Jake Elwes – Latent Space (2017). Special thanks to Anh Nguyen et al. at Evolving-AI for their research on GANs. In artificial intelligence, the term latent space; refers to the mathematical environment in which a neural network maps the patterns it has learned from visual data. After training on millions of images—trees, birds, faces, and beyond—the network begins to locate each category in distinct regions of this abstract space. By reverse-engineering these coordinates, the AI can generate entirely new, synthetic images that reflect what it has learned. Latent Space is a generative video work that invites us into this process. Using a neural network trained on 14.2 million photographs (from the ImageNet database), Jake Elwes visualises the AI’s journey not between fixed points, but through the in-between—the ambiguous zones where categories blur and meaning dissolves. The result is a hallucinatory, dreamlike sequence: an AI imagining the world from within, navigating the fuzzy boundaries of learned perception. At once beautiful and unsettling, Latent Space captures the uncanny emergence of machine imagination.

The token is the artist's certificate of authenticity and should not be traded separately from the physical work.