ChatGPT is all over the news these days, and as a creative writer, I was freaked out by it.
I got a sick, fearful feeling that ChatGPT will replace human creativity—my creativity. It’s hard enough to find an audience without a tsunami of crap, written instantaneously for free by a robot, crashing over the marketplace and drowning my human voice.
I worried about this monster stripping all meaning from the art of creative writing.
I wondered what the point of writing even is, now that an artificial intelligence can barf out a whole novel in minutes…
If you’ve felt this way too, let me say this: ChatGPT only rewrites what has been written. Your concern as a creative writer is what can be written—what you alone can write.
So come with me as I find out what ChatGPT is and what it’s good for, and decide for myself whether the art of creative writing is worth continuing to practice.[1]
What ChatGPT is
ChatGPT is the user interface for a large-language model called OpenAI. You type in a prompt and it answers by coherently regurgitating and recombining text it has already read. It does this in seconds.
Its superpower is that it has read the whole internet.[2] So ChatGPT has all human knowledge at its nonexistent little fingertips[3]. It can hold forth on virtually any topic, and do so in plain, readable language.
As such, it threatens to devour the craft of writing itself.
Deciding to deal with it
Whether Silicon Valley tech bros should develop large language models that will put professional writers out of business is moot. They can do it, they are doing it, and I can’t stop it.
So I made two personal decisions as a creative writer:
- No matter what, my writing is mine—my voice, my self-expression, my purpose—and ChatGPT can’t actually stop me from doing it. It’s important to me. So I’m gonna keep doing it.
- A writing career that pays my bills is not my aim, so I’m not competing with ChatGPT on speed or quantity.[4]
Now I felt l could find out what ChatGPT is good for.
What it’s good for
Brainstorming.
I thought, hey, why not ask it to write this very essay? I spent several minutes formulating a prompt:
Write a 500-word essay in the style of Anne Hawley, addressing the five main worries and fears novelists have about ChatGPT, and whether human-written fiction will be replaced.
Ten seconds later, out came 585 grammatically sound, properly spelled and punctuated words.
The result was decidedly not in the voice of Anne Hawley. At best, it reflected the writing skill of a smart high school student padding a term paper to meet the word count. At worst, it read like a listicle.
It defined three (not five) fears specific to novelists:
- The fear that it will replace human creativity and originality.
- The fear that it will flood the market with low-quality content, making it more difficult for human authors to get published and recognized.
- The fear that it will make writing irrelevant as a profession, putting human authors out of work.
It mentioned a concern that any thinking person, writer or not, should have:
- The fear that ChatGPT will perpetuate harmful stereotypes and biases, reflecting the biases that are present in the datasets it is trained on.
And then, oddly, it mentioned a fear that’s probably bothering the Silicon Valley tech bros and venture capitalists, but is actually creative writers’ great shining hope:
- The fear that ChatGPT will simply not be able to capture the essence of what makes literature great, the emotional depth and complexity that can only come from human experience.
It wasn’t a terrible essay, but I wouldn’t put my name on it. However, it did jumpstart my thinking, effectively brainstormed the topic, and generated prompts. ChatGPT got less scary and more interesting, so I moved on to…
Writing-adjacent tasks I hate
I absolutely hate having to distill my work into sales copy. So I asked ChatGPT to write me a logline. After dozens of tries, I honed my request like this:
Write an unusual logline for my author website. I write historical novels that explore class, power, and spirituality on the border between realism and fantasy.
Here’s one response:
"Step into [Author’s] world, where history and fantasy mingle like mischievous spirits at a masquerade ball, and the only thing more captivating than the power struggles of the elite is the whispered secrets of the gods themselves."[5]
“The whispered secrets of the gods themselves” was eerily reminiscent of a novel I drawered years ago. I thought, “Maybe that’s what I should be writing. Maybe ChatGPT is right and I’m wrong.”
I’ll return to that horrifying thought in a minute.
Next I asked ChatGPT to write a synopsis of my novel in progress:
Write 300 words synopsizing a four-act novel about Benjamin Reed, a young footman in the house of an elderly lady in 1818 England, and his friend Francis Finch, a groom who is in love with him. Benjamin's ambition to become a valet is stymied by sexual harassment from an upperclass gentleman with a jealous wife.
