In a recent report by the Bank for International Settlements (BIS), central banks worldwide have been cautioned about the inherent risks associated with their growing reliance on artificial intelligence (AI) tools. This nine-page report titled “Artificial Intelligence in Central Banking” sheds light on the use cases and potential pitfalls of integrating large language models (LLMs) into their operations.
Central banks, often considered early adopters of AI technology, have increasingly turned to AI models for a variety of purposes. These AI models have been instrumental in transforming information collection and processing, rendering many traditional human efforts nearly obsolete. AI models are now deployed for data sampling, cleaning, and matching, helping central banks streamline their operations with proven machine learning techniques.
One significant application of AI in central banking is in making informed decisions on monetary policy. By utilizing neural networks and random forest models, central banks gain access to real-time data for assessing inflation expectations and evaluating the effectiveness of their monetary policies. This data is sourced not only from traditional financial channels but also from the vast array of information available on social media platforms.
Extracting insights from vast data
As the report highlights, sifting through an enormous volume of data can be time-consuming and challenging, making AI models invaluable in extracting relevant insights efficiently. Central banks employ language models to summarize complex financial reports, track economic trends, and interpret interviews with business leaders and market experts. Moreover, these central bank language models (CB-LMs) have demonstrated an ability to predict market reactions to monetary policy announcements.
AI systems have also proven their proficiency in overseeing and supervising payment systems. They excel in identifying irregular financial transactions, a critical factor in combating money laundering and preventing cyberattacks. The Central Bank of Brazil, for instance, has recently introduced ADAM, a classification model designed to predict borrowers likely to default on their loans, enhancing their ability to mitigate financial risks.
Beyond traditional applications, central banks are also turning to AI systems to predict consumer behavior in response to the introduction of central bank digital currencies (CBDCs) and other financial developments. These predictive models provide valuable insights for policymakers to anticipate market reactions and adapt accordingly.
Challenges and risks of AI integration
While the benefits of AI integration are evident, the report underscores the challenges and risks that central banks face. One of the primary concerns is the potential for biased outputs stemming from the data used to train AI models. Furthermore, generative AI models, though powerful, require vigilant human supervision to reduce the likelihood of errors and inaccuracies.
In the short term, central banks will need to invest in equipping their staff with new AI skill sets to integrate AI systems effectively into their workflows. However, they are likely to encounter stiff competition from private financial firms in the recruitment of employees with advanced AI expertise. The salary disparity between public institutions and the private sector further intensifies this competition.
