2020-11-30
The Deutsche Bundesbank issued this policy discussion paper to outline supervisory expectations for the use of machine learning in the financial sector, emphasizing a risk-oriented and technology-neutral approach. The document argues that existing prudential frameworks are sufficient for most applications, rejecting mandatory prior approval for non-Pillar 1 models while insisting on rigorous validation, data quality controls, and human accountability for algorithmic decisions. It further addresses the 'black box' dilemma by advocating for explainable AI techniques and robust control processes to manage risks associated with model transparency and adaptive learning.