By David Langlieb
Have the robots taken over yet? Pardon my skepticism, but I turned 40 last year and find myself settling into a curmudgeonly distrust of new technology. I’m not a Luddite, but the truth is I’ve been temperamentally incredulous since second or third grade, so I’m trying to guard against it. The only thing more tedious than an older person complaining about technology is a late millennial complaining about technology.
It is in this spirit with which I’ve processed all the recent fuss over ChatGPT, its competitors, and the advances in artificial intelligence poised to revolutionize white-collar work the way that automation transformed blue-collar work in the latter half of the 20th century. This hits home for me personally, as we’ve been told artificial intelligence will automate the credit analysis business since I started underwriting small business loans in 2013. There are reasons to think it may. Credit analysis is partly math and not terribly sophisticated math. Software already exists that can instantly scan standard inputs like borrower tax returns, personal financial statements, and credit reports and spit out pages of metrics and ratios that would take hours to compile manually on paper or with the help of once cutting-edge technology like Microsoft Excel.
And yet, the labor market within banking and Community Development Financial Institution (CDFI) lending remains exceptionally competitive, with experienced loan officers in record high demand. The “problem” (if you want to call it that) is the same as it is in most industries: the robots can do some things quite well, but to the extent that judgment and creativity are needed to complete the job, there’s still nothing close to a satisfactory substitute for competent human beings.
First-year associates at white-shoe law firms once spent most of their time manually looking up old cases for precedents and citations. The advent of online databases and computer search engines radically transformed this thankless, time-consuming chore. But no one is hiring a chatbot to represent them in court. The closest example – a 2023 personal injury lawsuit brought in Manhattan in which two lawyers submitted a brief written by ChatGPT – resulted in six fictitious cases being included in the brief and the lawyers responsible being laughed out of court and sanctioned thousands of dollars in fines.
Credit analysis is only partly math, and a typical credit decision will account for at least a half dozen subjective factors – borrower character, breadth and quality of experience within a given industry, credibility of projections, and a good deal more. In a sense, the foundational principles of community development lending are hostile to automated underwriting anyway. We are taught, quite correctly, that understanding local markets and industries means finding the gaps in the market that the math can obscure.
Further, at its core, a loan approval process is a negotiation between at least two parties, and negotiations are essentially exercises in trust building. This is true even in the mission-based lending world, where we make every effort to treat our borrowers as partners rather than adversaries. Every loan agreement obligates both lender and borrower to a set of terms and conditions, with a process predicated on both sides living up to their obligations or at least attempting to do so in good faith. Trust is a hard-won and precious commodity in the community lending world, particularly for organizations like PAF which lend to historically disadvantaged communities overcoming centuries of discrimination and abuse from financial institutions.
The question then becomes, can a prospective loan applicant ever trust an automated lender?
I don’t see how. Any automated lending robot programmed by a lender will behave as the lender’s algorithm demands. Thus, a robot programmed by a lender is, by definition, going to pursue the interests of that lender as efficiently and ruthlessly as possible. Following that logic, the more sophisticated a publicly facing lender algorithm becomes, the more distrust it will engender.
Of course, the more complete answer is that yes, a loan applicant can be made to trust an automated lender, but only if that applicant has no other options. This is what is going on right now in the still-nascent fintech (online lender) industry, where the most technologically sophisticated tools in the financial services toolkit are deployed on the most powerless borrowers. A typical fintech lender takes some key inputs (net income, personal guarantor wealth, liquidity, credit score, etc.), runs it through an automated model, and potentially spits out an approval, often within 24-48 hours. But that approval will carry heavy fees and be priced at an exorbitant interest rate constrained only by the usury laws as applied in the lender’s state. Even this turns out to be thin protection – in many states, usury laws have not caught up to fintech lending, and certain lenders have found ways to charge rates up to and over 50% by calling their loans some other name or by incorporating in states with weak regulation.
