AI Didn’t Break the University. It Revealed What Was Already Broken
Within the past month, Stanford and Princeton ended honor codes that had governed student conduct for more than a century. Both universities will return to proctored in-person exams this fall after waves of AI-assisted cheating and collapsing trust in take-home assessment. Glasgow and other UK universities are doing the same. A recent preprint analyzing 3.2 million learning interactions found that under proctored conditions, students’ odds of answering correctly have dropped by 25 percent since ChatGPT’s release. In unproctored settings, scores rise. Students are outsourcing cognition. Institutions are discovering how little learning remained underneath the credential.
Carl Hendrick recently described these developments as “cognitive surrender” - a term coined in a Wharton working paper by Steven Shaw and Gideon Nave - and argued that the social contract underpinning the undergraduate credential has collapsed. He is right that something has broken.
I have argued in these pages before that AI itself is not the cause of the crisis; that blaming the chatbot for cheating and the LLM for loneliness is like blaming the calculator for poor math pedagogy. The deeper failure is institutional, and it preceded the technology. As I wrote here in March, the scarce resource in higher education is no longer knowledge but the human encounter itself. The most penetrating response to Hendrick’s essay confirms that and it did not come from another professor; it came from a 49-year-old graduate student named Norma Sancho, writing in the comments. Her reply has since circulated on its own, and it deserves to be read widely.
Sancho is enrolled in a private online education program. What she receives are videos, PDF study guides, and video-call classes where teachers lecture to students whose cameras and microphones are turned off. All assessments are multiple choice. There is no forum where students can meet, build networks, or experience anything resembling an intellectual community. When she began studying, she was excited. She took notes by hand. She wrote to the principal asking whether the school could create a space for students to exchange ideas. She received no reply. A few days later, she found herself discussing behaviorism with ChatGPT - not because she was lazy or un-motivated to engage with others, but because no one else was there.
Her question, addressed to Hendrick and to the rest of us in the academy: “What kind of economy and educational system has made a robot feel more viable, attentive, more dialogic, and more intellectually available than the institution charging students for a degree?” That is the question higher education has been avoiding.
Hendrick treats AI as the variable that broke the credential. Sancho treats it as the symptom. The chatbot did not create the void. It entered a void institutions had already made.
The labor market made the degree transactional. The institution withdrew the human encounter. AI arrived to fill the silence both had created.
The fashionable pedagogical answer, Wong and Qiu’s “think first, ChatGPT later”, presupposes a teacher who notices whether you thought. Sancho’s account is that the teacher has been engineered out of the equation.
People have asked me for years why I have stayed at Sarah Lawrence College despite more work and less money than I would have earned elsewhere. The honest answer is formation. Sarah Lawrence’s conference system pairs students with professors for sustained one-on-one work on their ideas and writing; the don system assigns each student a faculty advisor who follows them across their time at the college. The conference forces thinking into relationships. A professor who reads your drafts every week can tell the difference between genuine intellectual struggle and synthetic fluency. More importantly, the student can tell the difference too.
I see this in my own students. They use AI - of course they do - but when we talk about it in conference, they are nearly unanimous that they do not like it, do not trust it, and do not want it as a substitute for their own thinking. They reach for it when they are stuck or exhausted, and they feel worse about their work when they do. I have spoken with students at large research universities who describe the opposite condition: unseen, unheard, disconnected from any faculty member who knows them, navigating their education through learning management systems and graduate-student instructors who rotate out every semester. They are not lazy and they are not cynical. They are responding rationally to an environment that has told them their thinking does not warrant anyone’s sustained attention. This is the progressive tradition that Sarah Lawrence inherited from John Dewey: progressive in the early-twentieth-century pedagogical sense, not the contemporary political one: the conviction that learning happens through structured relationship, not content delivery. That tradition anticipated the chatbot crisis by a century, because it never accepted the premise that produced it.
