Ancient Language Decoded by an AI, What It Revealed Is Terrifying

0:00 where we will paste the text of the inscription and we will hide that word with question marks. In 79 AD, this ancient Roman city was also buried in volcanic ash by the eruption of Mount Vizuvius. But what was found here with the help of artificial intelligence could change the way we understand history. What if the voices of ancient civilizations were never really silenced, just waiting for the right machine to listen? Because that's what it feels like right now. where I said, "I think we can read everything inside the Herculanium scrolls without opening them." Ancient texts once thought lost forever, burned, buried, shattered, are finally starting to speak again. Unraveling the hidden history of these 2,000-year-old scrolls has required 21st

0:47 century technology. But here's the twist. They're not whispering lullabies. They're revealing secrets we might not be ready for. And the last ancient language on our list, it reveals the most shocking truth of all. Let's get into it. 20. The Danube script. You've heard of Kune form. You've probably heard of hieroglyphs. But the Danube script, that one's still flying under the radar. Archaeologists found strange symbols carved into pottery, stones, and figurines along the Danube river banks, modern-day Serbia, Bulgaria, and Romania. For decades, experts debated whether this was writing or just ancient doodles. Then AI got involved. Feeding the symbols into advanced pattern

1:34 recognition models, researchers noticed something insane. Structure, syntax, repeated sequences that mimicked language. Some of these markings are over 7,000 years old. If this really was a writing system, and all signs say it was, then the Danube script could be the oldest known form of written communication on Earth, older than hieroglyphs, older than Kunai form. No one knows exactly what it says yet, but AI has confirmed that it wasn't random. The arrangement, spacing, and symbolic repetition were intentional. So, who wrote this code while the rest of the world was still learning how to plant crops? And more importantly, why did it disappear without a trace? 19. The

2:19 Herculanium scrolls and the Vuvius challenge. For nearly 2,000 years, the charred papyrus scrolls found in Herculanium have been indecipherable. From prehistoric Europe to the shadow of Mount Vuvius. Another silent archive just woke up. When Vuvius erupted in the year 79, it swallowed a wealthy Roman library whole. The result? Hundreds of scrolls turned to charcoal, too fragile to unroll, too precious to destroy. For nearly three centuries, the scrolls of Herculanium sat unreadable. But in the spring of 2023, a digital firestorm was lit. The Vuvius challenge offered over $1 million to anyone who could use AI to virtually unroll the blackened papyrie. Enter Luke Ferriter. Just 21 years old.

3:07 He used machine learning to spot the first word, pfyras, meaning purple. That word cracked the seal. He teamed up with Yousef Nater and Julian Schilliger. Together, by February 2024, they had decoded over 2,000 Greek letters. First, it was just one, then two. Then, he made the breakthrough discovery that has won him the first prize in the Vuvius challenge. The scrolls revealed the voice of Filademus, an Epicurian philosopher, discussing life's greatest pleasures. This wasn't just a win for AI. It was a resurrection. Voices silenced by fire nearly two millennia ago are speaking again. And the world is listening. 18th Acadian Cooneyiform. Now we shift from the ashes

3:53 of Italy to the baked earth of ancient Mesopotamia. The Acadians left behind a sea of wedge-shaped impressions. Over 500,000 clay tablets stored in museums. Most of them are unread. Why? Because Aadian is brutally difficult to translate, only a few hundred people on Earth can do it. But in 2023, the game changed. Researchers from Tel Aviv and Ariel University developed an AI model trained on thousands of translated Acadian texts. It was written in the Babylonian language, which is a Semitic tongue related to the modern languages. Think Google Translate only for a 5,000-year-old language. The results fast, accurate, jaw-dropping. Another effort led by Steve at the AI Ununiform

4:40 Corpus processed over 130,000 texts, pulling from databases like CDLI and OCC. Even Germany joined the race with teams using 3D modeling and neural networks to recognize over 21,000 signs from nearly 2,000 tablets. While conspiracy lovers might chase myths of magic spells and ancient curses, most of the translations tell very real stories, trade agreements, tax logs, and even parts of the Epic of Gilgamesh. Thanks to AI, the oldest civilization on record just got a lot louder. 17. The Dead Sea Scrolls. You think you know the Dead Sea Scrolls? Think again. Discovered in the 1940s, these ancient texts are older than most

