Sustainability Initiative: Bank for International Settlements

Artificial intelligence (AI) partnership with central banks mines for climate data with wider potential supervisory applications

Climate change is causing worldwide financial risk, the true extent of which is unknown.

Project Gaia is an AI application developed by the Bank for International Settlements (BIS) and its project partners, the Bank of Spain, whose previous work it builds on, the Deutsche Bundesbank and the European Central Bank (ECB).

There are currently no global financial climate reporting standards, but central banks, supervisory authorities and financial institutions need higher-quality and more accessible data to model financial stability risks.

To help solve this, Project Gaia unearths climate-related data that is buried in financial institutions’ corporate reports, combining information contained in texts, tables, footnotes and figures pertaining to a single firm from multiple sources.

The BIS published the results of phase one of the initiative in March. ECB president Christine Lagarde says that, as Project Gaia “uses generative AI to decipher vast unstructured datasets”, it “has the potential to be a powerful tool” in assessing the economic effects of the climate crisis.

Phase one involved testing the application by inputting more than 2,000 documents from 187 financial institutions throughout 2018–22. The technology then used the data to create 20 key performance indicators (KPIs).

Raphael Auer, head of the BIS Innovation Hub Eurosystem Centre, says Gaia automates manual processes “that used to take weeks of tedious work into an easy-to-use IT application performing the same task in hours”.

Speaking to Central Banking, Cecilia Skingsley, head of the BIS Innovation Hub, highlights the benefits to central banks, supervisors and financial institutions. “Project Gaia demonstrates a use case for generative AI that improves the resilience of the financial system and contributes to the public good. It allows analysts to automate the manual work of extracting facts from corporate environmental, social and governance reports, turning weeks of tedious file-shuffling and reading through long tables into hours.” 

Gaia addresses the often-warranted scepticism towards generative AI by offering transparency: “Every data point comes with a direct link to the origin page and an explanation of how it was retrieved. Analysts are always seconds away from verifying the source with their own eyes.” A possible next step for Gaia could be to onboard more models, which would increase robustness and reduce the lock-in to a single AI vendor.

“In today’s data-driven world, the challenge isn’t just gathering information, it’s making sense of it,” Joachim Nagel, president of Deutsche Bundesbank, says. “By leveraging large language models [LLMs], Gaia empowers central banks to better navigate the complexities of sustainability reporting” and transforms “unstructured data into actionable insights”.

Gaia is also capable of extracting KPIs from non-English documents because the underlying LLM is agnostic to most major languages.

It is “an important breakthrough on the road to the understanding and adaptation of AI into the innovation culture of the eurosystem”, says Pablo Hernández de Cos, governor of the Bank of Spain.

The BIS Innovation Hub’s Eurosystem Centre and its project partners are sharing this gained knowledge with the central banking community.

Skingsley says central banks worldwide are exploring how the technology underlying Project Gaia – which is experimental in nature, similar to other Innovation Hub projects – could be used to examine a variety of datasets and employ it for other use cases.