Napoleon ordered a war missive to be written in code, creating a riddle unsolved for over 200 years, until AI deciphered it within hours. To decipher correspondences or bring to light ancient carbonized papyri, artificial intelligence (AI) is opening new avenues for historical research.
Carter Church, a 27-year-old artificial intelligence engineer, deciphered a 19th century letter addressed to one of Napoleon’s generals. The secret message from 1809, detailing the positions of French and allied troops, had been published in an online university archive.
Without digital tools, the task could have taken weeks, but an AI model developed by OpenAI, the creator of ChatGPT, completed it in just six hours, the engineer explains. “What AI mainly brings is perseverance,” he told AFP.
Long condemned to silence, the papyri of Herculaneum, charred by the eruption of Vesuvius in the year 79, are also beginning to reveal their secrets. Too fragile to be unrolled without risking being destroyed, the scrolls are scanned with X-rays then analyzed by AI, which reveals the writing and allows the contents to be read.
In June, researchers announced the first complete virtual unfolding of a papyrus held at the National Library of Naples. The revealed text is a “Stoic treatise devoted to ethics”, explains to AFP Federica Nicolardi, associate professor of papyrology at the University of Naples.
With the “Back in time” project, in France the National Institute for Research in Digital Sciences and Technologies (Inria) brings together historians, cryptographers and AI specialists to develop a tool specifically capable of transcribing and deciphering old encrypted diplomatic or military correspondence.
Thousands of documents encrypted to prevent a message intercepted by the enemy from being understood, composed of triangles, circles or invented signs, are lying dormant in European archives.
“There are entire centuries for which we are completely in the clouds,” explains Cécile Pierrot, cryptographer at Inria and scientific manager of the project, to AFP.
The researchers want to further improve the performance of the model, but with this technological advance, Cécile Pierrot hopes to provide “knowledge, a much broader perspective on history”.
“Facebook of the past”
In historical sciences, artificial intelligence is causing another upheaval: that of scale. The Midrash project, led by the École Pratique des Hautes Études-PSL, uses AI to transcribe millions of pages of manuscripts and thus constitute immense searchable corpora on Jewish culture.
“We transcribed 10 million pages of Hebrew manuscripts. It’s gigantic. There is no human being who could process such a quantity,” says Daniel Stökl Ben Ezra, director of studies in historical sciences at the École Pratique des Hautes Études in Paris.
The texts thus digitized feed giant databases allowing searches to be carried out in previously inaccessible corpora. The researcher compares this set to a sort of “huge Facebook of the past”. Historians can reveal connections between people, places, texts or ideas scattered across millions of documents. Thus, fragments of the Cairo Genizah, a set of Jewish manuscripts scattered around the world, could be brought together.
A methodological revolution which also affects the archives of Notre-Dame de Paris: as part of the e-NDP project, a team of researchers trained an AI to transcribe 14,600 pages in Latin of medieval registers from the cathedral chapter.
With a character recognition rate of between 88% and 94% depending on the volume, this work has notably brought to light information on the daily life of the cathedral, its work, its staff and even the care of abandoned children, explains historian Julie Claustre, who is leading the project.
Carrying out the same work manually would have taken “several decades,” she estimates. But “it is not just a saving of time”, on such corpora, automation “makes possible surveys which were not possible previously, it transforms the conditions of research”.
