> This is adjacent, but I’ve wondered how close we are to having these automated plagiarism detection systems start identifying large numbers of false positives

> This is adjacent, but I’ve wondered how close we are to having these automated plagiarism detection systems start identifying large numbers of false positives. Especially as they adopt machine learning, and they let their fuzzy matches get a little more aggressive. "how close we are" ...? The automated plagiarism engines have always had huge amounts of false positives. Essentially they are only useful for highlighting what they detected. Some of the things that I have seen them false detect: Direct cited quotes, common phrases and sentence structures. Factual explanations, e.g. a description of a specific aperatus which can only really be described in one way (even though exact wording differed). Reference lists, this was by far the most common reason for very high turnitin scores, when I used it turnitin was completely unable to pick up a reference list. The plagiarism scores of these systems are completely useless, I've found obvious plagiarism in work with scores of less than 3% (which is a very common score even if you would write an essay now) while I've had work with scores of >50% which were completely fine. Any teacher who uses the turnitin score for anything should be fired IMO, because they do not understand plagiarism.