Are AI-Powered Tools - and Cheating-Detection Tools - Hurting College Students?

A 19-year-old wrongfully accused of using AI told the Guardian's reporter that "to be accused of it because of 'signpost phrases', such as 'in addition to' and 'in contrast', felt very demeaning." And another student "told me they had been pulled into a misconduct hearing — despite having a low score on Turnitin's AI detection tool — after a tutor was convinced the student had used ChatGPT, because some of his points had been structured in a list, which the chatbot has a tendency to do."
Dr Mike Perkins, a generative AI researcher at British University Vietnam, believes there are "significant limitations" to AI detection software. "All the research says time and time again that these tools are unreliable," he told me. "And they are very easily tricked." His own investigation found that AI detectors could detect AI text with an accuracy of 39.5%. Following simple evasion techniques — such as minor manipulation to the text — the accuracy dropped to just 22.1%. As Perkins points out, those who do decide to cheat don't simply cut and paste text from ChatGPT, they edit it, or mould it into their own work. There are also AI "humanisers", such as CopyGenius and StealthGPT, the latter which boasts that it can produce undetectable content and claims to have helped half a million students produce nearly 5m papers...

Many academics seem to believe that "you can always tell" if an assignment was written by an AI, that they can pick up on the stylistic traits associated with these tools. Evidence is mounting to suggest they may be overestimating their ability. Researchers at the University of Reading recently conducted a blind test in which ChatGPT-written answers were submitted through the university's own examination system: 94% of the AI submissions went undetected and received higher scores than those submitted by the humans...
Many universities are already adapting their approach to assessment, penning "AI-positive" policies. At Cambridge University, for example, appropriate use of generative AI includes using it for an "overview of new concepts", "as a collaborative coach", or "supporting time management". The university warns against over-reliance on these tools, which could limit a student's ability to develop critical thinking skills. Some lecturers I spoke to said they felt that this sort of approach was helpful, but others said it was capitulating. One conveyed frustration that her university didn't seem to be taking academic misconduct seriously any more; she had received a "whispered warning" that she was no longer to refer cases where AI was suspected to the central disciplinary board.
The Guardian notes one teacher's idea of more one-to-one teaching and live lectures — though he added an obvious flaw:
"But that would mean hiring staff, or reducing student numbers." The pressures on his department are such, he says, that even lecturers have admitted using ChatGPT to dash out seminar and tutorial plans. No wonder students are at it, too.

The article points out "More than half of students now use generative AI to help with their assessments, according to a survey by the Higher Education Policy Institute, and about 5% of students admit using it to cheat." This leads to a world where the anti-cheating software Turnitin "has processed more than 130m papers and says it has flagged 3.5m as being 80% AI-written. But it is also not 100% reliable; there have been widely reported cases of false positives and some universities have chosen to opt out. Turnitin says the rate of error is below 1%, but considering the size of the student population, it is no wonder that many have found themselves in the line of fire."

There is also evidence that suggests AI detection tools disadvantage certain demographics. One study at Stanford found that a number of AI detectors have a bias towards non-English speakers, flagging their work 61% of the time, as opposed to 5% of native English speakers (Turnitin was not part of this particular study). Last month, Bloomberg Businessweek reported the case of a student with autism spectrum disorder whose work had been falsely flagged by a detection tool as being written by AI. She described being accused of cheating as like a "punch in the gut". Neurodivergent students, as well as those who write using simpler language and syntax, appear to be disproportionately affected by these systems.

Thanks to Slashdot reader Bruce66423 for sharing the article.

Read more of this story at Slashdot.