“It’s not just the problem of brazen cheating. In some ways, the more insidious threat LLMs pose to undergraduate learning is the promise of instant shortcuts.”
.
That’s Paul Sagar, Reader in Political Theory at King’s College London, writing at Unherd.
He continues:
Why struggle through that difficult article, why read that complicated book, why force yourself through the problem set, when the internet can just summarize it for you?
The answer to which is: because it is only through the struggle, the forcing, the wrestling with ideas for yourself, over the course of years, that you can truly train and develop your mind. Indeed, this is the reason university humanities degrees put such a high premium on writing. Writing is thinking. Until you have tried to put your ideas on the page, you never really know if you understand them and have them under control.
Unfortunately, the truth of these facts only becomes apparent with experience—which is exactly what undergraduates lack. You may also be surprised to hear that people in their late teens and early twenties tend not to be good at putting off immediate pleasure in exchange for distant reward. Traditionally, one thing that universities were good at was teaching young people this skill by forcing them to acquire it. (One learns by doing.) LLMs, however, pose a direct threat to this entire process. They are a quick-fix drug dangled before students’ noses whose true effects appear to be the stunting of intellectual development.
That’s the problem. What’s the solution?
By this point, it is abundantly clear that the only pedagogically robust response to LLMs in universities is at least a partial return to traditional methods. Reliance on online coursework has to be reduced; a significant return to paper and pen is required. This is the only way we can guarantee that students are not cheating in (all) their submissions. It is only by demanding that they prove their knowledge directly, in person, that we can incentivize them to go away and learn properly in their own time. Everybody in higher education knows this already.
But one problem with that solution is that it’s not clear that university administrators are on board with it. At KCL, Sagar says, faculty were told that “a return to having majority exam-based assessments was straightforwardly not an option.” In part this is owed to how exams are typically administered at UK universities; Sagar notes “we have more students enrolled than we could possibly fit into exam rooms during a finite exam period.”
And there are bigger worries. Universities have long marketed themselves on the basis of the improved employment prospects of their graduates. If many students are going to college solely for the sake of improving their job prospects, but the increasing use of AI by employers for white collar tasks decreases the number of job opportunities available to college graduates, students may not see college as worth it.
You can read Sagar’s piece here.
Meanwhile, Kyle Saunders, a professor of political science at Colorado State University, has developed a website, based on his research, that provides an analysis of the “institutional resilience” and “post-college market position” of US colleges and universities. The former refers to how well the institution can absorb financial and enrollment shocks, while the latter is about how well the institution positions graduates for the labor market ahead.
(Data for it comes from IPEDS institutional characteristics, finance, enrollment, and completions (2024), College Scorecard institution-level outcomes (most recent cohort), WICHE Knocking at the College Door (11th ed., 2024), O*NET Database (v29.0), NCES CIP 2020 to SOC 2018 Crosswalk, Anthropic Economic Index (August 2025), and Census Post-Secondary Employment Outcomes (PSEO).)
He provides a map of 1,556 US institutions of higher education—which Saunders refers to as “a strategic classification heuristic, not a ranking or a predictive model”:
At the website, the above map is interactive; you can click on each of the dots, which represents a different institutions.
Or you can just search for your institution. For example, here are the findings for my university, the University of South Carolina:
You can read more about Saunders’ methodology and search for your own institution here.
UPDATE: Earlier today, Inside Higher Education reported on how popular course management software Canvas now has agentic AI features “which can automate ‘low-value’ tasks for faculty such as rubric generation, content alignment and discussion reviews.” Reportedly, “In an effort to keep humans in the loop, it purposefully built guardrails into Canvas designed to prevent instructors from fully automating grading.” Yet:
Some education experts worry that integrating agentic AI into the classroom as a time-saving measure will give institutions leverage to increase class sizes and faculty workloads. “Eventually the question may become ‘If we have so many faculty just using agentic AI, what is their value and purpose?’” said Jason Gulya, a professor of English and media communications at Berkeley College…
“If a student knows that a message or rubric was created by AI, we need to think about what it does to the relationship between the student and the professor,” he said. “We’re going to ask students to do something difficult, and if we use this technology in a way that distances the educator and student, they’re not going to do that.”
And if students and instructors begin offloading too much of their work to AI agents, Guyla said it could eventually result in a classroom mostly void of human interaction and engagement.
“That’s absolutely possible if we’re not careful,” he said. “Ed tech is often pushing us toward that dead classroom theory. There’s a chance to rethink it, but it’s going to be on higher ed to do the heavy lifting.”
