3152 TechXchange_responsible AI for regulated govt use caes and IBM Z.pdf

《3152 TechXchange_responsible AI for regulated govt use caes and IBM Z.pdf》由會員分享，可在線閱讀，更多相關《3152 TechXchange_responsible AI for regulated govt use caes and IBM Z.pdf（30頁珍藏版）》請在三個皮匠報告上搜索。

1、October 21-24,2024Mandalay Bay Convention CenterLas Vegas,NevadaHow we apply responsible AI for regulated government use cases with Data&AI on IBM ZSession 3152ElpidaElpida TzortzatosTzortzatosIBM Fellow and CTO,AI on IBM Z&LinuxONEJothi Jothi PeriasamyPeriasamyPrincipal Data Scientist&Product Manag

2、er,State of CaliforniaPaul VincentPaul VincentITS III Architect,State of California#IBMTechXchangeNotices and disclaimersCertain comments made in this presentation may be characterized as forward looking under the Private Securities Litigation Reform Act of 1995.Forward-looking statements are based

3、on the companys current assumptions regarding future business and financial performance.Those statements by their nature address matters that are uncertain to different degrees and involve a number of factors that could cause actual results to differ materially.Additional information concerning thes

4、e factors is contained in the Companys filings with the SEC.Copies are available from the SEC,from the IBM website,or from IBM Investor Relations.Any forward-looking statement made during this presentation speaks only as of the date on which it is made.The company assumes no obligation to update or

5、revise any forward-looking statements except as required by law;these charts and the associated remarks and comments are integrally related and are intended to be presented and understood together.2024 International Business Machines Corporation.All rights reserved.This document is distributed“as is

6、”without any warranty,either express or implied.In no event shall IBM be liable for any damage arising from the use of this information,including but not limited to,loss of data,business interruption,loss of profit or loss of opportunity.Customer examples are presented as illustrations of how those

7、customers have used IBM products and the results they may have achieved.Actual performance,cost,savings or other results in other operating environments may vary.Workshops,sessions and associated materials may have been prepared by independent session speakers,and do not necessarily reflect the view

8、s of IBM.Not all offerings are available in every country in which IBM operates.Any statements regarding IBMs future direction,intent or product plans are subject to change or withdrawal without notice.IBM,the IBM logo,and are trademarks of International Business Machines Corporation,registered in m

9、any jurisdictions worldwide.Other product and service names might be trademarks of IBM or other companies.A current list of IBM trademarks is available on the Web at“Copyright and trademark information”at: Z IBM TechXchange/October 20242Businesses everywhere are navigating an increasingly complex gl

10、obal regulatory landscape with respect to AI Less than 60%of executives believe their organization is prepared for AI regulation1 72%of executives are choosing to forgo generative AI over concerns about AI ethics and safety11IBM Institute for Business Value,The CEOs Guide to Generative AI:Responsibl

11、e AIðics,2023IBM Z IBM TechXchange/October 2024Privacy,data,and AIregulations andenforcement activitiesare increasingPrivacy laws and regulations15 US states have passed comprehensive privacy bills,three of them will become effective in 2024.Indias Digital Personal Data Protection Act will become

12、 effective in 2024,including fines for non-compliance up to$30M.US Executive Order limits bulk transfer of specified types of sensitive data to certain countries.AI laws and regulationsThe EU AI Act was passed in March 2024 and includes fines of up to 7%of a companys annual revenues 7%of a companys

13、annual revenues for noncompliance.Canada,Brazil and Korea progress toward adoption of AI laws.A number of US states have already adopted AI laws,and AI laws are progressing in several others.US Executive Order directs new standards for AI safety and security.PrivacyAIData GovernanceMacro-environment

14、ChinaChinaPrivacy&Data Security Laws;AI regs15 15 U.S.U.S.StateStatePrivacy LawsExecutiveExecutiveOrder Order PI AccessRestrictions4IBM Z IBM TechXchange/October 2024AI systems considered as High-Risk systems according to EU AI Act-examplesHealthcare&medical devices:Diagnostics AI-powered medical de

