By Gabor Toth and GPT
The Grand Ducal Archive of the Medici, housed in Florence, Italy, stands as a monumental repository of early modern European history. It is one of the most significant collections of its kind, spanning an impressive length of shelf space and containing millions of documents. Generations of dedicated archivists have ensured that this collection is well-preserved, comprehensively inventoried, and remarkably accessible.
Efforts to enhance accessibility have been bolstered by initiatives such as the Medici Archive Project and the Euronews Project. Researchers involved in these endeavors have painstakingly cataloged archival materials and contributed to a comprehensive database accessible through the Medici Archive Online (MIA).
In the wake of these developments, I led a small-scale project last year, now published in the high-impact journal *AI and Society*, exploring how artificial intelligence (AI) could further unlock the treasures of the Grand Ducal Archive.
Harnessing AI: A New Frontier in Archival Exploration
Collaborating with experts from various institutions, we experimented with large language models (LLMs) to create an AI-enabled visual finding aid. Our initial step involved using LLMs to post-process the existing digital catalog data. This process included extracting names of individuals mentioned throughout the archives and identifying the actions they took, as recorded in thousands of documents. We also extracted place names and significant events associated with those locations, enriching the catalog with substantiated keywords.
Beyond data extraction, our team developed a visual platform to aid researchers in navigating this rich information. The platform allows users to browse key individuals and their associated actions, providing a more intuitive way to find relevant documents. A similar system applies to locations and events, offering a comprehensive tool for exploring the archives.
Our experiment demonstrated the potential of AI to increase accessibility to the Medici Archive significantly. With the visual finding aid, users can interact with the data in a more meaningful way, uncovering historical connections that might otherwise remain hidden.
The Challenges and Future of AI in Archival Research
While AI offers remarkable capabilities, it is not without limitations. Simple prompts and computational access allow for vast information extraction, but LLMs are not infallible. They can sometimes produce inaccuracies or "hallucinate," leading to false positives. As a result, while LLMs performed well in archival extraction, the accuracy was not flawless.
Given these limitations, our research remains experimental, and the visual finding aid is not yet available for public use. However, by reading our paper in *AI and Society*, you can explore how we've endeavored to harness LLMs to expand access to the Medici Archive.
Our ongoing exploration of AI in archival research marks an exciting frontier. While challenges remain, these advanced technologies promise to bring new insights to light, enabling historians and researchers to delve deeper into the storied past of the Medici and beyond.