In other words, LLMs show us that capitalism coupled with uncontrolled technological progress will not only kill us but will also make us dumb (before killing us). Not much of a surprise there.
“LLMs show us that capitalism coupled with uncontrolled technological progress will not only kill us but will also make us dumb (before killing us).”
Whereas *communism* coupled with uncontrolled technological progress will, one presumes, keep us healthy and intelligent, by some magic or other??
I wasn’t aware that capitalism and communism existed in this kind of dichotomy!
Naturally, they don’t — it was just a joking comment on the thoughtless way in which so many of the educated people who should know better mindlessly lump in capitalism with whatever they dislike, or give it as the cause of something they dislike, despite the lack of any clear connection between the two things, and despite the lack of any clearly better alternative to capitalism. It’s a sort of knee-jerk reaction that becomes contagious if people don’t call it out: if other people are constantly blaming capitalism for everything under the sun, well, capitalism must be bad, and we ought to fight against it. The more this passes without question, the dumber our conversations become.
Fair enough, but I think you’d have a hard time arguing that there’s absolutely no connection between the present system of global neoliberal capitalism and the proliferation of AI technologies.
If only AGT’s claim were simply that there is some connection between the present system of global neoliberal capitalism and the proliferation of AI technologies!
I mean, that was basically the thrust of what they said. They’ve acknowledged below that their rhetoric might have been a little hyperbolic, but the point is simply that problematic AI dependency is, to some degree and in some sense, an outgrowth of capitalism. Obviously a full picture of the connection would be more complicated than we have space for here.
There is probably scope for a robust Marxist analysis of the issues here but I think, even in the absence of one, what AGT said is just plain obvious. These technologies emerged within a thoroughgoingly capitalist landscape and much of their spread and use is dictated by the interests of the techno-capitalist class, which holds an even tighter grip on power now and, by virtue of that, shapes our relationship to this technology—whatever our opinions on the tech itself (from a purely technical or even philosophical perspective) are.
I mean, if they had emerged in a non-capitalist landscape (assuming you don’t like the word “capitalist”), they probably wouldn’t be optimized for engagement but would instead either be non-optimized and uninteresting, or be optimized for something else. But that “something else” probably wouldn’t be “edification.” It would likelier be something like “straightforward indoctrination.”
I meant to say, “assuming you don’t like the word ‘communist.’”
Maybe. But there are other options than capitalism+optimizing for engagement and non-capitalism+optimizing for indoctrination that I think are possible and desirable.
And that’s assuming that capitalism doesn’t optimize for indoctrination, among other things—which might be true in an environment where the capitalist class hasn’t radicalized. But that’s not the environment we’re in.
Of course it will! Not sure though that I made an argument for communism…
In so far as the argument in the article depends on the idea that students turn towards AI is in part as a response to an environment that makes the main incentive for getting an education one’s ability to get a wage paying position in a capitalist economy, it strikes me that we should expect any non-capitalist system, including communism, to at least dull that incentive.
It is far from a panacea to all influences or effects of AI, but it still seems relevant.
Point there. Not sure I see a single argument against what I said in the previous comments other than that I am dumb and have knee-jerk reactions. So, really, we don’t think that largely unregulated companies (in the US) driven exclusively by a profit motive and having almost exclusive control of potentially world changing technology will have or at least are likely to have a deterimental effect on our continued existence? I didn’t think I said anything controversial. But evidently, I did.
Of course, I said it in a satirical tone and with some rhetorical exaggeration and one can pick up on that, but otherwise, where is the major hair-rising crazy communist absurd claim?
It’s a little sad, really, to see people leaping so trenchantly to the defense of the present economic system. One does not have to be any species of Marxist to acknowledge that it is a deeply flawed model fraught with injustice. And, at any rate, the instinctual aversion to quote-unquote Marxism is tedious. If there are problems with that view, or set of views, they deserve to be tackled and discussed on their own terms.
You also don’t have to be morally depraved to think that a deeply flawed model fraught with injustice may still be better than any other system humans have so far developed for organizing billions of people in a complex and changing world! Many problems that we have under current capitalism are just as likely under other systems, or are replaced by other problems. The instinctual aversion to “capitalism” is just as tedious as the instinctual aversion to “Marxism”. The question is how to make the world a better place, not why capitalism makes us dumb and kills us.
