AI Reveals Secrets of the Past: Reading Ancient Inscriptions

AI on social media: a computer chip labeled AI. Fot. Igor Omilaev / Unsplash

Hundreds of ancient inscriptions have remained unreadable for thousands of years. Now artificial intelligence is filling in the gaps and adding context. Could AI help historians uncover secrets that have long been beyond human reach?

AI Reads Ancient Inscriptions: How Does Aeneas Work?

Ancient inscriptions — texts carved into stone, metal, wood, or ceramics — offer invaluable insights into legal, economic, social, and even ritual life in antiquity. Unfortunately, the passage of time has damaged or fragmented many inscriptions, making them difficult to interpret accurately and leaving gaps in our knowledge.

Enter AI. A large language model called Aeneas, recently presented in Nature, has been designed not only to interpret but also to provide context for Roman inscriptions.

You might also read: The End of Security: New AI Writes Bomb-Making Instructions

Secrets from Centuries Ago: What Can AI Do?

Aeneas was trained on nearly 200,000 Latin inscriptions, dated from the 7th century BCE to the 8th century CE, covering a broad geographical area from Portugal to Iraq.

Thanks to this dataset, Aeneas can:

  • reconstruct missing characters in inscriptions,
  • estimate the date and place of origin,
  • identify related inscriptions.

What makes Aeneas stand out is its precision: unlike other models, it was trained specifically on Latin data, which makes its predictions more accurate and less prone to hallucinations.

Why Is Aeneas More Accurate Than Other AI Models?

To test its performance, researchers conducted two case studies: Aeneas interpreted a complex inscription using Reflectance Transformation Imaging (RTI), a method that analyzes inscriptions through light behavior.

In the second, it analyzed an altar from Mainz, testing its ability to track changes over time and interpret inscriptions in context.

“In both cases, Aeneas was able to identify relevant epigraphic parallels and generate accurate predictions, formulating and presenting scholarly hypotheses in a quantitative way. These case studies highlight Aeneas’s versatility across different epigraphic contexts. Whether applied to an imperial monument or a provincial votive inscription, Aeneas mirrors the analytical process of an epigrapher, complementing traditional methodologies and producing accurate, meaningful conclusions,” the study notes.

Historians confirmed that Aeneas can integrate smoothly into academic workflows and provide transformative support for historical research.

A Breakthrough in History and Archaeology: What’s Next for Aeneas?

In short, the model performed well in diverse contexts, from imperial monuments to local inscriptions. Researchers highlight that Aeneas is a unique tool trained on a specialized Latin dataset, which makes it less prone to errors and fabrication compared to other AI systems.

For now, Aeneas has been made available to other researchers and students, but its future applications remain open. What is clear is that it represents a powerful new ally for historians.

“In conclusion, Aeneas significantly enhances collaboration between human experts and AI, in a mutually enriching exchange between the sciences and the humanities,” the paper states.


Read the original article: AI odsłania sekrety przeszłości. Będzie odczytywać starożytne inskrypcje

Published by

Patrycja Krzeszowska

Author


A graduate of journalism and social communication at the University of Rzeszów. She has been working in the media since 2019. She has collaborated with newsrooms and copywriting agencies. She has a strong background in psychology, especially cognitive psychology. She is also interested in social issues. She specializes in scientific discoveries and research that have a direct impact on human life.

Want to stay up to date?

Subscribe to our mailing list. We'll send you notifications about new content on our site and podcasts.
You can unsubscribe at any time!

Your subscription could not be saved. Please try again.
Your subscription has been successful.