- Execs at anti-cheating company Turnitin said they've figured out how to tell if a student is using ChatGPT for assignments.
- Turnitin said they were able to detect AI writing because it's "extremely average."
- This comes as the NYC Department of Education banned ChatGPT, saying it does not build critical thinking skills.
Since its launch in late November, ChatGPT seemed like a golden ticket for students looking to get out of take-home assignments.
The AI chatbot, developed by OpenAI, can quickly generate human-sounding answers to homework prompts, like "write a 200-word essay about the American Revolution in the style of a high school student," for example – and it's getting increasingly difficult to detect.
However, executives at anti-cheating software maker Turnitin say they've cracked the code.
The company, which works with thousands of universities and high schools to help teachers identify plagiarism, said it plans to roll out a service this year that can accurately tell whether ChatGPT has done a student's assignment for them.
"These models are trained at the sum of human knowledge, so they write extremely average. They are mad-lib machines that pick the most probable word in the most probable place. Humans are idiosyncratic… no person is actually exactly average," Turnitin's vice president of artificial intelligence, Eric Wang, said.
Many teachers have expressed concern that the bot is likely being used to help write students' assignments, according to Turnitin. One college professor recently told the New York Post he felt "abject terror" upon realizing a student had used it to write an essay.
"We're hearing from educators that they want different tools at their disposal, whether straight detection or other features to validate," Turnitin's chief product officer, Annie Chechitelli, said about ChatGPT.
Last week, the New York City Department of Education became the first school district to ban the use of the chatbot on its networks and devices.
"While the tool may be able to provide quick and easy answers to questions, it does not build critical-thinking and problem-solving skills, which are essential for academic and lifelong success," a department spokesperson said.
The ChatGPT challenge
Even Turnitin said it wouldn't be able to detect ChatGPT with 100% accuracy.
"I'm sure our first beta release will not be perfect, but it will be enough that we can start getting signal and learning from the instructors how to tweak it," Chechitelli said.
Zhou Yu, an assistant professor of computer science and Columbia University, said that for Turnitin's AI detector to be successful, it would have to constantly be updated and changed to keep pace with the rate the AI bot is learning. The most statistically likely word that AI generates today might be different than next year as OpenAI's product improves.
Nevertheless, Yu said she isn't as concerned about ChatGPT as others in the education world. Instead, she believes it can advance education when appropriately used as a tool: she is currently studying its effectiveness in teaching new languages.
"I don't think we should be afraid of it. It's the natural trend of progress, though, of course, there will be accidents," she said.
However, executives at Turnitin said they were concerned that OpenAI's product could revolutionize education altogether.
"Helping educators really grapple with how this technology is going to have to shift teaching, learning, writing, knowledge creation, and the concept of originality – that is really hard. Norms are going to have to shift," Wang said.