Old ruins and ancient artifacts have long intrigued archeologists. Such sites, and the treasures they contain, offer fascinating clues about past civilizations. Yet, over the centuries, the environment inevitably takes its toll and leaves many sites and objects in a state of decay. Reconstructing what a temple, village, or piece of pottery originally looked like can be extraordinarily difficult.
Fortunately, artificial intelligence (AI) is unearthing clues that allow archeologists to peer deeper into the past. Using algorithms, neural networks, satellite data, drones, lasers, LiDAR, photos, and an assortment of other tools, researchers are reconstructing sites and reassembling artifacts algorithmically. Enhanced imagery, three-dimensional (3D) models, and digital twins are reinventing the field.
“Humans alone can’t classify the vast amount of data that exists at archeological sites,” said Eric Poehler, a professor of classics at the University of Massachusetts, Amherst, and a digital archeologist who studies Pompeii, Italy. “With digital tools, we are able to work at a scale that otherwise wouldn’t be possible.”
Added Satoshi Tanaka, a professor in the College of Information Science and Engineering at Ritsumeikan University in Kyoto, Japan: “AI and 3D scanning techniques are helping us see what sites and structures looked like in the past. They are helping us protect cultural heritage and develop virtual reality models.”
Sites Unseen
Understanding what ancient temples and artifacts originally looked like is a daunting task. In many cases, it’s nearly impossible to discern the exact details of original structures, reliefs, objects, and art. Adding to the challenge: some critical elements might be hidden beneath the surface of the earth or remain inaccessible because the sites have been modified.
Old photos and archeological expertise only go so far. “Digital technologies and computer graphics began to appear in the 1980s and introduce the idea of virtual archeology,” said Maurizio Forte, a professor of classical studies, art, art history, and visual studies at Duke University. “Machine learning, AI, and newer forms of digital technologies are now revolutionizing the field. They are making it possible to see things and recreate landscapes and environments that are beyond the scope of humans alone.”
Increasingly, archeologists complement old photos with images captured by drones, satellites, and sensors using magnetometry, photogrammetry, LiDAR, and other sensing methods. For example, a drone can capture upwards of 5,000 images in 45 minutes, Forte said. AI can sort through these images—and associated data—to spot patterns, relationships, and associations that would otherwise go undetected.
Digital Relief
For example, Tanaka and a team of multinational researchers have developed a method that generates 3D digital twins from old two-dimensional (2D) photographs of damaged or obscured relief carvings. In the past, reconstructing depths precisely from photographs was impossible because researchers could not achieve the necessary level of accuracy. However, unlike 3D sculptures, carved reliefs possess shallow depths that represent fine details. Tanaka’s multitask neural network uses semantic segmentation, depth estimation, and soft-edge detection to extract soft-edge data from a 2D image to create a detailed edge map.
This means a single image can provide sufficient data to reconstruct an object’s original appearance, including structure and depth. “AI analyzes visible surface details, learns the patterns, and reconstructs the depth to generate a 3D relief,” Tanaka said. The team has also developed a complementary mathematical technique that pulls data from underground scans to recover 3D foundational information about structures.
“By integrating various types of data reconstructed from damaged or hidden structures with data representing the existing and visible portions of cultural heritage, we can create complete digital twins of a heritage site and realize immersive virtual reality and metaverse models,” he explained.
In 2024, he and his team tested the technique at Borobudur Temple, a UNESCO World Heritage Site in Indonesia. The study was part of an international collaboration between Ritsumeikan University in Japan, the National Research and Innovation Agency in Indonesia, and the Borobudur Conservation Office in Indonesia. Using a single 134-year-old photo, the researchers managed to recover 97% of the information about the original stone relief carvings, including portions obscured by modifications made during the Dutch Colonial era.
“Our computer visualizations and virtual reality tools provided a way to explore the site in new ways,” Tanaka explained.
Finding Structure
Sites like Pompeii are also getting a digital makeover. For example, Poehler has turned to AI to enable large-scale analysis and pattern recognition. His Pompeii Artistic Landscape Project uses machine learning to spot complex patterns, categorize artworks, and even reconstruct missing details that span more than 100,000 individual walls and over 10,000 pieces of art.
Poehler began working on the project with a low-resolution data set that contained approximately 70,000 images, including some in black and white. Later, the archeological site permitted researchers to access high-resolution images. Matching images in the two sets was next to impossible. “Without any naming schema, it would have been necessary to manually map and connect the two sets of data,” Poehler said.
Working with the University of Massachusetts data science center, a team of specialists assembled training sets. Eventually, the machine learning model reached a 98% level of accuracy. “A painstaking process that could have taken years took place over several months,” he said.
Poehler and his team have used the resulting output to look for patterns in wall paintings across the entire ancient city of Pompeii. The technology has deepened their understanding of the ancient site—and its art and culture. “We suddenly have an order of magnitude more information and evidence,” Poehler said.
Picture Imperfect
Forte, who studies Etruscan sites using digital tools and AI, says machine learning and AI are not perfect. “It is still an interpretation process. We cannot have 100% certainty that any system is able to represent the exact way an antiquity originally looked.” Nevertheless, the benefits for both quantitative and qualitative analysis are clear. AI and digital tools fill gaps. “In many cases, we’re able to explore ideas and think in new ways,” Forte said.
Digital archeologists envision a future in which AI further interconnects algorithms and artifacts by mashing up even more diverse types of data from photos, sensors, laser scans, semantic data, and other datasets. This integration of technology and humanities has the potential to reshape our understanding of the past and provide a more comprehensive picture of ancient civilizations and human history.
“AI and digital tools allow us to preserve and share the great treasures of the world in ways that weren’t possible in the past,” Tanaka said.
Samuel Greengard is an author and journalist based in West Linn, OR, USA.
