By SHANNON O. WELLS
Mary Boyd, a junior political science and communication student at Pitt, said the struggles of a Guatemalan friend’s mother to communicate in English got her thinking about the challenges of teaching and learning multilingual writing.
Entering the quick rise of generative artificial intelligence, or GenAI.
“What I found is that multilingual students largely used Gen AI for improving their lexical and syntactical complexity — quick translations for things,” she said during an April 22 forum about GenAI held on the Pitt campus.
“I actually found that ChatGPT and (Microsoft) Copilot and those other models are actually generally pretty effective in quickly producing translations, which was kind of surprising to me.”
But things get complicated pretty fast.
While multilingual students using generative AI models may find a quick fix by inserting their paper into the model and asking, ‘Can you improve my English?’, Boyd said the approach often comes up short in providing the cultural context required to explain nuanced English expressions and idioms.
“They’ll just say, ‘This is how you say this in English’ without any kind of cultural context,” she said, noting that English speakers and writers “kind of focus on the main point and the first part of the paragraph or the paper, or even to sentences themselves.
“That’s kind of ingrained in our culture, because we’re very much a direct culture, and that might be a little bit confusing today.”
In other cases, multilingual students might not even understand what the model is asking of them regarding help with organization or grammar.
“First in tutor training, we should really take a culturally linguistic and diverse approach,” she said, noting tutors may need to focus on “low-level concerns” such as grammar, syntax or lexical clarity, that they may have previously avoided “because they’re deemed as unimportant or less critical to the paper.”
“However, for multilingual students, these are at the foundation of their English learning acquisition process, so it’s really important that we take the time to actually explain those issues,” she added. “And that might mean that we need to have more proper training on how to actually teach these issues.”
Boyd shared her insights among several panelists during “GenAI Conversations across Pitt’s Campuses: Student and Faculty Voices on AI at Pitt–Oakland, Greensburg, Johnstown, and Bradford,” a forum held April 22 in the O’Hara Student Center.
Topics discussed included what Pitt students think about AI and platforms such as ChatGPT, how they’re using — or refusing — them, what they see in their classes, and what kinds of support do they want as they navigate the dynamic AI landscape.
As part of the panel discussion, Catherine Kula, associate director of the Pitt–Bradford Writing Center, said she was pleased with the well-attended faculty roundtables and a student forum she held to discuss using AI in classrooms and on class assignments.
“On the Bradford campus, individually, professors are talking about it,” she said. “I think this opportunity is really great, so we can all come together to talk about that.”
Birney Young, assistant professor of communications at Bradford, said he offered extra credit to students who completed a survey about their AI use in his classes.
“Because there was nothing. So I’m particularly grateful for … making this opportunity to actually hear from the Pitt students in the capacity that will be useful,” he said. “Not just my data set.”
Jessica Ghilani, associate professor of communication and affiliate scholar at Pitt–Greensburg, said the campus has “kind of menu of options” for syllabus statements in AI policy that was introduced this year, “and we’ve had a few workshops.”
Sean DiLeonardi, assistant professor of English at Greensburg, said the campus workshops have been successful.
“It’s all faculty-focused, that’s for certain. And I think one of the things that was really attractive to me — as I think it was for all of us — was the opportunity to do a student-focused study to see what student perspectives were on AI,” he said. “Because I think thus far we’ve been kind of, ‘How are we (as faculty) going to adjust to this? What are we doing in our classrooms?’
“And there’s been a lot less attention on how students are actually using these tools, and what are their thoughts and opinions on them.”
Game changer
Responding to a question of how much students know about AI and understand or misunderstand it, Kula said she finds students “definitely view ChatGPT as the new Google.”
“I think that we all kind of saw that on all of our campuses … along with sentiments like, ‘AI is so many different things. How can I have just one set definition?’ ” she said. “I also think that the way that students are using it is also kind of confusing. … I see a lot of misunderstanding around what AI is, what Gen AI is, the definitions of it and some of its applications.”
From the four focus groups at Pitt–Greensburg, Ghilani said she picked up that students “want to steer away” from generative AI, but it’s inescapable in many platforms where they are participants in social media. “So a lot of their definitional understanding about what it is came not only from ChatGPT or the usual suspects, but also from Meta AI or Snapchat’s integrated AI tool.”
On the Oakland campus, Jialei Jiang, teaching assistant professor in the Department of English, said the focus-group interviews of 31 students provided a “really exciting opportunity” for faculty to gain and share insights regarding generative AI.
Finding that students “definitely showcase their different degrees of understanding” related to AI, some said they would “believe AI before (they) would believe a doctor,” she said, prompting laughter from the audience.
Acknowledging the students who “may have overused AI or completely trusted what AI outputs,” she noted that seniors appear to be more “savvy and judicious about their uses of AI, which makes sense in terms of when they would have gone through school.”
Comparatively speaking, generative AI use by first-year students appears to be more “endemic.”
“Regarding misinformation and disinformation, students did show awareness of AI’s limitations, such as its ability to hallucinate, including producing factual inaccuracies and linguistic or cultural biases,” Jiang added. “So these experiences also reflected a cautious attitude toward AI tools, especially when used in academic contexts.”
‘Workload creep’
Elise Silva, director of policy research at Pitt Cyber, said students reported that “AI is viewed as more like a sophisticated and personal form of the internet — which I think is fascinating — where you’re free to search for information and answers.”
Silva, who said she was representing Mark DiMauro and Patty Michael from the Pitt–Johnstown campus, noted that ChatGPT was “by far the most often” mentioned element of AI.
“Something that came across in our groups … was kind of an uneven understanding of personal data and how it was or wasn’t being used, with some students being less concerned about it, and other students being a lot more concerned about things like misinformation or deep fakes, but not necessarily the data aspect of it,” Silva added.
Responding to a question about how professors and students can better connect and collaborate in the context of AI, Ghilani said for a couple assignments, to “minimize the amount of policing,” she asked students to write both a “thought process and a logistical process essay” to explain their creative approach.
“In this case, it was a visual product … using different software tools, some of which included generative AI. And I found that really rewarding, because it allowed for them to use it in a way that was transparent,” she said. “And we had this trust in each other in terms of disclosure —that that wouldn’t work for every assignment.”
As far as reviewing students’ AI use through public links, Ghilani said she’s concerned about the “workload creep that is always affecting those of us — especially with heavy teaching loads — that an additional thing that we have to police, an additional step to getting the grading done is always so much.
“We don’t have teaching assistant support on the regional campuses,” she added. “We don’t have the same kinds of resources to streamline that work. So if I have to also do this, on top of (AI checker) Turnitin, on top of doing all the other things, that’s going to only add to the time it takes.”
Shannon O. Wells is a writer for the University Times. Reach him at shannonw@pitt.edu.
Have a story idea or news to share? Share it with the University Times.