利用人工智能和量子計(jì)算增強(qiáng)美國(guó)安全部任務(wù)能力

August2024

ExpertInsights

PERSPECTIVEONATIMELYPOLICYISSUE

NICOLASM.ROBLES,ELIEALHAJJAR,JESSEGENESON,ALVINMOON,CHRISTOPHERSCOTTADAMS,KRISTINJ.LEUSCHNER,JOSHUASTEIER

UsingArtificialIntelligenceand

QuantumComputingtoEnhance

U.S.DepartmentofHomeland

SecurityMissionCapabilities

T

heU.S.DepartmentofHomelandSecurity(DHS)isthethird-largestcabinetdepartmentinthefederalgovernment,bringingtogethermultiplecompo-

nents,includingtheFederalEmergencyManagementAgency(FEMA),the

CounteringWeaponsofMassDestruction(WMD)Office,theU.S.CoastGuard(USCG),andtheU.S.SecretService(USSS),amongothers.Thesecomponentsarechargedwithcarryingoutadiversearrayofmissions:protectingtheUnitedStatesagainstterrorism,securingU.S.borders,securingcyberspaceandcriticalinfrastruc-ture,preservingU.S.economicsecurity,andstrengtheningdisasterpreparednessandresilience.1Tosuccessfullyachievethesemissions,DHSmustleveragetechnolo-giestothefullestextentpossible.

DHSemployswell-testedtechnologiestomanagethecomplexityandresourcethecostsofitsmissions.However,twopowerfulemergingtechnologies—artificial

2

Abbreviations

AIartificialintelligence

CBPU.S.CustomsandBorderProtection

CISACybersecurityandInfrastructureSecurity

Agency

DHSU.S.DepartmentofHomelandSecurity

FEMAFederalEmergencyManagementAgency

GPTgenerativepretrainedtransformer

GREGraduateRecordExamination

LLMlargelanguagemodel

MLmachinelearningNNneuralnetwork

PDEpartialdifferentialequation

PQCpostquantumcryptography

QCquantumcomputing

QKDquantumkeydistribution

QMquantummechanics

QMLquantummachinelearning

QSquantumsensing

QSVMquantumsupport-vectormachine

SVMsupport-vectormachine

TSATransportationSecurityAdministration

UAVuncrewedaerialvehicle

USCGU.S.CoastGuard

USSSU.S.SecretService

WMDweaponsofmassdestruction

intelligence(AI)andquantumcomputing(QC)—mighthavethepotentialtosignificantlyexpandthecapabilitiesavailabletoDHSinthefuture.AI—inparticular,itssubfieldofmachinelearning(ML)—isanumbrellaconceptofusingcomputerstorapidlysolveproblemsforwhichthedevelop-mentofalgorithmsbyhumanprogrammerswouldbecost-prohibitiveorotherwiseextremelydifficult(Murphy,2012).

QCattemptstoleveragetheprinciplesofquantummechan-ics(QM)toobtainquantifiableadvantagesovertraditionalcomputing,bothintermsofspeedandintheabilitytosolveverycomplexproblems.Unlikepreviousleapsintheprog-ressoradvancementofscience,suchasthenuclearprogramorthespaceprogram,whichwerestatesponsored,QCis,forthemostpart,incentivizedandpioneeredbyprivateandfor-profitcompaniesandbyacademicinstitutions(Parker,2021;Parkeretal.,2022).AIismorematurethanQCasadomain,andresearchinAIisdistributedwidelythroughacademiaandindustry.

Althoughthefullpotentialofthesetechnologiesisfarfrombeingrealized,DHScanpositionitscomponentstotakeadvantageoffutureadvancementsbyconsideringhowmatureQC-andAI-basedtechnologiesmightbeusedtoaffectDHSmissionoutcomes.Inthispaper,wearguethat

QCandAItools—iftheirpotentialisrealized—couldsupportDHSmissions,makingDHSmoreeffectiveandefficientandimprovingthelivesofDHSstaffandotherstakeholders.

