Artificial intelligence (AI) is opening new windows into the past. Machine learning and 3D modeling are revolutionizing the field of archaeology, identifying burial sites in satellite images, classifying ancient pottery fragments, locating shipwrecks in sonar images, creating 3D digital reconstructions of historical sites, translating ancient languages, and finding artifacts that are being sold illegally on the web.
Satellite Imaging to Identify Tombs
Dr. Gino Caspari is a research archaeologist for the Swiss National Science Foundation, and his work focuses on the ancient Scythians, a nomadic people from 3,000 years ago with a rich culture and commanding empire, centered in modern-day Crimea. The tombs of Scythian royalty contained great wealth, and so, they were popular targets for tomb robbers. Caspari estimates that over 90 percent of those burial sites have now been destroyed.
However, Dr. Caspari believes that there are still thousands of tombs across the Eurasian steppes, which extend for millions of square miles. He mapped burial sites using Google Earth images of territory in modern-day Russia, Mongolia, and Western China’s Xinjiang province. “It’s essentially a stupid task,” Dr. Caspari lamented. “And that’s not what a well-educated scholar should be doing.”
Dr. Caspari turned to Pablo Crespo, a graduate student studying Economics at the City University of New York, who used artificial intelligence (AI) to estimate volatility in commodity prices. The two worked to build a convolutional neural network to sort through and search the satellite images, which would place them at the forefront of archaeological analysis.
A convolutional neural network (CNN) analyzes information that can be processed as a grid, so it is apt for analyzing photographs and images. Crespo and Caspari’s CNN gives each pixel in the grid a rating based on how red, green, and blue it is. It analyzes small groups of pixels then larger ones, looking for matches or near-matches to the data it was trained to identify.
The two researchers worked for months and ran 1,212 satellite images through the CNN, looking for circular stone tombs and avoiding other similar structures, such as construction debris and irrigation ponds. Testing on an area of around 2,000 square miles, the system correctly identified known tombs 98% of the time.
Dr. Crespo shared that, “Creating the network was simple.” He wrote the code in under a month using Python. The team hopes that the CNN will be a new tool in archeologists’ arsenal to find new tombs and historical sites to gain new insights into past civilizations and protect them from looters.
Classifying Ancient Pottery Shards
Two archaeologists at the University of Pisa in Italy, Gabriele Gattiglia and Francesca Anichini, excavate Roman Empire-era sites, analyzing thousands of broken bits of pottery. The Romans made almost all containers, such as cooking vessels and the amphoras used for shipping goods around the Mediterranean, out of clay, so pottery analysis is vital to understanding Roman life and culture.
Dr. Gattiglia and Dr. Anichini estimate that only 20 percent of their time is spent excavating sites, and the rest is spent analyzing pottery, comparing shards to pictures in catalogs. “We started dreaming about some magic tool to recognize pottery on an excavation,” Dr. Gattiglia shared. The dream culminated in the ArchAIDE project, a digital tool, developed by computer scientists at Tel Aviv University in Israel and funded by the European Union’s Horizon 2020 Research and Innovation Program, which allows archaeologists to photograph a piece of pottery in the field and identify it with a neural network.
The ArchAIDE project involves digitizing print catalogs and training CNN to recognize different types of pottery. As of now, the system can identify a few specific pottery types, but as more records and photographs are added and the database expands, that number will grow.
LiDAR Mapping to Find Archaeological Sites in Madagascar
The technology LiDAR measures distances and senses objects by pulsing laser light and measuring its reflection with a special sensor. Dylan Davis, a Ph.D. A candidate at Penn State University’s Department of Anthropology developed an algorithm using LiDAR data to map the forest floor in Madagascar. The system identified over 70 confirmed archaeological sites across a 1000 square kilometer area, finding settlements and earthen mounds created by prehistoric North American populations.
