Gartner+生成式人工智能（如ChatGPT）对政府的影响-40页-WN6.pdf

《Gartner+生成式人工智能（如ChatGPT）对政府的影响-40页-WN6.pdf》由会员分享，可在线阅读，更多相关《Gartner+生成式人工智能（如ChatGPT）对政府的影响-40页-WN6.pdf（40页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、Gartner WebinarsGartner delivers actionable,objective insight,guidance and tools to enable stronger performance on your organizations mission critical priorities2 2023 Gartner,Inc.and/or its affiliates.All rights reserved.RESTRICTED DISTRIBUTIONEnhance your webinar experienceEnhance your webinar exp

2、erienceAsk a QuestionDownload AttachmentsWatch Again3 2023 Gartner,Inc.and/or its affiliates.All rights reserved.RESTRICTED DISTRIBUTIONConnect with GartnerBeyond the Hype:Impact on Beyond the Hype:Impact on Government of Generative AI,Such Government of Generative AI,Such as ChatGPTas ChatGPTBen Ka

3、nerSr Director Analyst4 2023 Gartner,Inc.and/or its affiliates.All rights reserved.This is all about ChatGPT5 2023 Gartner,Inc.and/or its affiliates.All rights reserved.This is all about ChatGPT6 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Generative AI small set of examplesImageDALL-

4、E 2CraiyonMaitu AILensa AIStable DiffusionRefaceMuseWomboVideoMake-a-VideoPoised AISynthesiaPictoryDeepbrainAISynthesysElai.ioSynths VideoVoiceAmazon PollyVALL-EMurfSynthesiaSynthesysListnrLovoRespeecherPlay.htSpeecheloSonanticResemble.aiTextLaMDA/BardGPT4ChatGPTGoogle translateAnthropicMetaAINVIDIA

5、DeepmindLLaMaEinstein7 2023 Gartner,Inc.and/or its affiliates.All rights reserved.What is a Generative AI Large Language Model?Neural network Many parameters Trained on large quantities of unlabeled text or other data Data generally from public datasets which introduces bias Self-or Semi-Supervised

6、Learning Emergent behaviors with unexpected results8 2023 Gartner,Inc.and/or its affiliates.All rights reserved.What does a Large Language Model Do?Has read a lot Is very articulate Can speak multiple languages Can draft rapidly Has no real-world experience Is somewhat nave May be able to produce in

7、teresting ideas Will learn from what they are given May or may not understand sensitivity and privacy9 2023 Gartner,Inc.and/or its affiliates.All rights reserved.And what might that remind you of?Has read a lot Is very articulate Can speak multiple languages Can draft rapidly Has no real-world exper

8、ience Is somewhat nave May be able to produce interesting ideas Will learn from what they are given May or may not understand sensitivity and privacy10 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Recommendation 1:Policy Use the hype around ChatGPT to highlight the need for policy Secu

9、rity and Privacy Internal Use External Facing Use Threat Assessment11 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Use Cases in Government12 2023 Gartner,Inc.and/or its affiliates.All rights reserved.What can it be used for?Generating TextAnalyzing TextManipulation and TranslationQuest

10、ion AnsweringManage Agency MissionReports,1stdraftSentiment Analysis reviews of media Resolving Administrative BacklogsCommunicationsPolicy questionsBack OfficeProcess Guides,ReportsRecruitment Processes,Complaint handlingReporting to multiple bodiesFOI requestsRamping on StaffITCode Generation,User

11、 GuidesDiscovery,functional analysis,code checkingRefactoring,legacy migration User SupportEngaging CitizensGuides,CommunicationSentiment Analysis,Empathic CommunicationTargeting marginalized constituentsHelp desks,Personalized Support and InformationOperationsGuides(internal and external),reporting

12、,summarizationTriage,prioritization,classification,redactionMultilingual support,outreachRamping on new staff,compliance,process adherenceUse cases and techniques are proliferating rapidly13 2023 Gartner,Inc.and/or its affiliates.All rights reserved.What can it be used for?Generating TextAnalysing T

13、extManipulation and TranslationQuestion AnsweringManage Agency MissionReports,1stdraftSentiment Analysis reviews of media Resolving Administrative BacklogsCommunicationsPolicy questionsBack OfficeProcess Guides,ReportsRecruitment Processes,Complaint handlingReporting to multiple bodiesFOI requestsRa

14、mping on StaffITCode Generation,User GuidesDiscovery,functional analysis,code checkingRefactoring,legacy migration User SupportEngaging CitizensGuides,CommunicationSentiment Analysis,Empathic CommunicationTargeting marginalized constituentsHelp desks,Personalized Support and InformationOperationsGui

15、des(internal and external),reporting,summarizationTriage,prioritization,classification,redactionMultilingual support,outreachRamping on new staff,compliance,process adherenceUse cases and techniques are proliferating rapidly14 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Specific Capab

