SACRAMENTO, Calif. — Graduation is a time of joy, but for the University of California, Davis graduate William Quartermann, his last semester has been tainted, he said.
He was accused of cheating on a midterm question using answers from the AI program ChatGPT, an accusation he denies.
“I’ve never opened ChatGPT to my knowledge,” Quartermann said. “I’ve never downloaded any machine learning program to my knowledge.”
The history major said it happened in February.
Emails provided by Quartermann, which were given to him by his UC Davis Student Services Judicial Affairs case worker, show his instructor ran his midterm paper through a program called GPTZero, which is designed to detect ChatGPT usage.
The program said his answers were likely written by a chatbot. Emails also show his instructor asked ChatGPT if it was the source of Quartermann’s answer to her exam question, which ChatGPT said it was.
Luckily for Quartermann, he had used Google Docs to write his answers.
“I was able to tell them here’s my edit history from my Google Docs, of my step-by-step process of answering the question,” he said.
Stacy Vander Velde is the director of the UC Davis Office of Student Support and Judicial Affairs, which handles any alleged cheating at the university.
Vander Velde said the misuse of AI is an evolving challenge for all universities.
“It’s very difficult to stay on top of it,” she said. “There’s new versions of AI-detection tools being developed as well as just new versions of ChatGPT.”
Her office from the start of the year to the end of their spring quarter has had around 100 alleged AI cheating referrals, Vander Velde said.
“At this point we are starting to see a pretty significant increase of referrals that allege use of AI,” Vander Velde said.
Turnitin is the main software Vander Velde said they use, but notes instructors are allowed to use others.
Both AI detectors claim to have a 99% false positive rate in detecting human-written papers on their websites and warn they should be used only as an indicator.
Vander Velde said they look at a variety of factors when determining if cheating has occurred, and through a multitude of factors her office was able to clear Quartermann of any wrong-doing.
He said along with his Google Doc history he was able to prove GPTZero’s unreliability by showing it flagged Martin Luther King Jr.’s famous “I Have a Dream” speech as AI written.
Quartmermann said the entire ordeal was traumatizing and felt his academic future was on the line throughout the process.
He said he wishes the instructor had come to him first before referring him to the judicial body.
“That’s the thing, I was worrying so much about this that my other classes suffered as a result,” Quartermann said. “She could have taken 10 minutes to ask me, ‘Can I see the edit history of your midterm to check that you actually went through and did it?’”
Making sure students are supported throughout the referral process is something Vander Velde said they do as much as possible.
“So we’re cognizant that our outreach to a student and making allegations of a policy violation can be very stressful,” Vander Velde said. “There are opportunities for students to have an adviser traveling along with them through the process. They’re there to provide support and kind of advise them along the way.”
Ultimately, Vander Velde said instructors have the discretion to speak with a student before a referral or not.
She also noted most students referred to her office admitted to using an AI program for their answers.
Quartermann said having your written history is something all students need to be conscious about.
“I hate saying use one particular program or another,” Quartermann said. “But use Google Docs or any program like that, that has a record of your edit history.”