Well, this week was… something else. Last week I didn’t put together a proper news overview because I went skiing, and while I was up in the mountains I’d occasionally check the news on BlueSky – and it wasn’t looking very happy. Elon Musk claimed all sorts of data and systems for himself, all development aid was put on hold, and research funding was decimated. I can’t just take a week off for vacation when everything’s falling apart! But anyway, on to this week’s AI news.

Boing Boing:
Boing Boing reports this week on a 2023 study showing that CAPTCHAs don’t work at all. I wanted to include it here, but it turned into a full-blown story, so I wrote a longer piece on it here (in Dutch).

The Verge:
Scarlett Johansson has made a strong call to the U.S. government to crack down on deepfakes after a video of her and other celebrities went viral. It’s unlikely anything will change overnight, but it’s great to see her continuously weighing in with her perspective!

Nature:
I also came across a fascinating paper in Nature this week that shows real-world math skills (like doing calculations at a market) don’t automatically transfer to academic math in the classroom – and vice versa! Not directly related to AI, but it is in a way, because anyone who thinks that chatting with AI will actually teach young people valuable skills is missing the point about the “transfer” of learning.

There’s also a another research study by my former colleague Advait Sarkar (among others) that will soon appear at CHI. Their research, involving over 300 knowledge workers, shows that using AI doesn’t stimulate critical thinking – it actually hinders it! In fact, heavy reliance on AI might even lower your problem-solving skills.

Wired:
Wired remains a beacon of light in these days of relentless tech coup stories from the Trump era, but this week I particularly enjoyed a light, thought-provoking piece on AI and dating apps. It may not have a lot of heft, but it does raise some interesting questions: is it okay to use AI to chat with a potential partner? And how do you handle those first dates “without” it?

Economist:
Online scams are, as expected, spiraling out of control – and they might already be as big a revenue source as online drug sales. This long article does a great job explaining how even the scammers themselves are often victims from low-wage countries.

NRC:
You may have seen it on LinkedIn already, but I also had my say with an op-ed in NRC about the (im)possibility of politically neutral AI! (in Dutch)

Good News, by Popular Request:
Here’s some uplifting news! English science museums have scanned and digitized their entire collections – that’s 500,000 objects! In an extensive report, they detail how they tackled the project. Or go ahead and explore the collection yourself – check out the stunning difference engine by Babbage or the Jacquard loom, for instance.

And finally:
In the U.S., it’s been decided that images created with AI—even those generated with complex prompts—cannot be protected by copyright (although they do note that this might change in the future).

It was another wild week of AI craziness!

The Guardian:
I’ve written before about all the madness in the UK, and this week they’ve launched something new over there —a tool powered by AI that helps Cabinet Ministers understand how people might react to their policies. And naturally, I can’t help but ask: do you really need AI for that? Wouldn’t it be better to just go out into your constituency and talk to people? That way, you immediately make people feel seen—and that’s exactly what you’d want as a politician (I hope so…).

Follow the Money:
Always delivering top-notch investigative journalism, -which is exactly what we’re going to need more of in the coming years. For instance, there’s this piece about an algorithm called Preselect Recidive that the police use to predict whether young people will slip up again! It’s an extremely disturbing article—one can easily imagine which groups would be affected the most. It’s like a Dutch version of Minority Report.

The Financieel Dagblad:
This week I even made a brief appearance in the Dutch economic and financial daily news outlet FD (Het Financieele Dagblad), with a delightful headline of mine: Programming Is More Than Just Typing Code (in Dutch). Zuckerberg’s idea to fire mid-level programmers makes no sense—good software requires thought, a consistent plan, and coordination. Moreover, it seems like nothing more than a diversionary tactic to silence those who criticize the anti-DEI measures, effectively muting their so-called “masculine energy.”

Bloomberg:
A long read on the influence of YouTube on Trump’s popularity offered a truly in-depth data analysis of 2,000 videos—totaling roughly 1,300 hours. What’s interesting about the analysis is that many top podcasters do talk about politics, yet they explicitly claim to be apolitical. They cover topics like sports betting, the gym, and meme culture, casually weaving in content about Trump that fits perfectly with the “locker room tough guy” vibe. It wasn’t until just before the election that most shows started interviewing explicitly political guests. They also target a male audience—only 12% of the guests in the analyzed shows are women, and according to the piece, these are also the people who voted for Trump (50% of men under 30, it claims). But what struck me most from the article wasn’t the data, but a quote from Mike Majlak: “The easiest route these days to viewership is by creating enemies” . These are men who understand the algorithm and know that it isn’t quality, but anger, that gives a show its cachet.

