MIT had just released the final report from its Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training. I got ahold of the report last night and then did a deep read. Once I read it, I realized this is a big deal!
Here’s why. MIT, the institution that helped build the foundation modern AI sits on, just told its own faculty that the way they teach and assess students needs to be rethought, top to bottom. Not tweaked. Rethought. Every class period. What gets taught, not just how it gets graded.
One paragraph that validates what I have been saying for the past year and a half. The committee was describing what actually becomes valuable once AI can write, code, and summarize competently on demand. A professor watching a student reason through a hard problem. A lab session. An oral defense. An argument at a whiteboard. One student catching a flaw another student missed.
Thinking in the presence of others - thinking out loud - without the aid of AI - this is where I can truly see if a student has mastered the material. I have been saying for some time that the coin of the new realm (in the age of AI) will be things that are distinctly human. And giving an oral defense, seeing students think out loud, is the way to verify what they learned.
And it’s not just this committee. Justin Reich, who directs MIT’s own Teaching Systems Lab, has been saying something similar for a while now: nobody knows how to teach in the age of AI yet, and the way forward isn’t waiting for certainty — it’s actual teachers running real experiments in real classrooms. I’ll be releasing a conversation with Justin soon where he makes that case directly.
We were solving the same problem from opposite ends
I’ve been building the Mastery Viva for a year, and MIT is saying we need to reintroduce the oral defense. MIT’s committee is made up of professors, researchers, and graduate students. People with sabbaticals, teaching assistants, and the freedom to redesign a syllabus over a summer.
I teach high school juniors and seniors. Kids every day, 6 classes. I teach both struggling students and advanced students. I usually eat lunch in my room while I tutor kids. MIT is one of the most selective universities in the world. This is not my reality.
But the interesting thing is that I, a regular classroom teacher, and MIT landed in almost exactly the same place. Their answer to AI eroding written work as proof of learning is oral defense, in-class work, and protecting the human moments a machine can’t fake. My answer, built out of necessity in an actual classroom, is the same three things — I just gave them names: Mastery Vivas, Analog Roots, and AI Engines used on purpose instead of by default.
That’s not a coincidence. When you actually want to know if a student learned something, not just whether they turned something in, you end up in the same room, whether you started at MIT or in a geology classroom in Houston.
Where I’d push further than they did
To their credit, the committee didn’t pretend to have this fully solved. They flagged proctored, in-person exams as the current best fallback. Not because it’s the best possible answer, but because most AI-detection tools are unreliable and feel like surveillance.
I’d agree with that as far as it goes, but I don’t think it’s an either-or. My students still take real tests — multiple choice, worked physics problems, the whole thing — in a locked-down, secure testing environment. I still believe there’s real value in a student sitting quietly and working through a problem, or simply knowing a fact. That part hasn’t gone anywhere.
What’s changed is that a piece of every major test is oral. Not instead of the written test. Alongside it. The written portion tells me whether a student can work a problem and recall what they need to know. The oral portion tells me whether they actually understand it, because they have to explain it to me, live, with no script.
I’m not trading one form of evidence for another. I’m adding a layer that written work alone can’t give me. This is the same layer MIT’s committee was describing when they talked about watching someone reason through a problem in real time.
What this means if you teach kids, not graduate students
Here’s the gap nobody at MIT has to close, and every K-12 teacher does. Their report is right. It’s also written for people with almost none of the constraints you actually live inside.
You don’t get a sabbatical to redesign your assessments. You get August.
You don’t have one section of twelve doctoral students. You might have five sections of thirty.
You don’t set your own curriculum. Someone above you did, and you’re accountable to it whether or not it accounts for any of this.
So if a watershed report from MIT is right, but useless without a plan that survives a real bell schedule, here’s where I’d start:
Pick one assessment this month and add a two-minute oral component to it. Not a redesign of everything you do. One assignment. One conversation per student, even a short one.
Protect one block of in-class time from devices, on purpose, this week. Watch what your students do with their hands and their attention when the option to reach for a phone is gone.
Don’t try to out-detect AI. Their committee said it plainly: the tools are buggy and feel like surveillance. Stop trying to catch AI. Start trying to hear your students think instead.
MIT didn’t build this because it was easy. They built it because the alternative — pretending the last two years haven’t happened — was worse.
I’ll be back in my room tomorrow at lunch, same as always, listening to a kid explain a problem out loud. MIT can call that a finding. I just call it Tuesday.
I appreciate the practical tips! Teaching in K-12 requires you to constantly tweak things rather than attempt an overhaul most of the time.