In this article, the authors discuss the increasing role of artificial intelligence (AI) in consumer lending (focusing on the risks it presents), assess how the Consumer Finance Protection Bureau has approached these risks compared to other federal agencies, and provide recommendations for banks that use AI in consumer lending on reducing the risks associated with AI.
Over the last several decades, banks have increasingly used algorithms to assist with underwriting consumer loans.1 Incorporating more data than ever before, AI has further enhanced the efficiency, speed, and personalization at which loans are made.2 The influx of AI has led consumer groups to express concerns with the Consumer Finance Protection Bureau (CFPB) on necessary protections for consumers. Commentators have noted that AI can pose risks, such as lack of transparency and explainability, unintended outcomes, and biases introduced from incomplete or historical data. Thus far, the CFPB has, among other things, created quality control standards for automated valuation models (AVMs) and issued guidance about the legal requirements lenders must adhere to when credit denials involve the use of AI.
Yet, consumer groups are urging the CFPB to go further and put forward formal guidance on how financial institutions should search for and implement less discriminatory algorithms in credit underwriting and pricing. Other federal agencies have identified the risks posed by algorithmic bias and have issued various forms of guidance. Ultimately, financial institutions must be proactive in mitigating and managing AI-related risks.
AI IN LENDING: CURRENT APPLICATIONS AND TRENDS
According to a recent report, the global AI lending market is expected to grow from $5.7 billion in 2022 to $32.8 billion by 2028.3 AI and machine learning (ML) have revolutionized the lending industry, offering new methods for assessing creditworthiness, making lending decisions, and enhancing customer service.4 Lenders have increasingly adopted these technologies to determine where to extend loans, assess a borrower's credit risk, and predict the likelihood of repayment.
One of the most significant applications of AI/ML in lending is in the development of lending models that streamline the credit underwriting process.5 These AI-driven models are designed to assess the risk of prospective borrowers defaulting on loan repayment, making the process more efficient and potentially more accurate than traditional methods.6 By analyzing various factors, such as bill payment history, unpaid debt, and outstanding loans, AI-driven scoring models can determine not only whether a loan should be granted but also an interest rate that is reflective of the applicant's credit profile.7 Moreover, the ability of ML to analyze vast and diverse data sets, including transaction data, allows lenders to uncover relationships and patterns that may not be evident in traditional models.
While predictive models have been used by lenders for decades (traditionally relying on statistical regression methods and data from credit reporting bureaus), AI/ML can provide a significant improvement.8 Rather than simply assigning weights to various variables to forecast an applicant's likelihood of defaulting or repaying a loan on time, ML models can continuously update themselves by identifying new patterns in credit conditions, thereby making more accurate underwriting decisions.9 This dynamic adaptability of ML models can lead to more precise consumer underwriting decisions, particularly as they can adjust to changing financial landscapes in real-time.
Another critical application of AI in lending is fraud detection and prevention.10 AI-powered systems are increasingly employed to detect fraudulent activities, such as fake identities, fraudulent loan applications, and identity theft. By analyzing large volumes of data, these AI systems can identify patterns and anomalies that may indicate fraudulent behavior, allowing lenders to prevent fraudulent loan applications before they are approved.11 This application of AI not only has the potential to protect lenders but also enhances the overall security and trustworthiness of the lending process.
In addition to these technical applications, AI plays a role in customer service, particularly through the use of AI-powered chatbots. These chatbots can handle a wide range of customer service functions, providing borrowers with instant support and answers to their queries. This not only improves the customer experience but also allows lenders to operate more efficiently by reducing the need for human intervention in routine inquiries.
