Daniel de Visé and Lexi Lonas
In four short months, the GPT family of artificial intelligence chatbots have upended higher education like nothing since the arrival of Wi-Fi connections in classrooms.
ChatGPT and its smarter, younger cousin, GPT-4, can create a realistic facsimile of a college term paper on command, or populate the answers to a midterm. At the start of the 2022-23 academic year, few professors had heard of it. They are learning fast.
“I think this is the greatest creative disruptor to education and instruction in a generation,” said Sarah Eaton, an associate professor of education at the University of Calgary who studies AI.
The impact of this quickly developing technology has sparked varying concerns across colleges and fields of study due to its implications for academic honesty and learning.
Not everyone sees this technology as an earth-shattering phenomenon, however. Some are excited about the implications it can have on learning.
“There just hasn’t been panic here on campus. In fact, the university is absolutely a wonderful place to consider all the implications both good and bad, and challenges and new questions raised by any kind of new technology, because we have people who are going to think about the problems from so many different angles and orientations,” said Jenny Frederick, executive director of the Yale Poorvu Center for Teaching and Learning and associate provost for Academic Initiatives.
Across universities, professors have been looking into ways to engage students so cheating with ChatGPT is not as attractive, such as making assignments more personalized to students’ interests and requiring students to complete brainstorming assignments and essay drafts instead of just one final paper.
Frederick conceded that at Yale, an Ivy League school with many resources at its disposal, it could be easier for the college to embrace the technology without fear.
At small schools, such as Texas Woman’s University, ChatGPT has provoked more hesitancy.
“I think the majority, the sentiment from the majority of my sort of academic network is one of sort of anxiety and fear,” said Daniel Ernst, associate professor of English at the school.
Texas Woman’s University held a workshop for faculty at the end of January regarding ChatGPT. Genevieve West, chair of the school’s Department of Language, Culture & Gender Studies, said she saw a generational divide at the event: Younger professors were more excited about the technology, while older professors voiced more concerns.
Since launching in late November, ChatGPT has already garnered 100 million users. Its use quickly surged on campuses in that time. In an informal and anonymous January poll, 17 percent of Stanford students acknowledged using ChatGPT in their fall finals. Most said they used AI only to brainstorm, outline and spitball. A tiny share said they submitted ChatGPT work as their own.
The rapid popularization of the new technology has sent both small and large schools scrambling to develop guidelines on how to approach it.
Stephanie Frank, an associate professor of instruction in religion and the humanities at Columbia College in Chicago, spent the last few hectic weeks on a task force to decide how the faculty should handle chatbots.
“The point of this was to get something out before midterms, which were this week for us,” she said. The task force issued a memo to faculty on Wednesday.
Columbia organized the work group after a professor caught a student “pretty flagrantly” using ChatGPT for answers on a quiz, Frank said. The professor canceled the next quiz, asking students instead to submit handwritten class notes. The same student provided handwritten notes that appeared to have been copied from ChatGPT.
Rather than set campus-wide rules, the Columbia College task force urged professors to make individual decisions about whether and when students should be allowed — or encouraged, or even assigned — to use AI.
Youngmoo Kim, a computer scientist, sits on a similar committee at Drexel University studying chatbots. The panel aims to issue guidance to the school by the end of March.
“We’re looking to put out guidelines for all of our faculty,” he said. “Not commandments.”
Kim expects the guidelines to be loose and broad because AI technology is evolving so quickly.
“The rules have changed in AI, not just in the past year, but within the past week,” he said. “If we put out some very strict guidelines right now, it will look silly. It will look silly probably within a few weeks.”
Not all schools have created official guidelines or even attempted to do so, and some have decided at this point that new rules are not necessary.
Justin Shaddock, the chair of the Honor and Discipline Committee and an associate professor of philosophy at Williams College, says his school handles suspected cases of cheating with a student panel and that will not change.
“We have this committee of students that are supposed to be the ones who sort of act as the judges to decide whether a given you know allegation of cheating really is cheating,” Shaddock said.
Policing plagiarism before ChatGPT was relatively easy: Plagiarism checkers can match a passage in a student essay to identical wording on Wikipedia, for example, catching the offender red-handed.
But ChatGPT doesn’t copy old words: It creates new ones. That complicates the task of catching cheats.
“It’s a lot harder with ChatGPT because it produces different answers every time. So there, it’s sort of more like, the professor has to put their question or prompt or whatever in the ChatGPT, maybe do it a couple times, and then try to show that there’s more similarity between the suspected essay and ChatGPT responses than there is between the suspected essay and sort of like a range of other student responses,” Shaddock added.
Spotting when a student is using AI-generated text in their papers might not be as easy as many professors think, especially in bigger classes.
One Canadian study, not yet published, found that “two-thirds of professors could not correctly identify texts written by AI,” Eaton said.
“When you get two-thirds of university professors failing a test,” she said, “we’re in a bit of a pickle.”
Professors are more likely to spot a chatbot cheat in a small seminar, a format that allows the instructor to engage with students about their work and grow familiar with each student’s style.
Despite all the concerns, many realize there is no stopping the evolution of technology and are looking at the bright side after the initial shockwave.
Laura Dumin, an English professor at the University of Central Oklahoma, administers a Facebook group of 2,000 faculty members to discuss positive uses for AI.
“I know that we are exhausted from COVID. We had to pivot once, and now we’re being asked to pivot very quickly again,” she said.
“Especially back in January and February, people were like, ‘We’re going to catch all of the cheaters,’” Dumin said. “And my thought was, You’re going to waste all of your time and exhaust yourselves.”
Instead, Dumin encourages professors to make peace with the technology and to find productive ways to use AI in the classroom.
Dumin’s own students now submit papers for three rounds of review: from their peers, from the professor and from the chatbot. “And it will give you feedback, tell you what it thinks about what you have written,” she said.