These works are created through the selection and placement of pixels from pre-existing images to form a digital collage. This process is applied to groups of images formed by an AI from the same prompt. Part of an ongoing series, my works are an exploration into the underlying structures of AI image generators and of the pixel as a carrier of information.

When an AI image generator is being trained, a mathematical construct known as Latent Space is created. In this multi-dimensional structure, each point in space represents an image and points that are closer together are images that are more alike. When prompted, the AI navigates to the area of latent space it strongly associates with that input - proceeding to decode the information at that point into a specific arrangement of pixels.

Whilst each pixel on its own is meaningless, in the correct context of others something recognizable emerges. This boundary between the purely abstract and the meaningful fascinates me, so the methodology used to create the work emphasises the granularity of digital images and the fundamentally abstract unit that composes them. With the functioning of AI image generators in mind, the amalgams of pixels I create represent sections of latent space - of the general arrangements that the AI associates with a distinct input.




The works also explore the parallels that can be drawn between the art of collage and the creation of latent space. Both use pre-existing materials in the creation of new forms and both bring questions of authorship to the fore. 

The specific way that pixels are placed in the collage retains the watermark present in the AI generated source material. In this, the process employed to create the work credits the AI in a way that the AI is unable to credit those responsible for its training data. In fact, even my own intervention at the end of a long line of authors (the AI’s creators, those responsible for the training data, the AI itself) is overwritten by the watermark.

The series’ focus on natural landscapes was initially intended to contrast with the synthetic and mechanical operations of the AI. Through the creation of the work it has become increasingly intriguing to consider how different, if at all, an AI’s understanding of nature is to our own. As we navigate the world are we simultaneously navigating our own latent spaces? Whilst the range of senses we have available to us is currently much greater than that available to an AI, the experience from those senses forms what could be described as training data.

From this perspective, given that the AI is trained by images made by a huge number of people, by attempting to represent latent space each collage implicitly represents how a collective has categorised an aspect of reality.




The landscapes are each emblematic of particular kinds of catastrophe: rampant deforestation exemplified by the Amazon, the pollution of rivers by the Ganges, the warming climate by the retreat of the Trift glacier. There is a peculiar awareness surrounding these problems. They are simultaneously known yet ignored. Given that Latent Space mirrors the biases of its training data, of the societies that feed AI’s information, is it then surprising that the images give no indication of such crises? Of a world on the brink of collapse?