There will be no long Hili post this morning (though of course Hili’s dialogue is below) due to medical exigencies, i.e., visits to eye doctor (I’m ok).
But here’s one issue for discussion, taken from a new op-ed in the NYT. Joe Cruz, head of the philosophy department at Williams College, where Luana works, has a piece called “What if writing isn’t central to thinking?” (article archived here). If his thesis is true, it will give students license to use AI for essays and exams, which in fact Cruz allows. He has a workaround for that, but it’s not a good one.
His thesis:
Most college professors I know believe that students learn to think — critically, creatively, reflectively, tenaciously — only by writing. At Williams College, where I am a professor, writing skills are among the very few general education requirements.
It’s no surprise, then, that universities see artificial intelligence as an existential threat, and have launched an arms race of countermeasures: more proctored essay exams; test-taking rooms shielded from the internet; requirements that essays be composed on platforms that track all edits, so the word-for-word genealogy can be verified. I’ve heard professors rejoicing when they see typos or grammatical errors in an essay, as if they proved that a student wrote it without any help from A.I.
But what if we are mistaken in treating writing as the highest cultivation of thought? The evidence that supports this belief is thin at best. Writing can be useful cognitive scaffolding for memory and self-regulated learning, but the notion that it is the best process for developing thinking skills is a fairly recent idea, traceable to the 1970s writing-across-the-curriculum movement.
His lack of evidence is merely a lack of relevant research. But Cruz finds support from none other than Socrates!
In Plato’s “Phaedrus,” Socrates offers a startling criticism of writing. Written words, he claims, are not the true children of knowledge, because writing cannot speak, answer questions or come to its own defense. [JAC: OF COURSE it can if the response is also in writing.] Only through conversation among minds, he suggests, does reason yield real insight. For the past two millenniums, this appraisal has seemed like a failure to appreciate the power of writing. But perhaps Socrates was right, and we think best by speaking, answering and defending of our ideas.
The problem here is that you can have back-and-forth discourses in writing (here’s one I had with Adam Gopnik about whether there’s truth in art.) And those are permanent, forcing you to think carefully, because once something’s published, it’s usually permanent.
But the essay gets worse:
I don’t rely on prepared scripts when I teach, and few of the teachers I most admire do. Instead, I engage my seminar students through messy, searching conversation, responding to their presence in the classroom to gauge whether we’re understanding one another. I want to make new connections in the immediacy of our back-and-forth. I want to recall a thread from two class meetings ago spontaneously and weave it into my improvised lecture while snapping into new ways of thinking about it in the moment.’
Here Cruz is talking about seminar classes—small discussion group with a back-and-forth in a small group. You can have those in philosophy, but not when teaching basic physics, chemistry, or biology. This doesn’t work with a larger class, of course. He does admit some use for essays, though how will he ensure that the essays aren’t written by AI?:
Writing assignments serve practical needs of colleges and universities. When we receive a batch of 10-page essays, we can take them home, read a few, put them aside and grade the rest when the opportunity arises. If our minds wander, we can reread the prose before us. If we need to justify a grade, we can cite the record. Writing is a portable, compact trace of mind that is more manageable in a bureaucracy than is a conversation. But none of these conveniences is about developing the best thought.
And here’s his solution: we must “think aloud together”!
Last semester I permitted my students to use A.I. tools on their final essays in accordance with specific guidelines. Each then had a 45-minute oral exam in which I could audit their understanding of their claims and push them further in their analyses. I was reassured by the results. In no case did I suspect that an A.I. had written a student’s paper in the substantive, objectionable way that we are all concerned about. The students were all too sharp on the ideas, too deft in reconfiguring their thoughts and extending them dynamically. Moreover, at least half of the class was more agile, more reflective and clearer during oral exams than they were in their papers.
I believe colleges can permit the use of A.I. writing tools without undermining the next generation of thinkers. But to get there, we need to embrace a different image of a university education — one in which ideas are expressed in real time and analytical thought unfolds as a debate or an oral defense or a group discussion. It will take courage to demand that students demonstrate active, participatory mastery of ideas. And it will take courage on their part to abandon the safe recursion of revision and narrative refinement. To be sure, there will be challenges of scale, and not every kind of institution will be able to adapt. The benefits, however, may be enormous.
