FanfictionAfterAI
Fanfiction After AI
What does fan culture reveal about the risks of asking our computers to do our storytelling for us?
Nothing stands in the way of the consumer-become-creator
Nothing stands in the way of your hollowing out the peach's pit And turning it into an apple
We are no longer passive
We are the new creators
The franchises of the future are our franchises
”Elon Musk created an AI that is stealing your fanfiction” is the kind of sentence that makes you want to log off for a while. But it’s a phrase I’ve been seeing a lot of on Tumblr, where online fandoms are reckoning with the arrival of new tools powered by artificial intelligence.
In recent months, fans of franchises like X-Men and Twilight have directed their ire at SudoWrite, an AI writing tool they allege has been scraping the fora and websites where creators post their fanfictions as a part of its training. There is some veracity to this claim. According to the company, SudoWrite uses several variants of GPT-3, a transformer model by OpenAI, which counted Musk as a co-founder, donor, and board member when it launched as a non-profit in 2015. (Musk is no longer affiliated with the company). Given that GPT-3 was trained on a dataset that consists of basically all web pages since 2008, fans say their fanfictions were likely scraped without their consent.
This is the primary contention that fan communities have with companies operating large language models (LLMs): that they “stole” fans’ creations in making their technology. In December 2022, a Tumblr user and fanfic writer named Vanya responded by writing that they would archive-lock all their work to protect it from being scraped. The post has 76,990 notes and counting.
There is, of course, something paradoxical about this. While it is easy to see why writers and artists would be unhappy about companies using their intellectual property without their consent, fans have a radically different relationship to “originality” and “authorship” than most cultural producers. Fan content is, by definition, derivative. It exists as an homage to something else and follows exacting genres and conventions. So why are fans so upset by tools that are essentially designed to perform the same function, to spin up new stories based on old ones? What does this tell us about why fans make this art to begin with? More importantly, what can it teach us about what we risk losing in a world where we ask our computers to do our storytelling for us?
The relationship between fandoms and AI wasn’t always this way. In 2017, a self-described “machine entertainment company” called Botnik Studios tasked one of its bots to read the entirety of the Harry Potter series and produce a new tome. The result, Harry Potter and the Portrait of What Looked Like a Large Pile of Ash, went viral and became a cult favorite in the fandom space. Part of its success was due to the inhuman nature of the robotic prose and its awkward formulations. Chapter 13, titled “The Handsome One,” explains that “Ron's Ron shirt was just as bad as Ron himself.”
But the programs have gotten exponentially better in the last six years, and their outputs no longer sound like the mumblings of a barely literate robot. In a December 2022 episode of the fandom-centric podcast Fansplaining, a caller named Steve expresses concern over the possibility of a fanfic generator and how it might usurp the creative spirit that drives and animates fan communities. “I’m in a bit of a spiral right now, reading a lot about these seismic advances in AI content generation and the impact on creative fields,” he says. “Will there still be the same impulse to write if these tools become so sophisticated that it’ll work on even the most niche prompts?”
One way to summarize Steve’s anxiety is this: AI tools threaten to automate away the participation that is at the core of contemporary fandoms. In Playing Fans, a seminal text in fan studies, Paul Booth, a professor of media and cinema studies at DePaul University, describes fandom as both an identity and a practice. On the one hand, a fandom is a community of individuals organized around their love of media, whether it be a novel, a franchise, or a band. On the other, it describes individuals who engage in fannish activities like the production of fanfiction and fan art. But another way of putting it is that fan communities exemplify what media scholar Henry Jenkins has called “participatory culture,” an alternative to the concept of consumer culture denoting an audience that is driven not by passive consumption, but by active reappropriation.
In a 2009 report called Confronting the Challenges of Participatory Culture, Jenkins, along with co-authors Ravi Purushotma, Margaret Weigel, Katie Clinton, and Alice J. Robison, defines a participatory culture as one:
- With relatively low barriers to artistic expression and civic engagement
- With strong support for creating and sharing one's creations with others
- With some type of informal mentorship whereby what is known by the most experienced is passed along to novices
- Where members believe that their contributions matter
- Where members feel some degree of social connection with one another (at the least they care what other people think about what they have created).
