AI & Archaeology · Classical Greece · Ancient Agora, Athens, Attica, Greece
AI may learn the design rules of ancient Greek buildings from Athens Agora records
Georgia Tech researchers have a $225,000 grant to build models that propose reconstructions of classical buildings, but their announcement reports no results yet.
Official announcement
Ancient Agora, Athens, Greece Researchers at Georgia Tech in the United States are building AI models to help reconstruct what ancient Greek buildings looked like. The models are meant to learn the design rules of temples, theatres, homes and other structures from excavation records. The project is at an early stage. The university's announcement describes no reconstructions or test results so far.
The project is a collaboration between Kartik Goyal and Myrsini Mamoli. Goyal is an assistant professor in Georgia Tech's School of Interactive Computing. Mamoli is an architectural historian and joint lecturer in the School of Architecture and the School of History and Sociology. Earlier this year they received a $225,000 grant from the National Endowment for the Humanities to support the collaboration.
Georgia Tech opens its announcement with the 2026 film adaptation of The Odyssey. Historians and classical scholars have argued over how accurately it shows ancient Greece. The university says the new models may soon weigh in on debates of that kind.
Records from the Athenian Agora
The data comes from the Ancient Agora in Athens, where Mamoli works in the summers. The American School of Classical Studies at Athens has excavated the site for almost 100 years. More than 180,000 artefacts have been found there. Every discovery since 1933 has a digitised record.
The Agora is where the political buildings of Athens emerged, and democracy was born, so it’s one of the most important sites in the classical world.
Myrsini Mamoli, architectural historian, Georgia Tech
Scholars reconstruct ancient buildings from the remains that survive, the excavation records and references in ancient texts. Mamoli says identifying all comparable structures and forming hypotheses about them takes years. Tracing design principles such as proportions and materials across the records is slow work.
Imagine if we could feed this data into an AI model that would automatically discover the connections.
Myrsini Mamoli, architectural historian, Georgia Tech
The approach rests on shape grammar. This is a visual, rule-based mathematical method that describes a style as a set of rules, which then generate designs in that style. Each architectural style has its own grammar. As a PhD student at Georgia Tech, Mamoli worked out these rules by hand to produce 2D visualisations.
Making these rules is time-consuming, so we hope with AI, a whole fresh approach can happen where AI can discover the rules by looking at the data at scale.
Myrsini Mamoli, architectural historian, Georgia Tech
Goyal has already used similar methods on early modern English printed books to identify where they were printed. He thinks the same process can work for classical buildings. Ancient texts describe fragmentary plans, and the models would infer patterns from them to propose reconstructions.
They make the model not just think about producing tokens or pixels, but also about the aesthetic and geometric features.
Kartik Goyal, assistant professor, School of Interactive Computing, Georgia Tech
No results yet, and a risk of bias
The announcement does not report how well the models perform. Goyal says he knows some humanities scholars worry that AI could compromise scholarly integrity. Large language models can struggle with cultural nuance. They also often insert biased hallucinations into their output.
If we run these models blindly, we risk incorporating undetected bias.
Kartik Goyal, assistant professor, School of Interactive Computing, Georgia Tech
Goyal says AI systems for the humanities need to be more transparent and easier to modify. Mamoli wants the models kept to a limited role as a tool.
I don’t see it replacing the scholarly work that is being done. I see it as assisting it.
Myrsini Mamoli, architectural historian, Georgia Tech
Mamoli has named one possible use. She works in Section Iota of the Agora, a disturbed area. It holds artefacts from all over the Agora, because people moved them there for secondary uses. She wants to know whether an AI model could map where these fragments came from.
What if an AI model could instantly create a map to trace that movement and show exactly where these fragments came from?
Myrsini Mamoli, architectural historian, Georgia Tech
Sources
- New AI Models Could Create More Accurate Visualizations of Classical Architecture (Georgia Tech College of Computing)
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…