Turns out, it’s pretty easy. AI detectors are awful at detecting human writing, but have a fairly decent track record at distinguishing AI written work.
Why does this matter? Well, depending on your writing style, usage of quotes, and stylistic choices, your 100% human-written essay may be flagged as AI generated. Bye-bye scholarship.
Whatever level of accuracy these AI detectors “claim” on their websites, slash that number in half. At least. Sure, they can detect AI, but they often flag AI usage incorrectly against pure, homegrown human writing.
So what’s the deal?
GradSimple tested AI detection reliability against human prompts, AI writing and prompts that were run through an AI humanizer – to give it more of that classic human imperfection. The results?
Out of five human-written prompts run through seven AI detectors, not a SINGLE prompt was detected as being 100% human by all of the detectors.
The same can be said of the AI-written prompts. When run through the same seven detectors, the results were all over the place. Some detectors returned “100% human” ratings, and some were able to return more accurate, consistent results.
Why it matters?
Well, think about that essay you’re doing for History 101: The History of Historical Stuff. You poured your blood, sweat, and tears (a lot of tears) into this mediocre masterpiece, and you’re finally ready to submit. You cited your sources, quoted your quotes, the works. Your professor has you turn in your essay through some fancy new AI detection software, and BAM.
Result: 50% AI generated. Academic probation, expulsion. Life over, scholarship lost, end of the world as we know it. Okay OKAY, kidding.
But really, this could seriously impact you.
Colleges should be mindful of the variability of AI checkers, especially with the uptick in plagiarism softwares now implementing AI detection. It’s often going to be a student’s word vs. the machines.
Professors, please, don’t be assholes.
Before marking a submission with a big fat zero and shipping your student off to academic dishonesty island, slow down, read the paper yourself, and have a chat.
Moral of the story? AI detection software sucks very often, as you can see, and is probably going to suck for a long, long time.
