LINDA DOUNIA REBEIZ


Linda Dounia Rebeiz is an artist and designer who investigates the philosophical and environmental implications of technocapitalism. She is inspired by science fiction, speculative design, solarpunk, and degrowth. 

Her work mediates her memories as alternative truths and evidence of excluded ways of being and doing. It is formed through the dialogue (and tensions) between analogue and digital mediums. 

In 2023, Linda was recognized on the TIMEAI100 list of most influential people in AI for her work on speculative archiving – building AI models that help us remember what we have lost. 



ONCE UPON A GARDEN
Genesis

Speculative Archiving
(2022—2024)


‘Once Upon A Garden’ is a digitally composited garden presenting a dystopian projection of a likely outcome of global warming. It depicts a world where humans now have to live with simulated images of plants and flowers because they have all disappeared from earth. It asks whether the preservation of nature is compatible with the goals of industrialised and industrialising societies, and in general is a reflection on what gets left behind in our pursuit of progress.
It made sense to create this work in collaboration with Artificial Intelligence from a conceptual standpoint. First, the rise of generative AI has uncovered gaping holes in how and where data about humanity is collected, stored, and preserved. There seems to be a violent unevenness in how much information AI has about non-Western ideas, contexts, people, and environments. This is evident in how common stereotypes and hallucinations are when interrogating generative AI tools today, be it language or image based.

Second, the conversation about how humans can coexist with AI is fascinating to me. So much of our fear of AI is based on its nefarious uses (deep fakes, privacy violations, etc.) and on the fact that it is projected to automate a significant amount of jobs done by humans today. While much of this is still true and worth denouncing, we are faced with the reality that it is here to stay, and that currently, much of our interaction with AI (consequently our relationship with it) is being shaped by for-profit companies who have no incentive to be transparent about their methods and aims.

Through this work, I wanted to explore if an artist from the Global South can develop a relationship with AI, given this status quo, and how might that relationship look. This is a question I explore in my practice more broadly.

‘Once Upon a Garden’ is ultimately a GAN (generative adversarial network) trained on thousands of images of extinct and endangered flora from the Sahel region, where I was born and live. It was created with the intent to be an installation that mimics walking through an artificial garden, and all the sorrow and beauty that this implies. It was developed in six stages: (1) investigation, (2) speculation, (3) compilation, (4) training, (5) curation, and (6) compositing.




Investigation
The first stage, investigation, began with a question: which flowers did I grow up seeing as compared to my mother, grandmother, and great-grandmother? The Kew Royal Botanic Gardens estimates that the rate of plant extinction today is 500 times what is was before the Industrial Revolution, and that we’ve love nearly 600 plant species in the last 250 years. This means, crudely, that my great-grandmother saw more flowers than my grandmother did. That my grandmother saw more flowers than my mother did. That my mother saw more flowers than I ever will. That I will see more flowers than my daughter ever will. And so on and so forth, but not for very long at the exponentially increasing rate of plant extinction.

I scoured the internet and local archives for lists of extinct and critically endangered flora species and focused on a little over 100 species from Sahel region. Then I set off to look for any potential photographic records of these flowers that might exist. I found very little. This launched me into a research about where the images of the internet come from and I learned that a great majority of data on the internet is contributed by about 4% of its users, most of whom are from Global North countries. It then became evident that I wasn’t going to find these plants online. So I turned to botanical archives instead. Most of the plants I was interested in were only preserved in herbariums and were from colonial expeditions records. There, I found the plants I was looking for — colourless and shrivelled, but present nonetheless.

