GoKawiilSome AI-detection tools used to analyse students’ work have been flagged as untrustworthy.Credit: Paul Guzzo/Getty
Last November, Lauren Jager, a chemistry undergraduate student at Idaho State University in Pocatello, was applying to PhD programmes when she noticed that some application portals warned students about using generative artificial-intelligence tools for their personal statements. They informed students that they would use detectors to sniff out applications that contained AI-generated text. The portals weren’t specific about which detectors they were using. But they were clear on one thing: “They said that if they felt that the personal statement had been written with AI, then they would disregard your entire application,” Jager says.
She didn’t think much of it — she hadn’t used AI at all — but a friend said they’d run their own statements through an AI detector on the Internet, just for safety. Jager decided to do the same with a few detectors she’d found online.
“They all came back at almost 100% AI,” she says. “I started freaking out.”
AI writing tools could hand scientists the ‘gift of time’
Looking back, Jager wonders why her essays were flagged. “I’m a chemistry major, but I was almost an English major. I grew up writing a lot and I used to study grammar books,” she says. “I think of myself as a good writer, but I write very much to the book, following all the rules specifically. Maybe that’s why it thought it had been written by AI.”
She ended up rewriting her statement entirely. And, instead of trying to write to the best of her ability, she says she wrote in a way to make sure it wasn’t flagged by the AI detector. “I was making it less perfect,” Jager recalls. When she ran her essay through the checkers again, they predicted that it was 30% AI-written. “I called that good enough and sent it in.”
When we last spoke to Jager, she had been accepted to do her PhD at the University of Utah in Salt Lake City.
Jager’s situation and cases like it are happening around the world. Universities that are struggling with a surge of written material that might have been generated, or heavily shaped, by AI are turning to AI-detector tools to help them grapple with the problem. And, just like other AI technology, such tools are far from perfect.
Educators and copycats have been trying to outsmart each other since the first Sumerian student copied their classmate’s cuneiform. In the modern Internet era, software provided by companies such as Turnitin, based in Oakland, California, detects text similarities on the basis of a vast corpus of previously published work. This has made it much harder for people to plagiarize text directly. In response, cheaters went online to find ghostwriters, paying others to produce their work for them. Many institutions tried in-person or timed exams to prevent this. But these come with their own set of issues, potentially disadvantaging certain groups, encouraging rote memorization and preventing students from demonstrating their ability to conduct deep research.
... continue reading