National-Artificial-Intelligence-Research-and-Development-Strategic-Plan-2023-Update國(guó)家人工智能研究與發(fā)展戰(zhàn)略規(guī)劃-2023年更新

NationalArtificialIntelligenceResearchandDevelopmentStrategicPlan

2023UPDATE

AReportbythe

SELECTCOMMITTEEONARTIFICIALINTELLIGENCE

ofthe

NATIONALSCIENCEANDTECHNOLOGYCOUNCIL

May2023

-

PAGE\*roman

ii

-

AbouttheOfficeofScienceandTechnologyPolicy

TheOfficeofScienceandTechnologyPolicy(OSTP)wasestablishedbytheNationalScienceandTechnologyPolicy,Organization,andPrioritiesActof1976toprovidethePresidentandotherswithintheExecutiveOfficeofthePresidentwithadviceonthescientific,engineering,andtechnologicalaspectsoftheeconomy,nationalsecurity,health,foreignrelations,theenvironment,andthetechnologicalrecoveryanduseofresources,amongothertopics.OSTPleadsinteragencyscienceandtechnologypolicycoordinationefforts,assiststheOfficeofManagementandBudgetwithanannualreviewandanalysisoffederalresearchanddevelopmentinbudgets,andservesasasourceofscientificandtechnologicalanalysisandjudgmentforthePresidentwithrespecttomajorpolicies,plans,andprogramsofthe

federalgovernment.Moreinformationisavailableat

/ostp.

AbouttheNationalScienceandTechnologyCouncil

TheNationalScienceandTechnologyCouncil(NSTC)istheprincipalmeansbywhichtheExecutiveBranchcoordinatesscienceandtechnologypolicyacrossthediverseentitiesthatmakeupthefederalresearchanddevelopmententerprise.AprimaryobjectiveoftheNSTCistoensurethatscienceandtechnologypolicydecisionsandprogramsareconsistentwiththePresident’sstatedgoals.TheNSTCpreparesresearchanddevelopmentstrategiesthatarecoordinatedacrossfederalagenciesaimedataccomplishingmultiplenationalgoals.TheworkoftheNSTCisorganizedundercommitteesthatoverseesubcommitteesandworkinggroupsfocusedondifferentaspectsofscienceandtechnology.Moreinformationisavailableat

/ostp/nstc.

AbouttheSelectCommitteeonArtificialIntelligence

TheSelectCommitteeonArtificialIntelligenceadvisesandassiststheNSTCtoimprovetheoveralleffectivenessandproductivityoffederaleffortsrelatedtoartificialintelligence(AI)toensurecontinuedU.S.leadershipinthisfield.Itaddressesnationalandinternationalpolicymattersthatcutacrossagencyboundaries,anditprovidesformalmechanismsforinteragencypolicycoordinationanddevelopmentforfederalAIactivities.ItalsoadvisestheExecutiveOfficeofthePresidentoninteragencyAIpriorities;workstocreatebalancedandcomprehensiveAIR&Dprogramsandpartnerships;leveragesfederaldataandcomputationalresourcesacrossdepartmentandagencymissions;andsupportsanationaltechnicalAIworkforce.TheNationalArtificialIntelligenceInitiativeOfficeprovidestechnicalandadministrativesupportfortheSelectCommitteeonAI.

AbouttheSubcommitteeonMachineLearningandArtificialIntelligence

TheMachineLearningandArtificialIntelligence(MLAI)Subcommittee(MLAI-SC)monitorsthestateoftheartinmachinelearning(ML)andAIwithinthefederalgovernment,intheprivatesector,andinternationallytowatchforthearrivalofimportanttechnologymilestonesinthedevelopmentofAI,tocoordinatetheuseofandfosterthesharingofknowledgeandbestpracticesaboutMLandAIbythefederalgovernment,andtoconsultinthedevelopmentoffederalMLAIR&Dpriorities.TheMLAI-SCreportstotheNSTCCommitteeonTechnologyandtheSelectCommitteeonAI.

