Could AI Cause a Global Financial Meltdown?
The adoption of AI by financial institutions could lead to algorithmic biases and regulatory compliance challenges, among other risks.
Innovation in artificial intelligence is unfolding rapidly, with the potential to revolutionize the financial services industry as we know it.
With the breakneck speed of development continuing apace amid a period of economic volatility -- most recently underscored by the collapse of banks in the United States and the rescue of major financial institutions abroad -- there are growing concerns over the role AI could play in further destabilization.
In May, Securities and Exchange Commission Chair Gary Gensler cautioned AI platforms could be central components of future “fragility” in the financial system and called for scrutiny of the use of Generative AI by financial institutions, lest it pose a “systemic risk” in the future.
“The risks and challenges associated with these emerging technologies are undoubtedly growing, posing concerns among regulators for their impact on existing financial systems,” explains Dennis Gada, executive vice president, global head of banking and financial services, at Infosys. “However, there are also several benefits to keep in mind.”
For example, financial institutions will gain efficiency and competitiveness, insights for better decisions, and improve client experience that will help them stay ahead in the game if they learn how to leverage AI.
“AI is quickly becoming an important technology in day-to-day work across the financial services sector, leading to higher productivity and faster delivery,” he adds.
AI and the Risk of Unknown Unknowns
Steve Sanders, chief information security officer for CSI, says the biggest challenge and resulting risk are both a result of the lack of insight into the decision-making of AI systems.
“Even when given well-defined parameters, AI systems can still generate unexpected responses,” he explains.
While one might think the financial system is based on facts and hard data, the latest economic challenges, which include the Federal Reserve’s efforts to prevent a credit crunch and fight high inflation, are one illustration of the complexity involved.
“The inputs and outputs of our financial system include an incredible number of situational scenarios and previously unencountered circumstances, which results in a very difficult training scenario for AI systems,” he says.
On January 26, 2023, NIST released the AI Risk Management Framework v1.0, and subsequently launched the Trustworthy and Responsible AI Resource Center on March 30, 2023, to support this framework.
“Experts agree that even if this voluntary framework were implemented, the opaque nature of closed systems will remain a challenge,” Sanders says.
He adds it is almost certain that AI algorithms will continue to become more complex as they further emulate human thinking.
“Even without the opaqueness, this creates a very difficult scenario for risk management,” he says. “AI systems will also continue to become more opaque unless transparency is forced through new regulations or laws.”
AI Finance Use Poses Challenges for Regulators
While AI applications for financial services raise challenges for regulators due to the unknown risks, Gada notes the same can be said for almost every industry applying this disruptive technology.
“Given the risks of using AI in financial services are more technical than those of other industries, there is heightened scrutiny among governing bodies, due to the sensitive nature of data, privacy and room for fraud,” he explains.
To circumvent these risks, regulators can implement AI auditing processes to provide standards, practical codes and guardrails for safe AI use in financial services.
He notes there are both upsides and downsides to regulating AI, and while regulation provides a framework for ethical practices, information security and consumer rights, it also poses a threat to the progression and innovation of this groundbreaking technology.
“Balancing regulation and innovation will be the best path forward to ensure safety, responsibility and value in deploying AI,” Gada says.
Sanders adds systems should be designed with a governance-first mindset, otherwise gaining control over the risks is going to be difficult.