Uncovering Ancient History with AI
Merton Tutor in Ancient History Professor Jonathan Prag is part of an ongoing project utilising AI in the field of Ancient History.
It is the role of Ancient Historians to try to work out what is missing in ancient stone inscriptions. Inscriptions have become damaged over time so there are gaps in the text. Uncovering the truth of the full text before it was damaged typically relies on years and years of experience studying these texts. Digital collections can speed things up somewhat, but this is still limited in its use.
The AI project began when one of Professor Prag’s Doctoral students was studying Ancient Sicily. They were in conversation with a Computer Science student who was also working with Google’s Deep Mind. They approached Professor Prag with the idea of using artificial intelligence (AI) to help ancient historians fill in the gaps in ancient Greek inscriptions on stone and other materials.
The resulting AI programme is Ithaca, a deep neural network for the textual restoration and geographical and chronological attribution of ancient Greek inscriptions.
Ithaca is also capable of attributing inscriptions to their original location with an accuracy of 71% and can date them to within 30 years of their likely original date. In one test, Ithaca was put to work on a series of important Athenian decrees, traditionally dated by historians to before 446/445 BCE. New evidence recently presented by historians suggests the 420s BCE as a more appropriate time period. Remarkably, Ithaca’s average predicted date for the decrees is 421 BCE, aligning closely with the new interpretation and demonstrating how machine learning might contribute to historical debates.
The project is an excellent illustration of the sort of collaboration between computer scientists (in this case Google’s DeepMind) and humanities scholars that is becoming increasingly common: it was led by Dr Thea Sommerschield (Marie Curie Fellow at Ca' Foscari University of Venice, and a former doctoral student of Professor Prag) and Dr Yannis Assael (Staff Research Scientist, DeepMind).
Professor Prag commented:
“The huge quantity of evidence from the ancient world, whether texts or objects, keeps on growing, and is increasingly beyond the scope of individual historians to master, even as we work to make sense of it and to make it more accessible. The application of AI to this data, as Ithaca demonstrates, presents incredible opportunities – ancient history has an exciting future.”
The team have found that the odds of Ithaca having the right answer are extremely high. It works best as an assistive tool for Historians, not as a replacement. Human judgements are still crucial in the process of restoring texts and it isn’t suggested that the model replaces a scholar, however Ithaca is able to process vast amounts of data very quickly in a way that humans never could.
Ithaca has faced some resistance from Academics in the field who are untrusting of this technology. It also sits alongside a broader issue in the humanities of what should be considered open data (and the quality of the available data). Datasets are commonly not shared in humanities publications in the same way that they are in scientific publications. However, a platform such as Ithaca relies on shared, open data, so that such methods can be reproducible. Academic culture has become one of the biggest challenges associated with this project.
After beginning with Ancient Greek texts, the team are now trying to apply the model to Latin inscriptions. Latin, however, has a lot of formulaic abbreviations. This presents a different challenge as the model must learn how the language has also been concealed in abbreviations.
Another strand of the project has developed as the team now tries to understand why sometimes the model misidentifies a pattern. Dr Sommerschield is now beginning a Leverhulme Early Career Fellowship at the University of Nottingham, aiming to use the connections implied by these mistakes to understand ancient Greek networks. One of the challenges with AI is understanding how it reaches its results, and that is particularly relevant for historians, since the connections in the data may offer clues for historical causation.