15、vices;AI in Emergency ServicesFinance&Insurance:Credit scoring Risk assessment and pricing in health&life insuranceEducation:AI systems used for grading in educational and vocational settings Energy&Utilities:AI systems for critical digital infrastructure managementTransport,automotive,air transport

16、ation:AI powered assistance and warning systems AI in Aircraft systems AI in Air Traffic Control Public sector:Biometric Identification Systems(law enforcement)AI systems for Emergency Services(classifying calls,dispatch prioritizing of police,firefighters,medical aid)Profiling during criminal detec

17、tions,investigations IBM Z IBM TechXchange/October 2024Principles for Trust and Transparency6The purpose of AI is to augment not replace human intelligenceData and insights belong to their creatorNew technology,including AI systems,must be transparent and explainable123ExplainabilityAn AI systems ab

18、ility to provide a human-interpretable explanation for its predictions and insightsFairnessEquitable treatment of individuals or groups by an AI system depends on the context in which the AI system is usedRobustnessAn AI systems ability to effectively handle exceptional conditions,such as abnormalit

19、ies in inputTransparencyAn AI systems ability to include and share information on how it has been designed and developedPrivacyAn AI systems ability to prioritize and safeguard consumers privacy and data rightsPillars of TrustIBM Z IBM TechXchange/October 2024IBMs Approach:Trustworthy AI ScaleA holi

20、stic and staged approach to establish scalable,sustainable,organizational AI Governance and AI Model Lifecyle GovernanceORGANIZATIONAL AI GOVERNANCEAUTOMATED AI MODEL LIFECYCLE GOVERNANCE STRATEGYPLANNINGWho?Business,AI Ethics Board,Data/AI leaders,Internal Policy&Regulations,CPO,CISOWho?Business,AI

21、 Ethics Board,Data/AI leaders,Internal Policy&Regulations,CPO,EcosystemDEVELOPMENT&DEPLOYMENTOPERATIONSWho?Dev teams,IT leaders,CDAO,Software&Data science leadersWho?IT leaders,MLOps teamsMONITORING&PORTFOLIO MANAGEMENT Business Outcomes&Model GovernanceBusiness leaders,MLOps teamsTwo critical compo

22、nents:Trustworthy AI PrinciplesAI Governance Policies,Processes&MetricsOperating ModelGovernance StructuresRegulatory and Risk AssessmentCentralised Platform for Model Lifecycle MonitoringTrustworthy AI Model LifecycleDefined Model OKRs and KPIsModel Onboarding,and Sustaining,ProcessesIBM Z IBM Tech

23、Xchange/October 2024Trustworthy AI LifecycleTrustworthy AI Lifecycle8AI DocumentationCapture facts about use cases,models and prompts throughout the lifecycleAI Evaluation&MonitoringModel Health|AccuracyDrift|Bias|ExplainabilityGenerative AI QualityCapture AI performance meta-dataDeploy approved AI

24、assetAI Risk GovernanceUse case&AI asset inventoryWorkflows|Risk assessmentsDashboards|Issue managementSync AI asset status and metadataDesign-time evaluation and explainabilityRun-time monitoring for assessment and business outcomesCapture development meta-dataCapture deployment meta-dataBuild(IBM,

25、AWS,MS,Google,Other)Deploy(IBM Z)Model OwnersModel ValidatorsAudit TeamsCompliance TeamsRisk Management TeamsData Privacy TeamsPrincipal Data ScientistsData Engineers(Citizen)Data ScientistsAI EngineersPrompt EngineersMLOpsMLOpsML EngineerIBM Z IBM TechXchange/October 2024Capture development meta-da

26、ta9Capture AI performance meta-dataDeploy approved AI assetSync AI asset status and metadataDesign-time evaluation and explainabilityRun-time monitoring for assessment and business outcomesCapture deployment meta-dataModel OwnersModel ValidatorsAudit TeamsCompliance TeamsRisk Management TeamsData Pr

