No wonder Robert Topinka found himself in a quandary (The software says my student cheated using AI. They say they’re innocent. Who do I believe?, 13 February). To test ChatGPT’s abilities and weaknesses, I asked it to write a short essay on a particular topic that I specialised in. Before looking at what it produced, I wrote my own 100% original short essay on the same topic. I then submitted both pieces to ChatGPT and asked it to identify whether they were written by AI or a human. It immediately identified the first piece as AI-generated. But then it also said that my essay “was probably generated by AI”.
I concluded that if you write well, in logical, appropriate and grammatically correct English, then the chances are that it will be deemed to be AI-generated. To avoid detection, write badly.
Prof Paul Kleiman
Truro, Cornwall