“I was crying during the meeting because it was so unexpected. I thought it was the last thing I’d ever have a discussion on,” a student who was granted anonymity due to potential academic and social repercussions said. After spending weeks researching and writing her history research paper, she was asked to wait behind after class and shown the results of an artificial intelligence (AI) detector on her teacher’s computer.
“It was some AI-detection, like GPTZero,” she said. “On it, you’re either red, which is like super-duper AI, or yellow, or green.”
According to GPTZero, her work was yellow, indicating a mix of human and AI writing, she said.
As AI writing tools become more powerful and accessible, both Upper School (US) teachers and students must adapt to new challenges surrounding academic integrity. The US history teachers follow a rigorous AI-detection process to ensure that students’ research papers are truly their own work. Each research paper is run through multiple AI detectors, including AIDetector.com and GPTZero. However, these tools can yield inaccurate results, suggesting that academically honest writing is instead generated by AI.
The conversation between the student and her teacher helped prove that she had not used AI assistance on her research paper.
“My teacher was like, ‘I’m not saying that I don’t believe you. I just want to hear what happened, your side and your process,’” the student said. “She’s one of those teachers who actually cares about her students and wants them to succeed, and she was able to understand what actually happened.”
The student did not face consequences, and her teacher attributed the AI detector’s results to a false positive.
“My history research paper was mostly based on statistics and numbers because my topic was a medical topic, so my teacher deduced that it was something to do with the quotations,” she said. “She said that those papers did tend to have triggers to the AI detector.”
According to History Department Head Dr. Stacey Spring, students have become increasingly likely to use AI tools to complete steps of the research paper process. As in this case, if a teacher suspects a student of using AI, they will meet with the student to discuss it.
“First, the teacher speaks to the student, goes through what’s there, and why they think it is a problem,” Dr. Spring said. “If there’s vocabulary that the student can’t identify or if they can’t explain their own argument or explain what was in a paragraph, then that is when we get into a situation where students could potentially be asked to redo work and the stated penalties could go into effect.”
While this process may help teachers catch instances of AI use, it can also lead to false accusations. Another student, who also wished to remain anonymous, similarly had her research paper flagged by an AI detection tool.
“I was sitting in the library, and my teacher came up to me and was like, ‘Hey, can we talk?’” she said. “He was like, ‘I don’t mean to accuse you, but this AI detector said that 70% of your paper was AI,’ and I was just in shock.”
The student said the discussion with her teacher felt disappointing and uncomfortable, especially since her Google Doc showed the many hours she had spent writing.
“I started crying because I was so overwhelmed,” she said. “I was a little frustrated because I spent all this time, and he saw that I had spent all that time, and then to be accused of using AI felt very invalidating.”
After examining the revision history, her teacher decided the writing was hers, and she did not face any consequences. However, she said the experience has impacted her relationship with her teacher.
“I think he lost a little bit of trust in me as a student, which feels very unfair to me, because if I didn’t do anything, then why should this affect our relationship?” she said. “I remember before that I could go and ask him about anything, and I wasn’t worried that he would be mad. But after that, I was very hesitant to talk to him.”
US History teacher Jessica Stokes recognizes that AI detection tools are imperfect and is hesitant to punish students when potential AI use is ambiguous.
“It’s sort of like how in The United States’ legal system, the accused are innocent until proven guilty,” she said. “If I can’t be sure, I would rather let somebody who’s guilty go free than injure a relationship with a student by making an incorrect accusation.”
Instead of relying on AI detection tools to combat AI-generated writing, Ms. Stokes focuses on modifying the research paper process.
“This year, I did oral defenses, where students would come in at wherever they were in the process and tell me what they’ve learned since we last met, how that relates to their research question, and then what next steps they need to take,” she said. “I think students became more invested in their topics and in what they wanted it to say.”
This summer, a committee of history teachers will discuss how the research paper process can be developed to address concerns around AI.
“We recognize that AI detectors are a reactive response and have a host of shortcomings,” Ms. Stokes said. “More to the point, though, we believe our job is to develop students’ critical thinking, synthesis and communication skills and to be able to fairly measure the degrees to which students have developed those skills. AI detection may be a component of that, but it is not the primary focus. The puzzle is much bigger than just that.”