I doubt that one has to save capitalism, as it exists today, to save humanity. It seems to be the contrary. Again. My remark was a short one and largely rhetorical and satirical. To say or imply that late capitalism is the best of all possible worlds really does not do justice to any of its alternatives that are not Marxist, communist or whatever. There are also many alternatives that keep private property. Even a decent liberal egalitarian or market socialist system would do much better than the present system.
The danger with abstractions such as capitalism, socialism, etc. which you philosophers know better than me I’m sure, is that they always incompletely characterize situations leading to distorted thinking (A. N. Whitehead taught me that). Regarding opening investigation into alternatives, Hannah Arendt’s On Revolution provides a rich analysis of the problems in the constituting of post-revolutionary governance in all the revolutions and post-colonial independence movements following the French Revolution of 1789. She sees a major problem in how we have constituted our our states and societies based on theories that emerged out of the “failed” French Revolution, whether they have abstractly called themselves capitalist, socialist or something else. Even for the United States, which is an outlier in constitution writing, she is able to refer to letters by Thomas Jefferson warning of the dangers of not constituting local forms of organization that could bring people and their communities into a direct role in governance, control and consent over the federal government (he’d argued for “wards” as the main actor). She sees parliamentarian or congressional forms of governance ruled by political parties in all cases as representing the “defeat of the people,” in whose name political parties tend to rule while constitutionally characterizing themselves as representing “sovereignty in the people.” She sees parties and parliament as opposed to the councils, assemblies (communes in France, soviets in Russia and China, etc.) and other forms of organization either traditionally embedded in societies or else arising through the course of revolutions, only to be destroyed by the political actors in the parties. I’ve seen in Nepal in the last thirty-five years the destruction of the traditional forms of organization known as “guthi” following the creation of parliamentary government by parties in 1991, with bureaucratic and corporate stripping of their wealth and displacement of their roles, and even before that the dismantling of “people’s governments” or jaana sarkar and their armies which had emerged during the revolution of 1950-51 by party leaders, which had left the those same party leaders bereft of any means to confront a means to confront the monarchical counter-revolution that occurred in the subsequent decade that was to last three decades, and since then a party autocracy that was overthrown by a mass youth movement this last winter, albeit without strong theoretical foundations leaving the constitution the same. Max Weber’s work furthermore provides an investigation of bureaucracy and bureaucrats, which in the second Index to the UCBerkeley version of his Economy and Society he described as “the real rulers of the modern state.” The anthropologist John Bodley saw what he calls small-scale states (if I recall properly) governed by localized forms of organization, which he sees as the antidote to the extreme centralization of the modern state and its state and corporate “bureaucratic despotism,” in Weber’s and Arendt’s words, or new form of “feudalism” in William K. Tabb’s book on global corporate governance.
“The instinctual aversion to capitalism is just as tedious as the instinctual aversion to Marxism.”
Agreed. But that’s not what I was talking about in my earlier comment, which was a response to the appearance of the former instinct in the conversation, not the latter. Your insistence that the latter also be taken into account betrays more about your own predilections than anything else.
And did you overlook the part where I acknowledged that one needn’t be any variety of Marxist to recognize the flaws of today’s prevailing version of capitalism? You seem to agree about that, but then you default to the disappointingly trite, and empirically unconfirmable, claim that there exists, and can exist, no better alternative to capitalism.
Whoops, switch around “latter” and “former.”
“ And, at any rate, the instinctual aversion to quote-unquote Marxism is tedious. If there are problems with that view, or set of views, they deserve to be tackled and discussed on their own terms.”
The 20th century showed up some problems with it, I think.
To repeat, I said nothing Marxist. To begin with, I didn’t criticize the system built on private property. I was criticizing unregulated late capitalist tendencies and institutional systems. To picture this as a dualism – today’s system vs Marxism or communism or whatever – is distorting my views as well as the possibilities. There are many entirely reasonable alternatives to today’s institutional system and none of them have anything to do with the things that had happened in the previous century.
Yeah I didn’t say, or imply, that there were no problems with Marxism. I said that if there are any, then we should do justice to the complexity of the situation, an aim to which the whole “The flaws of the USSR = a knock-down argument against Marxism in general” schtick is completely orthogonal.
I’ve got my popcorn, my candy, and I’m ready to watch all the LLM apologists rush in to explain why this is all totally inevitable, and there’s nothing we can do about it so we may as well embrace it.
I feel like I’ve responded to a version of this comment before. I’m definitely one of the resident “LLM apologists” here, but I completely agree with everything I’ve read of the Sagar piece. (I can’t read the whole thing.)
