AI Decoded a 3,700-Year-Old Babylonian Tablet - History Just Changed

Hi, it's Katrina. For thousands of years, the greatest stories of humanity have sat in silence, broken into millions of tiny clay fragments that no human could ever hope to piece together. But thanks to AI, the silence is finally over. Today, we are looking at five ancient tablets decoded by artificial intelligence, from hidden secrets in the Dead Sea Scrolls, to a discovery that is rewriting history books as we speak, the Song of Babylon. Let's dig in. The Song of Babylon. AI has just revealed a new secret from the past. A single clay fragment, no larger than the shard of a shattered dinner plate, sat for decades in a Baghdad storeroom. Like many other artifacts, it was catalogued, shelved, and forgotten. Its surface was etched with faint cuneiform wedges, the kind that usually signaled broken legal formulas or ritual instructions. Stripped of context, no one expected that it was actually music. This small piece turned out to be part of an estimated 3,700-year-old Babylonian hymn. It was a song that praised the city and language so rich and frilly, you would think Shakespeare wrote it. The fragment was one of many thousands of fragments photographed as part of the Electronic Babylonian Library. This is a long-term project led by Enrique Jimenez at Ludwig Maximilians University in Munich, in collaboration with scholars at the University of Baghdad. The aim of the project was to digitize the scattered remains of Babylonian literature and let algorithms compare shapes, match phrasing, and detect patterns across fragments that have been physically separated for centuries. Imagine millions of random fragments of cuneiform texts, all uploaded to a database. A human would have to look at each one, and there is no way they could memorize and connect them all. But AI can look at all of them at once and match the fragments that need to be together. Because on their own, most of these clay pieces are useless. A line here, a word there, a verb without a subject. Assyriologists have spent generations staring at fragments, hoping that memory or intuition might spark a connection. Now the software is doing it in seconds, and the first big breakthrough turned out to be a song. As the fragments snapped together digitally, scholars realized they were reconstructing a text that ancient Babylonians knew very well. The hymn appears in at least 20 separate manuscripts, meaning it was copied repeatedly over a span of about 500 years. It must have been a very popular song. Most of the clay fragments containing pieces of the song were found in Sippar, a city south of Babylon that functioned as a learning center. Based solely on the evidence, the hymn appears to have been used in schools. Children practiced its lines as exercises on pieces of clay, shaping the words with reed styluses long after the poem was first composed. It's got scholars wondering how a text so widely copied in antiquity could slip entirely through the cracks. Roughly two-thirds of the hymn has now been restored, amounting to about 250 lines. It opens with an outpouring of devotion for the god Marduk, the patron deity of Babylon. It's something you find in most Babylonian compositions. It was almost like a rule that they needed to begin by praising one god or another, you know how it is. After the praise to Marduk, the focus shifts and turns outward toward the city itself. The poem goes on to treat Babylon not merely as a place ruled by gods, but as a living paradise shaped by water, people, and the values they all shared. In the song, Babylon is described as flourishing with an abundance of wealth. The city is compared to a cultivated garden, a place where growth is exponential and everyone is living their best lives. It's basically one massive celebration of the city, and that has scholars wondering if it wasn't propaganda. In times of trouble, the government will do whatever it takes to assert its authority. This can be done any number of ways, but the most tried and true is through nationalism. If you can get the population riled up and patriotic for their homeland, it's a lot easier to send them to fight your war. This song, an overt praise to Babylon as the greatest city on earth, may have been a tool to get the Babylonians excited about their national heritage. It could have even been Babylon's national anthem, seeing as it appears to have topped the charts for five consecutive centuries. Even Taylor Swift can't pull that off, although maybe. We'd have to check in 500 years. The poem talks about the Euphrates River quenching the with herbs and flowers. It praises the free citizens of Babylon for their exemplary behavior, describing them as protectors of orphans and upholders of justice. Then the song does something rare. It goes on about how respectful the Babylonians were for the foreigners who lived among them. Rather than hate those who were different, it seems the Babylonians welcomed immigrants. The whole thing paints Babylon as a city where responsibility rather than dominance is respected. Babylon is presented as a place that defines itself by how its people treat the vulnerable. This is not what historians have been telling us about the