A college professor said he found an easy way to catch AI-generated plagiarism after finding phony citations in some of ChatGPT’s content.
“It’s very easy to identify the fake references,” said Terence Day, a physical geography professor at Okanagan College in British Columbia. “All you need to do, really, is to check them up on the internet.”
WATCH: COLLEGE PROFESSOR DETAILS HIS AI PLAGIARISM DETECTION METHOD
Day suggested that professors require students to attach a hyperlink for each reference included on a class work.
“That’s usually done in terms of what’s generally called the DOI, the digital object identifier, and that is a hyperlink,” he told Fox News. “You click on that. Does it exist? Does it not exist?”
Day detailed his detection method for fake AI citations in a peer-reviewed research paper published earlier this month in The Physical Geographer. He developed the approach after experimenting with ChatGPT and found it produced answers to his geography-themed questions with seemingly legitimate citations.
But upon further inspection, those AI-generated references turned out to be fake, according to the professor.
“The references and the citations associated with my inquiries … were unfamiliar to me,” he told Fox News. “So, I checked them out. And what I found was that they were all completely bogus.”
“I went to the home pages of the journals and I went through the volumes and the page numbers, and they were not there,” Day continued.
The professor said he entered some of the ChatGPT-cited journal titles into Google Scholar, a scholarly literature search engine, but they didn’t appear.
“I was a little flummoxed and tried one or two more — and more and more and more,” he told Fox News. “I never found one that was accurate, complete and existent.”
Day said all the citations he double-checked were seemingly falsified. He added that they were “presumably produced by the algorithm as part of a predictive process based on the … limited training that it has in a particular field.”
The professor said he felt his approach could more reliably detect plagiarism than competing AI software.
“There is a growing interest in terms of plagiarism detection software, which is capable of detecting a AI chatbot’s written material,” he told Fox News. “The problem is it only gives a probability that the material is plagiarized, and it cannot definitively state whether or not it was written by a person or whether it was written by the ChatGPT.”
Day said he believed his method’s simplicity decreased ambiguity around whether a student cheated.
“The advantage of the approach that I’m promoting here is the idea is just simply checking the references,” he told Fox News. “If the references exist, that … suggests that probably it’s genuine.”
“If the paper does not exist, then I’m sorry. You got caught,” the professor added.
To watch the full interview, click here.