16、ilities Human Services Protective ServicesIntegrated Eligibility Systems(IES)Rule-based determination of program eligibility for new applicants and enrollment of qualified applicants(aka clients,members,recipients,beneficiaries)into programs,or redirection(triage)to other resources.Eligibility redet

17、ermination of current clients based on changing circumstances.Enrollment of new service providers and revalidation of current service providers.Case ManagementIntake,Assessment,Service Plan Development,Referrals,Outcomes Tracking,CloseEligibility,Enrollment and Case ManagementProtective ServicesServ

18、ice Provider ManagementCare Delivery and Care CoordinationPayment SystemsEngage ConstituentsManage Agency Mission and StrategyData,Records and AnalyticsIntakeCase MonitoringPlacement MonitoringIntervention Support Augmented Risk-Based DecisionsReferral OrchestrationException and EscalationUpdate Cas

19、e InformationInbound Referral QueuesTriageEstablish Case InformationInvestigation SupportAssemble DataRisk Assess Case FactorsWorkflow ManagementEvidence ManagementPeriodic ReportingDynamic Case MonitoringEvent Audit LogsOutbound EventsAllocating PlacementTransition HandlingPeriodic Case UpdateFrom

20、A Business Capability Reference Model for Human Services G0078032515 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Integrated Eligibility Systems(IES)Rule-based determination of program eligibility for new applicants and enrollment of qualified applicants(aka clients,members,recipients,

21、beneficiaries)into programs,or redirection(triage)to other resources.Eligibility redetermination of current clients based on changing circumstances.Enrollment of new service providers and revalidation of current service providers.Case ManagementIntake,Assessment,Service Plan Development,Referrals,Ou

22、tcomes Tracking,CloseEligibility,Enrollment and Case ManagementProtective ServicesService Provider ManagementCare Delivery and Care CoordinationPayment SystemsEngage ConstituentsManage Agency Mission and StrategyData,Records and AnalyticsIntakeCase MonitoringPlacement MonitoringIntervention Support

23、Augmented Risk-Based DecisionsReferral OrchestrationException and EscalationUpdate Case InformationInbound Referral QueuesTriageEstablish Case InformationInvestigation SupportAssemble DataRisk Assess Case FactorsWorkflow ManagementEvidence ManagementPeriodic ReportingDynamic Case MonitoringEvent Aud

24、it LogsOutbound EventsAllocating PlacementTransition HandlingPeriodic Case UpdateFrom A Business Capability Reference Model for Human Services G00780325Specific Capabilities Human Services Protective Services16 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Specific Capabilities Law Enfo

25、rcement Prevent CrimeHelp Those in NeedKeep the PeacePrevent CrimeProtect Life and PropertyDetect and Apprehend OffendersMulti-ChannelCollabo-ration ToolsSocial MediaSocial Media ListeningCall CenterCampaign MgmtConstituent EngagementOperations SystemsData and ServicesAnalytics(Business Intelligence

26、,Geospatial)IoT PlatformSample Devices,Peripherals,AccessIoT(sensors,cameras,wearables)Mobile DevicesRadiosDesktop DevicesCommunications SystemsRMS9-1-1CADVoice RecordingInvestigation&intelligenceDEMSReal-time Incident/Crime CenterData/ManagementEnterprise Content ManagementDigital IDFrom Capabiliti

27、es Model for Law Enforcement G0077990017 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Specific Capabilities Law Enforcement Prevent CrimeHelp Those in NeedKeep the PeacePrevent CrimeProtect Life and PropertyDetect and Apprehend OffendersMulti-ChannelCollabo-ration ToolsSocial MediaSoci

28、al Media ListeningCall CenterCampaign MgmtConstituent EngagementOperations SystemsData and ServicesAnalytics(Business Intelligence,Geospatial)IoT PlatformSample Devices,Peripherals,AccessIoT(sensors,cameras,wearables)Mobile DevicesRadiosDesktop DevicesCommunications SystemsRMS9-1-1CADVoice Recording

29、Investigation&intelligenceDEMSReal-time Incident/Crime CenterData/ManagementEnterprise Content ManagementDigital IDFrom Capabilities Model for Law Enforcement G0077990018 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Prioritization19 2023 Gartner,Inc.and/or its affiliates.All rights res

30、erved.For proposed initiative:Score on:Business ValuePolicy/Societal OutcomeOperations EfficienciesRisk to ServiceFeasibilityTechnologyOrganizational(skills,governance,ability to adapt,operating model)External(data availability,regulation,acceptability)Business ValueFeasibilityLow20 2023 Gartner,Inc

31、.and/or its affiliates.All rights reserved.For each proposed initiative:Score on:Business ValuePolicy/Societal OutcomeOperations EfficienciesRisk to ServiceFeasibilityTechnologyOrganizational(skills,governance,ability to adapt,operating model)External(data availability,regulation,acceptability)Busin

