Finding a Way to Address Historical Loan Inequalities with Better AI: Learning from Success and Failure


Using the Reject Inference method (Ding & Wan, 2020) to teach AI to see unseen patterns can allow a more evenhanded distribution of loaned funds by allowing acceptance conditions to be more finely tuned to represent the population more fairly. Empirical experiments on the open public lending club dataset show us that a more holistically trained AI can allow us to break out of historically built prejudices in loan decisions. Encouraging organizations to apply Reject Inference methods can be a win-win situation for both loan lender and borrower by reducing Type I and Type II errors, helping us to take a first step out of our historical comfort zone.

the 42nd International Conference on Information Systems