What has happened is essentially a bifurcation within commercial lending – borrowers without existing bank or CDFI relationships and borrowers in dire straits without any negotiating leverage are compelled to rely on exploitative financial products. Meanwhile, better-positioned borrowers have the time and resources necessary to wait for a living, breathing human being to evaluate their loan application. It is a challenge to the CDFI industry broadly to make our processes as efficient and responsive as possible to discourage borrowers from taking out usurious fintech loans or maxing out their credit cards. That can (and should) involve the selective use of AI to streamline the parts of the underwriting, documentation, and loan approval processes where automation is most effective. And while “same day” loan approvals will never be achievable, our charge should, at the very least, be to provide borrowers with fast term sheets, realistic expectations, and frequent updates as their applications move through the credit analysis and approval process.
The Problem is Never the Technology Itself
The world of commercial lending I describe above is a harsh one, made only harsher by newer technologies better engineered to displace human credit analysts and maximize lender profits. This is at odds with the “techno-optimist” consensus about new technology over time – that, on the whole, we as a society benefit dramatically the better our technology gets. This viewpoint dismisses short-term problems – the unemployed workers who get automated out of their jobs or the under-resourced loan applicants ill-equipped to deal with exploitative financial algorithms – as a minor cost of progress. Recall the pinnacle scene in the film Other People’s Money, where Danny DeVito (memorably portraying leveraged-buyout artist Larry “the Liquidator” Garfield) discusses with shareholders his reasons for attempting to dissolve what he regards as an obsolete wire and cable company:
“You know, at one time, there must’ve been dozens of companies making buggy whips. And I’ll bet the last company around was the one that made the best goddamn buggy whip you ever saw. Now, how would you have liked to have been a stockholder in that company?”
True enough. But as a closing note of caution, consider the case of the cotton gin, helpfully referenced by MIT economics professor Daron Acemoglu and his colleague Simon Johnson in their recently published book Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. Their work thoroughly investigates how technological progress has impacted societies throughout history. It is a cautionary retort to the techno-optimists who promote AI as a superhighway to greater human flourishing.
As is widely understood, the cotton gin was revolutionary technology when Eli Whitney patented it in 1794. Even the earliest versions of Whitney’s gin were a dramatic technological improvement over the painstaking manual process of separating cotton fibers from the seeds in cotton bolls, and the widespread use of the cotton gin could have been an unadulterated net benefit to society at the time.
But as Acemoglu and Johnson remind us, the cotton gin was overwhelmingly responsible for sustaining and expanding slavery in the American South for the six decades following Whitney’s patent. Slavery was arguably a dying institution towards the end of the 18th century – in addition to imposing unspeakable misery on millions of people, the slave-centered economies of the Southern states were inefficient, backward, and primed for collapse as constituted. The cotton gin was such a dramatic technological advance that it made barbarism increasingly profitable, deepening the rot at the heart of Southern society. And it further implicated Northerners in the moral evil of slavery, as the prosperity of New England’s textile mills and the lucrative export of clothing would have been impossible absent the proliferation of inexpensive raw cotton from the South.
All this incentivized the perpetuation of American slavery. Nine new slave states joined the country after the invention of the cotton gin. The output of raw cotton doubled every ten years between Whitney’s patent and the Civil War, and by 1850, the majority of the global production of raw cotton was grown in what would soon be the Confederacy. A disturbingly straight line can be drawn from the patent of the cotton gin to the Civil War, the failures of Reconstruction, the horrors of the Jim Crow South, and the racial inequities that haunt the country to this day.
Of course, the cotton gin itself is an inanimate object incapable of imposing moral evil. It took a sick society, corrupted to its core – the moneyed plantation owners, the Northern businessmen, the politicians, the religious institutions, the schools, and the poor white underclass collectively – to make the cotton gin a tool of evil in the American South.
Given such history, it is helpful to consider all kinds of technological progress through this lens. Professors Acemoglu and Johnson remind us that new technology is only as wonderful as what we decide to do with it. The rapid advances in AI are simply the next challenge in the fight for human prosperity and our collective happiness; let us resolve to put it to work for better purposes than high-interest fintech loans.