| Madinah Erth |
2026 - present
Norman B. Leventhal Center for Advanced Urbanism &
Civic Data Design Lab
Sarah Williams (Principal Investigator)
Maysaa Sati (Project Manager)
Ashley Louie
King Abdulaziz City for Science and Technology (KACST)
Dr. Mohammed Alhamadi (Principal Investigator)
Haifa Alabduljabbar
Noura Alsomaikhi
Norman B. Leventhal Center for Advanced Urbanism &
Civic Data Design Lab
Sarah Williams (Principal Investigator)
Maysaa Sati (Project Manager)
Ashley Louie
King Abdulaziz City for Science and Technology (KACST)
Dr. Mohammed Alhamadi (Principal Investigator)
Haifa Alabduljabbar
Noura Alsomaikhi
Overview
As cities transform under globalization, identifying Intangible Cultural Heritage (ICH) is essential to ensuring that planning protects the everyday places and practices that shape a community’s identity. Yet living heritage is difficult to document. Participatory methods such as interviews and mapping workshops offer deep insight but are time- and resource-intensive, limiting how widely they can be applied.
Madinah Erth (إرث, “heritage”) examines whether artificial intelligence can complement these methods by identifying expressions of living heritage in social media. A collaboration between MIT’s Norman B. Leventhal Center for Advanced Urbanism (LCAU) and the King Abdulaziz City for Science and Technology (KACST), the project builds on the [Living Heritage Atlas | Madinah], extending its community-defined categories into a computational framework. Madinah, a city transforming rapidly through development and religious tourism, is the first case.
Approach
The project uses large language models (LLMs) to identify culturally significant practices in Arabic social media posts, beginning with X. Because keywords alone cannot distinguish a heritage practice from a casual or commercial use of the same word, the model draws on a layered knowledge base built from the Atlas data dictionary: definitions, everyday phrases, look-alike text that is not heritage, and cues that confirm it.
![]()
The Madinah Erth knowledge base. Four layers of contextual knowledge progressively narrow retrieval and increase confidence in identifying intangible cultural heritage.
A retrieval-augmented generation (RAG) classifier uses this knowledge base to sort posts into heritage, ambiguous, or noise. Researchers re-evaluate ambiguous cases to refine the model, and results are validated against a gold standard hand-coded by Arabic-speaking researchers. Classified posts are then linked to locations across the city.
![]()
The classification pipeline. Ambiguous results are re-evaluated and fed back to refine the model.
Early Findings
Preliminary results suggest LLMs can identify living heritage at scale and perform better when grounded in contextual definitions. But social media tends to amplify publicly promoted, recreational, and collective activities, while intimate, local traditions remain less visible, a reminder that data generated online differs from knowledge gathered through human interaction.
From Data to Planning
The first year focuses on building and validating the framework and comparing its results with the Atlas’s participatory findings. The second year translates these findings into an interactive platform developed with the Madinah Region Development Authority to support planning conversations. The project’s central contribution is showing how AI and participatory methods can work together: AI extends the reach of heritage identification, while direct engagement remains essential for planning that includes the traditions communities hold most closely.
As cities transform under globalization, identifying Intangible Cultural Heritage (ICH) is essential to ensuring that planning protects the everyday places and practices that shape a community’s identity. Yet living heritage is difficult to document. Participatory methods such as interviews and mapping workshops offer deep insight but are time- and resource-intensive, limiting how widely they can be applied.
Madinah Erth (إرث, “heritage”) examines whether artificial intelligence can complement these methods by identifying expressions of living heritage in social media. A collaboration between MIT’s Norman B. Leventhal Center for Advanced Urbanism (LCAU) and the King Abdulaziz City for Science and Technology (KACST), the project builds on the [Living Heritage Atlas | Madinah], extending its community-defined categories into a computational framework. Madinah, a city transforming rapidly through development and religious tourism, is the first case.
Approach
The project uses large language models (LLMs) to identify culturally significant practices in Arabic social media posts, beginning with X. Because keywords alone cannot distinguish a heritage practice from a casual or commercial use of the same word, the model draws on a layered knowledge base built from the Atlas data dictionary: definitions, everyday phrases, look-alike text that is not heritage, and cues that confirm it.
The Madinah Erth knowledge base. Four layers of contextual knowledge progressively narrow retrieval and increase confidence in identifying intangible cultural heritage.A retrieval-augmented generation (RAG) classifier uses this knowledge base to sort posts into heritage, ambiguous, or noise. Researchers re-evaluate ambiguous cases to refine the model, and results are validated against a gold standard hand-coded by Arabic-speaking researchers. Classified posts are then linked to locations across the city.
The classification pipeline. Ambiguous results are re-evaluated and fed back to refine the model.Early Findings
Preliminary results suggest LLMs can identify living heritage at scale and perform better when grounded in contextual definitions. But social media tends to amplify publicly promoted, recreational, and collective activities, while intimate, local traditions remain less visible, a reminder that data generated online differs from knowledge gathered through human interaction.
From Data to Planning
The first year focuses on building and validating the framework and comparing its results with the Atlas’s participatory findings. The second year translates these findings into an interactive platform developed with the Madinah Region Development Authority to support planning conversations. The project’s central contribution is showing how AI and participatory methods can work together: AI extends the reach of heritage identification, while direct engagement remains essential for planning that includes the traditions communities hold most closely.