AI wants to suck the life out of fan fiction
Maybe don't let it.
I understand, barely, the impulse to try and “complete” an unfinished novel by a famous author.
Over the years, a handful of dedicated Janeites have filled out the fragments of Jane Austen’s Sanditon, a draft manuscript of about a dozen chapters that Austen was unable to finish before she died. I’d classify this kind of “completion” as a form of fan fiction; these extensions and revisions and reworkings of Sanditon only exist because some people care deeply about Austen’s work. As someone who also cares deeply about Austen’s work, that makes sense to me, even if I’ve never been tempted to try writing my own version.
What I do not understand is the impulse, among evangelists of generative AI, to do the same thing with large language models — or rather, their impulse to convince others that this is desirable.
Churning out a “finished” version of Sanditon — or F. Scott Fitzgerald’s The Last Tycoon or whatever — isn’t a priority for these people. They seem more interested in vague handwaving about empowering humanity or talking about the apocalyptic risk of the machines they are building. Yet the idea that AI can serve as a ghostwriter for dead authors occasionally pops up as a justification for why OpenAI’s language models or services like Sudowrite (a tool marketed to creative writers) should exist.
For example, look at this commentary about GPT-3, published a few years ago in the journal Minds and Machines:
GPT-3-like tools will make it possible to reconstruct missing parts of texts or complete them, not unlike what happens with missing parts of archaeological artefacts. One could use a GPT-3 tool to write and complete Jane Austen’s Sanditon, not unlike what happened with an AI system that finished the last two movements of Schubert’s Symphony No. 8, which Schubert started in 1822 but never completed (only the first two movements are available and fragments of the last two).
What’s interesting about the analogy between unfinished art and broken artifacts is how nonsensical it is. If Jane Austen had completed Sanditon but only fragments of the finished work survived, maybe the comparison would have value. But there is no whole version of Sanditon drifting in the historical past, waiting to be excavated and touched up by AI tools.
The fact that this commentary is so untethered from reality reveals just how little the underlying literature matters to the writers of this article. An unfinished novel is not interesting to them because it might reveal something about the thinking or creative process of a beloved author; instead, it’s just a problem, a bug, an error in reality, and a large language model can solve it.
It’s not just AI researchers who promote this fantasy, though. Writers and self-proclaimed literature enthusiasts do it too. Novelist and nonfiction author Stephen Marche once ruminated in the pages of The New Yorker about the possibility of an “Austen machine” that could perfectly replicate her writing voice and be used to produce a full-length Sanditon. He acknowledges that fans have already written these kinds of continuations but complains that even the best ones feel too “contemporary.”
“The future would not have happened to the algorithm,” Marche comments, enigmatically.
Here again we have a fantasy of restoration. But in this case, the supposedly authentic Sanditon isn’t lost in the past. Rather, the completed novel is imagined to be somehow latent in Austen’s prior work. All we need is a language model trained on Pride and Prejudice and we can recover the ghost of Austen that still haunts her fiction. The author isn’t dead; she’s undead, and with generative AI, publishers can rip her spirit out of the text and put it back to work.
This is basically the opposite of writing your own version of Sanditon because you like Jane Austen. At best, fan fiction is a very personal form of creativity. Through writing, you can deepen your relationship with the stories and authors that have shaped your life. The value isn’t necessarily even in the finished product itself but in the process of further exploring a fictional world — or in the act of sharing your stories with a community of fellow enthusiasts.
AI writing tools already promise to take the writing out of creative writing. Should we be surprised that they also threaten to remove the fan from fan fiction?
Further reading:
Inside the AI Factory. New York Magazine
Readers Aren't Flocking to Chatbot Novels Just Yet. Counter Craft by