27、ivacy TeamsPrincipal Data ScientistsData Engineers(Citizen)Data ScientistsAI EngineersPrompt EngineersMLOpsMLOpsML EngineerConceptual mapping of potentially relevant provisions of the EU AI ActBuild(IBM,AWS,MS,Google,Other)Deploy(IBM,AWS,MS,Google,Other)AI DocumentationCapture facts about use cases,

28、models and prompts throughout the lifecycleAI Evaluation&MonitoringModel Health|AccuracyDrift|Bias|ExplainabilityGenerative AI QualityAI Risk GovernanceUse case&AI asset inventoryWorkflows|Risk assessmentsDashboards|Issue managementEU AI Act:Article 10 Data and data governance Article 15 Accuracy,ro

29、bustnessEU AI Act:Article 15 Accuracy,robustness Article 61 Post-market monitoring EU AI Act:Article 10 Data and data governance Article 12/20 Record keepingEU AI Act:Article 11 Technical documentation Article 13 Transparency and information to users Article 18 Documentation keeping EU AI Act:Articl

30、e 5 Prohibited AI practices Article 6/7 High-risk AI systems Article 9 Risk management system Article 13 Transparency and information to users Article 17 Quality management system Article 21 Corrective actions and duty of information Article 52 Transparency obligations Article numbers subject to cha

31、nge pending release of final EU AI Act text.The client is responsible for ensuring compliance with all applicable laws and regulations.IBM does not provide legal advice nor represent or warrant that its services or products will ensure that the client is compliant with any law or regulation.IBM Z IB

32、M TechXchange/October 2024State of California State of California Department of Motor Vehicles Disabled Parking Placard Fraud DetectionIntroductionIntroductionDMV identified Disabled Parking Placard(DPP)applicants submitting fraudulent data to obtain the DPP status and internal DMV employees approve

33、d ineligible DPP applicationsTo ensure DPP compliance standards,DMV wanted to leverage machine learning technology to prevent these types of fraudulent activities before they occur.1.For instance,the physician who certifies the DPP application is a specialized Obstetrician-gynecologist but the appli

34、cants underlying disease is diabetes.2.The applicant lives in Sacramento,CA but the certified physician is in Boston,MA3.The physician is not a board-certified physician4.The applicants last name and the physicians last name are the same and their addresses are also the same.5.An employee at the DMV

35、 issues DPPs for ineligible family members and bulk approvals for an organization(transportation company)with some understanding outside of the DMV business.#Business ScenarioBusiness ScenarioApplication Intake SourceApplication Intake SourceApplication Customer ScopeApplication Customer ScopeApplic

36、ation ScopeApplication Scope1Online(Web)Field OfficePostal mailAAATrade ShowIndividualsOrganizationsNew ApplicationRenewal11IBM Z IBM TechXchange/October 2024 Business Business ScenariosScenariosIdentify fraud from employee employee level behaviorIdentify fraud from transaction/applicanttransaction/

37、applicant behaviorIdentify fraud from medical providermedical provider behavior12IBM Z IBM TechXchange/October 2024Discover fraud among new applications for permanent disabled parking placardsSolution Architecture on IBM ZSolution Architecture on IBM ZModel DevelopmentModel DevelopmentModel Deployme

38、nt&ConsumptionModel Deployment&Consumption13IBM Z IBM TechXchange/October 2024What data is being used?What data is being used?Data fields retrieved from:DB2E_MVPDB2V_VVRM001V_AMIS_MAST,MAST_SM,and MAST_ST through DVM for z/OSFiltered on:TYPE_LICENSE=N1,EXPIRATION_DATE=2025-06-30,TYP_TRAN_CODE=E10 to

39、 get transactional data from active permanent Disabled Parking Placard issuancesactive permanent Disabled Parking Placard issuances.1.3 million records retrieved in 10 minutes into analytics environment on IBM Z1.3 million records retrieved in 10 minutes into analytics environment on IBM ZData Field