Babylonians. They are often described as the villains of the Bible, and of course as the people who invaded Israel, destroyed the first temple, and enslaved the Jews. This song paints Babylon in a completely different light. It shows them as respectful, peaceful, and wise. The hymn even speaks at length about women. This might be the biggest shock. Throughout all the surviving texts that have been rediscovered in Babylon, women appear in pretty much none of them. They pop up occasionally in legal and family roles, but as people they barely make an appearance. This hymn breaks that pattern by focusing on priestesses and their moral standing. They are portrayed as devout and discreet, as women who cleanse with pure water and kneel in prayer armed with supplication. The poem names three distinct classes of priestesses. Prior to this discovery, there were no known texts describing these priestess roles in such detail, or framing them as ethical figures within the city. The way the hymn makes it sound, all the classes of priestesses were revered by everyone. They were treated with the kind of respect typically reserved for priests these days. Experts are saying this particular section may be the most important part of the hymn. It gives us a glimpse into how Babylonians treated their female religious figures, and apparently it was pretty good. This brings me to the end of the hymn. Sadly, the final column of text is badly broken, and so nobody knows how the song ends. Around 100 lines are missing, and even after completing the rest, the AI program can't seem to crack it. Speaking of ancient documents, AI has found some more incredible information about the Dead Sea Scrolls. But first, I want to give a big shout out to Jenny Lee and Brent Price. Love you guys, and thanks so much for the support. If you are new here, be sure to check out our member program for behind the scenes and some extra content. AI Analysis of the Dead Sea Scrolls. The Dead Sea Scrolls might be even older than scientists thought. Decades after the world's most shocking religious texts were found in jars tucked away in the backs of caves in one of the driest deserts, the truth is finally coming out thanks to artificial intelligence. The Dead Sea Scrolls were found between 1946 and 47, when Bedouin shepherds stumbled upon clay jars inside caves through Qumran in Israel's West Bank. Inside were manuscripts unlike anything previously known. Over time, researchers identified roughly 1,000 texts written on animal skin, papyrus, and even hammered sheets of copper. Together, they formed a library of religious, legal, and communal writings from ancient Judea. Among them were the earliest known copies of texts from the Hebrew Bible. Versions of Genesis, Exodus, Isaiah, Kings, and Deuteronomy appeared in forms centuries older than any previously discovered manuscripts. In some people's opinion, the discovery of the Dead Sea Scrolls proves the Bible is nothing but a collection of ancient writings stuffed into a giant book, sort of like a compilation or an anthology. If the Dead Sea Scrolls are some of the earliest known texts from the Bible, but they differ from what is written in the Bible today, doesn't that mean the modern Bible isn't accurate? It's too big of a question to tackle, so let's stick to the current plot. Based on paleography and limited radiocarbon dating, scholars place the scrolls at being written in roughly the 3rd century BC. The problem is that paleography isn't always reliable. For it to work properly, it requires a person with expert judgment. Paleography is the technique of scholars studying letter shapes, spacing, stroke order, and stylistic habits, and then comparing them to dated inscriptions to come up with an accurate date. It is a discipline built on experience and pattern recognition, refined over generations, but because it relies on the human brain, there is always room for error. As a result, the age of many of the Dead Sea Scrolls has remained an educated estimate rather than a fixed fact. In an effort to clarify matters, a team of researchers turned to artificial intelligence. They developed a program called Enoch, designed to analyze handwriting patterns and associate them with time periods. The project was led by scholars at the University of Groningen in the Netherlands, working with help from the Qumran Institute. Scientists started by radiocarbon dating 27 manuscripts. They received seriously mixed results, with some fragments being younger than earlier estimates and several being older. What the team did next was use their newly dated samples as training material for their AI model. Enoch was fed high-resolution images of handwriting from 24 manuscripts, along with their radiocarbon dates. The program learned to associate subtle variations in letter forms within specific time ranges. The AI was learning to tell how old a specific manuscript was based solely on the physical ink on the page. I didn't even think something like that was possible, but here we are! After internal testing using 13 additional images from the same manuscripts, the researchers presented Enoch with a larger challenge. They asked it to estimate the dates of 135 manuscripts that lacked reliable dating. The program's