I think there are all sorts of useful and valuable things LLMs can do, but I also think they pose an extraordinarily dangerous temptation for students if we don’t do things to re-structure some features of assignment and education. I’m working with my university to set up a pilot of a computer lab where students log in to work on writing assignments on a computer without LLMs, but with all the readings, because I think we need people to practice long-form writing over several hours at a word processor, and not just 45-minute bluebook exercises, or yelling at AI companies but not doing anything to re-structure our assignments.
The forklift and the automobile didn’t kill the idea of lifting weights and running for exercise, but they did mean that we needed several decades of thinking how to re-structure our lives to properly invest in our bodies rather than coasting on machines. We probably still haven’t figured out the right solutions for those issues, though we are making progress – and it would be a mistake to say the machines are useless or should be eliminated, even if we acknowledge that we need a way for people to develop skills that the machine tempts you away from using.
“I’m here to wink at the fact that exchanges like this have happened before and probably will again in this thread.”
“Um, actually, I’m here to tell you that exchanges like this have happened before, and insist that they do so again in this thread!”
I assure you, I can yell at AI companies while also being proactive about restructuring my assignments. The difference in my approach is that I refuse to cede a single inch of ground to the highly tendentious and nefariously motivated idea that the percolation of AI technology into the classroom is inevitable. My efforts at restructuring revolve around preventing AI (ab)use, not “riding the tiger.” Like the assault rifle or the nuclear bomb, we might not be able to un-invent the technology, but we can absolutely advocate for its usage to be strictly regulated, and we can take measures to do so ourselves. The contrary idea is, forgive the hyperbole, neoliberal mythology.
While the problem is real and well-expressed here, attempting to go back to a writing-only mode surely cannot be the answer. Socrates regarded writing as atrophying to the mind, preferring the strenuosity of verbal debate – which indeed is a skill that many of us have lost, but humanity chose to take the gain of higher productivity and capability that writing enables. And how employable would our pen-and-paper students be in what seems to be an increasingly AI-permeated world? Wouldn’t it be better to guide students in the skill of using AI responsibly and discerningly – for which philosophers would seem better positioned than most other faculty? And then, dare one say it, help them take the gains of enhanced capability that AI can deliver?
I’d say that students who have the ability to read, write and think critically all on their own are rare and unique commodities, and therefore *more* employable, in a world increasingly permeated by AI slop producers.
When a philosopher uses “surely,” you can rest assured that what will come after won’t be justified.
Perhaps the most accurate comment I have ever encountered on this website.
Related: whenever a layperson says, “I’d argue that…” you know that what’s coming is a bald assertion.
I like your replacement of “surely” with “you can rest assured”.
The point, from the employers’ perspective, of using AI is that it increases productivity because it doesn’t require specialized skills to use it. You don’t need a subject matter expert, the AI can draft that document for you instead. They can replace knowledge workers with AI prompted by cheaper workers (cheaper either because each of them is paid less or cheaper because fewer of them are paid at all).
So, preparing students for a job market where the point is that there are fewer jobs for skilled professionals seems silly. Instead, students should be trying to get the skills that let them adapt to the changing job market circumstances they’ll face throughout their lives. Skills like thinking creatively, with clarity and precision, etc.
But those are skills best developed using traditional methods like reading and writing. Or at least those are the only methods which have been demonstrated (through long collective experience) to be able to teach those skills.
After acquiring those skills, students can turn them towards “using AI responsibly and discerningly” or whatever other task they set themselves.
> The point, from the employers’ perspective, of using AI is that it increases productivity because it doesn’t require specialized skills to use it. You don’t need a subject matter expert, the AI can draft that document for you instead.
There may be some employers who think this, but these are only one type of employer. When new tools are invented, there are usually two types of economic response, both of which are valuable. One response uses these tools to take something that had been expensive and artisanal and make it cheap and abundant, even if worse (like mass market clothing and IKEA furniture and fast food), while another response uses these tools to enable people to make new things that couldn’t be made at all before (like railroads and airplanes and electronic music and globalized cuisine). Both are valuable contributions to human well-being, though one enables opportunities for workers who don’t have long training, while the other enables new opportunities for workers who do have specialized skills.
I think there are opportunities for worthwhile use of AI, especially in research (unsurprising that an academic would think of those possibilities first).
But those worthwhile uses have nothing to do with how the industry is developing, the products it’s making, and the way it’s rolling them out. Chatbots and general purpose text, image, and video generation are pretty much the worst possible use case.
