A recent report from the Bank for International Settlements (BIS) Innovation Hub details Project Gaia, a collaboration between the BIS Innovation Hub Eurosystem Center and certain central banks in Europe. Project Gaia leverages generative artificial intelligence (GenAI) to analyze climate risks in the financial system.
The experiment showed potential in helping banks navigate the new reality of climate change by making it easier and faster to assess the risks that climate change poses to their business. The recent BIS report notes that this capability can extend beyond climate risk analysis in banking, offering potential applications in suptech and regtech.
Project Gaia used large language models (LLMs) to automatically extract climate-related data from publicly available corporate reports. This data could pertain to a company's carbon emissions, green bond issuance, and net-zero commitments. The Gaia architecture consists of structured and unstructured data storage, data processing pipelines, LLM integration and a user interface all on top of a cloud platform. The solution connects to an external LLM service via an application programming interface (API). The report explains a set of concrete design choices that allow the Gaia Proof-of-Concept (POC) to overcome technical challenges such as LLMs’ long response times, randomness (non-repeatability) in their responses, and hallucinations. These design choices are result of a thorough optimization process, which has entailed over 50,000 LLM queries to date.
The report describes the key steps of KPI extraction using LLMs, along with the design choices related to each step. The report notes that Project Gaia has proven the feasibility of climate-related financial risk analysis using AI and LLM. By tapping into public information readily available today, Gaia provides an efficient and reliable source of climate risk-related data. Furthermore, Gaia offers harmonized metrics despite the heterogeneity of naming and definitions across different jurisdictions and companies. For example, Gaia relies on each KPI’s definition rather than its name when searching corporate reports, thus it does not require harmonization of guidelines or regulatory requirements. This is crucial in instances where slightly different wording and definitions prevail for similar concepts (for example, Scope 3 emissions are sometimes referenced as financed emissions or indirect emissions). Thus, Gaia promises to overcome potential differences in official disclosure frameworks and offers much needed transparency and comparability of climate-related information.
The report notes that the significance of Gaia goes beyond climate data analysis, as the Gaia platform can be easily configured to extract new types of KPIs. Gaia offers a generic solution to extract desired KPIs contained in a predefined set of PDF reports. Within the financial sector, AI-based KPI extraction from large bodies of textual documents is being touted as a “game changer,” for example, in regulatory and supervisory use cases. As per the report, prospectuses for financial instruments is another area in which information is dispersed in the form of unstructured texts that can be leveraged. The Gaia PoC demonstrates the power of chowreating AI-enabled intelligent tools to automate existing workflows.
Related links:
LEARN MORE
Innovating with purpose
Moody’s is incorporating cutting-edge technologies, such as artificial intelligence, to help banks meet their existing challenges more effectively.