东盟人工智能治理与伦理指南-东南亚国家联盟（英）.docx

ASEANGuideonAlGovernanceandEthicsContentsExecutiveSummary03A. Introduction061. Objectives082. Assumptions083. TargetAudience094. Definitions09B. GuidingPrinciplesfortheFramework101. TransparencyandExplainability-112. FairnessandEquity123. SecurityandSafety134. Human-centricity145. PrivacyandDataGovernance-146. AccountabilityandIntegrity157. RobustnessandReliability16C. AlGovernanceFramework171. Internalgovernancestructuresandmeasures182. DeterminingthelevelofhumaninvolvementinAl-augmenteddecision-making-233. Operationsmanagement274. Stakeholderinteractionandcommunication41D. National-levelRecommendations46E. Regional-levelRecommendations53F. Conclusion58Annex A: AlRiskImpactAssessmentTemplate_60Annex B: UseCases731. AboitizGroup-742. EY763. Gojek784. CARE.AI805. SmartNationGroup(SNG)1Singapore836. MinistryofEducation,Singapore85ExecutiveSummaryWhattheGuideisaboutThisdocumentservesasapracticalguidefororganisationsintheregionthatwishtodesign,develop,anddeploytraditionalAltechnologiesincommercialandnon-militaryordual-useapplications.ThisGuidefocusesonencouragingalignmentwithinASEANandfosteringtheinteroperabilityofAlframeworksacrossjurisdictions.Italsoincludesrecommendationsonnational-levelandregional-levelinitiativesthatgovernmentsintheregioncanconsiderimplementingtodesign,develop,anddeployAlsystemsresponsibly.GuidingPrinciplesfortheFrameworkx/-/4KeyComponentsInternalgovernancestructuresandmeasures Multi-disciplinary,centralgoverningbody,suchasanAlEthicsAdvisoryBoard,tooverseeAlgovernanceefforts Developstandards,guidelines,tools,andtemplatestohelporganisationsdesign,develop,anddeployAlresponsibly Clearlylayouttherolesandresponsibilitiesofpersonnelinvolvedintheresponsibledesign,developmentand/ordeploymentofAlDeterminingthelevelofhumaninvolvementinAl-augmenteddecision-making/Conductrelevantriskimpactassessmentstodeterminelevelofrisk Threebroadcategoriesofhumaninvolvementbasedonlevelofrisk一human-in-the-loop,human-over-the-loop,human-out-of-the-loopMitigatingriskshelpsbuildtrusttowardstheacceptanceandgreateruseofAltechnologiesintheregionOperationsmanagement TheAlSystem1.ifecycleconsistsofvariousstagesandisoftenaniterativeprocess Conductrisk-basedassessmentsbeforestartinganydatacollectionandprocessingormodelling Mitigaterisksofunjustbiasduetoinsufficientlyrepresentativetraining,testingandvalidationdatasetsStakeholderinteractionandcommunicationDeveloptrustwithstakeholdersthroughoutthedesign,development,anddeploymentofAl ProvidegeneraldisclosureofwhenAlisusedinproductsand/orserviceofferings PutinplacemeasurestohelpemployeesadapttoanAl-augmentedworkenvironmentNational-levelRecommendations.NurturingAltalentandupskillingworkforceWorkcloselywithpublicandprivatesectorstoensurethatacountry'sworkforcecanadapttothenewwaysofworkingandpossessesenoughdigitalskillstointeracteffectivelywithAlsystems.SupportingAlinnovationecosystemandpromotinginvestmentinAlstart-upsWorkcloselywithpublicandprivatesectorstocreateasupportiveenvironmentforAldevelopment,wherecompaniesareabletoaccessandleveragedata,digitaltechnologies,andinfrastructure.InvestinginAlresearchanddevelopmentKeepabreastofthelatestdevelopmentsinAlandencourageresearchrelatedtothecybersecurityofAl,Algovernance,andAlethicstoensurethatthesafetyandresiliencyofAlsystemsandtoolsalsoadvanceinparallelwithnewusecases.PromotingadoptionofusefultoolsbybusinessestoimplementtheASEANGuideonAlGovernanceandEthicsDeploytoolstoenabletheimplementationofAlgovernanceinoperationsandensurethatdocumentationandvalidationprocessesaremoreefficient.RaisingawarenessamongcitizensontheeffectsofAlinsocietyRaiseawarenessofthepotentialrisksandbenefitsofAlsocitizenscanmakeinformeddecisionsabouttheappropriateuseofAlandtakeappropriateactionstoprotectthemselvesfromharmfulusesofAlsystems.Regional-levelrecommendations.SettingupanASEANWorkingGrouponAlGovernancetodriveandoverseeAlgovernanceinitiativesintheregionTheWorkingGroupcanconsistofrepresentativesfromeachoftheASEANmemberstateswhocanworktogethertorollouttherecommendationslaidoutinthisGuide,aswellasprovideguidanceforASEANcountrieswhowishtoadoptcomponentsofthisGuide,andwhereappropriate,includeconsultationwithotherindustrypartnersfortheirviewsandinput.AdaptationofthisGuidetoaddressgovernanceofgenerativeAlRisksinclude: Mistakesandanthropomorphism Factuallyinaccurateresponsesanddisinformation Deepfakes,impersonation,fraudulentandmaliciousactivities Infringementofintellectualpropertyrights Privacyandconfidentiality PropagationofembeddedbiasesGovernanceshouldinclude: Adaptationofexistingframeworksandtools Guidanceondevelopingasharedresponsibilityframework GuidanceonincreasingthecapacitytomanagerisksofgenerativeAl GuidanceonhowtodistinguishAl-generatedcontentversusauthenticallygeneratedonesCompilingacompendiumofusecasesdemonstratingpracticalimplementationoftheGuidebyorganisationsoperatinginASEANAcompendiumofusecasesshowcasesthecommitmentoftheseorganisationstoAlgovernanceandhelpsthempromotethemselvesasresponsibleAlpractitioners.UseCasesIllustrationofcomponentsoftheASEANGuideonAlGovernanceandEthicsthroughusecasesoforganisationsoperatinginASEANthathaveimplementedAlgovernancemeasuresinAldesign,development,anddeployment.GojekAboitizGroupSmartNationGroup(SNG)1SingaporeUCARE.AIEYMinistryofEducation,SingaporeA.IntroductionArtificialIntelligence(AI)isthedisciplineofmakinganalyticalmachinesintelligent,enablinganorganisationtofunctionappropriatelyandwithforesight.Unlikeothertechnologies,someformsofAladaptonitsown,learningthroughuse,sothedecisionsitmakestodaymaybedifferentfromthoseitmakestomorrow.Alandautomationhavebeenhottopics,bothfortheirtransformativepotentialandfortheircapacitytointroducenewopportunitiesbydisruptingoldmodels.SoutheastAsiaisnoexception.Alsystemsshouldbetreateddifferentlyfromothersoftwaresystemsbecauseofitsuniquecharacteristicsandrisks.CapabilitiesofAlsystemsfuelledbytechniquesevolutionandbreakthroughsarequicklyoutpacingthemonitoringandvalidationtools.ThedevelopmentofAlisalsodecentralisedduetolowbarrierstoentryandtheproliferationofopen-sourcetechnologies.GiventheprofoundimpactthatAlpotentiallybringstoorganisationsandindividualsinASEAN1itisimportantthatthedecisionsmadebyAlarealignedwithnationalandcorporatevalues,aswellasbroaderethicalandsocialnorms.Also,theASEANDigitalMasterplan2025(ADM2025)developedbyASEANMemberStates(“AMS”)envisionsASEANasaleadingdigitalcommunityandeconomicbloc,poweredbysecureandtransformativedigitalservices,technologies,andecosystem.Inthatcontext,theADM2025hasidentifiedEnablingAction(EA)2.7thatsuggeststhedevelopmentandadoptionofaregionalpolicytodeliverbestpracticeguidanceonAlgovernanceandethics.Inrecentyears,governmentsandinternationalorganisationshavebegunissuingprinciples,frameworksandrecommendationsonAlethicsandgovernance.ExamplesincludeSingapore'sModelAlGovernanceFrameworkInfocommMediaDevelopmentAuthority,"ModelArtificialIntelligenceGovernanceFrameworkSecondEdition"(21January2020)Vhttps:"WVvW.Ddpc.gv.sq-mediafilesPdPCpdf-filesresouce-foroqanisatio11aiSqmOdelaiqoVframeWOr2.pdf>2OrganisationforEnomicCo-operationandDevelopment,'RecommendationoftheCouncilonArtificialIntelligence"(22May2019)andOECD,sRecommendationoftheCouncilonAlVhttps:/IeqalinStrUmentSQeCdQrQeninstrumentsOECD-1.EGA1.