College instructor turns to typewriters to curb AI-written work

Then we did a university reform, partly with the excuse of aligning with the rest of the EU within the Bologna process (and I say "excuse" because that's what it was, because the politicians introduced some things with that pretense that weren't like that in the rest of the EU at all, and it was perfectly possible to comply with Bologna without doing them) and partly to copy the US/UK ways. And one of the pillars of that reform was continuous assessment, and evaluating coursework.

As a consequence of this, first of all working class students were royally screwed. Because suddenly it wasn't OK to just organize yourself to prepare the exam, you had to attend lots of sessions to earn points, which put students who work at a disadvantage. And second, passing by cheating became possible, even before LLMs. People tend to forget that before everyone got access to ChatGPT, some people had access to experts (family members, or even paying someone to do the work).

Now that this kind of cheating has been democratized and everyone can do it instead of just the most privileged with access to experts or money to pay them, people act all outraged. Although pretty much nothing is being done, except for using snake oil detectors, or sometimes increasing difficulty of assignments to make them LLM-proof (with which you screw the students who actually want to learn without LLMs).

They spent years indoctrinating us (professors) in training courses on how the old exam-based ways were wrong (the "Napoleonic" model, they called it... none of them seems to entertain the thought that maybe if it had been working essentially unchanged since Napoleon it wasn't that bad, and you need solid reasons to change it beyond "this is old so let's change") and the new ways were the bee's knees. Like in the Milgram experiment, it's difficult for people to back down and acknowledge that they have been wrong, even when the solution is obvious.

The best was when she barely unscrewed one of this big DIN connectors so at quick glance it looked fine, but wasn’t fully connected.

Lots of skills from those old days that have been lost/ignored in the pretence of productivity.

It's a shame that they are also way more susceptible to cheating with AI.

Writing papers is a useful skill to have. And many students aren't very good at that. I taught some classes during my Ph. D. and supervised some students with their master thesis and PhD thesis work. Many students get their degrees without that really getting addressed. At least Computer science degrees in the Netherlands just spend very little time on writing skills. You get students with high school levels of English and Dutch and that's it.

I learned to write properly only when I started my Ph. D. My supervisor made me do it right before he allowed me to submit papers for publication.

AI might actually be good for education long term. It will result in a more personalized approach, which I think is good. There are plenty of ways to test students that are more engaging and interesting for both teachers and students than some of the old ways. You can't fake knowledge when you do a verbal test. Or test people with a good old written exam.

And of course for teachers, you can automate a lot of the verification work. This can be a lot of work.

Which strikes me as a terrible way to teach and test programming skills. If you're teaching to program without so much as syntax highlighting, you're not preparing your students for anything that even remotely resembles the industry they aspire to work in.

Honestly, these days universities should probably find a way to incorporate AI into their teaching, rather than fight it. Anything else is betting that AI will not stick around, which strikes me as a hopelessly naïve bet. Especially for software development.

I don't pretend to have all the answers, I don't know how to teach systems thinking in a appropriate way either. But I'm pretty sure typewriters isn't it, unless your students are hoping to get hired by Ada Lovelace, it's just not going to be relevant.

But there were already heaps of problems with tech in education before AI.

My CS projects were often pretty free-form so in theory I could've just used AI - today, anyway. But a big part of the grade was a face to face interview where you actually had to talk about the code you wrote. Anyone lifting along with other people who didn't actually do any work would fall through then.

The only exception is that when I got into grad level classes we did have some big programming projects. But most of that programming happened on sparc stations, and it was actually just easier and more productive to sit at the machine in person with its nice big (at the time) display with all the other folks doing programming projects. Those machines had the standard dev toolchains provisioned that weren’t easy (at the time) to do on a dorm room Mac or windows computer.

I really think a lot of the ways we can reduce reliance on AI for thinking is to just set up systems where it’s not an inviting or rewarding option.

If my college is doing this, I cannot imagine how many others are also impeding on their entire goal: education.

Apparently you learn to double check your work!

But yeah, everything was hand-written. On sheets of paper with pencil. I even had to write x86 assembly out by hand for my CPU architecture class. Of course, laptops were available back then but not cellphones and certainly not LLMs, so cheating by electronic means probably presents a stickier wicket now than it did back then.