Well, if all learning occurred through Socratic dialogue between professors and students, that’s fine. But as we all know, that’s not the way it works. There are many courses with too many students How do you have a 45-minute oral exam if you have 600 students? And Cruz seems too credulous to me: “In no case did I suspect that an A.I. had written a student’s paper in the substantive, objectionable way that we are all concerned about. The students were all too sharp on the ideas, too deft in reconfiguring their thoughts and extending them dynamically.” I call bullshit on that. Did Cruz think of putting essays through Pangram? As I’ve learned, AI can write essays that are better than students can write, even Williams students.
In my career, I’ve written a lot and published a lot, both in the scientific literature and—more germane here—in the popular press, where I’ve placed over 190 pieces. I’ve learned that trying to express your ideas lets you know whether you really understand them, for sometimes, when trying to explain something, you realize that your mind is not clear on an issue, and you have to think it out. And for me, but perhaps not for many, I get great pleasure from writing, trying to produce prose that is incisive and, yes, beautiful. I have a long way to go, and will never get close to masters of this form, like Richard Dawkins in my field. But once written, a published essay is permanent—there for all to see—in contrast to the spoken word, which is evanescent. There is no substitute for an essay or article written and published, so long as the writing is not done by a computer program.
Please weigh in below, especially if you are a teacher of any sort.
**************
And a felicitous Hili dialogue. Meanwhile in Dobrzyn, Her Highness gives us words of wisdom—somewhat of a redundancy:
Hili: What do you think of artificial intelligence?
Andrzej: I get the impression that the better it gets, the better plagiarism gets too.
In Polish:
Hili: Co sądzisz o sztucznej inteligencji?
Ja: Mam wrażenie, że wraz z jej postępem poprawia się jakość plagiatów.
The AI concerns are tiresome in many ways –
I’d like to add something I have not heard come up – but is rather old – perhaps by starting with a name :
W. V. O. Quine
I happened to start reading about Quine (again) the other night, and followed many links in the Wikipedia bio. E.g. Quine–Putnam indispensability argument. Also in this web link sprawl was Justified True Belief. It occurred to me that Quine (et. al.’s) analytic philosophy is germane to the linear algebra – AKA AI – running rampant in the world’s minds. But I do not know specifically how – except that it appears to be distinct from Continental philosophy – e.g. Hegelianism.
So I could perceive the connection of a number of regular interests here on Why Evolution Is True – The Website!.
As for the classroom aspect : in hot take form :
Students using AI in class is like musicians using electric instruments for concerts.
That is all. 😁
Oh and :
PCC(E) : “I’m OK”
Big 👍
“Students using AI in class is like musicians using electric instruments for concerts.”
On the contrary, it’s more like musicians using recordings of other musicians in concerts.
Steve Reich comes to mind …
Hip-Hop / sampling…
Exactly.
Here’s an example of what I mean – a ‘short’/clip of a Rick Beato interview with Bumblefoot :
https://youtube.com/shorts/EOtbz1iPbwI?is=2QX5y8ULK97Y5dsS
… it is an amazing sound and I love it.
However, the combined sound was not literally played by the musician. Rather, the musician – with great skill – used electronic effects that take input sound to output a combined new sound from a single guitar in this case. There are other scenarios.
Contrast to :
An acoustic or classical guitar player au naturel in a concert hall — there’s a distinction with a difference.
I think there’s a parallel to see in this.
A bit of digression is fun as long as you don’t get yelled at for violating the Roolz. Here is a story I read in Grayling’s History of Philosphy. Schopenhauer was a bit jealous of Hegel and intentionally scheduled his classes to clash with Hegel’s. Hardly anyone turned up. They went to listen to Hegel.
Only the following sentence in your meandering response seemed to have anything to do with the OP:
“Students using AI in class is like musicians using electric instruments for concerts.”
