While AI and deep learning have, until recently, primarily been used to manipulate images and video, the technology is slowly being applied to other forms of media—especially novels. In the latest example of this trend, the novel Amor Cringe bills itself as deepfake autofiction—a work written entirely by an artificially intelligent system known as GPT-3 and writer K. Allado-McDowell.
Deepfake autofiction is an offshoot of deepfakes, a machine learning technique that uses artificial intelligence to insert Hollywood stars into pornographic videos. In contrast to traditional deep fakes, which superimpose a celebrity’s face onto another person’s body, we generate a fully computer-generated novel with no human involvement whatsoever.
We train our neural network on thousands of texts so it can learn and develop its own voice—just like your high school English teacher wanted you to do. Then, we feed it plot points from classic novels such as The Great Gatsby or Hamlet and let it loose.
The result is something like The Great Gatsby by F. Scott Fitzgerald but with added sex scenes featuring Tom Cruise as Jay Gatsby, Emma Watson as Daisy Buchanan, and Jennifer Lawrence as Jordan Baker (because why not?).
It’s not quite ready for prime time yet—the characters are still stiffer than a freshly pressed suit, but I think we’re getting there. Soon enough, all books will be written by machines!
We’re still in an age where we’re trying to figure out what deepfake autofiction is—what can it be and how is it different from traditional fiction? Because there’s certainly a lot that would seem familiar to someone who reads fiction.
But if you look at any kind of storytelling genre, whether it’s theater or film or TV or books, as soon as something new comes along, people have to try and define exactly what it is. Is it like anything else before? Is it not like anything else before?
And I think with deepfake autofiction, even though we know exactly what its DNA is—it’s been artificially generated by GPT-3—we don’t really know yet what its relationship to other kinds of literature is going to be.
So, I think that right now we’re in a stage where people are figuring out how they feel about it. Are they excited about it? Are they skeptical about it? What does their gut tell them when they read these stories and when they hear about them being generated by artificial intelligence versus written by humans? That sort of thing.
Does it matter that Amor Cringe was generated by an algorithm? Does it change how we read or interpret a work if we know someone else created it? Most literary scholars believe there’s no difference in analyzing or interpreting a novel produced by a computer, as opposed to one written with pen and paper.
The key question is whether you can detect any human influence on what appears on screen—and in Allado-McDowell’s case, she has full control over what goes into her stories. She says GPT-3 will never have its own ideas about where a story should go or which characters should appear; she decides those things herself based on her writing style and her own ideas about what makes for good literature.
I think people are going to see these works as something different than they would otherwise because they're aware of their origins, but I don't think that's necessarily bad, she says. It's actually kind of exciting.
Deepfake autofiction is just getting started. The literary form is brand new and full of potential, but there’s still plenty to discover in terms of genres, storylines, and content. While GPT-3 (the neural network behind Amor Cringe) has given us a good starting point for artificial intelligence that can write bestselling fiction, it’s up to creators like K. Allado-McDowell to take things further.
How will deepfake autofiction evolve? And what else can we expect from AI authors? We look forward to finding out.