The industrial revolution choked cities with smoke, polluted rivers and oceans, and set us on the path to catastrophic climate change. That there are some laudable uses for the steam engine doesn’t do much to make me think the way forward is to adopt the new technology as quickly and with as little foresight as possible. But that’s exactly the trajectory we’re on.
Yeah, we definitely shouldn’t be emphasizing speed over foresight. And we also shouldn’t be emphasizing total opposition over foresight. We should just do the obvious thing of trying to identify the good uses and the bad uses, and encourage the former and discourage the latter.
I’m sympathetic to most of this, but I want to insist that at least a good deal of IKEA products (not all, but lots) are not just cheaper than other options, but also actually better. Lots of the stuff they make is of really good quality, often better than things that cost (or did cost, before there was real competition) a lot more. There are some lessons here.
See my earlier comment for evidence of my prescience.
You AI apologists always seem to default to two tropes: The analogy (really a disanalogy) to the development of writing, and the ludicrous notion that AI is predestined to continue proliferating like some kind of force of nature, and so we should just accept it rather than doing everything in our power to combat it.
Two reasons: it isn’t hard to use AI to augment your skills (once you have some skills), and AI is changing rapidly, such that whatever we teach in four years won’t be relevant.
If all college teaches you is to be a prompt engineer of Chat, you certainly ought not to spend money on college, because prompt engineering just isn’t that hard (which isn’t to say that it can’t be done well.)
It’s not obvious at all that learning how to use chatbots enhances any capability (other than using chat bots). It is an assumption not based on good evidence that using a chatbot connected to an LLM is a valuable ability for everyone to acquire. The only reason why many university admins think it is is that they (or their governments) been marketed at by bodies incentivized by the prospects of money and power beyond measure.
In these discussions I really do think the burden of proof falls on those who want to take the line that learning how to use a chatbot well is a valuable skill. It’s certainly fun, and to many it probably seems as if they are doing cutting edge important work. But it could very easily be a mass delusion. I remember a philosopher commenting to me when people first started using email that one had to be careful: it can seem like you’re working when you’re typing emails at the keyboard, but actually just be sucking up a lot of time that could be used for real (i.e. creative) work. I think that’s true in spades when I see people in public with their heads in their phones, responding to various apps, many of which are connected to AI. I’m not even talking about social media here. People are doing this that and the other, responding to someone or something that has figured out how to exploit AI to be an additional middle man in the line of middle men between a buyer and a supplier of some needed service or product. So yes, in that way learning how to use a chatbot is useful. That way you can talk to the chatbot well while you hope to be connected to a hotel when you’re stranded in the airport or you can type to the chatbot well, when you’re hoping to have your job application considered by a person who shares a calling etc. etc.
Again:Burden of proof is on those who think it’s a useful skill for people in general to learn how to use AI
Everyone has little annoyances in their life that could be improved by a small website or app or other bit of code that is tailored to their particular circumstance. People who learned to code decades ago knew that, and were often writing little things for their daily life.
LLM chatbots give us all that ability, and it is useful to learn how to take advantage of it (especially since most of us are so used to just resigning ourselves to these little annoyances rather than recognizing them as things that can be fixed).
Nevertheless, I think this skill is far, far less valuable than the skills of writing and reading and thinking that LLM chatbots are tempting some students out of learning. Universities should generally be focusing on these traditional basics (as well as all the millions of disciplinary skills out there, some of which will soon involve LLM chatbots) rather than changing focus to LLM-use skills, as Matt seems to be suggesting.
I appreciate the observation that AI provides equal opportunity for those of us who do not know how to code. ChatGPT or Gemini will cheerfully spell out all the steps for building a website, for example. If you trust it.
Someone (not you I suspect, but another person) might mistakenly draw the conclusion that rolling out AI the rash way that is being done, with all the damage being done to education, is justified because it evens the playing field. It provides equal opportunity to non-coders. To build a more equitable society we should teach people prompt engineering and generally encourage AI use. (This would echo familiar terrain from the past – the dream many people had in the early days of the internet that less well-off people would have access to all the knowledge that the privileged people had/the government had.)
So I just want to say that the equal opportunity argument is not a justification.
Let’s say you have a country that has the draft – it drafts people belonging to Group A when there are wars. It hasn’t engaged in a war for a long time, a century, let’s say. Let’s also say that the country has recently undergone a thorough discussion and examination among all its citizens on the status of people belonging to Group B, including equal opportunity issues for group B people. Let’s say someone makes a long list of examples of failed equal opportunity, and that the country decides to rectify all the items on the list. One item is that Group B people have never been drafted or fought in a war. So the country decides to throw a war. It starts bombing some other country, killing that country’s citizens, endangering its own citizens through backlash, in order to give Group B people an equal opportunity to be drafted and fight in a war.