results aligned with the estimates of human scholars about 79% of the time. One manuscript, known as 4Q114, preserves parts of the Book of Daniel. Scholars previously dated its handwriting to around the time of the Maccabean Uprising in 165 BC, based on the script style and historical content describing the desecration of the Second Temple. When that manuscript was given to the AI Enoch, it estimated the scroll was written between 230 and 160 BC. Not perfectly accurate, but close enough to be correct. Remember, the scholars put it at 165 BC. It's the successful result that the study's authors are using as evidence that the tool can meaningfully perform paleographic analysis. However, that 21% of uncertainty is really bogging down the program. Enoch's creators claim their AI offers a way to examine handwriting on a massive scale, detecting patterns too subtle for a single human observer to track consistently. Not everyone is persuaded. Some paleographers view the results as interesting, but not that impressive. The idea that certain scrolls might be slightly older than the initial test isn't novel, but it's expected. It makes a great headline, though, but it's not shocking to the actual experts. It is to us. There's also the concern about how the results are being framed online. Claims that AI allows researchers to study the hands that wrote the Bible have been repeated in headlines across YouTube for months. Scholars are saying it's an overstatement. Even though the Dead Sea Scrolls are seriously ancient, they aren't originals. They are still copies of earlier writings. So, no, AI is not allowing researchers to study the hands that wrote the Bible. They are studying the hands that copied the Bible. AI has technically shown us that some of the Dead Sea Scrolls are even older than initially thought, but it isn't like they are thousands of years older. They weren't written 20,000 years ago by alien scribes. AI has proven they are merely a few decades or even a few measly years older. So while the headline is true, it just isn't as exciting as if the Dead Sea Scrolls had been written by Neanderthals or something. Now let's move on and meet a brand new AI archaeologist. Meet Aeneas, the AI archaeologist. An artificial intelligence model named Aeneas has entered the mainstream, tasked with decoding ancient Latin inscriptions as fast as Google Translate fixes your espanol. Aeneas is promising to estimate where each ancient Latin inscription came from, when it was carved, and even what its missing words might have been. Reading ancient Latin can be a nightmare, not because Latin is hard to translate, but because inscriptions are often found in poor shape. Chips, vanished letters, and broken fragments often make Latin inscriptions incomplete. Historians have traditionally filled those gaps by comparing fragments, memorizing stylistic quirks, and relying on their own expertise. Now Aeneas is promising to do it for free, instantly, and with minimal input from humans. So far, the results are alarmingly impressive. Aeneas was introduced to the world in a new paper published in the journal Nature. According to its creators, the model can analyze ancient Latin texts to produce a geographic origin, estimate a date of creation, and propose restorations for lost portions of text. This AI is behaving like a junior epigrapher with a photographic memory and infinite patience. It could change archaeology forever. If you're not sure what an epigrapher is, neither am I, but I'll try to explain it. Epigraphy is the study of inscriptions carved into durable materials like stone, metal, or clay. These texts form one of the most direct connections to the ancient world. Epigraphs record laws, dedications of victory, funerary statements, official decrees, and casual markers of daily life. They are also brutally difficult to interpret for a whole slew of reasons. Inscriptions change over time. Letter shapes evolve. Spellings shift. Regional habits emerge and disappear. Many inscriptions only survive in fragments. To make sense of one piece, scholars have to compare it to thousands of others. By searching for parallels in wording, layout, and formula, they can sometimes date a text and identify its source. But it's tough, overwhelming, and too much work for one human being. That problem is what led to Aeneas. The project was a collaboration between researchers at universities in the UK and Greece with help from developers at Google DeepMind. The team drew on three of the largest existing Latin epigraphic databases, compiling a dataset of 176,861 inscriptions. These dated as far back as the 7th century BC and as recently as the 8th century AD. About 5% of the almost 200,000 inscriptions were accompanied by images of the original objects. The rest was uploaded simply as transcribed text. By your usual archaeological standards, this dataset was insanely big. But it had to be if it was going to be the foundation for training the model. The technical side is even crazier. Aeneas operates through three separate neural networks, each focused on a specific task. One estimates where an inscription likely came from. Another predicts its date. The third attempts to restore any missing text. The system does not simply