AI model risk assessment Part 2: core content

The core of AI risk management framework provides the ability to have a dialogue 、 Understand and manage the results and actions of AI risk activities . The core consists of three elements ： function 、 Categories and subcategories . Pictured 5 Shown , Functional departments organize AI risk management activities at their highest level , To map 、 measurement 、 Manage and manage AI risks . In every function , There are categories and subcategories that subdivide functions into specific results and actions .

Map

Map Function to establish the context and apply AI RMF The attribute of taxonomy to design frame and AI System related risks . The information collected when performing this function provides information for decision-making about model management , Include preliminary decisions about whether an AI solution is appropriate or needed . Through income 、 Qualitative or more formal quantitative analysis of costs and risks , Compared with the current situation , Consider determining whether the use of AI is appropriate or reasonable . The companion document describes the practices associated with mapping AI risks . surface 1 Lists Map Categories and subcategories of functions .
Table 1

Measure

Measurement functions include analysis 、 Quantitative or qualitative evaluation , Track risks and their impacts . Risk analysis and measurement may involve uncertainty 、 Balance 、 consequence 、 possibility 、 event 、 Detailed consideration of control and effectiveness . An event may have many causes and consequences , And may affect multiple goals . Quantitative or qualitative measurement methods and indicators have developed rapidly , Qualitative and quantitative methods should be used to track risks . The accompanying documents describe practices related to measuring AI risks . surface 2 Lists the categories and subcategories of measurement functions .

surface 2

Manage

This function deals with mapped and measured risks , And manage it , To maximize benefits and minimize adverse effects . These are with AI RMF Risks associated with the attributes of the taxonomy . The decision of this function takes into account the environment and the actual and perceived consequences of external and internal stakeholders , Including the interaction between artificial intelligence system and the current world, as well as the potential benefits or costs . If the risk is considered tolerable , The management can deploy the system as is ; Deploy the system in a production environment that requires additional testing or other control ; If the risk is considered too great to be adequately addressed , The system should be completely deactivated . Like other risk management work , AI risk management must continue . Practices related to AI risk management are discussed in the supporting documents . surface 3 Lists the categories and subcategories of management functions .

surface 3

Govern

The governance function is developing 、 Cultivate and implement a risk management culture within the organization that deploys or acquires AI systems . Governance aims to ensure effectiveness 、 Identify consistently 、 Measure and manage risks and potential impacts . The governance process focusing on the potential impact of AI technology is the pillar of risk management . Governance should address supply chain , Including third-party software or hardware systems and data, as well as internally developed artificial intelligence systems . Governance is a function related to all other functions , It reflects the importance of injecting governance considerations into the whole risk management process and procedures . Governance is the continuous and inherent requirement of effective AI risk management throughout the life cycle of AI systems . for example , Compliance with internal and external policies or regulations is a common aspect of the governance function in risk management . Again , Governance provides a structure , Through this structure , AI risk management functions can be aligned with organizational policies and strategic priorities , Including policy and strategic priorities that are not directly related to AI systems . Supporting documents describe practices related to AI risk management governance . surface 4 Lists the categories and subcategories of the governance function .