Modifying model risk management practice in the era of AI/ML

Abstract

The model use of artificial intelligence (AI) and machine learning (ML) has caused unprecedented sensation around the wide applicability of these techniques. The rapid adoption of those alternative tools and methodologies by the heavily regulated financial sector, in areas that are outside the conventional credit lending and market participation, has posed significant challenges for model risk management professionals, including correctly defining AI and ML, properly establishing a governance framework, and, most importantly, effectively challenging AI/ML models. In this paper, the author attempts to describe the history of AI/ML, the evolution of key mathematical theories and modelling, commonalities and distinctions between statistical models and ML algorithms, and challenges of evaluation of some ML models. She discusses plausible solutions to practically address those challenges.