At an AI Academic Symposium, Federal Reserve Board Governor Lael Brainard compared the potential benefits and the regulatory risk of artificial intelligence ("AI") in financial services.
On the benefits of AI, Ms. Brainard focused on those for "credit invisible" individuals who may not have produced data that is usable by existing credit scoring methodologies. These individuals could be materially benefited by new methodologies that take account of a wider range of data.
Among the regulatory risks, Ms. Brainard warned of the "black box problem" of machine learning. She cautioned that human beings can input data into a model and can examine the outputs, but the process between the start and finish is opaque. Ms. Brainard explained that the lack of model transparency, combined with an algorithm's levels of "explainability," will affect banks in numerous ways. Ms. Brainard gave the following examples:
a compliance officer's ability to understand how a model affects operational risk management; and
a lender's ability to provide the required explanation to a consumer following a decision to decline consumer credit to them based on a machine learning model.
Ms. Brainard urged regulators to "provide appropriate expectations and adjust those expectations as the use of AI in financial services . . . evolve[s]." Ms. Brainard stated that the FRB is exploring additional supervisory clarity to facilitate the "responsible adoption" of AI, and encouraged stakeholders to provide additional feedback.