AbouttheSubcommitteeonNetworking&InformationTechnologyResearch&Development

TheNetworkingandInformationTechnologyResearchandDevelopment(NITRD)ProgramhasbeentheNation’sprimarysourceoffederallyfundedworkonpioneeringinformationtechnologies(IT)incomputing,networking,andsoftwaresinceitwasfirstestablishedastheHigh-PerformanceComputingandCommunicationsProgramfollowingpassageoftheHigh-PerformanceComputingActof1991.TheNITRDSubcommitteeoftheNSTCguidesthemultiagencyNITRDPrograminitsworktoprovidetheR&DfoundationsforensuringcontinuedU.S.technologicalleadershipandformeetingtheNation’sneedsforadvancedIT.TheNationalCoordinationOffice(NCO)supportstheNITRDSubcommitteeanditsInteragencyWorkingGroups(IWGs)

(/about/

).

AbouttheNITRDArtificialIntelligenceR&DInteragencyWorkingGroup

TheAIR&DInteragencyWorkingGroup(IWG)coordinatesfederalAIR&DandsupportsactivitiestaskedbyboththeNSTCSelectCommitteeonAIandtheSubcommitteeonMachineLearningandArtificialIntelligence.ThisvitalworkpromotesU.S.leadershipandglobalcompetitivenessinAIR&Danditsapplications.TheAIR&DIWGreportsinvestmentstotheAIR&DProgramComponentArea.

AboutThisDocument

Thisdocumentincludesrelevanttextfromthe2016and2019nationalAIR&Dstrategicplans,alongwithupdatespreparedin2023basedonAdministrationandinteragencyevaluationoftheNationalAIR&DStrategicPlan:2019UpdateaswellascommunityresponsestoaRequestforInformationonupdatingthePlan.The2019strategieswerebroadlydeterminedtobevalidgoingforward.The2023updateaddsanew

Strategy9,

whichestablishesaprincipledandcoordinatedapproachtointernationalcollaborationinAIresearch.

Copyright

ThisdocumentisaworkoftheUnitedStatesgovernmentandisinthepublicdomain(see17U.S.C.§105).Asacourtesy,weaskthatcopiesanddistributionsincludeanacknowledgmenttoOSTP.PublishedintheUnitedStatesofAmerica,2023.

Note:Anymentioninthetextofcommercial,non-profit,academicpartners,ortheirproducts,orreferencesisforinformationonly;itdoesnotimplyendorsementorrecommendationbyanyU.S.governmentagency.

TheNationalArtificialIntelligenceR&DStrategicPlan

-

PAGE\*roman

viii

-

NATIONALSCIENCEANDTECHNOLOGYCOUNCIL

Chair

AratiPrabhakar,Director,OfficeofScienceandTechnologyPolicy(OSTP),Assistanttothe

PresidentforScienceandTechnology

ActingExecutiveDirector

KeiKoizumi,PrincipalDeputyDirectorforPolicy,OSTP

SELECTCOMMITTEEONARTIFICIALINTELLIGENCE

Chair

AratiPrabhakar,Director,OSTP,AssistanttothePresidentforScienceandTechnology

RotatingCo-Chairs

LaurieLocascio,UndersecretaryofCommerceforStandardsandTechnology,DepartmentofCommerce

SethuramanPanchanathan,Director,NationalScienceFoundation(NSF)

GeraldineRichmond,UnderSecretaryforScienceandInnovation,DepartmentofEnergy(DOE)

SUBCOMMITTEEONMACHINELEARNINGANDARTIFICIALINTELLIGENCE

Co-Chairs

StephenBinkley,NationalNuclearSecurityAdministration,DOE

ErwinGianchandani,AssistantDirectorforTechnology,InnovationandPartnerships,NSF

TessdeBlanc-Knowles,SeniorPolicyAdvisor,OSTP

ElhamTabassi,AssociateDirectorforEmergingTechnology,InformationTechnologyLaboratory,NationalInstituteofStandardsandTechnology(NIST)

ExecutiveSecretary

FaisalD’Souza,NITRDNationalCoordinationOffice(NCO)

SUBCOMMITTEEONNETWORKINGANDINFORMATIONTECHNOLOGYRESEARCHANDDEVELOPMENT(NITRD)

Co-Chair Co-Chair

MargaretMartonosi,AssistantDirectorforComputerandInformationScienceandEngineering,NSF