5:26 modern religions, written by a group called the Essenes. But one scroll in particular, the great Isaiah scroll, was always thought to be the work of a single scribe. Then came 2021. AI researchers at the University of Groingan fed thousands of samples of the Hebrew letter Alf into a model. What did they find? Tiny, almost invisible variations. Not one scribe, but two. That changes everything. So, these are new pieces of the puzzle and we can add them to our greater picture. It means there may have been organized writing teams working in the shadows of the desert. And with AI able to pick up the emotional shifts in pen pressure and style, we're learning not just what they wrote, but how they felt. Forget the

6:12 myth of hidden Bible codes. The real story here is one of process, production, and human touch. Thanks to AI, we're not just reading ancient scriptures, we're meeting the people behind the pen. 16. The Indiscript. Back up to South Asia, and we find one of the great unsolved puzzles of archaeology. The Indiscript used by the Indis Valley civilization between 2600 and,900 years before Christ, it appears on seals, pottery, and tablets. But here's the catch. There's no Rosetta stone, no translation key, just rows of strange symbols. And a culture that historians again and again say is remarkable because of just how peaceful it was. But starting in 2009,

6:59 Professor Rajesh PN Ralph and his team tried something new. They built a Markoff model, a statistical tool that maps sequences, and fed it every known indiscription. The patterns it revealed too organized to be random. The script showed language-like behavior. That alone was groundbreaking. Then came AI. Deep learning models began recognizing recurring structures. Some researchers even reached over 80% accuracy in symbol segmentation. Still, challenges remain. Most inscriptions are super short, and no one knows what language the script was based on. Still, there's a million-dollar prize waiting for anyone who cracks it. And with AI closing in, someone might just walk away with both

7:45 money and history in hand. 15. The Marowoitic language. Let's follow the Nile South to a lesserknown kingdom, Maroa. From around 300 years before Christ to 400 after, the Maroic language flourished in what is now Sudan. It had two scripts, hieroglyphic for monuments and cursive for daily use. We've known about this language for over a century, but no one could decode it. No bilingual texts, no dictionary. It was a locked door. Now, researchers at MIT's CSIL lab are picking the lock. Their algorithms, previously successful with Ugaritic and Linear B are being retoled for meic. These models don't need translation. They detect structural logic, hunting

8:32 down grammar-like patterns and statistical relationships between symbols. It's still early, the data is thin, and the inscriptions are short, but AI is chipping away at the silence. And with every scan, the odds tilt in favor of discovery. So, what's next? Well, if machines can wake up Nubia's forgotten tongue, just imagine what they'll say when they hit Egypt. 14. The Protoelomite script. East of Mesopotamia, deep in ancient Iran, lies one of the world's oldest writing systems, Protoelomite. Dating back to around 3,100 years before Christ, these wedgel-like signs were carved onto clay tablets during a time when agriculture and citystates were just blooming. The problem, no one has ever truly cracked

9:19 the code. For more than a century, linguists stared at these strange symbols like static. This is the writing of the Elommites called ununiform. Only nobody knew what the writing meant. That changed when AI entered the scene. Oxford researcher Dr. Jacob Dah led efforts using reflectance transformation imaging, RTI, to capture ultradetailed images of the tablets. These weren't just pretty pictures. They were fuel for neural networks, allowing AI to recognize subtle grooves and glyphs too faint for the human eye. More recently, in 2025, researchers introduced algorithms that focused purely on deciphering the numerical systems, essential for decoding trade and economy

10:05 records buried in the tablets. The machines picked up patterns in how quantities were recorded, giving scholars context that had been missing for decades. Still, the full script remains a mystery. But now with every pixel scanned and every symbol classified, Iran's silent tablets are beginning to murmur. 13. Ancient Egyptian hieroglyphs. In the ancient world, there was a view that this life that we have, this physical life. When it comes to ancient writing, Egyptian hieroglyphs need no introduction. Intricate, majestic, everywhere. But decoding them has always been painstakingly slow until 2020. That's when Google Arts and Culture teamed up with McQuary University and others to launch

10:50 Fabricius, an AI powered tool that brings ancient Egyptian writing to the digital age. Fabricius doesn't just translate, it teaches. One mode walks users through the basics. Another lets them play with the script, but it's the work mode where things get serious. Here, AI uses pattern recognition powered by AutoML to scan and match hieroglyphs with known translations. streamlining what used to take months. Even better, Fabriius connects directly to a cloud-based hieroglyphic database. It offers multiple translation options for each symbol and draws on gardener's list, a foundational reference for Egyptologists. Sure, it's not perfect. Hieroglyphs include everything from sounds to ideas to grammatical clues.

