Why I Want to Punch AI Detectors in the Face, Repeatedly
Updated: May 30
Another year rolls by like it did billions of times. But, what makes this year different for millions of writers is that, and I write this with a lot of restraint, it has been a veritable AI-infested shit sandwich from start to finish.
At this juncture, people's eyes have already started rolling back into their sockets. "Great, another rant about AI from a writer. Who thought this was coming?" you're probably thinking. No, hear me out.
My stand on AI is pretty much the same today as it was years ago. AI exists, and it will take my job away if I don't pull my socks up. And, to be honest, I am fine with that. I've been writing for over a decade - my socks are pulled.
For me, AI is an evolution of technology that is inevitable and even exciting. I'm not some deluded old man who's shouting at the AI-generated clouds about how our lives were better in the "good ol' days" when there were no vaccines, starvation, torture, and the English Inquisition. Not in the least.
My stand on AI is pretty much the same today as it was years ago. AI exists, and it will take my job away if I don't pull my socks up. I'm not some deluded old man who's shouting at the AI-generated clouds about how our lives were better in the good ol' days.
My issue is with AI detectors. They don't. FUCKING. Work.
AI Detectors and How Companies Continue to Selling Shit That Doesn't Work
Do you know what the similarities are between AI Detectors and politicians? They're both tools that don't work (ba-dum tish).
That badly framed joke aside, AI detectors are definitely something to worry about for the writer, and they probably lie more than the average politician, which is saying something.
So, picture this - it's a warm sunny day and I'm saddled with the unenviable task of writing a listicle titled "10 Reasons Why You Need a Car Injury Lawyer." I motivate myself with the reminder that at least I'm not one of those children mining lithium that end up in some Insta model's iPhone.
Once I'm done, I pop it into the AI Detector "tool" with the self-proclaimed 99% detection rate (spoiler - it's CopyLeaks); it says my entire article is AI-generated.
I take a deep breath and rewrite the highlighted parts of the article. Nope. It still says my text is AI-generated. Unless, by some Shamalanaesque twist, I'm the supreme AI that I've been fighting all along, this is pretty much impossible. I've got minute-by-minute timestamps from my Google Doc history to prove my point. You could argue that I could have another screen open with ChatGPT suggestions on it that I could use, and that is a fair point. So my next mini-project is to record a video over my shoulder of me writing content and then feeding it to the AI detector, just so you can see it rejecting my article.
The only thing stopping me from doing this is probably my extreme social awkwardness and refusal to revise my COVID WFH wardrobe choices of vests and boxers.
So my next mini-project is to record a video of me writing an article and then feeding it to the AI detector, just so you can see it rejecting it. The only thing stopping me from doing this is probably my extreme social awkwardness and refusal to revise my COVID at-home wardrobe choices.
So, coming back to the "AI Content Detected" saga, it goes on for about three more attempts until I get the A-okay from this infernal block of code disguised as something that actually works. And the best part: the article read far worse than when it first started. The simple solution to getting detectors off your back is so counterintuitive that it would make Dr. House blush… just throw a bunch of random fillers at your text, break up your sentences, and hope for the best.
How and why does this happen? I don't know. If I were knowledgeable enough about programming and AI to answer that question, I really wouldn't have to write for a living. I could just copy stuff off Github, have AI generate code blocks, and call myself a programmer. Plus, I wouldn't need to subject myself and my code to scrutiny by some Snake Oil 3000 program that thinks it can recognize human writing patterns.
What Tools Should Do
I actually cringed when I wrote the subheading. "What tools should do"? For starters, it should fricking work, how about that? I honestly didn't see myself writing anything with a subheading like this, but 2024, and here we are.
"What tools should do"? For starters, it should fricking work, how about that? I honestly didn't see myself writing anything with a subheading like this, but 2024, and here we are.
But anyway, let's compare it to another tool I (and 99% of all writers) have used for close to a decade, Copyscape. For people unfamiliar with it, Copyscape has been the standard for plagiarism checking for writing. It's generally super reliable and highly accurate.
When I first started out, like any other writer, I did my fair share of "rewording" for my content. And when I get a bit too lazy with the writing, Copyscape flags my content for plagiarism. And it has been true EVERY time I've used it.
For me, Copyscape is a good example of how a tool is supposed to work. With enough experience, you generally have an idea of the result. If you plagiarize stuff, you can bet it will find out.
Contrast that with this… thing called the AI detector. It's like Forrest Gump's box of chocolates, but instead of treats, there are just big red buttons that activate various calamities, from getting hit by Category-5 hurricanes to being mauled by a hungry tiger.
Why AI Detectors Aren't Tools
Classifying AI detectors as tools when they're pretty much broken at every level is just wrong. Imagine if a carpenter's chisel worked like AI detectors. The next time your hammer comes down, you really wouldn't know if it pares wood or finger.
The worst part is that thousands of people around the world are affected by this. I see posts regularly on online forums and socials about how students' professors/employers shot down their work just because some program thought it was AI-generated.
I once ran a part of an article that I wrote in 2020 through an AI detector, and it flagged it as AI-generated content. This is probably the best proof I have of AI programs being horrendous at their job.
I've added a link so you can check it out for yourself -
That is a copy of the original file. What sort of accuser would I be if I didn't provide proof? So, here are the receipts -
The best part is, like any other scam, these companies are profiting off this broken product, and it isn't cheap. Imagine how devious you'd have to be to create a defective product that grabs people's money while being less accurate than a broken clock and invalidating the literal existence of millions around the globe. Yes, you're right, person in the back; dating apps are terrible, but we're bashing on AI detectors today, okay?
So What Happens Now?
Considering my track record of predicting things, I shouldn't venture further into the question. As a writer, I find it bad enough that AI takes away a significant portion of our income. Now, we have to put up with the fact that these "detectors" wrongly flag what some poor sod worked his ass off. What gets my goat is that these websites flaunt 99% accuracy, and nobody calls this shit out.
It's like we've taken a trip down memory lane to the era when doctors endorsed certain brands of cigarettes. No, I'm not kidding! And now, even in 2024, things haven't really changed. Apparently, people will do anything for a quick buck at someone's expense. Color me mildly shocked.
And one niche that is affected even worse than writing is academics. This area is where originality and innovation are paramount, and the consequences of being mediocre are particularly harsh if not career-ending. Many find themselves caught in the headlights where their efforts are overshadowed by the skepticism triggered by these overzealous AI detectors. It's just stupid how people today trust AI detectors blindly without even bothering to read what is submitted.
If you existed back in the 90s, there were these fun little astrology programs where you could type in your name and birthdate, and it would give you a "prediction." These AI detectors are pretty much the same block of code. The difference is that they cost money, and a lot of it.
To be honest, I'm really not sure where we're heading with these things, but it's nowhere closer to being more productive. The only hope is for AI detectors to get better, but will that happen anytime soon when they're already making tons of money off gullible people? Not likely. But like always, time will tell, I suppose.
Comments