Club de Madrid:
Eighteen former European leaders are calling on von der Leyen to take on Google and dismantle the AdTech sector. I didn’t really know the term “AdTech” until that excellent episode of Mystery Hype Theater 3000 last summer (the entire podcast is a must-listen!). What really stuck with me from that episode is that Google has very subtly shifted its goal—from “Hey, here’s a website where you might find what you’re looking for” to “Here’s the answer to your question,” which implies an entirely different kind of inquiry.

Nature:
Research involving almost 1,500 people published in Nature shows that working with AI changes your judgment and actually amplifies the biases already present in people. In fact, the effect of AI on people is greater than the effect of other people. The researchers write about “… a mechanism wherein AI systems amplify biases, which are further internalized by humans, triggering a snowball effect where small errors in judgment escalate into much larger ones”. This is a form of moral deskilling that Evgeny Morozov writes about in his book To Save Everything, Click Here. If you no longer have to think about what is correct (for instance, if you can no longer sneak past the metro turnstiles), eventually you’ll stop doing it altogether. It’s good that there’s comprehensive research confirming this—even if it’s a shame that such studies are needed when the outcome is so predictable.

NBC News:
An AI system designed to detect weapons on school premises (a dystopian idea in itself) didn’t work well in a recent school shooting because the shooter wasn’t properly in view of the cameras. This is a prime example of a technical fix for a problem that could just as easily be solved through legislation or standards (as is the case in the Netherlands).

A Few Minor Notes:
I’ll finish with a somewhat hopeful message: we’re hearing louder calls for more decentralized social media, for instance on 404 Media. It’s a bit unfortunate that BlueSky turns out to be a nicer alternative to Twitter than Mastodon—since Mastodon is arguably more democratic—but it’s something. Oh, and a heads-up! Microsoft might soon use all your texts in Word to train its AI. Don’t want that? Here’s how you can disable it.

One more thing, for comic relief… You can now create mind maps with ChatGPT. There’s plenty of excitement on social media about how it saves hours of studying and simplifies everything!!! But if that’s the case, then you really haven’t grasped the purpose of a mind map—the true value lies in processing complex information yourself.

This post was translated by Johanna Guacide.

I’m trying something new again! Instead of endlessly bookmarking articles because I think “I’ll come back to this later,” from now on I’ll start a blog post each week and update it throughout the week. That way, by the end of the week I’ll have a nice overview, and I’ll hopefully be able to easily find things without having to rely on big tech’s search (even a decent, small-tech tool like Pocket isn’t very searchable once you’ve accumulated 15 years’ worth of content). Let’s see how this goes!

TechCrunch:
OpenAI has quietly removed references to “politically neutral” AI from its policy documents. A striking twist—until now, OpenAI has consistently stressed its commitment to AI “alignment,” meaning AI that is beneficial for humanity (which, by definition, isn’t neutral!). Free speech is the prevailing sentiment in Silicon these days (just look at Mark Zuckerberg and Peter Thiel). In public, OpenAI is mainly campaigning to be allowed to collect more data without the usual hassles over copyright and the like.

New York Times:
Very different in tone but with a similar vibe is the much-criticized interview with tech investor Marc Andreessen. According to him, the leaders of big tech originally just wanted to be “good people”—and all the progressive moves they’ve made recently (like supporting same-sex marriage) were merely because they cared about being seen as virtuous by their peers; they didn’t truly believe in it (no worries, Trump & friends!). But when Biden (in his view) intervened too forcefully—possibly even hinting at AI regulation—it went too far. Suddenly, they had a change of heart and decided they should actually become Republicans (and apparently he even thought Hillary Clinton was Biden’s predecessor… a mistake the Times quickly corrected). It’s especially interesting to see how the “left” in the US was, in reality, very neoliberal and pro-business—and was widely seen that way. It makes you wonder how different the world might have been if we’d been truly left-wing (also here in the Netherlands). In a well-researched background piece, tech critic Brian Merchant explains how the Democrats essentially helped prop up the tech giants.

Futurism:
Tech startup School.AI has created a chatbot that lets you chat with Anne Frank. This isn’t just a fascinating technological development—it’s also a prime example of the deskilling I’ve always been wary of. Truly understanding what the Holocaust was is hard work, and it shouldn’t be simplified. Yet a chatbot can give the misleading impression that these profound subjects are just bite-sized chunks. For example, you could ask the chatbot whose fault Anne’s death was, and it would respond with some vague remark about how you really can’t pin the blame on anyone (see the screenshots here).