40、sData FieldsRO_RESIDENCE_CITYRO_ZIPRO_COUNTYEQUIPMENT_NUMBER(Medical Provider Number)DRIVER_LICENSE_OR_IDWORK_DTEOFFICE_IDTECH_IDEngineered Features Engineered Features(subset of features used in model)(subset of features used in model)cum_transactions_per_userrolling_7d_count_EQUIPMENT_NUMBERlag_7d

41、_days_EQUIPMENT_NUMBERmonthly_transaction_count_EQUIPMENT_NUMBERrolling_7d_count_TECH_ID_UNIQUElag_7d_days_TECH_ID_UNIQUEmonthly_transaction_count_TECH_ID_UNIQUEfft_day_office_magData fields were Data fields were encoded,and some encoded,and some features were features were engineered to perform eng

42、ineered to perform time series analysis.time series analysis.14IBM Z IBM TechXchange/October 2024Model PhaseModel PhaseData cleaning&feature engineeringIsolation forest model optimizationTest set performance&feature importanceAnalysis:TransactionsAnalysis:TransactionsTransaction patterns&trends by f

43、raud statusFraud distribution visualizationsAnalysis:TechniciansAnalysis:TechniciansFraud analysis for top techniciansFraud linkages:Technicians,applicants,providersAnalysis:Medical ProvidersAnalysis:Medical ProvidersFraud analysis for top techniciansProvider fraud trends&linkagesAnomaly/Fraud model

44、 Anomaly/Fraud model developmentdevelopmentand analysisand analysis15IBM Z IBM TechXchange/October 2024Machine Learning for Machine Learning for IBM z/OSIBM z/OS16IBM Z IBM TechXchange/October 2024Machine Learning for IBM z/OS-Enterprise Edition:The full-featured machine learning platform,training A

45、I models anywhere or on IBM Z and readily deploying those models on z/OS,co-located with enterprise applications,transaction data and business logic for in-transaction scoring in near real-time without impact to SLAs.Telum on-chip AI Accelerator and zIIP exploitation for AI inferencing at scaleScore

46、 transactions natively in CICS,IMS and Batch applications with near-zero latencyTrust and Explainability in every decisionAI-driven insights for accurate and real-time automationInfuse AI into products for a truly intelligent platform to propel your business forward Automate and scale legacy rules-b

47、ased systems with AI-driven insights and interpretabilityModular code for building ML solutionsGUI and code-driven approach for end-end model lifecycleSeamless upgrades to future versionsMachine Learning for IBM z/OSMachine Learning for IBM z/OSDeliver unprecedented inferencing performance for AI wo

48、rkloads on z/OSIBM Z IBM TechXchange/October 2024Why Machine Learning on IBM ZWhy Machine Learning on IBM ZRealReal-time insight at the point of transactiontime insight at the point of transaction In-transaction scoring(inferencing),supplemented by IBM Zs security,availability,and performance On-lin

49、e batch scoring,supplementedby IBM Z features and functionalityData gravity and integration with existingData gravity and integration with existingclient environmentsclient environments Data is kept in-place for higher throughputand lower latency Sensitive data remains encrypted and secure Completes

50、 an enterprise AI strategy by complementingcomplementing AI on distributed platforms“Consider Data as if it were a planet or other object with sufficient mass.As Data accumulates(builds mass)there is a greater likelihood that additional Services and Applications will be attracted to this data.This i

51、s the same effect Gravity has on objects around a planet.As the mass or density increases,so does the strength of gravitational pull.As things get closer to the mass,they accelerate toward the mass at an increasingly faster velocity.”-Dave McCrory18IBM Z IBM TechXchange/October 2024Deploy&ScoreHisto

52、rical DataDb2,VSAM,IMS,SMF,etc.Model Training JupyterHub IDEBusiness ApplicationsCICSIMSWASBatchModel ImportCloud and/or Cloud and/or x86 onx86 on-prem platformsprem platformszCXTrustworthy AI Monitors Train anywhere Train anywhere Deploy on IBM ZDeploy on IBM Zz/OSMachine Learning for IBM z/OSIBM Z