11:36 But Fabricius is leveling the playing field, making it easier for both scholars and curious minds to step into one of history's most iconic languages. What once took lifetimes now takes seconds. The pyramids haven't moved, but our understanding of what's written inside them just leaped forward. on akinchi rangarango. Jump halfway across the world to Easter Island and you'll find a language that defies logic. Rangorango, a script discovered in the 1860s carved into wooden tablets with a bizarre snaking direction. Terry Hunt, an archaeologist, has been studying them for over 20 years. Left to right, then right to left, like a mirror flipping after each line. Only about two dozen

12:23 tablets survive today. The culture that created them silenced before the 20th century. No Rosetta Stone, no living speakers, just wooden glyphs and a lot of speculation. For years, debates raged. Was it real writing or symbolic doodles? In the 1950s, Bhutenoff and Norzaf saw patterns suggesting calendars and family lineages. In the 1990s, Steven Fiser claimed it told creation myths. Most disagreed. Now AI is stepping in. Researchers are using machine vision to isolate individual glyphs, cataloging them across all surviving samples. These models are trained to detect similarities between Rangorango and known writing systems, not for direct translation, but to spot

13:08 potential linguistic traits. It's still far from solved, but for the first time, Rangorango has digital allies. Wooden tablets that once sat mute on museum shelves now pulse with potential. The past isn't always lost. Sometimes it just needs a better algorithm. 11th. The Ismian script. Meso America is a hotbed of ancient writing. You've got Mayaglyphs and Zapotch scripts. But what about the Ismian script? That one's the underdog. Also called epolmech. It shows up on artifacts like the Tux statueette and Lamohara Stala, dating between 500 years before Christ and 500 after. Some believe it's a legit writing system. Others say it's too limited to count. Scholars like Justine and Kaufman made

13:54 partial breakthroughs in the 1990s, but their ideas remain debated. So where does AI come in? While no major AI project has targeted Ismian directly, researchers at MIT's CSA lab have built models that can work on scripts just like this. Undeciphered, scarce, and ancient. These systems compare how language evolves over time using evolutionary linguistics to trace possible roots and structures. If applied to Ismian, these models could analyze symbol order, frequency, and morphology, offering clues about syntax, or phonetics. All that's missing is data, more inscriptions, more scans, more context. But the tools are ready. It's only a matter of time before someone gives AI the green light to finish what humans started. The

14:41 writing's on the stone. It's just waiting to be read. 10th Oracle Bone Script. Now, let's rewind to ancient China over 3,000 years ago. During the Shang Dynasty, rulers carved symbols onto turtle shells and ox bones. We're going to delve into the Shang Dynasty of ancient China, where psychics interpreted heated bones to help people make decisions, asking questions about war, crops, and fate. This was the Oracle Bone Script, China's oldest known writing. In 2024, a major leap was made. Haisu Guan and his team built OBSD, an AI system based on diffusion models. It didn't just scan these characters. It matched them to modern Chinese equivalents with astonishing precision.

15:26 The breakthrough came thanks to a massive image library called Hust OBC filled with over 140,000 glyph samples, some known, others still a mystery. The AI was trained on this data, learning to fill in gaps where carvings were damaged or faint. Another tool from Tencent, the Oracle Bones Corpus, gave researchers a user-friendly platform to study bone fragments side by side, finding overlaps and matches in real time. Oracle bones were meant to talk to the future. Ironically, it's the future through AI that's finally talking back. Although some oracle bones have been found from the Cho Dynasty, most of them are dated to the Shang dynasty. The kings of the Shang dynasty might never have imagined their secrets would one day be decoded

16:13 by code. Nine. Linear B and Yugaritic. Some ancient mysteries took centuries to crack. Others are falling in ours. Linear B was used in Mcinian Greece between 1450 and 1200 years before Christ. Linear B is one of two linear scripts. It was the earliest form of Greek writing and for decades nobody could make sense of it until Michael Ventress, an architect turned linguist, cracked it in 1952. Fast forward to 2019. A team from MIT, including Giaming Luo and Regina Barzel, built an AI model to do it faster. Their system translated over 67% of cognates shared words

