American colleges begin a new school year in a continuing crisis mode that began with the introduction of generative artificial intelligence tools four years ago.
Blue book sales have risen nearly 40 percent as professors return to timed, handwritten exams. The University of Chicago has banned AI use in undergraduate social sciences core courses and banned screens outright in law school classrooms. Berkeley’s law school forbids AI use on any work submitted for credit. Cornell just wrapped up a weeklong “unplugged” initiative encouraging students and faculty to put down their devices.
At the same time, Harvard College’s dean, David Deming, urged the opposite: “acceptance, or even encouragement” of AI use in assignments completed outside the classroom. Dartmouth’s president, Sian Leah Beilock, made a similar case in The Atlantic, warning that colleges risk “consigning themselves to irrelevance” if they don’t lean into AI technology.
The sense of confusion and crisis extends beyond the classroom. Just last week, Anthropic CEO Dario Amodei published an essay calling for AI companies to slow down just days after one of his own researchers resigned publicly and warned that the people building this technology “earnestly believe it could kill us all by the end of the decade.” Some see the warning as genuine alarm while others suspect a marketing gimmick.
Despite attempts at control, calls for acceptance, and fears for the future, the fight playing out in classrooms this fall may be aimed at the wrong target.
Deming’s message to Harvard’s faculty and incoming students argued that policing AI use was damaging trust between students and instructors, and that outright bans are unenforceable anyway. He described a “barbell” approach: encourage AI where it can “deepen learning,” and “AI-proof” the assessments meant to demonstrate real mastery.
Philosopher Anastasia Berg, writing in The Chronicle Review, pointed out that Deming’s barbell targets the humanities when he described “writing-intensive and/or project-based” classes with “papers too meaty to fit within a proctored exam window” as ripe for AI encouragement. Berg wrote, “There is no use of AI that can ‘deepen’ students’ learning of Plato, Machiavelli, Nietzsche, Marx, and Du Bois over and above an unmediated engagement with the original works.”
Deming and Berg offer two policy responses to AI in the classroom. But neither addresses the underlying question: what do we measure when we grade student work?
Education critic John Taylor Gatto argued that in the twentieth century industrial-scale classrooms replaced the mentoring model of teaching and learning. While some historical details are contested, standardized assessment tools have clearly become proxies for measuring how students engage with ideas. Learning has come to be defined by inputs like credit hours and seat time. Output measures include ability to correctly answer multiple-choice questions and write short essays that can be graded using rubrics.
Doug Lederman, editor of Inside Higher Ed, traces recent aspects of this history in a series of columns. Serious efforts to measure college-level learning go back at least to a 1984 federal study group that pushed colleges to develop new ways to assess learning. The Spellings Commission tried again two decades later. So did the Collegiate Learning Assessment, and the American Association of Colleges and Universities. None of it stuck.
Meanwhile, the average college GPA has climbed from roughly 2.7 in the late 1980s to about 3.15 today at the same time national measures like the ACT and NAEP have declined. That gap didn’t open because of AI. Our tools for measuring learning were already broken.
The battle for the role of AI in higher education this fall is being fought over when and where AI and other digital technologies should be used. This combat ignores the larger crisis: how can colleges verify that they are preparing students to become productive members of society in a digital age?
MIT’s Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training released a report this summer that does attempt to address larger issues. The report argues that educators should, “begin by defining the purpose of the learning experience itself: what students should come to know, be able to do, and learn to value. Assignments and assessments can then be designed to measure and encourage those outcomes, using AI if it’s helpful but not if it isn’t.” The report provides a short menu of policy options and encourages instructors to find ways to measure learning based on what they’re trying to teach.
Christine Nowik, an English professor at Harrisburg Area Community College, has been making a similar case from a very different vantage point: a 5/5 teaching load, one-person departments, none of the institutional slack a place like MIT takes for granted. She urges teachers to stop asking about AI detection and compliance and start asking what they want students to learn.
Humanities professors like Berg are sometimes painted as out-of-touch liberals when they argue for protecting unmediated engagement with hard texts. In a Washington Post opinion piece, Joshua Katz and Solveig Lucia Gold, senior fellows at the conservative American Enterprise Institute and the American Council of Trustees and Alumni, make a compatible argument from the right. Four years ago, majoring in computer science seemed like the fast track to job security. AI has now scrambled any reliable link between a major and a career. Students should pursue real depth and range because nobody can predict what will matter in the future.
When thinkers who agree on almost nothing else start converging on the idea that AI offers a chance to rethink the value of learning, that’s a stronger signal than either voice alone. Perhaps the industrial model of assessment has stopped working.
The argument over whether AI policy is too permissive or not permissive enough is a distraction from the much harder and more interesting question: four years into the AI era, what does an educated person need to know and how can we ascertain what students have actually learned?
Until somebody answers that question, arguing about blue books, digital devices, and AI detectors is just a skirmish in the real crisis of teaching and learning.
Discover more from Post Alley
Subscribe to get the latest posts sent to your email.
