmachinelearning)couldalsobeusedtooptimizenode-levelbehaviorsofablockchain.Inmanyblockchainimplementations–eachnodeissupposedtoindependentlycarryacompletecopyofthecollectiveledger….andeachnodeindependentlyvalidateseffectsofchangestotheledger.Thispatternisatthecoreofthecomputationalandresourcecomplexitytypicallyattributedtoblockchain.ThereisworkouttherethatlooksforAIto“optimize”whatneedstobestoredatanode–anddosointelligently(8).

Downsides:CouldAI+blockchaintogetherhurtus?

WhilethispaperoutlinespotentialbenefitsofleveragingAIandblockchain,itisworthalsopointingoutthatthesetwotechnologiescouldworktogethertocauseharm.Forexample,intheyettobedeterminedfuture,couldthesetwotechnologiesleadtothecreationoffullyartificialandsyntheticidentities,assetsandotherartifactsthatcouldthreatenourfinancialinfrastructure?

Andbothtechnologieshavebeencriticizedforbeingresourceintensive.Bitcoinhasbeenafrequenttargetofclimateactivists,particularlybecauseofthecomputationallyintensiveminingthathasanegativeenergyimpactontheenvironment(manyotherblockchainsarehighlyenergyefficient).

AI,withalargerelianceonhard-to-findGPUs,hasalsocomeundersomecriticismforbeingresourceintensive.Movingforward,astechnologyevolves,wewillseemoreenvironmentallyfriendlyiterationsofthissolutionset.

Conclusion

Fragmentedsolutions,un-interoperableplatformsandnon-standard/proprietarydataexchangesarethebarnaclescreatedinourindustry’sjourneytowardsdigitizationandtrulyimpedeourabilitytoachievealights-outdigitalmortgage.Everyparticipantinthemortgageproductionandservicingsupply-chainhasauditsandput-backrisksneedingcheckers-checking-checkers.AIoffersintelligentautomation,inlinerecognitionandreconciliationofdataandexception-basedworkflows.BlockchaincomplimentsAIwithbetterlineage,provenanceandtruth.Thisisaone-twopunch.Stayengaged,test,learnanditerate!

StayEngaged

OurindustrywillcontinuetoseesubstantivedevelopmentsinAIandBlockchainapplications.Thebestwayforustostayabreastofwherethesetechnologiescancreatevalue,andhowbesttowatchoutforpitfalls,istostayengaged.MISMOandMBAaregreatresourcesforourindustrytolooktowards.ThroughtheirCommunitiesofPractice,whitepapers,webinars,andeventsliketheAIForumattheFallMISMO2023Summitandmore,thesemember-ledorganizationsbringthelatestthinkinginAIandBlockchaintotheforeandcanbevaluableinguidingyourexplorationandinvestmentsintheseimpactfultechnologies.

References/Citations

Whatisartificialintelligence?Google.

/learn/what-is-artificial-

intelligence

Whatisblockchain?McKinsey.

/featured-insights/mckinsey-

explainers/what-is-blockchain

Blockchainexplained.MIT.

/ideas-made-to-matter/blockchain-

explained

HowAIisreshapingthemortgageindustry.ForbesFinanceCouncil.

/sites/forbesfinancecouncil/2023/03/09/the-future-of-mortgage-

lending-how-ai-and-humans-can-coexist/?sh=62fa32314a62

DigitalMortgage,What’sNext?

/industries/financial-

services/insights/documents/digital-mortgage-whats-next.pdf

CryptographymayofferasolutiontothemassiveAI-labelingproblem

/2023/07/28/1076843/cryptography-ai-labeling-

problem-c2pa-provenance/

Blockchainfordeeplearning:reviewandopenchallenges.

/article/10.1007/s10586-022-03582-7

AnAIBasedSuperNodesSelectionAlgorithminBlockchainNetworks.

/ftp/arxiv/papers/1808/1808.00216.pdf

Abouttheauthors

Prabhakar“PB”Bhogarajuhasover20yearsofmortgagetechnologyandconsultingexperience.PBisheadofstrategyforFinLockerLLC,aconsumerhomereadinessdataandanalyticsplatform.PriortohistimeatFinLocker,PBheldmultipleexecutiveleadershippositionsinproductdevelopment,datamanagement,anddigitalbusinessarchitectureatFannieMae.PBisanauthorandspeakerinvariousindustryforumsonemergingtechnologyapplicationsforourindustry.HefoundedanemergingtechnologyadvisoryfirmALE-Advisors,andcurrentlyadvisesseveralmortgageindustryfirmsontheirtechnologystrategyanddigitaltransformationinitiatives.PBisamemberoftheBoardofAdvisorsfortheResearchInstituteforHousingAmericaandChairoftheCertifiedMortgageBankersTechnologyCommitteeoftheMortgageBankersAssociation.

LeahPriceisanIndependentAdvisortoFinTech’s.PreviouslysheledthemortgageecosystemdevelopmentverticalatFigureTechnologies,ablockchain-basedfinancialservicescompany.PriortoFigure,LeahhelpeddrivemortgageindustryinnovationwithFannieMaeanditsDayOneCertaintypro