# As students embrace AI, universities scramble to keep up > Source: > Published: 2026-08-23 02:46:21+00:00 Inside a classroom at a California State University campus in Vallejo, six whiteboards line three walls, all covered in dry-erase markers. Four students cluster at each board, working through differential equations. No laptops or phones are visible. Taiyo Inoue walks through the room, hands clasped behind his back, watching each group in turn. He is an intense presence, with piercing brown eyes and graying stubble. His face is often fixed in a perpetual frown, his brows furrowed as if in confusion. He will not lecture today. He is not your typical college professor. Instead, he has posted his lectures on YouTube for students to watch on their own time. If they did not prepare for class today, he will know. A mathematics professor at Cal State’s Maritime Academy, Inoue had spent years lecturing from a podium and assigning problem sets for homework. By 2025, artificial intelligence was transforming the work his students submitted. Homework that once took students an evening suddenly began returning within minutes, error-free but pointless. Inoue had long been worried about [how AI might impact learning](https://www.deseret.com/opinion/2025/12/07/ai-education-role-in-schools/), but over time, his worries morphed into something much more unsettling. Students weren’t just using ChatGPT to write papers or cheat on tests, he realized. They were using it to think. “These new technologies allow students to completely and totally offload their cognition,” Inoue told me, “giving themselves up, their identity as students, totally and completely to machines. That’s the disaster scenario for higher education, because it completely dilutes the value and the signal that a college degree confers on its holders.” The advent of AI has already upended the classroom in just a few short years, and the stakes are existential. “It’s going to lead to the death spiral of higher education as an institution if we don’t do something about it,” Inoue says. “These new technologies allow students to completely and totally offload their cognition, giving themselves up, their identity as students, totally and completely to machines.” — Taiyo Inoue Inoue put his lectures online for the spring 2026 semester and moved the problem-solving into the classroom. Students now spend 150 minutes a week working through unfamiliar problems at the whiteboards, where he can watch the students reason and assess their participation in real time. Inoue’s bold experiment is an answer to something that becomes increasingly urgent with each passing year: Higher ed is in crisis. Universities are no longer debating whether artificial intelligence will transform learning. They are scrambling to respond. A recent survey found that nearly 90 percent of Harvard undergraduates used AI for coursework, and roughly a quarter said they used it instead of doing assigned reading. At the University of Reading in the United Kingdom, researchers quietly slipped AI-generated answers into real undergraduate exams. Human graders identified them just 6 percent of the time. Artificial intelligence isn’t just changing how universities teach. It’s forcing them to rethink why they exist. Colleges have long joined three functions: transmitting knowledge, validating mastery and forming character. AI is now pulling those functions apart. Information is becoming abundant and personalized, weakening the university’s traditional role as the provider of knowledge. Work completed outside the classroom can no longer reliably show what a student knows, forcing institutions to redesign how they certify mastery. And the habits universities have associated with formation — sustained attention, intellectual struggle and responsibility to other people — may disappear when the work that develops them is automated. The question is no longer simply how universities should use AI. It is which parts of a university can be automated without hollowing out the institution and damaging the society it aims to serve. “With each passing day a machine becomes capable — or seemingly capable — of doing something that until yesterday you and I thought only a human being could ever do,” said Adam Kronk, an ethicist at the University of Notre Dame’s Institute for Ethics and the Common Good. “What does it mean to be a human being? What does human flourishing look like? Why am I here? Why am I in school? It makes these questions much more acute and felt.” Universities cannot decide what to do with AI classroom by classroom, he argues, until they answer that larger question first. ### Proving what students actually know Every university is now trying to answer the same question: How do you know what a student actually knows? For many, the answer begins with reclaiming the classroom itself. Long before ChatGPT, educators had experimented with moving lectures online and using class time for discussion, practice and problem-solving. Artificial intelligence transformed that experiment into something far more urgent. Few universities have embraced that shift as deliberately as the University of Sydney. Years before ChatGPT, Adam Bridgeman, the university’s pro vice-chancellor for educational innovation, had begun moving lectures online. AI simply forced him to confront the question driving higher education today: If information can be delivered anywhere, what still requires a classroom? Only about 1 in 8 students at Sydney now attend traditional lectures, he told me. The university still owns the massive concrete auditoriums, but their original function has largely migrated online. “If I had better infrastructure we could do this better,” Bridgeman said. “But we have the infrastructure we have and so we have to make use of that.” California State University is