AI could solve the biggest archaeological mysteries

While relying on clever machines may lack the romance of crossing a desert by camel to discover unopened tombs or roaming through a jungle to discover an ancient temple, using artificial intelligence (AI) would have saved Indiana Jones the trouble of almost being crushed by a rolling stone, shot by mercenaries or being bitten by snakes while solving historical mysteries.

So, saddle up and take a tour of how AI is taking on some of the biggest unsolved historical cases, from deciphering scrolls and secret languages to dating mysterious remains.

In ‘The Last Crusade’, Indiana Jones reads an incomplete medieval stone tablet in Latin (with a little help from his father’s notebook) to reveal the location of the Holy Grail. Unfortunately, in real life, the solution doesn’t always present itself and information remains indecipherable even if it is set in stone, while delicate scrolls are unable to be opened without damage. That is, until the invention of AI.

Researchers at the University of Oxford, Ca’ Foscari University of Venice, and Athens University of Economics and Business are using a deep neural network named Ithaca, developed by Alphabet’s DeepMind, to study ancient Greek texts. Ithaca can not only restore missing text from inscriptions (so no need to risk life and limb finding Professor Henry Jones’ notebook) but also identify their original location with 71 per cent accuracy and establish the date they were written to within around 30 years.

Ithaca is trained on the largest dataset of Greek inscriptions. While standard language-processing models are trained using words – because the order in which they appear in sentences and the relationships between them provide extra context and meaning – Ithaca is also trained using individual characters as inputs, because ancient texts are often incomplete. The mechanism at the model’s core evaluates these two inputs in parallel, allowing Ithaca to evaluate inscriptions as needed, overcoming the challenge of damaged documents and missing chunks of text.

The team recently used Ithaca to redate a series of important Athenian decrees. Their age had been in dispute, but Ithaca dated them to 421 BCE on average, which aligned with new historical evidence that suggested they were probably written in the 420s BCE.

“This date shift has significant implications for our understanding of the political history of Classical Athens,” says Thea Sommerschield, Marie Curie Fellow at Ca’ Foscari University of Venice. “We hope that models like Ithaca can unlock the cooperative potential between AI and the humanities, transforming the way we study and write about some of the most significant periods in human history.”

Instead of the AI trying to put experts out of a job, Ithaca is designed to be used by researchers who can combine their historical knowledge with the tool’s assistive input. While Ithaca can achieve 62 per cent accuracy alone when restoring damaged texts, this jumps to 72 per cent with the help of historians.

“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, staff research scientist at DeepMind.

Machine learning is also being used to try and automatically translate long lost languages: those where we don’t know enough about their grammar, vocabulary, or syntax to be able to understand them.

British archaeologist Arthur Evans was at first baffled by a stone tablet he found in Knossos, Crete, in 1886 that bore inscriptions in an unknown language and turned out to be one of the earliest forms of writing. He collected more examples to realise he had found two scripts: Linear A, dating from between 1800 and 1400 BCE, when the Bronze Age Minoan civilisation ran the island, and Linear B, which appeared afterwards when the island was conquered by Mycenaeans from the Greek mainland.

For years, both were undecipherable, until a linguist named Michael Ventris cracked the code for Linear B, realising place names might provide a crib and that the language was probably an early form of ancient Greek. Computer scientists have proved AI’s ability to decipher lost languages by decoding Linear B, but so far, Linear A remains unbreakable to humans and machines.

The race is on to decode Linear A using AI. A team from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Yuan Cao from Google’s AI lab recently came up with a new system that has been shown to be able to automatically decipher a lost language without needing advanced knowledge of its relation to other languages, which offers hope in breaking Linear A, because nobody can work out the descendants of the ancient language.

The new system’s algorithm can determine relationships between languages and even assess the proximity between two languages. When tested on known languages it can accurately identify language families.

The team’s ultimate goal is for the system to be able to decipher lost languages using just a few thousand words, without connecting texts to related words in a known language, which is an approach known as cognate-based decipherment.

However, this is just one route. Other researchers are using computational analysis of the symbols of Linear A using natural language processing and data mining techniques to better understand the logogram language, in a bid to uncover interesting statistical and information-theoretic patterns even without deciphering the language, and they are making some progress.

While the content of Linear A remains a mystery, it seems that with the use of AI, the writing is on the wall (or stone tablet) that it will soon be deciphered.

“X never ever marks the spot,” Indiana Jones tells his students... before following Roman numeral clues including an ‘X’ in a Venetian library to locate a hidden passage. But when there’s no helpful clue, technology can help. Drones are often used to survey archaeological sites and would no doubt have helped Indiana Jones locate the ‘Well of Souls’ and other useful sites on his various quests for fortune and glory.

Experts from the British Museum and Iraqi archaeologists used a unique combination of remote-sensing work using high-resolution drone imagery, ground-truthing and conservation techniques to discover, identify and preserve buildings in Tello, Iraq.

