4 September 14 - 20, 2023 dallasobserver.com DALLAS OBSERVER Classified | MusiC | dish | Culture | unfair Park | Contents CheatGPT As artificial intelligence gets smarter, educators ask whether it will make students dumber. BY SIMONE CARTER M aya Bodnick, an undergraduate student at Harvard, received a report card over the summer that would make most parents proud. She’d brought back mainly A’s and B’s from the prestigious school: a solid 3.57 GPA. But Bodnick didn’t complete any of the assignments herself. The artificial intelli- gence bot ChatGPT did. Don’t worry: Bodnick was conducting an experiment, one that’s gone viral in the weeks since she published the results. The Harvard student had asked seven professors and teaching assistants to score her essays as they usually would, telling them — in an ef- fort to reduce grading bias — that they’d ei- ther been penned by her or by ChatGPT. In reality, the AI had written each re- sponse. “Right now, ChatGPT enables students to pass college classes — and eventually, it’ll help them excel — without learning, devel- oping critical thinking skills, or working hard,” wrote Bodnick in a piece for the Slow Boring Substack, where she also interns. “The tool risks intellectually impoverishing the next generation of Americans.” ChatGPT is a large language model chat- bot that allows users to request detailed re- sponses to specific questions and prompts — anything from, “What is the first letter of the supporting actor’s name in There Will Be Blood?” to, “Write a 500-word op-ed about recent developments in quantum physics.” The seemingly omniscient AI spits out the desired product in seconds. The chatbot isn’t even a year old but it has already evolved to become far more ad- vanced. A product of the latest era of the informa- tion age, ChatGPT, made by the company OpenAI, is evoking both wonder and dread among educators. It’s turned into something of a Rorschach test for teachers in Texas and across the country. Where proponents see progress and an effective tool for learning, naysayers warn that ChatGPT will atrophy critical thinking. It arrived on the heels of widespread pan- demic-induced learning loss. Now, critics are warning of a budding “cheating epi- demic.” Certain districts nationwide, including schools in Los Angeles and New York City, have chosen to ban the technology. North Texas districts are varied in their reactions. Fort Worth ISD indicated earlier this year that it had no plans to bar AI. In July, Plano ISD’s digital learning team held a vir- tual class on how ChatGPT can help teach- ers and students alike. School officials in Denton, meanwhile, have warned that ask- ing the technology to assist with homework amounts to plagiarism. Districts further south, like Austin and Eanes ISDs, also opted for bans. Dallas ISD told the Observer via email that it has “decided to embrace responsible use” of ChatGPT. Chief Academic Officer Shannon Trejo said the district would de- vise ethical application guidelines for stu- dents and staff to combat concerns about plagiarism. “Professional learning for teachers and staff, as well as training op- portunities for students will be incorpo- rated into a Digital Citizenship focus,” she continued. Rather than shunning the tech, some ed- ucators are working with ChatGPT as a teaching aid. It can compose handouts, syl- labi, lesson plans and grading rubrics, leav- ing teachers more time for higher-level planning. Others aren’t nearly as optimistic about its potential. Rena Honea, president of Dallas’ Alli- ance/AFT teachers union, fears that Chat- GPT allows students to avoid putting in the hard work. “We need them to be able to be strong thinkers, independent thinkers, criti- cal thinkers — take what they’re seeing and ask questions,” she said. “Don’t just take it as, ‘That’s the gospel. That’s the word.’” The way Honea sees it, kids need to ex- plore being creative and develop their own ideas. Otherwise, they’ll become too depen- dent on others. “Because when those people are gone or when those resources are gone,” she added, “what will they be left with?” T racy Everbach, a journalism professor at the University of North Texas, viv- idly remembers coming across an alarming social media post late last year. The video was from a former student who’d asked ChatGPT to respond to a specific prompt — something related to political and economic systems — and to cite sources only from within a certain time frame. The AI pumped out the essay in a matter of seconds. “When I saw that,” Everbach said, “I thought, ‘Oh, no. We are doomed.’” The clip alone seemed concerning enough to Everbach, who is also an aca- demic integrity officer at UNT. Then she read the post’s replies. “The responses to it scared me, too,” she said, “because students … were asking for tips on how they can use this without it be- ing found out.” There is evidence indicating that stu- dents are flocking to ChatGPT-like tech to help them cheat. In a March exclusive for The Daily Mail, one expert noted that “well over half of students are likely using AI tools” to do just that, but that the real num- ber could be even higher. It’s possible that some students may con- sult one AI tool to write their assignment and another to reword it, making detection that much harder. Bodnick’s Harvard report card isn’t the only proof of ChatGPT’s proficiency. The bot has passed difficult graduate-level ex- ams, including the bar. But ChatGPT isn’t always reliable. Some- times, large language models are known to “drift,” meaning that they’ve swerved from the prompt’s initial parameters in unantici- pated ways. One recent study from Stanford University and the University of California at Berkeley, for instance, suggests that the latest version got significantly worse at solv- ing certain math problems over a three- month span. ChatGPT isn’t exactly dependable when it comes to using legitimate references, ei- ther. It has at times concocted scholarly cita- tions thanks to a phenomenon dubbed “hallucination.” As an academic integrity officer, Ever- bach has encouraged professors to examine students’ sources: Do they seem credible? There are detectors available online, she added, but they, too, are flawed. UNT has purchased its own form of detection that’s fairly accurate. Still, Everbach urges faculty to read submissions closely if there’s a ques- tion of authenticity. Sometimes the chatbot will spew out nonsense or repetitive infor- mation, which students then turn in without a second glance. Cases of ChatGPT use soared during the spring semester in many colleges. But trying to confirm that a suspect assignment was aided by AI can eat up time, Everbach said: “It does take labor on the part of the faculty member to be able to figure these things out.” Rudi Thompson, associate vice president of digital strategy and innovation at | UNFAIR PARK | Alicia Claytor >> p6 Journalism professor Tracy Everbach fears how students will keep integrity in their reporting with new AI tools.