I'm sure they had some kind of submit your code as assignment and using testing as a way to grade the assignments.

The reason was less for myself and more because anything group related suddenly shot up in quality when the other individual work classmates were graded on couldn't be fudged.

My tests are almost 100% in person. Project work included, you can hand something in, but I'm going over line by line and ask what you did there.

I can do this, because while my school hasn't updated the tests yet, my classes are small and I can do all of them in-person.

It was just expected that you had a grasp of the literature enough that you could argue off-the-cuff in the exam setting, and then you were given leeway if you didn't have exact Harvard style notations to exact date/titles of referenced material.

Student turn-in a mid-term paper. A professor chooses a certain number of points from the mid-term paper and asks for explanations of these in long-hand. Pulling questions this way doesn't seem like it would take more time than a thorough reading of the text.

Oh, but paper reading has been delegated to a drudge you wouldn't trust with pulling question, oh how inconvenient. Which is to say the problems AI introduces to education are strongly related to much of work already being made mindless before AI appeared.

Doctoral candidates do this kind of thing all the time in qualifying exams, but that's after years of graduate school and fresh off doing nothing but reading 100+ books over the course of a few months.

No, High school students can do this. Well, they get impelled to do this. They can't do this now but that's a testament to current education.

Because using AI is the complete opposite of "I learned programming just to make tests easier".

By learning how to program solver, you not only learned how to program but also learned the method well enough to write it.

By pawning it off to AI to solve, you have learned nothing, not even how to prompt correctly as test questions are usually formulated well enough that AI doesn't need prompt massaging to get it.

You can use AI to get some knowledge about the problem (assuming you won't hit hallucination) but that's not what will happen when you use it for test.

And if you DO want to teach students how to use AI effectively, you can just have an AI class...

That is way to high recurring cost that many won't be able to afford. One could get a second hand calculator or even computer, and then additional resources needed was one's willingness. With mandating AI usage, we'd only increase the gap of haves and have-nots. I personally do not like the idea.

An LLM is a force multiplier only, not a replacement. It's a personal assistant to an expert. To use an LLM in a acceptable way, you still first have to learn how to do what it does yourself. I think your suggestion for people to be taught how to use LLMs is justified, but they should do so only after first being taught a no-LLM curriculum. I think this should be entirely after what the notion of an education was in pre-LLM times. Don't incorporate LLMs into our current education, instead teach use of LLMs after our current education.

It's an interesting debate, but I see several reasons not to.

1. As a human you need to learn gradually, e.g. in CS you need to learn the basics of programming before going into more complex stuff. If you embrace AI from the beginning, it can let you skip the basics (why would you code a simple 200-line program if the LLM can do it?) and then you don't have the fundamentals when you reach the more complex level where human thought is needed. It's a similar problem as firing the juniors (because AI can do their work) but then who will become senior if you have no juniors.

2. If you evaluate coursework with the expectation of students using AI, those who pay the $200 subscription will have an advantage over those who pay $20, and in turn over those who use free LLMs. The only way to make it fair would be to provide all students with the best available LLM.

3. While I have heard the analogy with calculators many times, I feel that LLMs are at a different qualitative level. The calculator doesn't really replace human thought, or if it does, it's only some very specific form of it. LLMs replace human thought in a very broad way, so I think they are much more dangerous for learning.

When the scientific calculator was invented, people could easily know what went into its production. As in what circuitry appears in them. You knew that if you bought it, it is yours. Want to program it? Grab a book and do this. The whole package would be a fixed price. You are in control. With AI? You are not at all in control. You rely on a big tech giant (or just like 4 useful ones) who is riding what people controversially still call an economic disaster. You are relying on a technology that is designed to very likely bait-and-switch you. As soon as you get too comfortable with AI, the big tech companies can just bump the prices up and you will not be able to say no. You rely on a technology that you do not control.

The comparison of AI to a calculator or any other technological advancement for students is apples and oranges for that reason.

Imagine giving a student a personal AI datacenter to carry with them. This may be more of a fair comparison.

PS Training students on using AI, especially for free, is setting them up for reliance on the big tech companies and the subscription model.

Why doesn't the essay class allow us asking your parents to write it for them? The art class, why not ask your parents to paint something for you? Geography, why not let ask your parents during a test?

