CROSSLUCID collaborate with AI to visualise the mysterious life forms of the future

The duo behind CROSSLUCID discuss the collective’s latest project Landscapes – a series of 5000 portraits created in collaboration with generative adversarial networks (GANs) to imagine the shapeshifting life forms of tomorrow

Lopsided skulls. Mercurial skin. Bionic body parts. The ageless entities in Landscapes look, at once, primordial and fantastical, familiar and foreign, of nature and artifice. The ongoing portrait series is the latest project of the Berlin-based interdisciplinary artist collective CROSSLUCID, created in collaboration with Generative Adversarial Networks (GANs) and two “data alchemists”, Martino Sarolli and Emanuela Quaranta.

“The outcomes of GANs typically seem uniform because the input used to train Artificial Intelligence (AI) models are mostly portraits of people who are white, gender-conforming and have conventional Western-looking features,” says Sylwana Zybura. Zybura is a founding member of CROSSLUCID, along with Tomas C. Toth. “It was exciting for us that such an incredibly large spectrum of characters emerged from the neural networks for our new project.”

The series began in November 2020 after Slanted magazine commissioned CROSSLUCID to create 5000 unique covers for an issue delving into the impact of AI on design and our daily lives. The project further speculates on the many underlying structures that inform our identity and how they might emerge in the life forms that will populate our future world. 

Landscapes builds off of the collective’s last photobook, Landscapes Between Eternities, published by Distanz Verlag in 2018, which visualises otherworldly figures and forms flourishing in a future that exists beyond binaries. The book investigates the many ways we construct identities and perceive bodily expressions through portraits of humans melding with props and shots of elusive objects.

CROSSLUCID met Sarolli and Quaranta in 2019 while working on a project with the Istituto Italiano di Tecnologia in Genoa, Italy. “We had many fruitful conversations with them about the potentialities of AI, specifically GANs, and shared similar goals in terms of exploring this field,” recalls Toth. For the new portrait series, CROSSLUCID employed datasets originating from Landscapes Between Eternities. They processed these through artificial neural networks, which are the computing systems fundamental to deep learning algorithms, which, as the name suggests, are inspired by models of the human brain. GANs, specifically, are a type of AI model that uses two neural networks, which compete with each other, to generate output.

“Normally, you need a big dataset, like a minimum of 10,000 images, to feed GANs so they can look for patterns and learn from them,” explains Zybura. “Of course, we didn’t have that. We initially had an extremely small dataset of 100 published images from the photobook. So we thought, ‘What would happen if we used test shots from our movement and texture studies as input?’ That’s when we decided to add around 200 images to the dataset that didn’t make it into the book or were error shots. From an artistic perspective, it was interesting to reveal so much of our creative process and include these behind-the-scenes images, but, for the GANs to be trained properly, it was also necessary.”

During the training period, which lasted about five months, the two data scientists sent CROSSLUCID a batch of new outcomes every couple of weeks. The artists observed that the initial rounds of portraits were abstract, barely resembling human forms. Although Zybura and Toth had to adhere to guidelines from the magazine publishers regarding the cover images’ compositions, they decided early on that it was important to let the training process unfold organically instead of trying to control the visual outcomes by driving it in a specific direction. 

“Ultimately, we didn’t want the final selection of images to reflect our aesthetic preferences,” says Zybura. “We were more interested in showing the mystery and unknowability inherent to the process of collaborating with GANs.”

Uncanny, diverse and appearing as if in perpetual motion, the 5000 images that currently compose Landscapes reminds us that AI is far from neutral. Indeed, many functioning neural networks don’t reflect the newest research or acknowledge that learning also happens beyond the brain. “The outcome of these AI models is completely related to their input,” says Toth, “To some extent, they can teach us about ourselves, our biases and the existing structures that govern our lives. We keep coming back to the outcomes – almost daily – interacting with them, reflecting on them and learning from their various shades and textures.” 

CROSSLUCID veers away from much of the discourse that conceptualises AI as a threat to humanity. Instead, the collective sees these emerging technologies as a part of the human ecosystem that can help us envision new possibilities of living and being in an ever-evolving landscape. 

@crosslucid