Experts Explain How AI Helps Preserve Historic Heritage, and Why People Still Lead
The Center for Heritage Conservation’s 2026 symposium brought researchers and practitioners together to discuss AI’s most useful roles and its limits in the field.
Panelists take part in a roundtable discussion during the Center for Heritage Conservation’s 27th Annual Historic Preservation Symposium Feb. 13-14 in Geren Auditorium at Texas A&M University.
Preservation teams now use artificial intelligence to record disaster damage, monitor archaeological sites and inspect building exteriors on a scale that was not possible 10 years ago. AI helps teams move faster, but people still decide how to move forward.
The shift inspired the 27th annual Historic Preservation Symposium, titled “Preserv[AI]tion,” hosted by Texas A&M University’s Center for Heritage Conservation (CHC) Feb. 13-14 in Geren Auditorium. Architects, engineers and researchers shared case studies from the field and discussed where AI fits in daily practice.
Speakers agreed that AI works best when teams use it for specific tasks. It can help teams collect and analyze large amounts of information, but the interpretive and ethical judgments at the heart of preservation work remain distinctly human.
Key Takeaways From the 2026 Historic Preservation Symposium
The 2026 Historic Preservation Symposium was curated by the CHC Director, Dr. Fabrizio Aimar. The day-and-a-half symposium included five presentations, a roundtable discussion and a student poster session. Speakers shared examples that showed how teams can respond faster after disasters, track changes over time and focus limited time and funding where they matter most.
Here are six key takeaways from this year’s symposium:
1. Education and Ethics Will Shape How the Preservation Uses AI
Engineer Stephen Lucy moderated a roundtable with Philip Chen of Annum Architects, Andrew Adkinson of GFF Design and Tom Whitmore of The Christman Company. Texas A&M architecture professors Andrew Hawkins and Benjamin Ennemoser also joined the discussion.
Speakers said AI could reduce some early-career tasks that teach young professionals to spot mistakes, judge risk and build good judgment through repetition and time in the field.
Panelists said preservation teams will get the most value from tools trained for a specific job, not general-purpose models. Ennemoser explained that general AI tools often know little about preservation because they were trained on limited preservation material.
Whitmore added that field experience still matters because many preservation decisions happen at the building surface, where skilled observation guides what comes next.
“I want people who work for me to be exposed to the actual work and the nuances of that craftsmanship,” Whitmore said. “How those decisions are being made right there at the face of the wall.”
Panelists also discussed using a tool that can search the Secretary of the Interior’s Standards for the Treatment of Historic Properties using only that text as its source. Adkinson said teams could use AI to review draft National Register nominations and flag missing information or contradictions, instead of writing the documents from scratch.
“The technology inherently is a yes-sayer,” Ennemoser said. “I like the ‘brutally critique’ approach way better.”
Stephen Lucy moderates the roundtable discussion during “Preserv[AI]tion,” the Center for Heritage Conservation’s 27th Annual Historic Preservation Symposium Feb. 13-14 in Geren Auditorium at Texas A&M University.
2. AI Can Speed Up Disaster Documentation When Time is Tight
Several speakers shared examples of times when communities needed fast information to prevent avoidable loss.
Dr. Joe Kallas, a UNESCO cultural heritage specialist and architect, described emergency documentation efforts after the 2020 Beirut port explosion.
His team documented 240 heritage buildings damaged by the earthquake using photogrammetry, a technique that uses overlapping photos to create a 3D model. Those models helped teams decide how to stabilize 60 high-risk buildings within one month.
“Too often, historic buildings are lost — not because they are beyond repair, but because we didn’t have the time or the right information at the right time,” Kallas said.
Kallas also presented a machine learning framework trained on post-disaster imagery that identifies multiple damage types on unreinforced masonry and maps what it finds in photos onto 3D building models. The system reached 88% accuracy when tested on photos it had not seen before.
3. Large Sites Use AI to Track Change and Prioritize Inspections
At complex sites, AI tools can help teams track conditions over time and decide where to send staff first.
Alessandra Zambrano, responsible for the Research & Innovation office and the engineering official at the Archaeological Park of Pompeii, shared a layered monitoring system designed for scale.
Pompeii has 1,221 buildings across 44 hectares and can receive up to 20,000 visitors per day. The site faces seismic risk, water infiltration and weathering.
The park uses satellite analysis to track ground movement and drone surveys to document roofs and drainage. Teams use tablet-based inspections to record damage type and severity in real time. Zambrano said a full-site survey completed in 2025 produced more than 70,000 inspection records.
Zambrano also described research partnerships that test robotics and automation, including a four-legged robot that repeats inspection routes and a project that uses AI and robotics to reconstruct shattered frescoes from stored fragments.
Attendees gather in Geren Auditorium for a presentation by Alessandra Zambrano during “Preserv[AI]tion,” the Center for Heritage Conservation’s 27th Annual Historic Preservation Symposium at Texas A&M University.
4. AI Can Help Find Heritage That Was Never Documented
Experts said that some of the highest-stakes preservation work begins before anyone knows what exists.
Dr. Bilge Kose is a landmarks preservationist with the New York City Landmarks Preservation Commission and an adjunct assistant professor at Columbia University. Kose addressed how teams often work in places where no one has formally documented heritage.
Her project, Preserve AI, began after the February 2023 Kahramanmaras earthquakes in Turkey.
The project uses GIS mapping tools and machine learning to flag buildings that appear consistently across decades of aerial photographs. It helps teams identify potentially historic structures when on-site access is limited.
“AI is not a replacement for preservationists,” Kose said. “It is a scout that scales our eyes. It processes the haystack to find the needles, the heritage sites waiting for our professional attention.”
5. Photo Tools Help, but Fieldwork Still Finds Problems AI Cannot See
Presenters cautioned that AI can help teams triage large image sets, but it cannot replace hands-on investigation or explain why a building failed.
Robert Kornfeld, a historic preservation specialist at Thornton Tomasetti, said AI can help teams sort photos and flag areas that need a closer look. His firm developed the T2D2 damage detection platform, which uses convolutional neural networks to flag damage conditions across large image inventories.
Kornfeld said the tool can identify what warrants a closer look, but it cannot diagnose causes, understand building type or recommend repairs.
He illustrated the limits with a project at the New York Central Building, where physical probing revealed severely corroded internal steel concealed behind terracotta columns. A professional identified the risk through hands-on investigation and interpretation of crack patterns.
Kornfeld also said drone-based facade mapping has already changed practice by creating accurate, zoomable records for buildings where teams once relied on street-level observation.
6. Preservation Still Depends on People, Meaning and Future Use
Philip Chen, principal of Boston-based Annum Architects, argued that preservation often succeeds when buildings stay in use. He shared projects that updated historic spaces while meeting new program needs, including a warehouse conversion at Springfield Technical Community College and work at Harvard Divinity School.
Chen said some limits have nothing to do with technology. He said tools can generate language that appears correct but misses cultural context and meaning.
“Human intelligence, at least now, is still important when it comes to processing meaning, processing values, processing morality,” Chen said.
Learn more about the Center for Heritage Conservationat Texas A&M University’s Department of Architecture.