The Salt and the Women uses — and reflects on — AI tools to create a short, experimental film about women connecting with mushrooms and machines through a cybernetic ritual. The text, read aloud by Amy (UK) through a text-to-speech program, speaks of watching this generated imagery as if it took place on the sand in front of her. The women of the story, like their uncanny-valley dwelling counterparts in the images, are confused for papier mache and wax. Mushrooms, bodies and wires become visually and poetically entangled. The pulse of motion between images suggests the organic, like breathing or waves, with a mechanical stutter. Sudden shifts, visual noise and peaceful pauses come forward as new “rituals” emerge on the shore.

The film speaks to an emergent medium of generativity and the desire to break free of the constraints of machines, emphasizing tensions between the machines that tell the story and the narratives they produce. The film score is produced by harnessing electrical signals from mushrooms and running them through an analog synthesizer, a system designed for for the album Worlding.

Using Stable Diffusion to create images, then asking questions about these images to GPT3, yielded a feedback loop with myself as a moderating influence. The film is chaotic, strange, and often uncomfortable. It ought to be. It’s a dizzying examination of the space between images, text, meaning and the desire for bodies to escape that entanglement. AI image generation tools struggle to depict human bodies; this aspect lead GPT3 to create explanations for these distortions when prompted. The contorted bodies were explained as fusing with a network of organic and digital materials, and the images produced by the prompts escalated that logic - which was, in turn, escalated by the GPT3 output.

This is the nature of what can sometimes be called “model collapse” when it exists within a single system. Too much of the same information coursing through a generative model leads it to break apart, repeat itself, lose context. This is a film about the loss of that context, a way of placing the viewer within whatever shreds of a narrative or visual structure endure.

The tensions between organic and digital worlds emerge as a theme in the narrative just as it emerged from the way it was made.

A Kind of Metaverse

The scenes GPT3 describes bear some resemblance to the “latent space” of Stable Diffusion’s conjured images. Because the systems produced images and described them, it was inevitable that the fiction would reflect the process used to generate that story.

Each image in this film, according to the emergent narrative, is a failed attempt at a ritual by non-existent women hoping to transcend one world for another. This ritual involves the fusion of digital and analog information streams — mushrooms and computers — but also the bending of bodies into contorted, inhuman arrangements. Through these contortions of bodies into cyborganic postures, the body is meant to enter a symbiotic network.

Each failure breaks the scene into ribbons, like threads grasping for the next arrangement. Using AI interpolation to move from one series of images to the next, we see the new “glitch” or “noise” state of AI images when that information strays too far from its closest adjacent pattern. It looks like the scene is melting, or being devoured by mushrooms: a happy coincidence of how these systems piece images together, finding connections like the wandering hyphae of a mycelial network.

The noise state in between transitions by interpolation, in which the AI “collapses” one image into another, is an exploratory state, seeking commonalities wherever it can. The result, to me, resembles the wandering hyphae of a mycelial network.

We watch a pulsing series of images cycle through possibilities. Each scene is a failure at making the women of the dataset real. We know that the women of AI generated landscapes “don’t exist,” but the women in the training data do. The AI women are generated like ghosts from the real bodies of the dataset. These are the products of the datafication of bodies through surveillance and the automated gaze. And don’t we all hope to escape that?

Every woman the AI dataset produced tended to be either unrealistically beautiful — rendered from celebrities and fashion models— or completely illegible: smudged and smeared like thumbprints on a painted portrait. Bodies were a disturbing, disconnected assemblage of parts, or sexualized images learned from pornographic material, which is then censored. This film is about the tension between the latent space - a space of possibility — and the weight of constraint by the underlying dataset.

In the end the ritual fades away and we, the viewers on the real side the screen, join the AI narrator watching a mushroom. The glimmer of a human hand emerges from the nest of wires and sticks.

AI images are inherently about the information they learn from, as if ripped apart from the real world and operating in a closed space. I was excited to tell a story that reversed that relationship, however abstract it was. It’s a story about the impossibility of breaking in or out of those systems, given the way they are designed.

On one layer this is a story - perhaps better to say a poem — about women on a beach creating strange combinations of machines and natural systems, contorting themselves to adapt to the demands of each new machine, all in the hope of finding ways to enter more intimately into those systems. In another interpretation, it’s a story about data and what it represents, posing questions about that data moving into the real world and the distortions that occur along the way.

Or perhaps, it’s all a mountain of nonsense. Working with AI tools to create work in this way is all about a reactive interpretation. I am trying to understand and project meaning onto images and texts. Asking GPT3 questions infers the meanings I want to find, and it offers me answers in response. AI is as much a cultural and psychological system, encoded as a system of text and image weights.

Layer these systems together and you create opportunities for stochastic resonance, in which noise aligns in ways that amplify a signal. The direct output of an AI system is one of categorization and assignment. Pair these systems together, however, and a human artist can activate meanings that may otherwise not be present. This is simple semiotics, really: place two objects beside each other and you create a new story. What changes is the technology and its affordances and how we, as artists, decide to use them.

AI images and text are a kind of noise disguised as signal: they don’t say anything precise, leaving room for interpretation. Poetry is also about the distance words have from precise meanings, but what steers those meanings are authorial intent: maybe emotional resonance.

With text generation the words don’t convey that intent directly. A human using these tools uses intent to shape something out of those words. The words suggest intent, through selection and arrangement, but could still be illegible to a viewer.

The film is experimental, the words and images loosely connected, so I crafted a story about navigating that noise. They occasionally come together in ways that imply richness, or at least, an invitation for the viewer to project richness into the interactions of images, words and sound.

The navigation of these affordances is, in some poetically removed way, the same as one’s escape from the confines of datafication, categorization. These texts and images speak to the elusive, probably impossible dream of fusing fluid experience with the binary logic of machinery: “the infinite struggle between bodies and information, identity and presence, technology and ritual, understanding and unknowing.”

You can watch the film through the link below.

Watch on Vimeo