Fumbling Toward an AI Policy

I’ve seen a common theme among faculty posting about AI: “We’re out on our own here! Why won’t the administration set a clear policy?”

In the abstract, it seems like a compelling question. But getting from “we need a policy” to “we need this policy” is much harder than it sounds.

Part of that is because AI is a rapidly moving target. In just a few years, it has gone from a lab model to a seemingly mandatory future, even as it functions on the ground as a paper-o-matic. It’s moving much more quickly than the policy-development process can, making it likely that any policy would be obsolete by the time it’s enacted. Yet it’s everywhere; just last weekend, my younger niece, who is halfway through high school, informed me confidently that “everyone uses it.”

Some employers seem to want AI fluency, though they don’t seem to know quite what that means, and now some firms are desperate enough for students who haven’t outsourced their thinking that they’re hiring humanities majors and training them on tech. It’s a return to a much older model—and, frankly, a welcome one.

A policy of a Great Refusal would hold real appeal to instructors who are (rightly!) tired of reading slop passed off as actual student work. But it would put students at a possible competitive disadvantage when they’re up against others who have done compelling work with it. It also presumes that AI can only be used for bad ends. Given how quickly it’s evolving, I’d be reluctant to declare that it could only ever do harm, even as I see tremendous harm currently being done. That would effectively preclude innovation. (I’ve heard of instructors having students proofread slop as an editing exercise; there’s something to that.) And it would force instructors to enforce rules that are so widely ignored as to call the legitimacy of enforcement into question. When culture and rules grow too far apart, culture wins.

The environmental objection to AI is real at this point, though I don’t know if it always will be. I could imagine data centers becoming much more efficient over time, at which point the environmental objection would be obsolete. If the real objection is to slop, then object to slop.

(Personally, I object to AI’s co-optation of the em dash. The em dash is a perfectly honorable writing tool. As Elle Cordova memorably put it, just because it got caught in ChatGPT’s teeth doesn’t make it bad.)

A Great Refusal would create serious issues in computer science classes and, increasingly, in allied health programs. As it becomes part of the practice of fields for which we’re preparing students, leaving it out would leave a gap in their preparation. It could even put ethics classes in a tough spot, since students would be asked to pass judgment on something they couldn’t try themselves.

A policy of “you can use it for suggestions and structure, but not for content” is likely much too nuanced and subtle to be enforceable.

A policy of laissez-faire is effectively no policy at all.

A policy of “it’s the future, so we encourage students to use it” effectively surrenders to the paper-o-matic. An old joke compares artificial intelligence to natural stupidity; my concern is that the former leaves the latter undisturbed. The point of writing, say, a five-page paper analyzing a marketing campaign, social movement or novel is not the paper itself; it’s the process of creating the paper. That’s where the learning happens. If the process is reduced to a prompt, that’s where the learning stops. It’s the difference between running a marathon and driving one. Yes, driving one is faster, but it doesn’t get you into shape. Letting marathoners drive cars defeats the purpose of a marathon.

A policy of letting each academic department set its own policy has the advantage of potentially distinguishing computer programming classes from sociology classes. But it also leads to incoherence on the ground, and a potential race to the bottom as departments compete for enrollments. And for very small departments, it’s not much different from no policy at all.

All of these assume that what AI looks like won’t change meaningfully for years. Given what we’ve seen over the last few years, that’s a bad bet.

Right now, plenty of institutions are taking the view that it’s better to wait for the dust to settle before making sweeping declarations. To someone exasperated by the paper-o-matic, I can understand why that might look like capitulation. But filling in the blanks of an actual policy—when it’s ubiquitous on the ground and the mechanisms are changing we speak—is somewhere between aspirational and hubristic. And even as I say that, I join the refusers in declaring that you can pry my em dashes from my cold, dead hands.