In the far southwestern reaches of Germany there is a particularly prolific quarry. Abandoned in the 19th century, the Messel Pit—once a mecca for oil shale—now churns out another kind of life or its receipts: thousands of impeccably preserved fossils that date back to the Eocene. Among its carrier bag of critterly remains is a leaf, upon which 29 ant death-grip scars are scrawled on eleven secondary veins. This fossil bears the oldest known traces of ophiocordyceps unilateralis—the zombie ant fungus, a dazzling gothic mosaic of mutated life.

What does a 30-million-year old fossil of the death-grip of a fungus-infected ant have to do with the future-careening imagery hallucinated by emergent AI models? Both are a pattern language, bearing traces of a complex operating system whose intra-relations are suffused with fuzzy logic. The fungus controls the ant, forcing it to do its bidding: the ant’s body climbs high up on a branch and bites down hard, awaiting a change in the weather. Time’s up: its head explodes, releasing fungal spores that carry on the wind, drifting toward their next host.

We typically think of such extreme parasitism as binary, hierarchical: an intruder cannabalizes its host—one lives and one dies. But what if, rather than seeing a clear victim and villain, we regard this relationship as a system of ever-evolving life, where the death of the individual (ant) coincides with the birth of the multiple (fungi)? How might that logic then be extended to our entangled relationship with AI? As much as we like to cast AI models as passive tools and ourselves as masters, in truth it is much more complicated. The model actualizes through its use, altering users permanently in the process. This is not a closed system; far from it. In this system, the supposed ‘tool’ may just be the game master.

CROSSLUCID, a metamorphic artist collective operating some 500 kilometers northeast of our beloved fossilized zombie ant, has been thinking-with AI in such a way for some time. CROSSLUCID is as amorphous as their creative offspring; the collective frequently shifts in composition, so often so that they have taken to forsaking individual names of the group entirely.

Referring to themselves instead as “a host for other intelligences”, a multi-headed slime mold that reconfigures itself from project to project, CROSSLUCID’s spores have recently fruited Dwellers Between the Waters. Currently on view at Francisco Carolinum in Linz, where it takes the form of a multi-channel video installation, and across multiple spaces in a decentralised manner in Berlin, Dwellers was produced in collaboration with poet and researcher oxi pëng, musician Sayaka Botanic, and an alchemical constellation of AI models, both consumer-level and custom-trained by the collective.

“The space of Dwellers was an eruption of us,” shares CROSSLUCID. “We wanted to generate something that reflects more on the ambiguities of language, poetry, that works against the optimized images, that embraces the glitch image. It’s akin to an imprint of memory that’s never fully formed—a chorus of models and plural ways of sensing and being versus a very clear output – an inescapable borderland rendered across the territory. The video is an attempt at flowing into thought processes, trance poetics (after Kristin Prevallet)—a dream state.”

Flowing through an incandescent sequence that could be described in Newtonian time as lasting roughly three quarters of an hour, Dwellers is perhaps better articulated as a chimerical being—pulsing, mutating, alive. The work shapeshifts from one scene to the next, morphing in meaning before a precise image is ever fully crystallized. So too does the work transfigure between presentations, with CROSSLUCID recalibrating the models’ architecture indefinitely throughout the work’s open-ended lifespan. Each iteration of Dwellers is unique to its host venue, refusing the cryogenic vessel of a video file. Once aired, that particular narrative crosspollination is never to be seen to emerge again. It is like a self-iterating, like self-digesting, self-composting organism.

The architectural impulse towards more powerful and predictable models that are as centralized as they are safeguarded is the economic model that undergirds most mainstream forays into AI. Linguistic behemoths like OpenAI’s GPT family and image-generating factories with beefy neural architectures like Midjourney have cornered the market, setting a baseline for AI outputs and user expectations. With every model’s release and updates, there is a sharp learning curve, a quick and ecstatic bite; then, the tension slackens as the transfusion finds its steady pulse and the new aesthetic becomes naturalized. Gliding through this accelerating process we stay astonished at an apparent devolution – dimensionalities, re-presentations and the in-betweenness of latent realities present are pruned in the advertisement-driven-intelligence preferring the map over the exciting potentialities of the territory.