Ourpredictionsarecontingentonwhethersuccess-fulquantumMLalgorithmscanbediscovered(i.e.,shownmathematicallyorempiricallytobeadvantageousovertheirclassicalcounterparts)andonwhethertheycanrunsmoothlyonpracticalquantumdevices.2Bothissuesarethesubjectofveryintensecutting-edgeresearch.

Inthispaper,webrieflyexplaintheconceptsofQCandAIandthendiscusspotentialapplicationstoDHS’smis-sions.WeconcludethepaperwithrecommendationsonhowDHScouldbestpositionitselftoleverageQCandpre-pareitsworkforce.

3

QuantumComputingandArtificialIntelligence

Astwofieldsofscienceandtechnology,bothAIandQChavegainedextremepopularityinadditiontotheiraccep-tanceinthescientificcommunity.AIhasprovedtobeavaluabletoolinmodernscienceandcomputing.QC,whichitselfisasubfieldofthewiderquantuminformationsciencediscipline,isatanearlierstageofdevelopmentthanAIisbutisstrivingtocatchupwithitsAIcousin.Inthepastdecadeorso,therehavebeenmanyattemptsatmergingthepromisedadvantagesofQCintothefieldofAI,although,todate,theseattemptshavemetwithmixedsuccess(SchuldandPetruccione,2018;seealsoBiamonteetal.,2017).

QuantumComputing

QCattemptstoleveragequantummechanicalphenomena,suchassuperposition,entanglement,andinterference,toobtainquantifiableadvantagesovertraditional,orclassi-cal,computing.QM—thetheoreticalbasisofQC—isoneofthemostsuccessfultheoriesof20th-centuryphysics,withexperimentaltestsverifyingitsvaliditytoincrediblepreci-sion(GriffithsandSchroeter,2018;Sakurai,1994).QMisafundamentaltheoryofnaturethatdescribesthesubatomicworldinwhichclassical(i.e.,Newtonian)physicsfails.Forinstance,inQC,thefamiliarnotionofaninformationbitbeingexclusivelyoff(0)oron(1)nolongerholds.Aquan-tumbit,knownasaqubit,existsinasuperpositionofoffandonsimultaneously—onlyuponmeasurementofaqubitisitforcedtotakeadefinite0or1valuewithspecificprob-abilities,therebycollapsingintoabit.

Thissuperpositionallowsforquantumparallelism,whichistheabilityofquantumcomputerstoevaluateafunctionformultipleinputvaluessimultaneously.Thekeytomanyprovenspeedupsinquantumalgorithmsispre-ciselythisparallelism(DeutschandJozsa,1992;NielsenandChuang,2010).Indeed,thisisagamechangerbecausethesolutionstoverycomplexproblemscannowbeencodedinaregistryofqubits,andresearcherscanextractthedesiredsolutionorpropertiesfromthesequbitsinacontrolledway.

Toproceedwiththisextraction,researchersmustturntheirattentiontootheruniquenotionsthatdonothaveclas-sicalcounterparts,suchasentanglementandinterference.3OneoutstandingexampleofsuchaspeedupandextractionprocedureisShor’salgorithmforprimefactorization,whichfindsprimefactorsofanintegerwithasuperpolynomialspeedup—animprovementoverthebest-knownclassicalalgorithms.4Thishasveryseriousimplicationsincryptog-raphy;asaconsequence,theNationalInstituteofStandardsandTechnologyisstudyinganewarrayofpostquantumcryptography(PQC)algorithmsthatdonotdependoninte-gerfactorization.5

However,theadvantagesofQCoverclassicalcomput-ingarenotstraightforward.Incertainsituations,ratherthansupplyingasuperpolynomialspeedup,QCprovidesamoremodestquadraticspeedup.Awell-studiedsearchalgorithmknownasGrover’salgorithmissuchaninstance.ManyofthequantumalgorithmsthatwediscussinthispaperfallintothequadraticspeedupcategorybecausetheyarederivativesofGrover’salgorithm.Effectively,thismeansthat,ifaclassicalalgorithmrequiresNiterationstoproducearesultwithacertainaccuracy,aquantumalgorithmcould

producethissameresultinonlyO(√)iterations,thereby

providingaquadraticspeedupinruntime.6

4

Thenextgenerationofoperatingsystemsshouldbeabletodeterminewhichtasksshouldbesolvedbyclassicalprocessingunitsandwhichtasksshouldbeoutsourcedtoquantumprocessingunits.

