Future Paracosms and Their Infrastructures

When I was a child, my mother told me to “think of something pleasant, think of a forest” while trying to fall asleep. So I did, and over the years the forest I imagined expanded into an intricate ecosystem, complete with human protagonists, fantastic species bonded by symbiotic relations, futuristic technologies, and the epic story of one character, a reflection of myself. Instead of being simply “pleasant,” from the beginning the fantasy was also dark, sometimes dystopian and eventually sexual. Later I learned that such delirious projections are called “paracosms” and are not uncommon. I learned that some of the most famous paracosmic dwellers include the Brontë siblings and James Cameron. Eventually I even learned that some psychiatrists decided that such adult world-building can be considered a mental disorder, referred to as “maladaptive daydreaming.”

I never grew out of my fantasy, and I’m still invested in world-building on a daily basis, especially when feeling overwhelmed by reality: I usually revisit different “scenes” or events, make adjustments, or follow the dialogues as if watching a film. If I was given a choice, I would fully move my life into this fantasy universe. In spite of the occasional cruelty of my inner world, it has great advantages: it is absolutely enchanting, tailored for my preferences, and most of all private. I never share its details with anyone and I’ve never used any of its elements in my films—until recently. In my short film The Book of Flowers, a somewhat unimportant detail of my private earth’s biology—the fact that flower cups are gigantic and humans are the main pollinators of plants—became the premise of a whole new story.

The Book of Flowers is an animated film combining a classical narrative format and nineteenth-century organ music with visuals generated using Stable Diffusion, a latent neural network that consists of a text-encoding model and a diffusion image generator. Freely released to the public last year and developing at a shocking pace ever since, Stable Diffusion uses text prompts to generate images. After deciding to surf the first wave of AI-powered video generation, Nathan Gray, Ewa Polska, and I started to experiment and adjust our method to the new tools, which are updated with new functions every week. Because the available models were trained on still images, they generate excellent photorealistic pictures of flowers, but are unable to convincingly reproduce the movement of ripening pearl-seeds and growing machine tentacles. In the end, the film was made using both text and image prompts—in this case sequences of 16 mm time-lapse animation. The footage served as a template for movement and depth, while the original was completely overwritten with imagery generated from text: frame after frame created as a separate picture.

I describe here the details of our creative process because it hints at the major shifts that may soon consume the film industry, the cultural-production sector, and its economies (my expertise is in moving-image media, but obviously this is not the only field that is undergoing a revolution). After an intense time developing a work with AI tools that are themselves still developing, it’s hard not to be filled with anxiety and awe. It feels as if humanity is standing on the edge of a well, ready to dive in head first—a well that could turn out to be a wormhole leading to another world, or a dark pit filled with tar. One thing is for sure: this well seems to deepen every day, forcing the diver to redefine themselves and change their biology in order to survive the swim. In the near future, amateurs and professionals alike will be offered infinite possibilities in film creation. But is the human brain and human society creative enough to work with AI?

At the moment, it seems that in no time one will be able to generate Hollywood-quality visuals without having access to expensive production equipment. One new method of making a movie without any knowledge of filmmaking is to use existing films as templates. The significance of this phenomenon goes beyond the disturbing trend of deepfakes and the global “suspension of belief” produced by the forgery that machine-learning enables.1 Think of someone deciding to watch Workers Leaving the Factory but as an inter-franchise anime, mashing up Sailor Moon, Evangelion, and Berserk; In the Mood for Love starring their own neighbor and with a different ending; Tár where the protagonist is not an abusive classical-music conductor but the director of a contemporary art institution in their hometown; a fertility clinic ad presented by a pleading child who resembles the viewer and their secret lover; even a social media feed but shot with an 8 mm camera and with everyone appearing in reels changed to fit a certain age group or race, according to demographic factors or the preference of the user. The technology to create moving-image entertainment completely tailored to each viewer is almost here.

The next step is, of course, that there is no central narrative. Instead, professional creators supply the audience with characters, settings, and stories that could be freely shaped into anything they desire, according to their most private dreams. Imagine that fan fiction becomes the largest and most popular field of culture.

“Each new medium both expands and shatters the human ego, showing us more of the universe and then promptly reducing it all to us,” writes Bogna Konior in her The Dark Forest Theory of the Internet.2 While Web 2.0, as analyzed by Konior, “creates the illusory integrity of the world and the self,” AI has the potential to create the illusion that each self is integrated into a different world. It provides everyone with sufficient means to build their own paracosm, engendering a society of daydreamers whose experience is atomized and singular, the notion of collectivity acquiring a completely new meaning. A cultural experience, instead of being shared, becomes a process in which elements can be either individual or collective; the circulation of “affective capital” becomes fragmented and seemingly decentralized, despite the centripetal force of the market pulling everything into its orbit.3 The most reliable communicator is still natural language—but only because it is the language used in coding, in the form of text prompts.

