Like all fields, finance is rapidly being transformed by emerging AI technology. McKinsey reports that generative AI could add an annual $200 to $340 billion in value to the global banking sector, largely through increased productivity. A May 2023 survey of about 75 CFOs at large organizations found that almost a quarter were actively investigating uses for generative AI in finance. Overall, the potential annual value of AI and analytics for global banking could be as high as $1 trillion. In short, there is plenty of excitement around FinTech—or “Financial Technology”—right now, excitement which serves as the momentum behind CSAIL’s latest research initiative, FinTechAI@CSAIL.
However, the application of new technologies to finance comes with important challenges, ethical concerns, and creative opportunities to consider. Professor of Finance in the MIT Sloan School of Management, Principal Investigator at CSAIL, and faculty lead of FinTechAI@CSAIL Andrew W. Lo has dedicated his career to these challenges, studying the nature of intelligence—both human and artificial—to better capture human behavior in markets, creating technological solutions to industry problems, and offering new ideas to help fund risky but important ventures like cancer research and fusion. Professor Lo wants to leverage AI and computer science to help small-portfolio investors and use the power of these incredible new technologies to tackle some of the biggest financial and economic challenges of our time.
FINDING HIS INTEREST
Professor Lo’s love of computer science began in high school when he was given access to an IBM 1620 and HP 2000E at the Bronx High School of Science. Those early computers were enormous—the size of commercial refrigerators—with only a fraction of the capabilities of a modern iPhone. However, a young Professor Lo was hooked, learning FORTRAN, BASIC, and Assembler so he could work on larger projects, like creating an “expert system” that could play Monopoly optimally. This was his first introduction to AI as well as to Monte Carlo simulations, both of which he still works with today.
Professor Lo went on to earn a bachelor’s degree in economics from Yale University and then a PhD in economics from Harvard University. He began his academic career as an assistant and then associate professor of finance at the University of Pennsylvania’s Wharton School. In 1988, he joined the MIT faculty where he now studies and teaches financial economics. He is the Charles E. and Susan T. Harris Professor at the MIT Sloan School of Management, Director of MIT's Laboratory for Financial Engineering, and the faculty lead of FinTechAI@CSAIL. He’s also put theory into practice, launching a hedge fund in 1999 from which he stepped down as chairman at the end of 2021, and currently serves as chairman and chief investment strategist for a healthcare investment management company he co-founded in 2019.
Over the course of his career, Professor Lo says he’s seen tremendous progress in the fields of computer science and economics separately and in their convergence. He says, “some of the core ideas in econ—utility theory, game theory, and risk analysis—are now being studied by computer scientists in the context of AI and data science, and the reverse is true for a number of younger economists who are applying ideas in statistical learning theory, cryptography and differential privacy, and machine learning to a number of economic problems. The two fields have much to learn from each other.”
Indeed, Professor Lo himself has taken part in the evolution of these ideas with many notable publications, such as his award-winning book Adaptive Markets: Financial Evolution at the Speed of Thought, which blends knowledge from a variety of fields, including AI, to apply the theory of evolution to markets. In earlier work, Professor Lo had discovered that markets were not, in fact, random and could be forecasted to some degree, a finding which “flew in the face of much of modern financial theory.” This led him to work on a new theory which explained the markets as “highly competitive, much like an ecosystem with predators and prey.”
As with his book, Professor Lo is taking an interdisciplinary approach to his current research, which includes integrating computer science innovations into economic and financial practices to craft novel methods for understanding human behavior, apply AI tools in industry, and to support non-experts in risky investment ventures through the use of ML.
FINANCE AND COMPUTER SCIENCE: AI, MEGAFUNDS, CLINICAL TRIALS, AND MORE
In his recent podcast with CSAIL Alliances, Professor Lo explains how we’re not that far from an LLM that would be “able to dispense reliable, accurate, customizable, and trusted advice in a number of different domains, including financial investing.” This has the potential to be broadly beneficial, especially for small-portfolio investors who might not be able to afford a traditional financial advisor but could gain a lot from an AI system that could offer inexpensive, personalized advice on common financial pitfalls. Such a system could also benefit financial advisors, allowing them to reach more customers and focus on cases that need more creative or unique solutions.
Unsurprisingly, there are challenges to overcome before an AI model might provide real-world financial advice to consumers. First, the advice has to be accurate, a problem Professor Lo believes is close to being solved. Current LLMs are already capable of passing the standardized tests used to certify financial expertise and will continue to improve as the models get better. Second, the advice must be customized to the specific circumstances of an individual, which Professor Lo says, “is still about a year or two away, but my collaborators and I are optimistic that we’ll reach that point shortly.”