Asaresult,quantumcomputersarenotall-purposecomputersthatwillsomedayreplaceordinarycomputers.Quantumcomputerswilllikelybeemployedprimarilyinthemost-taxingoperationsandthosemostpronetocre-atingbottlenecks(e.g.,Kothari,2020).Indeed,onecouldthinkofquantumdevicesasbeingpowerfulenginesinalargechainofprocesses.Therefore,orchestrationacrosscomputingapproacheswillbekey.Thenextgenerationofoperatingsystemsshouldbeabletodeterminewhichtasksshouldbesolvedbyclassicalprocessingunits(includinghigh-performancecomputersandgraphics-processingunits)andwhichtasksshouldbeoutsourcedtoquantumprocessingunits.Evenoncequantumcomputerswork,therewillprobablybeabreakevenpointatwhichthequan-tumcomputerisworthwhileonlyfortasksthatarebiggerthansomethreshold.

HardwarealsoplacesconstraintsonQC.Severaltech-nologiesandengineeringparadigmsexisttoproducework-ingqubits:superconductors,iontraps,photonics,annealers,neutral-atomtraps,silicon-spinqubits,and(morespecu-latively)topologicalqubitsandnitrogen-vacancycenters.Thesetechnologies(exceptforannealers)sharethesamearchitecture,whichisknownasuniversalgate-basedcom-

puting.Currentdevicesfromprivate-sectorcompanies,federallyfundedresearchanddevelopmentcenters,anduniversitiesproduceverynoisyqubits,soqubitoperationsworksuboptimallyandslowly.Thismeansthat,evenifanalgorithmproducesatheoreticaladvantage,realizingthisadvantagepracticallyisstilldifficultbecausethedevicesarenotyetrobustenough.Certaintechniques,suchaserrormitigationanderrorcorrection,canhelpundothenoisetowhichqubitsarepronebytheirquantumnature.However,thesetechniquesarenotfullydeployableyetandsometimesadduptotheglobaloverheadofthealgorithm,therebyreducingitseffectivenessinsomecases(see,e.g.,GoogleQuantumAI,2023;Mandelbaum,Steffen,andCross,2023;Stamatopoulosetal.,2020;andWoernerandEgger,2019)forcertaintheoreticaloverheadsnotrelatedtoerrorcorrection.Devicesthatareimperfectareknownasnoisy,intermediate-scalequantumcomputers.Annealers,ontheotherhand,haveadifferentarchitecturealtogetherthatisnotgatebasedanddoesnotperformuniversalcalculations,butitexcelsatdiscreteoptimizationandoperationsresearchproblems.7

5

Insum,wenotethefollowingaboutthecurrentstatusofQC:

?TheonlyknownQCalgorithmforAIandMLisGrover’salgorithm.

?ThetheoreticalspeedupofGrover’salgorithmismodestandmightwellbewashedoutbyallneces-saryhardwareoverhead.

?OtherQCalgorithmsforAIandMLmightariseinthefuture,butwhethertheywillisstillunknown.

QuantumComputingandMachineLearning

ThesuccessesofAIarewelldocumented,andAIhasbecomeanindispensabletoolinmoderncomputing,whetherforcommercial,military,orsecurityapplications,asillustratedinKrelina(2021)andQuantumWorkingGroup(2021).ItthusbecomesnaturaltoaskwhetherQCcanfurtherboostMLbyprovidingadvantagesoverclassicalcomput-ing.GiventhesuccessofQMinphysicsontheonehandandthesuccessofMLincomputingontheotherhand,theexpectationsofquantumML(QML)are,ingeneral,dis-proportionatelyhuge(SchuldandPetruccione,2018).ButalthoughthecommercialandbusinessimplicationsofQMLarenowbeingexploredandaddressed,theresultshavenotyetmatchedtheexpectations.