But truly grasping the Holocaust is challenging—and it’s a lifelong endeavor. This summer, I read about Hitler’s attire, which shifted my perspective, and just this week I watched the film A Real Pain (highly recommended), which again changed the way I think. Understanding such a vast and complex history takes time, effort, and a willingness to consider different angles. Was it also the fault of those who worked in the camps? Of those who betrayed Jews? Even of those who didn’t help? Some questions have no clear answers, and pretending otherwise is a problem in itself.

This post was translated by Johanna Guacide.

In this thorough piece, The American Prospect explains that while some people might hope that universities will save democracy, but that that might be tricky since they themselves are not at all democratic.

This piece reminded me of a lecture I gave right before the Christmas break for students about AI & education, in which I casually said that I am an anarchist (I will explain the context of that one in a later post!) One of the students got kind a angry about this and said: “what does that mean, do you want the world to have rules?”

I must admit I was caught by surprise by this question but it was a great one cause it made me think (maybe I assumed that students would also be or at least sympathize with anarchists?) so my anwser might have been a bit half-assed, but I have thought about it a bit more since and I think I can formulate it pretty well now.

VU, my university, is capable of having an opinion, for example on Ukraine, fossil fuel or gender equality. so how do those opinions come to be? Sometimes they come from the board of the university, lower people with power (deans, heads of big institutes), sometimes from even higher levels of power (all unis together), sometimes from special working groups, but as an employee I dod not choose any of those people in power. So that is not democratic at all, which I find problematic, because I do have to live with decisions of the VU either because people ask me about it, or because they affect my working life.

We could have elections for deans, rectores, department heads etcetera, which would be more democratic, but, for me, not enough since a lot of unexpected things might happen in a few years (like the war in Ukraine or Gaza) on which opinions needs to be formulated, and contrary to political parties that often have a history of voting and clear set norms, if I vote for a candidate un the uni context, how do I know they actually hold my opinion often enough?

So why not have all people of the university, professors, non-scientific staff, students deliberate together, without power structures? You can imagine all sorts of systems, a random selection of people each time (like jury duty), digital systems in which larger groups people discuss or vote etc. If I say I am an anarchist, I mean that I want universities (and ultimately: society) to make decisions in a non-hierachical way, independent of systems of power. And if that feels weird, scary and chaotic, I can only say: is our current system not weird, scary and chaotic?

We had a teenager over for New Year’s Eve, and one of his biggest hobby is to explain to me and my husband (“boomers” as he calls us even though we are millenials) what terminally online kids do these days; which words and memes and emoji are still in use. And this is how my final conversation of the year 2024 with came to be about the distracted boyfriend meme (which the teenager finds totally boomer).

I remembered then, that I had read on the internet a while ago (turned out to be 2018 haha, when the teenager was 10) that there is a Charlie Chaplin version of this meme (watch the whole film Pay Day that this is from on YouTube)

It is of course not said that the mem creators were ripping off Charlie Chaplin here, since people on Twitter came up with several older paintings and even a tapestry with similar images. But what struck me was what I said next, without even really thinking about it.

I said “I am happy I saw that meme before AI, because now I wouldn’t be sure of it was real”. Even if I could have found the whole movie the still comes from, that too would have been very easy to create with AI nowadays, and it would have costs me a lot of time to dive in. I am pretty sure that in 2018 I did not give a second thought to it, I just saw the image and could realistically assume it to be real.

I can’t bear to think of the extra work that we now all to carry out when sharing picture or video or audio, or the fact that people might refrain from sharing funny things for fear of fake stuff.

Ow the internet we have lost!!

Next week I will visit SPLASH to present a paper titles “A case for feminism in programming language design”, co-authored with Ari Schlesinger. A preprint of the paper can be found at the end of this post, if you want to check it out, or in the ACM digital library.

I fully understand that this paper title will create friction, people will surely be upset at the mention of feminism in the space of programming languages, and evenmore so because we suggest that PL design can use some feminism, but I hope this short post will help people understand how to read the paper, and what the back story is, so that they can understand better where we are coming from.

Rejecting feminism

Last year, I gave a talk about this idea for a group of young female CS students, and I asked who of them identified as feminist, and almost all of them raised their hands. When I was their age, I surely would not have identified as a feminist (so I guess we are making some progress), and to be honest I am not even sure why.

