Prior to December, if students were assigned a five-paragraph essay on writer and women’s rights activist Mary Wollstonecraft, they would need to read her writings, synthesizing that information, and craft an original essay. Since OpenAI’s release of ChatGPT, students need only type the phrase “five-paragraphs on Mary Wollstonecraft” into the ChatGPT website and marvel as the essay’s text scrolls across their computer screen at the rate of 828 words-per-minute.

Designed to respond to text-based queries and generate natural language responses, ChatGPT is part of the broader field of artificial intelligence known as natural language processing, which seeks to teach computers to understand and interpret human language, according to OpenAI’s website. But given the speed at which students can now access a complete draft of an essay on any subject, it's no wonder school districts, colleges, and universities are scrambling to make sense of this powerful chat technology and looking for ways to stop students from using it to plagiarize.

An All-Out Ban

Almost immediately following ChatGPT’s release, school districts from New York to Los Angeles banned the platform. Unless individual schools specifically request access to the platform, New York City Public Schools has restricted the use of ChatGPT, says Jenna Lyle, the district’s deputy press secretary. She called the ban a temporary step in response to a powerful new technology. “We want our students to be exposed to the latest technologies and our schools to be at the forefront of innovation,” says Lyle. The district is currently engaging a set of stakeholders, including educators, students, and the tech community, to determine how it can best leverage the strengths and innovation of these technologies to support students’ futures, she says. For now, educators can request to have the platform allowed on their school’s computers.

Similarly, the Los Angeles Unified School District preemptively blocked access to the ChatGPT model on all district networks and devices to “protect academic honesty while a risk/benefit assessment is conducted,” according to an LA Unified spokesperson. “In the meantime, we will continue to provide robust and relevant training and instruction in digital citizenship and computer science education for all school communities,” the spokesperson added. 

Earning a B- at Best

In higher education, where the stakes are higher and students have access to a broader range of technologies and networks, New York University has not restricted access to ChatGPT. Instead, it issued a three-page document to its professoriate offering recommendations on how they can use ChatGPT, and similar generative writing tools, in their classrooms. The document was developed and issued so NYU students know where their professors stand on the use of these technologies, and faculty doesn’t presume their students know whether ChatGPT is or isn’t allowed.

David Levene, professor of classics at NYU, will not allow his students to use ChatGPT as part of his program and has developed a strategy to prevent his students from using it to plagiarize their assignments. After experimenting with the software for some time, Dr. Levene created exercises that are “not amenable to ChatGPT,” he says. First, he developed a rather basic essay prompt and ran it through ChatGPT to gauge the platform’s performance. The software produced an essay that was at best a B-. “I then added more terms and conditions to (the prompt), things that had to do with my course,” says Dr. Levene. “I eventually came up with complex prompts after several iterations and ChatGPT couldn’t produce a passing essay.”

ChatGPT is trying to please the user by producing what it “thinks” the user wants, says Dr. Levene. “It’s like a bad student, in a way. It will take shortcuts. If I put enough iterations and conditions in the prompt, I can beat it just by the complexity of the requirements,” says Dr. Levene.

Those students searching for the path of least resistance to course completion will use ChatGPT to produce sub-par essays and accept the B- it produces. “It’s really hard to guard against that,” says Dr. Levene. “If all they want to do is pass, they might just about get away with it.”

But for students and professors desiring to author scholarly works, ChatGPT will disappoint every time. Recently, Dr. Levene wrote and presented a paper at Columbia University on the Roman historian Livy. Again, he pushed ChatGPT’s capabilities to the limits by prompting it to produce a scholarly work based on the question his paper attempted to answer. “What it came up with was absolutely disastrous. It actually invented things that were not in the text of Livy and claimed they were in the text of Livy. It gave references to Livy that don’t exist,” says Dr. Levene. Not only was the paper not scholarly; it wasn’t worthy of an undergraduate.

ChatGPT lacks the capability to judge quality because it’s not scanning scholarly texts to answer a question. “It’s looking for other things that other people have said (about a subject) whether good or bad,” says Dr. Levene with a laugh.

Using ChatGPT to Reimagine Education

Despite its shortcomings and its ability to tempt students to plagiarize, professors and teachers still may want to integrate ChatGPT into their curriculum. Those who do have several options. They may want to instruct students to prompt ChatGPT to generate a rough draft on a subject, then have their students present both the draft and their suggestions on improving it. The drawback to this strategy is that students will not learn the process of thinking and writing simultaneously. Nor will they learn to structure an original argument. Another idea that’s been floated is having students prompt ChatGPT to write pro and con essays on the same subject and then debate the virtues of the resulting essays.

Kerry McDonald, senior education fellow at the Foundation for Economic Education and author of Unschooled: Raising Curious, Well-Educated Children Outside the Conventional Classroom, calls ChatGPT “an exciting new AI tool that educators should embrace. It enables a quick, if incomplete, synthesis of ideas and can expand our knowledge base, in much the same way that Google and Wikipedia do,” says McDonald.

McDonald feels that students cheat and plagiarize because “the content is insignificant or irrelevant to them. When children are able to direct their own education, in pursuit of their own goals and while following their own passions, they rarely find cheating to be advantageous. But when students are sitting in compulsory classrooms with a top-down curriculum that may hold little meaning for them, cheating can seem desirable,” says McDonald. “ChatGPT creates classroom conflict that may have more to do with conventional classroom dynamics than technology. Hopefully, ChatGPT will prompt more educators to reimagine education in a more bottom-up, learner-driven, non-coercive way where cheating would serve no purpose for students," says McDonald.

The Game of Cat and Mouse Continues

When students gained access to the internet in the mid-90s, it didn’t take them long to use search engines to find fully-researched papers and pass them off as their original work. But teachers and professors had access to the same search engines their students did, and they used them to find those papers too, thus thwarting plagiarism. ChatGPT, however, could take this game of cat and mouse to a new level. The internet was not as fundamental a challenge to the mechanics of education as ChatGPT could potentially be, assuming programmers will continue to improve it, says Dr. Levene.

Dr. Levene has yet to read a student paper that he suspects was written by ChatGPT, but one of his teaching assistants did. “It showed signs of being written by a machine rather than by a student. It was a really, really, really bad draft. So, he told the student to go back and do something better. That’s as close as I’ve gotten so far,” says Dr. Levene. ChatGPT’s writing style is so distinctive that it’s a kind of give away. “I can recognize ChatGPT’s writing style, and I’d be instantly suspicious of something that came in with that writing style,” says Dr. Levene.