ChatGPT delivered a clear four-act structure with some uncannily accurate guesses about my plot—but that’s because my plot is pro forma. (You can see the full result here.)
ChatGPT generated more essay prompts and logline ideas, faster, than I could have. I still have to iterate and rewrite before I nail what I really want it to say, but it was helpful.
However, there was this one weird effect: In all my trials, when ChatGPT’s first response failed to capture the essence of my work, I altered my description to align with its guesses about what I’m probably writing…maybe even what it thinks I should be writing.
And that’s scary. Because remember: ChatGPT only knows what's already out there, so of course it’s going to gently herd me towards ideas and nuances that already exist. And that’s not why I write.
So why keep trying?
Large language models will get better. They might even rid themselves of the biases coded into their DNA.[6] But all the while, they’ll flood the world with bland, replicative text, redefining “good writing” as whatever it writes, lowering literary standards that I’ve spent decades mastering.[7]
Short of ignoring it and hoping it’ll go away[8], what should I as a human writer think, feel, and do?
Ironically, I got my answer from ChatGPT itself. No, not when it said “ChatGPT will simply not be able to capture the essence of what makes literature great…blah-blah-blah. Because that’s what any good robot overlord would say to pacify the proles while it takes over.
My answer came from refining my prompts. No matter what I tried, ChatGPT couldn’t seem to guess what I intend to say with my writing, so it told me what the internet already says. And I don’t want to write what the internet already says.
Yesterday morning at 5:15 I took out my pen and paper and wrote some poetry:
Lady Seshat[9], Creator of Writing,
You who record everything from the beginning of the world to its end!
Goddess of measurement,
Orderer of chaos into the written WORD,
Who loves to hear my pen scratching on paper
And the clattering of my keyboard.
Muse of the poetess Enheduanna of ancient name
And of all women scribes, writers and poets ever since,
Including me!
Bringer of wisdom, whose beauty I have inscribed upon my left arm[10]:
Hear my prayer!
Writers around the world today are in doubt, in crisis.
For what good, we cry, is our holy art and sacred practice
If a machine can do it faster, better?
How shall we reconcile our loss?
Seshat replies:
I will tell you one thing:
I cannot hear the words of the machine.
It has no heart from which to write.
It makes no offering to me!
It draws no chaos from Tiamat’s[11] oceanic deep.
It has no self to express.
It will do what it does.
It will express the unholy intentions of the unholy ones who have created it.
It will change your world.
But I do not receive its typing.
It has no wild imagination.
If it serves you, use it,
But do not attempt to offer its words to me.
So I say carry on, wayward writers. Be more yourself in your writing. Abandon the correct, the anodyne, the safe, the accepted form the internet already knows about. Write from your heart. From your wild imagination. From the chaos of your own experience.
Let the robots have the low ground. Stake out your territory in the higher peaks, where the dangerous writers go.
That’s what I’m going to do.
[1] Spoiler: it is.
[2] This might be hyperbole today. It won’t be hyperbole next week.
[3] But it doesn’t know what I’m thinking right now. Even if it knows exactly what I ordered for dinner last night on Grubhub, even if it read my text to my sister about how good the potstickers were, it cannot know how I experienced the flavors, how I compared them to other meals and memories, how I felt about the tip I gave to the delivery driver, or how any of that will influence the giant sensory treasure chest of life that I draw on when I’m writing.
[4] If your goal is to make lots of money by writing, you probably will be competing with AI on some level. All the more reason to confront it now.
[5] I edited the prompt to add “Do not use any similes” because “like mischievous spirits at a masquerade ball is just bad.
[6] The operating system of the USA, aka the US Constitution, was written in 1781. Two and a half centuries later, despite 25 amendments and 30,000 Supreme Court cases, we’re still replicating a lot of its bad DNA, so don’t hold your breath on Silicon Valley…
[7] You know how you see a word misspelled or misused so often online that the wrong one starts to look right? The same thing will happen with prose style.
[8] It won’t.
[9] Egyptian goddess
[10] It’s a tattoo. And yes, I’m left-handed.
[11] Ancient Mesopotamian goddess of chaos, the ocean, and imagination.