What happens when machines interpret memory?

That question isn’t rhetorical. It stood at the heart of our recent collaboration between the Estonian National Museum and media artists Varvara & Mar we reported a few weeks ago. From the museum’s point of view, we weren’t launching another new exhibit in the space of the permanent exhibition at the ENM. We were staging a live experiment: a digital installation where a generative AI model interprets fragments from Estonia’s ethnographic archive with the visitor as a vital link in the chain. Quantum est in Libris by Varvara & Mar, the resulting artwork, doesn’t just display heritage. It reanimates it.

Beyond preservation

Most museum professionals see digitization as preservation: converting collections into bits and bytes for longevity and access. In the middle of these everyday tasks, they likely do not have much time to think about how the pathway from this daily work to the (digital) heritage futures looks like.

What are the opportunities ahead, and whether the strongly normative, but also hopeful dominant narrative saying that AI holds immense potential for revolutionizing museum experiences, is the road to be taken? Already at the beginning of digitisation of the heritage, there is a silent, but largely overlooked principle: digitality isn’t neutral. It reshapes what heritage is, not just how it’s stored.

When archives go digital, they enter new contexts. Their meaning becomes fluid, reinterpreted not only by human researchers but increasingly by algorithms. This is not just media change, it is a change of media logic. Our notions of memory, interpretation, and even authority are up for negotiation.

Quantum est in Libris makes this tension and set of unanswered questions visible to those, who want to see. At the exhibition, visitors read aloud handwritten snippets from early 20th-century field diaries, descriptions of everyday life, material surroundings, lived experiences, beliefs and more. The text they read aloud is passed into a sculptural screen interface, and transformed into generative visualizations. The AI facilitating this process has no prior knowledge of Estonian heritage and the visual aspects of ethnographic museum collections. Its “understanding” is algorithmic and pattern-based, derived by an invisible set of data. On the one hand, we could say that AI is culturally agnostic, ie not referencing to any specific culture. On the other, AI is not culturally neutral by design and trained on large datasets of likely English-speaking (American) cultures. Also, ethical frameworks, embedded into the system, are not universal.

Looking into that friction between the local, particular, experiences and expressed, and the generic, the handwritten and the algorithmic, is the main “task” we would like to give to the museum visitors, experiencing Quantum est in Libris from the museum side.

The Quantum archive as metaphor

Our museum team, collaborating with the artist, started to refer to the starting point of the installation: the ethnographic archives of the Estonian National Museum as a kind of “quantum archive.” Archival meaning in the digital realm is not fixed until observed. Until activated by human voice and realized by the machine logic, the archival fragments forming the dataset of the installation, exist in a state of potential.

The metaphor of quantum archives is provocative but useful for thinking about heritage as dynamic, unstable, shaped by context, interface, and interaction.

Thinking of any archive as a quantum archive challenges assumptions embedded in traditional museum work: that archives are stable, that interpretation should be expert-led, that preservation is a form of neutral safeguarding. Instead, Quantum est in Libris suggests that heritage is co-created by institutions, technologies, visitors, and the unforeseeable outcomes of their collaboration.

Cultural memory in an AI loop

Should we trust machines to interpret cultural heritage? What do we gain and risk with by letting algorithms “show” the past? Is it ultimately, disconnecting heritage from its communities, who are the original owners of the archived knowledge, and also one’s who are expected to guide the interpretation of it, according to the principles of new museology, but also dominant heritage frameworks, such as Faro convention, establishing the principle of heritage as human-centric.

On the one hand, AI could offer a chance to uncover new patterns, stories, or connections hidden in massive data sets: this is, what is praised in many of the papers exploring and advocating the application of AI in heritage management and interpretation. It can give dormant collections new life, surfacing unseen perspectives. But it also carries the risk of flattening cultural nuance, over-interpreting noise, or reinforcing biases built into training data at the cost of original heritage voices.

From the museological point of view, the power of Quantum est in Libris lies in its ambivalence. It doesn’t resolve these tensions; it performs them. It invites visitor reflection, not suggesting instruction. What does it mean when generative visuals, created by a machine unfamiliar with a culture, are the output of everyday, intimate, local voices? How do these images relate to memory or identity?

Toward participatory archives

Quantum est in libris was born from a desire to ask questions we don’t always stop to consider while digitizing: What is lost when we translate memory into data? What new forms of agency emerge when machines enter the conversation?

By disrupting the permanent exhibition with an active, participatory layer, Quantum est in Libris offers the visitors to become mediators of meaning. Their voice becomes the trigger to animate the archive. On the other hand, it ceases to listen to them, as captivating imagery in the screen is …algorithmic.

In both situations, the museum ceases to be a storage facility and becomes a memory machine. We speak of heritage as something to be preserved, protected, kept safe. But what if it’s also something to be disturbed, reshaped, and reimagined? The past is not just behind us. Is it now also something we continue to code, every day?