Off-Label

A few years ago my doctor prescribed me something for a purpose it was never actually approved to treat. She mentioned it almost in passing: “This is a little off-label, but there’s good reason to think it’ll help.” The phrase stuck with me. As it happens, somewhere between a fifth and a third of prescriptions written in this country are off-label, which means the drug is being used for a condition, a dose, or a patient the FDA never formally blessed. That isn’t malpractice. That’s just medicine. Physicians do it all the time, and mostly they do it well, because the alternative, waiting for a definitive trial on every possible use, would mean letting people stay sick while the evidence catches up.

I keep coming back to that phrase this fall, because “off-label” is the most honest description I can find for what every college in the country is currently doing with AI.

Consider the gap. This week Inside Higher Ed ran a piece with a headline I haven’t been able to shake: the research on AI’s benefits for education is, in the words of MIT’s Justin Reich, “almost nonexistent.” Not thin. Not mixed. Nonexistent. Meanwhile the California State University system has spent seventeen million dollars putting ChatGPT in front of its students, more than seven hundred thousand AI licenses have been sold across higher education, and the Digital Education Council’s latest global survey (forty-five thousand respondents, so not a rounding error) found that eighty-eight percent of students are already using AI to learn. We have prescribed this drug to nearly everyone. We are still waiting on the trial.

Here is the part that should make us squirm a little. We are, professionally, the most evidence-obsessed people alive. We make undergraduates cite a source for every claim. We won’t let a colleague pilot a new gen-ed requirement without an assessment plan and a committee and two years of data. We require review-board approval to so much as survey our own students about their sleep. And then we hand those same students a technology with, and I want to be precise here, no rigorous evidence that it improves learning, and in some studies (the methodologically sound ones, according to Ivy Tech’s Patrick O’Neill) evidence that using it without guidance actively erodes the critical thinking we were hired to build. We set a bar for everyone in the building except the thing we spent the most money on. All right, maybe not the most money. But more time, attention, and effort than any disruption before it.

So the tempting move, the one I see a lot of very smart colleagues making, is to conclude that we should not be doing this at all. Pull the licenses. Wait for the science. Come back when there’s a Cochrane review.

I don’t think that’s right, and the off-label metaphor is exactly why. Doctors don’t refuse to prescribe off-label. They can’t; the evidence never arrives fast enough, especially now, when the models change (as Stacey Alicea of the Research Partnership for Professional Learning points out) every one to three months, which is to say faster than any study can be designed, funded, run, and published. Even in medicine, most off-label prescribing runs ahead of solid evidence, and we still don’t call it malpractice, because what makes it responsible is the discipline around it, not the certainty underneath. If we insist on waiting for the definitive trial, we will be waiting forever, and our students will spend those years learning to use these tools from a YouTuber instead of from us.

What doctors do instead is prescribe off-label responsibly, and responsible off-label practice is a real discipline with real rules. It’s worth stealing all of them.

Off-label prescribing requires informed consent: you tell the patient this use is experimental. The classroom version is telling students, plainly, that we are figuring this out together, that the tool is unproven, and that “the AI said so” is not yet a thing anyone on Earth can stand behind.

In my teaching, I don’t require students to buy textbooks and use only OER and library-accessible materials because data and experience told me that a $50 textbook can be a big deal for my students. I teach primarily online because my students are working professionals and this is the modality that students selected repeatedly over the in-person alternatives. I have often had my teaching career treated as off-label, with some variation of “online learning isn’t as good as in-person” quietly creating a sub-class of both learners and credentials. The research has been clear for over a decade: what matters is the preparation of the instructor and the design of the learning experience, not the modality. Online can be as good or better, when it’s built well. Essentially, good teaching happens everywhere, and there are great students everywhere too.

As an online teacher, my job is to set students up for a world where AI is both a real tool for their professional lives and a standing temptation to swap the learning process for an AI-generated learning product.

Off-label prescribing requires monitoring: you watch the patient, you don’t just write the script and walk away. The institutional version is local evidence. We do not have to wait for MIT to run the national trial. Every one of us is sitting on a data-generating machine called a course, and the honest thing to do is treat this semester as the study: measure what actually happens to student work, to student learning, to the roughly one in five who now say the work is harder without the tool, and be willing to report results we don’t like.

And I don’t blame students for being tempted by distractions or by the allure of AI when their resources are feeling thin, their imposter syndrome is too loud, or they just don’t have it in themselves to hammer out another discussion board post. I hold them accountable when I encounter what I consider to be misuse of any tech or tool, of course, but that doesn’t mean I can’t relate. And as a thought experiment for you department chairs and directors out there: ever been tempted to ban laptops and phones from your staff and faculty meetings?

And off-label prescribing requires a willingness to stop. If the drug isn’t working, you take the patient off it. The scariest sentence in all of this is the one no procurement office wants to say out loud: we might have to unwind a seventeen-million-dollar decision. Except that isn’t a hypothetical. Some of Cal State’s own faculty and students are already refusing to use the tool their system bought for them, and there’s an open petition to cancel the contract. The system’s answer, so far, has been to renew it. Responsible practice means building the exit ramp before you need it, rather than defending the purchase because it was expensive. “We already spent the money” is a sunk cost, not a study.

None of this is abstinence and none of it is cheerleading. It’s the boring middle path that actual professionals take every day when they have to act before the evidence is in, which, if we’re honest, is most of the time. Teaching has always been a little off-label. We adopt, we watch, we adjust, we occasionally admit we were wrong. The only thing that’s new is the price tag and the pace.

So I’d retire two questions this fall. “Is AI good for learning?” is unanswerable right now, and pretending otherwise, in either direction, is the actual malpractice. “Should we be using it at all?” is already moot; eighty-eight percent of our students answered it for us. The question worth asking is the one every good doctor asks over an off-label script: are we doing this carefully? And on the whole, I bet we are. But the response is piecemeal, inconsistent, and chaotic while being discussed monolithically. Used poorly, students will get very little out of the learning experience, but implemented carefully, students can show you their work and allow us to be more process over product.

← All essays · Also published on Substack.