Historian leads project to design new AI architecture for humanities research

Artificial intelligence might be the wave of the future, but Stanford historian Giovanna Ceserani hopes to use it to better understand the past.

Ceserani is leading an interdisciplinary project to create an AI framework that can help interpret complex historical materials, containing not only text written at a specific time, but also other handwritten marks, drawings, and graphical notations made over time, even centuries. Interpreting these types of materials and making connections among other sources requires careful evaluation—something currently beyond the ability of large language models. 

“We want AI to read sources the way historians do,” said Ceserani, professor of classics in the School of Humanities and Sciences. “Treating archival documents as ‘deep sources’ means attending to not just the words and the characters on the page, but also the many layers of meaning that a source carries.”

The project brings together a multidisciplinary team of historians, physicists, computer scientists, developers, and designers to create the AI architecture. The goal is to address some of the problems of using LLMs, which are a bit of a black box, by designing a new architecture based on scholarly practices of transparency and accountability. This means a historian working with the new AI framework will be able to trace the origin of the evidence and interpretations for each step taken in the research process.   

The effort, named SETS, is based on set theory, the branch of mathematics that deals with collections of objects. The overlapping circles of Venn diagrams are a familiar application of set theory. Using the theory in the context of reading thousands of subtle handwritten marks on historical documents requires something much more complex. But if successful, the SETS approach will greatly accelerate and expand historical analysis and, with it, our understanding of the past. 

Ceserani is collaborating with co-investigator Sebastian Ahnert from the University of Cambridge in the U.K., who is leading architecture development, and co-investigator Michele Mauri of Politecnico of Milan, Italy, who will lead the interface design.

One of the potential benefits of this new AI system would be the ability to find new connections among sources. For instance, if a historian discovers the meaning of a certain pencil mark on one document, SETS could search through hundreds of previous archival pages for similar marks and apply the same reasoning. 

“It could help us make new discoveries that we might not otherwise find,” Ceserani said. 

The system would also make it possible for historians to easily pick up where another has left off. Part of the plan for this new AI architecture is meticulous tracking of all the steps it takes to arrive at a particular reading or connection.

After the initial development stage, the team plans to test the SETS approach on three case studies involving the work of Stanford historians. These include research into the governmental records of slave liberations in 19th-century Senegal, led by Richard Roberts, professor of history, emeritus; investigations of probate inventories in the18th-century Ottoman empire, led by Ali Yaycıoğlu, associate professor of history; and Ceserani’s investigations into the records of arrivals and departures in 18th-century Venice.

Each of the case studies attempts to understand human journeys across geography and generations. The sources they access also contain highly complex networks involving family background and relationships, finances, and employment as well as travel origins and destinations.

If the researchers can show that the SETS approach works with these case studies, they believe the AI system can be applied not only to other historical work, but also to any other research involving interconnected, layered data sets in the humanities and even in the sciences.

The SETS project had its genesis at Stanford with a Propel Grant from the Office of Vice Provost and Dean of Research. Then in December 2025, it received two years of support from Schmidt Sciences, a philanthropic organization, which granted a total of $11 million to 23 projects that bridge AI and the humanities.

“I was so glad to see this full initiative from Schmidt Sciences,” Ceserani said. “It really shows the value of investing in research on the relationship between AI and the humanities, and in a very deliberate way, it looked for projects that benefits both fields.”

Acknowledgments 

The SETS team also includes: Diyi Yang, Stanford associate professor of computer science; Nicole Coleman, the former research director of the Humanities + Design lab at Stanford’s Center for Spatial and Textual Analysis; and Allen Romano, a former Stanford doctoral student.