The European Central Bank said on Thursday that it was exploring ways to use artificial intelligence to better understand inflation and support its oversight of big banks, but stressed that these efforts were still in the early stages.
The central bank is looking into how it can use large language models, similar to ChatGPT, for various purposes, Myriam Moufakkir, the bank’s chief services officer, wrote in a blog post. This includes preparing summaries and briefings that could be used to assist policy and decision-making; making the bank’s public statements easier to understand; and analyzing and comparing documents provided by banks.
The bank already uses machine translations to communicate in many languages with people across the eurozone. The bank is assessing the use of artificial intelligence in nine projects.
“We will continue to investigate the possibilities and challenges of using A.I.,” Ms. Moufakkir wrote. Those examples “are only the tip of the iceberg” of possible uses.
The central bank’s key task is to set interest rates for the 20 countries that use the euro currency, but it also supervises the bloc’s largest banks, employing vast amounts of data. Artificial intelligence provides “new ways for us to collect, clean, analyze and interpret” that information, Ms. Moufakkir wrote.
As an example, A.I. can help automate the time-consuming process of sorting data needed for economic analysis. Insights from A.I. could feed into analysis for monetary policy, but decisions, such as on interest rates, rest “in the hands of humans,” that is, the members of its governing council, the bank later explained.
A.I. can also be used to help the central bank better understand inflation, the blog post said. The bank already gathers real-time data on individual prices for products; it wants to use A.I. to structure all that incoming data and improve the accuracy of bank’s inflation analysis.
These efforts come after the European Central Bank and other central banks were caught off guard by the strength and persistence of recent inflation. As policymakers have raised interest rates rapidly to ease price pressures, they have also reviewed forecasting models and questioned their assumptions about how prices move.
Other central banks are also exploring how to use A.I., sharing knowledge at conferences in recent years and building on the existing use of machine learning. On Friday, the Federal Reserve Bank of New York plans to host a conference on the uses of generative artificial intelligence for economists.
Late last year, the Bank of England said it was using artificial intelligence to analyze large data sets, which could help forecast economic growth, trouble at banks or financial crisis. The British central bank also said it was exploring whether it could use A.I. to analyze news articles and improve economic forecasting, or to create other indicators that track economic trends more quickly than traditional statistics.
Central banks aren’t traditionally on the cutting-edge of advanced technology. But no organization wants to be left behind as A.I. becomes more accessible and governments scramble to regulate it.
Still, central banks are treading lightly as the debate rages over the benefits and risks of the technology.
A.I. can be very useful for central banks in certain areas such as risk management, where there’s a lot of data and relatively simple repeated actions or decisions, said Jon Danielsson, a co-director of the Systemic Risk Center at the London School of Economics. And he expects A.I. to be increasingly used in routine economic analysis.
“The danger for central banks is for macro problems,” such as a financial crisis, Dr. Danielsson said. “The events are very infrequent, crises happen only rarely and crises are unique, which means it is really hard for A.I. to train on past crises.”
“So the risk of A.I. use for macro problems is that it ends up making a catastrophic decision,” he said. Central banks could also be fed misleading advice by A.I. that didn’t fully understand the nature of certain problems, he added.
The European Central Bank is taking a cautious approach toward A.I., and considering data privacy, legal constraints and other ethical issues including transparency and accountability, Ms. Moufakkir said. But the intention is to “accelerate” its adoption so the bank can be “modern and innovative.”
The E.C.B. has also been working with the Bank for International Settlements, a bank for central banks, on how to use large language models in analyzing climate disclosures by companies.