生成式人工智能治理模型框架（英）.pdf

ExecutiveSummary3Accountability6Data9TrustedDevelopment12andDeploymentIncidentReporting16TestingandAssurance19Security21ContentProvenance23SafetyandAlignmentR&D26AIforPublicGood28Conclusion31Acknowledgements32FurtherDevelopment34CONTENTS3MODELAIGOVERNANCEFRAMEWORKFORGENERATIVEAIGenerativeAIhascapturedtheworld’simagination.Whileitholdssignificanttransformativepotential,italsocomeswithrisks.Buildingatrustedecosystemisthereforecritical—ithelpspeopleembraceAIwithconfidence,givesmaximalspaceforinnovation,andservesasacorefoundationtoharnessingAIforthePublicGood.AI,asawhole,isatechnologythathasbeendevelopingovertheyears.PriordevelopmentanddeploymentissometimestermedtraditionalAI.1TolaythegroundworktopromotetheresponsibleuseoftraditionalAI,SingaporereleasedthefirstversionoftheModelAIGovernanceFrameworkin2019,andupdateditsubsequentlyin2020.2TherecentadventofgenerativeAI3hasreinforcedsomeofthesameAIrisks(e.g.,bias,misuse,lackofexplainability),andintroducednewones(e.g.,hallucination,copyrightinfringement,valuealignment).TheseconcernswerehighlightedinourearlierDiscussionPaperonGenerativeAI:ImplicationsforTrustandGovernance,4issuedinJune2023.Thediscussionsandfeedbackhavebeeninstructive.Existinggovernanceframeworksneedtobereviewedtofosterabroadertrustedecosystem.Acarefulbalanceneedstobestruckbetweenprotectingusersanddrivinginnovation.Therehavealsobeenvariousinternationaldiscussionspullingintherelatedandpertinenttopicsofaccountability,copyrightandmisinformation,amongothers.Theseissuesareinterconnectedandneedtobeviewedinapracticalandholisticmanner.Nosingleinterventionwillbeasilverbullet.ThisModelAIGovernanceFrameworkforGenerativeAIthereforeseekstosetforthasystematicandbalancedapproachtoaddressgenerativeAIconcernswhilecontinuingtofacilitateinnovation.Itrequiresallkeystakeholders,includingpolicymakers,industry,theresearchcommunityandthebroaderpublic,tocollectivelydotheirpart.ThereareninedimensionswhichtheFrameworkproposestobelookedatintotality,tofosteratrustedecosystem.a)Accountability—AccountabilityisakeyconsiderationtoincentiviseplayersalongtheAIdevelopmentchaintoberesponsibletoend-users.Indoingso,werecognisethatgenerativeAI,likemostsoftwaredevelopment,involvesmultiplelayersinthetechstack,andhencetheallocationofresponsibilitymaynotbeimmediatelyclear.WhilegenerativeAIdevelopmenthasuniquecharacteristics,usefulparallelscanstillbedrawnwithtoday’scloudandsoftwaredevelopmentstacks,andinitialpracticalstepscanbetaken.EXECUTIVESUMMARY1TraditionalAIreferstoAImodelsthatmakepredictionsbyleveraginginsightsderivedfromhistoricaldata.TypicaltraditionalAImodelsincludelogisticregression,decisiontreesandconditionalrandomfields.Othertermsusedtodescribethisinclude“discriminativeAI”.2ThefocusoftheModelAIGovernanceFrameworkistosetoutbestpracticesforthedevelopmentanddeploymentoftraditionalAIsolutions.ThishasbeenincorporatedintoandexpandedundertheTrustedDevelopmentandDeploymentdimensionoftheModelAIGovernanceFrameworkforGenerativeAI.3GenerativeAIareAImodelscapableofgener