The site of the ancient Sumerian city of Girsu  – one of the earliest known cities in the world – which is at least 4,500 years old, had been ravaged by destructive 19^th-century excavations and 20^th-century conflict, but drone imagery offered a fresh perspective, leading to the discovery of a lost 4,000-year-old palace of the kings of Girsu and precious artefacts, including clay tablets and seals.

While drones and other technology such as ground-penetrating radar are already proving successful in unearthing buried treasure, AI is typically faster at sifting through images, aerial photos and satellite data than humans, to spot potentially interesting archaeological sites from above and shed light on mysterious sites.

The Nasca Lines in southern Peru have puzzled experts since they were discovered in the 1920s. They depict shapes of varying complexity etched into the terrain, some several hundred metres in length, but nobody knows why there were created. By uncovering more of these mysterious formations, archaeologists hope to piece together clues about their existence.

In recent years, satellite-based or drone-based remote hyperspectral sensing and imagery have helped researchers discover hundreds of these figures thought to have been made between 100 BCE and 300 CE, covering an area of about 500 square kilometres south-east of Lima.

A research group led by Professor Masato Sakai of Yamagata University, Japan, and Peruvian archaeologist Jorge Olano, recently discovered 168 new geoglyphs of humans, camelids, birds, killer whales, felines and snakes on the Nasca Pampa, using field studies informed by high-​resolution aerial photos and drone images.

The team collaborated with the IBM Thomas J Watson Research Centre to conduct an AI-based study of the distribution of Nasca geoglyphs. Johannes Schmude, research staff member at IBM Research, wrote in a blog post that the experts used IBM PAIRS Geoscope, which is IBM’s cloud-based AI technology for scaling geospatial analytics to large and very complex data sets.

“For Yamagata’s research, PAIRS brings the unique ability to analyse massive and disparate geospatial and temporal datasets from a number of sources, including layers of lidar data – which is used to sense and examine the surface of the Earth – alongside drone images, satellite visuals and geographical survey information, to help reveal new lines and formations,” Schmude wrote.

Professor Sakai is preparing to publish details of AI-based geoglyph images, according to the university and it will be fascinating to see what the technology has added to the team’s work.

While Indiana Jones operates on instinct, most archaeologists favour precision, especially when dating remains and artefacts. Radiocarbon dating has proved to be a game-changing technology since Willard Libby proposed it in 1946, and it led to him winning the Nobel Prize  for chemistry in 1960. The technique compares three different isotopes of carbon found in all living things and allows experts to date organic material younger than 50,000 years, based on the chemical reactions the body exchanges with the environment after death. However, only around half of corpses can be dated using this method because it requires much organic material that may not have survived.

“Unreliable dating is a major problem, resulting in vague and contradictory results,” says Eran Elhaik, senior lecturer in population, medical and evolutionary genomics at Lund University in Sweden. To solve this problem, he has developed a method that uses AI to date genomes via their DNA “with great accuracy”.

Called Temporal Population Structure (TPS), it identifies changes in the patterns of mutations that increased and decreased over time. For example, the lactase persistence mutation that allows adult humans to digest milk has only increased in frequency since the Neolithic period. By looking at this mutation, the method can distinguish periods that predate or postdate the Neolithic.

“Combining time-relevant mutations allows us to distinguish historical periods by the exact pattern of these mutations. We can look at the DNA of each ancient individual and represent them in terms of these temporal patterns, much like we would do for ancestry, only now we represent individuals in terms of time,” Elhaik explains.

His team trained machine-learning algorithms to associate these temporal patterns with time, so that it can be used to date a skeleton by calculating temporal patterns and asking the machine learning – which was already trained on thousands of skeletons – to date a new ancient individual.

In the study, published in Cell Reports Methods, the research team analysed approximately 5,000 human remains, from the Late Mesolithic period (10,000-8,000 BCE) to modern times, to prove their method, demonstrating that all of the samples they studied could be dated with a rarely seen accuracy.

“We show that information about the period in which people lived is encoded in the genetic material. By figuring out how to interpret it and position it in time, we managed to date it with the help of AI,” says Elhaik.

The researchers do not expect TPS to eliminate radiocarbon dating but to be a complementary tool that could prove especially helpful when there is uncertainty involving a result. They are now using their method to date Viking skeletons and test the predictions against radiocarbon dating, as well as developing an “even more powerful method”.

In ‘The Dial of Destiny’, it looks as if Indiana Jones has come out of retirement (again) and still favours his old techniques, including wielding a whip to ward off the competition. AI tools are not being designed to put archaeologists like Jones out of work but to speed up and automate some of the more mechanistic parts of the job such as dating bones and opening delicate scrolls. AI could also help join up projects and analyse big data in a way that even the most talented archaeologists cannot, to pick out geographical and historical patterns.

While Indy may think the days of waking up in the morning and embarking upon a new adventure “have come and gone”, he is proved wrong in the new film, and wrong again if he thinks AI will rob real archaeologists of the chance to risk life and limb in the jungle or delve deep into historical mysteries underwater. After all, “we do not follow maps to buried treasure”, making human ingenuity and inspiration irreplaceable, even if AI can help point experts in the right direction.