The Congressional Research Service ("CRS") identified risks arising from the use of machine learning ("ML") in the underwriting of consumer loans.
CRS highlighted the benefits of using artificial intelligence ("AI") and ML to improve consumer loan underwriting by (i) improving efficiency and performance, (ii) reducing costs and (iii) increasing accuracy in consumer underwriting decision by identifying patterns such as changing credit conditions. CRS also identified potential risk introduced by ML models, including (i) a lack of explainability regarding why programs make certain decisions and (ii) "dynamic updating," which is when ML models gradually evolve without oversight.
CRS identified several policy challenges regarding the use of ML models for credit underwriting, such as:
- a lack of "algorithm transparency" which may hinder the ability of (i) regulators to ensure that ML programs make decisions in compliance with applicable laws and (ii) lenders to provide applicants with a reason for denying their loan application;
- the risk of ML models having (i) training data biases due to being developed using a flawed data set or (ii) historical biases which could cause discrimination against protected classes; and
- privacy and cybersecurity concerns caused by ML models’ access to sensitive consumer financial data.
Commentary
A regulator could assert that every data set is flawed in that it reflects events in the past at time when there was racial bias, or that current data is inherently biased. Further, large AI models are by their very nature not transparent. So, this CRS report is a warning to lenders that they may be breaking the law just by their use of AI. This demands a response by policymakers, at least to provide guidance to lenders on how they can assure compliance.
