Not long after University of NH students returned for the spring semester, Prof. Nora Draper led an unconventional exercise with her students using ChatGPT.
Draper, an associate professor of communication, asked the software program to write a rap about UNH campus life. Not just any rap, but one written in the style of the Beastie Boys, a group that formed 25 years before some of her students were born. The results were impressive. The software dutifully created a passable tune that incorporated lyrics about “The Whit” (Whittemore Center) and “HoCo” (Holloway Commons).
As the class discussed the implications of the software, particularly around the potential for misinformation, she asked ChatGPT to write her professional biography. Much of this information could be found on her UNH faculty page. Even so, the program struggled. It convinced itself she was a health communication scholar (she isn’t). And when asked to find citations of her scholarly work, the program just made them up entirely, complete with fictitious co-authors.
These exercises showed the potential and power of artificial intelligence (AI) chatbots like ChatGPT, along with some glaring limitations.
“What ChatGPT is good at is style and form,” Draper says. “It is good at mimicking the generic and stylistic elements of whatever you are asking it to do. If you ask it to write an advertisement or an essay or a haiku, it can mimic the style and form pretty well.
“And when you look at this profile it wrote of me, it was very instructive of the limits of this kind of technology,” she continues, noting that it can create things that sound correct but, in many cases, aren’t. An attorney found that out the hard way in a well-publicized case where the attorney used ChatGPT to craft a motion only to have the court discover the cases cited in the motion were entirely made up by the AI program.
For better or worse, AI has been a part of our lives for a long time in myriad ways. That includes your Alexa assistant and Siri on your iPhone. Technology that uses your face to unlock your phone, finishes your thoughts in an email or text message and answers questions about an online clothing order are all based on AI.
Even so, AI technology burst into the collective consciousness last fall with the release of ChatGPT. The powerful software can do many things a human can. Based on a series of prompts input by a human, the software can scour the internet to quickly come up with coherent answers to questions and requests. It can discern context, remember prior inquiries and learn from its mistakes. It can also engage in conversation.
This is all possible through large language models. According to NVIDIA, the advanced chip maker whose share price has soared amidst the AI tech boom, these are “deep learning algorithms that can recognize, translate, predict and generate content using very large datasets.”
In other words, these AI programs can quickly review vast amounts of data on the internet, understand that material and then use that information to answer the question or perform a task requested by a human.
Universities Respond
Few colleges anywhere have embraced AI like Southern NH University. The college has about 4,000 students on campus in Manchester and another 175,000 around the country who take courses online. SNHU President Paul LeBlanc says its staff and faculty have already engaged with AI in many ways and are looking to do even more. The university hired preeminent AI researcher George Siemens to explore ways to integrate AI into all levels of university operations.
“We’ll be working with him to take a clean-sheet of paper approach to the question of, ‘What does it look like to build a university that is AI-powered and human centered?’” LeBlanc says.
At its core, ChatGPT and similar programs under development by other tech giants represent little more than a technological advance, not unlike the light bulb, printing press or the combustion engine. But in practice, the technology can and will have important ramifications on college campuses in NH and around
the world.
“It was really toward the end of the fall semester [that ChatGPT came out], and … in my conversations with students, it spread really quickly throughout the student population,” says Michael Blackman, the dean of students at UNH.
UNH administrators and faculty responded quickly too, in part by trying to understand the potential effects of ChatGPT in the classroom. The university looked at its existing policies around plagiarism and found they effectively covered automated writing tools such as ChatGPT. The university reiterated these points in an email to the UNH community last January as the spring semester got started.
Any large university like UNH encounters reports of plagiarism throughout the year, and it has a process for addressing those as they arise. Last spring was a little different in the types of plagiarism cases that emerged. The number involving ChatGPT grew as the semester wore on, Blackman confirms.
With the summer break now fully under way, UNH officials are revising policies involving cheating and automated writing tools like ChatGPT to make the expectations transparent and clear. The new rules will be finished in time for the fall semester.
“At the heart of it, plagiarism is plagiarism, whether that is borrowing from someone else’s paper, Wikipedia or ChatGPT. Our policy has always said you can’t represent ideas that aren’t yours as yours without citing them,” Blackman says. “In the fall, we will explicitly state … in our academic honesty policy to make it clear this behavior will be prohibited.”
He adds that faculty are provided training in ways to detect the use of AI and access to software programs that can scan assignments to look for possible help from AI programs.
Professors, too, are using the arrival of ChatGPT and similar programs to revisit the types of assignments they give students. At SNHU, LeBlanc says instructors are focusing where possible on project-based learning. They are asking students to show what they can do with what they know, rather than just recite something from memory.
Draper has taken a similar approach. She has looked closely at the assignments she’ll give students this fall to ensure they cannot be completed exclusively using ChatGPT. “If something can be pretty easily done by ChatGPT,” she says, “then we need to rethink the types of skills we expect from our students.” n
Current Issue - May 2024