And this doesn’t seem to hold. I need to know how to play an instrument to play an electronic keyboard, or shred on the electric guitar. AI would be akin to a robot doing it for me.
The product of “AI” is entirely mathematical as it is the output of linear algebra.
Quine, Putnam, et. al.’s work is in the philosophical understanding of mathematics, and naturalized epistemology – a “systematic attempt to understand science from within the resources of science itself”.
Philosophy of science is philosophy enough as Quine put it.
This^^ is from the Quine bio on Wikipedia.
I found it not unrelated to “AI”. But thanks for the withering criticism.
Methinks thou givest far too much importance to linear algebra.
Yes, AI models rely heavily on large matrix operations. A single layer may contain millions of parameters, and processing one token [word] involves applying many such layers, alongside attention and other operations. The cost therefore reaches billions or more arithmetic operations per token. Generating text repeats this process for each output token. And training the beast is vastly more expensive because the model must adjust billions of shared parameters over enormous quantities of training data. [Slightly edited output from GPT-5.6-Luna in response to my more verbose and much less accurate (but funnier 🙂) first draft]
But, and here finally is my point, all that number crunching does not show itself in the output. It’s purely in the infrastructure, not in the result. And FWIW, philosophically speaking, the entire universe has been described as “entirely mathematical”. Your move 🙂.
You know, I am plenty old enough to have seen various technological revolutions that have replaced older implements in teaching. I was the last generation to learn how to use a slide rule. We were taught in the 3rd grade, but in the 4th grade we had calculators. I am sure that was met with a lot of consternation among the adults, but I certainly thought that calculators were way cool. I expect the calculator does take away math skills from students, especially in the area of solving a problem that includes parentheses. But what is done is done.
AI is far more than a calculator when it comes to developing young minds. In seconds it does their thinking, writing, punctuation, and their (increasingly accurate) citations of sources. With AI, the student is passive and is never made to increase their weights or their reps in the mental gym. Look, I am well into my 60’s, and this comment took me almost 10 minutes to write. I didn’t need to take this long, but I enjoyed improving parts of every sentence, and I even backed up and re-directed its trajectory more than once. Not that I necessarily needed to. The circulation in my brain has pleasurably increased, and although my comment here is not going to be as thoughtful as other comments that will soon appear, I am still experiencing satisfaction. I did this.
Mark – re: ” … I am still experiencing satisfaction. I did this.”
Thank you, well said. Your second paragraph nailed it on the head – for me – regarding writing and AI.
And let me emphasise the mental value of revising. I prefer doing several passes through my near-final draft, because each pass unearths at least one thing that cries out for improvement. Developing that sort of intuitive aesthetic sense requires practice, but ISTM even at the most basic level of proofreading (which is surprisingly error-prone) and copy editing, one starts to develop it.
The closed captioning “bloopers” that occur quite frequently on broadcast tv may be a basic example of why and how we should not rely on AI too heavily. If a person were to take AI instructions as “gospel” while engaged in a project involving potential hazards (electrical, chemical, mechanical, etc.), they could possibly earn a Darwin Award in a flash. Perhaps some day, when “Super Intelligence” (as our “Leadership” has instructed us to call it) becomes “Super-Dee-Duper Intelligence”, everything will run smoothly.
Ugh, Writing may not be “the highest cultivation of thought,” but being able to write in a way that accurately reflects your thoughts and that readers can understand is critical. Now, as suggested, Socratic dialogue can achieve much the same thing, but not on the scale that a modern society requires. Even Socrates wasn’t teaching all of Athens, and the sum of human knowledge at that point in our history was relatively small. The other thing is that writing and reading go together, and it’s not clear that one can abandon writing and not negatively impact reading skill. I assume that Cruz isn’t saying that we should also learn (about things other than ourselves) as opposed to displaying our learning via the Socratic method. I can’t imagine the time it would take to convey the knowledge associated with any modern academic discipline to even a few students. Writing conveys and preserves knowledge; it’s fundamental.
