We are constantly producing the past. We are factories for the past. Living past-making machines, what else? We eat time and produce the past. Even death doesn’t put a stop to this. A person might be gone, but his past remains.

―from Time Shelter (2023) by Georgi Gospodinov

I must confess that I am obsessed with the mechanics of how we think of/produce the past. Not in the nostalgic sense but more so in a clinical sense. I’ve always been fascinated with the question of how we remember what we remember, why we remember it and what we make of that remembrance. After all, what is the stuff memories are made of, if not Combrian madeleines? I’m obsessed with how the things that have come before shape us, individually and collectively. So, when I chanced upon a Barcelona-based studio that works on information technology research, Domestic Data Streamers' project Synthetic Memories—'an initiative that aims to recover lost or undocumented visual memories using AI image generation'—I was instantly intrigued.

As I have often argued, memories are the bedrock of stories (or how humans make meaning), they make us and they make how we are in the world. And here was an experiment that was working with the same hypothesis. As Pau Garcia, CEO & Founding Partner of the practice, succinctly put it in a conversation with STIR, “Memories are the architects of our identity and visual memories shape our sense of self, our sense of belonging to a specific place.” While there is a lot to be said about this idea of preserving a sense of self, and more broadly identity that is implied in the project—especially in a world that is gearing towards the erasure of populations that do not fit into the long narrative arc of their history—what's more fascinating is the use of AI, and the project’s claim to alleviate the effects of memory loss through their work.

Probing into the nature of our memories, the choice of AI for the project, the mechanics of generative art (or photography in this case), and how we may harness technology to create solidarities and proffer help to actual patients form the substance of the conversation I engaged in with Garcia. Born out of the need Garcia saw to humanise digital technology, the project began in 2022, where the team worked with migrant and refugee communities from Bolivia and Korea who had settled in Sao Paulo. The project worked with these people to create images of scenes they recalled that may not have been photographed or recreate scenes from the past they wanted to preserve. From there, the project has reached four countries to recover personal stories that are threatened with erasure due to several reasons.

It recently won a grant from the Fundación BITHabitat’s 2023 call for urban innovation under the theme "the proactive city". With this project in particular, the use of technology serving as an intergenerational bridge seems to be a way to improve co-existence in public space in Barcelona, by fostering camaraderie through a person’s life and their stories.

Matters of Memory

When asked about the project, Garcia was extremely deferential, talking about his apprehensions with the technology and wanting to add a critical perspective on its use. While many so-called AI aficionados have used AI to create what in their words is original or see it as a way to optimise otherwise ‘dreary’ processes of data collection and analysis, here the idea was to harness its power to be kinder, to show how it can create bridges between past and future generations. Dreams not of electric sheep, but of actual people and their stories. The core project, called Synthetic Memories—which Garcia states is supposed to highlight the artifice of the memories that are created—creates photographs using older generative models such as DALL-E 2 and Stable Diffusion to act as ‘memory vectors’ allowing people to connect to the images and recall their experience.

While I must admit that I read this with a sense of scepticism since I believe that there is more of an experiential quality to memorialising something that cannot fully be captured in an image, Garcia responds by talking about the responses he has received from participants, and how positive they tend to be. On further probing, he reiterates the fact that the images are supposed to be highly subjective.

Delving into this process of creating these dreamy images that mimic our memories, the conversation delved into the mechanics of the process, with my apprehensions on the use of the technology, the biases such digital mediums have garnered in which databases used, and hence if at all there is diversity in the responses? Admittedly, the question of data bias and diversity is perhaps the most vital when thinking about and working with artificial technology. But to detail the mechanics; the process for synthesising someone’s memory starts from an interview that's conducted with the participant through which prompts are generated for specific experiences and stories.

Simultaneously, images are generated from these testimonies that are either made part of a larger archival exhibit (with permission from the participant) or given to them as keepsakes. What I specifically think is interesting is how through this reiteration, the team created a prompt archive that details the minutiae of each story and is a fascinating map of the images that have already been created.

