Artificial Intelligence: A Knockoff Bag in Designer Clothing

by Sharon Kayne

When you work in a creative industry—say, you are a novelist—you have enough to worry about. Like: will I ever find an agent or publisher? If I go it alone and self-publish will I sell more than 50 copies? Does my writing really suck but are my friends and family just telling me it’s great because they love me? I’ve worried about each and every one of those things. And, as if that wasn’t enough, I now have to worry about AI. Specifically, that some stinking AI program will “determine” that my work is likely AI generated.

For the record: I don’t use AI for anything when I’m writing. Sure, there’s the spelling and grammar check that’s part of the MS Word program, and I use that. Sadly, it’s very often wrong, so I also have to ignore it quite a bit. But I’ve turned off Word’s annoying Co-pilot feature—the one that suggests rewrites you haven’t even requested. I write and edit my own blogs and novels, just like I wrote and edited all sorts of nonfiction pieces as a professional writer/editor for 30 years. It’s not only what I do—it’s also what I love to do. I’ve known I was a writer since I was six. Why would I outsource my vocation to a machine? Particularly a machine that learned by plagiarizing the work of other writers? The whole idea makes me ill.

To be clear, I’m not terribly worried that AI will start writing best-selling novels and put human authors out of work. Even though there’s likely a large market for the kind of generative slop AI would put out, there are still enough discerning readers out there who will reject that crap. No, my real AI fears popped up when I read about Nigerian author Jerry Adedayo Falade. His was a story that debut authors only dream about. Until it became a nightmare. Not only did he secure agents (a proposition with worse odds than winning the lottery and being struck by lightning), but they managed to get him a reported $2 million publishing contract (that may or may not have included film rights) for his debut novel, “Call Me, I’ll Hide the Body.” While such a deal would be a great plot for a book or movie, in real life it’s unheard of.

Sadly, someone—probably at the publishing house—ran Falade’s book through an AI detection program, which determined that his book was likely AI generated. The publishing contract was cancelled, his agents dumped him, and his dream was deflated. For his part, Falade maintains that he didn’t use AI. In fact, he says he has pre-AI text messages to prove it. He also says he may file a copyright infringement suit against whoever fed his book to the AI collective, and I hope he does. I also hope he wins a mint.

The whole idea that you can use an AI program to determine if something was written by a human seems backwards to me. It’d be like ascertaining if the new handbag you bought was a designer original by checking it against known knockoffs. How, exactly, would that work?

As it happens, I recently had the opportunity to speak to someone who teaches robotics and AI at a university. I asked him about the veracity of having AI detect itself in someone’s novel. He admitted that AI detection programs are “problematic.” He also mentioned something about “AI ethics.” I told him I thought that was a contradiction in terms.

Since I was only able to speak to this person briefly at a social event, I turned to the web for more in-depth research. I asked Google how AI detects AI writing. Google’s AI Overview told me, “AI detectors look for mathematical and statistical patterns in text.”

My response was “huh?” Is that some robotic way of describing a writer’s “voice”?

No. Turns out, AI detectors look at three main aspects of writing: perplexity, burstiness, and stylometry. (I kid you not.) Perplexity is whether your word choices are statistically likely. You know how this works: When you’re typing something into your phone or computer, it will give you word prompts for where it thinks you’re going. Apparently, the more perplex your word usage is, the more likely you are human. (The professional writer/editor in me must take this moment to point out that “perplexity” is not the best label for this aspect of writing. If your word choice is perplexing, your readers will have a hard time following you. “Predictability” or “complexity” would be better labels. But we’re talking about a damn computer program that thinks it can write, so never mind!)

If, like me, you’re scratching your head at the words “burstiness” and “stylometry,” allow me to elaborate. Burstiness is about sentence length and structure, which should be varied. Stylometry is all about punctuation habits and repetitive vocabulary words. Allow me to demonstrate some good burstiness and stylometry: fuck AI. Didn’t see that coming, did you?

Well, if you’ve read any of my previous blogs about AI (here and here), you may have seen that coming. Clearly, I don’t like AI. I thought I was safe from it as long as I didn’t use it, but it turns out it might bite me in the ass at some point anyway, by wrongly deciding my own, original work is AI-generated. And that’s one more thing I shouldn’t have to worry about.

I post to my blog every Sunday. Follow me on Facebook, Instagram, or Substack, or sign up for my monthly newsletter (here) to be informed of my new posts. Also, please follow me on Amazon, Goodreads and BookBub.

Buy my historical romances, This Restless Sea, The Green Silk Gown, and Variations on a Romance, at Amazon.