Ever since the release of ChatGPT in late 2022, there has been much handwringing about the role of AI in fiction. Writing a coherent and enjoyable story is an incredibly difficult thing. A long process that's often very frustrating. But not for ChatGPT. The large language model (LLM) can crank out paragraphs of reasonably well-written fiction with the most barebone prompt. As a writer, it's easy to feel disheartened.
Many famous writers have come out against AI over the last few months. In his newsletter The Red Hand Files, after a reader had asked ChatGPT for lyrics in the style of Nick Cave, the legendary singer-songwriter himself called AI replication as travesty. Cave wrote that ChatGPT can never create a genuine song because it cannot suffer. It has no inner being and so it can only ever imitate.
And perhaps this is true. Perhaps ChatGPT or any other LLM will never be able to write a Nick Cave song as well as Nick Cave does. But that's beside the point. We're talking about one of the world's best modern lyricists here. Brilliant, intense, irreverent. What he creates isn't what most people create. The question isn't whether ChatGPT can write something that's better than what the very best write, it’s whether ChatGPT can write something that's good enough you wouldn’t know an AI wrote it.
Turns out it absolutely can. In early May 2023, a writer named Aiden Marchine published Death of an Author, a murder mystery novel that was almost exclusively written by AI tools. Even The New York Times admitted that it was halfway readable. It's not hard to imagine a nearby future where AI can write novels that are perhaps not as good as the novels of our very best writers, but that are otherwise just as good as the average human-written book.
Okay, this is all still pretty disheartening.
But is it really? It all depends on how you look at it. If you believe AI will be the end of human fiction, then that's what it will look like. If you believe AI will make it impossible for you to make a living writing books, then that's how it will be. But there's another perspective: AI doesn't spell doom for human writers. It just means change. As with any new groundbreaking technology, the winners will be those who embrace the new. And that's what I tried to do with Broken Dreamscapes.
After my initial skepticism and mild disgust at how easily ChatGPT cranks out answers, I decided to use it for my first piece of fiction on Wormhole Stories. To be clear, every scene of Broken Dreamscapes I wrote myself. Painstakingly. Sentence after sentence for hours. Even for someone who wants to embrace AI as a writing tool, it felt dirty to have it write the actual fiction. I'd be tricking you, the reader, and that's the last thing I wanted to do. Besides, I enjoy playing with language to write something that's unique, quirky, me. So I’ve written everything myself.
However, ChatGPT has been tremendously helpful for everything around the actual writing. It's at its best when I give it a long prompt with a particular idea, a tiny seed, and ask it to give me some suggestions that build on that idea. For example, I knew I wanted Broken Dreamscapes to be about some kind of device with weird powers in a science-fiction world. I asked ChatGPT for a few interesting settings and eventually it gave me the Dream Nexus.
But here's what's important: ChatGPT gave me many other suggestions too. Most were garbage. Even its suggestion about the Dream Nexus wasn't very good at first, but it was good enough that it got me thinking about it. I asked it to elaborate, came up with my own ideas, added those to my prompts, and slowly we began building something that I thought was good enough to create a story about.
Here’s another example. During the final scene of Broken Dreamscapes, I wanted the offdream community to have a mural on a factory wall. Instead of breaking my head over what it could be, I asked ChatGPT to give me some suggestions. Because everything about Broken Dreamscapes is in one long chat, ChatGPT had all the context it needed to come up with a few suggestions, one of which was the large tree with mechanical roots, which I liked enough to put into the story.
It was a collaborative process. If I hadn't come up with the idea of the mural on the factory wall, there wouldn't have been one. And it wasn't always that easy either. I'd spend hours going back and forth with ChatGPT because I was dissatisfied with its answers. But the process of trying to clarify what I wanted to the AI actually helped me understand the story better too.
It makes me think of this: During his interview with Anderson Cooper, the legendary music producer Rick Rubin went viral when he said he couldn't play a musical instrument, couldn't work a soundboard, and doesn't know anything about music. Cooper was baffled. When he asked what Rubin got paid for, he said it was the confidence in his taste and his ability to express how he feels.
Although not as extreme, there's a similarity with how I felt using ChatGPT for Broken Dreamscapes. It's no longer purely about creation, about coming up with ideas from absolute scratch. It's more about curation, about choosing the ideas that you like and that make sense within the story.
Again, I want to emphasize that I did all the writing myself. That still felt like the right thing to do, and it often took the story in unexpected directions without the help of ChatGPT. But whenever I was trying to figure out how the story could continue or whenever I got stuck and needed a particular idea, I went back and forth with ChatGPT until we landed on something that I was excited to write about.
Is this blasphemy? Some writers will probably believe so. And perhaps this will never produce truly original fiction, as Ted Chiang argues in his New Yorker essay. But I believe it can. Where ChatGPT is the tinder, the human mind is the spark.
Very thought-provoking :)