17:00 between linear B and modern Greek. That's not just good, that's groundbreaking. Now look at Yugaritic, an ancient Semitic script from modern-day Syria. In 2010, another MIT team led by Ben Snder and Kevin Knight created a machine learning model that deciphered yugaritic by aligning it with Hebrew in mere hours. Both projects used comparative linguistics AI matching word roots, grammar, and meaning across time and languages. This proves one thing. With the right data, even the most ancient tongues don't stand a chance. AI isn't just catching up to the past. It's outpacing it. Eight, the Nazca lines. Some writing doesn't use words at all, just like the Nazca lines. In 2024,

17:45 a team from Yamagata University in Japan, working with IBM Research, used AI to uncover 303 new geoglyphs in Peru's Nazca desert. A giant picture called a geoglyph of a hummingbird etched in stone and sand by ancient Peruvian peoples. These massive earth drawings, some as small as 10 ft, others stretching hundreds, were created over 2,000 years ago by the Nazca people. Previously, researchers spent nearly a century identifying 430 geoglyphs with AI over 300 in just 6 months. The key was training image recognition software on aerial photos, teaching it to spot patterns humans missed. Some depict animals, birds, cats, whales. Others

18:32 look like people or ceremonial scenes, and many were placed along ancient footpaths, suggesting they were meant to be seen from the ground. The Naza lines have long been a source of wonder, but there remains much uncertainty about their origins and meanings. Not the air that debunks the alien theory, but it doesn't make the lines any less mysterious. Why were they made? What did they mean? A I might not answer those questions yet, but it's helping us find the clues a whole lot faster. Even in the most barren landscapes, ancient people left signs. All we needed was a machine to see them. Seventh Inca Kipus. Sometimes language doesn't look like writing at all. High in the Andes, the Inca didn't leave scrolls, tablets,

19:18 or carved glyphs. Instead, they left threads, literally known as kipus, these knotted strings were their version of a ledger, a map, and maybe even a story book. For centuries, we thought they were just counting tools, color-coded cords used to log taxes, livestock, or census numbers. And in some ways they were. But what if that was just the surface? Enter the new wave of researchers and machines. Anthropologist Sabine Highland dug into Spanish colonial records and surviving Kipus, suggesting they might be more than calculators, maybe even phonetic, maybe even narrative. AI has started helping map out the subtle patterns where knots are tied, the spacing, the string placement. Using statistical modeling,

20:03 algorithms are revealing structures too complex to be random. Could these knots hold syllables? Could they tell history the Incas never wrote on walls? We're closer than ever to answering that. The Inca spoke through wool and code, and AI might finally be fluent, one thread at a time. Their silence is unraveling. Six. The Palumsts. Not all secrets are buried in dirt. Some were buried under other secrets, erased, overwritten, and hidden in plain sight. Enter the Palumsts, ancient manuscripts, where the original text was scraped off and replaced with something new, was about to begin on a rather small and unassuming manuscript. And for centuries, what lay beneath was lost forever. That changed when

20:48 multisspectral imaging and AI joined forces. One legendary example, the Archimedes palest. Beneath a 13th century prayer book, researchers found lost mathematical writings from the Greek genius himself, unseen for over a thousand years. In 2021, a team led by Anna Stinska and David Messenger developed a new AI technique. Using deep generative networks, they trained machines to isolate the overwritten layers. The result? hidden texts became legible again without damaging the original page. This method is now being applied to hundreds of palumests from Bzantine treatises to medieval medical scrolls. While myths talk about magic spells and secret societies, the real

21:36 power lies in rediscovered knowledge, science, philosophy, astronomy. AI didn't just read what was erased. It brought the forgotten back to life. Bashinji nushu dil in the hills of southern China a hidden language once whispered between generations of women nushu literally women's writing was invented by yao women in Hunan province who were denied formal education for centuries they wrote songs stories and sorrow on fans and paper away from the eyes of men by the 21st century nushu was dying then came 2025 Dartmouth college launched newu rescue, an AI initiative to revive the script using minimal data using only 35 sample pairs.

