Historians have long clashed over when “Res Gestae Divi Augusti,” a monumental Latin text, was first etched in stone. The first-person inscription gave a lengthy account of the life and accomplishments of Rome’s first emperor. But was it written before or after Augustus, at age 75, died in A.D. 14? Some experts have put its origin as decades earlier.
Known in English as “Deeds of the Divine Augustus,” the text is an early example of autocratic image-burnishing. The precise date of its public debut is seen as important by historians because the emperor’s reign marked the transition of Rome from a republic to a dictatorship that lasted centuries.
Artificial intelligence is now weighing in. A model written by DeepMind, a Google company based in London, cites a wealth of evidence to claim that the text originated around A.D. 15, or shortly after Augustus’s death.
A report on the new A.I. model appeared in the journal Nature on Wednesday. It makes the case that the computer program can more generally help historians link isolated bits and pieces of past information to their socially complicated settings, helping scholars create the detailed narratives and story lines known as history.
The study’s authors call the process contextualization. The new A.I. model, known as Aeneas, after a hero of Greco-Roman mythology, specializes in identifying the social context of Latin inscriptions.
“Studying history through inscriptions is like solving a gigantic jigsaw puzzle,” Thea Sommerschield, one of the researchers, told reporters Monday in a DeepMind news briefing. A single isolated piece, she added, no matter how detailed its description, cannot help historians solve the overall puzzle of how, when and where it fits into a social context.
“You need to use that information,” Dr. Sommerschield said, “to find the pieces that connect to it.”
In the Nature paper, the authors note that roughly 1,500 new Latin inscriptions come to light every year, making the new A.I. model a potentially valuable tool for helping historians to better illuminate the past.
In an accompanying commentary in Nature, Charlotte Tupman, a classicist at the University of Exeter in England, called the A.I. model “a groundbreaking research tool” that will let scholars “identify connections in their data that could be overlooked or time-consuming to unearth.”
The DeepMind researchers trained Aeneas on a vast body of ancient inscriptions. They used the combined information from three of the world’s most extensive Latin epigraphy databases: the Epigraphic Database Roma, the Epigraphic Database Heidelberg and the Epigraphik-Datenbank Clauss-Slaby. The third of those archives, based in Germany, holds information on more than a half-million inscriptions.
The model could then analyze a particular text and link it to similar examples in that body of information. The final readings consist of sets of probabilities on the text’s likely age and site of its geographic origin, and make predictions for likely candidates to fill in an inscription’s missing parts.
The Google scientists also surveyed 23 epigraphers — specialists who study and interpret ancient inscriptions. The A.I. model aided the vast majority of them in locating starting points for their research as well as improving confidence in their subsequent findings.
To study the Augustan text, the DeepMind researchers linked it to subtle linguistic features and historical markers. For instance, the model found close parallels in a proclamation by the Roman Senate in 19 A.D. that honored an heir of the emperor’s dynasty.
With the model’s unveiling, DeepMind says it is making an interactive version of Aeneas freely available to researchers, students, educators and museum professionals at the website predictingthepast.com. To spur further research, it says it is also making the computer code of the Aeneas model public.