- made tasks easy that were a necessary prerequisite for advanced math (basic arithmetic), but not what the lesson was supposed to be about

- could in theory also let students skip over what they were supposed to be learning (applying the correct operations in the correct order to solve a problem) but doing so would require programming or getting a program from someone else, which the teachers probably figured was a high-enough hurdle to accept the risk

Hence, scientific calculators helped teachers by removing unnecessary friction.

Meanwhile, current LLMs

- will happily attempt to do the student's entire homework for them

- cannot reliably be restricted in functionality to leave the part the students are supposed to do themselves to the student

Hence, LLMs undermine teachers by removing necessary effort.

Sure, in theory LLMs could enable even more focused lessons by removing even bigger unnecessary frictions (e.g. in history class, have a LLM scour a large collection of primary sources to exhaustively list passages mentioning a certain topic), but students cannot be trusted to use them this way.

Hence, teachers are trying to use all kinds of tricks to ensure that what they wanted to teach actually passed through the student's brain at some point.

I would say that definitely shaped me in a way where I rarely bother with the underlying details and tend to focus on how high level abstractions interact. [2]

[1] German "Mathe-LK", we could chose specializing in two things, for me it was math and computer science, the later being quite novel back in 2003. [2] I _do_ tend to specialize in things, but e.g. for LLMs or GLMMs, while I do have the capability to understand the technical details, I just don't bother.

With the calculator analogy, I think that the calculator automates executing certain algorithms (like multiplication, etc.), but using AI takes away some (most?) of the thinking.

They. Are not. The same.

Have you ever known people to commit suicide, kill, or give themselves rare diseases because of their calculators? How about people dating their calculator and going batshit for a software update?

Not to mention learning to do on your own is a useful skill to teach you to think, and an essential skill to (as you suggest) verify answers. People not understanding how things work is exactly why they take bullshit output from an LLM as gospel.

I also note that such arguments tend to be profoundly selfish and self-centred. Your anecdote happened to have an outcome you enjoyed and benefitted from, but I bet that wasn’t the reality for all your colleagues. Just like you are glad for the calculators in your class, some other student may be glad for the lack of them in theirs and it may be the reason they got into their field of study.

Here’s one possible scenario: After graduation, you (or somebody else) shares the program with friend, with a promise to not to share further. Soon enough, it’s on everybody’s calculator. What did real educational thing for you, is just cheat where one needs to press the right buttons and get the right answer. This completely destroys the educational purpose, but significant amount of people just don’t care and want to get a pass.

Yes, there always is a counter weapon by teachers: for example, to point to random line and ask to explain and whatever, but this is not (always) scalable.

I’ve seen this in reality in college, when there was a cs/database course final project implementation, written in Delphi (very popular at a time in xussr), that was passed from year to year, that the professors and ta were so fed up, that I got almost auto pass because I wrote mine in C++…

——

To summarize - the overinreasing amount of pure slop is seen everywhere. Regular multi-thousand line prs where author didn’t even bother to look into code, written by ai. Just prompt -> commit, push, or. Nobody wants to deal with that

Same is happening here - u it’s not to punish people who use tool in proper context, it’s to filter out people who just don’t give a fuck.

The other analogy is taking a forklift to the gym. Sure you lift weights, but you don’t really do any exercise to develop your own muscles.

AI automates a significant chunk of the exercises. So you are left with people who didn’t build any mental muscles.

This would be bad enough, but it’s worse because AI severely benefits experts who have build mental reflexes/taste and can judge / verify output with minimum information.

You would give the brains of the younger generation to American tech oligarchy, a class of people openly hostile to the principles of the democratic rule of law. If you want to see the damage actors like Fox News et alii alone can do, just take a look around in the US. Now imagine them taking over the parenting and teaching role; you wouldn't need gerrymandering if you can control people's beliefs.

Because this makes a subscription a requirement for education, and thus advances the grift that is subscriptions, rent-seeking and dependence on a service. This isn't something we should ingrain into our children from an early age.

Calculators were buy once, use forever. Subscriptions to slop generators are a long term dependency and I want my children to not be exposed to that until they can decide for themselves.

You need to be able to do both things. We don’t need to make it a choice.

Back in college, my assembly class was in MIPS (incidentally taught by Professor Larus of SPIM fame). I remember slogging through writing the assembly to compute factorial and saving registers and then dealing the frame pointer and the stack pointer.

