AI in a World of Discipline

AI is changing things fast, and no more so than in education. Should we be worried?

When ChatGPT first came on the scene, it was immediately obvious that this was an immense technological advance. To test its capabilities back then, I fed the AI some questions from a very tricky (students would say, rather, deeply unfair and outright nefarious) exam I used to give in my MBA course on firm theory. The results were impressive. The AI aced the exam and was able to summon up accurate reasoning to explain its answers. Clearly, something big was happening.

Two years later, how do things stand? I now do all my teaching in the undergraduate classroom, so I don’t have to worry about the AI affecting my MBA exam results. In fact, I don’t use examinations of any kind in my undergrad courses, since, as I see it, they are incompatible with the learning objectives of my courses, which, it is important to stress, are all in the humanities. (STEM classes are a different species altogether when it comes to such things. And as an aside: why is most AI anxiety coming from the humanities? Where are the biochem professors worried about AI-generated exam answers? A quick search suggests they’re...fine? I’d have thought they’d be just as susceptible to the problem of AI, but anyway….) Instead, I ask my students to write essays. A lot of them. A typical student will, over the course of the semester, produce between 10-15k words for my course. And some end up writing much more.

Now, AI tools are very good at taking exams. But they are even better at writing essays. There is no doubt that AI can produce an intelligent, well-informed, and coherent paper on any topic I can dream up. In fact, AI can probably come up with as good or better topics than I can ( something I haven’t yet tested, being rather chary about the result).

And so, like almost everyone else teaching in higher education, I am seeing firsthand the effects of these tools in my classroom, and it would be naïve to think that my courses are not vulnerable to the impact of AI. In fact, my courses are especially vulnerable because of the focus on extensive written work as the basis for grading student performance.

For the first year of our new AI reality, I was (perhaps a bit willfully) naïve. I figured I had a pretty good sense of when an essay had been produced by AI, since there are many telltale signs, and I just wasn’t seeing them in the overwhelming majority of the submitted work. Yes, every so often someone would submit an essay that was clearly AI-generated, but this was at most 5% (maybe 10% on a bad day). And, yes, perhaps I missed one or two along the way, but an advantage of requiring a lot of written work is that it creates multiple opportunities for detection. So I figured that my course structure offered sufficient built-in protection from wanton opportunism such that I did not need to scramble my entire classroom philosophy to solve the problem of a few bad apples.

But straight out cheating is not, in fact, where AI is happening – and I am becoming less naïve now. Starting this academic year, I have introduced a very modest guardrail into my course: I require my students to append a statement to their essays, indicating whether they used any AI tools to help them and, if so, how. No one is penalized for using AI (unless they straight up have the AI write their essay for them), they just have to be honest about it; about half acknowledge the use of AI in their work.

That is a lot more than I thought were relying on these tools in Year 1, and perhaps uptake has increased from year to year, as with any technology. But it is consistent with reports about undergraduate usage elsewhere, so obviously there was a lot more AI happening in the background that I wasn’t seeing. While I was generally catching out those students who had the AI generate their work for them, most people are using the tools, let’s say, collaboratively. They are writing (most of) what they submit, but AI is offering a major assist. Here’s a typical statement I now get from students.

I acknowledge I used AI to assist me in crafting my essay in the following ways:

- To create an outline for this essay

- To give me a summary and the key ideas the book mentions

As an example of what that looks like, I hopped over to ChatGPT, gave it one of my essay prompts, and then asked just a couple of questions about the key ideas. The result? A well-informed and detailed outline of an essay, which required only minimal input on my part. I would still have to write it up, but the difficult part was not done by me, and my knowledge of the text would not have to be very deep at all. To turn it into essay format, I have been given more or less pinpoint instructions on how to proceed. (You can see it here.)

Of course, with this primer in hand, I could then return to the texts and read through the material closely to come up with my own thoughts. And to be clear, some students do, in fact, just that. But I could also just take what the AI has generated and fill it out. And some students are doing that, too.

