Novel-Writing Chatbots Are Scary, But They Have One Lesson to Teach Human Writers Now by Andromeda Romano-Lax

One area where I believe it would be unwise to rely solely on AI-generated text is in the field of creative writing, particularly in the writing of novels. Writing a novel is a complex and nuanced process that requires not only the ability to produce coherent text but also a deep understanding of human emotions, psychology, and the human experience.

If that paragraph seems stilted and bland, it’s because it was written by ChatGPT after I fed it a prompt to write a skeptical blogpost about why we don’t want AI writing novels.

Still, it’s comprehensible.

Of course, the AI was just trying to please me by saying what I wanted to hear. While I clicked over to a fresh page to start this blogpost, it continued generating text in an open browser window, explaining how it didn’t have the same depth of insight as a human and that if I left it to ChatGPT to write novels, literature might suffer from homogenization.

(I could almost picture the AI mind behind this generation exercise, chuckling as it typed, “You just wait and see, dummy. One more year and you won’t be able to tell the difference between me and Stephen King.”)

Meanwhile, ChatGPT is already producing novels, including several hundred for sale on Amazon.

Does this give you the same feeling of dread it gives me?

I published a 2018 novel, PLUM RAINS, in which a robot named Hiro is able to show more empathy and attention to his elder client than a distracted human nurse.

The day is coming, friends. But let’s return to the subject of creative writing.

On Twitter the other day, I read this tweet by a suspense novelist:

TEXT: I asked ChatGPT to write a short story “in the style of author Andrea Bartz” and now I’m EXTREMELY CREEPED OUT. It’s not a perfect mimic, but on a sentence level, it sounds like another writer impersonating me.

Of course I had to open up ChatGPT for the first time and give it a try myself.

The result unnerved me, but I also recognized that ChatGPT has a tendency to default to hackneyed language and plot.

ChatGPT excels at cliches. We—the human writers of the world—shouldn’t.

For now—because very soon AI will be much smarter—we could play around with ChatGPT not as a way to generate story or voice, as some articles seem to be suggesting, but as a way to remind us how not to write.

Let’s see how ChatGPT responds when asked: “write a scene from a memoir set in Alaska.”

I remember the cold biting into my bones as I stepped out of the plane onto the snowy tarmac in Anchorage, Alaska. The wind howled around me, blowing my hair into my face and causing my eyes to water. I had never experienced anything like it before.

I had come to Alaska on a whim, seeking adventure and a change of scenery. I had always been drawn to the rugged beauty of the state, with its towering mountains, sprawling glaciers, and untamed wilderness. And now, here I was, standing in the midst of it all, feeling both exhilarated and terrified at the same time.

Groan. Recognize any of those phrases? I do. And I promise never to use the phrases “rugged beauty” or “untamed wilderness” again, or to open a script or novel with someone stepping out onto the tarmac. (Guilty!)

Given the prompt “write a thriller set in rural Alaska,” ChatGPT provided a novel-like opening, but it was dry, written in summary mode.

I told it: use that same prompt but include more scene, description and dialogue. Don’t use summary.

The next response was better. Frighteningly better. Following a few lines of summary, ChatGPT wrote:

She walked into the local diner, where the locals gathered for breakfast every morning. The small restaurant was dimly lit, and the scent of freshly brewed coffee filled the air. Kate took a seat at the counter and ordered a cup of coffee from the friendly waitress, who was happy to chat with her.

“Have you heard about the fisherman who was found dead?” the waitress asked, her voice hushed.

I tried to retune the prompt several times by telling the AI to avoid all cliches and omit common words. If anything, the story opening became more cliched (my emphasis, below).

Kate stepped gingerly through the snow-carpeted streets of Eagle, an isolated hamlet nestled deep in the heart of Alaska’s untamed wilderness. Her keen senses picked up the rich aroma of freshly brewed coffee wafting from a quaint eatery that beckoned invitingly to her.

Even when I specifically told the AI to avoid so many adjectives, it piled them on, almost as if it were in a panic.

For now, our jobs are safe, writer friends.

But maybe not for long, especially as chatbots continue to use mass data to imitate specific writers, becoming more intelligent in the realm of voice as versus content.

Even if you have no desire to play around with ChatGPT, consider this. You’re working on a scene in your story, novel, or memoir. What’s the most obvious way this would open? Don’t do it. What is the most obvious gesture or metaphor you should employ next? Avoid it! Where should your plot go next? If readers—or ChatGPT—can guess, then keep thinking!

Are there particular genres that are more cliché-ridden than other genres? Maybe we should sacrifice those to the robots first, and specialize in that which can’t be so easily mimicked. The future of humanity depends on it!

 

Andromeda Romano-Lax’s sixth novel, forthcoming in 2024, is The Deepest Lake. Meanwhile, she is trying to write her next thriller as fast as possible, hoping to finish it before the robots take over. She is also a book coach who enjoys helping clients avoid cliches whenever possible.

5 thoughts on “Novel-Writing Chatbots Are Scary, But They Have One Lesson to Teach Human Writers Now by Andromeda Romano-Lax”

  1. Thanks for the reality check Andromeda. Truth meets fiction, AI style. You were onto something with Plum Rains. Sending this to my book group that read it.

  2. Hey there Andromeda–
    I enjoy reading your occasional essays here.

    As you can imagine, I’m keenly interested in the emergence of AI art and literature. I’ve been plinking away at a short story involving an AI author for years now, and the real world tech seems to be outpacing my plot ideas.

    There is one CHAT-GPT experiment I’m considering conducting. Whenever a story idea strikes me, I jot it down in my writer’s notebook. Usually I manage to lay down the story beginnings and sometimes the middles, but not often the endings. Over the years I’ve collected dozens of these, and while some of them are eventually fleshed out into completed stories, most still languish in their green notebooks.

    So I’m contemplating feeding the beginnings and middles into the AI and ask it to finish the stories in my style. If the prompts given are original, will the endings also be original? I’m eager to find out. Depending on the results, I may quadruple my literary output (in collaboration with a machine), or I may decide to quit writing altogether and get into an as-yet AI-proof medium, like ice carving.

    My email is the same, in case you’d like to drop a note.

    cheers,
    David

  3. thanks Andromeda, for all your insights/assists in this space – appreciate this one tremendous
    sharp angle on the obtuse future(now).
    All the more juice to make it new.

    your project idea is interesting David – machine finisher. side by side Examples cd offer something, as cd ice carving , ur stories wd no longer be able to check “i’m not a robot” as i must now

    1. Andromeda C Romano-Lax

      Thanks for commenting, Sean! That “I’m not a robot” checkbox keeps gaining in meaning as we head toward that day when the robots will be doing even MORE for us.

  4. Andromeda C Romano-Lax

    Thank you so much for sharing this experiment with us, David! I’m so curious how it will turn out. And I’m so glad to be back in touch with you! Please tell us all more when the time comes.

Comments are closed.

Scroll to Top