For centuries, the study of cuneiform—the world’s oldest known writing system—relied on painstaking manual transcription, with scholars dedicating years to deciphering individual tablets. You needed a lot of work and a great deal of expertise to translate cuneiform. Now, artificial intelligence is making the process much simpler.
An AI-driven system, ProtoSnap, developed by researchers at Cornell and Tel Aviv University, can recognize and reconstruct cuneiform characters with remarkable precision, even accounting for variations in writing styles across different regions and time periods.
This breakthrough is already expanding the number of deciphered texts, shedding new light on the economic and social history of ancient Mesopotamia. But that’s just the start of it.
From Clay to Code
Unlike modern alphabets, which rely on a limited number of characters, cuneiform has over a thousand signs that evolved across different regions and time periods. It’s an archaeological nightmare. The same word, written centuries apart, can appear drastically different, making it an immense challenge even for human experts.
“Even with the same character, the appearance changes across time, and so it’s a very challenging problem to be able to automatically decipher what the character actually means,” says Hadar Averbuch-Elor, a computer scientist at Cornell Tech.
To solve this, the researchers turned to a diffusion model, a type of generative AI that aligns an image of a cuneiform character with a known prototype, snapping it into place like a puzzle piece. The result: a system that can copy, reproduce, and recognize cuneiform inscriptions faster and more accurately than ever before.
Out of an estimated half-million cuneiform tablets sitting in museums, only a fraction have been fully translated. By automating much of the transcription process, ProtoSnap could vastly expand the body of available ancient texts, providing new insights into Mesopotamian law, economy, and daily life.
“At the base of our research is the aim to increase the ancient sources available to us by tenfold,” says Yoram Cohen, an archaeologist at Tel Aviv University. “This will allow us, for the first time, the manipulation of big data, leading to new measurable insights about ancient societies—their religion, economy, social and legal life.”
A New Era of Decipherment
ProtoSnap is just the latest in a growing list of AI breakthroughs reshaping how historians read the past. In 2023, a team of computer scientists and papyrologists achieved what once seemed impossible: they extracted readable Greek text from a papyrus scroll that had been buried under volcanic ash for nearly 2,000 years.
These scrolls, discovered in the ruins of the Roman city of Herculaneum, had remained a mystery since their discovery in the 18th century. Fragile and carbonized, any attempt to physically unroll them resulted in their destruction. But AI provided another way.
Using machine learning techniques, the Vesuvius Challenge team trained an algorithm to recognize the faintest traces of carbon-based ink hidden within the layers of papyrus. The outcome is a strip of Greek text, glowing against the digital background, revealing entire passages unseen for millennia.
“It was incredible,” recalls Federica Nicolardi, a papyrologist at the University of Naples. “I thought, ‘So this is really happening.’”
A New Window in the Ancient Past
The same pattern is playing out across different civilizations and languages. AI models have successfully restored damaged Greek inscriptions, translated ancient Akkadian tablets, and even predicted the origins and dates of previously undated texts.
Oxford University’s Ithaca model, for instance, has already helped historians resolve long-standing debates in classical studies. In one case, researchers used the AI to redate a series of Athenian decrees. Based on historical evidence, some scholars had suggested a later timeframe for these texts—around 420 BCE instead of 446/445 BCE. When put to the test, Ithaca independently confirmed the date 421 BCE, lending weight to the revised historical narrative.
“Although it might seem like a small difference, this date shift has significant implications for our understanding of the political history of Classical Athens,” says Jonathan Prag, Professor of Ancient History at the University of Oxford.
Meanwhile, Korean researchers are using AI to process massive archives of Hanja, the ancient writing system used in Korea and China. These archives, which document the reigns of 27 Korean kings over five centuries, contain an overwhelming amount of information. AI is now making it possible to detect patterns in governance, economy, and diplomacy that might have otherwise gone unnoticed.
The flood of newly accessible texts is not without challenges. Neural networks can sometimes generate misleading translations, and machine hallucinations—where AI fills in gaps with plausible but incorrect content—remain a risk. Experts stress that AI should complement, not replace, human scholars and humans should have the last world
Despite these hurdles, AI-driven decipherment is opening doors once thought to be permanently sealed. With the right tools, historians may one day reconstruct lost libraries, decode languages that have been silent for millennia, and ask new questions about civilizations long gone.
“We believe machine learning could support historians to expand and deepen our understanding of ancient history, just as microscopes and telescopes have extended the realm of science” says Yannis Assael, a research scientist at DeepMind.
The ancient world is still speaking. We’re just now learning how to listen.
