AI & Archaeology · Prehistory to the Middle Ages · New York University, New York, New York, United States
NYU and Tübingen build open-source AI to reassemble broken stone tools, pottery and bones
Funded by Schmidt Sciences, the team will gather at least 100,000 real and simulated broken objects, and archaeologists will check the reconstructions the AI proposes.
Official announcement
New York University, New York, United States Engineers and archaeologists at New York University are building open-source AI tools that piece broken stone tools, pottery and bones back together. The software is also meant to fill in what is missing. The aim is to turn fragments that sit unmatched in collections into evidence about how people lived in the past.
The project is called "Fragmented Legacy: Turning Broken 3D Materials into Archaeological Interpretations". It includes partners at the University of Tübingen in Germany and has a three-year grant from Schmidt Sciences. It is one of 19 projects chosen for this year's Humanities and AI Virtual Institute (HAVI) awards. The announcement describes a funded plan and reports no results from the new models yet.
Why most broken finds stay unmatched
Stone tools, pottery, animal bones and human skeletal remains often come out of the ground broken. Putting the fragments back together is called refitting. It is slow, so most collections are never systematically refitted. Reported success rates run from about 1 percent for ceramics to between 4 and 26 percent for bones.
A single match can carry a lot of information. Suppose a stone flake from one part of a site fits back onto a core found in another. Researchers can then learn how people moved through that space or how tools were made step by step. A refit can also show whether the site's layers were disturbed later.
Refitted animal bones can show how an animal was butchered and how the meat was shared. Refits of human fossils can reveal important anatomical features. Without refits, many of these questions go unanswered.
Refitting is fun, but we rarely have time to do it right. In museums, one can often only look at one drawer at a time, so it's impossible to lay out an entire collection at once. Refitting objects between sites to show direct relationships between them has only ever been done twice.
Radu Iovita, professor in NYU's Department of Anthropology
Simulated breaks, then checks by archaeologists
The team will first build a collection of at least 100,000 real and digitally simulated broken objects. It will pair 3D models with photographs, descriptions and catalogue data. The simulated breaks follow how stone, clay and bone actually fracture.
Next, the researchers will develop AI models that work out where each fragment belongs. The same models will generate a plausible version of the complete object. They will reason across shape, images and text, so that archaeological knowledge can rule out unlikely reconstructions.
Archaeologists will stay involved throughout. They will compare competing reconstructions and teach the system which ones hold up. The rebuilt objects are proposals, and the announcement gives no accuracy figures or comparison with human refitters.
My lab teaches machines to understand 3D shapes in the real world, where things are messy and incomplete. Broken archaeological material is one of the hardest versions of that problem. The pieces are worn, many are missing, and there's no picture on the box. Our job is to build AI that can reason through that uncertainty, and our archaeology colleagues tell us whether the answers make sense.
Chen Feng, Institute Associate Professor, NYU Tandon School of Engineering
Feng is founding co-director of the NYU Center for Robotics and Embodied Intelligence (CREO). He leads the project with Jing Zhang, a CREO Faculty Fellow. They work with Iovita and Scott Williams, both professors in NYU's Department of Anthropology. Raiko Krauß, professor of Central and Southeast European Prehistory at Tübingen, is also on the team.
Leiden University, Arizona State University and the Max Planck Institute for Evolutionary Anthropology are collaborators. The project grew out of work between Feng's AI4CE Lab and Iovita's lab. That collaboration began about four years ago with LUWA, which used AI to classify microscopic wear on stone tools. Zhang, then a postdoctoral researcher in both labs, connected the two groups.
In 2024 Zhang asked Iovita about archaeology's most pressing technical problem, and Iovita named refitting. That led to GARF, which reassembles broken objects. It also led to CRAG, which reassembles objects and fills in missing pieces.
Archaeologists are like Sherlock Holmes, piecing together clues to understand the past. We want Fragmented Legacy to be their Watson. It can match fragments and suggest possible reconstructions, so researchers have more time to interpret the evidence and ask new questions.
Jing Zhang, CREO Faculty Fellow, NYU
All datasets, model weights and benchmarks will be released under permissive open-source licences. University researchers can use them directly. So can cultural resource management firms, which carry out much of the archaeology required ahead of construction in the US and abroad. The team expects the methods to be useful in other sciences and in practical repair and restoration.
Sources
- NYU Researchers Are Building AI to Piece Together Broken Archaeological Finds (New York University)
Spotted a mistake? Tell the editor about a spelling error or a fact that is wrong. We correct stories openly.
Readers' picks show on the front page.
Follow @AncientHeraldHQ on X for the day's finds as they come in.
Comments
House rules. Stay on the subject, be kind to other readers and say where your facts come from. No advertising and no personal attacks. The editor reads along and hides comments that break these rules.
Loading comments…