Navigating Model Risk Management in the Age of AI

This week, Christophe Rougeaux from TD and Hua Julia Li who led Model Risk Management at 美国道富银行 in her prior role discussed navigating model risk management in the age of AI at the QuantUniversity AI Fall school. This session was moderated by Sri Krishnamurthy, CFA, CAP

Main Themes:

• Evolution of Model Risk Management (MRM): The conversation centered around the rapid evolution of MRM in the last 5 years, particularly due to accelerated digitization, increased AI adoption, and the emergence of GenAI.
• Adapting MRM to AI: The discussion explored the challenges and strategies for adapting traditional MRM frameworks to effectively address the unique risks associated with AI and GenAI, including explainability, ongoing monitoring, and bias detection.
• AI Use Cases and Applications: The panelists shared various innovative AI use cases emerging across industries, including automation of operations, customer experience enhancement, fundamental investment research generation, automated model documentation and validation, and compliance training.
• Communication and Education: The panel emphasized the importance of clear communication and tailored training programs for different stakeholder groups, including model developers, end-users, and risk management teams.
• AI Governance and Collaboration: The need for establishing robust AI governance frameworks, promoting cross-functional collaboration, and breaking down silos between MRM, cybersecurity, IT, data governance, and legal/compliance departments was highlighted.

Key Takeaways:

• MRM Capabilities Need to Evolve:"MRM is really rethinking their capabilities ... new types of algorithms ... large variety of use cases ... speed to market"
• Holistic Risk Management Approach:"Not thinking anymore as model risk management only, but wearing this risk management hat and looking at the risk holistically as a second line"
• Traditional Frameworks Need Enhancement:"Focus on enhancing our model risk management framework ... areas like ongoing monitoring, explainability"
• Operational Efficiency is Key:"Achieve efficiency, right? Operational efficiency ... our models have a long development and deployment cycle ... it's no longer true"
• Importance of Education and Transparency:"Communicate clearly what are the expectations from risk management ... training ... showing by example"
• "Laying out all the key limitations ... documenting that in a clear manner ... translate that into a language that is understandable by anyone"
• Breaking Down Silos and Fostering Collaboration:"Promote the enterprise-wide AI governance ... model risk management needs to work with other risk disciplines ... everyone has a seat at the table"
• Tailored Training for Different Stakeholders:"Tailor to groups of stakeholders? with different backgrounds explaining in a clear manner to users what could be the risk limitations of AI"

Future Implications:

The evolving landscape of AI and GenAI necessitates continuous adaptation and upskilling within MRM functions. Key areas for future focus include:

• Developing specialized AI/GenAI MRM expertise: Recruiting data scientists with AI expertise and providing tailored training to existing staff.
• Refining existing frameworks and processes: Integrating AI-specific considerations into existing MRM frameworks, particularly for model risk assessment, validation, and ongoing monitoring.
• Establishing robust AI governance structures: Defining clear roles, responsibilities, and policies for AI risk management across the organization.
• Leveraging technology for automation and efficiency: Exploring and implementing AI-powered tools and solutions to streamline MRM processes.

Recommendations:

• Organizations should proactively assess their current MRM capabilities and identify areas for enhancement to address AI-related risks.
• Investments in education and training programs are crucial for upskilling MRM professionals and educating stakeholders across the organization.
• Collaboration and communication across different functions are vital to establish a robust and effective AI governance framework.

You can watch the full session here:

Next week,

Join us next week for a master class on Deploying Large Language Models in Production: Practical Considerations and Best Practices? with Uday Kamath, Ph.D. from Smarsh!

Best,

Sri Krishnamurthy,CFA

QuantUniversity