Should Algorithms Determine Financial Eligibility?
By AG CPA Co. | Perspectives on Finance & Technology |
Credit decisions — whether to extend a loan, at what rate, and under what terms — have historically involved human judgment operating within a regulatory framework designed to prevent discrimination. The Equal Credit Opportunity Act, the Fair Housing Act, and the Community Reinvestment Act each reflect a belief that access to credit is too consequential to be left entirely to unconstrained discretion. The question now before regulators, researchers, and the public is whether replacing human discretion with algorithmic decision-making advances or undermines that goal.
The answer, it turns out, depends heavily on which algorithm, applied to which data, with what oversight.
The Promise
The case for algorithmic credit assessment is not difficult to make. Human loan officers carry biases — sometimes explicit, often implicit — that research has consistently shown produce discriminatory outcomes. A 2024 study published in MIS Quarterly examined a major bank's adoption of an AI-enabled credit scoring model for an underserved population of over one million customers. The AI model incorporated what researchers called 'weak signals' that traditional rule-based systems ignored, providing more accurate estimates of individual creditworthiness. The result was reduced reliance on group characteristics that had previously led to financial exclusion.
This is statistical discrimination working in reverse: a more precise model that sees individuals more clearly than a blunt instrument that relies on proxies. When AI genuinely improves prediction accuracy at the individual level, it can expand access to credit for people whom traditional models categorically exclude.
The Problem
But the optimistic case requires conditions that are not always present. AI systems trained on historical lending data inherit the biases embedded in that history. A model trained on decades of lending decisions in which Black and Latino applicants were systematically denied credit will find patterns in that data and reproduce them — not because the algorithm intends discrimination, but because discrimination is what the training data reflects.
A 2024 Urban Institute analysis of Home Mortgage Disclosure Act data found that Black and Brown borrowers were more than twice as likely to be denied a loan as white borrowers with comparable credit profiles. A University of California Berkeley study found that African American and Latino borrowers are charged nearly five basis points more in interest rates than credit-equivalent white counterparts — amounting to $450 million in additional interest payments annually. Whether algorithmic systems are improving or extending these disparities depends on how they are designed and what data they use.
Research from the University of Illinois at Urbana-Champaign has further identified differential predictive accuracy by gender in widely used credit scoring models. Credit scores were found to be more accurate predictors of default among male borrowers than female borrowers across most loan types — an embedded asymmetry with real consequences for how creditworthiness is assessed.
The Regulatory Response
Legislators are beginning to respond. Hawaii introduced SB 59, which prohibits discriminatory 'algorithmic eligibility determinations' — defined as any determination based in whole or significant part on an algorithmic process using machine learning or AI to determine access to credit or other important life opportunities. Illinois amended its Consumer Fraud and Deceptive Business Practices Act in 2024 to expand oversight of AI used in consumer creditworthiness determinations, with the amendment taking effect in early 2026. The federal landscape shifted considerably when the Consumer Financial Protection Bureau was effectively shut down in early 2025, but state-level activity has increased in its absence.
The Right Question
The debate over algorithmic financial eligibility is sometimes framed as a choice between human judgment and machine judgment. That framing is probably not useful. Human judgment has a documented record of discriminatory outcomes. Algorithmic systems have a documented record of encoding and amplifying historical bias. The productive question is not which is better, but what design, oversight, transparency, and accountability requirements make either one acceptable.
For businesses and individuals navigating financial systems, what matters is whether the criteria used to evaluate them are fair, explainable, and contestable. An opaque algorithm that produces adverse outcomes without explanation offers little recourse. That gap — between the system's power and the individual's ability to understand or challenge it — is where the most important policy work remains to be done.
REFERENCES
Li, C., Wang, H., Jiang, S., & Gu, B. 'The Effect of AI-Enabled Credit Scoring on Financial Inclusion.' MIS Quarterly 48(4), 2024.
Urban Institute. Analysis of Home Mortgage Disclosure Act Data. 2024.
UC Berkeley Study on FinTech Lending, cited in Kennedy Human Rights Center. 'Bias in Code: Algorithm Discrimination in Financial Systems.' August 2025.
Goodwin Law. 'The Evolving Landscape of AI Regulation in Financial Services.' June 2025.
Gies College of Business, University of Illinois. 'New Research Reveals Widespread Bias, Inefficiency in Credit Scoring.' November 2025.