“Literature is that neuter, that composite, that oblique into which every subject escapes, the trap where all identity is lost, beginning with the very identity of the body that writes.” (Barthes)

“We are faced with the problem of error.” (Quine)

The editor of the Philosophers’ Magazine, Daniel Kodsi, recently defended AI-supported writing as not essentially different from traditional approaches. Letting AI draft a text for you is like using a dictionary or conferring with an editor or following a style guide. “Writers have always relied on external supports,” Kodsi reminds us: “teachers, editors, friends, books, style guides, dictionaries, templates, stock phrases, genre conventions, and prior texts absorbed over years of reading. AI is simply another such support.” In fact, our resistance to AI is grounded in a myth (and even “cult”) of writing as “self-expression,” which Kodsi thinks we do well to abandon:

the basic purpose of writing is not to display the unaided workings of an individual mind. The point of writing is to communicate: to convey information, make arguments, issue instructions, tell stories, record events, or produce understanding in a reader. Once that is clear, the importance of asking who first generated a given sentence diminishes. What matters first is whether the writing says something true, useful, illuminating, or worth reading.

Kodsi does grant that there is a “narrow range of cases” where writing might be understood as “fundamentally self-expression” (though he is quick to remind us that here, too, the shoe only fits “imperfectly”.) In my reply to his post, I upped the ante a bit, suggesting that there is a broad class of cases where we want writers to “display” the “workings” of their minds and where we are, in fact, interested in their ability to do this “unaided”. This is a kind of writing that I teach and coach, namely, academic writing.

Kodsi disagreed. Academic writing, he suggested, is also mainly communicative; its aim “is to advance knowledge by putting claims and arguments into public circulation.” Here it doesn’t matter who is making the claim. “What matters is the research, not the researcher.” Moreover, published research today is not a pure representation of what one or several researchers had on their minds at the time of writing, put the result of a complex processes involving a variety of actors. “Research assistants, editors, peer reviewers, and colleagues already shape academic work in extensive ways, and none of that is thought to invalidate it.”

I countered (as I had already suggested in another thread) that the so-called replication and criticism crisis is rife with scandals in which blame for errors (and outright fabrications) is deflected onto, precisely, research assistants and co-authors. While I grant that the contributions of research assistants, editors, and peer-reviewers is a normal and accepted part of academic writing, this happens on the background of a presumption of authenticity, authorial intention. Anyone who has let a chatbot generate a text that expresses (even quite accurately) their own ideas knows the sadness of lost authenticity. Signing your name to it feels fraudulent.

But there is a deeper issue here, which arises, not in cases of research misconduct, but precisely in the cases were it conducted normally. Kodsi suggested I elaborate a bit on it because it didn’t seem obvious to him. This post is an attempt to re-articulate my intuitions about the essence of academic writing, why AI can’t do it, and why you shouldn’t let it. As you’ll see, I could be wrong.

In science, publication is not just a means to “circulate” ideas and discoveries. In fact, I would argue that good ideas and real discoveries will find receptive audiences willy-nilly, whether or not they are published in reputable journals. They will circulate as working papers, seminar and conference presentations, and in personal emails. And they will be implemented in subsequent scientific research and technological development. The main purpose of academic publication is not to communicate research results, but to expose those results (and the ideas they embody) to the criticism of qualified peers. This requires the researcher to present them in ways that are open to interrogation by other knowledgeable people, both pre- and post-publication. In a word, publishing our results (in conventional academics form) makes us corrigible. It tells our colleagues, not just what we found, but how we found it, so that any mistakes we may have made along the way will be noticed by people working with the same methods, framed by the same theories.

Here the researcher does in fact matter as much as the research. Research is embodied in social and material conditions, including the actual flesh-and-bone bodies of researchers. It is the researcher whose methods must be corrected and whose instruments must be re-calibrated if they are making serious errors (or, in those hopefully rare cases, if they are committing outright fraud). “Publish or perish” is a defensible norm precisely because we want our epistemic authorities, whether they are speaking in the classroom or to the media, to be under continuous “peer review”. We want the people who speak confidently about how the world works to also be speaking carefully and clearly to people who know something about it too. In short, it matters whether the words on the pages that they sign their names to were “generated” by their own minds, not by computing statistically plausible passages of prose on the subject, even when the claims made are perfectly true.

