In order to create new images, artificial intelligence must first be trained using what is known as a “data-set”. These data-sets (commonly built of archives of 10,000+ images) are built by human beings with human biases. If data-sets do not feature the diversity of humanity, neither do the resulting images or programs that go on to use them. This leads to higher levels of inaccuracy when this technology is used on people of colour and gender non-conforming individuals; who can then be denied access to social services, healthcare or fundamental human rights when such technology is used to make decisions.

CROSSLUCID is a duo of artists: Sylwana Zybura and Tomas C. Toth, aiming to build new and inclusive data-sets that envision a future beyond these human biases. Using their own dataset of images from their previous work, Landscapes Between Eternities, the duo (alongside the “data-alchemists” Martino Sarolli and Emanuela Quaranta) trained their own Generative Adversarial Network. This GAN is a program where two digital processes compete, one producing fake data, the other checking it for its likeness to “reality” until an outcome is reached. In this new series of images, new forms take shape, creating a place to imagine a future where the body and its expressions are free from strict codes of behaviour.

As one might predict, after 5 months of training, their GAN began producing new and uncanny identities - bodies from beyond our imagination. They stand in stark contrast to the output of the conventional GAN (trained by white, western and gender-conforming images). Rather than the usual apocalyptic discourse surrounding AI, CROSSLUCID shows us the shimmering possibilities of AI in understanding the self within technology's ever-evolving landscape.