32、ess ValueFeasibilityLowThe Tech is only 1/6 of the challenge!21 2023 Gartner,Inc.and/or its affiliates.All rights reserved.LowAvoid!?Caution!Selectivelyinvest!Business ValueFeasibilityWhere things will end upWhich are feasible?Which result in real return?Which look nice,but in reality.22 2023 Gartne

33、r,Inc.and/or its affiliates.All rights reserved.Recommendation 2:Prioritize InitiativesAppraise which business capabilities can benefit from which Generative AI capabilitiesAggregate valueValue to missionReduction in operational resourcesReduction in risk of failureEnsure feasibilityTechnical feasib

34、ilityOrganizational FeasibilitySocietal and Regulatory Acceptability 23 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Recommendation 2:Prioritize InitiativesAppraise which business capabilities can benefit from which Generative AI capabilitiesAggregate valueValue to missionReduction in

35、operational resourcesReduction in risk of failureEnsure feasibilityTechnical feasibilityOrganizational FeasibilitySocietal and Regulatrory Acceptability The most common cause of failure in major changeLLM Technology is immature and changing rapidly AI is particularly sensitive to negative news24 202

36、3 Gartner,Inc.and/or its affiliates.All rights reserved.Challenges and Risks25 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Considerations Large Language Models learn from existing dataThey will incorporate biases from their source dataThe data is not live and may be out-of-dateThey ha

37、ve no conceptual model of the desired outcome The data provided may be seriously misleading Much of the material used may be copyrighted The products of Generative AI cannot be copyrighted under US guidance The implications on liability have not yet been legally tested in any jurisdiction Unlike sea

38、rch results,A chat response usually gives ONE optionIf that response is wrong or dangerous,who is liable?26 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Recommendation 3:Mitigate RiskEnsure that all staff understand the limits Generative AI is not an infallible oracleProvide a safe spa

39、ce for experimentationPeople are already experimenting!Harvest good ideasHigher value/risk combinations should deliver in stages:Start with Human in the Loop the human is accountableScale to automation only when the residual risk is low enough to be accepted by execs and legal Continue to monitor de

40、velopments27 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Use Outside Government 28 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Government are Not the Only UsersExternal users will not go direct to government sitesBenign UsesPersonalised guidanceThrough government proces

41、sesTourism itinerariesResponses to government campaignsBetter communication by marginalized/low education communitiesChallengesInformation may be out of dateInformation may be inaccurateWork with external service providers to improve their ability to serve their customer setAfter all it offloads wor

42、k from government services!29 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Government are Not the Only UsersExternal users will not go direct to government sitesMalign UsesAdministrative FloodingComplaints,FOIDeepFake AttacksSocial EngineeringCorruption of process,exposure of informati

43、on,redirection of funds,provision of licences/permitsImpersonation of public figuresPoisoning the WellDeliberate misinformation through the external modelsCode GenerationViral attacks generated at scale Monitor developments and build plans30 2023 Gartner,Inc.and/or its affiliates.All rights reserved

44、.Government are Not the Only UsersSomething to bear in mind:Public ChatGPT,Bing,Bard can be used by anybody with a mobileIt is very useful for generating drafts or summarizing textGovernment employees will be using it,and will bypass your controls unless you give them a safe space in which to do so

45、31 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Recommendation 4:Keep abreast of threats Involve external parties in discovery Engage with Generative AI service providers to learn how to optimize/control responses Build mitigation plans for likely attack vectors Deepfakes Poisoning the

46、 well Administrative Flooding.?32 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Recommendations33 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Recommendations Many executives,politicians and media are excited ChatGPT is only one large language model There are many other us

47、es of AI This is an opportunity Value to PurposeMissionCostsRisk Reduction FeasibilityTechnicalOrganizationalAND Societal Provide a safe space for experimentation Deliver in stages Start with human-in-the-loop Scale when residual risk is shown to be acceptable Involve external parties in discovery E

48、ngage with providers to control responses Build mitigation plansExploit the hype Exploit the hype to drive policyto drive policyPrioritize Prioritize InitiativesInitiativesMitigate RiskMitigate RiskKeep Abreast of Keep Abreast of ThreatsThreats1 12 23 34 434 2023 Gartner,Inc.and/or its affiliates.Al

49、l rights reserved.RESTRICTED DISTRIBUTIONAsk the speaker Ask the speaker 2023 Gartner,Inc.and/or its affiliates.All rights reserved.Discover tools and techniques to enhance your IT and business strategiesLearn more: this years conference,youll learn:Examine the opportunities and risks in adopting em

50、erging and innovative technologiesChallenge how you think about leadership and discover new approaches to lead11 13 September 2023|Gold Coast,Australia16 19 October 2023|Orlando,FL6 9 November 2023|Barcelona,Spain13 15 November 2023|Tokyo,Japan28 30 November 2023|Kochi,IndiaThe Worlds Most Important