22:24 GPT4 Turbo was trained to translate Nushu into Chinese. It achieved nearly 50% accuracy, a major leap for a script no modern AI had ever seen before. Another project, Ai Nushu, simulated how women might have evolved the script across generations. It wasn't just about linguistics. It was cultural preservation. These weren't just letters. They were lifelines. Songs of love, grief, resilience. With AI amplifying their voices, Nushu is no longer a fading whisper. It's becoming a song the world can finally hear. Four. The Voinich Manuscript. Some mysteries don't whisper, they mock. Enter the Voinich Manuscript. A 600-year-old book that

23:09 looks like it came from another world. It's filled with bizarre plants that don't exist. zodiac signs drawn by an unsteady hand and nude women bathing in cosmic bathtubs. But the real puzzle, the language, it's like nothing ever seen before. No alphabet, no origin, no translation. In 2018, scientists at the University of Alberta ran the manuscript through a custom AI algorithm. the fact that the manuscript is written in a way that nobody else could read it. Their hunch it might be scrambled Hebrew stripped of vowels and syntax. The model claimed that up to 80% of the words had Hebrew equivalents. One sentence even translated as she made recommendations to the priest man of the house and me

23:56 and people promising maybe. But when Hebrew scholars reviewed the output, it unraveled fast. The grammar was off. The meanings were jumbled. Even Google Translate couldn't help. A year later, British scholar Gerard Cheshure dropped another theory. Protoromance, a forgotten language used by medieval nuns to write a guide on women's health. Again, scholars raised their eyebrows and their objections. AIS taken its swing. So have humans. Yet, the Voyic manuscript just smiles back with silence. Maybe it's nonsense. Or maybe it's the most well-hidden voice of them all. Third, Mayan glyphs. When it comes to visual complexity, Mayan glyphs take the crown. Spirals within spirals, shapes within

24:41 shapes. For centuries, decoding them meant years of visual study and guesswork. However, a recent project called artful algorithms, led by West Virginia University, is rewriting the game. The decipherment of my hieroglyphic writing only took place within the last uh 30 years approximately. Combining art history and computer science, they trained AI to isolate and classify individual Mayan glyphs from massive databases of scanned stonework. Another project, segmentation of Maya hieroglyphs, fine-tuned foundation models to identify and extract glyphs with pixel level precision. This helped machines distinguish overlapping symbols critical for understanding compound meaning. These AI breakthroughs are helping finish what linguists like Yuri Kurrosaf

25:28 began in the 1950s when he cracked the phonetic elements of the script. Now, instead of years, identification takes seconds. Some glyphs speak of gods, others sacrifices, but most tell stories of everyday life, families, calendars, crop cycles. I doesn't just read these stories, it gives them back to us. Two, the Cypro Manoan script. Over 2,000 years ago on the island of Cypress, a mysterious writing system appeared. The Cypro Manoan script written between 1550 and 1050 before Christ has baffled scholars for over a century. Some say it's one script, others say it's three. Recent studies using unsupervised deep learning suggest a radical possibility.

26:14 It's all one language. By training algorithms to analyze the visual structure of signs, researchers uncovered consistent rules and stylistic features across all known samples. No translation yet, but structural harmony suggests a unified origin. Even more ambitious was the use of seans, convolutional generative adversarial networks to analyze cypromanoan next to other bronze age scripts such as clay balls, cylinders and tablets and votive stands which bear cypromanoan inscriptions have been found. The results hinted at potential relationships with linear A and early Semitic languages. For now, Cypro Manoan remains unread. But with machines narrowing the field, a breakthrough feels closer than ever. Cypress hasn't

27:01 given up its secrets yet. But AI is asking the right questions. First, Atruskin. Tucked in the shadows of Rome, the Atruscans were once the rulers of central Italy. They built cities, temples, and tombs long before Caesar drew breath. But their language, that's still one of Europe's biggest mysteries. We've found thousands of inscriptions on urns, mirrors, monuments. Enough to fill volumes, but not enough to read them fluently. Why? Because at Truskin is a language isolate. There are no known relatives, no Rosetta Stone, and only partial translations. That's where artificial intelligence enters the ruins. By feeding neural networks a massive corpora of Atruscan texts, researchers are detecting statistical

27:47 patterns invisible to the naked eye. AI can flag recurring suffixes, mark potential verb forms, and even suggest meanings based on how words cluster together. It's not guesswork. It's structured linguistic modeling. Machines don't get tired. They don't overlook a stray symbol. And that matters when you're trying to reconstruct grammar from inscriptions carved over 2,000 years ago. The payoff, cultural depth. These aren't just phrases. They're windows into religion, trade, rituals, and identity. For centuries, the Atruscans spoke from the grave with no one to answer back. Now, with every computation, their voice gets clearer. And one thing's for sure, history isn't done talking, and neither are we. Don't forget to like, drop a comment with your

28:32 thoughts, and trust me, it only gets deeper from here.