Identifying AI in Student Papers: No Ethical Use in Academia
Is Turnitin making student papers available to train AI?
I am a doctoral student and teaching assistant and frequently have the pleasure of marking student assignments. The reliance on AI in my most recent cohort was overwhelming. While it is easy to tell which articles are entirely AI generated (e.g. buzzwords, lack of context, word salad), those that show effort from the student to work with the generated text create a more insidious problem.
While I certainly have tried to find a case for AI in my own work, there appears to be no way to do so that is both successful and ethical. While it may be possible to mask the use of AI in the first, the bar of ethicality seems to be inherently impossible to overcome.
My university (and most others) uses Turnitin1 which, for the uninitiated, is a software that detects plagiarism by comparing the assignment against their database of various academic texts. Now I'm an easygoing marker, but I take plagiarism personally. Academia thrives on sharing ideas, not presenting others' hard work as your own. Put more simply Copying is Not Theft, as long as you don't claim you created it yourself.2
Now Turnitin is not without controversy, though this has increasingly been ignored as it became the inevitable law of the land. Turnitin, like ChatGPT, relies on large datasets of existing data that it hoovers up from the Internet, academic databases and, most controversially, student papers. That is, every submission analyzed in Turnitin is added to the database and will be compared to future submissions. This is controversial because students' original works are added to the database without their notice or choice.3
And here's where the problem of using AI, even to write drafts, enters.
In my most recent student cohort (a first-year course of over 300 students), I noticed many papers with high Turnitin scores. Far higher than in previous years. This score identifies what percentage of the paper comes from other sources. In some cases, this will indicate common phrases or idioms which don't necessarily need to be cited. It is also common for Turnitin to "catch" items from the reference list. Of course, none of this matters if the student cited the quoted material properly.
However, in my own attempts to use ChatGPT I find that, when asked for references, it frequently identifies unrelated ones or else fabricates them entirely. Reports indicate that this effect, known as "hallucinating," is increasing.4 This alone would be good enough reason to avoid AI like the plague, but my experiences with my most recent cohort of students suggest something more insidious.
In this cohort, I noticed that Turnitin detected direct quotes from other student papers, written at universities around the globe, in a significant number of submissions. Now one, or even a handful, with a quote from a paper at, say, the University of Wellington in New Zealand would be an oddity, but dozens or even hundreds? I neither believe that this is happening by pure coincidence, or that my students are systematically and knowingly stealing from students at universities around the world. Frankly, the coordination and organization required would almost be enough for me to congratulate them.
No, I suspect what is happening is that Turnitin has contracted with companies like OpenAI to train services like ChatGPT on their existing databases, which is mutually beneficial, as it would enable Turnitin to credibly promote their use of AI to detect plagiarism (Chechitelli, 2023). This means that a student may use ChatGPT to write a first draft and go through significant work to edit and focus the article, clearly making the piece their own, while being completely unaware of the fact that they are still plagiarizing. This happens because ChatGPT, like any AI, doesn't create; it merely compiles snippets of various related texts into a whole. Unless the student then changes every single word in the resulting document, they are inevitably going to plagiarize. And because ChatGPT is apparently pulling data from both public and private sources, the author cannot even determine what may have been plagiarized, particularly since ChatGPT does not seem to be able to reliably communicate where and what they are quoting.
Unfortunately, however, I can't blame ChatGPT for academic misconduct, only the individual that submitted the article.
And you may say that it hardly matters if the article answers the posed question, but I would answer that submitting others' ideas as your own, even if you are unaware that you are doing so, is robbing both the originator of that intellectual work, and yourself of the education you are paying for. It is hard to write an academic article, especially for a first-year student. It's supposed to be. The point is that you have to keep doing it to get better at it, and this will never happen if you let AI do the hard work for you. This is no less true in the day-today life of non-students.
If you're using AI to write for you, you're asking it to steal from others, and unless you change every word, it's going to keep being theft. But even if you don't care about that, you're robbing yourself of the opportunity to get better at something, and isn't that the whole hacker ethos?
If you've had similar experiences, or have other academic hacker related concerns, please reach out to the address above.
References
- Turnitin. (n.d.). www.turnitin.com; Chechitelli, A. (2023, Jan 13). Sneak Preview of Turnitin's AI Writing and ChatGPT Detection Capability
- Question Copyright. (2010, Apr 2). Copying Is Not Theft (Video)
- Vanacker, B. (2011). Returning Students' Right to Access, Choice and Notice: A Proposed Code of Ethics for Instructors Using Turnitin. Ethics and Information Technology, 13, p. 327-338.
- Murray, C. (2025, May 6). Why AI "Hallucinations" Are Worse Than Ever Forbes (Mirror)