-0449>.However,therehasnotyetbeenanintergovernmentalcommonstandardforAlthatdefinestheprinciplesofAlgovernanceandprovidesguidanceforpolicymakersintheregiontoutiliseAlsystemsinaresponsibleandethicalmanner.IntheprocessofdraftingthisGuide,existingAlgovernanceframeworksandguidelinessuchasUNESCO'sRecommendationontheEthicsofArtificialIntelligenceUnitedNationsEducational,ScientificandCulturalOrganisation,"RecommendationontheEthicsofArtificialIntelligence"(23November2021)Vhttos:UneSdoC.unesco.orqaric48223Df00003811374EropeanCommission"EthicsGuidelinesforTrustworthyAl"<https:diitaCstratess.ec.euroca.euenIibrarVethics-uidelines-trustworthv-ai>andEU,sEthicsGuidelinesforTrustworthyAlwerereferenced(8April2019).TheASEANGuideonAlGovernanceandEthicsaimstoempowerorganisationsandgovernmentsintheregiontodesign,develop,anddeploytraditionalAlsystemsresponsiblyandincreaseusers,trustinAl.1. ObjectivesThisdocumentservesasapracticalguidefororganisationsintheregionthatwishtodesign,develop,anddeploytraditionalAltechnologiesincommercialandnon-militaryordual-useapplications.ThisGuidefocusesonencouragingalignmentwithinASEANandfosteringtheinteroperabilityofAlframeworksacrossjurisdictions.Italsoincludesrecommendationsonnational-levelandregional-levelinitiativesthatgovernmentsintheregioncanconsiderimplementingtodesign,develop,anddeployAlsystemsresponsibly.TheASEANGuideonAlGovernanceandEthicsencompassesabroadrangeofconsiderationsthatneedstobetailoredtotheorganisationsdesigning,developing,anddeployingAldependingonthenatureoftheindustry,complexityofthetechnology,andassociatedrisksoftheAlsystems.1.ocalgovernmentauthoritiesareencouragedtorefertothisGuidewhenformulatingtheirpoliciesandapproaches.Whileadoptionoftheframeworklaidouthereinisvoluntary,thisGuidecanhelporganisationsbuildtrustamongstakeholdersandthepublicaswellasaligntheirAlpracticeswithinternationalstandardsandbestpractices,amongothers.OrganisationsareencouragedtorefertotheguidelinesinthisdocumenttounderstandhowtoassesstherisksassociatedwithAlandtakemeasurestodesign,develop,anddeployAlresponsiblyinthecontextoftheirorganisations.ThisGuideismeanttobealivingdocumentthatshouldbeperiodicallyreviewedandassessedbyrelevantASEANsectorialbodies,inconsultationwithindustrypartners,toensurethatitisuptodatewiththelatestregulationsandadvancementsintheAlspace.UpdatestothisGuidemaybepublishedsubsequentlytokeepupwithevolutionsandgrowthingovernanceandstandards.2. AssumptionsAlsystemsneedtobemanagedholistically,includingitsecosystemandallcomponents-humanoperator,InternetofThings(IoT)1robotics,traditionaltechnology,vendors,etc.InadditiontotherecommendationssetoutinthisGuidepertainingtoAlgovernance,organisationsarealsoencouragedtofollowandrefertointernationalstandardsandbestpracticesinrelatedfieldslikeinformationsecuritymanagementsystems(ISMS),datamanagementandgovernance,softwaredevelopmentandtesting,cybersecurity,IoT,etc.Developersanddeployersneedtoadheretoapplicablenationallawsandregulations,includingsectorspecificlawsandconstitutionswhendesigning,developing,anddeployingAltechnologies.TheASEANGuideonAlGovernanceandEthicsdoesnotreplaceorsupersedeanyexistingorupcominglawsandonlyservesasaguideforresponsibledesign,development,anddeploymentofAlintheregion.BeforedeployingAl,itisalsoimportantfordevelopersanddeployerstoconsidertherelevantlegalandregulatoryrequirementsoftherespectivecountrieswheretheAlsystemswillbedeployed,aswellastheusecontextforlegal,policy,andregulatoryconcerns.Giventhefast-evolvingspaceofAl,developersanddeployersofAlsystemsshouldbemindfulofthelateststateoftheartvis-visgovernancetoolsandtechnologiesandconducttherelevantassessmentstoevaluatethefeasibilityandusefulnessofthesetoolsintheimplementationofAlgovernancepracticesinthedesign,development,anddeploymentofAlsystems.3. TargetAudienceTheASEANGuideonAlGovernanceandEthicsisanASEAN-endorsedframeworkfororganisationsintheregiontorefertowhendesigning,developing,anddeployingtraditionalAltechnologies.Itprovidesguidelinesandrecommendationsforadiverserangeofindividualsandorganisationsalongtheentirevaluechain.TheseincludeAldevelopersanddeployers,academicprofessionals,andeveryonethatisinterestedinutilisingorscalingupAlsystems.TheGuidealsoincludessectionsonnational-levelandregional-levelrecommendationsthataremoretargetedtowardspolicymakersinASEAN.4. DefinitionsThedefinitionsofseveralkeytermsusedinthisGuidearesetoutbelow.ASEAN:referstotheAssociationofSoutheastAsianNations.Itisapoliticalandeconomicunionof10memberstatesinSoutheastAsia,whichpromotesintergovernmentalcooperationandfacilitateseconomic,political,security,military,educational,andsocioculturalintegrationbetweenitsmembersandcountriesintheAsia-Pacific.ArtificialIntelligence(Al):isanengineeredormachine-basedsystemthatcan,foragivensetofobjectives,generateoutputssuchaspredictions,recommendations,ordecisionsinfluencingrealorvirtualenvironments.Alsystem:isamachine-basedsystemthatiscapableofinfluencingtheenvironmentbyproducinganoutput(predictions,recommendationsordecisions)foragivensetofobjectives.Itusesmachineand/orhuman-baseddataandinputsto(i)perceiverealand/orvirtualenvironments;(ii)abstracttheseperceptionsintomodelsthroughanalysisinanautomatedmanner(e.g.,withmachinelearning),ormanually;and(iii)usemodelinferencetoformulateoptionsforoutcomes.Alsystemsaredesignedtooperatewithvaryinglevelsofautonomy.Deep1.earning:isasubsetofmachinelearning.Itisverylooselybasedontheinformation-processingarchitectureofthebrain-albeitfarfrommatchingitsability-enablingsystemstofindwaysofrerepresentinginputdatathatfacilitatemakinghighlyaccuratepredictions.Deployer:isanentitythatusesorimplementsanAlsystem,whichcouldeitherbedevelopedbytheirinhouseteamorviaathird-partydeveloper.Developer:isanentitythatdesigns,codes,orproducesanAlsystem.Machine1.earning:isasubfieldinAlwherealgorithmslearnbyidentifyingpatternsandcorrelationswithindatausingstatisticaltechniquestoenhanceperformance,allwithoutbeingexplicitlyprogrammed.User:isanentityorperson(internalorexternal)thatinteractswithanAlsystemoranAl-enabledserviceandcanbeaffectedbyitsdecisions.GuidingPrinciplesfortheFrameworkB.GuidingPrinciplesfortheFrameworkThesevenguidingprinciplesbelowhelptoensuretrustinAlandthedesign,development,anddeploymentofethicalAlsystems.TheyalsoprovideguidanceonhowAlsystemsshouldbedesigned,developed,anddeployedinwayswhichconsiderthebroadersocietalimpact.1. TransparencyandExplainabilityTransparencyreferstoprovidingdisclosureonwhenanAlsystemisbeingusedandtheinvolvementofanAlsystemindecision-making,whatkindofdataituses,anditspurpose.BydisclosingtoindividualsthatAlisusedinthesystem,individualswillbecomeawareandcanmakeaninformedchoiceofwhethertousetheAl-enabledsystem.ExplainabilityistheabilitytocommunicatethereasoningbehindanAlsystem'sdecisioninawaythatisunderstandabletoarangeofpeople,asitisnotalwaysclearhowanAlsystemhasarrivedataconclusion.ThisallowsindividualstoknowthefactorscontributingtotheAlsystem'srecommendation.InordertobuildpublictrustinAl,itisimportanttoensurethatusersareawareoftheuseofAltechnologyandunderstandhowinformationfromtheirinteractionisusedandhowtheAlsystemmakesitsdecisionsusingtheinformationprovided.Inlinewiththeprincipleoftransparency,deployershavearesponsibilitytoclearlydisclosetheimplementationofanAlsystemtostakeholdersandfostergeneralawarenessoftheAlsystembeingused.WiththeincreasinguseofAlinmanybusinessesandindustries,thepublicisbecomingmo