THE CFPB's ROLE IN SHAPING AI LENDING REGULATIONS
The CFPB is no stranger to issues regarding AI in lending. On June 24, 2024, the CFPB announced it had approved a new rule regarding the use of algorithms and AI for home appraisals and valuations.12 The rule, formally titled, "Quality Control Standards for Automated Valuation Models (2024)," was promulgated by the CFPB, the Federal Reserve, the Federal Deposit Insurance Corporation, the National Credit Union Administration, the Office of the Comptroller of the Currency, and the Federal Housing Finance Agency.13
AVMs are being used with increasing frequency as part of the real estate valuation process.14 To protect the credibility of such models, the CFPB's new rule requires the implementation of quality control policies by mortgage originators and secondary market issuers designed to, among other things, protect against data manipulation and conflicts of interest, and comply with nondiscrimination laws.15 The final rule will become effective next year.16
The CFPB has also issued several updates. These include, for example, the agency's position on combatting digital redlining in the mortgage market;17 an interpretive rule against algorithmic marketing by creditors in advertising to consumers;18 and guidance defining "abusive" AI technologies by creditors that have the potential to harm consumers in financial markets.19
In 2023, the CFPB issued guidance about certain legal requirements that lenders must adhere to when using artificial intelligence and other complex models.20 The guidance mandates that creditors list the actual reason for the credit denial, or change of credit conditions, when taking adverse actions against borrowers, so customers are protected from both potential arbitrary or discriminatory denials, and so customers' future chances for obtaining credit are less impacted.21
In an interagency statement, the CFPB, along with the Civil Rights Division of the United States Department of Justice (DOJ), the Federal Trade Commission, and the U.S. Equal Employment Opportunity Commission, reiterated its commitment to working with these other agencies to monitor discrimination perpetrated through automated systems.22
Finally, the CFPB has also emphasized through guidance the need for creditors using AI and ML to find less discriminatory alternatives in their automated processes, generally, and continuously test their models so that the risk of discriminatory lending practices remains low.23
CONSUMER GROUPS' LETTER TO THE CFPB
In a June 26, 2024 letter to Rohit Chopra, Director of the CFPB, the Consumer Federation of America and Consumer Reports detail the "urgent need for regulatory clarity and certainty" regarding LDAs in credit underwriting and pricing.24
In general, these consumer groups are asking the CFPB to develop guidance for financial institutions on how to search for and implement LDAs in credit underwriting and pricing. The consumer groups stressed that the guidance should be flexible and not overly prescriptive while still providing necessary clarity and protections for consumers.
The groups raised the need for clarity on lenders' obligation to search for and implement LDAs to mitigate discrimination, particularly in the context of disparate impact. The groups emphasized that this obligation should be viewed as a fundamental part of compliance with existing anti-discrimination laws. They suggested that the CFPB issue supervisory guidance that elaborates on the duty to mitigate bias and to identify and implement LDAs.
Specifically, the CFPB's guidance should, the groups contend, cover operational aspects of developing and comparing LDAs and outline appropriate metrics for fairness when testing for disparate impact. The groups want broad guidance on the steps companies should take throughout the model development process to minimize disparate impact. The groups further recommended that companies be required to document their efforts in searching for LDAs, as this documentation would play a crucial role in demonstrating compliance, and that the CFPB offer guidance on considerations for determining the viability of an LDA. To foster a more robust approach to mitigating discrimination, the groups suggested that the CFPB encourage financial institutions to utilize model multiplicity – a practice where different models performing the same task are compared, and those models causing disparate impact are removed and replaced.
Beyond LDAs, the groups called for the CFPB to issue guidance on supervisory expectations and best practices for adopting anti-discriminatory approaches. They noted that the CFPB's existing practice of publishing supervisory highlights, which identify problematic practices within the market, has proven effective in proactively altering business practices. The groups propose that similar highlights, particularly those illustrating examples of effective model testing programs, could be highly beneficial without being overly prescriptive.
Lastly, the groups requested that the CFPB issue an advisory opinion on the adoption of ML technologies. This opinion would, they argue, help ensure consistency and that robust efforts are made across the entire market to prevent discriminatory practices.
THE RISKS OF AI IN LENDING
According to a 2023 survey of senior credit risk executives from twenty-four financial institutions, twenty percent have already implemented at least one AI function in their credit risk operations, with sixty percent expecting to do so within a year.25 Among the respondents' top concerns on the rapid rollout of AI in lending was the need for a framework to avoid model risk issues, such as "transparency, audibility, fairness, and explainability."26
Research has been conducted concerning algorithmic systems' ability to perpetrate societal biases.27 Data may reflect historical biases – in a lender context, these may include, for example, unfair underwriting records stemming from historical systemic racism28 and/or the lack of credit history data on mortgages for minority groups.29
Further, the "black box" nature of certain automated systems' programming can prevent transparency that automated systems are operating without bias.30 Developers, creditors, and lenders alike are thus left unsure whether their automated systems are fair. 31
The International Monetary Fund (IMF), too, found that oversight should exist so that data used to train Generative AI (GenAI) models in financial industry processes is not only complete, but does not reinforce underlying, embedded biases that exist in the data used to train AI/ML models.32 This concern, along with the potential for creditors to over-rely on AI without consumers realizing as much, requires the "appropriate human judgment" to complement GenAI-based lending models, according to the IMF.33
Finally, the cost to invest in the necessary hardware, software, and expertise to train and deploy AI systems can be significant, especially for smaller lenders.34
OTHER STANCES ON RISKS OF ALGORITHMIC BIAS IN AI
The CFPB is not alone in assessing the risks of algorithmic bias in AI. Other regulatory and governmental agencies are attempting to address the potential discriminatory impact of AI.
