This morning I received the copy-edited version of a paper I recently submitted to a journal. This is always cause for celebration, another step towards publication accomplished! But this time the copy-editor wasn’t human, it was an AI.

We have undertaken a light text edit using the tool Paperpal Preflight; please check all edits to ensure you are happy with the changes made.

Right off the bat I noticed they had Americanised all my spelling. That’s annoying, especially for a European journal. I checked the journal’s author guidelines: write in English, they say. It doesn’t specify which English.

There’s the usual annoyances that I’ve experienced with human copy-editors too: every “like” was changed to “such as” and every “so” to “thus” or “therefore”. Some journals do require that sort of language, but I don’t like it. I don’t like having my “don’t” changed to “do not” either, but OK.

They copyedited direct citations. Hey, that’s not cool. And honestly, a half-decent copyediting tool should be able to identify a block quote and skip copyediting it.

And it adds mistakes. The name “Perez y Perez” is corrected to “Perez and Perez” which is certainly not correct.

The word “demonym” is changed to “denomination”, which completely changes the meaning:

Some of the corrections were just strange. “Aer indicatesmeans air in all three languages.” Why would you change means to indicates in that sentence? It diffuses meaning. It added a “the” in front of ChatGPT. I have never seen ChatGPT referred to as “the ChatGPT”.

The irony here is that the paper that Paperpal butchered is about how generative AI normalises and homogenises our writing. Which is exactly what Paperpal was doing to my writing.

Paperpal is one of many AI writing assistants being touted specifically for “academic writing”. It’s obviously mostly meant for students, with tools that both check for accidental plagiarism and paraphrase text for you. The unspoken suggestion is that you can copy and paste text and it will paraphrase it so the plagiarism isn’t detectable. It has a disturbing feature called “Citation Generation” that allows you to “Find generate, and cite papers in 10,000+ citation styles.” I’m not sure I want my AI to be generating citations for me – does it just make it up at random, do you think, or does it try to find a real paper and just make it sound relevant? The website promises both US and UK spelling and style, so perhaps the journal I had submitted my paper to simply didn’t take the time to check the box for UK English.

It feels like a huge waste of time to have to go through each of hundreds of tiny changes to my text to reject most of the edits. There were a few good edits. But almost all of them were either homogenising my writing by making it more American and more blandly impersonal, or they were actually adding mistakes to my work.

At least I now have an extra example to add to the paper of how generative AI homogenises the way we express ourselves.

The paper will be out soon so you can read it for yourself!