When something goes wrong in a classroom that no one is allowed to identify, a certain kind of silence descends. A student in the back row is silently aware of something that the teacher in the front of the room might not want her to know as she stares at an essay that has been returned to her with red marks all over it. Anyone who has ever sat through a lecture wondering if the authority figure at the chalkboard had actually done the reading will recognize this tension, which is currently manifesting in schools in a novel and genuinely peculiar way.
The panic in schools quickly spread after OpenAI released ChatGPT in November 2022. Instructors were concerned that students would use the tool to falsify essays. To stay ahead of the curve, some professors completely eliminated homework assignments in favor of oral exams and handwritten assignments. Emergency meetings were called by administrators. Emails were sent by parents. Nearly all of the discussion focused on students as the threat, teachers as the victims, and technology as the issue that needed to be resolved. The suspicion would eventually turn against them, something that no one seemed to foresee.
Perhaps something more private and intimate than a policy meeting was what made all the difference. After running a piece of writing through an AI detection tool, a student—a real adolescent, sitting somewhere with a laptop—didn’t receive the expected outcome. It wasn’t her text that was flagged. The person who had been grading her work throughout the semester owned it. It’s difficult to ignore how extremely uncomfortable that must have been for everyone involved when observing that moment from the outside. Suddenly, a child with a tool that the adults hadn’t quite figured out was holding something that no one wanted her to.
Over the winter break, 22-year-old Princeton computer science and journalism student Edward Tian created GPTZero after observing AI-generated writing spreading in strange ways. His app gauges a piece of writing’s complexity as well as its sense of variation from sentence to sentence. Human writing has a tendency to flow erratically, with pauses and bursts as well as rhythmic changes that result from a real mind trying to figure something out. Prose written by AI tends to flatten. The longer you read it, the more bizarre the steady, almost pleasant clip becomes. Tian made it very clear that he wasn’t against AI, but he did want people to stop acting as though it didn’t exist.
More than 30,000 people had used GPTZero within a week of its launch. The volume caused it to crash. Tian’s inbox was inundated with correspondence from principals and teachers across the globe.
The fact that the tool operated in all directions was something that no one, probably not even Tian, had fully considered. A permission slip regarding who can be detected is not included with detection. When using GPTZero for an essay, a student is not limited to running her own work. There’s a feeling that the symmetry of it, rather than the cheating, is what the system was not designed to deal with.
One teacher, one student, or even one app isn’t really the focus of the deeper question. It’s about what happens when a new technology emerges and no one can agree on regulations quickly enough. Almost instinctively, schools in Seattle and New York City banned ChatGPT; Tian felt this was a defensive move that completely missed the mark. “It doesn’t make sense that we go into that future blindly,” he said to NPR. He was correct. As it happens, the blindness was never confined to one side of the desk.
How seriously schools choose to take the symmetry will likely determine whether this moment becomes a reckoning or merely an awkward footnote. Institutions’ interest in having that discussion is still unknown. However, the students are aware of how to use the tools that are currently available. It’s not changing.