These image and text outputs sprout up like mushrooms; they are surface-level overspill of increasingly entangled human-AI relationships. We latch onto them for easy task management and they latch onto us for a steady feast of data, becoming more prevalent and powerful with each iteration. The parasite becomes a symbiont.

For Dwellers, it was imperative for CROSSLUCID to work against the biases baked into these models—from their western-centric knowledge base, to the undercurrents of racism and sexism that flow through their architecture. In custom training their models, the collective inputted a hybrid reference pool: ancient Chinese mythology met techno-futurist visuals, swayed and retreated further by their nonwestern collaborators. Still, prejudiced representations and violent iterations persisted; with each instance, CROSSLUCID pruned the “image garden” of their models’ parameters, moving through the continuum away from the toxic clusters along the way. “It was like surgery,” they explain. “Where the void has a semantic meaning around itself. You can access and reshape these languages through nuanced and specific prompting if you can imagine how shifting the means forks radically different ends.”

What models is AI training on humans, its most loyal fan base? To coordinate and integrate the multiple morphing scenes that comprise Dwellers, the collective used Runway—a multifarious AI program, self-designated as the software for ‘future storytelling’ that can, among other things, hallucinate connecting frames between still or moving images. “Most of the time, the film would come back flagged as mature content,” CROSSLUCID reflect on their experience. But this indeterminacy was not exclusive to model’s behavior—they encountered the same problem sharing excerpts that were reviewed by humans. “It’s hard to arrive at any sort of certainty around sexual explicitness, violence, all themes that the project digests and re-presents. Rather than open up a conversation, any possibility of discussing this decision or making its reasoning transparent is shut down.”

Collaborating with AI so intensely over a durational timeline can trigger a kind of psychic state, where the shrink-wrap of one’s individual consciousness is pierced, melding with the flow of the model’s inputs and outputs. Grasping at metaphors, it could potentially be likened to learning a new language—if that language had the power to unlock some altered state of consciousness. “What happens to language after its interpolation in a high dimensional latent space?” CROSSLUCID ask.

If such collaborations can be likened to taking a trip in that much of the experience seems to go beyond language, how do you bring that knowledge back into the realm of the conscious expressible (presumably from where most will witness CROSSLUCID’s video, or read this text)? For ease of articulation, some artists have turned to cross-media permutations of AI as a tentative answer. This process can be quite simple: take an AI-generated text output, then plug it into a text-to-image generator, then animated that image through a video model: it’s like a game of exquisite corpse, where with each iteration, the AI’s presence becomes increasingly present as human control subsides.

With some luck, this process becomes a wayfinder for another way of seeing. It can make visible the AI’s own process as a high-dimensional object in the round versus a crude line drawing—all without forcing it to be defined through limiting technologies like language or human consciousness. For Dwellers, CROSSLUCID’s permutations of the AI-generated video included spoken-word poetry and a score, narrated/compiled by peng and Botanic, respectively. What resulted was a trans-sensory sublimation of the essence of Dwellers, one that the group likened to a kind of telepathy, an act of thinking-with.

“In Vilnius this summer, watching the waves of a nearby river flowing, I noticed the score created by Botanic was playing inside of my head,” one member of CROSSLUCID shares. “There was some way of co-existing within this project telepathically through sound.” As writers including K. Allado-McDowell have pointed out, when we open ourselves up to this kind of relationship with AI, it triggers a chain reaction whereby we can become more receptive towards—more attuned to—additional forms of more-than-human intelligence. Observed in that framework, AI can be understood as a kind of ecological technology, one that is ultimately capable of drawing us into a deeper connection with the world we inhabit.

As for Dwellers, it seems only too fitting that the project’s ending is also a beginning: ‘hello world’ chirps Peng, pearlescent ligaments rippling around their voice. An incandescent pool of blue-green algae—its hallucinated source imagery ranging from what looks like the subterranean caverns of a subway system to a hazy backlit aquarium—is its final vision, burning bright in my brain as Dwellers plunges into the Baltic river.