The model Sancho describes - the video lecture, the disabled camera, the multiple-choice assessment, the unanswered email - is now the default experience of higher education for so many students. The residential seminar with a professor who knows your name is the exception, marketed at premium prices. Sancho is not complaining that her program failed to be Sarah Lawrence. She is observing that her program scaled away the human encounter and continued to charge for it.
This is where Hendrick’s frame falls short. The deeper rupture is older than ChatGPT. The university began outsourcing teaching to administrative apparatus, contingent labor, and asynchronous content delivery long before any student opened a chatbot. The credential had already begun hollowing out decades ago. Generative AI did not produce that hollowness. It made it visible.
Three things follow.
First, detection is not formation. Proctoring software, AI-detection tools, and elaborate redesigns of assignments to thwart language models are measurement fixes. They can tell us whether learning happened. They cannot make learning happen, and they cannot rebuild what the institutions have spent thirty years disinvesting from.
Second, human-scale education is now a scarce institutional good. The case for the formation-centered college is stronger than it has been in a generation, not because such institutions are morally superior, but because they are doing what the market still wants and cannot get elsewhere. When the only responsive presence in a student’s intellectual life is a chatbot, the seminar table becomes a scarce good. The “last mile” argument about distinctively human expertise that Hollis Robbins has been making presupposes a first mile, and the first mile is where the conference and the don do their work.
Third, students still want intellectual attention; institutions stopped providing it. They did not suddenly stop wanting human engagement. Many stopped expecting it, because the institutions they paid quietly scaled it away while continuing to market “transformative education” at premium prices. Sancho’s unanswered letter to her principal is the document the academy has to reckon with, not the proctoring software.
The chatbot answered her because the institution would not. That is the indictment and no amount of assessment redesign will lift it.
This is not just a pedagogical problem; it is a public-health one. Nearly two-thirds of American college students report feeling lonely, according to a 2025 Active Minds survey, and the Surgeon General declared loneliness a national epidemic in 2023. Students are not only outsourcing cognition. They are outsourcing intellectual companionship. They did not arrive in this condition by accident. They arrived because the institutions that were supposed to form them stopped showing up.
The obvious objection is that the model I am defending is available to a few thousand students at expensive residential colleges and almost no one else. That objection is correct as a description and wrong as a verdict. The formation-centered model is rare and expensive because institutions chose, over decades, to make it so. The lecture hall replaced the seminar, the learning management system replaced the lecture hall, asynchronous video replaced the live lecture, and now the chatbot is replacing the cognitive work itself. Each step was sold as an efficiency that would preserve the educational core. The chatbot is the terminal point of a trajectory the institutions have been on for forty years. Reversing it requires concrete choices: capping introductory section sizes so faculty can know students by name, restoring faculty advising loads with accountability attached, hiring tenure-track teachers rather than expanding administrative apparatus, and rewarding the labor of formation in tenure decisions the way research output is rewarded. Public institutions did this once, at scale, with state funding that has since been withdrawn. The retreat from the human encounter was a choice. So is its reversal.
AI did not create the crisis of intellectual abandonment in American higher education. It revealed it. The institutions that built the chatbot-shaped vacuum cannot fix it by deploying better detection software. They can only fix it by rebuilding what they spent thirty years dismantling: the relationships between teachers and students that made the credential mean something.
Some institutions still know how to do this. Sarah Lawrence is one of them. There are others -— small colleges, honors programs inside larger universities, individual professors who hold the line on what teaching actually requires. The question is whether the rest of higher education has the institutional honesty to admit what it has been selling, and the will to start selling something else.
The chatbot is not the enemy of the university. It is the mirror. What higher education sees there is its own long retreat from formation, and the bill for that retreat is now coming due.
Norma Sancho’s question is not rhetorical: what kind of system has made a robot feel more attentive than the institution charging her for a degree? Higher education now has to answer it. The chatbot did not create the crisis. It revealed what institutions had already stopped providing - intellectual attention, formation, and human presence. Whether universities are willing to rebuild those things is the only question left.