I think it was mostly that feminism was presented to me as a group of whiners. We had education, abortion, voting rights? What more could they want really? I bought into the ‘lean in feminism’ of the 90 and zeroes: if you simply work hard, you have the same opportunities as men, so just keep your head down and work. Of course it did not help that in my CS program there were, literally, only a handful of women (I was one of 2 out of about 120 students), so I already stood out like a sore thumb, I was surely not going to attract more attention by complaining about that.

So it people reading this and reject the whole idea of feminism and ‘complainism’, I get that!
But over time, my thinking changed.

By the way, the image of feminists being complainers is not something that appeared out of thin air, but was constructed by people opposing feminism, as you can see in imagery from those days, for complaining about the vote, something most people would now feel is… a reasonable thing to want.

Feminism is about examining systems

A thing I do, naturally, is to ask why things are the way they are. Maybe that is a result of my upbringing, where my dad’s parenting slogan was: “We will decide that for ourselves”, when people would have opinions of what we should or should not do. I also hate injustice and inequality.

Over time, I learned that a core tenet of feminism is to ask why things are the way they are, and by that to examine how we can make them more fair.

Now, that really resonates with me, because over time I have developed a few questions about why the PL community is the way it is. The question that I had did not have so much to do, per se, with the lack of women in PL, but about systems: what do we value and why?

Turns out, feminism has a lot to say about systems of power, because systems of power are exactly what feminism has been studying and changing (like: who gets to vote, have a bank account, have power over their own bodies etc.)

What is a programming language?

One of the PL community questions that has been bugging me for a long time is what is and what isn’t a programming language. As long time followers know, I used to work on spreadsheets—I did very cool things like build refactoring tools for spreadsheets—but the only thing people generally cared about was telling me over and over that spreadsheets are not programming languages. And it never became clear why that is. Argumentation that could easily be refuted (yes, spreadsheets are Turing complete, thank you very much) did not help in any case: Saying that spreadsheets are code is outside of the Overton window of acceptable PL opinions, I learned over and over again.

But why? In many other cases, the definition of programming languages is fluid, I remember when Python was not a programming language, but a scripting language, very different. And UML is, I think, universally not seen as a programming language, but it is featured in the book Masterminds of Programming which features “exclusive interviews with the creators of several historic and highly influential programming languages.”

So the way we construct what is a programming language is social, groups decide what is in and out, and if you are out, like spreadsheets, and thus like in early in your career, you cannot participate in the world of PL. If we want to study this phenomenon, we cannot do that in the realm of PL itself, you will need theories about how social constructs work, and that is where feminism can help!

How do we study programming languages?

The other question that has been rolling around in my brain for years, is how we study programming languages. Ever since I met Stefan Hanenberg at ECOOP 2010, when he was working on attempts with Andreas Stefik to get PL to do more user studies, I have wondered why studies with human subjects are so rare in PL. In software engineering (at least before the field was eaten by LLMs….) a broad variety of research methods were in scope, from formal proofs to corpus analysis of code and issue trackers, to observations, interviews and theory building. PL is arguably similar to software engineering in that they both aim to improve te state of software by creating things that people could potentially use to improve that state with. Why are the research methods then so different?

My exploration of feminism has helped me, more than anything else I have tried in the last decade, including explaining to many people that yes qualitative work is also hard and valuable. If you do not understand where they are coming from, what systems of power their beliefs are rooted in, this does not do a lot.

I hope this intro will help people that were confused about this work to place the paper better and to enjoy reading it more!

Part 1: Lane Assist

We recently bought a car, a fancy new car with fancy options like lane assist: a feature that keeps the car in its lane by moving the steering wheel for you. In theory, it is a great feature. In practice, it sometimes fails. Not often, but often enough it will decide it wants to really take an exit, or it will get confused when a road is wide and alternate left and right and left and right, after which I turn it off. For a while. And then I turn it on again.

Part 2: Goodnotes

On a totally different end of the spectrum of software, I am having similar issues. I use an app called Goodnotes, and Goodnotes is my life. I use it all day every day, for making slides, for my todolist (don’t ask me why I don’t use an app for that, I have tried them all and this works for me), and for reading papers. I don’t exaggerate if I say I have been using this tool 10 hours a week for the past decade.

Goodnotes too has implemented AI features, like Universal Object Selection:

Like lane assist, in theory this is pretty amazing. When you are editing and you want to move text or drawing, in the old days, you’d have to switch to lasso tool, select, and go back to pen mode. Now you can just stay in pen mode to move stuff. Similarly, there is the Scribble to Erase.

Very cool and useful, but they occasionally don’t work! If you write a bit hasty, text will be seen as scribble and erased when you write it, and sometimes the scratches don’t register, you will do a few of them and Goodnotes will crash. It is so annoying. I could just use the Lasso and Eraser tool, like I did for years. But I don’t.