Formany,AImeansMLforbigdata.Thisis,however,oneoftheapplicationsofAIforwhichQCistheleastuseful.AnyapplicationofquantumalgorithmsforthattypeofAIisprobablystillfarinthefuture,giventheneedforhardwareresources(memory,gatespeed,andotherconceptswedis-cussinthispaper)andbecauseitisnotknownyetwhetherQCwouldspeedupthatkindofAIeveninprinciplebecauseofsuchissuesasdata-loading,aswediscusslater.

AssumptionsUnderlyingThisPaper

Astheprecedingdiscussionillustrates,manytechnicalchal-lengeswithAIandQCremaintobesolved.Despitetheseissues,ingeneral,thispaperdoesnotfocusontimelinesorcurrenttechnologyreadinessbecauseAIandQCarestillintheprocessofmaturing.OurviewsonhowquantumtechnologiescouldenhanceAItechniquesarelargelyinde-pendentoftheunderlyingtechnologyusedtoproducethequantumdevicesonwhichtheseQMLalgorithmsaregoingtoberun.Instead,forthispaper,weassumetheexistenceofaworking,orclose-to-working,fault-tolerantquantumcomputer,focusingonwhattheexistenceofsuchatechnol-ogycouldmeanforDHScapabilities.

DHShaspubliclyexpressedinterestinquantumonlyforPQC(DHS,2022).OurviewssuggestthatDHScouldexpandtheseinterestsintootherquantumsubjects,suchasQCandquantumsensing(QS).Informedbyourinvestiga-tions,literaturereview,andprofessionalexperience,weputforwardintheconclusionasetofideasandrecommenda-tionsthatcouldassistDHSinleveragingQCsuccessfullytoprotecttheUnitedStates.

U.S.DepartmentofHomelandSecurityMissions

PerTheDHSStrategicPlan:FiscalYears2020–2024(DHS,2019),thedepartmenthassixprimarygoals:

?Counterterrorismandhomelandsecuritythreats.

?SecureU.S.bordersandapproaches.

?Securecyberspaceandcriticalinfrastructure.

?PreserveandupholdU.S.prosperityandeconomicsecurity.

6

?Strengthenpreparednessandresilience.

?ChampiontheDHSworkforceandstrengthenthedepartment.

Foreachofthesegoals,weprovidesomespecificexam-plesofhowAIorMLcouldaffectDHS’scapabilities.TheconclusionsinthispaperarebasedonourfamiliaritywiththescientificliteratureonQCandAIandonourpreviousandongoingpeer-reviewedscholarship.Whenpossible,wemaketheconnectiontoQMLanddevelopthepotentialbenefitsofprovidingquantumbooststoMLtasks.Notallinstancesoftechnologies’impactwillbepositive,and,inafewinstances,classicaltechniquesaremorethanenoughtoprovidetheneededcapabilitiesorQCsimplyfailstodeliveradvantagesoverclassicalmethods.

Asacaution,weemphasizethatourattemptsatfindinginstancesofprofitableusesofQMLhavenotbeenexhaus-tive,sotherecouldbeotherexamplesorsituationsinwhichquantumadvantagescouldbeimportantforotherDHSactivitiesthatarenotcontemplatedinthispaper.

Beforeproceeding,wementionthebalancethatmustbeachievedtoproduceapaperthatisinformativewithoutbeingexcessivelytechnical.WestrovetodescriberealisticideasandscenariosinwhichAIandQMcouldbemergedtoalleviatethecomputationaltasksthatDHScomponentsmustcompleteaspartofperformingtheirduties.Moreover,thereisnoshortageoftechnicalsourcesinwhichquantumalgorithmsandroutinesarecarefullyelaborated,andwerefertheinterestedreadertoBarnett(2009);Hidary(2019);NielsenandChuang(2010);RieffelandPolak(2014);Scherer(2019);SchuldandPetruccione(2018);SteebandHardy(2018);andWong(2022).However,thesesourcestendtoemphasizethequantitativeaspectsofthesealgorithmsand

largelyignorepotentialapplicationsinindustry,military,andsecurity.