output an answer. It also lists comparable examples from its dataset, showing the user which inscriptions influenced its conclusions. The transparency is central to how the tool is meant to be used. By showing the scholars which inscriptions helped it reach its conclusions, the human can then make the jump and link different fragments that might be part of the same text. The model is already publicly accessible through a Google DeepMind website called Predicting the Past. You can visit it right now. Well, after you watch this video, of course. Just don't expect to upload your own inscriptions. Users must manually transcribe the text they wish to upload, line by line, and estimate how many characters are missing in damaged sections. Only then does Aeneas begin its analysis. So, unless you have a random Latin inscription on a chunk of clay at home and know precisely what it's missing, Aeneas probably isn't for you. But it wasn't designed for random folks at home. It's for researchers who want to resurrect every Latin inscription they set their eyes on. And that's exactly what they're doing. AI and the Indus Valley There's no getting around it. The Indus Valley script is unbeatable. It just can't be cracked. But AI is still going to take its swing. For more than a century, the written language of one of humanity's first great empires, and India's origin story, has sat at the edge of understanding. Visible, but unreadable. Carved into seals, tablets, and fragments of pottery, this script belongs to the Indus Valley Civilization, a vast Bronze Age society that flourished around 4000 years ago across what is now Pakistan and Northwest India. The ruins of the civilization's cities, like Harappa and Mohenjo-Daro, reveal advanced urban planning, standardized weights, and long distance trade. Every single thing that has emerged about the Indus Valley paints them as hugely progressive. They appear to have fought in no wars, had no kings, and lived in what may have been the world's first utopia. Seriously, the more you learn, the more you will find things that will shock you. There is evidence that they were incredibly advanced, possessed sophisticated technology, and maybe even surpassed the Egyptians in their wisdom. Yet we can read hieroglyphics. I mean, I can't, and you probably can't, but there are people who can. On the flip side, nobody can read the Indus Valley script. And that sucks big time, because this civilization may have figured out the secret, peace, and prosperity. They had money, they had gorgeous architecture, fabulous art styles, and the most stubborn script ever. To this day, no one has read even a single word from the Indus Valley. At this point, it's almost hopeless. Or is it? With the rise of AI and LLMs, artificial intelligence is now being asked to do what generations of scholars could not. The question is whether modern computation can push beyond the limits of human intelligence. It's time to place your bets. All of a sudden, scientists have been given some incredible motivation to get this done. Tamil Nadu Chief Minister M. K. Stalin announced a $1 million reward for anyone who can successfully decipher the Indus Valley script. The prize money reflects the belief held by some scholars that the script may be linked to early Dravidian languages, which are still spoken across southern India today. Decoding the script and getting the prize money isn't just about technology, ancient writing, and human history. It's also become about Indian pride. If scientists can prove the Indus Valley script is connected to Dravidian, it would reshape how South Asian history is understood. It's become a national race to unlock India's ancient secrets. But even with money and modern tools on the table, the Indus script remains uniquely resistant to code crackers. The problem begins with the material itself. About 4,000 Indus inscriptions are known, but most are extremely short. It's strange because most fragments consist of just four to five symbols. Those could be letters, numbers, or ideas represented as images. Nobody knows. The longest examples rarely exceed a dozen signs. And that is nuts, because in most other early empires, inscriptions were huge. Egyptian inscriptions covered whole stelae, as if the scribes were rambling in a chat room high on Mountain Dew at 3am and couldn't stop typing. The bottom line is that there's simply not much text to work with. Worse, there are no bilingual inscriptions. When scholars cracked Egyptian hieroglyphs, they had the Rosetta Stone as the blueprint. The Rosetta Stone had the same text in multiple languages, allowing scientists to crack hieroglyphics because the exact same thing was written under it in Greek. And we understood Greek. The Indus Valley offers nothing comparable. There is no translation key to speak of, and no clear references to known historical events or rulers. Then there are the symbols, which, let's just say, are beyond frustrating. Nobody knows what the symbols even represent. The Indus Valley script could be logographic, with symbols standing in for entire words. Or it could be syllabic, phonetic, or something else entirely. Without knowing the basic structure, even identifying where a sentence begins or ends is