Michigan universities embrace AI, but struggle to regulate student use
Video: PhD students at MSU develop smartphone app to help patients communicate with physicians
Video: PhD students at MSU develop smartphone app to help patients communicate with physicians
Michigan State University Ph.D. student Sara Rezarimanesh didn’t grow up with ChatGPT or any other generative artificial intelligence tools, but she now works in MSU’s Human Augmentation and Artificial Intelligence Laboratory.
Rezarimanesh and her colleague, Niloufar Eghbali-Zarch, are developing an AI-integrated app that allows patients to have a personal health advocate always with them. The app was developed with the use of a Large Language Model that was fed medical knowledge.
Rezarimanesh, 24, and Eghbali-Zarch, 34, described themselves as “self-taught” in generative AI and LLMs, unlike current undergraduate students. Without this, the women said they could learn and understand code without relying on the tools and feel comfortable using the tools to develop their app.
However, some of the undergraduate students Rezarimanesh teaches in an introductory coding class rely solely on generative AI to complete their assignments, she said. Because they never had the chance to learn how to code without AI, those students can’t tell when the AI’s output is wrong and aren’t learning how to fix it, she said.
Students who are relying on an imperfect program to do the work for them aren’t learning, said Phillip Snalune, co-founder of Codio, an AI-powered tech skill and development platform based in Cambridge, Massachusetts, that has partnerships with universities nationwide, including MSU.
As more Michigan universities lean into AI, through initiatives, institutes or university-specific tools, most still lack clear university policies and standards on how generative AI should be used. Instead, university leaders are leaving the regulation of AI up to individual departments and professors.
But the differing faculty opinions on AI mean some students are exposed to a patchwork of guidance on the tools, which can lead to confusion and leave some students out of a tool they’ll likely use in their future careers, according to faculty who’ve adopted the tools into their classes.
The faculty’s fears about students using the tools to cheat are valid, said Grand Valley State University Assistant Professor of Writing and Interdisciplinary Studies Alisha Karabinus. But she said she was concerned about students experiencing a sort of “whiplash” going from class to class and each having different rules on the use of AI.
“I think that the real issue isn’t that it could be considered a plagiarism machine,” Karabinus said. “The real issue is that our students are living in a very complicated reality.”
The varying needs of different faculty and programs, along with the ever-changing nature of AI, make it difficult to enforce a standardized approach to using AI in the classroom, said Ravi Pendse, vice president for information technology and chief information officer at the University of Michigan.
“One thing we don’t want to do, and something I wouldn’t advise, is to have a cookie cutter approach for everything, because the kind of AI that one might need and the kind of exposure to AI tools that one might need in marketing might be different than what one would need in English,” Pendse said. “What’s critical is the idea of AI literacy.”
What AI use looks like in classrooms
Rezarimanesh’s professor, Mohammad Ghassemi, said he wants his students to think of AI as a “learning partner.”
Ghassemi is the head of the lab where Rezarimanesh and Eghbali-Zarch work. He said the point of their work is to find where the capabilities of humans and those of artificial intelligence meet and how they can solve problems together that neither would be able to do as effectively on their own. This idea, he said, could be how higher education looks to integrate AI.
“AI systems are phenomenal at telling us how we can achieve what we’re interested in,” he said. “But what AI systems are not as good at is the what and the why.”
The pressures of the workforce make it necessary for students to have an understanding of how to use the tools, he added. And when used effectively, can be a benefit for the students.
“(AI) is basically a very cheap, high-quality tutor that can help be a catalyst for ideation,” Ghassemi said. “When you enter the workforce, unless you’re in a very strange organization, people are not going to throw away productivity by stopping you from using an AI, too. It would be as silly as us saying we shouldn’t use the spell checker in Microsoft Word because people are going to be slower at spelling words or we shouldn’t use a calculator, because people are going to forget their multiplication tables.”
Ghassemi encourages students to use AI to help them as they complete homework and projects. However, there’s a “gotcha” with this, he said.
Ghassemi or one of his teaching assistants, like Rezarimanesh, will have a brief call with each student a couple of times each semester. The students are questioned on what the code they built looks like, why they made the decisions they did and other questions related to the design process.
This way, the student can show how the decisions they made were their own or whether it’s clear they relied on AI to complete their assignments, he said.
“The reason I do these exams is because when you enter the industry, even if you’re using an AI agent to code what you’re doing, you’re still going to have to be able to explain to your peers, who are human beings, why you did what you did and what problems you think are necessary to go after next,” Ghassemi said.
Grand Valley’s Karabinus said she allows her students to use AI in class unless she makes it clear otherwise. But in her classes, particularly her undergraduate digital studies course, she said she takes time to show how AI cannot be relied on all of the time and encourages her students to cross-reference the information they receive.
Karabinus works through several “stress tests” with generative AI tools. Often, she said she finds the tools break down after repeated questioning and would give conflicting information. This is a common flaw in the systems, she said.
“When I use it, I want to test ‘What are responses going to be?'” Karabinus said. “‘What is this going to look like? How is somebody going to look at this?’ And I will often prepare materials for my students in which I will say, ‘This is what a generated document in response to this homework prompt looks like, and it will always look the same.'”
Concerns over use, calls for guidance
Karabinus cautions her students against relying on the generative AI to complete assignments for them that they don’t already understand how to do themselves.
“I’ve always presented (AI) as a double-edged sword,” she said. “This is a tool, and it can be good for some things. But we also have to be mindful about the things that it’s not good for, the implications of use and the problems that come with not knowing where the information comes from.
