How to prevent AI loans from making financial transactions

even more unjust to women.

It’s common knowledge that when women borrow money from salespeople, they typically get less favorable loan terms than do males. That was validated by a recent investigation on loan procedures in US vehicle dealerships. On the other hand, it has long been present in mortgages and bank lending globally.

According to academic research, salesmen may give women harsher terms because they assume that since they are less knowledgeable about the market, they will be less able to determine whether they are receiving a fair deal.

Another possibility is that women face consequences for lacking the assertiveness of men.

The subject of how artificial intelligence (AI) may impact this as it becomes more prevalent in lending is one that is becoming more and more urgent. While banks and other lenders may be reluctant to discuss the degree of their use of generative AI and machine learning in lending, these technologies are undoubtedly being used behind the scenes and will only grow in significance over the coming years.

One may assume that AI could lessen the discrimination against women in lending by, for example, eliminating sales agents’ prejudices. Indeed, a recent study conducted by my research group suggests that things could grow worse. What causes this, and is there a way to prevent it?

Our analysis of over 50,000 auto loans in Canada revealed further instances of discriminatory lending towards women. The typical method for comparing loans in the field of credit research is called “expected utility.”

This gauges the extent to which a loan helps a borrower by taking into account variables such as interest rate, loan approval probability, and salesperson effort. We discovered that women’s predicted usefulness of loans was 68% less than men’s.

We examined how machine learning could improve the fees that lenders pay to salesmen for arranging loans for car buyers in order to get a sense of how artificial intelligence (AI) can affect the automobile industry, which is currently in the early stages of embracing it.

Commissions are a major source of income for dealerships and have a significant impact on the loan pricing decisions made by sales representatives in the auto industry.

In a perfect scenario, using AI in this process would allow you to automate loan pricing, eliminate the need for salesmen, and do away with their commissions. The truth is that dealerships make so much money from commissions and there is enough competition among lenders for them to just go elsewhere for business.

Therefore, it is doubtful that the loan commission model will alter, either in the auto sector or in consumer lending in general.

Lenders can instead take advantage of this by using machine learning to optimize commissions, which will incentivize sales people to choose loan rates that will increase the lender’s projected profits and put in enough effort to persuade the consumer to accept the offer. We discovered that lenders might increase their earnings by 8% by doing this. Naturally, clients pay a price for this. We discovered that in this case, consumers’ expected utility of loans decreases by 20%.

But when we looked at borrowers by gender, we discovered that women’s falls were 42% while men’s were only 17%.

Although we did not test for the specific cause, it is reasonable to assume that the AI made this worse by believing that women are more tolerant of inferior offers than males, given that the back data was “contaminated” with bad loan arrangements for women.

The substitute

This validates long-standing concerns among industry observers that artificial intelligence may lead to increased lending discrimination against women and other groups that are given less favorable conditions, like some ethnic minorities.

One could argue that the wise course of action for lenders would be to completely avoid AI.

However, we questioned if a compromise would be feasible. Could we improve the trade-off between profits and social justice by encouraging lenders to employ AI more responsibly?
In order to test this, we trained the machine-learning algorithm in our study to maximize revenues without negatively affecting the projected value of loans for women. Put otherwise, men’s utility only declined. We discovered that lenders may still make a 4% profit increase under this constraint.

This suggests that AI has the potential to safeguard underprivileged populations and assist lenders when applied wisely. To those who would prefer that artificial intelligence remain out of financial services, it may be wiser to recognize that it is here to stay and instead use it as a tool to improve lending equity.