The latest version of ChatGPT has only been around for a few months. But Aaron Romoslawski, the assistant principal at a Michigan high school, has already seen a handful of students trying to pass off writing produced by the artificial-intelligence-powered tool as their own work.
The signs are almost always obvious, Romoslawski said. Typically, a student will have been turning in work of a certain quality throughout the year, and then “suddenly, we’re seeing these much higher quality assignments pop up out of nowhere,” he said.
Romoslawski and his colleagues don’t start with a punitive response, however. “We see it as an opportunity to have a conversation.”
Those “don’t let the robot do your homework” talks are becoming all too common in schools these days. More than a quarter of K-12 teachers have , according to a recent survey by study.com, an online learning platform.
What’s the best way for educators to handle this high-tech form of plagiarism? Here are six tips drawn from educators and experts, including a , two education technology nonprofits focused on building students’ literacy skills.
1. Make your expectations very clear
69ý need to know what exactly constitutes cheating, whether AI tools are involved or not.
“Every school or district needs to put stakes in the ground [on a] policy around academic dishonesty, and what that means specifically,” said Michelle Brown, the founder and CEO of CommonLit. 69ý can decide how much or how little students can rely on AI to make cosmetic changes or do research, she said, and should make that clear to students. She recommended “the heart of the policy [be] about allowing students to do intellectually rigorous work.”
2. Talk to students about AI in general and ChatGPT in particular
If it appears a student may have passed off ChatGPT’s work as their own, sit down with them one on one, CommonLit and Quill recommend. Then talk about the tool and AI in general. Questions could include: Have you heard of ChatGPT? What are other students saying about it? What do you think it should be used for? Discuss the promises—and potential pitfalls—of artificial intelligence.
“One of the big concerns right now is that teachers want to encourage curiosity about AI,” said Peter Gault, Quill’s founder and executive director. Strict discipline at this point “doesn’t sit right with teachers where there’s a lot of natural curiosity here.”
Romoslawski uses that approach. And so far, he hasn’t had a student try to use ChatGPT on an assignment twice. “We’ve gotten to the point where it’s a conversation and students are redoing the assignment in their own words,” he said.
3. If students use ChatGPT for an assignment, they must attribute what material they used from it
If students are allowed to use ChatGPT or another AI tool for research or other help, let them know how and why they should credit that information, Brown said. Since users can’t link back to a ChatGPT response, she suggested students share the prompt they used to generate the information in their citation.
When Romoslawski and his colleagues suspect a student used ChatGPT to complete an assignment when they weren’t supposed to, he also brings up citation, in part as a way into the conversation.
“We ask the students ‘did you use any resources that you don’t cite?’” he said. “And often, the student says ‘yes.’ And so, then it creates a conversation about how to properly cite and attribute and why we do that.”
4. Ask students directly if they used ChatGPT
Don’t beat around the bush if you suspect a student may have used AI to cheat. Ask them in a very straightforward way if they did, CommonLit and Quill say.
If students say “yes,” Romoslawski likes to get a sense of why. “More often than not, the student was just struggling on the assignment. They had a roadblock. They didn’t know what to do,” he said. “They were crunched for time, because we’re a high-achieving high school and our students are taking some pretty rigorous courses. This was their third homework assignment of the night and they just wanted to get through it.”
If the student says “no,” but you still suspect them of cheating, ask if they got other help with the assignment. If they still say “no,” explain your concerns by pointing out differences between the work they turned in and their previous writing, CommonLit and Quill suggest.
5. Don’t rely on ChatGPT detectors alone to determine if there was cheating
There are a number of tools—including one from OpenAI, ChatGPT’s developer—that purport to be able to distinguish an AI-crafted story or essay from one written by a human. But most of these detectors don’t publish their accuracy rates. And those that do are ineffective about 10 to 20 percent of the time.
“You can’t fully rely on that as the sole proof of academic dishonesty,” Brown said.
6. Make it clear why learning to write on your own is important
69ý in general, and particularly students who take advantage of AI to cheat, need to understand what they are missing out on when they take a technology-enabled shortcut. Educators should try to persuade students that learning to write on their own will help them reason and think, or be critical to future job success, Gault said.
But others will need a more immediate incentive. The strongest argument one teacher came up with, according to Quill’s Gault? Tell students that learning to write will make them more persuasive, and therefore, “you can convince your parents to do what you want.”