Cruz’s approach is only possible in fields with rapidly declining enrollments. His whole Socratic approach makes no sense when trying to teach why demand curves slope downwards or why electrons have spin. 45 minute oral exams means that we would still be giving exams to students who took our courses a year and a half ago.
This is not a very good advertisement for philosophy as a discipline.
There is an old joke that is aimed at philosophers. I won’t narrate it here, but the gist is where an administrator was kvetching that a science professor was spending too much money to do research. So why can’t they be like people in the math department, where all you need is pencil, paper, and a waste basket? Better still, why not be like those in the philosophy department, where all you need is pencil and paper?
And if you were as good as Socrates, not even that.
Snap!
And regarding the efficacy of high-AI learning, I’m reminded of the common Soviet era saying, “We pretend to work, and they pretend to pay us.” And I guarantee you that in the class struggle, the students overall are far more motivated to cut corners than the educators are to educate. E.g., seeding errors in AI output is trivially easy; just add to the prompt something like “Include a few clangers in spelling, grammar, and logic. Include some student-level humour (but no fart jokes).” Try it.
I struggle to take seriously anyone who would write “For the past two millenniums” — but maybe he threw that in to prove it wasn’t written by AI?
LOL
That plural sure is incongruous with a philosophy prof. expounding on classical thinkers, but it’s not wrong.
One interesting form used by Wittgenstein is that of a dictated text. In thinking through an issue, he made written notes, and developed his ideas through these notes. At various times, he found it useful to dictate a text in which the ideas were worked into a more or less continuous structure. When Wittgenstein gave lecture-classes, he worked over the material using his notebooks between the lectures, then lectured without notes, but the material is close to that in the notes, though he also responded to students who had questions, including Alan Turing.
A well-written essay is clear, compelling in its arguments, and only as long as it needs to be. For the vast majority, this is a learned skill and not instinctive. Thus, across 40 years of university teaching, I have mostly encountered student essays that are ‘messy’. Yes, they fundamentally answer a question or viably develop a proposition. However, they can be painful to read. Long run-on sentences, with a bucketload of embedded clauses. Multiple redundant sentences making the same point. Tangential excursions that detract or obscure the main points. And, flowery thesaurus intrusions. In short, a quite acceptable product relative to the task at hand, but a horrible read.
I don’t have any of my written essays from my undergrad years, but they were probably as good (and as painful to read) for my professors. Forty years on, I am better but still not always clear and definitely could be more succinct. I find that I am a far better editor for the writings of others than my own. Thus, still whenever a colleague, friend, spouse, co-author, or reviewer reads my “perfect” prose there are always edits and suggestions for improvement that astound me in how obvious they are!
The few times I checked my own writing into and AI app, it generally makes very few suggestions. Yay, I’ve actually learned something over the years! There are, however, always a couple of suggestions that are clearly improvements (and corrected typos). But I never just cut and paste, because there are also some suggestions that would subtly change what I want to say.
Thus, overall I would argue that if students are using AI as a grammatical editor, then more power to them. This certainly makes life easier for their readers. And hopefully, they are learning from the AI suggestions how to be better writers – just like I learned from all the humans who edited me over the decades. Since AI is really a very recent development, my actual teaching experience is limited to one class. I allowed complete AI use for the students, and that use was indeed to only make editorial improvements in the writing.
One point in support of Cruz. The vast majority of students will never write a 10 page essay on anything after graduating. But they ought to, and need to, be able to think clearly and verbally express themselves every day of their lives! I’m still undecided whether AI will help develop this or just replace it. Could go either way.
Well, tell me how you think AI by itself will make students into better speakers, improving their abilities to express themselves clearly and in an engaging way.
I’m serious; I’d like to know! In Cruz’s essay, it is the TEACHER, not AI, that is helping the student to think and speak.
Yes, the teacher is not a potted plant in the room. Cruz makes his students verbally defend/propose answers. That definitely requires some thinking, both beforehand and during. It was important to me in my one class experience with AI to have the students be able to recognize that an AI answer to a posed question could be both well written and not particularly thoughtful. However to your question, becoming an effective writer does not equal good speaker (e.g., Stephen Jay Gould in my experience). That’s a different skill and I don’t see AI replacing actually practicing and doing it. I think the papers I published are clearly and hopefully persuasively written. However, I cannot imagine a better way to clear a room than to read one verbatim out loud.