53、 IBM TechXchange/October 2024Explainable AI with Explainable AI with OpenScaleOpenScale andand AutoAIAutoAI20IBM Z IBM TechXchange/October 2024OpenScale Monitoring:OpenScale Monitoring:Recommended Workflow for Trustworthy AI ModelsRecommended Workflow for Trustworthy AI ModelsUnsupervised Anomaly De

54、tection ModelHistorical Transactional DataExpert Prediction ValidationKnown Fraudulent DPP HoldersValidated DPP Fraud DatasetSupervised Fraud Detection ModelAutoAIoAutomatically train AI models with AutoAI oExplain AI predictionsoMonitor for bias and drift Work with domain experts and investigators

55、to create curated fraud dataset1.Anomaly Detection2.Trusted Dataset Creation3.Trustworthy AI 21IBM Z IBM TechXchange/October 2024Identify suspicious data records which maybe indicative of fraudAutoAI Example Experiment for DPP FraudAutoAI Example Experiment for DPP FraudFrom provided dataset,AutoAI

56、automatically performs data preprocessing,hyperparameter optimization,feature engineering,and model evaluation to create an optimized AI model ready for deployment.22IBM Z IBM TechXchange/October 2024AutoAI Example Experiment for DPP FraudAutoAI Example Experiment for DPP FraudReview accuracy,precis

57、ion,and recall to assess models.23IBM Z IBM TechXchange/October 2024OpenScale Monitoring:Example Fraud Prediction Explanation for OpenScale Monitoring:Example Fraud Prediction Explanation for DPP Initial Issuance with Provider Number A69869DPP Initial Issuance with Provider Number A69869Explain mode

58、l predictionsShow the most influential featuresExplain in natural languageAvailable API for prediction explanations24IBM Z IBM TechXchange/October 2024OpenScale Monitoring:Example Fraud Prediction OpenScale Monitoring:Example Fraud Prediction Explanation for DPP Initial IssuanceExplanation for DPP I

59、nitial Issuance2006 days since this medical provider last signed a DPP application25IBM Z IBM TechXchange/October 2024OpenScale Monitoring:Example Fraud Investigation OpenScale Monitoring:Example Fraud Investigation for DPP Initial Issuance(Medical Provider)for DPP Initial Issuance(Medical Provider)

60、DPP holder 270146D with 2016-05-03 DPP issuance date has A10572 as the medical provider on their DPP application26IBM Z IBM TechXchange/October 2024OpenScale Monitoring:Example Fraud Investigation OpenScale Monitoring:Example Fraud Investigation for DPP Initial Issuancefor DPP Initial IssuanceLarge

61、distance between medical provider and DPP holderUnlikely medical provider who graduated in 1944 signed application with 2016-05-03 issuance date27IBM Z IBM TechXchange/October 2024Key Business BenefitsKey Business BenefitsAutomatically detect anomalous applicant behavior for 100%of initial DPP appli

62、cationsQuickly detect and uncover new fraud methods and patterns From AI explanation,investigators understand where to best direct their investigative efforts Reduce government revenue loss and ensure accessibility for disabled citizens 28IBM Z IBM TechXchange/October 2024Experience more here at IBM

63、 TechXchangeIBM Z IBM TechXchange/October 202429Come to our IBM Z and LinuxONE Sandbox#850 Come to our IBM Z and LinuxONE Sandbox#850 Experience the world famous plexi and lego!Engage our SMEs,demos,AMAs,community&skills A snapshot of some of the great topicsA snapshot of some of the great topicsAI

64、for mainframe app dev:#2341,3059,3766,1983New chip set,simplification&zNext:#3761,3063,3520Mainframe AI assistant for ops:#3767,3060,1950Threat detection&Cyber Vault:#2559,3773OTel,IntelliMagic&Instana:#3768,3762,1729 All things Data&AI:#3799,3213,1969,3319,1969DevOps,testing&more#1186,3212,3058Skills#3214,1889,1890IBM Z and LinuxONE content in TechXchange catalog30