Part 3: But I can’t stop!

A sane person would just stop using these stupid features (my husband stopped using Lane assist soon after we got the car, mostly because EU law dictates you have to hold the steering wheel and it warns you of that every 3 seconds with flashes and beeps), but I can’t because, well maybe this time they will work! Like a person in a bad relationship, who keeps going back to an abusive partner, I keep thinking “maybe this time it will be better”.

Soon, all software will implement all sorts of “smart” features, which are AI features, and if people are more like me than like my husband, UX will get a lot worse for most people.

A tool that I have been using for years, whose limitation that I had to switch from pen to eraser to selector never ever irritated me, because that is just how software works, now bugs me all the time, with a bug that is not even a bug, because of course an AI feature can’t work all the time, it is statistics after all. UX has now become a Skinner Box, and who does not like a sip of “OMG this just works” sugar water?

In a previous post I have tried to describe the reasons I see being used to use LLMs in CS education: 1) professionals use them, 2) LLMs can replace parts of teaching and 3) students will use them anyway so we as teachers have to somehow deal with that.

What I am missing a lot in the current discussions around LLMs are reasons to *not* use them! This too is visible from the invitation for the panel that “discussion will revolve around opportunities, challenges, and potential solutions”. In that description the only (somewhat) negative word is challenges. The things I am describing in this post aren’t challenges, things to be addressed, but fundamental issues that cannot be fixed.

So let’s dive into some, fundamental, issues that curb my enthusiasm for LLMs in many applications, including education.

Why not 1. Climate impact

According to recent research the BBC, generative AI will soon exert as much CO2 as the whole of the Netherlands. So simply said, we put all those solar panels on roofs, and all those windmills in the sea only to have it all nullified by software. And software to do what? To cure cancer? To end hunger? No software to generate cat videos and to save us from reading API documentation. On a planet that is burning and drowning, do we really find this to be the right reason to make it so much worse?

I think a very fun paper to write (maybe one day I will if I have the time) is to calculate the carbon footprint of the most recent ICSE, not in the amounts of CO2 we are burning to fly there (which are much, but can, in my eyes, be justified by science being a social process) but with all the LLM training and querying. Is it worth it?

Why not 2. Exploitative business practices

I no longer buy fast fashion, because I can’t explain to myself that I am willingly participating in the exploitation of people, in supporting their terrible working conditions (while others benefit of their labor) Instead I buy second hand, or I make my own clothes. Everyone of course is free to decide for themselves what they find ethical consumption, but using LLMs, whether you like it or not, is supporting the continuous exploitation of labor in the developing world.

In addition to exploiting underpaid and overworked content moderator, I feel LLMs are also exploiting me, personally. The Hedy repo contains maybe a hundred thousand lines of code, which I made public so that people could learn from it. Our EUPL license states, for example, that a licensee can “modify the Work, and make Derivative Works based upon the Work” which I am totally ok with, if it is done by a person, for example if someone wants to make Hedy Javascript version, they can absolutely copy my grammar and reuse the transpiler where applicable.

But open source licenses were never really designed to prevent AI usage (in retrospect, they should have!) and the EU license that we use states that “Those rights can be exercised on any media, supports and formats, whether now known or later invented, as far as the applicable law permits so.”

Does that media include gen AI? I am not a legal scholar, so I don’t really know (and I believe that in this case the jury is still out, quite literally, in a few law suits) Maybe it violates the Attribution right that states that the license information should be shared with the code, which clearly is not happening with LLMs.

But the law does not decide what I find morally correct, we all know that many things that were immoral were legal, and I feel gobbling up my source code, repacking it, separate from its intended context, and then selling it for profit, violates the informal contract I had in mind when sharing the code.¹

Why not 3. Bias in output

Several recent studies have shown that LLMs exhibit large amounts of bias: simply ask GPT who made Hedy and it will not be me, but a man. Of course a logical closing of a sentence about who made a programming language is a male name, and that is just scratching the surface. Brilliant and genius are associated with men, and written text that uses African American forms of English are judged to be more lazy and dirty that white coded English. Do we want the limited progress that we have made in diversifying CS to be nullified by algorithms that will present students with 10 white men if they are about who contributed most to programming?

Why not 4. Leaning into LLMs will lead to deskilling of teachers, and diminish the value of the profession of the teaching profession

The last few decades have seen immense growth of universities; the university I went to more than doubled in size in the last 20 years (5000 students when I went there, 12.000 now). In the Netherlands, this can be attributed to two factors: 1) more international students as more BSc and MSc programs switch to English as language of instruction, and 2) more people that are eligible for higher education since more people follow “academic high school” (VWO).

