MIT Says AI Can Finish Most Undergrad Assignments. Now It Wants to Reinvent the Classroom.

MIT's Ad Hoc Committee on AI in Teaching, Learning, and Research Training published a final report on August 13 with a finding that has no real precedent at a major research university: generative AI "can produce credible solutions and provide reasonable responses to almost any written assignment in our undergraduate curriculum, including essays, math and science problems, proofs, and coding assignments."

That sentence is not about cheating. The committee, co-chaired by professors Eric Klopfer and Samuel Madden, treats it as a definitional change to what an undergraduate degree certifies. If the unsupervised written artifact no longer holds as evidence of the student, something else has to do that work. The report's answer is to put the work where the student is, in the room, in real time, watched.

The report calls the moment a "watershed" and rejects "patches and duct tape." The accompanying letter from MIT's leadership makes the same point in less academic language. The eight principles that open the report, including "Be humble" and "Be bold," are less a curriculum than a posture. Stop defending the homework. Start defending the credential.

That posture shows up in the pedagogical menu the report names. Oral exams. Hand-written essays. In-class discussion. Commonplace-book note-taking, the old practice of recording what a student actually read and thought, by hand, in their own book. More hands-on assignments. The committee is explicit about why: AI detection software is, in its words, "quite unreliable." Policing a student's process is treated as a cost the institution should not keep paying. Better to redesign the assignment so the artifact is not the point.

The cultural case is harder. The report observes that in less than three years, measured from the late-2022 release of ChatGPT, office-hours attendance has dropped, online discussion participation has thinned, and study groups in dorms and libraries have thinned with them. None of this is measured the way a grade is measured, and the report does not pretend it is. It treats the in-person, supervised, low-stakes practice of being a student as the thing the degree is supposed to certify, and warns that piece is eroding.

MIT is not the only place making this argument out loud. The University of Chicago Law School has rolled out what it calls an "AI strategy" that bans phones and laptops in first-year courses, on the theory that the first-year classroom is where a lawyer learns to think on their feet and that the device on the desk is now a different kind of risk. Princeton dropped its century-old Honor Code after an AI cheating scandal, a more aggressive move that concedes the old unsupervised compact no longer holds. Both moves were reported alongside the MIT findings. Both institutions are betting on the same thing MIT is: that the supervised, in-room performance is the artifact worth certifying.

The trade-offs are real and the report does not hide them. Oral exams grade real-time reasoning but can be biased and are harder to make accessible. Hand-written work is harder to outsource but is not the same signal as a defended essay. Banning laptops protects discussion but presumes a classroom of discussants. Commonplace books reward a particular kind of student. None of these choices are free. The report's argument is that pretending the homework is still the homework is no longer free either.

The next test is whether MIT actually moves. The report is a posture, not a policy. Course-by-course changes will land in departments, on syllabi, and in the next round of teaching assignments. The wager is that what an undergraduate degree certifies is no longer the unsupervised written artifact. It can be the supervised performance, if the institution is willing to build around that.