I would submit that there is something off-base here. This is not a good way to make a fairer society. It seems that it would be better not to have a war, than to give Group B people equal opportunity to die on the battle field.
Same with rolling out AI in the way that is currently being done: not a good way to create a fairer society.
I do think that students should learn something about the use of AI tools, and there will probably soon be more and more such skills that people in any particular field should learn. But I don’t think that teaching AI skills should be any more central than teaching e-mail skills, file management skills, database skills, or calculator skills. The central skills should still be reading, writing, talking, listening, thinking, being creative, and the like.
As a teacher, what I don’t understand is other teachers who don’t see generative AI use by students as an existential threat to the entire project of undergraduate education. Did they not have the experience of growing and developing new and valuable skills through struggling with precisely the things AI gets today’s students out of? How can so many teachers STILL be so on board with the ‘use AI in your classes’ and ‘online courses are as good or better than in-person’ school of thought?
All of the explanations I can come up with are depressing, unfortunately. At the very least, the gap between those of us raising the alarm and others who dismiss it seems really hard to bridge.
I agree; the disconnect is very disheartening. I can only assume the AI celebrators will be grinning and cheering as they empty their offices, after their department gets shuttered because AI has made them “obsolete”.
I don’t think that the problem is the use of summary. Accurate summary is a good place to start. I think that the problem is that students tend to go no further once they have the summary. We aren’t going to get rid of AI’s ability to summarize, but we can think about ways to encourage students to go further than the summary.
The mental exertion of generating one’s own initial summary is a preliminary step to expanding and deepening one’s understanding. In my experience, those unwilling to do the former on their own tend to undervalue the latter as well.
I don’t start by generating my own summary when approaching a new literature. I usually read Phil Compass and SEP articles, and book reviews, and only after having done that, decide what to dive in on.
I agree that actually learning how to read in a way that enables you to produce a useful summary is an extremely important skill that students should learn. But I also think that summaries are useful artifacts that aren’t just exercises for their creator – they actually produce something valuable, and having easy access to huge numbers of them on any topic you want is a good thing. (This is why we celebrate Phil Compass and the SEP, rather than thinking that they are awful sites that devalue education.)
“I don’t start by generating my own summary when approaching a new literature. I usually read Phil Compass and SEP articles, and book reviews, and only after having done that, decide what to dive in on.”
We’re talking about two different things. Simply orienting yourself in the literature is one thing (even then it’s advisable to use the curated advice of experts, rather than a tech billionaire’s toy), but what do you do after you’re done situating yourself? The responsible researcher reads the “primary literature” and then proceeds with what we’re calling the work of summarizing.
“I also think that summaries are useful artifacts that aren’t just exercises for their creator – they actually produce something valuable, and having easy access to huge numbers of them on any topic you want is a good thing.”
Perhaps I need a more detailed explanation of this evaluative claim, but on the face of it I disagree. The point of textual exegesis is to arrive at a reconstruction of the author’s positions which will be as charitable and accurate as possible. Typically, this requires several rounds of reading and revisiting, and the “summary” phase is valuable only insofar as it contributes to and facilitates the more rigorous interpretation. And, as I said earlier, the exertion of generating a summary functions as a launch pad for further, more granular, inquiry. Skipping steps to speed up the process is actually counterproductive, and, in my opinion, a little irresponsible.
I’m not sure what steps you think are being skipped. It’s absolutely best to start with human expert-written summaries of all the fields that might be tangentially relevant, to orient yourself for your own reading that will come later. But we don’t have expert-written summaries of all fields. The fact that we can now get chatbot-written summaries of a bunch of fields quickly and easily is a benefit that partially fills this gap.
There’s still the really hard work in education of getting students to understand that there is value in reading the originals, rather than just summaries, and it’s true that this has gotten harder now that summaries are so easily available. I don’t think anyone in this thread is denying that.
Every step in the traditional exegetical process – including summarizing – is important, no matter how tedious it might seem. We call it “rigor” for a good reason. As much as this might gall you, in most instances it is simply the case that the process cannot be expedited without compromising on quality, and certainly not for undergraduate students who are still developing reading competence. On principle, anything that purports to let you do less reading is bad for your long-term academic health.