Even though more staff were hired, the growth has made professors more overworked, not only because of the number of students but also because of a lower level, international students will not command English as well as Dutch people do Dutch in many cases, and more students eligible for uni will mean, like it or not, lower levels of prior knowledge. Plus of course a highly competitive academic field (esp. outside of the natural sciences) means that demands on scientific work come on top of teaching duties.

This situation creates very fertile soil for (gen) AI: if I have to grade 80 essays in a day, or if I don’t have time to update my powerpoint slides with new research, using AI suddenly seems like a reasonable or even necessary. But grading or preparing isn’t a purely productive activity, I would argue that it cannot be optimised or made more efficient, because the goal is not only to grade the essays, the goals is also to learn from what students are submitting to improve teaching, and the goal of making slides is not to make the slides, but to read a few more papers about my field and update the slides with those I find will have value for the students.

Leaning into the idea that LLMs can do the deep thinking work required will inevitably lead to less skilled teachers that are no longer learning form their students’ mistakes and from new research. Also, it will hamper activism of professors against work pressure, which traditionally has been relatively successful. In a pre-GPT era, having to grade 80 essays in a day might have led to people going on a strike (students and professors) but now that it is “possible” to use an AI, the problem is not so visible in the direct sense, only in a slow (but sure) erosion of the profession.

Soon, the Netherlands will have a right wing government, and if the polls are any indication, so will the EU, and probably the US again after November too, and those governments hate science and education and want to budget cut the hell out of us all. If we, the scientists, are already saying AI can replace us, even if we are careful about what it can and cannot do, it will be used as a reason to reduce funding even more, and we can all easily predict, without an AI, where that will lead. This holds especially true for computer scientists, who will be asked more than other about their opinions (while probably being impacted less)

Addendum

Why so few objections? This is part of a longer set of posts that are upcoming, but I am reflecting on the field A LOT lately. I am a bit of an academic nomad, going from PL to SE to CSed to PL and most recently I am doing some work in the history and philosophy of science applied to programming, mainly because I am so curious about why our field is as it is. Why am I often the only person talking about climate impact and bias? While there are, of course, a gazilion reason, a paper I read recently showed that being a social activist is a strong detractor for studying computer science², and so it being artistic. So (compared to the general public) already very few people that care about social justice are entering our field, and then or source our culture does a great job at making care less about others.

I know I sound like a broken record for my regulars but a field that has a Von Neumann medal, named after a guy instrumental in killing hundreds of thousands of civilians does project some values on the inhabitants of that field (although many, like me for a long time, might be utterly unaware of his involvement, which is a bit of an excuse, but also another sign that we just don’t care).

—

1. It is also somewhat disorienting to see a paradigm shift happening in real time. I vividly remember the fury with which professors, when I was young, hated Microsoft, because they were making MONEY off of SOFTWARE. Even if Linux did not work well and their community was toxic as hell, there is one thing that is worse and that is running a profit.
   
   To see a whole field do a 180 and suddenly be exited about CoPilot, which is not only software for profit, but it profit from open source software, is… something ︎
2. Sax et al., Anatomy of an Enduring Gender Gap – Journal of Higher Education 2016. ︎

This Friday I will be in a panel at the SEN symposium (which I was participating in 2 years ago too, talking about autograders)

The organizers told me that the “discussion will revolve around opportunities, challenges, and potential solutions” and I was allowed to discuss one statement. Since they only gave me 7 minutes (and no slides :() I thought it’d be fun to elaborate a bit more here.

My chosen statement is:

Why

What has been bugging me in the discourse around LLMs in CSed a lot, is that the core sentiment seems to be: let’s use them, because we can (in a great contemporary Dutch saying that can’t really be translated apty: “niet omdat het moet, maar omdat het kan”). We don’t discuss specific problems, just solutions, we see that in the description of the panel too 1¹.

As the kids would explain it!

But what do we even think LLMs can offer? Let’s look at the reasons people give in favour or the use of LLMs.

Why 1. Professionals will use LLMs, so students will need to (learn how to) use them too

I actually don’t hate this argument, I think it is a pretty good point: we need to prepare our students for professional careers. However… LLMs are quite new, and we don’t really know how professionals use them yet (although of course academics are now performing one study after another to better understand their use in practice), and we do not know whether in a few years, when our current students graduate, the situation will be the same.

