Three-dimensional scanning and artificial intelligence are opening new avenues for preserving threatened Arab heritage, from creating accurate digital copies of artefacts and sites to suggesting visualisations of their missing parts.
But the expanding use of these technologies requires researchers to draw a clear distinction between what has actually been documented, what has been inferred from evidence, and what algorithms have produced as a visual possibility rather than a historical fact. Over thousands of years, stone inscriptions, temples, palaces, murals and manuscripts have preserved aspects of societies’ lives and beliefs.
Wars, earthquakes, storms, natural forces and looting have damaged many sites and artefacts, however, making the preservation of what remains of the Arab world’s heritage dependent on institutions’ and researchers’ ability to document it before further deterioration or loss. Laser scanning, photogrammetry and three-dimensional modelling have made it possible to convert an artefact into a set of digital data retaining its measurable dimensions and details.
This data can be used for study, comparison, display and restoration, and remains available even if the physical original is later damaged. Artificial intelligence adds another stage to this process: it can analyse and link images, documents and scans, then produce visualisations of damaged or missing elements.
An Egyptian-British centre for heritage digitisation
But its ability to produce a convincing appearance does not necessarily mean that the result is based on archaeological evidence. This makes documenting the source of each element and the degree of certainty associated with it an essential part of any virtual reconstruction project. In Egypt, the use of these tools took on an institutional form with the launch of the Egyptian-British Centre for Artificial Intelligence and Virtual Heritage Technologies, “BECAIVE”, in Cairo during August 2026.
The centre was established through a partnership between Britain’s University of York and Egypt’s National Research Institute of Astronomy and Geophysics. It focuses on using artificial intelligence and advanced digital technologies to preserve heritage and antiquities, study and introduce museum collections and archaeological sites, and engage the public with them.
It is also developing digital frameworks to help link artefacts to the sites from which they came and to the related historical contexts. The digital model therefore becomes more than a three-dimensional image prepared for display; it becomes part of a broader system for gathering, studying and reusing information in heritage research.
Tanit XR documents Tunisian heritage
This approach makes it possible to treat an artefact as part of a site, a historical period and a network of relationships, rather than as an object separate from its surroundings. In Tunisia, immersive artist Ines Saied, founder of the global non-profit Tanit XR community, is digitising artefacts and sites from Tunisian heritage. The project uses extended reality, three-dimensional scanning and immersive storytelling to preserve threatened heritage and present it in digital formats.
Saied says the primary value of digitisation lies in protecting information associated with artefacts, because antiquities located in open sites remain exposed to changing weather, climate change, looting and other dangers. When an artefact is scanned in three dimensions and preserved digitally, the project team has an accurate copy that is not exposed to the same physical risks surrounding the original.
According to Saied, the benefits are not limited to preserving data. Digitisation also allows access to artefacts that may be difficult to display inside museums. She notes that some digital documentation can now be carried out using phones, expanding the scope for individuals and volunteers to participate in collecting heritage-related data.
Saied cites the archaeological site of Neapolis in the Tunisian city of Nabeul, near the area where she grew up. Storm Harry exposed parts of the site, leaving them more vulnerable to damage and prompting the Tanit XR team to scan and preserve them digitally.
Saied believes that three-dimensional models can reopen heritage to the public and researchers and make it easier to share Tunisian heritage with the world, turning it into a field for cultural exchange. Digitisation can also reconnect younger generations with their heritage by involving them in studying artefacts and preparing them for digital display. Her team works with volunteers to study artefacts, improve digital models and prepare them for use in interactive experiences.
Controls on reconstructing missing parts
This work combines documenting an artefact’s physical form with presenting it in environments that allow the public to interact with it, rather than limiting access to its presence at an archaeological site or in a museum gallery. But dealing with missing parts remains one of the most sensitive aspects of digitisation and virtual reconstruction.
Saied stresses that preserving an artefact as it is must remain the priority. Her team therefore does not automatically fill gaps that appear during scanning. If data is missing, it remains missing in the original documentation copy. When an interactive experience or display material requires a missing part to be reconstructed, the team ensures that viewers are told that the part does not belong to the original scan data.
Tanit XR uses a different, simple colour for reconstructed elements, rather than making them visually identical to the original parts in a way that could blur the distinction between documentation and supposition. The teams also retain two versions of the digital artefact: a raw version representing the scan results as they are, and an enhanced version prepared for use in interactive experiences.
This separation allows the original data to be revisited rather than replaced by the processed version, which may include adjustments linked to display or improved appearance. Preserving raw data becomes increasingly important as tools develop, since it can be reprocessed in the future with more advanced software to produce better models.
Saied says that, during a previous conversation with a specialist at the Smithsonian Institution, she was advised to retain the raw images of every artefact that could be photographed, so they could be used again as three-dimensional modelling technologies advance.
Applying this practice on a large scale requires greater resources and support, according to Saied, particularly because much of the work relies on volunteers. Preserving and organising raw files, improving models and preparing them for interactive use require ongoing time, technical capacity and storage. The digital model’s function is not limited to virtual display.
Saied explains that three-dimensional scanning can serve as a reference in restoration work because it contains measurable details and allows measurements to be taken and compared with the artefact’s original form. But she stresses that the accuracy of the results depends on the type of scanning tool used, and that the final restoration decision remains the responsibility of specialists. Archaeologist Laura K.
Harrison, president of Archaeo Consulting and former director of the Axis 3D Laboratory at the University of South Florida, affirms the practical value of this type of data. She says detailed three-dimensional scans, together with the raw data and accompanying metadata, provide measurable references that are highly useful for conservation and restoration work.
