I never used to check my writing with an AI detector before joining Medium. And even when I first published on the platform, I didn’t check.
But then I read a few stories from writers who indicated their writing, or some of it, had been flagged as AI by an AI detector checker. I thought that I had better check my stories just to see.
For the most part, they pass with a big zero.
But occasionally, I receive a return that tells me that anywhere from 3% to 10% is most likely AI. My usual reaction is, “Really?” Often, it is only one line, one sentence, or at most a paragraph. Occasionally, it may consist of just a few words.
I ask myself why would the detector think that an occasional line in a story of several hundred words, or even a few thousand, is most likely AI-written? What is it about the words that makes the detector algorithm think that it has been written by AI? And typically it will say, 100% certain.
For example, the line below.
Of the two, I prefer the astronomical summer because it lasts longer!
It was part of a story I wrote about the two end dates of summer in the UK, meteorological and astronomical. I wrote it for the reason given; the end date of the astronomical summer is later than the other date.
It was a simple enough sentence. Anyone could have written it. But the detector thought that AI wrote that line. It didn’t make sense to me. Was it because I had used a fancy word like ‘astronomical’? Or maybe it was the exclamation mark at the end? Perhaps AI did not think that a human would write that way, to emphasise being happy?