Image: The Song Yuan xue'an, an 18th-century work widely recognised as one of the most comprehensive studies of Chinese Neo-Confucianism ever written, is converted into colour-coded word embeddings in two-dimensional space.

This presentation examines the impact transformer-based machine learning brings to the historian’s craft beyond familiar but limiting framings. These framings usually oscillate between dismissals of “AI slop” and unrealistic expectations of articles and monographs produced at the click of a button. Moving past this polarized discourse, the presentation asks how, in an era when historical research is conducted primarily through digital mediation, the transformer paradigm reshapes established digital historical practices.

The talk begins with an earlier phase of my work on digitally mediated multiscale exploration, or what I refer to as “digital (re)reading.” This approach introduced data structures into relatively unambiguous historical records and showed how digital environments, when coupled with the careful application of computational methods, can reveal latent structures and long-range connections. At the same time, it emphasized maintaining interpretive agency firmly with the historian.

Building on this foundation, the presentation turns to the ways in which transformer-based models and systems extend and challenge the premises of these earlier approaches by enabling what I term “algorithmic reading.” In addition to accelerating workflows, algorithmic reading enhances the historian’s capacity to engage heterogeneous and semantically complex sources—from literary Sinitic texts to born-digital web archives. It supports a range of tasks, including transcription, extraction, and semantic retrieval, while also opening new historiographical possibilities. In this sense, it expands the range of historical questions that can be pursued across both close and distant modes of inquiry, enabling AI-assisted research that remains both interpretively nuanced and grounded in evidence and context.

The talk concludes by outlining new directions that point toward the emergence of autonomous, agentic AI researchers. It raises a more fundamental question as to how far such systems may further reshape the boundaries of historical practice. More specifically, the talk considers the possibility of agentic AI historians connected to curated knowledge bases, as well as online forums in which such systems autonomously engage in complex and sustained historiographical debates, with or without human historians in the loop. If discussions that once unfolded over decades can be compressed into a matter of months, what might this mean for the future of the historical profession?

Speaker:

Javier Cha is a digital historian and medievalist who uses technology to explore the East Asian world a millennium ago and also aims to prepare the historical profession for an era of data abundance and automation. His research spans the translation of primary sources in classical Chinese, machine-assisted historical methods, and experimental projects that address the challenges posed by big data and artificial intelligence in the humanities.

Cha is currently Assistant Professor of Digital Humanities at the University of Hong Kong. Previously, he was a founding member of the Leiden University Centre for Digital Humanities and taught at the interdisciplinary College of Liberal Studies at Seoul National University. He has secured multiple competitive grants and fellowships from the Hong Kong Research Grants Council, the Academy of Korean Studies, the Korea Foundation, the University of Hong Kong, Seoul National University, and the Social Sciences and Humanities Research Council of Canada. Most recently, he has been selected as one of the inaugural recipients of the Schmidt Sciences Humanities and AI Virtual Institute (HAVI) award for the “Playing Heaven” project, which aims to leverage transformer-based machine learning to develop sound computational methodologies for the intellectual and cultural history of early modern Neo-Confucianism.