Creative writing in the age of AI

Years ago, if you wanted to write a short story, you needed a compelling idea, some time, and lots of space for revisions. Over time, you might walk away with a good piece of work. Sometimes not. But a thoughtful writer grows with the process.

This is no longer case. Simply go to Chat GPT, write a simple prompt and next thing you know, you are writer although you didn’t write anything.

The big question, is the generated material any good? Take the prompt: “write a horror short story in 900 words about a woman tracking down a serial killer in a graveyard”.

Here’s the first paragraph from the returned result:

A dense fog rolled over the old cemetery, its tendrils wrapping around ancient tombstones and obscuring the path that twisted through the graves. The moon hung low in the sky, a pale eye that watched as Elise Monroe made her way through the graveyard. She clutched a flashlight in one hand, its weak beam slicing through the mist, and a worn leather-bound journal in the other. The journal had been her father’s, and now it was her guide.

I have a character and a setup. But it’s boring. I don’t mean to be dismissive or come across elitist. I don’t profess to be a literary great, or even a literary adequate, but that opening is all sauce, no cheese.

It’s algorithmically creative, but very artistically flat. The first sentence alone is rejection worthy. Here it is:

A dense fog rolled over the old cemetery, its tendrils wrapping around ancient tombstones and obscuring the path that twisted through the graves.

Does that convince you to read the rest of the story? Are you persuaded to finish the current paragraph? Are you getting invested? For me, that’s a big fat no. By the end of the first paragraph, I’m already thinking about my laundry. How about this:

Elsie’s flashlight cut through the fog just like the razor that had cut through her best friend’s neck. The blinding white mist pooled between the headstones like Jenny’s blood had pooled between the tiles of that bathroom floor.

Perfect? No. Clunky? Sure. It does set up character, setting, and conflict; all wrapped in a disturbing metaphor. I’m not saying it’s great or even good, but it’s effective. It does the job getting you to the next sentence. And if I can keep you moving through the beginning of my story, I might be able to keep you moving through the rest of it.

Yet, the AI version is almost there. One might think AI will replace traditional writing, but I don’t think so.

Not thinking-machines, just machines

At some point, scientists in the future will be able to break down the concepts of intelligence, imagination, and most like likely, the creative process. This will be a sad day, but for now, it’s a magical process. It’s beautiful process. It’s defines our humanity.

However, when it comes to modern artificial intelligence, the content generation process is well known and understood. Even when the machine does things that we don’t expect, it’s not because the machine is “self aware”. Rather, it’s just a demonstration of probability.

AI chip manufacturers and tech bros will explain AI as if it is a machine like Hal 9000, but AI is really just a Price is Right Plinko machine. Plinko was (or is?) a game where contestants drop a disc through a large pegged board. The disc traverses down the board, moving side to side from the pegs. Contestants win a prize depending on the disc’s landing point. It’s all random.

At the end of the day, AI is just a giant complex Plinko machine

AI is like this game. Think of the disc being dropped into the machine as the prompt. The pegs represent the actual data, otherwise known as the model. The model defines relationship between pieces of data. Think of each row being a different piece of data. When the disc traverses the machine, gravity and physics determines its route. In the case of AI, gravity and physics is replaced with statistical probabilities. With each row, the disc assembles all the related data and then produces the final result at the bottom.

When you submit a prompt, you don’t actually see the disc move through the machine. You give it a prompt and you get an answer. There’s no magic or thought. You are working with a machine no different than your car or toaster oven.

The key point: the data model provides both the accuracy and the depth of the information. Needless to say, AI companies are snorting public data off the internet like rockstars at an all-you-can-snort cocaine buffet. This raises questions about copyright and plagiarism since the AI is using the actual sources without explicit permission.

AI and storytelling

When a machine “creates” a story, it’s using existing stories as templates. AI is only as good as the data being used. For large language models, this means hundreds of thousands of sources. Until we know the exact sources being used, I can only guess that there is a wide assortment of literary sources being used along with an ocean of, shall we say, low quality content.

The end result is mid-range generation. You may be able to target a style say based on Dickens by your prompts, but the actual storytelling mechanics may be lacking. If you are really enthusiastic about generating stories based on “the greats” such as Hemingway, you can actually create your own model based just on Hemingway’s stories.

Granted, creating your own model is not guaranteed to create a good story. Probably not even a passable story. You can spend a lot of time tweaking AI to produce the best possible story, but you are essentially abdicating the creative process to a machine. You might make money off in the short term, but you won’t keep an audience in the long term. After all., the audience can do the same thing.

The best thing to do is focus on your own writing versus on story generation.

Just another tool?

Believe it or not, I actually enjoy using AI. I use it for things like SEO and post metadata. I’ve used it at times to generate images and actually use it on my day job. So it’s definitely useful in a variety of contexts.

I DO NOT use it for is my actual writing. All writing on this blog in written by me. Batteries not included. I maintain a “clean room” approach to working with AI. I use it for a variety of tasks, but I never copy and paste nor do I rewrite AI generated content. All my writing is original.

(Photo by cottonbro studio)

That said, just because I won’t use AI, doesn’t mean other writers are the same. You can expect lots and lots AI generated content. In fact, the internet is already flooded with it. I imagine in five years, the vast majority of online content will be AI generated. It’s “too convenient” and “too cheap” for the big players to not use it.

It’s already affected the writing market. Clarksworld, a famous science fiction short story magazine, shut down submissions last year due to all the AI generated submissions. Self publishing stores such as Kindle Unlimited are being filled with AI generated books.

Where to go from here?

Honestly … who knows? This feels like uncharted territory. I’m sure people felt the same about the steam engine, but this is different. This is art. This our collective soul. In a way, embracing AI art erases our place in the world. Japanese animator Hayao Miyazaki put it rather bluntly:

I am utterly disgusted. If you really want to make creepy stuff, you can go ahead and do it. I would never wish to incorporate this technology into my work at all. I strongly feel that this is an insult to life itself.”

– Hayao Miyazaki

While I don’t feel as strongly as Miyazaki, I don’t think he’s wrong. To abdicate our dreams to the machines leaves us as devoid as discarded toaster ovens. So think long and hard before you leverage AI in your art. Embrace life with your own hands and leave the machines to handle the taxes.

By Brian Moakley

Brian Moakley is a writer and editor who lives amongst the quiet hills in New England. When not reading tales of high adventure, he is often telling such stories to all who will listen.

Related Post

Leave a Reply

Discover more from Jezner Blog

Subscribe now to keep reading and get access to the full archive.

Continue reading