More importantly, perhaps, composing our thinking in critically open paragraphs that make explicit claims about the world, and support, elaborate, or defend those claims, shapes our thinking, preparing it to be corrected (or corroborated) by peers we respect. That is, scholars who do their own writing are establishing precisely those logical connections between their ideas that will let them more efficiently update what Quine called their “web of belief” in light of counter-evidence to any one of them. Prosing our worlds, we might say, weaves the web in ever finer threads that allow changes to be made precisely, without, except in the extreme (but also easily identifiable) cases, having to discard a whole theory and replace it with another.

We expect academics to be thus prepared for criticism. Not just willing to hear it, but able to respond to it, sometimes by changing their minds. And we don’t want these changes to result in chaos and confusion; we don’t want our knowledge of the world to be turned on its head very often. (We don’t want “progress through revolution” all the time!) We rely on our institutions of higher education (including academic research) to keep the ebbs and flows of our stream of consciousness orderly, to keep our questioning within reasonable limits.* This is simply what it means to be academically “literate”; and we expect this of, precisely, academics themselves — not just their editors or machines.

Kodsi’s argument tends in the same direction as those who declare, on Barthes’ dubious authority, “the death of the author” in order to liberate the reader from the intentions of the writer and make what they will of a text, except that Kodsi seems more intent on liberating writers from the moralizing demands of their readers. As a mode of (or, as Susan Sontag suggested, a stance against) literary interpretation, I’m all for it. But, for academic purposes, I would deploy Foucault’s proposed “author function” (an application of his idea that a discourse is shaped, in part, by its “enunciative modality”) with a straighter face and insist that our authorities commit themselves to recognizable roles. Not least, to assume the position of a subject in the discourse, an author — or even a potential author, a peer, a hypocrite lecteur, mon semblable, mon frère — is to open yourself to questions from other knowledgeable people. The prose of the world prepares our minds for critical discourse. Academics should be better prepared than most to be wrong. It is by writing that we prepare our bodies to face our errors.

_______
*As I was writing this sentence I couldn’t help but think of Kafka’s little allegory in “The Great Wall of China”:

“Try with all your might to comprehend the decrees of the high command, but only up to a certain point; then avoid further meditation. A very wise maxim, which moreover was elaborated in a parable that was later often quoted: Avoid further meditation, but not because it might be harmful; it is not at all certain that it would be harmful. What is harmful or not harmful has nothing to do with the question. Consider rather the river in spring. It rises until it grows mightier and nourishes more richly the soil on the long stretch of its banks, still maintaining its own course until it reaches the sea, where it is all the more welcome because it is a worthier ally. Thus far may you urge your meditations on the decrees of the high command. But after that the river overflows its banks, loses outline and shape, slows down the speed of its current, tries to ignore its destiny by forming little seas in the interior of the land, damages the fields, and yet cannot maintain itself for long in its new expanse, but must run back between its banks again, must even dry up wretchedly in the hot season that presently follows. Thus far may you not urge your meditations on the decrees of the high command.”

“If its object were scientific and philosophical discovery, I do not see why a University should have students.” (John Henry Newman, 1852)

In the coming years, I expect, the purpose of higher education will be a recurring topic of discussion among academics. What are students supposed to learn at university? How are they to be improved? What do we want them to become better at? The answer will have to take seriously the fact that we have the authority only to impart and certify knowledge, to arm them only with ideas and “the authority of right reason,” as Scotus Erigena put it. We may only teach and examine our students. Why should students submit to this authority? What do they gain from it? What does society gain from demanding it of them?

An alternative, after all, is fast coming into view. Students are learning to use artificial intelligence to “support their learning,” which, unfortunately, is often indistinguishable (sometimes to themselves and sometimes, it seems, even to their teachers) from avoiding the problem of learning altogether. I’ve pointed out before that when Bertrand Russell suggested that a “system of notation” could almost serve as a “live teacher,” and “a perfect notation would be a substitute for thought,” he was probably not imagining the “stochastic parotting” of large language models. But they certainly do seem capable of substituting for a great deal of the thinking we’ve traditionally demanded of our students. We seem to be on the verge of automating the very problem of knowledge. Our students have been given an epistemic workaround.

Now, I have served for about two decades as a writing coach at a major European business school and during that time I have developed what I think is a healthy, working epistemology. I don’t claim to have solved all the problems that philosophers have raised in regards to the nature of knowledge, but I have a confident grasp of what we mean, or should mean, when we say that we are “knowledgeable” people and are helping our students to become likewise “able”. As I never tire of telling them, to be knowledge-able is both to be able-to-know things and en-abled by that knowledge to do things they would not otherwise be capable of doing. Knowledge is a competence that is manifest in a performance. In what, then, does that performance consist? What should university students become increasingly better at doing? What does it mean to be able to “know something” for, let us say, “academic purposes”?