CounterTerrorismandHomelandSecurityThreats

ThefirstDHSmissionistocounterterrorismandhomelandsecuritythreats.Thismissionhasfourobjectives:

?Collect,analyze,andshareactionableintelligence.

?Detectanddisruptthreats.

?Protectdesignatedleadership,events,andsofttargets.

?CounterWMDandemergingthreats.

QCandMLtogethercouldhelpDHSaccomplishthesegoalsinanyofseveralways.

Collect,Analyze,andShareActionableIntelligence

DHSaimstodevelop“timelyandactionableintelligencetoaccuratelyassessandpreventthreatsagainsttheUnitedStates”(DHS,2023).AchallengeforprovidingaccurateandactionableintelligenceistheglutofinformationthatDHScomponentsreceive.DHS’sintelligenceanddomainawarenessoperations,includingthoseintheOfficeofIntel-ligenceandAnalysis,theUSCG,andtheNationalOpera-tionsCenter,mustidentifythreatsbysiftingthroughtensofthousandsofvesselsoperatinginU.S.waters,thousandsofflightsinU.S.airspace,andthousandsoftipsandalertsfil-teringupfromstateandlocalpartners,almostallofwhichareinnocuousnoise.AlthoughDHSreceivesahugevolumeofinformation,itdoesnotcollectthatinformationopti-

7

AchallengeforprovidingaccurateandactionableintelligenceistheglutofinformationthatDHS

componentsreceive.

mally,pullinginalotofnoiseinawaythatrisksmissingimportantsignals.

MLapproaches,potentiallyenabledbyQC,couldtrainonthesedataandhelpDHSintelligenceagentsmorequicklyandaccuratelyidentifytheneedleinthehaystackofinformationtheyhandleeveryday.Largelanguagemodels(LLMs)arealreadyadeptatintegrationandanalysisoflargedatasets,asevidencedbytheperformanceofGenerativePretrainedTransformer(GPT)4onnumerousbenchmarkexams.QCcouldhelpoptimizeintelligencecollection,suchasfromUSCGpatrolsorU.S.CustomsandBorderProtec-tion(CBP)searches,tobetterinformintelligenceoperations.Furthermore,likewedowithproblemsinquantumchem-istry,quantumfinance,andgraphtheory,weexpectthatLLMscouldbeusedtodesigntailor-madequantumarchi-tecturesfortheseintelligencecollectionproblemsbyusingpriorknowledgefromtherelevantresearchcommunities.

DetectandDisruptThreats

AnothercomponentoftheDHSmissiontocounterterror-ismandhomelandsecuritythreatsistodetectanddisruptthreats,suchasthroughtheactionsthattheTransporta-tionSecurityAdministration(TSA)takestosecureairportsandairplanes.Asstatedearlier,matureQCshouldbeable

torapidlyoptimizepatrol,search,andscanstrategiesatcheckpoints,atcriticalinfrastructurelocations,andalongthebordersandapproaches.Theseinnovationswouldaidnotonlyinimprovingdatacollectionforfutureintelligencedevelopmentbutalsoindetectinganddisruptinganyactivethreatsinthepresent.

ProtectDesignatedLeadership,Events,andSoftTargets

TheUSSShastheprimaryroleinprotectingleadership,events,andsofttargetsforthedepartmentinmostcases,althoughtheFederalProtectiveServiceandtheOfficeofHomelandSecuritySituationalAwarenessalsoplayrolesintheprotectionoffederalbuildingsandevents,respectively.Akeytaskforeachofthesecomponentsistoconductriskassessments(ofevents,facilities,andpersonnel)tooptimizethelevelofprotectionthateachreceives,givenlimitedpro-tectionresources.