Again, we shouldn’t conflate “summarizing,” which involves actually reading a text and trying to extract its main contentions, with “orientation,” which merely involves discovering which texts one should proceed to read, in a certain subfield.
the key word here is “accurate” and this is still precisely what llm chatbots are absolutely not.
Have you actually used recent LLM chatbots to test their abilities? They’re now producing things that are more accurate than a lot of Wikipedia pages that still manage to be useful, even if neither is as good as an actual human subject-matter expert (who will also have some inaccuracies, some of which are more relevant to your needs than others).
indeed i have and i find that llm chatbots are simply not trustworthy *unless* you actually already know the thing you’re asking about and can prompt accordingly.
i take the person to whom i replied to be saying that there’s nothing wrong with summary per se, and therefore if the llm can accurately summarize, the issue here isn’t summary as such.
my reply is meant to suggest that we simply can’t rely on even the very best llms to accurately summarize for the lay reader because they consistently make errors due to the way in which they are built. though the problem of grievous hallucinations has been decently mitigated in many cases, the fundamental issue remains that without knowledgable oversight, llms do and will continue to confidently state things that aren’t true, will “misread” sources, and so on and so forth.
the issue isn’t whether llms can’t be right. they can be and sometimes are. but they are wrong *enough* of the time to be highly problematic for even basic search queries. if your preferred media outlet produced the same error percentage as the llm chatbots do, you’d not trust them.
when we tie this to education, the problem becomes clear. if twenty years ago our textbooks consistently contained fundamental errors and confidently stated misinformation that no one corrected and our students believed to be true, we’d be failing our most basic jobs as educators. but llms do this every day and all the time. the error rate is flatly unacceptable *for teaching purposes* that have to with anything beyond the most basic facts about the world (and even then…).
so when our students don’t read the assigned texts and ask the llm instead, and they they don’t actually write the paper or discussion post or even outline themselves about the thing they didn’t read there is, i’d argue, no educational value here.
does that mean that ai is useless or that knowledgable people can’t use it well? of course not. but should our students trust it? should any lay person trust it?
> they are wrong *enough* of the time to be highly problematic for even basic search queries. if your preferred media outlet produced the same error percentage as the llm chatbots do, you’d not trust them.
> should our students trust it? should any lay person trust it?
I think these are difficult questions in social epistemology! Some philosophical discussions of expert deference seem to act as though human experts are more reliable than they actually are, and as though there is a qualitative difference between experts and non-experts, rather than a continuum.
We’ve been through this with newspapers, cable news, wikipedia, and so on, with new layers of intermediaries that give less accurate but more easily available access to expert ideas, and each causes problems and also causes benefits.
We have to figure out what skills in using the new tools help us maximize the benefits.
They are far more accurate than 99% of college students in the United States I will guarantee you.
And how are these college students ever going to improve their own abilities if we let them keep outsourcing them to machines?
We absolutely shouldn’t let them outsource for most of their work, just like we shouldn’t let them write papers that only cite Wikipedia rather than the philosophical papers that contain the arguments they are responding to.
But we also probably shouldn’t just assume that there is *no* point in *any* educational process where access to Wikipedia or an LLM is helpful.
Anyone who agrees that “[b]y this point, it is abundantly clear that the only pedagogically robust response to LLMs in universities is at least a partial return to traditional methods” should join other profs by posting their course on https://certifiedaifreeskillsandknowledge.org/
I appreciate a good grass roots accreditation project as much as the next person. But I’m curious why you also intend the database to include instructor-reported (or also student-reported?) student enrollments. At minimum, this strikes me as costly to do well, because it requires a procedure to link up names and nicknames and typos across multiple courses to generate an output per interested, accessing user. Meanwhile, isn’t it sufficient to simply encourage students to state on their resumes/vitas that they enrolled in courses certified as AI free, perhaps with the link to the proof that their semester’s indexed course was certified as AI free?
I don’t see why everyone who thinks that we need at least a partial return to traditional methods would think we need 100% elimination of non-traditional methods.
I might have missed an explicit statement to that effect, but I don’t see it as an entailment of the general point. 100% elimination of non-traditional methods on an ad hoc course-by-course basis in a larger curricular setting isn’t 100% elimination simpliciter.
Meanwhile, 100% elimination at the individual course level *is* an administratively cheap way of creating an honest method for a student to signal their own effort (given opportunity) to deliberately seek out such coursework. And that could wind up being a valuable good in the hiring end of things, at least in certain occupations or else for market differentiation among applicants.