First, if you are knowledgeable about something you are able to make up your mind about it. Invoking the long tradition of Western epistemology, I usually characterize this competence as the ability to form a justified, true belief. Given a series of experiences or a set of materials, a knowledgeable person is able to reach a determination of what is going on and what it means. They can convert what William James (and, later, Thomas Kuhn) called “the bloomin’, buzzin’ confusion of experience” into sterner stuff: propositional attitudes about stable facts, and theories that arrange concepts into orderly frameworks. Obviously, even very knowledgeable people can be wrong about particular facts and this is why I have increasingly come to emphasize that knowledge isn’t so much being right about things as being able to change your mind about them in an orderly fashion. On the path “from stimulus to science,” as Quine puts it, “we are faced with the problem of error” at every step. Knowledgeable people face this problem, let’s say, squarely.

But forming a belief, no matter how true or justified, is not sufficient to be considered knowledgeable in an academic setting. You must not only know what you’re talking about, you must be able to talk about what you know. To know something at university is to be able to hold your own in a discussion about it with other knowledgeable people. For students, this should be understood as an ability to discuss topics raised in class with their fellow students. It is sometimes argued that “academic writing” lacks “relevance” for students because it doesn’t relate to their “lived experience” or “real life”. Against this, I have long tried to argue that the “academic situation” offers an entirely meaningful context for rhetorical engagement. The classroom is a microcosm of the discipline and the students are learning to address their peers and respect their criticism. In that sense, students are apprentice scholars.

But nor is that all there is to knowing things “for academic purposes”. Knowledgeable people are not just able to make up their minds and speak their minds; they are able to write it down. The test that I recommend is whether you are able to compose a coherent prose paragraph about something you know in under half an hour. More specifically, are you in an epistemic position to decide at the end of one day to write a paragraph a start of the next about something you knew was true last week? This is not just a good test of whether you actually know the thing you think is true, it is also a good skill to have in itself. It’s nice to know that you can write half a page about anything you know in half an hour. Doing it regularly should give you a certain confidence, not least that confidence which comes from regularly experiencing that you do, in fact, know things to be true.

It is this competence that I hope the university will continue to represent and this confidence (which is not without humility) that I hope we will continue to instill in our students. To that end, I never miss an opportunity to plead for a shift of focus in our assessments, from homework assignments back to on-site, off-line, invigilated written examination. A university student who has taken a course (or is in the middle of a term) should be able to answer a relevant 20-word question given 4 hours and 1000 words (with relevant accommodations for disability) in a way that can be easily graded by their teacher. The ability to write, say, five coherent prose paragraphs is a good proxy for whether the student is able to make up their mind on the subject they are studying, speak their mind, and (of course) write it down. I would certainly argue that the plain in-ability (that the exam conditions of course make, precisely, plain) to write such an essay should call into question their claim to be a “knowledgeable” person on the subject in question. Higher learning is manifest in such basic skills.

[I generally don’t like this kind of stunt. But I was able to generate this essay in under 10 minutes using Microsoft Copilot. While it reused some of the language from my prompts, all the paragraphs (and the title of the post) were generated out of whole cloth by the AI. I had a brief exchange with it about why it is so good at generating prose about Hamlet, and then I asked it to write a five-paragraph essay using this fact to argue for on-site examination. I then simply asked it to add two paragraphs (the fifth and sixth paragraphs in the final version) about how (a) students would be incentivized to write at home and (b) how homework could still be used, giving only indicative grades. I have done no editing of the final text at all. I would say it is 100% in line with my views on the subject. I’m posting it here mainly to document where we’re at today. Copilot’s parting comment was, of course, to offer more help: “Would you like me to tighten this into a formal academic style with references, turn it into a persuasive op-ed, or create a version optimized for a university policy proposal?” I left it there.]

In the age of artificial intelligence, the integrity of literary education faces an unprecedented challenge. Large language models (LLMs) have made it remarkably easy for students to produce polished essays on canonical texts without engaging deeply with them. Nowhere is this more evident than with Shakespeare’s Hamlet, a play that has generated centuries of commentary and analysis. Ironically, the very text we most hope students will wrestle with is also the one they can most convincingly fake an understanding of using AI. This reality demands a reconsideration of assessment methods, and on-site, invigilated written exams offer a compelling solution.

The problem begins with the nature of LLMs and their training data. These models are built on vast corpora of text, and Hamlet occupies a privileged position in English-language discourse. It appears in countless essays, scholarly articles, and online discussions, giving AI systems a rich interpretive reservoir to draw from. As a result, generating a plausible essay on Hamlet requires no genuine engagement with the text—only a prompt. When students can outsource interpretation so easily, traditional take-home assignments lose their ability to measure understanding.

