AI detection

With all this hoo-hah about AI, a market has sprung up for "AI detection." Unfortunately many writers submit 100% original work only to find that AI detectors consider it "50% AI-written" or "50% likely to be AI-written" with the numbers widely varying; it's general consensus that the detection software is useless. One will call it 80%, another 20%; they are so wrong, so often, that they might as well be picking the numbers out of a random number generator.

This leads me to wonder what the heck the schools can do about this. I'm sure AI is generating a lot of homework and getting away with it. But should a school invest in "AI detection?" I don't think so.

But it leads me to my ultimate question: How DO you know what's AI generated? If AI is as smart as we are, and uses language that we've been using for years, and always uses the language that is most commonly used to express something - then, it's kind of like us, in the sense that it's creating language the same way we are. It seems to me that if you track this down carefully enough, and let AI create enough, there will really be no difference anyway. So you can't tell what's AI-generated and what's not. You can't be sure of anything.

Somebody mentioned em-dashes, as if AI kind of had a preference for them, so they've stopped using them altogether so that the AI detection software doesn't scoop them up. But wait a minute! Why would AI prefer em-dashes, or anything, more than the rest of us? That's really a style question, and AI doesn't really have a style, does it?

Just thinking out loud. This "detection" software, after it determines you have written AI-like material, then offers to "humanize" it, making it less like AI. For a fee, of course.