revamping its entire approach to education in response to AI. Rather than trying to wall the technology off, the nation’s largest four-year public university system is redesigning teaching, assessment and workforce preparation around the assumption that every graduate will use AI. In January 2025, the 23-campus system signed a $16.9 million agreement with OpenAI, placing ChatGPT in the hands of hundreds of thousands of students, faculty and staff. Inoue had begun by asking what students must still do for themselves. CSU began by distributing the tool. AI simply forced him to confront the question driving higher education today: If information can be delivered anywhere, what still requires a classroom? Inoue defends the decision as a matter of equity. At roughly $1.60 per user per month, CSU could provide the same enterprise version, including stronger data protections, to students who could not afford individual subscriptions. “I am not fond of a classroom of AI-haves and AI-have-nots,” he said. Critics worried CSU was mistaking technology for equity. The university introduced AI without evidence that it improved mastery, retention or independent reasoning. Wealthier students would still have professors who knew them, challenged them and pushed their thinking. Students elsewhere might increasingly be asked to learn with machines. CSU treated access to AI as the urgent divide. The more consequential divide may be access to people. Universities are beginning to split into two camps. One sees AI primarily as a tool to expand access and personalize learning. The other believes the more AI enters education, the more valuable human instruction becomes. The future of higher education may depend on which vision prevails. Bridgeman says the choice is a false one. If students’ needs will always exceed society’s capacity to provide deep, personalized instruction, AI may be part of a solution to scale opportunity. But that will only work if universities can determine what their unique offering is, while also figuring out how to deliver it and measure results. This past spring, Brown University saw how easily AI can call into question what a student ostensibly learns by taking a course, whether taking a course even makes sense and why professors even exist. After students expressed anxiety about gathering in a classroom following a campus shooting, economics professor Roberto Serrano allowed them to take the midterm at home. He had taught the course for nearly two decades, and midterm grades typically averaged between 65 percent and 80 percent. On the take-home exam, the average was an astounding and improbable 96 percent. Serrano then moved the final back into the classroom and warned that if the gap between a student’s midterm and final scores was excessive, the final would determine the entire course grade. Eighteen students dropped the class, and nine others remained enrolled but did not take the exam. Among the 59 who did the final, the average score was 48.6 percent, the lowest Serrano had ever recorded. At the University of Sydney, Bridgeman believes that trust has to be rebuilt from the ground up. “We have to know what our graduates can do,” he said, “and they have to be able to do what we say they can do.” But he also notes with equal emphasis that graduates must be prepared for a world in which AI use is expected. Whatever answer emerges, he argues, must take both of these into account. Sydney’s answer is a “two-lane” model. In Lane 1, students prove mastery under secure conditions through in-class writing, proctored exams, oral defenses and other controlled assessments. Lane 2 assumes AI. The principle is simple: Professors must validate what students need to know and cannot prohibit what they cannot prevent. One lane measures what students can do on their own. The other prepares them for the world they’ll actually enter. “It’s at the degree level that we need to make sure what the student can do,” Bridgeman said. “That’s where the planning has to happen, not at the individual Chemistry 101 level.” Every department must identify the knowledge and skills graduates are expected to possess — and how students will prove they have them. A writing student may defend an essay in an oral audit. An engineering student may demonstrate a laboratory procedure or explain how a technical problem was solved. The redesign took about 18 months of debate. It also came with a cost. In one writing course, roughly 600 students each complete a 10-minute oral audit to demonstrate they understand the argument they submitted and can explain how they developed it. AI may make information cheaper. Trust still requires human time. The University of Chicago Law School sees the problem in more philosophical terms. Its concern is not only whether students can prove mastery, but whether the difficult work required to achieve it has value in itself. This summer, Chicago Law announced a pilot policy removing electronic devices from core first-year classes, keeping exams in person and offline, and requiring students to discuss substantial research papers orally with professors. William Hubbard, who chaired the committee that developed the strategy, acknowledged that AI’s shortcuts can be valuable when professional efficiency is the goal. Education is different. Those same shortcuts can be “very, very damaging,” he said, because “the whole point is to do things the hard way — because that’s how you learn.” For Hubbard, that difficulty prepares students for work that cannot be reduced to producing a correct answer. Lawyers must interpret incomplete facts, explain their judgments, respond to challenges and accept responsibility for the consequences. A lot of legal work is not simply about getting something done, he said. It’s about a human being getting something done. The friction is therefore