Google DeepMind is now using AI to improve understanding of ancient Roman history

Scientists at U.K.-based DeepMind, the Google LLC laboratory that specializes in advanced artificial intelligence, have developed a solution to decipher ancient Roman texts that are often damaged and difficult to interpret.

The solution is Aeneas, an AI software model named after a mythical Greek and Roman hero. There are currently a plethora of ancient objects bearing inscriptions — many of the words beaten up after centuries of wear and tear, many fragmented or defaced. Aeneas is now being used to fill in the missing pieces and build on contextual clues to transcribe the full text and hopefully give it a place in time.

It’s believed about 1,500 such objects are discovered each year, whether decrees from emperors or perhaps the epitaphs of slaves. Given the Roman empire spanned two millennia and thousands of square miles, accurately understanding what the object means and when the text was written is a huge challenge to experts in epigraphy — the study of ancient inscriptions.

In this regard, Aeneas is now being hailed as “groundbreaking.” DeepMind researchers trained the AI on some of the world’s largest Latin epigraphy databases, going through millions of inscriptions, a vast collection a human eye could never hope to assess and compare among each other. To understand inscriptions, historians usually rely on these “parallels.”

A paper in the Nature journal explained human experts have heretofore relied on “laborious manual searches or string-matching techniques,” but since it can be so time-consuming, they usually have to focus on regional and chronological specializations. DeepMind says the software can now be fed an inscription and it takes just a matter of seconds before the AI finds a parallel and can speculate meaning or time.

“Aeneas greatly accelerates this complex and time-consuming work,” explained DeepMind. “It reasons across thousands of Latin inscriptions, retrieving textual and contextual parallels in seconds that allow historians to interpret and build upon the model’s findings.”

Putting the AI to the test, researchers gave it a Latin text etched in stone whose origin has for years been contested. The text, “Res Gestae Divi Augusti, Emperor Augustus,” is a first-person account of the emperor’s achievements, but experts have never been sure when it was written, whether before or after the emperor died in AD 14.

Aeneas got to work and was able to make relevant parallels from various imperial legal texts. “Aeneas produced a detailed distribution of possible dates, showing two distinct peaks, with one smaller peak around 10-1 BCE and a larger, more confident peak between 10-20 CE,” DeepMind said.

In a further test, the AI analyzed inscriptions on a votive altar discovered in today’s Germany. It assessed subtle linguistic parallels with other ancient inscriptions and was able to compare it with another votive altar in the region.

“Those were jaw-dropping moments for us,” said a historian who worked with the researchers to develop Aeneas.

Photo: Wiki