Can You Really Identify AI on Medium?
Is reporting users going too far?
There have been suggestions on Medium that we should report AI generated content that isn’t clearly labelled as such. I understand where this is coming from, but I wonder how easy it is to be 100% sure that something is written by AI?
I was once asked if a piece I’d published was written by AI, and it wasn’t. You can’t reliably tell just by reading a piece. Digging around in someone’s profile might give you some insight to whether an account is automated or not, but that’s time consuming and it’s not fool proof either.
I write for a wide range of different magazines in different styles. Just because one of my articles is more formal than, say, a personal blog or rant that I wrote the day before, doesn’t mean it’s not my work. It just means it’s a different style of article that was originally written for a different market.
What I write for Medium is often less formal than the stuff I write for magazines because the magazines usually couldn’t care less what I think about anything, so the personal element is taken out of it.
I’ve written for trade magazines, lifestyle magazines, and newspapers — they all require different styles.
We are a community of writers from different paths in life, with different interests and backgrounds, and different skills and strengths. This means there are bound to be different styles of writing on Medium, even from the same author.
News reporting will inevitably be more formal than a rant about someone’s annoying comment. But it doesn’t mean it’s AI.
And there-in lies the problem with vigilante reporting. You can report someone for for ‘breaking the rules’ — for using AI without declaring it. But you can’t tell for sure what’s been generated by AI — and you might be wrong.
Some stuff with bulleted lists and typical GPT Chat layouts might look like AI, but before you press the ‘report’ button to dob someone in for using AI without declaring it, can you be 100% sure the author didn’t write it themselves?
Sometimes I think something looks like AI, but then I realise the stilted prose is actually *so bad* it’s more likely to have been written by a human — because AI would do a better job of stringing a sentence together.
AI writing is sometimes better than human writing, particularly if the human author does not speak English as their first language, so they’re trying to write in an unfamiliar tongue. Credit to them for trying, because I couldn’t write anything decent in French. But the point is, a piece of work being flawed doesn’t necessarily mean it’s generated by AI.
It’s impossible to know what is AI generated for sure. That leaves us with a real conundrum, which makes it virtually impossible to effectively ban AI from the platform, or to demand compliance that it’s clearly labelled when used.
© Susie Kearley 2023. All Rights Reserved.
More from me…