not an inefficiency surrounding education. It is part of the education. Or is it? If AI can write papers, take tests and give lectures, why couldn’t it also introduce the friction required to learn? That’s what Panos Ipeirotis at New York University’s Stern School of Business is trying to figure out. He calls his system Viva, or as Ipeirotis put it, “fighting fire with fire.” Viva conducts personalized oral examinations on students’ semester projects and the course’s core concepts. Students may use AI freely to research, write code or build their projects, but they must then explain and defend the work aloud before an examiner that adjusts its questions to the strength of their answers. In Ipeirotis’ approach, an AI examiner conducts the oral test, while multiple large language models independently assess the transcript and review one another’s judgments. Students may take the exam repeatedly throughout the semester. Each attempt draws different questions, and stronger answers invite harder follow-ups. Rather than using AI to make academic work easier, Viva uses it to demand more from the student. “The system says, ‘Oh, you are answering this very nicely, so let me probe deeper,’” Ipeirotis said. One student put it more bluntly: “The better I do, the ceiling rises.” Students initially regarded Viva’s shifting standard as unfair, but by its second semester, a slight majority accepted it. Ipeirotis also invited students to challenge their grades by identifying errors in the exam transcript, but none of the appeals he reviewed showed that the system had misjudged an answer. Ipeirotis is candid about its limits. Viva remains vulnerable to cheating, and he has asked students to try. The most effective exploit used a second phone to transcribe the examiner’s question, send it to another chatbot and feed the answer back. Ipeirotis expects that high-stakes versions may still require human proctors. He is equally candid about scale. “I built something boutique for myself that I know how to use,” he said, and he is uncertain whether it would work as well for another professor or across an institution. ### The value of doing things the hard way The real risk of AI isn’t that students stop learning. It’s that they stop becoming. Shannon Vallor, a philosopher of technology at the University of Edinburgh and a former Google AI ethicist, has spent years studying what happens when machines assume tasks that once exercised human capacities. She calls the net result “moral deskilling.” Automation, she warned, can weaken not only our ability to perform particular tasks but the habits of judgment developed by performing them. The mass adoption of generative AI, Vallor argues, has placed that danger on an unprecedented scale. “We’ve essentially conducted an unregulated mass social and psychological experiment on the entire human population without ethics approval, without safety testing,” she said. “We have already unleashed this technology at scale in society.” The risk is greatest when machines begin exercising capacities for which no substitute is obvious. “It’s one thing to take away our ability to make shoes for ourselves,” Vallor said. “It’s another thing to automate away our ability to make our own moral decisions.” “It’s one thing to take away our ability to make shoes for ourselves. It’s another thing to automate away our ability to make our own moral decisions.” — Shannon Vallor This taps into something universities have long taken for granted. Judgment is not information a person simply possesses. It is a capacity built through repeated acts of attention, deliberation and choice. Practical wisdom, empathy, courage and self-command are not downloaded or memorized. They are cultivated. “Young people need time to develop skills of critical thinking,” she said. “They need time to develop the capacity to make complex moral decisions and they need practice doing that.” When AI supplies the interpretation, drafts the argument or resolves uncertainty, it may preserve the appearance of competence while quietly eroding the habits that competence depends on. The concern extends beyond reasoning. Those same habits are formed in conversation, disagreement and the experience of encountering people who resist, misunderstand or challenge us. Vallor is careful not to overstate what research has yet shown about AI’s role. But she worries that if machines increasingly mediate how young people think, decide and communicate, they may also weaken capacities no algorithm can supply: the patience to deliberate, the willingness to revise one’s views and the ability to pursue a common good with people unlike ourselves. That kind of practice requires more than assignments. It requires places where people think together, disagree, explain themselves and discover the limits of their own understanding. Katie Day Good initially brought writing back into the classroom because she needed to know who had produced it. What she recovered was larger than authorship: trust between teacher and student, physical presence and the experience of developing an idea before other people. For years, the communications professor at Calvin University in Michigan had assessed students in film and media studies with the standard take-home analytical essay. Then ChatGPT arrived. As AI-generated prose became harder to recognize, Good began questioning who had written the papers. “Very quickly I realized that my approach to assessment would no longer work,” she said. “The take-home essay was no longer a viable, at least for me, primary form of assessment.” Good moved writing into the classroom in spring 2023. Students now spend three hours writing analytical essays by hand, surrounded by classmates wrestling with the same uncertainty. What returned surprised her. Surveillance was the least interesting thing