The big hurdle to launching financial advisor LLMs, though, is the need for the advice to be trustworthy and ethical. Professor Lo explains, “In the domain of financial advice, the industry has a term to describe what we need to achieve: fiduciary duty. A ‘fiduciary’ is a legal term that describes an individual who is required to put your interests ahead of theirs when giving you advice. Can we create an LLM that is able to be a fiduciary in the eyes of financial regulators? We hope to make progress on this challenge within the next three to five years, but there's no doubt that this is the last hurdle before AI becomes truly integrated into our financial lives.”
Another way Professor Lo is applying AI to finance is by using AI tools to predict clinical trial outcomes, offering more accurate estimates of a drug candidate’s probability of success to help investors evaluate risk. He explains that, lacking medical expertise, potential investors in biomedical innovations are often scared off because they can’t evaluate the risks of the opportunities in front of them. Using AI, Professor Lo and his collaborators have been able to “provide generalists with a clearer picture of the risk and reward of clinical investments,” which brings more money into the field and fuels critical, life-saving research. Professor Lo says he has received so many inquiries about implementing these ideas commercially that he co-founded QLS Advisors, a firm which “produces these forecasts for a variety of industry partners and biopharma stakeholders.”
In a similar vein, Professor Lo has piloted what he calls “megafund” programs, which combine concepts like portfolio theory, risk management protocols, securitization, and financial simulation to “reduce the risk of early stage drug discovery and improve its returns so that investors will want to put more money to work in this important endeavor.” The idea behind megafunds is pooling multiple potentially risky endeavors together into a portfolio, investing in many experimental compounds simultaneously to increase the chance that a few of the ventures will succeed and generate more than enough profit to make up for those that don’t.
For Professor Lo, this work is more than theoretical. After losing several friends and his mother to cancer, he was frustrated by the fact that people were still dying of a disease “we’ve been studying for decades and know so much about.” Digging into the process of cancer research, he discovered the so-called “Valley of Death,” or the stage of R&D between the conceptualization of a drug and in-person clinical trials where funding is famously difficult to come by. He believes megafunds and ML predictive engines can be combined for greatest effect in this area, adding, “we're poised for yet another dramatic advance when LLMs are applied to journal articles, patent filings, patient medical records, and other text-based information that previously had to be digested by human experts.” His group is also broadening these concepts beyond medicine to fund other risky but important endeavors, like fusion energy.
LOOKING AHEAD: DEMOCRATIZED ACCESS, PERSONALIZATION, AND MORE
Speaking generally, Professor Lo says, “FinTech is at an inflection point, thanks to recent breakthroughs in hardware, software, telecommunications, and AI. In particular, the introduction of large-scale LLMs like ChatGPT has now democratized access to AI so that even non-specialists can benefit from these powerful tools.” To him, this democratization is particularly exciting for applications like fundamental stock analysis, which has until recently been left to human analysists digesting huge volumes of company-specific news. But imagine, for instance, AI-powered investment co-pilots or truly personalized indexes which might allow for broader participation in the market and widespread potential gains. Professor Lo goes on to describe how, “we're entering a new renaissance in FinTech where really sophisticated AI will start being applied to some of the most difficult financial problems, problems that previously required large numbers of highly trained financial analysts. Some areas ripe for this kind of innovation include credit analysis, risk management, algorithmic trading, fundamental analysis, portfolio management, and even financial advice. These advances will make markets more efficient, but they will also bring more capital to be deployed, which in turn will stimulate growth.”
While Professor Lo often wishes that this progress would “happen at a faster pace,” he acknowledges that there is reason to be cautious. Mistakes in finance are often costlier than in other industries, so the rollout of AI tools must be vigilant and deliberate. He explains, “the common practice of letting your customers help debug your software doesn't really work in FinTech.” There is also concern about the “excesses” that might come along with AI-powered growth, like what led to the crisis of 2008. Certainly, any AI tool has the potential to be misused and abused and Professor Lo doesn’t believe we can evade those dangers entirely. But he believes, “the solution is not to avoid using these tools, but to understand how and why they create potential problems and then address those problems directly.” This means crafting legislation, regulation, and guardrails on the AI programs being designed and launched as well as understanding human behavior and motivations. Ultimately, robust research is needed on nascent AI systems and their potential impact in the market. But through the diligent work of scientists like Professor Lo, we can enter “a new age of AI where we can enjoy their talents while protecting ourselves against their dangers.”
Learn more about Professor Lo on his website or CSAIL Page.
Learn about FinTechAI@CSAIL on the CSAIL Alliances website or contact Lori Glover at lglover@mit.edu to get involved.