For example, on May 12, 2022, the DOJ released a technical assistance document entitled, "Algorithms, Artificial Intelligence, and Disability Discrimination in Hiring,"35 describing how algorithms and AI can lead to disability discrimination in hiring, including against potential applicants with disabilities.36
The Securities and Exchange Commission's (SEC) Investor Advisory Committee (IAC), too, published guidance on establishing a framework for ethical AI functions for investment advisors.37 The IAC noted that the SEC has authority to monitor the industry's use of technology to provide investment advice and should do so to promote equity and ensure fair and equal access to markets.38
Finally, the Office of Management and Budget (OMB) published a memorandum this past March mandating the mitigation of algorithmic discrimination in AI processes.39 The memorandum established guidance for federal agencies to create AI governance structures, advance AI innovation, and manage risks from the government's use of AI. The OMB's key guidance included detailing a blueprint for an AI Bill of Rights that includes protection from unsafe systems, algorithmic discrimination, abusive data practices, and access to human alternatives40 and discussing an AI Risk Management framework to establish evaluation and monitoring of government AI systems.41 The OMB's memorandum builds off an Executive Order passed by President Biden in October 2023, directing similar objectives.42
MITIGATING AI-RELATED LENDING RISKS
To mitigate the risks associated with AI in lending, there are several steps financial institutions can take related to ongoing model oversight, transparency, and comprehensive evaluation.
First, by consistently and routinely monitoring the results of their ML models, financial institutions can fine-tune those models as necessary. Regular observation ensures that the models are performing as expected and allows for timely adjustments to address emerging issues.43
Another tool for financial institutions is to demystify the "black box" of certain ML models.44 Taking that step will allow financial institutions to understand the underlying algorithms and decision-making processes for those ML models. This transparency will not only assist with regulatory compliance but also demonstrate to stakeholders and consumer groups how the models are supervised and how the results are produced.45 Establishing clear communication about how AI-based credit decision-making works can enhance trust between borrowers and lenders, ensuring that all parties understand the basis of the credit assessments.
Additionally, banks should pay close attention to how alternative data sources – such as internet searches, shopping habits, punctuation in communications, and hobbies – impact the outcomes of their models.46 By scrutinizing the influence of such alternative data, banks can avoid incorporating variables that may lead to biased or misleading results.
Developing and comparing multiple models to perform the same task is another effective strategy. By evaluating which model is most successful in minimizing disparate impact, financial institutions can identify and implement the most effective solution while discarding less effective models.47 This approach improves fairness and enhances the overall performance of the lending process.
Peer group performance benchmarking can also be beneficial. Financial institutions that assess their lending performance against their market peers, analyzing application numbers, denial rates, and withdrawal rates, particularly in minority areas can help identify any disparities and ensure that lending practices are equitable across different demographic groups.48
Finally, conducting thorough market studies can also help. Gaining a deep understanding of the market areas being served, including the demographics and specific characteristics of those areas, enables financial institutions to tailor their AI models more effectively to the needs of their target populations and to address any potential biases that may arise from regional or demographic factors.
Employing all of these recommendations may not be necessary for all financial institutions, but a combination of them can assist financial institutions with managing the risks associated with AI in lending. Implementing a defensible AI compliance system will become increasingly important as regulators and consumers continue to focus on AI and as AI continues to be integrated into lending and other processes.
Footnotes
1 Congressional Research Service, Automation, Artificial Intelligence, and Machine Learning in Consumer Lending (May 10, 2023), available at https://crsreports.congress.gov/product/pdf/ IF/IF12399.
2 See id.
3 Revolutionizing Lending: The Impact of AI on Modern Financial Services (June 19, 2024), available at https://manvsdebt.com/revolutionizing-lending-the-impact-of-ai-on-modern-financialservices/.
4 Congressional Research Service, Artificial Intelligence and Machine Learning in Financial Services (Apr. 3, 2024), available at https://crsreports.congress.gov/product/pdf/R/R47997.
5 Akash Takyar, AI in Loan Underwriting: Use Cases, Architecture, Technologies, Solution and Implementation, available at https://www.leewayhertz.com/ai-loan-underwriting/ (last visited Aug. 25, 2024).
6 Akash Takyar, AI-based Credit Scoring: Use Cases and Benefits, available at https://www. leewayhertz.com/ai-based-credit-scoring/ (last visited Aug. 25, 2024).
7 Aaron Klein, Credit Denial in the Age of AI (Apr. 11, 2019), available at https://www. brookings.edu/articles/credit-denial-in-the-age-of-ai/.