Harrison cites her involvement in sharing three-dimensional lidar scans of Tampa Union Station, designated a National Historic Landmark in the United States, with an architectural firm working to restore the building’s windows according to historical standards. The architect extracted precise measurements from the scan and used them to prepare the architectural drawings for the new windows installed later.
The value of the data also extends to monitoring sites over time, according to Harrison. She considers it important to retain a complete record of everything remaining at a site, including the built heritage and the surrounding geographical or urban context when it is relevant to understanding and protecting it. Repeated scans at regular intervals can reveal changes affecting sites.
Lidar devices can detect differences at millimetre level, changes that may be difficult to identify with the naked eye or from photographs alone. Comparing successive scans makes it possible to monitor cracks, movement or gradual deterioration.
Harrison describes laser scanning and three-dimensional documentation as proactive conservation measures because they create a digital archive that may survive longer than the physical sites themselves. Those sites face multiple threats, including armed conflicts, industrial expansion and climate change. Nevertheless, Harrison stresses that digital copies do not replace or substitute for original heritage sites.
Their function is to support interpretive analysis and comparative research and make knowledge available to future generations, while the original site’s physical and historical value remains separate from the value provided by its digital documentation. Harrison points out that buildings and three-dimensional records cannot, on their own, tell the history of a place.
These records remain silent without the contextual information provided by stakeholders and communities connected to the sites, with their perspectives, memories and histories. She therefore believes that documenting sites should be accompanied by collecting the stories and memories associated with them, including oral histories.
Museums, exhibitions and virtual collections can incorporate this human dimension into digital copies, so that heritage is not represented solely through stone and geometric form. Alongside documenting what exists, artificial intelligence expands the scope for visualising elements that are no longer standing.
Archer project tests AI reconstruction
The Archer project is prominent in this context. It is a European research project launched in 2019 to use artificial intelligence and deep learning to automate some stages of the virtual reconstruction of architectural heritage. The project seeks to move beyond traditional digital reconstruction methods, which are time-consuming and require substantial manual intervention, towards models that analyse architectural patterns and styles and help produce visualisations of damaged or missing elements.
This does not eliminate the researcher’s role, but allows multiple possibilities to be examined more efficiently. During its various stages, Archer used technologies including GANs to reconstruct two-dimensional elements and transformer models to process text and images. The project reached newer techniques for three-dimensional reconstruction, such as Gaussian splatting and SuGaR.
But the importance of these experiments lies not only in producing realistic-looking images or models. It also concerns the question of the limits of historical evidence when parts of a site are missing or have not been fully documented. A model may produce a visually coherent result even though some of its details lack material or archival support.
This dilemma emerged in a study that tested the use of artificial intelligence to reconstruct missing parts of the Temple of Bel in Palmyra, Syria. The research framework used, known as HARF, was originally developed in the context of the Bamiyan Buddhas and was then tested on a limited basis on the temple in Palmyra.
The test relied on documentary information concerning the temple’s dimensions, the type of stone used, the Corinthian order, its layout, inscriptions and decorative programme. 32 people from several disciplines took part in assessing the results. Architectural elements such as Corinthian columns received more positive ratings than others, while inscriptions, ornaments and surface details proved more difficult.
The study does not indicate that artificial intelligence failed to produce convincing images. Rather, it shows that visual plausibility alone is not evidence of historical accuracy. Details that appear consistent with a building or period may be probabilistic outputs created by the model, without being documented in physical remains or historical records.
The researchers stress that the outputs of generative models are probabilistic and should not be treated as evidence-based historical reconstruction unless they are supported by physical remains, archival documents or recognised principles of historical reconstruction. The study also does not establish that the Temple of Bel can be fully reconstructed archaeologically; it tests, on a limited basis, HARF’s ability to generalise between different sites.
Paolo Clini, a professor at Italy’s Marche Polytechnic University, director of the MAND Lab and a specialist in artificial intelligence and digital cultural heritage, says virtual reconstruction of lost heritage existed before the emergence of current artificial intelligence applications.
But Clini believes that the new tools have strengthened the ability to link scattered evidence, including historical photographs, drawings, archival documents, scans and surviving remains. They have also enabled researchers to explore different reconstruction possibilities in less time and more efficiently than some traditional manual methods.
He stresses the need to distinguish between using artificial intelligence to extract information from sources and using generative artificial intelligence to suggest missing elements. He says that the form produced by the system, however convincing it may appear, is not in itself evidence of what actually existed.
According to Clini, researchers must state the basis for each element within a reconstruction: whether it is directly documented, inferred from the available evidence or represents a hypothetical visualisation. They should also document the sources used and artificial intelligence’s role in building the model, and show which parts remain uncertain, particularly when more than one possible visualisation exists.
He stresses that details generated by algorithms should not be treated as independent historical evidence. Digital models should also remain open to review and modification whenever new evidence emerges, rather than being presented as a final and fixed version of the past.
For threatened or destroyed heritage in the Arab world, Clini believes these tools can support the study of lost architectural relationships, comparison of different historical periods and the planning of conservation work. They can also provide access to heritage for communities whose members are unable to visit the original sites.
Clini links the success of these projects to involving local researchers, heritage institutions and communities connected to the sites. Digital reconstruction concerns not only architectural form, but also the meanings and memories that a place holds for those who lived around it or whose history is connected to it. He also cautions against broadly applying the term “digital twin” to different virtual models.
He believes the term should be limited to a model that maintains an ongoing connection with data from its physical counterpart, while a model that reimagines a destroyed building should clearly be described as a virtual reconstruction. According to these applications, digitisation provides a means of preserving what remains of heritage and making it available for study and access even if the original is damaged or lost.
Artificial intelligence can suggest images of what no longer exists, but it does not turn possibility into historical fact. Therefore, the more convincing models become, the greater the need to show the boundaries between documentation, inference and algorithmic suggestion.