This erosion of authenticity matters because literary study is not merely about producing words; it is about grappling with ambiguity, forming original insights, and developing interpretive judgment. Hamlet is famously complex, inviting questions about action and inaction, mortality, and meaning. These questions cannot be answered by parroting consensus views—they require intellectual struggle. If students bypass that struggle through AI, they miss the very experience that makes literature transformative. Assessment must therefore shift from evaluating polished output to observing the thinking process in real time.

On-site, invigilated written exams restore this integrity by creating conditions where students must rely on their own interpretive resources. Without access to AI, they are compelled to engage directly with the text and articulate their ideas under pressure. This format does not eliminate preparation or external influence, but it ensures that the final performance reflects the student’s own reasoning. In doing so, it preserves the core educational value of literary study: the cultivation of independent thought.

Moreover, this kind of exam would incentivize students to practice writing at home. Knowing that their grade will depend on their ability to produce coherent, insightful prose under time constraints, students will naturally rehearse the skills required for success. They will write practice essays, refine their arguments, and internalize the rhythms of literary analysis. In this way, the exam becomes not just an evaluative tool but a motivational one, encouraging sustained engagement with both the text and the craft of writing.

Finally, homework can still play a role in this model, provided it is not graded in a way that affects the final mark. Assigning essays for indicative feedback rather than summative assessment allows students to experiment, receive guidance, and improve without the temptation to outsource their work to AI. Teachers can use these assignments diagnostically, identifying areas of weakness and tailoring instruction accordingly. By decoupling homework from high-stakes grading, we preserve its pedagogical value while neutralizing its vulnerability to technological shortcuts.

In conclusion, the rise of LLMs has exposed a vulnerability in traditional assessment practices, particularly for canonical works like Hamlet. If we want English majors to truly understand literature rather than merely simulate understanding, we must design exams that cannot be outsourced to machines. On-site, invigilated written exams are not a nostalgic return to the past—they are a necessary adaptation to a future where authenticity is at risk. By embracing this change, we safeguard the intellectual rigor that defines the study of English.

On Saturday, I wrote about one objection to my “Prompts and Conditions” post that came up during my faculty development course on “Teaching Writing in the Age of AI.” This post is about another one, which begins with something I say at the end of that post:

Surely, we can say what a math student at any level should be able to do without a calculator, or what a history student should be capable of without the internet. AI generalizes this problem. In each field, we must decide what students should be able to do on their own.

A participant reminded me that the calculator analogy has long been discredited as a helpful analogy for understanding AI, especially if our aim is to limit reliance on it. After all, everyone relies on calculators these days, and very few exams ban their use. This hasn’t caused any kind of catastrophe for education. We have simply changed the way we teach and learn math.

I think it’s worth looking into this claim in some detail, I should say. After all, it is my impression that many high-stakes exams — like the American SAT — have very specific rules for calculators that enforce limits on the functions that are allowed. I’m pretty sure this has been the case for as long as calculators have been available; their use is in governed by policy. But the general idea that math instruction hasn’t banned them altogether is of course true, nor have we kept teaching the same “old” things is. In any case, talking about “what a math student should be able to do without a calculator,” the participant suggested, was like asking what a carpenter should be able to do without a screwdriver. The whole point of learning the craft is learning how to use the tools.

I immediately liked this way of putting it because when he mentioned carpentry I thought he was going to talk about power tools, but the problem, of course, arises already at the level of saws and hammers and screwdrivers. We may as well start there. Would I say, “Surely, we can say what a carpenter’s apprentice should be capable of without a screwdriver”? As it happens, I would answer yes. But I must first emphasize that I have not said that students and apprentices should be examined only without their tools. I have said they should also be examined without their tools and that, in any program of instruction, there must be some set of skills that can be examined this way. It’s not either/or, but both, separately.

In the case of the carpenter’s apprentice, I suggested that someone who is able to use the standard toolkit will also be (and should also be) able to talk intelligently about how they would go about a particular task, without holding any of the tools in their hands. Also, an apprentice woodworker can be sent into the woodshed to pick out some boards that would be ideally suited to making a particular piece of furniture. This requires no tools, only a good grasp of the materials themselves (a “feel” for them, if you will). It might also be worth seeing if they can “eyeball” rough dimensions, i.e., whether they have realistic intuitions about size and space.