8 Supra note 4.
9 Overview: The Use Of Machine Learning For Credit Underwriting: Market & Data Science Context, available at https://finreglab.org/research/overview-the-use-of-mahine-learning-forcredit-underwriting-market-data-science-context/ (last visited Aug. 25, 2024).
10 Ravi Sandepudi, The Banker's Guide: Using AI for Fraud Detection (Mar. 11, 2024), available at https://effectiv.ai/resources/fraud-detection-using-ai-in-banking/.
11 Maciej Markiewicz, Risk Reducing AI Use Cases for Financial Institutions (June 13, 2024), available at https://www.netguru.com/blog/risk-reducing-ai-use-cases-financialinstitutions.
12 Consumer Financial Protective Bureau, Quality Control Standards for Automated Valuation Models (June 24, 2024), available at https://www.consumerfinance.gov/rules-policy/ final-rules/quality-control-standards-for-automated-valuation-models/.
13 Quality Control Standards for Automated Valuation Models, 89 Fed. Reg 64538 (Aug. 7, 2024).
14 Consumer Financial Protective Bureau, Agencies Issue Final Rule to Help Ensure Credibility and Integrity of Automated Valuation Models (Jul. 17, 2024), available at https://www.consumerfinance.gov/about-us/newsroom/agencies-issue-final-rule-to-help-ensurecredibility-and-integrity-of-automated-valuation-models/.
15 See id.
16 Consumer Financial Protective Bureau, Agencies Issue Final Rule to Help Ensure Credibility and Integrity of Automated Valuation Models (Jul. 17, 2024), available at https://www.consumerfinance.gov/about-us/newsroom/agencies-issue-final-rule-to-help-ensurecredibility-and-integrity-of-automated-valuation-models/.
17 Consumer Financial Protective Bureau, Director Chopra's Prepared Remarks at Justice Department Interagency Event in Newark, New Jersey to Highlight Efforts to Combat Modern-Day Redlining (Apr. 19, 2023), available at https://www.consumerfinance.gov/aboutus/newsroom/director-chopra-remarks-justice-department-interagency-event-combat-modern-dayredlining/.
18 Consumer Financial Protective Bureau, CFPB Warns that Digital Marketing Providers Must Comply with Federal Consumer Finance Protections (Aug. 10, 2022), available at https://www.consumerfinance.gov/about-us/newsroom/cfpb-warns-that-digital-marketing-providersmust-comply-with-federal-consumer-finance-protections/.
19 Consumer Financial Protective Bureau, CFPB Issues Guidance to Address Abusive Conduct in Consumer Financial Markets (Apr. 3, 2023), available at https://www.consumerfinance. gov/about-us/newsroom/cfpb-issues-guidance-to-address-abusive-conduct-in-consumer-financialmarkets/.
20 Consumer Financial Protective Bureau, CFPB Issues Guidance on Credit Denials by Lenders Using Artificial Intelligence (Sep. 19, 2023), available at https://www.consumerfinance. gov/about-us/newsroom/cfpb-issues-guidance-on-credit-denials-by-lenders-using-artificialintelligence/.
21 See id.
22 Federal Trade Commission, Joint Statement On Enforcement Efforts Against Discrimination And Bias In Automated Systems (Apr. 2024), available at https://www.consumerfinance. gov/about-us/newsroom/cfpb-issues-guidance-on-credit-denials-by-lenders-using-artificialintelligence/.
23 Brad Bowler, CFPB Puts Lenders & FinTechs On Notice: Their Models Must Search For Less Discriminatory Alternatives Or Face Fair Lending Non-Compliance Risk (Apr. 5, 2023), available at https://ncrc.org/cfpb-puts-lenders-fintechs-on-notice-their-models-must-search-forless-discriminatory-alternatives-or-face-fair-lending-non-compliance-risk/.
24 Letter, Jennifer Chien, Senior Policy Counsel, Consumer Reports & Adam Rust, Director of Financial Services, Consumer Federation of America, Urgent Call for Regulatory Clarity on the Need to Search for and Implement Less Discriminatory Algorithms (June 26, 2024), available at https://advocacy.consumerreports.org/wp-content/uploads/2024/06/240626-CR-CFA-Statementon-Less-Discriminatory-Algorithms-FINAL.pdf.
25 McKinsey & Company, Embracing generative AI in credit risk (Jul. 1, 2024), available at https://www.mckinsey.com/capabilities/risk-and-resilience/our-insights/embracing-generative-aiin-credit-risk.
26 See id.
27 Barocas, Solon and Andrew D. Selbst. "Big Data's Disparate Impact." 104 California Law Review 671, 2016, available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2477899.