This process of cultural content being atomized into building blocks is not unique to the development of machine learning technologies: it has long lurked in the film, game, and music industries. Recently I have been working on a feature film script, a science-fiction period drama, supervised by an established production company. When I shared my doubts about the economic sense of my decision to set the film in Australia with the supervisor, I received this reply: “Finish the script, and if the emotional relationships between the characters are established well enough, you can always move the story somewhere else.” The specificities of culture, era, ethnicity, and geography, while prized and sought-after by the film industry, are transferable, and this transferability of specificity is not unique to the film industry.

***

In my excitement I didn’t wait for the rocket to cool after its descent through the planet’s atmosphere, but jumped out at once and shouted:

—Excuse me, is this by any chance the Highest Possible Level of Development?!

No answer. In fact, they paid no attention to me at all. Somewhat taken aback by this show of utter indifference, I looked around. The plain shimmered beneath the square sun. Here and there, things stuck out of the sand, things like broken wheels, sticks, bits of paper and other rubbish, and the inhabitants lay any which way among them, one on his back, another on his stomach, and farther on was one with his legs up in the air.4

In one of his short stories, which may come across as far too jocular for today’s reader, Stanisław Lem presents a vision of a civilization that has achieved the “HPLoD,” a state of homeostasis. Its representatives spend their whole lives entranced, daydreaming—“scratching, yawning, gargling.” If the magic of artificial intelligence is key to HPLoD, it may be that humanity is not ready to use it. In fact, it has never been more difficult to imagine humanity achieving a state of homeostasis. The multiverse of paracosms requires a supportive infrastructure, and, in the era of ever-increasing wealth inequality, impoverished classes may easily become the mechanical turk automating someone else’s dream, or, even worse, they might become entranced in a knockoff paracosm—an uncreative, psychophobic nightmare.

While the main issues being discussed in the context of AI-based cultural content—the disruption of labor markets and the originality of artworks—require attention, I would like to look at them from a different angle.

The problem of labor disruption is complex and will require many adaptations, though it probably won’t leave anyone unable to pursue their creative calling. Content creation will still need a huge amount of professional and skilled input. But I have no doubt that AI will be used as a pretext for jobs to be offshored, made more precarious, and generally devalued. The large-scale layoffs that started in 2022 in tech and media super-conglomerates came early and had nothing to do with developments in technology, but rather with the inherent nature of rapacious neoliberalism. Yet it is a labor scare that (again) dominates the coverage of AI in the media, both conservative and progressive. “We need to strike on computers,” sang Johnny “Guitar” Watson in 1984, as if smashing the box of an unlucky Commodore with its 64 KB of RAM could protect workers more effectively than guillotining the rich. The process of turning everyone into a magician could easily go astray in the current ecosystem of power distribution.

It’s also hopeless to expect current cultural content-making systems, dominated by mega-scale resellers, to be ready to radically redefine copyright and create new models of cultural content circulation. Current discussions about copyright orbit around the question of messing with the “original work” and appropriating artists’ work for training data sets. While it is obviously not right to profit from somebody else’s work, I personally feel like all works are in some sense derivative and there is no work whose originality can be, or should have to be, proven in the first place. In some sense, I even see the movement that is trying to block “original” works from being used in machine learning as increasing the quantification and tokenization of cultural content—something that goes against, I suppose, the common desire of those who are protesting. In today’s economy of large-scale file sharing, copyright is only enforceable by those with the resources to police it at scale, i.e., entertainment monoliths.

In the process of allocating venture capital to various forms of cultural content creation, the artist has become a separate currency to their work, and each of the work’s components has the prospect of becoming a discrete currency. Alongside this process, new technologies are gaining the ability to personalize content, so that the cultural landscape could change into an industry of world-building, with each experience constructed from atomized components. What worries me more than the potential change in the status of creators is a situation where the response of copyright lawmakers leads to this new cultural industry being locked in a dreadful, uncreative stagnation, where production and distribution corporations are free to patent elements like personality traits, distinctive gestures, or certain emotional states.

The content generated by AI that is based on existing material is not a “filter” but always a new creation, a new set of pixels. I can imagine an absurd situation where AI tools are used to determine if a legal percentage of copyrighted material has been merged into someone’s filmic dream. As a reference, in 1956 a new law was introduced by the Central Committee in Poland stating that paintings could have a maximum of 15 percent abstraction. It’s hard to conceive of measuring such a thing without the use of learning machines.

Whatever gets patented in culture becomes a frozen asset of an otherwise freely developing story; “story” here is understood as the oldest and most basic human technology of information storage.5 What will happen to the story once it is deconstructed into building blocks, once its experience is personalized, once its components are patented? It is this clash of technologies and their economic landscapes—story vs. artificial intelligence—that might produce the most important shift for humanity.

***

—Pardon me, but if I’m not mistaken, you gentlemen have been fortunate enough to achieve the Highest Possible—

The words died on my lips. He didn’t seem to hear me at all, for he was wholly taken up with what lay on his knees, which happened to be his very own face, removed somehow from the rest of the head and sighing softly as he picked its nose.6

This text was inspired by conversations with Nathan Gray, Giacomo Mercuriali, and Sam Samiee, who may not share the same reflections as myself..