(I am sometimes told horror stories by teachers of quantitative methods about students who do not immediately recognize that a calculation they have let a spreadsheet carry out is off by three orders of magnitude and even in the wrong direction: positive when they should be negative, negative when they should be positive. It is worth having students estimate calculations, without a calculator, simply to make sure they have a realistic sense of the thing they are calculating.)

I think it is true that we must accept AI into writing instruction just as we have accepted calculators into math instruction. We can’t burry our heads in the sand (or, perhaps more precisely, require our students to tie their hands behind their backs). But, just as we can require an apprentice to be able to tell us how they plan to go about a project before they pick up a tool and show us what they’re capable of, we can just as reasonably at least require university students to tell us how they would use AI to solve a writing problem. But we can go further.

My participant suggested that I was ignoring what we know about “embodied cognition” (and we might add “extended mind”). But I am absolutely on board with those sorts of views. We exist in an environment of tools and machines, which not only help us to get around, but shape our very being. I will insist, however, that our environment also includes other people and the language we use to communicate with them. Our words, as Heidegger pointed out, are part of the “equipmental contexture” of our existence, our being-with-others. Teaching students how to write good prose by themselves is very much a way of helping them embody their knowledge.

I want to stress that my point is that AI “generalizes” the issue. (Indeed, Silicon Valley keeps promising us something the call AGI: “artificial general intelligence.”) With minimal prompting, AI is increasingly able to simulate almost any academic competence at least “passably” (deserving a C, let’s say, or what we call a 7 in Denmark.) If universities are to maintain their assessment integrity, we need to find a way to make sure that the actual bodies of the students are capable of something in particular, something that reflects 3 to 5 years of study. And that means we have to come up with some things we can test whether their bodies can do.

I’m running a faculty development course this month called “Teaching Writing in the Age of Generative AI” and it has led to some interesting discussions. I asked the participants to read and reflect on my post, “Prompts and Conditions,” which I wrote a couple of years ago, when the challenge that AI might pose for higher education was just coming into view. I think the basic idea of the post was agreeable to most of the participants, but a number of them had some interesting reactions to what I would call the “rhetoric” of the post. I want to address two of them in particular in a post each.

One of the participants noted a parenthesis at the end of what I took to be a very practical remark about setting deadlines. Here’s what I had written:

If we imagine classes are held on Mondays, students can be given the take-home prompts at the end of the class and submit their essays on Friday (there’s no need to ruin their weekend).

“Doesn’t this suggest that our students don’t want to learn? Why would we presume that studying on the weekend ruins it?” asked the participant. He recalled that when he was a student he was happy to spend his evenings and weekends studying and, indeed, that he felt that he was expected to do so by his teachers. Is this something that we have suddenly abandoned? (And, we might add, does AI force us to do so?)

Now, I must say that I had not expected anyone to take this remark as seriously as that. The idea of “ruining” a weekend by doing school work was only intended as a lighthearted gesture at the priorities of young people. But perhaps playing to these priorities is ill-advised; and perhaps it comes off as condescending, even to the students who have them. It’s always worth thinking about the rhetorical effects of our pedagogical strategies.

In defending my choice of words at the time, I did point out that he was taking a rather hard line against another kind of concern that teachers often express for their students: not everyone has the luxury of devoting their entire lives to school while they are attending university. Many have jobs on the side; some even have families to tend to. “Ruining the weekend” may be more existential for some students than merely skipping a night on the town. This seemed to elicit some nods in the room, including from my critic.

In any case, it’s important to remember that deadlines are always somewhat arbitrary and are likely to occasion both procrastination at first and consternation at last among some students. So I spend a lot of time teaching (and coaching) students (and faculty) to plan their work in orderly, half-hour “moments” of composition so that they can comfortably meet their deadlines without having to miraculate a text at the eleventh hour. For the same reason, whenever it is up to me, I like to place that eleventh hour before noon on a Friday, rather than midnight on a Sunday. It’s just a good way to signal that you may as well get the work done during the working week, as part of your regular day-to-day program of study. It keeps the task of doing an assignment in proportion.

Like I say, I don’t want to dismiss the concern about the rhetorical force of talk about “ruining” our students weekend by asking them to study. But perhaps it is precisely the question of whether what they do during their “free” time is chosen or assigned. A student that wanted to read a book you have suggested or do some writing of their own can still have that hope “ruined” by poor planning and ostensibly “generous” deadlines.

I don’t claim to have a definitive take on how to talk about school-life balance with students and how important we should presume (whether in our thoughts or in our speech) learning is to them. As an occasion to give it some thought, my participants remark is well taken. I’m happy to hear more thoughts in the comments below.