Hong Qu

Impact of AI on Fair Lending Laws and Practices

Project Lead: Hong Qu

 

OVERVIEW

A decade has passed since the subprime mortgage crisis and the Fintech industry is starting to embracing alternative data and machine learning to improve predictive models that have the capability to determine the creditworthiness of consumers with low credit scores.  Compared to traditional credit-risk and lending practices, these new financial products and services–buttressed by fledgling data sets and enigmatic algorithms–pose a new set of challenges for regulatory oversight and risks to consumer welfare.

We will apply the AI Blindspot discovery process created at the Assembly program at the Berkman Klein Center and MIT Media Lab to investigate the potential benefits and shortcomings of new financial service offerings from startups such as Upstart, with emphasis on their compliance with three legal mandates: 1. Equal Access to Credit Act’s call for guarding against algorithmic bias; 2. Consumer Credit Protection Act’s stipulation for transparency and explainability; 3. Dodd-Frank Act’s prohibitions against unfair, deceptive, or abusive acts or practices.

 

APPROACH

This project seeks to engage with the relevant social groups in the consumer loan marketplace to identify blind spots that might arise from unconscious bias and historical inequalities over the course of developing, deploying and operating AI systems that automate consumer lending processes.  We will delve into issues such as discrimination by proxy, disparate impact, transparency and explainability, and control over personal data.  

Sample questions we plan to address are:

• How should regulators develop monitoring and enforcement tools to promote fair lending by AI systems?
• What types of alternative data will likely lead to discrimination by poxy and reinforce historical inequalities?
• How might we ensure that data collection processes respect the right to privacy and informed self-determination of all borrowers, especially those 19 million Americans who are credit invisible?
• For those consumers who have been denied credit, what are their rights to explanation, appeal, and collectively challenge the fairness of the credit-risk model?

PROCESS

This proposed project will:

1. Review academic literature, industry trends, and public policy to analyze the relevant social groups’ power dynamics and incentives.
2. Conducting lightweight ethnographic observation and interviews with major stakeholder groups such as tech company product teams, regulators, academic experts, consumer advocates, and impacted populations in marginalized communities.
3. Propose a mechanism to monitor disparate impact by testing, documenting, and reporting on unfair, deceptive or abusive practices.
4. Write and publish a primer on the interplay of consumer access to credit, legal and regulatory policies, and the technical basis for alternative data and machine learning models which enable predictive modeling of credit risk.
5. Share project findings with journalists in trade, local, and ethnic media.

• PhD student, Network Science Institute at Northeastern University
• Research Fellow, Malcom Wiener Center for Social Policy at Harvard Kennedy School
• Adjunct Lecturer, Harvard Kennedy School
• Research Director, The Council on the Responsible Use of Artificial Intelligence, Belfer Center
• Assembly Project Fellow, Berkman Klein Center