The Dual Edge of Artificial Intelligence in Finance: Opportunities and Threats

Artificial intelligence (AI) is poised to revolutionize the finance industry, offering innovative solutions and efficiencies. However, as Gary Gensler, the chair of the U.S. Securities and Exchange Commission (SEC), points out, without adequate oversight, this transformative technology could lead to significant risks for investors.

Addressing Oversight Concerns in AI

This past summer, Gensler took a pioneering step by proposing regulatory measures for AI. His focus is on addressing conflict-of-interest concerns stemming from financial firms using predictive data analytics. Recently, he has reiterated the importance of minimizing AI-related risks to protect investors.

The Dangers of Dependency on Unified AI Models

In a conversation with The New York Times, Gensler highlighted the dangers of the U.S. relying on a limited number of foundational AI models. This reliance could replicate the conditions for a financial crisis, especially given the interdependencies across the economic landscape and the phenomenon of “herding,” where individuals respond similarly based on shared information.

A Transformative Yet Potentially Perilous Technology

During an interview with Bloomberg, Gensler referred to AI as “the most transformative technology of this generation,” while cautioning about its potential to provoke crises in 2028 or 2032.

Academic Insights: MIT’s Deep Learning Research

While serving as a professor at MIT Sloan, Gensler examined these issues in detail in a pivotal 2020 paper, “Deep Learning and Financial Stability,” co-authored with then-research assistant Lily Bailey, who currently serves as a special assistant at the SEC. This paper highlights five pathways through which widespread adoption of deep learning could exacerbate financial instability.

Data Dependency as a Vulnerability

Gensler and Bailey pointed out that significant data consolidation across various sectors could pose risks. For instance, platforms like Google Maps and Waze dominate traffic data, creating a scenario of uniformity that can lead to dangerous outcomes.

According to the authors, “Models built on the same datasets are likely to generate highly correlated predictions that proceed in lockstep, causing crowding and herding.”

The Risks of Model Design

The design of AI models can lead to systemic risks comparable to those witnessed during the 2008 financial crisis. The authors draw attention to the overreliance on credit agencies like Standard & Poor’s, Moody’s, and Fitch in underwriting complex financial products.

Furthermore, the unique attributes of deep learning models can make them more sensitive, resulting in a higher likelihood of black swan events. When different models coordinate to optimize outcomes, they may execute identical strategies, amplifying market volatility.

The Regulatory Gap in Deep Learning

Gensler and Bailey argue that existing financial regulations will likely fall short in tackling the challenges introduced by deep learning. Variability in the adoption of deep learning across financial entities poses challenges, with fintech startups often adopting technologies faster than their larger regulated counterparts.

Tiered Adoption Leading to Regulatory Arbitrage

This uneven pace of adoption may result in regulatory arbitrage, wherein certain financial activities could shift towards less regulated entities, raising new concerns about oversight and accountability.

Algorithmic Coordination: A Hidden Threat

One of the critical concerns raised by the authors is algorithmic coordination leading to increased interconnections among firms. If different financial institutions’ models communicate and align, it heightens the likelihood of herding behavior altogether.

Moreover, regulatory mechanisms designed to detect algorithmic coordination might only reveal these interactions after harmful outcomes have occurred. The inherent complexity of deep learning models makes it challenging for regulators to identify and manage such coordination proactively.

The User Interface Challenge in Finance

AI has found widespread use in user interface applications, including automated advice and recommendations for investing, lending, and insurance.

Gensler and Bailey warn that the standardization of virtual assistant technologies—like chatbots providing investment recommendations—could foster herding behavior among clients, potentially across entire market sectors. Many prominent financial institutions have invested significantly in their proprietary assistants, further compounding this issue.

Conclusion: The Path Forward

Despite the vast potential of AI to enhance client services, the technology can also obscure accountability during crises, as noted by the authors. To safeguard investors, a comprehensive understanding of AI, along with robust regulatory frameworks, is indispensable.

Read the full research: “Deep Learning and Financial Stability”