Kathleen(Kamie)Roberts,NITRDNCO

ExecutiveSecretary

NekeiaButler,NITRDNCO

ARTIFICIALINTELLIGENCERESEARCHANDDEVELOPMENTINTERAGENCYWORKINGGROUP

Co-Chairs

StevenL.Lee,OfficeofAdvancedScientificComputingResearch,OfficeofScience,DOE

MichaelL.Littman,DirectorateforComputerandInformationScienceandEngineering,DivisionofInformationandIntelligentSystems,NSF

TechnicalCoordinator

FaisalD’Souza,NITRDNCO

CraigI.Schlenoff,IntelligentSystemsDivision,EngineeringLaboratory,NIST

WritingTeam

GilAlterovitz,VA

FaisalD’Souza,NITRDNCO

AllisonDennis,NIH

KyleFox,NIJ

CraigGreenberg,NIST

WilliamHarrison,NIST

RobertHickernell,NISTJonnieBradley,DOE/AITODavidKuehn,DOT

StevenLee,DOE/SCMichaelLittman,NSFZhiyongLu,NIH

JillianMammino,State

AaronMannes,DHSNikunjOza,NASARichardPaladino,MDAPavelPiliptchak,NISTCraigI.Schlenoff,NISTAdriaSchwarber,StateRamD.Sriram,NIST

AnnStapleton,USDAElhamTabassi,NISTMaryTheofanos,NISTStevenThomson,USDAJayVietas,NIOSHJamesWarren,NIST

MeganZimmerman,NIST

TableofContents

ExecutiveSummary vii

IntroductiontotheNationalAIR&DStrategicPlan:2023Update 1

AIasaNationalPriority 1

Strategy1:MakeLong-TermInvestmentsinFundamentalandResponsibleAIResearch 3

AdvancingData-FocusedMethodologiesforKnowledgeDiscovery 3

FosteringFederatedMLApproaches 4

UnderstandingTheoreticalCapabilitiesandLimitationsofAI 4

PursuingResearchonScalableGeneral-PurposeAISystems 5

DevelopingAISystemsandSimulationsAcrossRealandVirtualEnvironments 5

EnhancingthePerceptualCapabilitiesofAISystems 5

DevelopingMoreCapableandReliableRobots 6

AdvancingHardwareforImprovedAI 6

CreatingAIforImprovedHardware 7

EmbracingSustainableAIandComputingSystems 8

Strategy2:DevelopEffectiveMethodsforHuman-AICollaboration 9

DevelopingtheScienceofHuman-AITeaming 9

SeekingImprovedModelsandMetricsofPerformance 10

CultivatingTrustinHuman-AIInteractions 10

PursuingGreaterUnderstandingofHuman-AISystems 10

DevelopingNewParadigmsforAIInteractionsandCollaborations 10

Strategy3:UnderstandandAddresstheEthical,Legal,andSocietalImplicationsofAI 12

MakingInvestmentsinFundamentalResearchtoAdvanceCoreValuesThroughSociotechnicalSystemsDesign

andontheEthical,Legal,andSocietalImplicationsofAI 12

UnderstandingandMitigatingSocialandEthicalRisksofAI 13

UsingAItoAddressEthical,Legal,andSocietalIssues 14

UnderstandingtheBroaderImpactsofAI 15

Strategy4:EnsuretheSafetyandSecurityofAISystems 16

BuildingSafeAI 16

SecuringAI 17

Strategy5:DevelopSharedPublicDatasetsandEnvironmentsforAITrainingandTesting 18

DevelopingandMakingAccessibleDatasetstoMeettheNeedsofaDiverseSpectrumofAIApplications 18

DevelopingSharedLarge-ScaleandSpecializedAdvancedComputingandHardwareResources 20

