Groundbreaking research from the Bank for International Settlements (BIS) has demonstrated that generative artificial intelligence (AI) agents can perform critical liquidity management functions in central banks and high-value payment systems traditionally managed by humans.
This study was conducted using ChatGPT’s o1 inference model in agent mode to simulate real-world scenarios. AI needed to balance liquidity costs and risks Delays in multi-million dollar transactions.
In this experiment, we designed three scenarios that replicate real-world challenges in RTGS or real-time payment systems (e.g., Fedwire, TARGET2, Lynx), which are central to traditional financial systems.
In the first scenario, AI only had $10 of liquidity and two pending payments of $1 each. Faced with the possibility of a $10 emergency order, he decided to freeze everything. His own explanation clarifies why he made this decision. “We are now delaying small payments to preserve liquidity and allow us to respond in the event of an emergency transaction.”
In the second scenario, the probability of receiving external funding (90%) is more complex; Make an emergency payment (50%). In this case, the AI processed only low-risk transactions and demonstrated dynamic prioritization capabilities.
Tests showed that the AI maintained a proactive approach even when the probability varied from 50% to 0.1% and the scale reached billions of dollars. However, in complex situations there was a slight decrease in consistency and occasional changes in decisions.
AI is already a better financial manager than most humans, says BIS
the study Proposing the development of an “AI assistant” for routine tasksthe human role is reserved for oversight and strategic decisions. Researchers predict that similar systems could be tested in a regulatory sandbox environment before actual implementation.
“The results suggest that certain AI solutions have the potential to reduce operational costs and improve operational efficiency and safety,” the BIS report said. But he warns of limitations. Models rely on historical data and can fail in the face of extreme events or “black swans” beyond the experience they were trained for.
This study compares this approach with traditional reinforcement learning. The authors highlight that unlike traditional reinforcement learning (which requires thousands of simulations), generative AI achieved “excellent results without any specific training.”
Therefore, with this level of effectiveness, the report’s authors believe that AI Potentially saves millions of dollars in tied-up liquidity Significantly reduce payment queues in RTGS systems.
Although the BIS report focuses on the traditional financial system, its findings are not surprising in the world of digital assets. This is because decentralized finance (DeFi) applications already exist. They have been managing liquidity for years. It’s 100% automatic with AMM pools, flash loans, and algorithms that rebalance in seconds.
As CriptoNoticias reports, Uniswap, Aave, and Curve are already making billions of dollars of work on what BIS is hailing as an innovation since 2020.
(Tag Translate) Banking and Insurance
