How to Prevent AI Loans from Making
Financial Transactions Even More Unjust to Women
The Gender Bias in Traditional Lending
It’s well documented that women often receive less favorable loan terms than men. A recent investigation into U.S. auto dealership lending practices confirmed this ongoing bias. However, this issue extends beyond cars; it exists globally in mortgages and bank loans as well.
Academic studies show that salespeople might assume women lack market knowledge. This perception makes them more vulnerable to unfair terms. Additionally, some women are penalized for not displaying the same assertiveness as male borrowers.
Enter AI: Hope or Harm?
As artificial intelligence becomes more prominent in lending, it raises an urgent question: Will AI reinforce or reduce these biases? Banks and lenders remain secretive about how extensively they use AI and machine learning. Still, their use is expanding rapidly.
In theory, AI could reduce discrimination by eliminating human bias. But research by my team indicates the opposite could occur. Our study found that AI might worsen gender disparities in lending outcomes.
Discrimination in Canadian Auto Loans
We analyzed over 50,000 auto loans in Canada. The study found clear signs of discriminatory lending against women. Using a metric called “expected utility,” which evaluates a loan’s benefits to a borrower, including interest rates and likelihood of approval, we discovered a troubling gap. Women’s expected utility from loans was 68% lower than men’s.
How AI Could Exacerbate the Problem
We explored how AI might impact commission structures for auto loan salespeople. These commissions significantly influence pricing decisions and are crucial income sources for dealerships. Ideally, AI would automate pricing, reducing human involvement and eliminating commissions. However, dealerships profit heavily from these commissions, so the model is unlikely to change.
Instead, lenders could use AI to fine-tune commissions. This would incentivize salespeople to offer loans that maximize lender profits. Our findings showed lenders could boost profits by 8% this way. However, this comes at a cost to consumers.
The expected utility of loans for consumers would drop by 20% under AI-optimized commissions. Alarmingly, the decline for women was 42%, compared to 17% for men. AI systems trained on biased historical data may have assumed women are more accepting of poor loan terms.
A Growing Concern
These findings confirm fears that AI may amplify discrimination in financial services.
Avoiding AI Isn’t the Answer
Some might argue that lenders should avoid AI altogether. However, this approach ignores the technology’s growing influence. Instead, we explored whether AI could be used more ethically.
A Fairer AI Alternative
We retrained our machine learning model to maximize profits without reducing the expected utility of loans for women. Under this condition, only men’s utility decreased. The result? Lenders still saw a 4% increase in profits.
This finding offers a path forward. When designed with fairness in mind, AI can both protect vulnerable groups and support lenders’ financial goals. Rather than rejecting AI, it’s wiser to adopt responsible practices that prioritize equity.
#AILoans, #GenderBias, #FairFinance, #AIInBanking, #MachineLearningEthics, #WomenInFinance, #FintechEquality, #ResponsibleAI, #BiasFreeLoans, #FinancialInclusion,