Responsible A.I.-based Credit Scoring – A Legal Framework

Abstract

The paper proposes a legal framework to evaluate emerging FinTech methodology based on alternative data and machine learning to score borrowers. Instead of conventional variables, novel methods rely on information gathered from social networks, “digital footprints”, mobile phones or GPS data. Correlating these with repayment of loans is promoted as triggering precise predictions of probability of default. Borrowers profit if their profile falls outside of classic scoring checks but performs well under the new regime. Borrowers are disadvantaged if the new methods entail disparate impact for groups which are protected under anti-discrimination laws. Additionally, data may be collected without their consent or used in a way they don’t understand. Two contributions to the debate are submitted. Firstly, a comparative assessment of EU and U.S. data protection and anti-discrimination laws suggests what might qualify as responsible A.I.-based scoring. Secondly, public and private enforcement mechanisms are explained Artificial Intelligence, Credit Scoring, Biased A.I., GDPR, Discriminative Lending Practices, A.I. compliance, Scoring and Banking Regulation, ECOA, FCRA, FICO score