My first thought on reading Cruz’ essay was the effect his teaching method would have on the substantial proportion of students who aren’t classroom stars. Unless we’re dealing with a small seminar of the best and brightest, most class wide discussions probably involve a majority of students throwing out a sentence or two and a verbal minority explaining, elaborating, and expanding on this. While the result might seem like it’s an entire class equally involved, not everyone is actually engaged in the full process.
Writing, however, isn’t a group project. It’s personal. You can’t implicitly rely on the mental and verbal skills of your classmates to boost your understanding, which I suspect might be a temptation somewhat similar to using AI.
I’m not sure that writing is the “highest cultivation of thought” but it’s the best way I know of to express one’s thoughts without ambiguity. By “writing,” I don’t mean the written word itself. The written word is the product of writing. What I mean is the process of writing, from which the written text emerges. Every choice of word is purposeful. Every path of logic is tested. Much of what it means to think at all is engaged. Writing is a stringent test of one own thoughts and beliefs, and—if one is open to change—the process of writing can even change one’s own mind.
I know that people struggle with writing. Much of this comes from impatience: not choosing every word, not rereading, not moving text around, not eliminating the unnecessary. (I am not immune.) But as a methodology for creating clarity—for myself and for others—writing seems like a form of magic.
Amen.
Writing can be made more difficult by teaching it as a required chore, not as an opportunity. Similarly for reading beyond just functional literacy. And for many other things (even sex, at least according to Monty Python).
Whenever I actually try to put my thoughts in writing, I discover all kinds of flaws in those thoughts. The writing process forces me to organize my thoughts better and arrive at stronger conclusions.
So unless I am some kind of mutant, I believe this applies to many other people too.
You remind me of my two favorite quotes about writing:
“If you’re thinking without writing, you only think you’re thinking.”
— Leslie Lamport
“Writing is nature’s way of letting you know how sloppy your thinking is.”
— Guindon
Both of those resonate with me when I struggle to express what I want to say succinctly and clearly.
Great quotes!
I’m an engineer and I need to be able to write clearly and succinctly in order to communicate our team’s ideas. I’ve been using AI over the past couple years to support this. At first, it was as an editor. I’d do the research and then ask AI to “summarize this into a few key bullets.” I was in control and knew the information, it was my helper.
Now I ask it to do the research and summarize it, and I skim through the results to make sure it’s correct. I’m not as informed as I used to be, which on one hand allows me to get more throughput in terms of volume of work, but I’m starting to realize that I’ve lost something in terms of deep understanding.
For writing Python though it’s great. I never used Python in the past for analysis of large datasets (I did simple analysis using Excel macros or charts), so now that I can do that I do feel like I’m more able to do deeper analysis of data than I could before.
My opinion: students should have to write their own papers, messy as they may be. That’s what the teacher is for – to guide their improvement into becoming passable writers. That said, if I were a prof, I might have AI do the grading!
Whenever I have to write something substantial or important, I usually take a long walk to get my thoughts together; only then do I start writing. Admittedly. i will take walking-and-thinking breaks again and come back to the writing, so perhaps the two feed off each other. However, I am not convinced that, for me at least, the writing process is the most important part to organizing and clarifying my thinking. I do re-read and heavily edit my writing but is this improving the clarity of my thinking or just better conveying my thoughts to the reader? I’m not sure.
Do other commenters here begin by taking long walks?
I used to take long walks when thinking through something that required a long chain of logic. Something about the step, step, step cadence that helped me stay on course and keep my mind from wandering. Otherwise I don’t take walks, but I will ponder the problem off and on until I think I have something useful to say. (Sometimes I fail. :-)) In other words, I think before committing words to screen. And then I edit. Ruthlessly.
