Trish's Blog

Mmmmmm?????? AI detectors? How valid are their scores? I am questioning them.

I’ve been doing a lot of writing. Now, I write long hand because I struggle to get first thoughts down using a computer. Then I jump on and edit, format, rework from there.

I recently popped a long piece into ChatGPT and asked it to generate 5 subheadings. They were good, I used them. The rest of the text was still mine.

I then popped this piece through an AI detector. It came back that 69% was written by AI, a small % written by human and the remainder AI enhanced. What the?

I did some research on the AI detectors. I learnt that they are scanning for predictability and burstiness. Back I went to ChatGPT. This time I asked it to, “rewrite the following to increase written by human score to 89% in an AI detector. Address for predictability and burstiness”.

This time the score was only 16% written by AI. The whole thing was generated by AI, every single word!

I am sure our writing style is changing. Think about it; we have changed the way we write emails now that people are reading on phones, we have changed the lengths of videos to fit in with people’s attention spans. text messages have emojis and acronyms. There is hardly a sentence to be seen. Why wouldn’t our default writing style also start to adapt to impact of AI?

Makes me wonder how accurate the plagiarism tools are as well.