the classroom gave back. Students watched one another think. Ideas took shape in public. The essay became less a product than an event: sustained attention, visible effort and vulnerability shared in a common room. Good no longer tries to police every use of AI during preparation. Because students must ultimately explain and defend their ideas in person, the important work still belongs to them. The classroom becomes one of the few places where students and professors can be confident they are encountering another mind. Human presence restores more than confidence in authorship. It forms students by making them answerable to other people. ### What education is supposed to make of us But what kind of person is higher education trying to form? Roosevelt Montás, a professor at Bard College who founded Columbia University’s Freedom and Citizenship program, believes the answer is someone capable of governing both a democracy and himself. He calls that capacity self-government. Every July, 45 rising high school seniors gather around seminar tables at Columbia University to argue about Socrates, Frederick Douglass, James Baldwin and the meaning of freedom. Many hope to become the first in their families to attend college. For a month, they read difficult books slowly, debate questions with no settled answers and discover that education can be something more than preparation for a career. “If you put an experience like that in front of these kids,” Montás told me, “will it do something for them like what it did for me? ... And it does.” The program recreates the education that transformed Montás himself. After immigrating from the Dominican Republic to Queens at age 11, he found two volumes of the “Harvard Classics” in a pile of discarded books and struggled through Plato’s account of the trial and death of Socrates. A teacher noticed him reading in the hallway and began staying after school to discuss Plato, college and the wider world. “Finding him was exactly what I needed,” Montás recalled. “The thing that’s transformative about Plato and great books in general is that they offer a set of provocations for you to grapple with.” — Roosevelt Montás That teacher helped Montás reach Columbia, where the Core Curriculum turned an accidental encounter with great books into a sustained education. Students from every discipline wrestled together with enduring questions of justice, freedom, beauty and the good life. Montás earned his doctorate there, joined the faculty and eventually directed the Core. He fears that education has become harder to find. Most universities still require courses outside a student’s major, he says, but the larger purpose has faded. General education has become a collection of disconnected requirements rather than a coherent attempt to educate the whole person. “There is no organized, cohesive education aimed at the whole person,” Montás said. Liberal education, he argues, should help students order their own lives, examine their desires and exercise judgment rather than merely react to impulse or pressure. It should also prepare them for democratic life: to make arguments, hear opposing views, resist propaganda and deliberate with people unlike themselves. “The project of self-governance has two sides,” he said. The personal side requires organizing one’s “inner life” to maximize freedom and autonomy, while the collective side requires the capacity to “empathize and conceive of a common good.” Artificial intelligence threatens that project because it can bypass the struggle that produces understanding. A chatbot can summarize Plato or identify the themes of a novel. It cannot supply the experience of wrestling with a text, testing an interpretation against other people and discovering that the questions have become personal. “The thing that’s transformative about Plato and great books in general,” Montás said, “is that they offer a set of provocations for you to grapple with.” The growth is in the grappling, he argues. For nearly two decades, Montás has tried to recreate the chain of encounters that changed his own life. He launched Freedom and Citizenship at Columbia in 2009 with 15 low-income high school students. The program now brings about 45 rising seniors to campus each summer for intensive seminars on freedom, justice and citizenship, followed during the school year by mentoring, college guidance and civic projects. Its model — difficult books, close discussion, sustained mentorship and practical help entering college — has spread through the Teagle Foundation’s Knowledge for Freedom network to more than three dozen institutions, including Yale, Princeton, New York University, Washington University in St. Louis, Baylor University and the University of Guam. The expansion of Freedom and Citizenship beyond Columbia reveals both the promise and the limit of Montás’ argument. Information can be distributed at enormous scale, and mastery can be tested through increasingly sophisticated systems. Formation of character is harder. It requires Socratic dialogue on difficult questions, sustained attention, and teachers and students willing to become answerable to one another. “Education,” Montás said, is “a thing that happens from person to person.” The principle holds whether the subject is Plato or differential equations. In Vallejo, Inoue’s students are back at the whiteboards. The lectures can travel over the internet, and machines can supply explanations in seconds. But students must still work through the problem, defend what they know and accept corrections from someone standing before them. Artificial intelligence is separating the functions universities once joined. What remains distinctly theirs is the work of turning information into knowledge, knowledge into judgment and judgment into a life lived responsibly among other people.