Brad DeLong's Grasping Reality cross-posted a post from Counting Atoms

Brad DeLongSep 16 · Brad DeLong's Grasping Reality

Chad Orzel aptly dubs “AI” a “roiling cauldron of linear algebra,” & then argues that stochastic parrots doing bag‑of‑words prediction raise real issues—but the hard look should target the mismatch between what higher-education students want and what they get. In physics courses, the “infinitely easy to cheat” problem has existed for a generation. Teachers have adapted. Non‑STEM instructors, IMHO, should make sure students know a lot is riding on their becoming become good front‑end notes in the Anthology Super‑Intelligence of the Collective Human Mind—learning grammar, rhetoric, logic, research, & orienting knowledge—& handle grade‑motivational clean‑up via oral exams on students’ written papers. Straightforward, IMHO, for all except the absurdly entitled:

The Problem of the Bag of Words Is Not Evenly Distributed

The solution won't be, either

Chad Orzel

Sep 16

READ IN APP

black and white robot toy on red wooden table — Photo by Andrea De Santis on Unsplash

The theme of the 2025-26 academic year is clearly “Fretting About the Bag of Words,” as I can’t seem to go more than a couple of days without some new thinkpiece or faculty-email-list discussion of the roiling cauldron of linear algebra currently being marketed as “AI.” This one from Tyler Austin Harper (late of Bates College, now of The Atlantic) is a real doozy, though. After the usual anecdata about the need for radical measures to prevent students from cheating with chatbots, he really dives into the “radical” end of the pool:

At the type of place where I taught until recently—a small, selective, private liberal-arts college—administrators can go quite far in limiting AI use, if they have the guts to do so. They should commit to a ruthless de-teching not just of classrooms but of their entire institution. Get rid of Wi-Fi and return to Ethernet, which would allow schools greater control over where and when students use digital technologies. To that end, smartphones and laptops should also be banned on campus. If students want to type notes in class or papers in the library, they can use digital typewriters, which have word processing but nothing else. Work and research requiring students to use the internet or a computer can take place in designated labs. This lab-based computer work can and should include learning to use AI, a technology that is likely here to stay and about which ignorance represents neither wisdom nor virtue.[…]
Colleges that are especially committed to maintaining this tech-free environment could require students to live on campus, so they can’t use AI tools at home undetected. Many schools, including those with a high number of students who have children or other familial responsibilities, might not be able to do this. But some could, and they should. (And they should of course provide whatever financial aid is necessary to enable students to live in the dorms.)

You think you have the will to fight “AI,” he’ll show you what will really is:

It shouldn’t be a surprise that I find this proposal overwrought, given that I’m on record as saying that I think this problem isn’t that different than what we’ve been dealing with in STEM for decades. And that goes to my core issue with a lot of the discussions of “AI” (and a huge range of other academic policy issues).

That is, I think that most problems in academia, particularly those relating to teaching and learning, are fundamentally local. The process of education involves an individual professor and the specific collection of students who happen to be in their class(es), and the relationship between those is something that has to be negotiated among that group¹. Individual students and individual faculty will differ in their tolerance for various behaviors— class attendance and participation, deadlines and late penalties, etc.— and what works well in one context may fail miserably in another.

For a relatively non-controversial example of what I mean (an example I repeat a lot, so sorry if you’ve seen it before), in my first year teaching at Union I asked one of my senior colleagues for advice, and he said that in order to make the most of our small classes he made it a point to “break the fourth wall” and move out into the middle of the room while lecturing. That sounded good to me, so I tried it for a while, but after a week or two I noticed that whenever I moved out into the room, the students I was closest to were visible uncomfortable and would sort of cower away from me.

The explanation here is that my colleague is a bit below average height for an American man, while I am 6’6” and at that time weighed around 280lbs². When he walked out among the students, it was non-threatening and inclusive, but when I did the same thing, they were terrified that I might fall on them. I went back to lecturing from a spot closer to the board, and everybody was happier.

My problem with the Butlerian Jihad approach to AI is that it’s a call to impose a global solution on a local problem. If you want to insist on hand-written blue-book exams in your own classes, that’s great— Godspeed. If you want to mandate that I have to give blue-book exams in my classes, though, we’re going to have a problem. And that’s even before getting to things like banning wi-fi— if we’re going that far, let’s also turn off the water so students can’t ask to go to the bathroom lest they cheat in there.

The concept of “academic freedom” is frequently invoked in discussions of higher education, and is somewhat subject to abuse in the form of stretching the definition³. It seems to me, though, that one of the core elements that are indisputably central to academic freedom as a norm is that individual faculty control their individual classes⁴. As long as they’re not violating actual laws or standards of professional ethics, faculty should have wide discretion to set and enforce whatever policies they think best regarding assignments, attendance, and appropriate use of technology in class.

Again and again, though, we have faculty responding to issues that bother them by calling for global policies that will be a huge pain in the ass for everybody else— we need a campus-wide ban on laptops in class, or a strict policy forbidding food and drink, or any of a host of other things that seem to me to be clearly within the purview of individual instructors. Frustratingly, this often comes from people who are quick to cite “academic freedom” in response to anything impinging on something they like to do⁵. A college-wide ban on modern technology is into “destroy the village in order to save it” territory, when it comes to academic freedom.

The threat of “AI” systems is not equally menacing to every class— as noted above, I don’t think it’s qualitatively different for many STEM classes than the decades-old problem of solution sets existing out “in the wild” on the Internet. Accordingly, the measures required to deal with it will be different in different courses. And that, in turn, means that the policies and precautions needed to regulate the use of “AI” are solidly in the realm of things that individual faculty should be free to adjust to best suit their particular class and their particular preferences.

I do agree that the rise of bag-of-words systems poses some deep and significant issues that will require taking a hard look at what it is that we do on a fundamental level. But I don’t agree at all with the idea of trying to hold back the technology via scorched-earth methods. That’s not the answer. If anything, it’s a desperate attempt to avoid confronting the real issue, which is the mismatch between what students want and what institutions of higher education are currently providing.

Kind of teasing a different piece that I don’t currently have the time to write, I guess, so here’s a button if you want to see whether I ever manage it:

And if you want to argue that a ban on thinking machines doesn’t go far enough and needs to be expanded to cover Mentat disciplines as well, the comments will be open: