The AI made mistakes, according to the humans, flagging an estimated 350 publications as questionable when they were likely legitimate. That still left more than 1,000 journals that the researchers identified as questionable.

That’s quite a large error rate.