Regarding bias and control, as Garcia mentions, the images are trained on context-specific biases while the controls are placed to ensure that certain details are perfect. However, flipping through some of the images, it’s disconcerting how homogenous they seem without being able to refer to the textual prompts. While there are certain details here and there, many seem interchangeable. This is another apprehension I have always held about AI-generated art; its non-specificity. According to Garcia, in this case, it acts as a boon, with the grainy and blurry effects mimicking a dream state, allowing people to fill in the gaps and reconstruct the past in their heads exactly as they lived it.

Memories of Matter

As Garcia mentions, the most recent component of the project is the physical Citizens’ Office of Synthetic Memories at the Design Hub Barcelona (DHub). Open from May 17 to July 28, 2024, the ‘speculative project’ was conceived to allow anyone from the local community to access the project and its infrastructure. The office includes a ‘specialised team trained in using this new technology, as well as in the process of recreating these types of memories.’ From what Garcia talks about, several participants are older, with families coming with their matriarch/patriarchs to relive their childhoods and life stories. People want to remember happy things, Garcia underscores.

He continues, “The idea of the office was to open it to any citizen, which includes migrant communities. [For them] it was a cathartic way to explain all the things they didn't or couldn't collect or they had lost. In that sense, I have realised that even if people are in a very distressing situation, they remember things that are mostly fun or happy. This is something that I like because people talk about traumatic events and situations to total strangers in this office. There is this space of intimacy, where people talk about things that happened to them long ago but were very relevant. And that's why I think these images that we are generating have so much power because they are truly backbones of the identity of so many people.”

The spatial design of the Citizens’ Office that the studio was responsible for, invokes the feeling of being in a bureaucratic office, from the rows of file cabinets to the costumes to even the computer systems. It’s quite a tongue-in-cheek gesture, meant to evoke a nostalgic reaction and to signify that AI is not a big scary technology, but far more human than otherwise thought of.

The Politics of Memory Matters

While the community aspect of the project and their almost cheeky use of AI is novel, perhaps the most speculative part of the project involves the pioneering work they are doing with Alzheimer’s scientists. As Garcia explains, “This is not a cure for dementia or Alzheimer's or Parkinson's or any cognitive disease that is going against our cognitive abilities. What reminiscence therapy does, which is where our methodology could be integrated, is reduce the number of cases of depression.” He explains that one of the side effects of memory loss is a sense of anxiety because one starts to lose their sense of self. The therapy is meant to create a safe space for patients, and what memory-based reconstructions help with is acting as memory vectors. They allow people suffering from memory loss to feel a sense of calm by creating a spark of recognition. It allows people suffering from an almost disturbed loss of self to come back to themselves, a sensation akin to looking for your glasses all over the place and then finding them on top of your head. This could never have been thought of without the ingenuity and human compassion.

Whereas previously, debates have centred on the authenticity and authorship of technologies like Midjourney to create “original artworks”; by giving them the moniker of a person’s memories, we circumvent the need for these. What makes the conceptualisation more fun is that for just a moment, participants can relive their lives through the mystical magic of AI. That being said, while the project’s goal to ‘expand our society's capacity to connect, empathise, and build bridges between different generations and experiences’ is commendable, I come away with questions. The project certainly seems to engage local communities but the conversation has not completely dissuaded my apprehensions on the issue of data bias in the responses, or the effectiveness of visual stimuli—albeit the community archive does a noteworthy job of critiquing contemporary historiography in reconstructing our selves. Really, what do we gain from foregrounding the stories of these people, whom history seems to have forgotten?

Bilal remembers a protest he took part in during the Arab Spring. He was 18-years-old. The protest took place on a highway near the city where he lived. There were many people with orange flags, symbolising the civilians who had been hurt by the war. Above the highway where the protest was taking place, there was a bridge. On it, a group of drivers appeared and started throwing stones at the people protesting. Driven by adrenaline, Bilal sought refuge under one of the stopped cars. For him, that moment was important, as it was crucial to defend their rights and achieve peace in their homeland.

And I think, it is exactly this; the representation of a moment of loss, the ability to document voices that would otherwise not be able to find one that makes this work worth writing about; and the notion that AI could become a tool to synthesise our lives as we lived them. After all, at our most human, we are just stories.