Here is what we do know:
Programmers don’t use LateX, they use Word or maybe Markdown.
Programmers use GitHub.
Programmers refactor code.
Programmers use debugging tools.
Programmers perform code reviews, programmers use CI/CD tools, programmers deploy code on the Cloud, etc, etc.

So if we are shaping our curriculum by the future usage of professionals (which again, I do not object to at all) why don’t we teach all the above tools, and why don’t we stop immediately with teaching and promoting LateX, a system proven to be worse than Word and not used in industry. If we do the one but not many of the others, what does that tell us about our real motivations? Could it maybe be the case that we think LLMs are very very cool (like LateX) and let that cloud our judgements? Seeing how little actual tools and practices from industry are being taught, I think this is a reasonable hypothesis.

Why 2. LLMs can replace teachers, so we must explore how to use them

The next underlying reason that people rush to use LLMs in education, is that they, fully or partly, believe that LLMs can teach students, rather than professors. Let me firstly throw in this amazing comic by Doug Savage.

I like teaching! I like grading, and making powerpoint slides, not only because they are simply part of my job, but also because grading gives me unprecedented insight into my own teaching, and preparing slides helps me outline my thinking. I don’t want to automate those inherent parts of teaching, and neither should you. If you want to automate teaching, why are you a teacher??

However, me saying that “grading is teaching” and “preparing is teaching” creates interesting discussion. What is teaching? Saying that AI can automate some parts of teaching leads to deep philosophical questions about education which I feel most CS professors are not equipped to properly discuss, because of the lack of philosophical and educational theoretical grounding.

Because in order to discuss what LLMs might mean for “education” we need to define education, and that is harder than you think.

To connect this to point 1, what even is the goal of CS education? Is it to train programmers? Is it do train future scientists? To train programming “though leaders”? ²And should education (any education, or CS in particular) be about teaching skills, or “ways of thinking”, or about giving people the vocabulary and mental tools (like math) to deal with all sorts of issues.

It has been possible, for a very long time, to learn most content of a CS undergrad program by yourself, when I was a kid in the 90s I learned a lot from books from the library, then with YouTube, MOOCs and now (people are saying) with LLMS. Yet, registration numbers for CS programs are soaring! So we must offer (at least in the views of 18 year olds and their parents) a value. What is that value? I would say the value is context, camaraderie with fellow students, connections with teachers, and learning what you did not know you did not know. None of these are automatable with a machine, so what do we even think the LLMs are doing?

And more important thing is not only what the LLMs are doing, but also for whom. In the school of education, where I teach pre-service CS teachers (lerarenopleiding, in Dutch), we tell our students that there is no such thing as a good intervention. Any decision you make while teaching is good for some students, but worse for others. More explanation is good for kids with lower prior knowledge. More group work is good for students in a dominant group.

So who are we designing CSed programs for? For kids that already know they like programming and what to become programmers, or for people that don’t know what programming means? We have to constantly make trade-off.s For example, about equity. If in a given class of 50 students, 10 people are confused and 10 people are bored, who do we care about most? Who do we address and help first? Because we don’t talk about these type of high level design goals, many people design their teaching for people “like them” (likely to be excited about all sorts of technology, including programming and LLMs)

Answering questions like these (which is hard and messy and imprecise) needs to come before deploying any kind of tool. We are already seeing results (very much in line with what learning theory would predict) that students with more prior knowledge are helped most by LLM use, so who are we helping and who are we disengaging (even more)? If you answer with “for all students” you lack an understanding of teaching theory.

Why 3. Students will use LLMs to do their homework!

As I said two years ago at the SEN symposium, I am excited to change introductory programming to be less about programming, and to get rid of autograders. When LLMs came along, I had some level of optimism that this would be the end of programming exercises involving a lot of syntax, but the opposite has been true. We are now leaning into the fatalistic notion that LLMs in CS education are inevitable, and refraining from further reflection upon our teaching.

Let’s dive in a little bit more about what exercises for programming courses usually look like. I think more or less like this:

We give students a prompt, and they have to make a plan how to solve the exercise, choose the right concepts and combine them together in a running and working program. These individual steps, I believe but I base this belief in cognitive science, represent different types of activities. Steps 1 and 2 mainly use the Working memory, fitting a solution in your head and weighting different options. Step 3 mainly uses your long term memory, you will have to remember the right syntax to correctly implement your idea.³

Despite the fact that these are really different things, we grade only the final product, which students struggle a lot to to arrive at, sometimes because their plan was flawed, something because they don’t command the syntax well enough.