Speculation
The second stage, speculation, was concerned with plugging the visual gaps in my investigation. I took the botanists’ annotations from the herbarium pages and used them as prompts in DALL.E (the generative AI tool by Open AI) to generate the closest visual estimations of what the plants I was interested in might have looked like. Detailed annotations were more successful prompts but I used all the generated results to stay faithful to the goal of this stage: a closest estimation of something that was never captured is the best we will ever get. This stage was painful for me. The more I speculated with the help of DALL.E the more it dawned on me how complete the loss of these flowers was. Some of the flowers I was interested in were critically endangered, which meant, they could still be growing somewhere in the plains of the Sahel. Yet, there was not a single photographic evidence of their existence. It was as if they were already gone.


Compilation
The third stage, compilation, was about creating a database of images to start training the GAN. At this stage, I had thousands of images: some that I took myself with flowers that I could find during my travels in Senegal, some that I found in my research online, and some that I imagined from herbarium annotations using DALL.E. It was important to make sure the database had between 10 to 50 representative images for each of the flower species I was interested in. They also had to be resized to all have the same dimensions and aspect ratios for training.


Training
The fourth stage, training, was relatively simple: feeding the database to the GAN, choosing the number of training steps (with checks along the way to evaluate the progress and fidelity of learning), and waiting for training to be completed. I wasn’t expecting perfect results from the model. I knew that it was impossible to generate photorealistic flowers from the model because so much of its data was created by the imagination of another unsupervised AI model. I knew there would be distortions and lossiness, but I was ok with that. It was an apt metaphor for the ongoing disappearance of our natural world, and how thorough that disappearance was in places where we don’t care to remember what’s left and what’s been lost.



Curation
The fifth stage, curation, was about combing through 10,000 outputs from the GAN to find the images that would be part of the garden. At this stage, I used feeling as method of selection: Which flower called to me and why? How did the flower made me feel? What did the flower remind me of? I recorded my observations as I went along this journey for my own sake. It was like working through fresh grief in the worst possible way — by going through an old, decaying, and hazy photo album of the person you lost.



Compositing
The sixth stage, compositing, was where the gardens were formed (or planted). Individual flowers were animated and brought together into different combinations. At this stage, I wanted to, as much as possible, mimicking some aspects of a natural environment such as light and wind conditions. Both these considerations were taken into account for the animations. I also wanted to create a separate set of individual flowers, representative of the general population of flowers of the GAN, that were meant to be contemplated on their own. On those flowers, I encoded and overlaid some information such as the growth of surface area dedicated for growing cereal crops and rearing farm animals in the last 100 years. I wanted to contrast the spectral remains of what could have been flowers with what they might have been replaced with in the absence of sufficient digital memory.

Installation
The final stage, installation, is an evolving and context-specific process. The garden and its flowers have to be in active conversation with the spaces and people they travel to. Therefore the installation changes to adapt to the context of the space and the intent of the exhibition. This body of work has been projected, projection-mapped, displayed on a variety of screen types, and in some cases made physical. It’s important to keep this stage active as the body of work grows and to account for how a specific audience and a space might need to receive it.






This project helped me grieve the loss of flowers and plants I never knew existed and will never see in my lifetime. It also helped me answer the question as to whether an artist from the Global South can develop a relationship with AI, given this status quo, and what that relationship can look like. To me this relationship is not just possible, it is necessary.

It’s paramount that we, artists from marginalised and mis/un/represented backgrounds relentlessly layer our realities over those ascribed to us, with the aim to fill the gaping holes. First, it was history books. Today, it’s AI. For to “make injustice the only measure of our attention” is to accept internal erasure and surrender to non-existence.

Our realities will continue to unfold no matter what. We are going to exist anyway, comforted by the belief that there is room for our defiant narratives in the world. That there is power in telling the stories we want to tell, even if no one listens, even if we have to brute-force the tools at our disposal to tell them. That for us to exist in freedom, all realities — endless realities — can and should be true at the same time.
This body of work was commissioned as a special project for Art X Lagos’s 2022 edition and has since also been exhibited at Miami Art Week with Refraction. It is licensed for public display to Artpoint. 






This project has 3 chapters so far, explore them here:

HOME  NEXT
2024, DAKAR, SENEGAL