nothing but guesswork. And this is where artificial intelligence enters the picture. Unlike human researchers, AI does not need intuition or cultural familiarity to begin working. It excels at pattern recognition. The one thing that AI can do better than anything is recognize patterns. Machine learning models can scan thousands of inscriptions and detect regularities that are difficult for a person to see with their human eyes. And the AI can do it about a million times faster. So far, scientists know that certain symbols appear frequently at the beginning of inscriptions, while others tend to only appear clustered at the end. This seems to suggest there is some structure to the language, which hopefully AI can pick up on. AI is also helping classify symbols, identify positional constraints, and test hypotheses about reading direction. Reading direction is a big one, because for those of us who speak English, we assume people are reading top to bottom and left to right. But go to China and there is a completely different reading order. Perhaps in the Indus Valley, people read bottom to top, right to left. It's these kinds of things that can make deciphering language from scratch impossible. But with AI able to test thousands of samples in an instant, things are looking up. An AI program could theoretically piece together all the specifics of the language. However, AI won't be able to come up with a meaning. Pattern detection is not the same as understanding. AI can tell researchers that a certain symbol often follows another certain symbol, but cannot tell them whether that symbol means king, grain, river, or nothing. Meaning requires context, and context is precisely what the Indus Valley script lacks. There are no surviving explanations, no associated myths, and no inscriptions describing themselves. They were one of the most successful, advanced, and peaceful empires in world history, yet scientists don't even have the name of a single king or god. And AI isn't going to magically sprout meaning where there is none. This is the hard limit that AI runs into. Machine learning models learn from examples, but the Indus Valley has no examples, so AI has no reference points. It might be able to puzzle out the mechanical parts, but it will be a human at the end of the day who decodes the meaning to the words. Even so, many researchers believe the organizational power of AI could be instrumental in reaching a point where a human can produce meaning. By narrowing down which interpretations are mathematically plausible, AI can eliminate vast numbers of bad theories. It can test whether proposed readings behave like real languages, or if the scientists are just grasping at straws. It can reveal whether the script is phonetic or symbolic based on statistical behavior. But if AI cracks the code, who gets the million bucks? AI deciphers the Epic of Gilgamesh. The Epic of Gilgamesh is officially getting the AI treatment. One of the oldest stories ever written down, an epic so epic it literally has epic in the title, is being scanned in a way it has never been scanned before. For those a little rusty on their Mesopotamian literature, I'll get you up to speed. The Epic of Gilgamesh is not a book in the modern sense. Unlike most traditional books, this tale exists only because countless clay tablets were pressed with wedge-shaped marks and then broken, buried, scattered, looted, misfiled, and rediscovered across thousands of years. It was written, or carved, an estimated 4,000 to 5,000 years ago, and it has survived all of these eons and bits and pieces, which scholars have painstakingly put back together. For thousands of years, no one on Earth could read it. But in modern times, cuneiform was cracked, Babylonian was decoded, and the Epic came to life. Now, an artificial intelligence system is helping scholars stitch the story back together faster than ever, in a way that could reveal parts of the tale never before heard. The project is called the Fragmentarium, and it operates at the Institute for Assyriology at Levig Maximilian University in Munich, Germany. At its center is Enrique Jimenez, who I mentioned before, a professor of Assyriology who studies some of the oldest written texts humanity has ever produced. In his spare time, Enrique likes to stand in darkened rooms before enlarged images of cuneiform tablets, holding fragments that once carried royal boasts, prayers, warnings, and messages of grief. Enrique is the guy you want decoding the tale of Gilgamesh. He has joined a long lineage of scientists who were each obsessed with the same thing. For a long time, clay tablets covered in strange wedge marks sat silently in museum collections and storerooms, their meaning lost. That changed in 1872, when a British museum employee named George Smith became the first modern scholar to read the Epic of Gilgamesh. His translation was revolutionary. His painstaking work decoding the cuneiform script revealed a flood story that predated the Bible and introduced the world to a heroic narrative older than Homer. All these years later, Christians are still mad about it. Translating the Epic of Gilgamesh was the first step to proving that many of the Bible's most famous stories were a little bit