Jenkins describes the key activity of fans as “poaching,” a term he borrows from the French Jesuit priest and philosopher Michel de Certeau. Poaching is the act whereby a viewer takes a piece of media and selects which parts of it they like and which parts they don’t, such as when a fan of The Office makes a compilation of their favorite Michael Scott moments. It differs from passive consumption in that the poacher is exercising a degree of agency over the content. Through poaching, Jenkins notes, fans become “producers” themselves, remixing slivers of an original text to create alternative storylines or imbue the material with new meaning.
Importantly, when fans poach, they displace the original creators as the central authority on a text. Often, this behavior is motivated by a desire to see one’s identity reflected in the franchises one loves. Harry Potter, Naruto, Twilight, and other franchises that are popular on sites like FanFiction.Net tend to lack broad representation, which explains why many fanfictions alter existing narratives by queering them or inserting a marginalized character into the story. Often, content of this kind serves as an implicit critique of mass consumer media and its formulaic representations of race, sex, gender, and other intersections, with fan creators effectively ramming diversity into the stale wasteland of monoculture.
But fanfiction isn’t just about creating an alternative version of an existing story. It’s also about building relationships with others who share the same passion. On fanfiction repositories like the open-source AO3, authors and their readers engage in constant, peer-to-peer conversation, posting a steady stream of comments and reviews under published stories. Often, a reader will even request a specific story, with their own select parameters, for an author they admire to compose. Ultimately, what emerges is a circular loop where fans come together to discuss the works they love, creators transform them, and fans respond to the transformation. Participatory culture takes the form of a virtuous cycle, grounded in a reciprocal relationship between author and fan.
And therein lies the trouble with AI: on the podcast, Steve expresses concern that advances in AI could lead fans to consume content produced by programs rather than by their peers. For example, a fan looking for a story centered around their ideal “ship” (where two characters from a franchise fall in love) could now just prompt SudoWrite or ChatGPT to write it, thereby removing the need for fanfic authors and interrupting this virtuous cycle. The generative dialog that powers the fan ecosystem breaks down, shifting the fan experience from something communal and participatory to an insular exchange between human and machine.
This perceived disruption of the social dynamics of fandom points to another point of contention between fan creators and artificial intelligence programs: consent. Fanfic authors like Vanya claim they never allowed companies to scrape the content they produce, even when that content is often a permissionless derivative of copyrighted IP. And while fan creators and AI models could both be said to engage in practices that are appropriative in nature, fans concerned with questions of consent often point to two important distinctions between the two.
One is the aforementioned participatory culture in which these fan creations are embedded. The other is that, unlike the authors and entertainment corporations that own the IP underlying these franchises, most fans do not attempt to derive a profit from their creations. They poach because they love these franchises and the meaning and community they have co-created around them.
In the decades since fan culture began migrating to the internet, that love has yielded an immense volume of content, with little in the way of an economic return. On FanFiction.net, fans have created and posted more than 14 million stories since 1998. That does not include the illustrations, memes, GIFs, and other content that fans have produced alongside them. However, because fanfiction tends to utilize copyrighted material as its source, almost none of the people who create it receive any compensation, despite driving an extraordinary amount of attention to the franchises themselves.
Or, as the anonymous author of an essay titled “Fan Fiction and the Bourgeoisification of Creativity” laments, what these fans are really doing is performing a staggering amount of unpaid labor—participants in a form of creative production in which “surplus value is extracted by the corporations that own the franchises around which the fan action occurs.” Fan labor, they write, “is unpaid public relations, content generation, and advertising work for the franchise.” This creates a tension that is central to the relationship between fans and their beloved content: Their love for these franchises, the very source of their empowerment, is also the thing that makes them vulnerable to exploitation.
And yet, this willingness to create in the absence of economic incentives is what makes the fan community so unique: it is a world driven by care and the desire to communicate. “They don’t do it for money. That’s not what it’s about,” says Lev Grossman, a novelist and occasional fanfiction writer. “They’re fans, but they’re not silent, couchbound consumers of media. The culture talks to them, and they talk back to the culture in its own language.”
