Managing model risk for Machine Learning and Artificial Intelligence Models

With modern advances in technology, computing power and algorithmic research, machine learning (ML) and artificial intelligence (AI) models are becoming more commonplace within various industries, such as banking and finance. While ML and AI models may prove to be beneficial in terms of increased accuracy and automation, it is critical for users and developers to be aware of the model risk associated with these types of models. Model risk can usually be defined with respect to (a) model uncertainty risk (the impact of incorrect model outputs), and (b) model use risk (the inappropriate use of models). Up until now, many industry practitioners have made use of model risk governance frameworks that were originally developed for "traditional" models (simpler models that are based on established, longstanding methodologies which typically involve extensive manual interventions by humans). However, recently published frameworks, such as the EU AI Act and the AI Risk Management Framework by NIST, suggest that traditional model governance frameworks may need to be adapted to account for the complexities of ML and AI models. In this talk, we will discuss the thought process and approach behind the establishment of a model risk governance framework for ML and AI models.