Underwater Robot to Find & Digitally Reconstruct Shipwrecks
In marine archaeology, travel to sites is often expensive and difficult, and divers cannot spend too much time underwater for the risk of severe pressure-related injuries. Chris Clark, an engineer at Harvey Mudd College, addresses both of these challenges: an underwater robot makes sonar scans of the seafloor, and a neural network searches the images for shipwrecks and other important underwater sites.
For the past few years, Clark has worked with Timmy Gambin, an archaeologist at the University of Malta, to scan the floor of the Mediterranean Sea surrounding the island. In 2017, it identified a World War II dive bomber off the coast of Malta, and using the sonar scans from the underwater robot, Clark and Gambin created a 3D digital reconstruction of the site.
AI Translates Ancient Languages
Researchers at the University of Chicago’s Oriental Institute and Department of Computer Science worked together to bring an AI system: DeepScribe, that decodes tablets from ancient civilizations to life. Archaeologists began incorporating computer programs into the study of ancient documents in the 1990s, but complex cuneiform characters and the 3D shape of tablets limited the technologies’ efficacy. The University of Chicago’s computer vision model was trained on more than 6,000 annotated images from the Persepolis Fortification Archive and a ‘dictionary’ of over 100,000 individually identified characters, and now, it can interpret the inscriptions on ancient tablets with close to 80% accuracy.
Similarly, Google’s DeepMind developed the neural network PYTHIA to ‘fill in’ missing inscriptions in ancient Greek on the damaged surfaces of stone and ceramic artifacts. PYTHIA, named after the Oracle at Delphi, “takes a sequence of damaged text as input, and is trained to predict character sequences comprising hypothesized restorations of ancient Greek inscriptions,” the researchers said.
Finally, Chinese researchers used a similar CNN to understand an ancient language found on 3,000-year-old tortoise shells and ox bones and gain insights into oracle bone morphology, a keystone in ancient Chinese culture.
Identifying Illegal Sale of Human Bones Online
Shawn Graham, Professor of Digital Humanities at Carleton University, modified Inception 3.0, a CNN developed by Google, and trained it on 80,000 images of human bones to now scan the internet and identify images and information related to buying and selling of human bones. Many countries, including the United States, required by law that human bones in museum collections should be returned to the person’s descendants, but there is an illegal online market violating these regulations.
“These folks who are being bought and sold never consented to this,” Dr. Graham said. “This does continued violence to the communities from which these ancestors have been removed. As archaeologists, we should be trying to stop this.”
Shawn Graham has since joined the Alliance to Counter Crime Online, an organization identifying illegal online activity and combatting human rights injustices, such as the ivory trade and sex trafficking.
3D Digital Modeling Preserves History
In 2010, French architect Yves Ubelmann took a picture of a small village in Afghanistan. Two years later, when he returned, the site was destroyed, and an elderly man, who remembered Ubelmann, asked for the photo. “The picture is the only link I have to my personal history,” the man said, and he encouraged Ubelmann to share the image with others. This small interaction catalyzed the creation of Iconemn, a Paris-based company creating 3D digital models of historic landmarks “threatened by war, conflict, time, and nature.”
Preserving history is urgent in countries like Syria, where all six of its UNESCO World Heritage sites, including centuries-old temples, mosques, citadels, bazaars, and tombs, were damaged by years of war and violence.
Iconemn’s drones capture thousands of images across 20 countries, and machine learning algorithms, powered by Microsoft AI, stitch those photos together into high-resolution 3D models to access the level of damage, reconstruct parts of the site, and immortalize its history.
Describing Iconemn’s work, “It’s a way to keep history alive,” says Ubelmann. “If you don’t know where you come from, you don’t know where you go.”
Archaeologists are in a race against the consequences of human activity and global rising sea levels, deforestation, and other ramifications of climate change to identify, protect, and study historical sites. If we cannot find these locations quickly enough, this history may be destroyed and thus, lost forever. Algorithms can sort through thousands of satellite images at a time and identify new historical finds, and advanced 3D modeling can digitally reconstruct artifacts and maps of archaeological sites, preserving them for generations to come. Further, virtual reality can allow us to step into the world of the past and put ourselves in the shoes of our ancient ancestors.