One of the other students had access to a DecStation, wrote the program in C, and ran gcc -S to get the MIPS assembly from it. However, the compiler realized the for loop (and tail call) optimizations and instead of making it a recursive function (and help us practice and understand $fp) write it with a jump instruction instead.

Aside from getting a 0 on that homework, they struggled with the next assignment that presupposed understanding of how write function calls.

---

You could argue that learning C makes needing to learn assembly irrelevant (and MIPS is even less relevant today than back in the 90s). But for learning in school, it's not about the assignment but rather the journey that one takes to get to that assignment and learn from it.

Being able to check the answers that are provided to someone requires the understanding of what goes before and beneath the answer itself.

Writing the assignment in C when you're learning the before and beneath of "this is how assembly works" means that when you later take the complier class you won't be able to debug if your code generation is incorrect.

Working with an AI as the primary tool for learning problem solving keeps the person at the higher level. There is some foundational level that a person needs to learn without relying on an AI to do it for them.

The AI and other abstractions of the underlying problem do allow us to work with more complex problems. Would you trust a bridge built by an engineer who built it with AI and didn't understand the underlying math and physics themselves?

This is especially at issue in college where some students are taking classes to get the requirement out of the way, some are taking classes for that directly - an accountant is taking the math class to do math with the numbers, while the engineer is taking the math as a prerequisite for physics which is a prerequisite for a material science class which is a perquisite for a soil mechanics class.

If you don't understand the various foundational levels without using AI then trying to identify where the AI (or any other tool) got it wrong isn't something that you're necessarily able to do.

Finally, nobody seems to know what teachers are actually discussing. To your assertion:

>So I can't help but wonder whether schools are going about this all wrong. Rather than banning the use of AI and trying to catch students who are cheating, why aren't they creating schoolwork that requires AI?

Many teachers do create schoolwork that requires AI. Many teachers ban it. Everybody's trying to work out policy (to the detriment of other policy discussions, particularly the new ADA landscape). Many ED departments are captured by AI vendors - AI is a normal technological competency for ED majors at different levels. It's not that the discussion is not "are students going to work with AI?" The discussion is "how do we teach?" which is what the discussion always is.

But policy is a part of that. Admin will have guidance and policy statements, and each instructor will as well. Students, who get thrown off balance if they have two teachers with different nav bars in the CMS, want clarity of policy: can I use it? is a different question than should I use it? But "should I use it?" is the much more relevant question for instructors. The instructors passed through the 90s/00s/10s blissfully unaware of anything that was happening in these fields.

>These tools are not going to cease to exist.

Which tools, precisely? Because I'd assert your 200-dollar-or-whatever tier that runs out of tokens on Monday does not functionally exist for most students. I don't know what happens at MIT but Penn State satellite campus students aren't whipping up agentic solutions to "I have a summer online course with discussion boards." They're just plugging that shit into whatever chatbot they have on their phone. Honestly, most online courses aren't even worth that effort, but: different discussion.

The only reason there is pervasive student use is because someone made it free. The CoPilot window that comes with the basic tier of 365 and all the other in-app copilots are what my students have; the Google Docs stuff exists, and the Grammarly stuff exists, and the best way to "ban" it (which is just fine as an approach) is to make it even slightly more expensive. If someone does that, yeah, I think some of these products might cease to exist.

>I think we're doing them a disservice by trying to give them the same education that was given to those from previous generations

Students graduate into a world where all kinds of stuff is going on and all sorts of ideological forces shape what they encounter in schools. "AI" more a dominant element in that force than the calculator was because calculators didn't have hundreds of billions of dollars of investments, and I'm not sure I ever knew what the prevailing political project of Texas Instruments was. Maybe TI's CEO had a manifesto; tbf, I do remember IBM's corporate culture being strong enough to drive cultural change. But it would be naive to not recognize that these products are made by people antagonistic to the idea of education as a compulsory and public good.

World war I was very much that.

However, they didn’t remove the embedded revision history in the .docx file they submitted, so that went about as well as you can expect.

oh look there is a llm trained on key loggers to spew slop at your personally predicted error rate; bonus if it identifies to USB as keyboard.

I graduated in 2020, so I've only gotten to see the changes secondhand through friends and family who are teachers, and through my sibling who graduated a few years after me. But the difference is staggering.

