AI-Driven Financial Chaos: A New Threat to Market Stability

Financial stability in the age of artificial intelligence (AI) may depend on more than just the underlying economic situation or riskiness of strategies. According to a new study by economists at the European Central Bank, different types of AI architecture can lead to vastly different outcomes.

The researchers simulated a scenario where mutual fund redemptions resembled a classic bank run. Instead of human agents making decisions, they used two types of AI: Q-learning and large language models (LLMs).

Q-learning AI agents learn from experience and tend to act in unison, resulting in either everyone staying invested or rushing to redeem mutual fund investments. This can lead to a bank run-style panic even if the fundamentals don't justify it.

On the other hand, LLM agents reason like textbook rational investors, but their differing interpretations of the situation cause them to fail to coordinate. This leads to unpredictable partial withdrawals from mutual funds.

The study's findings have significant implications for policymakers, who need to understand how different AI architectures affect investment behavior. The authors recommend that regulators should be aware of the types of AI tools used in trading rooms and bolster traditional risk management tools to respond to the speed and scale at which AI agents operate.