borrowed from even older civilizations. The Epic of Gilgamesh has its own flood story, almost identical to the one in the Bible, but starring a Mesopotamian guy named Utnapishtim instead of Noah. Sadly, George Smith did not live long enough to see how huge his discovery would become. He died just four years later at the age of 36, while traveling through the Middle East searching for additional fragments of the poem. I have to say, at least he died doing what he loved. Since then, generations of scholars have continued Smith's work. Archaeologists unearthed new tablets during excavations, while museum staff rediscovered fragments forgotten in storage. Authorities recovered looted artifacts and returned them to collections. Slowly, so painfully slowly, the story expanded. Each new line, each new deciphered tablet added a smidgen more to the overall story. But what a lot of people don't realize is that the Epic of Gilgamesh is not entirely complete. The process of translating fragmented clay tablets was never very efficient. Identifying fragments depended heavily on a scholar's memory, their intuition, and to be honest, luck. Every now and again, a scholar finds a fragment in a drawer in some dusty storeroom that changes everything, but it's pure luck. Now, with the invention of the Fragmentarium, historians are hoping to streamline everything and make translation way easier. The Fragmentarian system is built around a massive digital database of cuneiform fragments. It uses an algorithm capable of analyzing ancient texts far faster than any human researcher. By comparing characters, phrases, and patterns across thousands of fragments, it can identify matches that would otherwise take years to find, if they were found at all. The Fragmentarium does not only focus on famous texts like Gilgamesh, it processes all kinds of Babylonian writing. For example, legal documents, bills of sale, administrative records, and personal correspondence all pass through the system. While these texts may appear mundane, they provide critical context for all of Babylonian literature. They reveal how people spoke, traded, worshipped, and organized their lives, which in turn helps scholars understand literary works like Gilgamesh. I think you'll be surprised to learn just how long this system has been active. AI really blew up in 2020 and has been steadily seeping into every corner of our lives, whether we want it to or not. But the Fragmentarium has been online since 2018, before AI was a buzzword. In the last few years, the Fragmentarium has matched lines from the epic to texts from about 1,500 tablet fragments. In the several centuries before artificial intelligence was hired, scholars had made such connections for only about 5,000. That is 5,000 for humans over a span of centuries and 1,500 for AI over the span of about six years. Humans are winning, but AI is going to catch up quickly. It's not just about speed, but scale. The AI-based reconstruction process relies on databases that collect thousands of characters and textual variants. With that information available instantly, all known versions of a Babylonian text can be analyzed together and used in context. This is a change that's affecting the entire field. The Fragmentarium is part of a larger shift in how ancient manuscripts are studied. In 2023, researchers at Tel Aviv University unveiled a similar project that worked with Mesopotamian cuneiform. Another AI system developed at the University of Kentucky is helping scholars read carbonized scrolls burned during the eruption of Mount Vesuvius. I did a deep dive on those scrolls in a recent video, so be sure to check it out after this one. And if you don't want to miss any more videos, be sure to subscribe and hit that notification bell. The Epic of Gilgamesh tells the story of Gilgamesh, the demigod king of Uruk, and his companion Enkidu, a wild man created by the gods. Together, they kill Humbaba, the guardian of the forest. The gods respond by killing Enkidu. Grief-stricken about losing his best friend, Gilgamesh sets out on a journey to find his ancestor Utnapishtim, the only man to survive the catastrophic flood that destroyed the rest of humanity, hoping to learn the secret of immortality. There's a lot more to it than that, but I don't want to spoil it for you. 5,000 years later, the Epic of Gilgamesh is still as compelling, dramatic, and tear-jerking as it was when it was first written. And now, thanks to the Fragmentarium, the story is growing. Newly identified lines have added scenes between Gilgamesh and Enkidu, scenes that add extra moral tension to the narrative. There is also a new chapter that takes place after the pair kill Humbaba, in which they travel to Nibor to visit the god Enlil, something that was never known to exist before. As the project continues, let's all say a little prayer to Marduk that the full, complete, unabridged version of the Epic of Gilgamesh is found. Maybe then we'll get a big-screen adaptation. Thanks for watching! Do you think the Indus Valley script will ever be decoded? And if it is, what secrets do you think the text will reveal? Let me know in the comments. Be sure to subscribe and I'll see you next time for more videos about ancient history. Bye!