There is an ongoing trend for book bloggers and reviewers to run novel content through an AI checker and then flag the work according to what the checker generates. 

This is wrong on so many levels.

For the sake of this article, I will ignore the copyright breaches necessary to feed a novel’s content into an AI checker and concentrate on another fundamental that the reviewers and bloggers either do not know or choose to ignore.

Note to reviewers: To copy a novel’s content (which you must do to run it through a checker), you need the written permission of the owner; otherwise, you are in breach of copyright.

Big Data companies train their generative AI algorithms using pirated content of existing novels. So, what does this mean? Put in laypeople’s terms, it means the checkers are checking the content of new works against the content of old works used to train the models. They check for trends common in the data lakes used by AI, trends that all creatives will use to a greater or lesser extent. Any check against AI will flag halfway decent work or work in the data lake because that is what generative AI is—a lake of examples it uses to generate (not create) text. This is completely illustrated by both Pride and Prejudice and the American Constitution being flagged as 100% AI-generated.

What’s the message?

While the checkers are unreliable, we should probably leave AI policing alone and concentrate on convincing the relative legislative bodies to ensure those who use AI to generate creative work are transparent. Given time, checks and balances will be put in place (I hope), but current AI checkers are unreliable.

All of our novels are written, edited, and packaged by humans. This is our pledge to readers.