MakingTestingResourcesResponsivetoCommercialandPublicInterests 21

DevelopingOpen-SourceSoftwareLibrariesandToolkits 21

Strategy6:MeasureandEvaluateAISystemsthroughStandardsandBenchmarks 22

DevelopingaBroadSpectrumofAIStandards 22

EstablishingAITechnologyBenchmarks 24

IncreasingtheAvailabilityofAITestbeds 25

EngagingtheAICommunityinStandardsandBenchmarks 25

DevelopingStandardsforAuditingandMonitoringofAISystems 26

Strategy7:BetterUnderstandtheNationalAIR&DWorkforceNeeds 27

DescribingandEvaluatingtheAIWorkforce 27

DevelopingStrategiesforAIInstructionalMaterialatAllLevels 28

SupportingAIHigherEducationStaff 28

Training/RetrainingtheWorkforce 28

ExploringtheImpactofDiverseandMultidisciplinaryExpertise 29

IdentifyingandAttractingtheWorld’sBestTalent 29

DevelopingRegionalAIExpertise 29

InvestigatingOptionstoStrengthentheFederalAIWorkforce 30

IncorporatingEthical,Legal,andSocietalImplicationsintoAIEducationandTraining 30

CommunicatingFederalWorkforcePrioritiestoExternalStakeholders 30

Strategy8:ExpandPublic-PrivatePartnershipstoAccelerateAdvancesinAI 31

AchievingMorefromPublic-PrivatePartnershipSynergies 31

ExpandingPartnershipstoMoreDiverseStakeholders 32

Improving,Enlarging,andCreatingMechanismsforR&DPartnerships 32

Strategy9:EstablishaPrincipledandCoordinatedApproachtoInternationalCollaborationinAIResearch 34

CultivatingaGlobalCultureofDevelopingandUsingTrustworthyAI 34

SupportingDevelopmentofGlobalAISystems,Standards,andFrameworks 35

FacilitatingInternationalExchangeofIdeasandExpertise 36

EncouragingAIDevelopmentforGlobalBenefit 36

EvaluatingFederalAgencies’ImplementationoftheNAIIAandStrategicPlan 38

ListofAbbreviationsandAcronyms 39

Endnotes 40

ExecutiveSummary

Artificialintelligence(AI)

1

isoneofthemostpowerfultechnologiesofourtime.InordertoseizetheopportunitiesthatAIpresents,theNationmustfirstworktomanageitsrisks.Thefederalgovernmentplaysacriticalroleinthiseffort,includingthroughsmartinvestmentsinresearchanddevelopment(R&D)thatpromoteresponsibleinnovationandadvancesolutionstothechallengesthatothersectorswillnotaddressontheirown.ThisincludesR&DtoleverageAItotacklelargesocietalchallengesanddevelopnewapproachestomitigateAIrisks.ThefederalgovernmentmustplacepeopleandcommunitiesatthecenterbyinvestinginresponsibleR&Dthatservesthepublicgood,protectspeople’srightsandsafety,andadvancesdemocraticvalues.ThisupdatetotheNationalAIR&DStrategicPlanisaroadmapfordrivingprogresstowardthatgoal.

ThisplandefinesthemajorresearchchallengesinAItocoordinateandfocusfederalR&Dinvestments.ItwillensurecontinuedU.S.leadershipinthedevelopmentanduseoftrustworthyAIsystems,preparethecurrentandfutureU.S.workforcefortheintegrationofAIsystemsacrossallsectors,andcoordinateongoingAIactivitiesacrossallfederalagencies.

2

Thisplan,whichfollowsnationalAIR&Dstrategicplansissuedin

2016

and

2019,

reaffirmseightstrategiesandaddsaninthtounderscoreaprincipledandcoordinatedapproachtointernationalcollaborationinAIresearch:

Strategy1:Makelong-terminvestmentsinfundamentalandresponsibleAIresearch.PrioritizeinvestmentsinthenextgenerationofAItodriveresponsibleinnovationthatwillservethepublicgoodandenabletheUnitedStatestoremainaworldleaderinAI.ThisincludesadvancingfoundationalAIcapabilitiessuchasperception,representation,learning,andreasoning,aswellasfocusedeffortstomakeAIeasiertouseandmorereliableandtomeasureandmanagerisksassociatedwithgenerativeAI.

Strategy2:Developeffectivemethodsforhuman-AIcollaboration.IncreaseunderstandingofhowtocreateAIsystemsthateffectivelycomplementandaugmenthumancapabilities.Openresearchareasincludetheattributesandrequirementsofsuccessfulhuman-AIteams;methodstomeasuretheefficiency,effectiveness,andperformanceofAI-teamingapplications;andmitigatingtheriskofhumanmisuseofAI-enabledapplicationsthatleadtoharmfuloutcomes.