AnexampleofsuchariskassessmentistheSpecialEventAssessmentRatingsystem,whichdetermineswhatfederalprotectiveassistanceisneededforprivateevents.Theseassessmentsarecurrentlyconductedusingamixofdataanalysisandhumanjudgment,buttheadditionofQCandMLcouldintegratemanymoredatafeedsintothe

8

analysis,thusprovidingamorenuancedandoptimizeddistributionoffederalresourcesandpersonnel.ThiswouldallowDHStoprotectmoreevents,facilities,andpersonnelandprovidebetterassistancetothoseitcurrentlyprotects.

Duringprotectionoperations,thereareadditionalopportunitiesfortheapplicationoftheseemergingtechnol-ogies,includingtheclassificationoftargetsofinterest(e.g.,ataprotectedevent).Forinstance,noisy,intermediate-scalequantumdevicescouldbeusedtotrainaquantumcircuitforclassificationtasksusingexponentiallyfewerparam-etersthanaclassicalneuralnetwork(NN)wouldrequireforthesametask,withapparentlyminimalreductioninperformance(Schuldetal.,2020).Anotherexampleisthatfault-tolerantQCdevicescouldbeappliedtospeedupclas-sificationtasksbyutilizingthemanyquantumalgorithmsforlinearalgebraicroutinesthathavebeendevelopedintheliterature(Cao,Romero,andAspuru-Guzik,2018).Thesecouldbeexecutedeithercentrallyatacommandcenterbasedonsensorfeedsor,inthemoredistantfuture,attheedgebythesensorsthemselves.

Toidentifytargetsquicklyenoughthatactioncanbetakentopreventdangertoleadershipandevents,searchalgorithmsmusthavesufficientlylowrunningtimeandbeusableincombinationwithclassificationalgorithms.Thequantummechanicalpropertiesofinformation,includ-ingentanglementandsuperposition,havethepotentialtoquadraticallyreducetherunningtimeofsearchalgorithms.LLMscanbeusedtodesignnovelquantumarchitecturesthataretailoredtovariousproblems,soanLLM-designedquantumarchitecturetailoredtotheproblemofsearchingforpotentialtargetswouldbeespeciallyusefulinprotectingdesignatedleadership,events,andsofttargets.

CounterWeaponsofMassDestruction

DHSworksto“deter,detect,anddisrupttheuseofweaponsofmassdestruction(WMD)andhealthsecuritydangersasearlyinthethreatpathwayaspossible”(DHS,2019,p.16).ThisincludesemplacingdetectioncapabilitiesatportsofentryandacrosstheUnitedStatesandworkingwithinter-nationalpartnerstosecurepotentiallydangeroussubstancesandprecursors.

ImprovementsindetectioncapabilitiesforWMDusingQCcouldenhanceDHS’sabilitytodisruptWMDpathwaysathomeandabroad.OneapproachtointegratingQCcapa-bilitieswouldinvolvesendingdatafromclassicalsensorstoacentralizedQCcapability.However,thiswouldnecessi-tatethetransformationofdatafromclassicaltoquantumsothatthedatacouldbeusedinaquantumalgorithm.Thistransformation—usuallytermedloadingdataontoaquan-tumdevice—isanexpensiveprocess.Ontheotherhand,ifthedatawerealreadyinquantumform,suchasdatacol-lectedfromaquantumsensor(Krelina,2021;QuantumWorkingGroup,2021),and,ifaquantumalgorithmcouldbedeployedalmostimmediatelyonthisdata,thedata-loadingproblemcouldbebypassed.ThisQSwouldallowCBP,theCounteringWMDOffice,andotherstodetectchemical,biological,radiological,andnuclearthreatsmorequicklyandmoreeffectivelyandtobetterresolvealarmsinthefield.Althoughthismergingconceptisstillexperi-mental,suchacapabilitybeingevenpartiallyrealizedcouldsignificantlyboostthebenefitsofdetectingthesetypesofthreatsatportsofentryorinmetropolitanareas.

AnotheraspectofDHS’scounter-WMDeffortsishorizon-scanningforthreatsfromemergingtechnologies—includingpotentialthreatsresultingfromtheuseofQC,ML,andAI.Forinstance,quantumalgorithmsmightbe

9

AnLLM-designedquantumarchitecturetailoredtotheproblemofsearchingforpotentialtargetswouldbe

especiallyusefulinprotectingdesignatedleadership,events,andsofttargets.

employedtoacceleratethesynthesisofpoisons,nerveagents,biotechnologies,anddrugsthatareharmfulorillegal.

