Read the January charge and the August report side by side and you notice something that most coverage skipped past: MIT asked for a policy and got a confession.
The charge was crisp. Three bullets, handed down by the Chancellor, the Provost, and the Chair of the Faculty: assess current AI use, identify innovations in teaching and assessment, propose an AI use policy. A tractable brief with a deliverable at the end.
The committee fulfilled it. And then, early in the report, it set the brief aside: "what we learned as a group quickly convinced us that the Institute community, particularly the faculty, must tackle a set of deeper questions about the structure, meaning, and value of an MIT education."
That is not scope creep. That is an institution discovering, in real time, that the policy it commissioned cannot be written until someone settles what it is protecting.
That is not scope creep. That is an institution discovering, in real time, that the policy it commissioned cannot be written until someone settles what it is protecting.
Last week I left an open thread: is any school willing to treat comprehension — the thing AI most quietly erodes — as explicit curriculum rather than as a compliance problem? This week MIT answers, and the answer is instructive precisely because it is not a fix. It is a call to action. The distinction is the whole story.
Start with what the committee could actually produce, because it is real and it is useful.
There is a four-option syllabus menu, complete with traffic-light icons: Unrestricted, Limited (support tool only), Required, and Strictly Prohibited. Every class picks one, states it, and explains the rationale. This is good hygiene. It is reportedly aimed at ending the confusion students reported about inconsistent, unjustified rules across the curriculum.
There is, reportedly per the FAQ, the AI-Light and AI-Heavy pathway idea — the suggestion that departments offer rigorous variants so students can move through a curriculum on different terms with the technology.
And there is the most concrete remedy of all: every subject "should include a regular in-person social component." Not lecture. Structured collaboration — group problem-solving with a TA, graded participation, feedback discussions built around a rubric.
Now set that remedy against the wound it is meant to close. In under three years, reportedly per the FAQ, AI technologies have driven major shifts including "decreased attendance at office hours, reduced participation in online discussions, and... a drop in in-person study groups in dorms, libraries, and other study spaces" — the last of which the FAQ notes was heard anecdotally.
The p-set, the study group, the office hour — the tacit machinery of an MIT education — thinned out in thirty-six months. The committee's answer is to mandate, subject by subject, that students be put back in a room together.
That is a genuine intervention. It is also, read honestly, an admission that the organic version of the thing died faster than the institution could notice, and that the fastest available replacement is a scheduling requirement.
Now hold the deliverables the committee could produce against the ones it could not.
It could specify the menu. It could mandate the social component. It could not answer whether an MIT education still means what it meant three years ago. That question — "the structure, meaning, and value of an MIT education" — it hands to "the faculty" and "the community."
The committee is explicit that these are the deeper questions, and explicit that they are not its to resolve. The committee — reportedly faculty from every school, students, the head of Libraries, five months of weekly meetings — and the load-bearing conclusion is that someone else, later, must decide what the degree is for.
The reflexive reading is that this is a failure of nerve. A blue-ribbon committee flinching from the one question that mattered.
I want to argue the opposite. The deferral is structurally correct, and the reason it is correct is worth naming carefully.
There is a formalization, reportedly set out in a single independent-researcher preprint, called the generation-governance impedance mismatch. Its claim, in the software context, is this: "implementation generation now occurs at machine speed, while behavioral governance remains constrained by institutional deliberation." The two are not just different speeds. They are "orthogonal in function." Generation produces artifacts. Governance establishes what is permissible.
Import that frame to a university and the committee's predicament snaps into focus.
The generation side — students using AI to produce p-sets, essays, code, the erosion of the study group — moves at the speed of a model release. The governance side — deciding what a residential education is for and rewriting the curriculum accordingly — moves at the speed of faculty deliberation. Committees, votes, semesters.
These are not the same clock, and no amount of urgency reconciles them. A deliberative body cannot author meaning faster than the machine erodes the thing it deliberates about. That is not the committee being slow. That is the structure of the problem.
Which is why the honest move is buried exactly where it is. The committee could produce the fast, mechanical layer — the menu, the mandate, the pathways — because those operate at the implementation layer, where you can move quickly. It could not produce the slow, definitional layer at machine speed, because that layer is orthogonal to speed by nature. So it named the question and passed it up the chain to the only body authorized to answer it.
Read that way, the deferral is the report's most truthful sentence. The committee is telling you which of its problems are tractable at the speed events demand and which are not — and refusing to fake the second kind.
But truthfulness has a cost, and the cost surfaces when you look at the surrounding language.
President Kornbluth reportedly calls this "a watershed moment." The report insists "this is not an optional exercise" and "there is a lot to do." The committee lists eight guiding principles, the first two of which are reportedly "be humble" and "be bold."
Notice what those are. They are values, not rules. "Be humble" adjudicates no case. "Be bold" selects no menu option. They are the vocabulary an institution reaches for when it knows it must respond and knows it cannot yet decide.
And notice the concession folded into the landscape section, almost in passing: other institutions are grappling with the same questions, "but no one seems to have it all figured out." A watershed, an urgency framed as non-optional, and an open admission that nobody — not even the school "deeply identified with the birth of AI" — has the answer.
Put it together and the uncomfortable reading emerges. Five intense months of the most concentrated deliberative effort MIT could mount, and the output that matters most is a set of questions and a set of values. The institution's fastest available response is already the slow variable.
The "responsibility to lead" language — the claim that MIT is "uniquely positioned to lead higher education in this high-speed, high-stakes adaptation" — reads differently in that light. Leadership implies you are ahead of the thing, setting its direction. But a committee that spends its authority naming values and forwarding questions is not ahead of the change. It is downstream of it, describing the water it is already standing in.
That is what the call-to-action framing conceals and reveals at once. A call to action is what you issue when you cannot yet issue instructions. It converts the gap between machine speed and deliberative speed into a rhetorical posture — energetic, principled, urgent — and buys time by naming values instead of setting rules.
None of this makes the report a bad document. It is, if anything, an unusually honest one. It did the fast work it could do, and it declined to fake the slow work it couldn't. The committee, one might argue, saw the impedance mismatch clearly and reported it faithfully — though the framing of that mismatch is mine, not language the committee itself used.
The trouble is not with the committee. The trouble is that "faithfully reporting the mismatch" is itself the finding. When the strongest institutional response available is a confession that the deeper questions remain unanswered and must be answered by someone else, later, the institution has told you where it stands relative to the technology.
Not leading it. Answering it. And answering, by its own admission, on a clock it does not control.
So the open thread I left last week closes with a harder question than the one I opened. It is no longer whether a school will treat learning's erosion as explicit curriculum. MIT plainly will — the mandate for in-person social components is exactly that, curriculum as ballast against drift.
The question is whether the definitional work — what the degree protects, what it is for — can be done at all on a timeline that matters, by a body that must remain human-authorized while the erosion runs at machine speed. MIT's committee looked at that question, and its most honest act was to hand it back.