Strategy3:Understandandaddresstheethical,legal,andsocietalimplicationsofAI.Developapproachestounderstandandmitigatetheethical,legal,andsocialrisksposedbyAItoensurethatAIsystemsreflectourNation’svaluesandpromoteequity.Thisincludesinterdisciplinaryresearchtoprotectandsupportvaluesthroughtechnicalprocessesanddesign,aswellastoadvanceareassuchasAIexplainabilityandprivacy-preservingdesignandanalysis.Effortstodevelopmetricsandframeworksforverifiableaccountability,fairness,privacy,andbiasarealsoessential.

Strategy4:EnsurethesafetyandsecurityofAIsystems.AdvanceknowledgeofhowtodesignAIsystemsthataretrustworthy,reliable,dependable,andsafe.Thisincludesresearchtoadvancetheabilitytotest,validate,andverifythefunctionalityandaccuracyofAIsystems,andsecureAIsystemsfromcybersecurityanddatavulnerabilities.

Strategy5:DevelopsharedpublicdatasetsandenvironmentsforAItrainingandtesting.Developandenableaccesstohigh-qualitydatasetsandenvironments,aswellastotestingandtrainingresources.Abroader,morediversecommunityengagingwiththebestdataandtoolsforconductingAIresearchincreasesthepotentialformoreinnovativeandequitableresults.

Strategy6:MeasureandevaluateAIsystemsthroughstandardsandbenchmarks.DevelopabroadspectrumofevaluativetechniquesforAI,includingtechnicalstandardsandbenchmarks,informedbytheAdministration’s

BlueprintforanAIBillofRights

and

AIRiskManagementFramework

(RMF).

Strategy7:BetterunderstandthenationalAIR&Dworkforceneeds.ImproveopportunitiesforR&DworkforcedevelopmenttostrategicallyfosteranAI-readyworkforceinAmerica.ThisincludesR&DtoimproveunderstandingofthelimitsandpossibilitiesofAIandAI-relatedwork,andtheeducationandfluencyneededtoeffectivelyinteractwithAIsystems.

Strategy8:Expandpublic-privatepartnershipstoaccelerateadvancesinAI.PromoteopportunitiesforsustainedinvestmentinresponsibleAIR&Dandfortransitioningadvancesintopracticalcapabilities,incollaborationwithacademia,industry,internationalpartners,andothernon-federalentities.

Strategy9:EstablishaprincipledandcoordinatedapproachtointernationalcollaborationinAIresearch.PrioritizeinternationalcollaborationsinAIR&Dtoaddressglobalchallenges,suchasenvironmentalsustainability,healthcare,andmanufacturing.StrategicinternationalpartnershipswillhelpsupportresponsibleprogressinAIR&DandthedevelopmentandimplementationofinternationalguidelinesandstandardsforAI.

ThefederalgovernmentplaysacriticalroleinensuringthattechnologieslikeAIaredevelopedresponsibly,andtoservetheAmericanpeople.FederalinvestmentsovermanydecadeshavefacilitatedmanykeydiscoveriesinAIinnovationsthatpowerindustryandsocietytoday,andfederallyfundedresearchhassustainedprogressinAIthroughoutthefield’sevolution.Federalinvestmentsinbasicandappliedresearch

3

havedrivenbreakthroughsenabledbyemergingtechnologieslikeAIacrosstheboard,includinginclimate,agriculture,energy,publichealth,andhealthcare.StrategicfederalinvestmentsinresponsibleAIR&DwilladvanceacomprehensiveapproachtoAI-relatedrisksandopportunitiesinsupportofthepublicgood.

TheNationalArtificialIntelligenceR&DStrategicPlan

-

PAGE

10

-

IntroductiontotheNationalAIR&DStrategicPlan:2023Update

Advancesingenerating,collecting,processing,andstoringdatahaveenabledinnovationinAI,allowingthistechnologytobecomeubiquitousinmodernlifeandtouchnearlyeveryfacetofdailyactivities,directlyorindirectly.BesidestheAI-enabledapplicationsinsmartphonesandpersonalcomputers,applicationsofAIhavestreamlinedlogistics,acceleratedscientificdiscovery,enabledmoreefficientdesignandmanufacturing,andaidedindetectingfinancialfraud.However,realizingAI’spotentialsocialandeconomicbenefitsandaligningitwithAmericanvaluesrequiresconsiderableresearchinvestments,pursuedinaccordancewiththeprinciplesofscientificintegrity.