So, is this an acceptable use of AI?

With more nostalgia than reluctance, I state the answer is straightforward and inevitable: it has to be. Because this is what AI does extremely well, and this is just one of the thousand and one ways that my 18 and 19-year old students will be incorporating AI into their world for the rest of their lives at all levels.

AI is an incredibly powerful adjunct to our cognitive landscape. So when you are facing a cognitive task, why wouldn’t you use it? AI is not a calculator, helping people perform a mechanical task. It is a tutor, a scholar, a library, an assemblage of knowledge so vast it far surpasses what any one person could ever hope to know. It can guide us to conclusions that we may not have perceived, and it can draw inferences and make comparisons that are genuinely insightful, because it is drawing from a compendium of human knowledge that is itself the source of those insights. AI is the modern equivalent of an Ancient travelling to Alexandria and confronting, in the Great Library housed there, all the knowledge of the world in one vast repository – revelation just one scroll – or one prompt – away.

In short, if AI can help a student write a strong, insightful, thoughtful essay, why shouldn’t she use it?

Our educational answer to that is because she is relying on the AI to do work that she herself is supposed to do: to read and think for herself, to work through concepts and find points of connection and comparison. In other words, if you shift the cognitive work over to the AI, you do so at the serious risk of your own cognitive development. The classroom as a site for developing creative and critical thinking, better reasoning skills, and more articulate and sophisticated argument, all those things are obviated to a significant degree if it is the AI, and not you, that is providing the intellectual and creative heft in your own work.

But is it cheating?

In other words, does someone who acknowledges they used AI to summarize texts, isolate key ideas, and structure their essay for them – have they cheated? Are they getting an unfair advantage over someone who decides to forego such AI tools. And going forward, how many will remain willing actually to forego them? A 2024 survey of Harvard undergrads found already that “35% of students are worried their peers will use generative AI to gain an unfair academic advantage in class.” And is it unreasonable that students will increasingly feel essentially compelled to use AI for their coursework precisely because of the advantages that it brings, and hence the fear of underperformance if they don’t?

This is the divide that is happening now in education, an AI curtain has descended that separates those who see its use as a form of cheating, and those who see it as an inevitable tool that must be incorporated into the future of knowledge, and thus into the future of education. If you are in the former camp, then the result will be mistrust, suspicion, and surveillance of your students and the creation of AI-free zones (proctored exams written longhand, e.g.) to thwart the miscreant. If you are in the latter, you are likely somewhat overwhelmed by the need for a total rethink of how we teach, and even what teaching, what higher-education is designed to achieve. (There is a third group: people who are retiring soon, and are thus grateful they won’t have to pick a side. Looks in mirror.)

But this divide is, I think, rather a false one. Because it is not really about the philosophy we bring to our teaching. Rather, it is about how we evaluate our students capacity to learn what we are teaching. And largely overlooked in this debate is the role played by grades, considered in the abstract.

Grades are, essentially, a disciplinary mechanism. (Not just grades, in fact, all types of review structures, like employee performance.) They create norms and standards, reward good behaviour and punish bad behaviour. And while grades can create positive incentive, they also cause a lot of fear and anxiety. The grade is a system of classification that divides between success and failure, highlighting worth, value, and competence from their opposites by use of a number or a letter. The widespread phenomenon of grade inflation is not so much a failure of modern academic rigour as it is an implicit acknowledgement of the underlying problematic, even inadequacy, of grades themselves. Foucault, in his book Discipline and Punish, noted the characteristics of what he called disciplinary power and the way it operates:

The art of punishing, in the regime of disciplinary power, is aimed neither at expiation, nor even precisely at repression. It brings five quite distinct operations into play: (1) it refers individual actions to a whole that is at once a field of comparison, a space of differentiation and the principle of a rule to be followed. (2) It differentiates individuals from one another, in terms of the following overall rule: that the rule be made to function as a minimal threshold, as an average to be respected or as an optimum towards which one must move. (3) It measures in quantitative terms and hierarchizes in terms of value the abilities, the level, the 'nature' of individuals. (4) It introduces, through this 'value-giving' measure, the constraint of a conformity that must be achieved. (5) Lastly, it traces the limit that will define difference in relation to all other differences, the external frontier of the abnormal [Which in the case of grades is outright failure].

