The University of Minnesota expelled a third-year Ph.D. student in January after he was accused of using generative artificial intelligence (AI) on an exam. He appears to be the first student expelled from the University over AI.

MPR News was the first to report on Haishan Yang’s expulsion. Yang’s lawsuit against the University claims he was wrongfully accused of breaking the University’s scholastic dishonesty rules by using AI on his preliminary exam last August.

Yang also filed a complaint with Minnesota’s Department of Human Rights in February for discriminatory treatment based on his national origin.

Yang came to the University to pursue a Ph.D. in health economics after earning his first doctorate in economics from Utah State University. He said he hoped to pursue research as a professor in the U.S.

Growing up in Quanzhou, a southern Chinese city with a population of more than 8.7 million, Yang’s first language was Southern Min, a Chinese dialect.

Yang said he used AI like ChatGPT to check his grammar as a non-native English speaker but decided against using it for his preliminary exam.

University documents Yang shared with the Minnesota Daily show four faculty graders were doubtful that one section of Yang’s exam was written without assistance from AI. The professors pointed out an acronym not used in preparatory reading for the exam and other details that were included in one of Yang’s essay responses as signs of ChatGPT use.

The faculty graders input exam prompts to ChatGPT, compared its answers to Yang’s and shared the results at the hearing.

“I was struck by the similarities between the two that seemed extremely unlikely to be coincidental,” professor Peter Huckfeldt wrote in a letter for the hearing. 

Yang alleges one professor, Hannah Neprash, modified parts of the ChatGPT-generated response by removing a summary paragraph, excluding a header and changing bolded text to plain text. 

He also said she did not present a shareable link to the ChatGPT answer and instead exported a PDF shared with the other professors reviewing Yang’s exam. Yang filed a defamation lawsuit against Neprash in January.

Neprash declined to comment based on student privacy laws. A Feb. 21 filing said that changing the boldness of text and removing a summary paragraph are “trivial differences, even if intentionally made.”

Roxanne Krietzman, assistant director of student advocacy for the University’s Student Advocate Services, said during the hearing the question under suspicion of plagiarism represented only 6.6% of Yang’s total exam. Krietzman said AI detection on other questions within Yang’s exam reported percentages below 30% likelihood of using AI, compared to the section given an 89% likelihood.

Yang also said methods for detecting AI use are not as reliable for analyzing the work of scholars who are not native English speakers.

The concept of AI detection is based on how words are used and how often, said Louie Giray, an assistant professor of communication at Mapúa University in the Philippines. Non-native English speakers commonly write in simple words and simple sentences that flag an AI detection program looking for abnormal writing patterns, he said.

Non-native English speakers may also use more basic transitions, such as “In conclusion” and “However,” Giray added.

Giray has written about the limitations of detecting AI plagiarism and found that detection programs lack context, such as with non-native English speakers.

A study from the University of Washington found that Turnitin’s AI detector found 25% false positives when analyzing work that did not use AI.

But many scholars advocate for human-AI collaboration in detecting plagiarism, Giray said.

“I don’t think that in the future there’s going to be a fool-proof AI detector,” Giray said. “AI hyperactively advances, and detection cannot keep pace with it.”

The University of Minnesota declined to comment on the lawsuit, citing federal and state privacy laws. University spokesperson Jake Ricker also declined interviews on behalf of the faculty members named in Yang’s lawsuit.

The University’s Student Conduct Code definition for scholastic dishonesty includes “the unauthorized use of online learning support and testing platforms.”

If the lawsuit settles in his favor, Yang said he would finish his program at the University but shift to a separate track. 

Now, Yang has been traveling in Africa and said he has been interviewing for faculty positions in economics around the world.

“I’m still going to try to survive, try to find a new opportunity,” Yang said.