I like to either walk or run to get my thoughts together. It’s an automatic process most of the time. The thoughts will appear and then I work around the ideas. I used to run or walk with headphones and listen to podcasts or music, but I found that my brain was working on other things and as a result I’d miss much of the discussion. So now it’s just me and my brain, and a notepad waiting in the garage when I get home.
Writing encourages the thinker to think clearly and precisely, because what you’ve written is now convenient for editing in a way that saying something isn’t. Even recording what you say and playing it back the next day, like writing something and editing the next day, isn’t as convenient, and that convenience tends to maximize the clarity and precision of thought.
By writing, what is often meant is typing on a computer keyboard. There are important differences between this form of writing and writing by hand on paper. With a computer it is easier to correct mistakes and move text around, something that is harder to do when writing on paper. Writing on paper demands a greater initial clarity of thought.
Constraints help.
I was reminded of a series of introductory books for which the publisher’s constraint was that the author not refer the reader to other sources — no telling them to look in another book for information to understand the current material. Maybe a further reading list was allowed.
Trying to express an idea through speech imposes interesting constraints as well, especially for mathematical ideas. A professor of mine used to fall asleep in the middle of the weekly colloquium. She invariably woke up to the applause at the end. But I wondered what a speaker could do to prevent her nodding off in the middle. Cymbals.
Talking to oneself is also a good method of clarifying one’s ideas. When done in public or at the workplace, it has the added benefit of keeping people away from you.
Except for the men in white coats.
This is my comment posted in the NYT:
“Writing Isn’t as Important as We Think.”
Neither is reading, perhaps? Mr. Cruz and other college professors contribute to the development of students who cannot write and do not read anything, but somehow think critically. They cannot express their thoughts, but this is apparently fine.
That you succeed in your aspirations is why I look forward to seeing what you have to say every day.
Writing is not essential to thinking. It can be a useful exercise in self-examination, as when you try to put thought on paper and discover a misplaced confidence in your understanding. The writing showed the gap; it did not show how to bridge it. Neither will any writing instruction in grammar, mechanics, style, or organization if the problem is with the thought rather than the expression.
The Socratic method can also provide this examination, but as others note, it isn’t appropriate for lecture halls with hundreds of students. But neither are graded essays. Do students in disciplines that are neither writing-intensive nor seminar-heavy not learn how to think? I would ask the same of skilled craftsmen, artists, pilots, surgeons, and many others whose core professional skills seldom demand putting words on paper.
Even where writing is assigned, I wonder how often it does the work people ascribe to it. Those who reflect fondly on learning in their writing-intensive courses how to think and how to express their thoughts strike me as having had an education that is far from typical today. I assume many of them learned the basics in high school; that is no longer a safe assumption. It also seems that they are disproportionately from among the more elite colleges and universities, where classes are smaller, resources greater, and students have more ability to teach each other and themselves.
But what of the common model that has overworked and underpaid adjuncts or instructors on 4/4 and 5/5 teaching loads with 80-100 essays to grade for every writing assignment? Do we really believe the assignments are adequate and the feedback sufficient? Perhaps we long ago opened the door to AI replacing the work of both students and instructors.
Maybe that’s why G_d created TAs.
I am not a teacher and will defer to those who are with respect to what happens in the classroom. I agree with much of what has been said above and will note only that to me Cruz’s dismissal of writing is appalling. His Socratic comparison is outdated and overrated. My personal experience is limited, of course, but I attended law school back when a fair amount of Socratic method teaching was still employed. Useful, yes, but hardly a mechanism that diminishes the utility of writing.
While my comment here should not be viewed as an example, like many others I often use writing to engage with a topic I wish to address, to hone my thoughts, and to attempt to obtain some measure of eloquence. My intellectual abilities, such as they are, would be far less without the leavening power of writing. I shudder at the thought of the “educated” people of the future relying on a software program to express themselves through the written word.
See C.M. Kornbluth’s 1951 sci-fi story The Marching Morons. Back then it earned an award for fiction; maybe these days it’ll get another one for prophecy.
https://en.wikipedia.org/wiki/The_Marching_Morons
Prof Coyne. I have read much of your work and much of Dawkins’ and you seem to me to write with comparable skill.