Central banks are increasingly exploring and using generative artificial intelligence (generative AI) to support their activities, such as economic and monetary analysis, statistical production, financial supervision, and payment oversight. Large language models (LLMs) in particular – a class of generative AI that can analyze and produce text by mimicking human intelligence – have unlocked unprecedented opportunities to deal with textual data, of which central banks are both heavy users and producers.
Continuous and rapid innovations are further expanding the capabilities and quality of generative AI’s outputs. First, methods such as retrieval-augmented generation (RAG) can improve the accuracy and reliability of LLM-produced content, for example by drawing on verified sources and specialized knowledge. Second, a growing area of interest is the use of small language models (SLMs), which are more limited and cheaper versions of LLMs tailored to specific tasks. Third, and more generally, agentic AI systems can enhance task automation and assist decision-making in business processes, for example by orchestrating complex workflows.
With the proliferation of new data solutions and competing work priorities, one question for central banks is how to make the most of these promising yet often embryonic AI advances in an efficient, effective, ethical, and safe way. Experience so far has highlighted the importance of adequately managing innovation, not least to balance its benefits with the associated risks. Three distinct areas of focus have emerged:
- Governance Frameworks: The first relates to the design and implementation of proper governance frameworks for managing the associated risks and harnessing AI effectively and responsibly. Central banks have already taken significant steps in this endeavor, leveraging their long-standing expertise in data issues. Yet AI governance initiatives are still evolving, often reflecting the need for new skills and organizational strategies to support organization-wide innovation.
- Technical Aspects: The second area of focus is on technical aspects, reflecting the need to secure adequate IT resources as well as to adopt interoperable data processes and systems to further advance the use of generative AI both within and across organizations. Open software tools and well-established information standards, such as the Statistical Data and Metadata eXchange (SDMX) initiative, can play a decisive role in this regard. They can ensure that data are used properly by AI and, more broadly, preserve the quality of the reference information produced by authoritative sources, such as central banks.
- Collaboration and Knowledge Exchange: Finally, promising yet often fast-paced technological advancements serve as a good reminder of the importance of cooperation and knowledge exchange across central banks as well as with other stakeholders involved in the data ecosystem. Such collaborative initiatives can yield significant benefits, including sharing best practices and raising awareness of the challenges posed by AI. They also help to optimize the use of resources in the central banking community by facilitating opportunities for co-investment in areas such as data-sharing techniques or open-source software to make data AI-ready.