Last spring, Brown University economics professor Roberto Serrano gave his first take-home midterm in his course, ECON 1170: “Welfare Economics and Social Choice Theory.”
Sitting in his office in Providence, Rhode Island, this summer, he recalled how he immediately suspected “massive cheating,” because the grade distribution was, in his words, “absurd.”
“While the historical average of these midterm exams in previous editions of the course ranged between 65 and 80, the average this time was 96,” said Serrano.
Forty students got perfect scores. Serrano and his teaching assistant found patterns they say matched Chat-GPT-generated answers.
Later, in class, Serrano says he confronted his students and asked them what they were doing at Brown.
Their response?
“Silence,” he recalled, shaking his head.
So Serrano told them the final exam would be in-person and closed-book.
“A very sizable number of students dropped the course right after I made that announcement, including 22 students that had scored 100,” he said. “The average score in the final went down to 48 from the original 96.”
Serrano’s case illustrates a growing challenge for colleges: AI is upending higher education, changing not just how students do their work, but what it means to teach, learn and assess what students know.
So far, much of the debate has focused on cheating — students using AI to write papers, solve problems, complete assignments or take-home tests.
Surveys show most students now rely on generative AI for everything from applying to college to getting through it. Some educators are leaning in, treating AI as a teaching partner. Others are leaning out, doubling down on in-person, discussion-based learning, blue books and oral exams.
Nearly two-thirds of professors at Harvard now say the widespread use of AI has had a negative effect on their courses. And a new report from the Massachusetts Institute of Technology warns overreliance on AI can weaken memory and cause “cognitive surrender.”
“It’s rare that a successful institution has to take a fresh look at many aspects of its mission,” said MIT President Sally Kornbluth. “But I’m convinced that the opportunities and risks generative AI poses for our model of education and research now constitute such a watershed for MIT — and for all of higher education.”
That leaves professors and college leaders scrambling to respond and wrestling with a broader concern: If AI can do the difficult work for students, will they develop the ability to think critically, solve problems and struggle through something they don’t immediately understand?
Without higher standards, stronger assessment and more oversight, some professors worry AI could erode the foundations of learning.
Jeffrey Cohen, dean of humanities at Arizona State University, says colleges can’t catch every student who uses AI to cheat. Instead, Cohen says, professors need to design courses that demand students take the initiative to learn.
“I won’t say it’s an easy road,” Cohen said, laughing. “AI itself keeps changing in what it’s capable of doing, and it feels like we’re always trying to keep a little bit ahead of it.”
ASU was one of OpenAI’s first institutional partners, and the university has integrated the technology into much of its coursework and student support services.
Cohen says ASU has adopted “a positive-mind frame” around the technology rather than trying to “batten down the hatches.” Other schools, including the University of Chicago, the University of California, Berkeley, Cornell University and Columbia University, have banned the technology in certain courses and for certain assignments.
At Brown, administrators are not banning AI and Serrano describes the university’s initial response to alleged cheating on his mid-term as “meek.”
“The relevant committee didn’t even acknowledge the submission of the case,” he said.
Brown administrators say the faculty-led academic conduct committee is now investigating Serrano’s case.
“I think we’re seeing this around the nation popping up,” Brown Provost Frank Doyle told GBH News.
But Doyle says the Ivy League university isn’t simply trying to ban or police AI use. Like MIT and other universities, it’s studying how students are using it.
An internal survey this summer found that more than half of Brown undergraduates — and about two-thirds of PhD and medical students — say they intentionally use generative A.I. tools daily or weekly.
“So it’s quite prevalent across the student body,” Doyle said, noting the university offers computer science courses that bring these tools into the classroom. “But we also have a burgeoning number of courses that are harnessing classical disciplines to put AI under the lens.”
One new interdisciplinary humanities seminar called “Reading the Large Language Models,” for example, examines human-like writing.
Until now, individual professors at Brown largely set their own AI rules. So the university has created customizable syllabus templates to make those expectations clearer to students tempted to lean on the technology to do their work.
The bigger challenge now is figuring out what material and skills students should be assessed on — and how.
“Assessment has been broken before AI,” said former Southern New Hampshire University president Paul LeBlanc. “Like COVID, it takes negative trends and accelerates them.”
LeBlanc is teaching a new course this fall at Harvard called “Designing An AI-Powered University.” He says the answer isn’t simply banning AI, but emphasizing the skills technology can’t easily replace.
“We need to give students a more powerful cognitive toolkit,” LeBlanc said. “That includes critical thinking, their interpersonal skills, creativity, sensemaking, empathy.”
At MIT, where much of this new technology was first developed, roughly half of undergrads report using AI daily. The faculty-led report released in August concluded the technology has created “major shifts in campus culture.” Students are isolating themselves in their dorms, skipping office hours and communal studying. The report says taking answers from a chatbot can contribute to what professors call “the illusion of learning.”
MIT Chancellor Melissa Nobles says that’s why MIT is trying to make sure technology doesn’t replace the human connection at the heart of learning.
“A big part of the MIT intellectual culture is working together in community and collaboration. The P-sets,” she said. “Hard problems require people working together, and that means human-to-human interaction.”
Nobles says AI shortcuts can short-circuit that process because struggle — and even failure — are part of learning.
“We want to make sure that energy is rooted in the students learning themselves and doing the hard work,” she said. “And we have to make sure that that happens with AI and without it.”
Nobles, a political scientist by training, says the core mission of a university is to generate ideas and teach students not what to think, but how to think.
An existential threat to higher ed?
Some see AI as an existential threat, not just to higher education and the humanities, but to humanity itself. But universities have survived some profound technological and social changes in the past.
In his book, “The Uses of the University,” Clark Kerr, the first chancellor of the University of California, Berkeley, noted that about 80 institutions in the Western world established in the 1500s still exist today. Seventy are universities; the others include the Catholic Church and a handful of parliaments.
So today, while politicians and AI leaders warn about doomsday scenarios, some professors are bullish about the ability of colleges to adapt.
ASU’s Jeffrey Cohen, a self-described “medievalist” and “higher ed optimist,” says colleges will survive threats posed by the technology. And he says blaming students for being lazy or less dedicated than previous generations is hardly new.
“Way back in the Middle Ages, the monks who were teaching the novices would say, ‘Novices these days! They don’t read anything. They don’t learn their Latin. They’re just not diligent.’”
Cohen says faculty should redesign assignments and classrooms rather than simply assume students are mailing it in.
“It’s going to take a little more work and a little more affirmation of what we’re doing well,” he said, suggesting colleges avoid going “down the road of ‘here’s why we’re on cheat patrol all the time.’”
Ahead of the fall semester, Brown professor Roberto Serrano met with his new teaching assistant in his office on a recent morning, reviewing some new class policies.
Serrano isn’t banning AI. But homework will now count for zero percent of the final grade. Each lecture will include an ethics question, and he’s adding an oral exam using new webcam software designed by Brown engineers to detect cheating.
“The software will be able to tell if the student, instead of looking at the camera, is looking at another screen to get the answer,” Serrano explained.
“So they will provide the answers?” his teaching assistant asked.
“Yes, they will provide the answers! You will get the full recording,” Serrano said.
Serrano says colleges need to adapt quickly before AI changes the nature of learning itself.
“AI could be very useful,” he said. “It’s like another encyclopedia, but we cannot have a civilization in which technology is driving the car and humans are in the backseat. No, no. The humans should be continuing to drive the car.”
“We cannot choose to become idiots,” he said.
Otherwise, Serrano says, colleges risk normalizing cheating and cognitive offloading — and allowing technology to do the hard work of thinking and learning for students.
