France is scanning its cathedrals, restoring its paintings with AI film, and digitizing 25 million documents. It's calling it preservation. It's a backup.
There are two Notre-Dames now.
One is made of stone and oak and lead, and it nearly burned to the ground in 2019. The other is made of data, and it already existed before the fire — which is the only reason we know exactly what the first one looked like. The second Notre-Dame is the one you can walk through on a laptop. The first one is the one that matters.
Livio De Luca, research director at the CNRS, led this wild, sprawling team that built what they call a "digital twin." It's not a pretty render. It's a collaborative platform, a kind of Google Earth for Notre-Dame. You click on a beam in the roof frame and you get everything attached to it — dendro dating, analysis of charred wood, 3D scans, old photos. It's designed to keep absorbing information forever. Not an image. A memory palace made of point clouds and metadata.
The timing is what kills me. Art historian Andrew Tallon had already laser-scanned the whole cathedral back in 2010, every tilt of every stone. Without that scan, the reconstruction would have been guesswork. With it, the digital twin wasn't just a record. It was a guide.
And then the Louvre did almost the opposite. Not preserving what's there, but showing you what used to be there.
Since February this year, you can point your phone at six masterpieces everyone walks past and watch them come back to life. The Kore of Samos — this Greek marble girl who's been headless and armless since 1881 — is suddenly in color, based on the Louvre's archives of fabrics and pigments, holding whatever she was holding. The bust of Akhenaten, broken for thousands of years, is complete. The Code of Hammurabi, this black slab of cuneiform, becomes something you can actually read. 44% of Louvre visitors are under 26. The head of mediation basically said, we have to meet them where they are.
In Tours, they just finished something that started in 2022: 530 medieval manuscripts, some over 1,200 years old, digitized and put online. The oldest is from the year 800. One of them is the Jeu d'Adam, one of the first plays ever written in French. The library in Tours burned in 1940. Only part of it was saved. The director of the project, André Binggeli, said something I can't get out of my head: "Digitizing is allowing memory to be preserved for centuries to come. It's particularly relevant today with the wars happening, libraries being destroyed in the Middle East, in Africa. We had the case in Tours, the library was destroyed in 1940 and only a part of the collections was saved."
And then there's SOCFACE. Probably the biggest use of AI on an archive anywhere in the world: 25 million documents digitized, handwriting recognition turning them into a database of 400 million people, public on the FranceArchives portal. Census lists from 1836 to 1936 — twenty censuses, a century of French lives on paper. Not just preservation. They want to actually study it. How families moved, how jobs changed, how villages emptied out.
The Inria one is even crazier. CoMMA. They built a specialized AI that transcribed and translated 32,763 medieval manuscripts in four months. That would take a paleographer years for one manuscript. 3.33 billion tokens. And they made a deliberate, stubborn choice: train the model to correct nothing. Not abbreviations, not scribal mistakes, not flipped letters. Golden rule: "Correct nothing." Stick to the raw document. Don't extrapolate.
The art restoration piece is where it starts to hurt my brain.
A researcher at MIT, Alex Kachkine, came up with a method where AI analyzes a damaged painting, makes a digital restoration, prints it on an ultra-thin polymer film, and lays that film over the original. The film can be dissolved if conservators need to see the damaged original underneath. He tried it on a 15th-century oil painting that was wrecked. It restored 57,000 little losses in 3.5 hours — 66 times faster than traditional retouching.
It's reversible. It's non-destructive. It respects every ethics code in conservation. But it also means the thing you see is the AI version on top, and the real painting — the one the artist actually touched — is hidden underneath it. So when you walk away, which one is the "real" painting? The one under the film, or the one you just looked at?
I don't think the people doing this are naive. They know.
The French Ministry of Culture published its AI strategy in 2025, five axes, including "responsible uses of AI" and "mobilizing AI for heritage." It talks about evaluating whether AI should even be used, controlling environmental impact, sustainability, fighting bias, explainability. They're thinking about it.
But here's what I can't stop thinking about. There's this phrase going around: "Once digitized, it can be more durable than stone."
I get what they mean. Stone crumbles. Stone burns. Stone gets bombed. A file, in theory, lasts forever. But only if the servers keep running. Only if the format doesn't go obsolete. Only if the company that makes the software doesn't go bankrupt or decide to pivot to something more profitable.
The historian Valérie Schafer said it better than I ever could: "We are building archives that depend on private technologies and energy-intensive infrastructures. Preservation has become paradoxical — it demands constant renewal."
The Ministry's own numbers say it. Digital is about 4% of France's carbon footprint. Data centers using more and more energy and water. "More durable than stone" only works if you ignore that digital preservation needs electricity every second of every day, and money every year, and political will that never wavers. Stone doesn't need any of that. A manuscript in a cool, dry room can sit there for 500 years and survive. A server farm needs to be cooled, powered, patched, guarded.
So where does that leave us?
I think digitization is not preservation. It's a backup. It's insurance. It's making sure that if the original is destroyed, we still have something. And that something matters. The student in Dakar can browse the Louvre's Egyptian wing. The researcher in Tokyo can walk through a Gothic chapel in 3D. A kid in Tours can look at a manuscript from the year 800 on her phone. That's real democratization. That's real.
But democratization has a cost. The cost is that the thing being democratized — the original, physical, irreplaceable thing — gets pushed to the back. The copy becomes how most people experience it. The original becomes the backup.
And then one day the server farm goes dark, or the format dies, or the company decides it's not profitable, and we look up and realize we spent so much time scanning the stone that we forgot to keep it standing.
I don't have a solution. I don't think there is one clean one.
I just think we need to stop pretending digitization is the same as preservation, and a digital twin is the same as the thing it's twinned with. They're not the same. They do different things.
And if we're not careful, we'll end up with a world where everything is accessible and nothing is real.
The house is on fire. And we're making a really detailed 3D model of it instead of putting it out.
Notre-Dame digital twin, digitization, cultural heritage, AI art restoration, Louvre, augmented reality museum, Livio De Luca, CNRS, Andrew Tallon, laser scanning, Kore of Samos, Akhenaten, Code of Hammurabi, Tours manuscripts, Jeu d'Adam, SOCFACE, CoMMA, Inria, MIT restoration, Alex Kachkine, polymer film restoration, French Ministry of Culture, AI heritage, digital preservation, Valérie Schafer, archive digitization, medieval manuscripts, FranceArchives, cultural heritage AI, memory preservation, digital archive, 3D scanning, opinion
Notre-Dame digital twin
is digitization preservation
AI art restoration ethics
Louvre augmented reality
digital preservation problems
cultural heritage AI
why digitization isn't preservation
SOCFACE project
CoMMA medieval manuscripts
AI restoring paintings
Comments
Post a Comment
If you want any topic to be written message me.