AI Deciphers Ancient Hammurabi Tablet with 98% Accuracy
AI Deciphers Ancient Hammurabi Tablet with 98% Accuracy
A new artificial intelligence (AI) system has accurately read an ancient Hammurabi tablet with 98% precision, marking a significant step forward in translating some of the world’s earliest written laws.
The system successfully identified the cuneiform signs carved into a clay tablet containing the opening line of Hammurabi’s Code, written around 1754 BCE in Mesopotamia. The breakthrough, shared in a May 7, 2025, study on the public research site arXiv, could help unlock thousands of untranslated tablets stored in museums.
The code of Hammurabi
The Code of Hammurabi is one of the oldest and most well-preserved legal texts from the ancient world. Commissioned by King Hammurabi of Babylon, the code consists of 282 laws carved into a tall basalt stele, written in Akkadian using cuneiform script. These laws covered a wide range of topics, including trade, property, family matters, and criminal justice, and were based on the principle of retributive justice—often summarized as “an eye for an eye.” The Code of Hammurabi aimed to bring order, fairness, and predictability to Babylonian society, and it offers valuable insight into the values, social structure, and legal practices of the time.
Researchers trained AI on thousands of images
The study was conducted by researchers Shahad Elshehaby, Alavikunhu Panthakkan, Hussain Al-Ahmad, and Mina Al-Saad at the University of Dubai. Their work focuses on using modern image-recognition tools to identify the wedge-shaped symbols used in one of the earliest writing systems in history.
To train the system, the team fed it over 14,000 high-quality images representing 235 distinct cuneiform signs. These symbols, once pressed into clay using a stylus, recorded laws, trade documents, and religious texts in ancient Mesopotamia.
The tablets are small—often the size of a human hand—and reading them by sight remains a slow, manual process. Few experts can read cuneiform today, and copying each tablet by hand can take hours. The AI system offers a much faster alternative.
The top-performing model was nearly flawless in testing
When tested on withheld images, the system’s best-performing model, based on a version of EfficientNet, misread just one symbol in every ten thousand. On the real artifact, it achieved 98% accuracy, misidentifying only two characters out of every hundred. A second model tested in the study showed lower accuracy at 89 percent.
AI Achieves 98% Accuracy in Deciphering Ancient Babylonian Law Tablet
Researchers have successfully used machine learning to transcribe a 3,700-year-old Babylonian clay tablet dating to the era of Hammurabi. The AI model, trained on thousands of cuneiform signs, accurately… pic.twitter.com/wFkHp2I4J1
— History Content (@HistContent) July 9, 2025
The researchers say these results are promising for museums and universities looking to digitize their collections. Many institutions house large numbers of unread tablets from Mesopotamia, Syria, and Anatolia. Fast, reliable transcription could make these texts more accessible to scholars and the public.
Technology could support deeper historical analysis
Beyond translation, the system could also support historical research. By comparing how symbols changed across regions and periods, scholars may be better able to date the tablets and trace their geographic origins. This could lead to more accurate reconstructions of early civilizations.
Plans to expand to other ancient scripts
The Dubai-based team now plans to improve the system further by combining multiple AI models and refining its ability to process damaged or burned tablets. They also hope to apply the same approach to other writing systems, such as Egyptian hieroglyphs, once large enough image sets become available.
As AI tools continue to evolve, researchers believe they will not only speed up translation but also help uncover long-lost stories etched in clay thousands of years ago.