And note his conclusion:

The perpetual penalty that traverses all points and supervises every instant in the disciplinary institutions compares, differentiates, hierarchizes, homogenizes, excludes. In short, it normalizes.

Grades are a literal number (or letter) that attach to your performance. They define your value; are often indelibly attached to your sense of self-worth; and are perceived as a defining feature of the kind of life you will get to lead. Grades do more than evaluate — they normalize. They compare, differentiate, rank, and exclude. No wonder they scare the shit out of people. A lot of people who score badly on exams do so, not because they lack knowledge, haven’t studied, or are underprepared (although that’s certainly a good deal). No, they do poorly because an exam is a terrifying device, and they shut down psychologically precisely because we have put so much emphasis on the disciplinary value of the result.

As professors in a classroom, we like to think of ourselves as educators. But in fact, we are disciplinarians, because we are at the frontline of the disciplinary society. More than to educate, our primary social function is to assign grades. We may not see it that way for ourselves, but our students definitely see it that way (or at least a large majority of them do). And their perception is our reality, whatever conceits we might like to dish out about the noble function of education.

Therefore, we have to care about the integrity of those grades. It matters to us if students cheat, and thereby earn a grade they don’t deserve. If you think about it, why should we care? Aren’t such people really just cheating themselves? Well no, not in a disciplined society, because the grade is a social good that purports to measure the value of human capital and thus exists within a grander social context. (I like to remind my students that the actual impact of their grades over their lives will turn out to be vanishingly small. But perhaps I am being optimistic, and it doesn’t matter anyway, because that is how they have been socialized to think throughout their youth, and nothing I can say will change the social value of the grade.)

So, the problem of AI in the classroom is not really about the learning function, it is about the grading function. AI facilitates a measure of attainment that we see as unearned and thus unwarranted by our social disciplinary standards, because we think through the lens of grades themselves. To wit, in a recent New Yorker piece about how AI is impacting the humanities classroom, the Princeton scholar D. Graham Burnett’s takeaway from reviewing AI-generated content based on his own classroom material was: “O.K. Respect, I thought. That was straight-A work.” The grade thought-structure is deeply ingrained.

But I am not dumping on Burnett’s essay, because I think he makes an essential and vital point. In a world of AI, he notes

You can no longer make students do the reading or the writing. So what’s left? Only this: give them work they want to do. And help them want to do it. What, again, is education? The non-coercive rearranging of desire.

That, at first glance, seems like an incredibly naïve, even pie-in-the-sky conclusion that only a professor of humanities at a top-tier institution like Princeton, teeming with grade-skipping, hyper over-achievers, could utter. Why? Because our disciplinary society is reliant on coercion. You can’t place “non-coercive desire” into the disciplinary fold, except under exceptional circumstances. (Like maybe, just spitballing here, a Princeton seminar?)

In other words, the rise of AI has created at the educational frontline something of an aporia within our approach to, and really, the very meaning of education itself. If we stop being naïve about AI in the classroom (i.e. students will be using it, full stop), then the result is we have to end up being naïve about education, thinking that a classroom becomes a place where students will commit to the work that they actually find interesting. That might work if your baseline model is a seminar of 20 students discussing the work of Li Bai. But what about the real world where you have classes teeming with students taking some entry level statistics course. “The non-coercive rearranging of desire?” LOL.

