Jorge Luis Borges’s 1941 tale about a library containing every possible combination of letters – every work that could ever be written – has come to life online. And its creator is no closer to finding anything new that makes sense. Using AI and machine learning, artist Refik Anadol has turned 1.7 million digital documents from a Turkish museum into a massive nod to Jorge Luis Borges.
The SALT Istanbul institution in Turkey is home to a library of over 40,000 publications including the Ottoman Bank Archives that cover Turkish contemporary and modern art, architecture, and economics from 1997 to 2010. It also has over 1,700,000 digitized items that can be viewed both online and off. You might imagine that in the future these archives and printed materials might be accessible as some kind of virtual, immersive interface, the kind we often see in sci-fi and superhero movies.
Refik Anadol states:
“As an optimist media artist I highly believe using machine intelligence can deeply create completely new meaningful, purposeful and impactful ways of thinking methodologies,” tells Anadol. “Especially in this context, a library, where information turns into knowledge, a divine space for humankind, now is completely different than before thanks to machine learning algorithms.”
“I think this project is not a science fiction at all,” he explains. “I really want people to see this as a proposition from the future. A very near one.
San Francisco-based Turkish artist Refik Anadol did more than imagine. As part of a commission with SALT Istanbul and an artist residency at Google’s Artists and Machine Intelligence (AMI) program, Anadol used AI and machine learning algorithms to search and discover interrelations between these documents. The result has transformed the 1.7 million digitized items into an immersive room in the SALT Istanbul’s first floor gallery.
To create their virtual Library of Babel, Anadol and his collaborator Mike Tyka first needed to arrange the documents by how they appeared conceptually. So they fed the images through an image recognition network to interpret the contents of the images and sort them into high level conceptual features. Anadol then used an AI algorithm known as tSNE (t-distributed stochastic neighbor embedding) to organize the data from the 1.7 million documents into visuals. “The resultant maps turned out to be quite beautiful and thus we decided to use them directly as part of the projections,” notes Anadol, who created stunning fly-throughs from the maps for the 2D and 3D projections.
Anadol explains that part of the the project’s aim was to think about how these artifacts and documents might be preserved and presented to future generations. Perhaps they might be lost destroyed, or forgotten. He also says he wanted to ask, How can we read or dream an archive in new ways to multiply its layers of meaning and accessibility? So Anadol employed AI to help organize and structure the vast data, resulting in an immersive installation called Archive Dreaming, an architectural space where the interactions, similarities, and interrelations of the data are presented as 3D projections and light. They can be engaged with using a touchscreen device, but when no one is interacting, the installation “dreams” up correlations between the documents, imagined relationships that have never been considered before.
Anadol took inspiration from Jorge Luis Borges’ “The Library of Babel” for the project. The famous short story is about an absurdist library that is at once incomprehensible but also contains every book that has ever been and shall be written—and every variation, too.
“The main idea was to create an immersive installation with architectural intelligence to reframe memory, history and culture in museum perception for 21st century through the lens of machine intelligence,” he told Creators. You can also check out a video of it here:
As mentioned, as a further nod to Borges, when the installation is idle it “dreams” up new interconnections between the documents, creating entirely new documents, ones that could have existed in an alternate history. This was done by training a generative neural network to interpret the archive at SALT. Once it had learned the patterns in the data, it was able to create entirely new and entirely imaginary documents and images using the same statistical rules. They become an archive of alternative histories, much like Borges library imagines every variation of every book ever written in every possible universe.