It also explains the contention with large language models. Simply put, the idea that an unknown corporate interloper could appear out of nowhere and appropriate their creations for profit is antithetical to the ideology of the community in which those creations were produced. Why should capitalists and their scrapping reap the benefits of fans’ labor of love?
If fans write fanfic not to claim authorial first-hood, or to make money, but rather to be part of a community, then it’s unlikely that they’ll give up the practice simply because AI has arrived on the scene. “We live in a world where there are already 313 Dean/Castiel high school AU hurt/comfort fics — and yet people were still inspired to write/update two more this week,” writes fandom analyst Destination Toast. “People are not going to stop creating new fanworks just because the AIs are increasingly able to join in and create more.”
Still, we can begin to trace the outline of what the future of fandoms will look like, along with the role that new mechanisms for value capture and creation will play in them. Over in the blockchain space, projects like Azuki and Bored Apes allow owners to create and sell everything from hoodies and watches, to animated series, based on their NFT avatars, driving publicity for these collections in a way that benefits everyone who purchases them. Nouns holders vote on proposals to fund projects that proliferate the Nouns brand, while fans take advantage of the project’s permissive CC0 licensing model to make their own Nouns-related content. As web3 thinker Drew Coffman explores in a recent video, this paradigm is replicable for any new franchise that seeks to give owners a vested interest in growing the value of a project, while allowing anyone to participate in its cultural proselytization.
The problem with this pseudo-organic growth model, however, is that it tends to confer all the power to the owners. Because of the profit incentives at play, the artist David Rudnick has noted, the content becomes promotional and is clearly legible as “shilling,” the product of a system that is “capable only of a circular, sycophantic mode of generation.” And when non-owner fans participate, they are unlikely to receive any of the value associated with the labor they input. They’re essentially no different from the Harry Potter fan helping to promote their favorite franchise for free.
Likewise, there are areas where the affordances of AI could be said to converge with the goals of fandoms. Writing is difficult, and tools like SudoWrite could help individuals who face hurdles or impediments in writing move from reader to author—to make their voices heard. Machine learning models also open up new opportunities for collaboration between creators and fans. Recently, Grimes released a tool called Elf.Tech, which invites fans to input a song of their own creation and then outputs a version sung in the artist’s own voice. With Grimes’ permission, fans can distribute songs created using the tool, with the artist and co-creator sharing a 50/50 split of all royalties generated by the song. Like the earlier Holly+ project by Holly Herndon, it reveals that these tools can be used to produce new spaces of exchange between artist and fan, perpetuating the virtuous cycle and even allowing fans to share in the value they are helping to create.
Unfortunately, the future could also look much less idyllic. We are already seeing efforts from creators to demand more transparency from the owners of these models, to allow artists to opt out of scraping, and to put limits on how entertainment companies use and abuse AI tools to save money at creators’ expense. Without these steps (and, ultimately, regulation), it’s not hard to imagine a world where a record label could use AI voice models to create and release songs without consulting the living artists they were trained on, or where streaming TV shows are written by generative AI instead of rooms full of writers. Such a culture, where creators are replaced in whole or in part by generative systems, would all but completely rupture the fan-creator relationship. It could hardly be called a culture at all.
Fans remind us that to create an active community around art is to let them become creators in their own right, without bounds and restrictions. To build a participatory culture, as Jenkins and his co-authors defined it, is to encourage creation—first, by allowing it, and, second, by creating tools that facilitate it. But any tool that aspires to facilitate creativity will fail in its aim if its result is to replace either the artist or the fan creator with a machine, thereby severing the relationship between them. The tools that stand the test of time will be the ones that uplift voices on both sides of the artist-fan divide and empower them to be creative together.
What fandoms teach us is that affect and care—not capital—are central to culture. Without them, there is no culture, only content. Only through thriving fanbases will franchises prosper, and to enable more fans to participate is to enable culture to flourish. Participation is culture’s true virtuous loop.