Programming education Is NOt aBouT SYntAx!!!!
_{(but we only check if you can “do the syntax”)}

The reasons I think are threefold: 1) This is how we all learned, so we think it works, 2) because most CS professors lack the educational theoretical vocabulary to distinguish the steps, and 3) because we can check code easily, so we must (“niet omdat het moet maar omdat het kan”). And not only do we only grade syntax, we also only explain syntax. We never explain to students how they should make a plan, evaluate a plan, choose concepts etc.

In my opinion the deepest issue that we currently have in programming education, is that we do not split these steps into separate exercises and assignments. It would be trivial to do, give students 20 prompts and just have them write down a high-level plan, so they practice this skill, and then transform the plans into lists of concepts, and then to syntax. Only after practicing those skills in isolation, they should be combined. Everything we know about learning supports this way of teaching (in math, we practice many concepts and skills in isolation before combine them; in language learning we do the alphabet, vocabulary and grammar separately) and yet we do not do this.

Because doing this creates the question of how would we grade the natural language plans and lists of concepts? My answer would be that multiple choice is actually a fine strategy, but maybe there even LLMs would make sense, because we have a problem in which natural language needs to be processed! (There are different reasons to not use LLMs though, but at least this is a LLM like problem).

Yes instead of adapting our teaching methods and finally stopping the tyranny of autograders… we keep doing the things we do, but students have access to LLMs and now they are learning even less.

As I also wrote about last week, when I was a young assistant professor, and I was complaining about cheating (pre-GPT) a wise old professor once told me: “When students cheat, *you* are being unreasonable.” I of course thought that was a silly thing to say, the students were just being lazy!! But, in retrospect, this colleague made a great point, we are asking unreasonable amounts of learning in an introductory programming course, so of course students are going to use help, their peers, the internet and now LLMs. If we would make our education more reasonable, there would be less “cheating”.

Do you think students would immediately go to LLms if we’d simple ask them to explain in words how to find the longest alphabetical substring, after they had seen a few similar examples in the lecture? I don’t think so.

—

1. Not to pick on the organizers! This pattern is common everywhere. ︎
2. I have talked before (in Dutch) about the opaque goals of CS in higher education: https://www.agconnect.nl/carriere/arbeidsmarkt/waar-is-een-universitaire-informaticaopleiding-voor ︎
3. In reality of course there is no linearity in this process, people go back and forth between steps. In this example a student might only realize at the implementation stage that a for loop does not give you easy access to the next letter, needed for this assignment and then switch to a while loop. ︎

I see this argument “why stop students if we can’t check it anyway” so much, so let’s dive in a bit!

You can’t stop students from doing X

Firstly, we have been saying things like this for decades, if not longer. “Don’t collaborate on homework”, “Do this exercise without a calculator”, or when I was a student “Don’t use Wikipedia”.

Why do we see so much more fatalism in algorithm use then in collaboration or other tools? Why do we give up so quickly and assume students will “cheat anyway”? One reason I think is the narrative that AI companies are pushing: there is no escaping the AI wave, everyone will use these tools in the future, all the time. That creates a situation in which teachers assume that all students will use AI in their daily life also, so, why prohibit them now.As they say “resistance is futile”?

Is it though? Haven’t we all been quite successful at prohibiting students in elementary schools to use calculators, even though professionals use them all the time? Most educators agree kids should still learn the tables of multiplication even though they can easily be automated. Kids complain about that 9as they do) but we all keep this line: you need to be able to do small calculations in your head; the whole tower of math skills leans on that!

Why do students cheat (with or without AI)?

A wise old professor once told me, when I was complaining about cheating in the pre-GPT era: “When students cheat, *you* are being unreasonable.” I of course thought that was a silly thing to say, the students were just being lazy!!

But in retrospect, this colleague made a great point: Students come to university to learn, they might be lazy a bit, but if many of them are, is your assignment clear enough? Is it doable? Are there many other deadlines? I think if we critically examine CS ed from a learning perspective, it is totally reasonable for them to cheat, and it we’d make it more reasonable, they would not cheat (as much).

Professors set the norm, even if students don’t follow them

Much as with “don’t collaborate on this”, we can set norms even if we know damn well not all students will listen. This is because setting rules also set norms. Contrary to popular belief, laws often shape, not follow what people think! For example, when gay marriage was introduced in the Netherlands, a minority of people were in favor, and the law helped increase support. After all, what is allowed, must be good, and what is forbidden is not.

Students doing an exercise with AI, knowing it is not allowed, will feel like their are breaking a norm, and will thus differently about their work, and that matters.