AI and Byzantine Studies: Structures of Knowledge at Scale

The central methodological goal of the ERC project RELEVEN (2021–26) has been to represent data about the history of the medieval Near East, including Byzantine history, in a way that meets the needs of the scholars who study that history – this means encoding the conflicting accounts of events, room for doubt concerning facts, and different interpretations of surviving texts in a way that brings the contingency of our knowledge directly into the data structure, rather than leaving it as an exercise for the user. While the need for such a data structure is clear, its unavoidably increased complexity compared to ‘normal’ data collection practices means that collection of data into RELEVEN-style structures is time-consuming and easy to misunderstand; as a result, the feasibility of our model’s use remains an open question. Fortunately for our scholarly community, technology has not stood still. As unfeasible as a reliance on machine transcription of medieval manuscripts was fifteen years ago, it is almost commonplace today. Similarly, the development of machine learning methods and language transformer technology that have resulted in today’s GPT models may be just in time to come to the aid of desperate Byzantinists with fiendishly complex needs for the representation of their source material as tractable and analysable data. In the round table discussion, I will focus on experiments with knowledge extraction from source texts, its ‘trustworthiness’, and the effect that these affordances might have on how we do Byzantine studies in future.