It's a shame that humans find a way to cheat ourselves out of things that benefit us by over "optimizing" the wrong things.

I teach at two universities in Japan and occasionally give lectures on AI issues at others, and the consensus I get from the faculty and students I talk with is that there is no consensus about what to do about AI in higher education.

Education in many subjects has been based around students producing some kind of complex output: a written paper, a computer program, a business plan, a musical composition. This has been a good method because, when done well, students could learn and retain more from the process of creating such output than they would from, say, studying for and taking in-class tests. Also, the product often mirrored what the students would be doing in their future lives, so they were learning useful skills as well.

AI throws a huge spanner into that product-based pedagogy, because it allows students to short-cut the creation process and thus learn little or nothing. Also, it is no longer clear how valuable some of those product-creation skills (writing, programming, planning) will be in the years ahead.

And while the fundamental assumptions behind some widely used teaching methods are being overthrown, many educators, students, and administrators remain attached to the traditional ways. That’s not surprising, as AI is so new and advancing so rapidly that it’s very difficult to say with any confidence how education needs to change. But, in my opinion at least, it does need to change at a very fundamental level. That change won’t be easy.

It's still a new tech so I'm not surprised a lot of teachers have different takes on it. But when it comes to education, I feel like different policies are reasonable. In some cases it's more likely to shortcut learning, and in other cases it's more likely to encourage learning. It's not entirely one or the other.

Nice idea. What class and what work are you doing then?

How do you know you actually learned, instead of being fed slop by the AI that isn't true at all? If you didn't study, then I doubt you'll really know if the AI is lying to you or not. I have to wonder if your teacher will too, sounds like they have kind of checked-out from actually teaching.

You still do all the same things, and they are graded, but this doesn't affect your final grade. Instead, you need to pass a threshold to enter the exam, which is then graded.

The US isn't so amazing at this, it simply can be done better. Recognizing where you can improve and from whom you can learn is a great first step to ACTUAL improvement.

What a narrow set of skills to send into your economy.

I don’t know why you’d need to teach anyone to code with LMM’s though. If you are a CS major (or any reasonably intelligent person) and can’t figure it out on your own, well, workd needs bakers and carpenters still you know?

What I am learning is mathematics, which what all CS including AI and machine learning, really is. ”Mastering AI” for me means building your own models and AI applications and undestanding linear algebra, multivariable calculus and the propability theory behind it all.

When we had to add another core to keep up with Moore’s Law?

No, it shouldn’t. I’m not bearish on AI but it shouldn’t replace any part of a classroom where the objective is to learn and communicate in a new language (German). The typewriter argument is memorable and interesting - the article points out the lack of editing forces kids to slow down and think about their writing, as well as iterate through multiple drafts. It’s not a nostalgia thing, they’re not old enough to have ever used one before.

I could see an argument for adding on a new class for GenAI, agents, context engineering or what have you, but considering how behind current US curriculums already are and how quickly the AI field moves, I can only see this ending in wasted time and money: even an up to date class will be stale by the time it’s over. Kids will end up learning this anyway outside of the classroom, no use lecturing them on something they’ll already know.

You don’t give first graders a calculator because they will always have one in their pocket- they end up just inputting numbers in a magic box and not learning how to do this manually which will destroy their future mathematical education. It’s about the same with AI.

Former (second-generation) college professor, here. I find it almost impossible to be cynical enough about the US education industry.

This statement is more defensible after removing “only”. If it “only” hurt the cheaters, there would be no need to police cheating at all.

And they'll do it with all the 'unnecessarily high stakes' and 'risk of unconscious bias' and 'not truly representative' problems that written exams have; and a bunch of extra problems too.

Gyms aren't redundant because tractors exist.

We're doing these students a major disservice making them live in the old world. It's our fault for being inflexible, but their world is going to be wholly different and we should just embrace that.

I remember having to split bar tabs by hand at the end of the night without a calculator. The waiter delivers a hand-written bill, and we'd calculate how much each person owed, splitting drinks individually and food equally, recalculating tax and tip to derive each person's total. We'd pool the cash in the middle of the table (with people taking their own change as they go where possible), then counting all the cash and making sure it still included sufficient tip. All after many pints. I do NOT miss those days.

Oh