负责任的人工智能与治理：如何应对不断变化的环境.pdf

1、JoAnn Stonier,Mastercard Fellow,Data&AI December 20204How to Navigate the Changing EnvironmentResponsible AI&GovernanceWSAIQ24,Main StageOur Dynamic WorldDECEMBER 5,2024Elements of Responsible AIResponsible AIGovernanceDataOrganization ReadinessSecurity LeadershipEngagementGovernance5DECEMBER 5,2024

2、 GovernanceDataDataRights,Privacy,IPQuality&FitnessBias,Fairness&OutcomesAccuracy/ValidityCompletenessConsistencyTimeliness/RecencyIntegrityUniquenessFitness for PurposeReliable/UsefulRepresents the population set related to the outcomeSource of DataRights ObtainedPrimary&Secondary Uses Ownership Ri

3、ghts and Limitations Personal information privacy compliance Sensitive data PETs Licenses Ownership of AI Generated contentWhat is the purpose of the project/product/service What is the data source how was it obtained what gaps does it have naturally missing geographic or demographic groups that are

4、 important to the outcomesWhat other bias human,storage,training,analytic bias has been introduced via the process and needs to be addressHow are the models observed observational AI?Are the outputs viewed in light of outcomes to be achieved?Training/Red teaming/reviews/audits Requires highly traine

5、d staffSecuritySecurityAttacks are exploiting vulnerabilities at every interaction stage of Gen AI:PromptsRAGTool OutputModel ResponseAttackers are motivated to access and steal proprietary data,generate malicious content/codeThe introduction of AI Agents adds another layer of complexity as these au

6、tonomous systems can interact with different environments and make independent decisions creating a new threat landscape Organizations will need to continue to invest in robust security measures to monitor their infrastructure as well as their partnership eco-systems Organization