Theseemergingtechnologiescouldalsohavethepoten-tialtocounterthesamethreatsthattheyunleash.Forinstance,QCandAIcouldbeusedtodesigndrugstocoun-tertheeffectsofWMDandotherthreats.ManyadvancesindrugdesignhavecomefromAI—specifically,deepNNsandsupport-vectormachines(SVMs),whichuselargedatasetswiththousandsofmoleculardescriptors.BecausetheseMLalgorithmsarecomputationallyexpensive,therehasbeenarecentpushtousequantumcomputerstoaccelerateMLfordrugdesign.Forthistowork,thesetofmoleculardescrip-torsmustbecompressedforusewithaquantumcomputer. Recentresearchhasuncoveredamethodforcompress-inguptohundredsofthousandsofmoleculesforusewithSVMsanddata-reuploadingclassifiersonaquantumcom-puter(Batraetal.,2021).KushalBatraandhiscolleaguesconsideredsetsofmoleculardescriptorsrepresentingcoro-navirusdisease2019(COVID-19),plague(Yersiniapestis),andtuberculosis.Otherresearchinthisareahasexploitedthefactthatquantum-gateparameterexplorationoffersanadvantageoverNNparameterexplorationbecausetheprobabilisticnatureofquantumsystemsenablesgeneration

ofmoleculesthatwouldnotbeexploredbyaclassicalgen-erativeadversarialnetwork(Lietal.,2021).ThisideawasusedtodevelopnewQMLtechniquesfordrugdiscovery,includingaquantumgenerativeadversarialnetworkthatlearnspatternsfromthesetofmoleculardescriptorsandgeneratessmalldrugmoleculesandaquantumvariationalautoencoderthatperformsaprobabilisticsearchtogeneratelargedrugmolecules.AlthoughDHSwouldnotnecessarilydirectlyemploythesemethodstogeneratenewdrugsandcures,itcouldbenefitfromthemandcouldpreparetohelpdistributetheminanemergency.

SecureU.S.BordersandApproaches

DHShasacriticalmissiontosecureU.S.bordersandenforcecustomsandimmigrationlaws.Thismissioniscomplex,inlargepartbecauseofthesheersizeoftheinterfacebetweenU.S.bordersandtherestoftheworld.Forinstance,CBPactivelymonitorsthousandsofmilesofterritorialbordersand328portsofentry(CBP,2023),whiletheUSCGpatrols4millionsquaremilesofterritorialwatersandexclusiveeco-

10

AsAIadvancesfurther,theneedforahumanpilotmightberelaxed,anduncrewedsystemscouldactastrue

resourcemultipliersforCBP’slimitedhumancapital.

nomiczones(NationalOceanicandAtmosphericAdminis-tration,undated).

SecureandManageAir,Land,andMaritimeBorders

Giventhescopeofitsmission,CBPhasaworkforcechal-lengealongboththenorthernandsouthernborders:CBPhastoofewagentsconductingtoofewpatrolsacrosstoomuchborderarea.Currentoperationsarepersonnelinten-sive,requiringhumanpatrolsbetweenportsofentryatallhoursofthedayandnight.ThisworkforcerequirementhascompoundedbecauseofCBP’sdifficultyinretainingborderagents(see,e.g.,Gambler,2019)andtheincreasednumberofrefugeefamiliesattemptingtocrosstheborder,whichdivertsCBPresourcesawayfromlawenforcementandtowardmigrantaidandarrest(Morgan,2019).

Tocounteractthisshortage,CBPhasbeguntoemployautonomouscapabilities,andadvancesinQCandAIcouldfurtherempowerthesesystems.Since2012,CBPhasusedlargerMQ-9uncrewedaircrafttoconductchange-detectionsweepsalongthesouthernborder(CBP,2022).Inaddition,CBPhasalsobeguntousesmall,uncrewed