The battle between AI and academic integrity has reached a fever pitch on college campuses across the United States. With the rapid adoption of generative AI, a new arms race has emerged, pitting students against professors and AI detection tools.
Anxiety levels are high as professors fear students may cheat their way through college, leading to the widespread use of AI detectors to scan student papers. However, these detectors have faced criticism for their unreliability and tendency to flag non-native English speakers, causing emotional distress and legal battles for students who feel wrongfully accused.
In response, a new breed of AI tools called "humanizers" has emerged. These tools aim to scan essays and suggest changes to make the text appear less AI-generated. Some are free, while others require a monthly subscription. Students use these tools to avoid detection or to ensure they aren't falsely accused of AI usage.
As AI chatbots continue to advance, companies like Turnitin and GPTZero have upgraded their detection software to catch writing that has been humanized. They've also introduced applications for students to track their writing history, but humanizers can bypass these measures by allowing users to copy and paste text.
"Students are now trying to prove their humanity, even if they've never used AI," says Erin Ramirez, an associate professor of education. "We're in an endless spiral."
The heightened anxiety about cheating has propelled a competition between AI detectors and writing assistance programs. It's a never-ending cycle, impacting even those who don't want to use AI and those who must police its use.
"If we write properly, we're accused of being AI. It's ridiculous," says Aldan Creo, a graduate student studying AI detection.
The stress from these accusations has driven some students to drop out. Brittany Carr, a long-distance student at Liberty University, received failing grades on assignments flagged by an AI detector. Despite providing evidence, she was still required to take a "writing with integrity" class and sign an apology statement.
"It's a weird feeling. The school is using AI to tell us we're using AI," Carr said.
To avoid further accusations, Carr ran her material through Grammarly's AI detector, changing sections until it confirmed human authorship.
"I'm writing just to avoid flagging AI detectors. It feels like my writing lacks insight," she said.
After the semester, Carr decided to leave Liberty, unsure of her next steps.
At the heart of these conflicts is a disagreement over what constitutes excessive AI use in homework. Independent analyses of AI detectors show varying accuracy, with some studies finding high false positive rates and others suggesting near-perfect accuracy.
"The understanding of what's acceptable is fragmented, and it's becoming more so with the growing number of tools," says Edward Tian, CEO of GPTZero.
AI detection probability scores are often misinterpreted, leading to false positives. Users may not understand that detectors flag text likely generated by AI or that text must be a certain length for effective evaluation.
Turnitin and GPTZero emphasize that research showing flaws in their detectors is outdated due to rapid model evolution and software updates.
"The better the writer you are, the more AI thinks you're AI," Ramirez says.
All parties agree that professors should have conversations with students after software flags their work to prevent false accusations. However, this takes time, especially with large class sizes.
"It creates extra, uncompensated labor for us," says Morgan Sanchez, an assistant professor of sociology.
The humanizer industry is booming, with Turnitin viewing them as a threat to academic integrity. The company issued a software update to detect text modified by humanizers.
"The most important question is not about detection but about where the line is," says Annie Chechitelli, Turnitin's chief product officer.
GPTZero also advises against using its detector to punish students.
Ramirez, who studies AI use in K-12 settings, says anyone relying on a detector hasn't put their own work through it.
"I put my own papers into AI detectors, and it flags me at 98% every time, even though I didn't use AI," she says.
Some students believe universities should stop using AI detectors due to false positives. An online petition calling on the University at Buffalo to drop the software received over 1,500 signatures.
Kelsey Auman, a graduate, started the petition after fighting to prove she didn't use AI. She now runs her papers through multiple detectors before submission, creating more anxiety when they incorrectly flag her writing.
"It's like going down the rabbit hole. I'm making myself crazy," she says.
Tricia Bertram Gallant, director of academic integrity at UC San Diego, advises faculty to realize the difficulty of ensuring students don't use AI in unsupervised assessments.
"Don't bother trying to ban AI or prove its use. You'll spend more time doing that," she says.
Bertram Gallant wishes more pressure would be put on the government to regulate AI and the academic cheating industry, and on tech companies to prevent students from using their products to cheat.
"We keep focusing on what academic institutions need to do to fix problems they didn't create," she concludes.
