A new AI tool can write convincing high school essays. Chicago teachers wonder what to do next.

ChatGPT is trained on real writing so it generates human-sounding text. Should schools ban it or embrace it?

Associated
A ChatGPT prompt is shown on a device near a public school in Brooklyn, New York, Thursday, Jan. 5, 2023. New York City school officials started blocking the impressive but controversial writing tool that can generate paragraphs of human-like text. Peter Morgan / Associated Press
Associated
A ChatGPT prompt is shown on a device near a public school in Brooklyn, New York, Thursday, Jan. 5, 2023. New York City school officials started blocking the impressive but controversial writing tool that can generate paragraphs of human-like text. Peter Morgan / Associated Press

A new AI tool can write convincing high school essays. Chicago teachers wonder what to do next.

ChatGPT is trained on real writing so it generates human-sounding text. Should schools ban it or embrace it?

WBEZ brings you fact-based news and information. Sign up for our newsletters to stay up to date on the stories that matter.

A new artificial intelligence tool that can generate a convincing high school essay is dominating teacher conversations in the Chicago area and across the country, leaving teachers and administrators wondering how to respond.

Using the new tool ChatGPT, a high school student can type, for example, “write a five-paragraph essay about the role of guilt in Hamlet” and receive an original finished product in seconds.

This leaves teachers and schools in a predicament — how to get students to turn in their own work in a time when it’s easier than ever to cheat.

The two largest school systems in the U.S., New York and Los Angeles, have already blocked access to ChatGPT on school computers and networks. But Chicago Public Schools, like many school districts locally, has not yet made a decision one way or the other. In a statement to WBEZ, they wrote, “Artificial Intelligence has significant educational value so we need to take a measured approach before considering blocking it.”

ChatGPT, which was launched on Nov. 30, was created by the San Francisco-based research lab OpenAI, which also created the art-generating AI DALL-E. It’s trained on millions of pages of actual writing by humans, and it can imitate different styles of writing. A teacher and a children’s author alike were fooled by the AI’s imitations of fairy tales written by fourth graders. Researchers at Northwestern and the University of Chicago found that its imitations of research paper abstracts even fooled scientists about a third of the time.

Since it generates writing that sounds human, ChatGPT makes it easier than ever for students to turn in an essay they didn’t actually write. Plagiarism-checking software like TurnItIn compares writing to a database of existing content, but ChatGPT creates something new.

The most accurate way to catch AI-written text so far turns out to be another AI. A 22-year-old college senior at Princeton named Edward Tian created GPTZero, which analyzes characteristics like word predictability and variation of sentence lengths to guess whether the writer was a human or a machine. Still, it’s only a guess.

ChatGPT in the classroom

Among local teachers and administrators, views on this new tool are across the map. Some are labeling any use of AI cheating, while others are looking for ways to embrace it. Some think concerns about cheating with this imperfect tool are somewhat overblown.

As school resumed this month, Keir Rogers, principal of Hersey High School in suburban Arlington Heights, wrote in a note to parents that students caught using AI would be treated as cheaters. But his district, as well as other large schools in Harvey and Park Ridge, say they have no set policy on ChatGPT yet.

Phil Cantor, a science teacher at North-Grand High School in Humboldt Park, is skeptical of trying to ban AI. Instead, he thinks teachers can find ways to give AI-proof homework. Assignments that ask students to make their own connections to a topic are not only harder to fake, but also matter more to students.

“If we’re giving assignments that kids don’t care about, why should they not just use AI?” Cantor said. “That’s another way to AI-proof things — make it an assignment the student actually wants to do.”

Cantor also thinks a first draft from ChatGPT could eliminate hurdles for English language learners and students with learning disabilities, who may have a hard time communicating what they know. He says it would be a mistake for CPS to block access to it.

Katherine Czajka, who teaches history at Lincoln Park High School, also says she’s not overly worried about ChatGPT for now. In her class, her students’ essays are handwritten in class. She’s played with the software and says it can, when prompted, generate good examples of historical thinking and even a decent study guide for the AP US History exam. Still, she thinks ChatGPT’s potential to make factual errors means it’s not quite ready for use in classrooms.

“This isn’t something I’d be assigning to students anytime soon,” she Czajka.

Another educator sees how ChatGPT makes mistakes humans wouldn’t.

“A colleague of mine asked ChatGPT if a pound of beef weighs more than a pound of pork, and ChatGPT said, ‘Yes, the pound of beef weighs more,’ ” said Barbara Di Eugenio, a professor of computer science at the University of Illinois Chicago. Di Eugenio thinks some kinds of assignments, like those that require a lot of citations from primary sources, could be particularly AI-proof.

But she believes ChatGPT could become part of the essay-writing process, as long as students are taught to use it responsibly. She says she would allow her students to use ChatGPT to generate a first draft, “but then make sure everything is correct, and rewrite it in your own style — in your own voice.”

Brady Gunnink, who teaches English at Jones College Prep in the South Loop, sees the potential to use ChatGPT in his own teaching, including a lesson on rhetoric this month. His students will analyze Martin Luther King Jr.’s letter from the Birmingham Jail and try to imitate King’s writing style. Then, they’ll have ChatGPT complete a similar assignment to see which parts of King’s style the AI notices. “What does [ChatGPT] seem to think Martin Luther King Jr. sounds like?” he said.

Gunnink thinks AI is here to stay in schools.

“Students have already told me that they’re using ChatGPT,” he said. He thinks trying to punish students for using AI will simply become an arms race of better and better chatbots. “I’m interested in a more collaborative model,” Gunnink added. “This is something that got dropped on all of us, and I’m interested in working with my students to examine it together.”

For the time being, though, teachers and students can’t use ChatGPT, even if their districts haven’t blocked it. Its servers are currently at capacity, meaning most people will have to wait to try it until interest dies down — or its parent company, OpenAI, grows.

Char Daston covers education for WBEZ. Follow him @behindthissky and @WBEZeducation

This story is updated to correct the spelling of teacher Brady Gunnink.