Word Count: 974 | Reading Time: 4 min
An article summarising the state of AI in publishing and what the writing community can do to protect our spaces from encroachment.
Bookstagram, Booktok, and Threads (Breads?) are roiling in response to the recent cancellation of a 7-figure book deal over suspected AI use. A week prior, the debate was over this article in The Atlantic accusing a best-selling Indie-published novel of using AI. While many insisted the article offers little proof, others noted that the more widely publicised accusations seem to dog young Black authors in particular. All of the controversy rests on a bedrock of this very uncomfortable question: Can AI really write well enough to pass for human creativity?
It has been a difficult year for the writing community: the cancellation of Shy Girl, a seemingly AI-generated story winning at least one prestigious writing prize, this Apex Magazine announcement, and this article from the editors at Bona Books admitting that they almost published two AI-generated short stories.
Those of us with algorithms attuned to the topic have been practically drowning in news about AI showing up in creative spaces. News that has become exhausting to consume. It feels as though everything on the internet may be AI: every short story, every novel after 2023, every poem, every article, every cute animal video, every news story, every graphic. All of us have now experienced that sickening lurch when you are jolted out of your innocent enjoyment of something by a sudden suspicion, and then (because the alarms are blaring) to see more and more ‘proof’ as you consume, until, before you know it, you are strapped into the latest outrage cycle, unable to look away.
Who benefits from this? The social media apps that run on farming our rage? The tech-bros trying to sell us their unreliable AI detectors? The publishing executives trying to replace us? The people who do admit to using AI to write and cannot stop gleefully salivating over its supposed inevitability? Surely not the mostly POC writers who find themselves in the eye of almost every storm. Definitely not the small teams running many of the indie presses and magazines that put in so much unpaid labour and unrequited love into this work.
Perhaps the literary world needs this reckoning. The Commonwealth controversy especially has started important conversations around the state of prestigious prizes and journals, the quality and tone of the stories they prefer, especially from writers from the Global South. Established editors and writers have perhaps been too confident about their abilities to instantly detect ‘slop’, and thus have failed to engage with AI enough to be able to not only catch it, but to create robust, replicable, non-AI systems to help filter it out.
The many ‘use-cases’ of AI, hawked constantly on every platform, make things even murkier. Some writers admit to using it to brainstorm; others say that they use it for research. Detractors argue that the machines are unreliable at both, hallucinating frequently and pushing ideation down the already well-worn paths of their training data.
One thing is clear across the book community: the writers must be human. No one wants to read something no one could be bothered to write. It must then, naturally, be turtles all the way down. Since human writing can and should only be evaluated by human editors, slush-readers, judges, and publishers, we must also accept that humans make mistakes. If we want to resist the tech-bros encroaching into the management of our spaces, and the enforcement of our policies, then we must accept that some AI writing will get through. That does not let anyone off the hook, however. It is, after all, the editor’s job to recognise voice, to show discernment, to know when they’re reading a good writer, not just good writing. They should also actually be talking to each other; relying on collective experience and second opinions.
The team at Bona Books has made an excellent start. Their article demonstrates accountability, transparency, and actionable steps that other publishers can take to make sure they are doing their best to not let AI slip past. Many of the tells their article describes have been around since the earliest models. An overuse of specific phrases, grammar, and punctuation can be trained out or edited away, but LLMs do not live in the real world. They cannot create interiority, exhibit specificity, or build a consistent world. They cannot mimic a writer who lives the story before putting it on paper.
Unfortunately, one thing AI is actually good at is scale. It is overwhelming our digital spaces and our nervous systems as we try to play detective instead of focusing on our craft. Worse still, feeding the outrage cycle allows more tech CEOs, this time as purveyors of slop detectors, to market off of our despair. Panagram, an AI company trying to cement a position as the definitive AI text detector tool, has recently been, despite backlash, integrated into Substack. It has also raised nine million dollars in venture capital in July.
Ultimately, writers are not in competition with the machine because creativity is not within the machine’s remit. We are also not in competition with each other. Solidarity at this point in time means more than a refusal to cede control over our carefully developed skills and voices to CEOs who want to sell our own work back to us. It also means returning to our experiments, our failures, my submissions, our observations, our internal worlds, and our life outside the screen. It means returning to our writing communities and perhaps overhauling our publishing systems while supporting the human teams who run them. It means reading, savouring, and reflecting on each other’s writing, and, slightly rephrasing Joseph Fasano, to love the work so we can participate in all this, the miraculous task of living.