“Every response is a whole new world for the machine,” Karabinus continued. “It’s not required to tell us the truth. It’s not required to be consistent across responses.”
Rezarimanesh said she experienced some frustration when talking with students about their use of AI. She had seen students submit code for a programming-heavy class that wouldn’t work, and she held a session where she went over some of the submissions. She said she was stunned that the students couldn’t tell when there was a clear mistake.
“If (my students) cannot tell me what’s wrong with the code, they’d shouldn’t use (AI),” Rezarimanesh said. “(AI) doesn’t have common sense. I use an example of asking an LLM how you should fix a broken boat to cross a river, and what the LLM does is it suggests drying the river to walk across instead of telling you how to fix your boat.”
UM Professor of Comparative Literature and German Studies Silke-Maria Weineck said she was extremely concerned about her students. Generative AI is “strictly forbidden” in her class, and she believes some of her students have submitted work they asked AI to write for them.
However, some of Weineck’s students are more critical about the tools than the university’s leadership, she said. She said she’s heard from students that they’re hesitant to use it at all or worry that their job prospects will disappear as employers turn to the cheaper generative AI to fill entry-level roles that were historically filled by recent college graduates.
UM’s university-specific tools are also a point of contention for her and other faculty who oppose the use of generative AI in their classrooms. The tools, first introduced in August 2023, weren’t rolled out with guidance for faculty and students on how to responsibly and ethically use them, Weineck said.
Leaving the faculty to their own devices in regulating them has led to conflicts with students. UM is being sued by a student for being accused of AI use and denied disability accommodations when she tried to appeal.
The ongoing dispute about how generative AI should be used on campus led to UM’s Faculty Senate approving a resolution, introduced by Weineck, to urge the university to provide stronger guidance and policies for the tool’s use.
The resolution was approved with 86.1% faculty support. It read in part that the university must create a comprehensive generative AI strategy that offers faculty, students and staff the opportunities to learn how to use AI ethically and addresses other concerns, such as its impacts on the environment, the amount of spending on the tools and the resulting loss of skills among students.
Generative AI is an extremely resource-intensive tool. Data centers, needed to do the computing the tools require, use billions of liters of water each year and rely on fossil fuels. In 2024, U.S. data centers used 560 billion liters of water, approximately as much as 833,000 American households, according to research by Northern Arizona University’s School of Informatics, Computing and Cyber Systems, where researchers are tracking the facilities’ water use throughout their supply chain.
What universities can do
Like UM, Michigan State and Grand Valley are among the universities in Michigan that have rolled out expanded initiatives meant to train students in the use of AI.
At MSU, the Green and White Council, created by President Kevin Guskiewicz, recently released an initiative called AI-Ready Spartans. The initiative would “build (students’) digital competencies by empowering faculty to lead the integration of AI across the undergraduate experience, while partnering with industry leaders to provide employer insights, with a goal of ensuring every graduate has ample opportunity to learn how to use AI effectively and responsibly.”
MSU’s policies on generative AI are stronger, but still leave most of the decision-making up to professors. The university requires that generative AI tools “should not be used to deliberately fabricate, falsify, or misrepresent information; impersonate others or oneself; or create deceptive content, except when explicitly authorized for instructional or research purposes within a controlled and approved environment.”
Grand Valley recently received $1 million in federal support for an initiative to unite faculty experts and partners in industry and the public sector to design responsible, trustworthy AI systems. Like MSU, the university prohibits several uses of AI, including “malicious use.” Specific use within classrooms is left up to the faculty.
Oakland University is set to build a data center and AI institute on campus, pending board approval. Wayne State University’s board approved the creation of the Institute for AI and DAta Science (AIDAS), which would build on existing university AI research.
UM’s Pendse acknowledged why faculty were frustrated and said that after the Faculty Senate’s resolution passed, he assembled a committee to answer the questions it raised.
“The idea is that when the work is done, there’ll be guidance coming out of this committee, providing ideas and suggestions for faculty members on how to think about this technology,” Pendse said. “Right now, it is possible that depending on the class you’re in and the discipline, you may get inconsistent advice and guidance.”
Pendse was adamant that university leadership does not need to require the choices individual professors make in their classrooms.
“I personally am not in favor of mandating something and making everyone use (generative AI),” he said. “Instead, we’d provide very thoughtful guidance to faculty members and encourage them to consider using that guidance, which will be coming from their faculty peers.”
Codio’s Snalune said he’d be very surprised if all universities didn’t eventually standardize the use of AI and expand opportunities for students to learn the tool. However, he said, no policy could be developed without the concerns of privacy, ethics and governance being addressed first.
MSU’s Ghassemi said there could be a place for a “general education” style class that undergraduates are required to take on the use of generative AI, so they’re entering the university with the skills needed to responsibly use AI.
“I know lots of universities, including MSU, are thinking about how to train students very early in their academic career, maybe as part of the core curriculum, to understand how they can use AI to really grow throughout their professional journey,” he said. “I think you want to do that right when they enter. … Increasingly, they’re going to have to collaborate with AI, whether they like it or not.”
And understanding responsible use can lead to the kind of work being done by Rezarimanesh and Eghbali-Zarch, Ghassemi said. The app developed by the women in tandem with generative AI will soon be tested with actual patients at the Henry Ford Hospital in Detroit, a close partner with the university.
“AI can help us learn and solve problems that maybe weren’t solvable before,” Eghbali-Zarch said. “The tools we build are not intended to replace humans, but to enhance their decisions or help them do their jobs better.”
satwood@detroitnews.com