The reverse: the classroom remains a place to enforce a set of coerced tasks (reading, writing, exam-taking) all of which, because it can be so effectively circumvented via AI, requires surveillance systems of the type that would make the East Germans proud. But that means not only reinforcing the disciplinarian over the pedagogical function of education. It also means doubling down on the very mode of knowledge acquisition that is falling into an AI-driven desuetude, since that requires reliance on crude tools (exams, basically) that tend to measure as much – or more – memorization and cool-headedness under pressure than they do any real ability to think or engage critically and creatively.

But if such controls are needed because in a class of, say 100, only 20% will have enough non-coercive desire to do anything (read, attend lecture, write something, study stats, etc…), isn’t the real problem: why are the other 80% even there?

Of course, the answer to that is also rooted in our disciplinary society. For most, the benefit of a university education is not what it provides for the self in some abstract sense of self-actualization, nice as that sounds. Instead its value lies in externalization, via labour markets and the ability to acquire more generally higher forms of social capital.

In other words, why get a university degree? It’s to get a better job, make a higher wage, have better career prospects, have better life outcomes, and be better perceived in your social milieu. That’s not everyone, to be sure. But that’s a lot of people. Including many who attend Professor Burnett’s seminar at Princeton, since the value of a degree from Princeton is precisely in the (justifiable) perception that it leads to, if not fame, then certainly higher-than-average fortune. Which is why, and just a hunch here, all the undergraduates at Princeton were fanatically grade-obsessed in High School. I mean, that’s how you get into Princeton.

We are all socialized under this order. The disciplinary model is intertwined with our understanding of ourselves; it is part of what Foucault termed the episteme of our industrial and post-industrial modernity. And a feature of this or any other episteme is that it places limits on what we can conceive of as even possible. We’re trying to solve the problem of AI in our classrooms using toolsets and, more importantly, mindsets that the presence of AI itself is rapidly making obsolete and dysfunctional. No wonder we’re confused.

So, one interesting thing about the AI revolution, at least as we are perceiving it in the lived realities of our classrooms, is it is forcing us to confront the limits of our current epistemic understanding of our society. The core aporia in education – grades that AI makes largely irrelevant, but a praxis where grades are the core measure of educational attainment – is perhaps a leading edge, but nonetheless just one part of a larger transformation being ushered in by AI.

As I see it, the paradigm shift into an age of Artificial Intelligence is fast approaching, and with it the reliance on a disciplinary machinery developed in the 19th century and perfected in the 20th will, it seems reasonable to predict, prove increasingly unfit in the 21st. If students of the future encounter a classroom that is NOT based on coercion, but is instead better tailored to what they actually want to spend their time doing, well I doubt they will complain. But what that looks like? And how many of them there will be? No idea.

As for me, I will generally stick with my essay-based grading scheme, imperfect and vulnerable as it is. But I do plan to introduce more heterogeneity in the course evaluation to give students more options about how they can fulfill their coursework. The one-size-fits-all model of evaluation (midterm, final exam, e.g.) while convenient, sacrifices autonomy for normalization – which is the whole point in a disciplinary environment. Your ‘A’ can be rationally compared to her ‘B+’ and his ‘C-’. Of course if I were teaching ‘Introduction to Quantum Electrodynamics’ (i.e. the class my father taught for many years), that would not be possible since how much autonomy can you have in quantum electrodynamics (honest answer: no idea, but guessing little). But I teach philosophy and history, so that’s not a problem.

Overall, I still think there is a value in giving students the space to learn how to longform their ideas through writing essays and then giving those essays a grade. Some, maybe many, will doubtless use AI to circumvent the need to think for themselves. But many others will not, and it seems a shame that they may lose out, just so I can maintain the integrity of a disciplinary tool that is rapidly losing its meaning. So while I don’t claim any special insights on how education needs to adapt to deal with AI, I do think that a good place to start is to be adaptive in our thinking about what it means to take – and more importantly to pass – a class.

Well, this went on much longer than I intended, so if you’ve managed to make it to the end, many thanks for reading. Please share any thoughts you have and throw me a like!