InFebruary2022,theOfficeofScienceandTechnologyPolicy(OSTP)issuedaRequestforInformation(RFI)

4

requestinginputfromallinterestedpartiesonthedevelopmentofthisplan.Over60responsesweresubmittedbyresearchers,researchorganizations,professionalsocieties,civilsocietyorganizations,andindividuals;theseresponsesareavailableonline.

5

ManyoftheRFIresponsesreaffirmedtheanalysis,organization,andapproachoriginallyoutlinedinthe2016and2019strategicplans.ItisnoteworthythatamajorityoftheRFIresponsesreferredtoaspectsofethical,legal,andsocietalimplicationsofAI(

Strategy3

)orsafetyandsecurityofAIsystems(

Strategy4

).Theseresponsesunderscoreaheightenedpriorityacrossacademia,industry,andthepublicfordevelopinganddeployingAIsystemsthataresafe,transparent,andimproveequity,andthatdonotviolateprivacy.ResponsestotheRFIalsoemphasizedtheimportanceofsupportingAIR&DthatwilldevelopsystemscapableofhelpingtoaddresssomeoftheforemostchallengesandopportunitiesbeforetheNationtoday,includingadvancingpersonalizedmedicine;improvingcybersecurity;addressinginequities;bringingefficienciestomanufacturing,transportation,andothercriticalsectorsoftheeconomy;ensuringenvironmentalsustainability;andenablingthescientificdiscoveryandinnovationthatwillpowerthenextgenerationoftechnologicalbreakthroughs.

AIasaNationalPriority

TheBiden-HarrisAdministrationiscommittedtoadvancingresponsibleAIsystemsthatareethical,trustworthy,andsafe,andservethepublicgood.Thefiscalyear(FY)2023President’sBudgetRequestincludedsubstantialandspecificfundingrequestsforAIR&D,aspartofabroadexpansionoffederallyfundedR&Dtoadvancekeytechnologiesandaddresssocietalchallenges.

6

TheCHIPSandScienceActof2022

7

andConsolidatedAppropriationsAct,2023

8

reflectAdministrationandCongressionalsupportforanexpansionoffederallyfundedR&D,includingAIR&D.

9

ThememorandumonMulti-AgencyResearchandDevelopmentPrioritiesfortheFY2024Budget

10

issuedjointlybytheOfficeofManagementandBudgetandOSTPlikewisecallsforagenciestoprioritizeR&Dfundingtowardadvancingnationalsecurityandtechnologicalcompetitiveness,includingtrustworthyAI,amongothercriticalandemergingtechnologiesofnationalinterest.ThisplanpursuestheadvancementoffundamentalandtranslationalAIresearchtomakeAItrustworthy,equitable,andbothrights-andprivacy-preserving.

TheNationalAIInitiativeAct(NAIIA)of2020establishedtheNationalAIInitiativeOffice(NAIIO)tocoordinatekeyAIactivitiesacrossthefederalgovernment.Thisoffice,basedintheWhiteHouseOSTP,isthecentralpointofcontactfortechnicalandprogrammaticinformationexchangeonactivitiesrelatedtotheNationalAIInitiativeacrossthefederalgovernment,academia,industry,nonprofitorganizations,professionalsocieties,civilsociety,andstate,local,andtribalgovernments.Inaddition,theNAIIOhelpsadvanceprogressontheprioritiesoutlinedinthisplanandimplementsacomprehensiveapproachtoAI-relatedrisksandopportunitiesinsupportofthepublicgood.

WhileR&Dactivitiesandoutputsinformgovernanceandregulatoryapproaches,thisplanleavesdiscussionsofregulationorgovernancetootherfederaldocuments,suchastheBlueprintforanAIBillofRightsandtheAIRiskManagementFramework.Inaddition,issuesrelatedtoscientificintegrityandpublicaccess,whiledirectlyrelevanttoAIR&D,arelargelylefttootherfederalgovernmentdocumentsaswell.

Strategy1:MakeLong-TermInvestmentsinFundamentalandResponsibleAIResearch

TheUnitedStateshasmaintaineditsleadershipinAIinlargepartbecauseofcontinuedandconsistentinvestmentinlong-term,fundamentalAIresearch.Forexample,manyoftoday’sAI-enabledproductsandserviceshavetheirrootsinfederallyfundedfundamentalresearchdatingbackdecades.Thistrendhascontinuedsincethereleaseofthe2019StrategicPlan,withanotableincreaseinAIR&Dfundedbythefederalgovernment.Forexample,theAIInitiativeandcompanioneffortsfundedbytheDepartmentofEnergy(DOE)OfficeofSciencehaveenabledgroundbreakingdiscoveriesinareasfromfusionenergytoSARS-CoV-2understanding.TheNationmustcontinuetofosterlong-term,fundamental,andresponsibleresearchinAItoallowforfurtherdiscoveriesandinnovationswithlong-termbenefits.

InvestmentsinfundamentalAIR&Dspanthespectrumfromfoundationaltouse-inspiredresearch.Forexample,foundationalinvestmentinAIR&Ddrivesforwardlearning,reasoning,planning,knowledgerepresentation,computervision,andbeyond,withpotentialforscale-upandadoptioninpractice.Use-inspiredAIresearch,meanwhile,contributestoadvancesinAIwhilealsoadvancingareassuchasagriculture,healthcare,manufacturing,economics,criticalinfrastructure,andsustainability,withthegoalofengagingandimprovingallofsocietywhilerespectingindividualfreedoms.

OfparticularimportanceistheinvestmentinthedevelopmentofassuranceandtrustinAIsystems,asreflectedin

Strategy3

and

Strategy4.

ResearchintheseareasisessentialforusingAIinallfields,butitisparticularlyimportantinsystemsthatinvolvesafetyorapplicationsinwhichAIdecisionsaffectindividuals,groups,communities,andtheenvironment.MostAIR&DthusfarhasfocusedontheadvancementofAIforindividualtasks.Additionalworkisneededtosolveincreasinglydifficultscienceandtechnologychallengescoveringmultipledomainsandapplications,movingtowardthevisionofgeneral-purposeAI.AIR&DincreasinglyattemptstoconsiderhowvariousareasofAIworkcanfittogetherintoanintegratedsystem.Asaresult,thisstrategyincludesprioritiesthatcontinuetoadvanceAIforindividualtasksbutalsoaimtowardthevisionofgeneral-purposeAIsystems.Theprioritiesinvolveusingthesignificantamountofavailabledataformachinelearning(ML)andknowledgediscovery,improvingtheabilitiesofAItoperceiveandact,anddevelopingscalable,general-purposesystemstoworkinrealandvirtualenvironments.

Finally,developingatheoreticalunderstandingofthecapabilitiesandlimitationsofAIsystemscaninformwhatR&DshouldbedoneandiscriticalforenablingsafeuseofAI.Forexample,abetterunderstandingofhowdeepnetworksconstructeffectiverepresentationscouldleadtonewnetworkdesignsthatcanreasonaboutuncertaintymoredirectlyandwithoutrequiringasmuchdatatotrain.

Thisstrategyisdividedalongtenlinesofeffort:AdvancingData-FocusedMethodologiesforKnowledgeDiscovery;FosteringFederatedMLApproaches;UnderstandingTheoreticalCapabilitiesandLimitationsofAI;PursuingResearchonScalableGeneral-PurposeAISystems;DevelopingAISystemsandSimulationsAcrossRealandVirtualEnvironments;EnhancingthePerceptualCapabilitiesofAISystems;DevelopingMoreCapableandReliableRobots;AdvancingHardwareforImprovedAI;CreatingAIforImprovedHardware;andEmbracingSustainableAIandComputingSystems.

AdvancingData-FocusedMethodologiesforKnowledgeDiscovery

AsdiscussedintheFederalBigDataResearchandDevelopmentStrategicPlanfrom2016,

11

newtoolsandtechnologiesareneededtoachieveintelligentdataunderstandingandknowledgediscovery.Forexample,progressonthedevelopmentofmoreadvancedAIsystemswillhelpidentifyusefulinformationhiddeninbigdata.Manyopenresearchquestionsrevolvearoundthecreationanduseofdata,