That’s a useful point – having some 100% AI-free classes could be a good thing, though I don’t know how to structure this thoughtfully rather than just driving some students and professors to polarize themselves into doing this for all of their classes or none of them.
I think it is important to recognize that so many (and I would guess most) professors make it easy to cheat using AI, which is really unfair to those students who do not cheat. Nor is all such cheating counterproductive to learning. Students, especially in more difficult majors, often reasonably choose to cheat with AI because they save time on less valuable assignments by doing so and so can devote more time to preparing for difficult in-class exams, etc. Thus, not all cheating means that one gains less from the class. I know of many specific instances of honest students who suffer this sort of unfairness. One of my students, for example, took a Physics course in which she was in the upper quartile for all in-class exams and the lower quartile for all of the lengthy and difficult homeworks that could be aced using AI. The professor would have to be an idiot not to know that all those 100s on the homeworks were achieved through cheating. But because of the cheating, my student didn’t do nearly as well relative to her classmates as she should have when the final grades came out. Class after class is like this in this student’s experience, and it is demoralizing for her. I respect this student’s honest, but, in my opinion, if professors do not address cheating, they have no right to complain that students cheat.
I would take slight objection to the last sentence – we have the right to complain about cheating no matter what we do. But I completely agree with the central point – if one thinks that there is value in having students not use AI for certain assignments, then one is a bad teacher if one doesn’t incentivize students to not use AI for those assignments!
We used to be able to just coast on the fact that general education requirements force students to take certain classes even if they think (incorrectly) that the class is valueless, and then coast on the fact that students have to do the assignment if we assign it. That’s been taken away from us, but many people are still just coasting.
I suppose that professors have a right to make ridiculous complaints, but if I were a student who cheated only because professors made it easy to cheat and because I would be put at an unfair disadvantage by not cheating, I would laugh at the professor who complained about my cheating.
Kaila
I think you are being a little naive about the nature of cheating. One does not just cheat when it is convenient. Rather, one accustoms oneself to cheating, and in time becomes a cheater (as one becomes a smoker who can no longer control the habit). Think the ring of Gyges.
But if all the other shepherds are using their rings, what is Gyges to do? Give up his chance for a royal internship or preserve his soul? I suppose he should preserve his soul, but it’s still not fair.
What about those of us who like Dr Sagar, are forced by our universities to use assignments that are trivially easy to cheat on? (Not a rhetorical question, i would really like to know)
I don’t have a general answer, but if the assignments are difficult or time-consuming, I might consider telling my students to cheat to even the playing field.
In much of the UK, the students’ grades will be 100% determined by one take-home exam / essay
Yeah, I don’t know anything about UK universities. I am really talking only about US universities
Who would have thought that higher education would come down to Blue Books versus supercomputers.
The showdown of the century!
Saunders’s website is really cool.
Meanwhile in mathematics:
Watershed Moment for AI–Human Collaboration in Math Twenty-first-century Fields Medal proof formalized for the first time.
And even in film:
How Gossip Goblin is Becoming the Auteur of the AI Age.
But philosophers? Nah, they just can’t figure out how to use AI. Well, some of them just can’t figure it out. Universities will still exist, and they’ll still have philosophers, but not philosophers who can’t figure out how to use AI.
I dunno man, maybe some of us just have principled objections or genuinely see no good use case for our current situations? Maybe some of us think ‘x is happening’ ≠ ‘x is good’? Maybe some of us are interested in that second part, about whether it’s good? And maybe we’re a little skeptical of that, and have reasons why? Maybe even we are open to using AI in some cases but object to the particular ways in which our bosses think it should be used? Maybe there’s more to this than that we “can’t figure it out”? Maybe???
Some philosophers can’t figure it out (e.g. our Luddite friends above). But lots of other philosophers are trying!
Your eagerness to introduce unnecessary and counterproductive changes to our pedagogical practice simply because the tech oligarchs have convinced you it needs to happen doesn’t make you a visionary. And our unwillingness to do so doesn’t mean we’re incapable of it, as this dismissive comment implies.
I would take this comment more seriously if it contained any original contentions, but this is nothing but the same old AI fatalism that we’ve seen over and over again. Maybe you and the billionaires whose PR campaigns you’ve imbibed will succeed in freeing humanity from the burden of thinking, but if that does happen it will not be because of some metaphysical necessity, nor for lack of opposition from those of us who care.
Fully online asynchronous instruction is doomed to go fully automated.
The good news is that many students don’t want this type of instruction.