How a Professor Caught Students Cheating with AI
A history professor at Alcorn State University caught students cheating with AI by embedding hidden instructions in an exam. This incident highlights the challenges of AI use in academia.
In the age of artificial intelligence, the educational landscape is rapidly changing, with many students leveraging tools like ChatGPT to enhance their academic performance. However, this trend has also sparked a significant challenge for educators, as demonstrated by a recent incident involving students at Alcorn State University.
Jason Gibson, a history professor at the university, shared his innovative approach to tackling the issue of AI-assisted cheating in a TikTok video. During a midterm exam, he cleverly embedded invisible white text within the exam instructions that directed an AI model to generate nonsensical responses. This tactic enabled him to identify 32 out of 35 students who had resorted to using AI to complete their exams, resulting in their partial failure due to unauthorized assistance.
The Prevalence of AI in Academia
The trend of using AI tools in education is not limited to the United States. In Germany, a survey conducted by the University of Darmstadt found that over 90% of students utilize AI tools like ChatGPT at least occasionally. Among these, more than two-thirds use AI to clarify understanding, while 52% rely on it for text analysis and creation, and nearly half for translations. Similarly, a study by OpenAI revealed that college students in the U.S. are the most frequent users of ChatGPT compared to other demographics. To avoid detection while cheating, some students have begun incorporating deliberate typos or using simplistic prompts to make AI-generated results appear more natural.
The unauthorized use of AI poses significant challenges for educational institutions worldwide. In response, the University of Chicago Law School has implemented a ban on laptops for first-year students, aiming to foster critical thinking skills. The dean noted that some professors had been somewhat complacent by assigning tasks that could easily be completed using tools like ChatGPT. Instead of outright banning AI, the school plans to redesign its curriculum to clearly delineate which skills students must develop independently and where AI usage is acceptable. This new strategy includes laptop-free lectures, supervised exams, and oral defenses of major assignments to ensure students can articulate their responses independently.
A Unique Approach to Detection
Gibson's decision to confront the issue of AI in exams was marked by his unconventional method. In the midterm assignment focused on the Industrial Revolution, he included a hidden instruction that prompted AI tools to produce random digressions about Madagascar. Most of his students copied the assignment into a chatbot, generating their entire response without verifying the coherence of the output. This technique, known as prompt injection, is typically associated with cybercriminals manipulating AI tools.
In another TikTok video, Gibson showcased some of the most egregious examples of AI misuse among his students. He emphasized that his intention was not to humiliate them; rather, he aimed to educate them on how he caught their cheating and offered them the chance to contest their grades, although only two students chose to do so. "It doesn’t bring me satisfaction to see students fail. I don’t intend for you to fail – that would be strange," he stated in his video. He also highlighted a fundamental issue: "We don’t have established best practices for dealing with AI in academia. We’re all just trying to maintain a certain level of academic integrity."
