AI Detection Tools: Can They be Trusted?

Sixty-eight percent of teachers in the U.S. regularly use AI detection tools on their students’ work, according to a survey done by the Center for Democracy and Technology. In recent times, AI has been a concern in schools. Due to this, AI detection schools have quickly risen in popularity. They can help alert teachers to students engaging in academic dishonesty via AI, which has become an issue as of late. But, the issue arises when AI detection tools are wrong. Can we really trust AI detectors?

I tested out three AI detectors by inputting the prologue of Romeo and Juliet, which was written around 360 years prior to the initial invention of AI, and about 430 years prior to AI such as ChatGPT being released and becoming easy for widespread personal use. ZeroGPT, one of the most popular AI detectors, flagged the text as AI written all three times that I checked it. I also tested various AI detectors on unedited texts written by ChatGPT, the AI detectors I used did fairly well at detecting the use of AI. However, when I just asked ChatGPT to sound more human, the AI detectors suddenly started detecting far lower percentages of AI usage, despite the fact that it was still entirely written by AI and unedited. While the AI detectors were sometimes right, they still had flaws and false positives just in my minor amount of testing. Many teachers believe AI detectors to be flawless and trust them 100%. This can be due to the teachers not fully understanding how this technology works. So, how does it work?

AI detectors are simply AI meant to determine if a piece of text sounds like AI. They’re not some sort of magical tool. We all have surely heard stories about AI just making up fake sources and studies that never happened; so why do we trust AI detectors so much? Even the people behind large AI detection tools admit that it’s flawed. Turnitin, an AI detection software used by many schools, said that “[their] sentence-level false positive rate is around 4%.” When the AI detectors I used properly detected AI, it was all pieces of text that I think anyone could look at and determine that it was AI. Why? Because no one talks like that. Students using AI know this, and so they often edit their AI-written schoolwork to slightly reword things to sound human.

Students have mixed impressions of the usefulness of AI detection software. Some appreciate the possible useful capability of these tools. Senior Finn Graziano said he hasn’t used AI detection software, but it could be helpful. “I know a lot of people really don’t like the idea of AI, and teachers could easily use it to pick out any cheaters,” Graziano said.

Other students were more critical of the AI detection tools, such as sophomore Ginny Henry. “I like the idea of using AI as a tool, but it shouldn’t be something we depend on all the time because it’s literally a robot,” Henry said. As for AI detectors, “AI detectors are ‘whatever,’ it gives teachers more time I guess. But, they are still stupid. I have mixed feelings about AI.”

Patton Parkinson, a junior, pointed out the potential for AI detection software to prompt a false accusation of academic fraud being placed on a student’s record. ”It sucks, not gonna lie,” Parkinson said. “Because you can type out an essay and it’ll say it’s AI. Then you get a zero, like what? And then if someone uses AI it says it’s 100% human-written.”