28 Neil Bhutta, Aurel Hizmo & Daniel Ringo, How Much Does Racial Bias Affect Mortgage Lending? Evidence from Human and Algorithmic Credit Decisions (Oct. 2022), available at https://doi.org/10.17016/FEDS.2022.067.
29 Rashawn Ray, Andre M. Perry, David Harshbarger, Samantha Elizondo, and Alexandra Gibbons, Homeownership, racial segregation, and policy solutions to racial wealth equity (Sept. 1, 2021), available at https://www.brookings.edu/articles/homeownership-racial-segregation-andpolicies-for-racial-wealth-equity/.
30 Federal Trade Commission, Joint Statement On Enforcement Efforts Against Discrimination And Bias In Automated Systems (Apr. 2024) https://www.consumerfinance.gov/aboutus/newsroom/cfpb-issues-guidance-on-credit-denials-by-lenders-using-artificial-intelligence/.
31 Id.
32 Ghiath Shabsigh, El Bachir Boukherouaa, Generative Artificial Intelligence in Finance: Risk Considerations (Aug. 22, 2023), available at https://www.imf.org/en/Publications/fintechnotes/Issues/2023/08/18/Generative-Artificial-Intelligence-in-Finance-Risk-Considerations537570.
33 See id.
34 Jennifer Chien, Adam Rust, Urgent Call for Regulatory Clarity on the Need to Search for and Implement Less Discriminatory Algorithms (Jun. 26, 2024), available at https://advocacy. consumerreports.org/wp-content/uploads/2024/06/240626-CR-CFA-Statement-on-LessDiscriminatory-Algorithms-FINAL.pdf.
35 U.S. Department of Justice Civil Rights Division, Algorithms, Artificial Intelligence, and Disability Discrimination in Hiring (May 12, 2022), available at https://www.ada.gov/resources/ ai-guidance/.
36 Id.
37 Christopher Mirabile, Leslie Van Buskirk, Establishment of an Ethical Artificial Intelligence Framework for Investment Advisors (Apr. 6, 2023), available at https://www.sec.gov/files/ 20230406-iac-letter-ethical-ai.pdf.
38 Id.
39 Shalanda D. Young, Advancing Governance, Innovation, and Risk Management for Agency Use of Artificial Intelligence (Mar. 28, 2024), available at https://www.whitehouse.gov/ wp-content/uploads/2024/03/M-24-10-Advancing-Governance-Innovation-and-Risk-Managementfor-Agency-Use-of-Artificial-Intelligence.pdf.
40 The White House, Blueprint for an AI Bill of Rights, available at https://www.whitehouse. gov/ostp/ai-bill-of-rights/ (last accessed Aug. 25, 2024).
41 National Institute of Standards and Technology, AI Risk Management Framework, available at https://www.nist.gov/itl/ai-risk-management-framework (last accessed Aug. 25, 2024).
42 The White House, Fact Sheet: President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence (Oct. 30, 2023), available at https://www.whitehouse.gov/ briefing-room/statements-releases/2023/10/30/fact-sheet-president-biden-issues-executive-orderon-safe-secure-and-trustworthy-artificial-intelligence/.
43 Unleashing the Power of Machine Learning Models in Banking Through Explainable Artificial Intelligence (XAI) (May 17, 2022), available at https://www.deloitte.com/global/en/ our-thinking/insights/industry/financial-services/explainable-ai-in-banking.html.
44 See id.
45 Aparna Dhinakaran, Overcoming AI's Transparency Paradox (Sept. 10, 2021), available at https://www.forbes.com/sites/aparnadhinakaran/2021/09/10/overcoming-ais-transparencyparadox/.
46 Joel Rickman, How Alternative Data Can Help Expand Opportunities and Allow Banks to Acquire New Account Holders, available at https://www.bai.org/banking-strategies/howalternative-data-can-help-expand-opportunities-and-allow-banks-to-acquire-new-accountholders/.
47 Daniel Johnson, The Use of AI for Less Discriminatory Alternative Models in Fair Lending (Feb. 29, 2024), available at https://www.treliant.com/knowledge-center/the-use-of-aifor-less-discriminatory-alternative-lda-models-in-fair-lending/.
48 David Deckelmann, What is the Benefit of Peer Benchmarking?, available at https:// livecusurvey.com/benefit-peer-benchmarking/ (last visited Aug. 25, 2024); Andy Barksdale, 3 Simple Strategies to Avoid CRA, Fair Lending, and Redlining Risk (Sept. 10, 2014), available at https://www.ncontracts.com/nsight-blog/cra-fair-lending-compliance-redlining-riskmanagement.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
