In all seriousness, the best use I've seen for AI writing is to create bad fanfics faster than humans can. And even then they struggle. Now, maybe in the future they will be a concern, but right now they seem borderline useless.
I disagree.
I think for relatively high level work (final year and postrgraduate), AI might struggle. However, if someone's smart about using it, it's extremely hard to tell. For instance, you can constrain the AI: tell it to write an essay on a certain topic using certain references. It'll get the references right, and the information will probably be accurate. There are things in terms of detail and sophistication the AI is not good at, but in the first couple of years of a bachelor's degree, it'll probably generate something that will at minimum pass, and could even be marked as good.
There are giveaways of AI text. You can get odd repeats of information. The sentences are usually short and in very plain English with minimal to no errors in spelling, grammar and sentence construction, so if you have examples of the student's written work that's in a different style, it might stick out. This includes that they default to US spelling, so if you're in a country that doesn't use US spelling, that can be a giveaway. And so on.
There's also an element of structure. If you look at two AI-generated essays on the same assignment, they tend follow almost exactly the same structure - and I mean very, very similar. However, whether you can take action against a student is hard to say: the risk may be that you penalise a student with a very formulaic style, and I'm not sure misconduct processes will support holding a student guilty of misconduct because the style and structure of their essay is very similar to another's (and just like an AI would produce).
www.rollingstone.com
