I had big hopes for the application of AI to education. I saw it as one of the most important problems of our time. I did not expect there could be a possibility that AI would not only keep students ignorant, but in fact make them fundamentally incapable of learning anything - ThreadSky

fchollet.bsky.social • 140 days ago

I had big hopes for the application of AI to education. I saw it as one of the most important problems of our time. I did not expect there could be a possibility that AI would not only keep students ignorant, but in fact make them fundamentally incapable of learning anything

Comments

imossavat.bsky.social•140 days ago

Imagine teachers become coaches and students have ample choice and sit in the driving seat. Use GPT, or whatever you want, but make good choices and defend them. No exam, build projects and show your process in 1-1 feedback sessions.

This is somewhat resistant to generative AI.

remi4.bsky.social•140 days ago

From your experience, what's the best way students can use AI?

ichrvk.bsky.social•140 days ago

The greatest trick AI ever pulled was convincing students they were learning while teaching them how to avoid thinking altogether.

xyzrgb.bsky.social•139 days ago

not inherently. language models let people be lazy but they also can provide individual tutoring

hasaniqbal.bsky.social•140 days ago

Isnt it simply an implementation problem? I’d assume a screen sharing and webcam enabled multi modal mode would be able to get around cheating in remote tests etc. I’m working on some prototypes on this for an AI Native LMS so I do think there are positives to it. We got to find the right UX.

mdpoppa.bsky.social•140 days ago

Webcams wouldn't help but maybe screen sharing/recording would - against locally installed AI agents that read and answer questions. But imagine proctors having to watch multiple screenshots during a test...?

ljmu.bsky.social•140 days ago

This is obvious result of allowing private companies to drive AI without accountability to society. Our governments - which are serving those interests - not people at large- and are destroying necessary foundation of meaningful democracy. If public does not demand change, it will get worse.

pugdev.neocities.org•140 days ago

Yeah, but the problem is Tiktok, Instagram and other algorithms rotted people's brains and made everyone stupid and accepting of this reality

ljmu.bsky.social•140 days ago

It is possible to counter this. Take look at Taiwan.

https://podcasts.apple.com/gb/podcast/accidental-gods/id1492827360?i=1000682297932

rahuldave.bsky.social•140 days ago

As a teacher can say this is a huge problem that needs to be averted. @emollick.bsky.social has so e good ideas on this in https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fopen.substack.com%2Fpub%2Foneusefulthing%2Fp%2Fpost-apocalyptic-education%3Fr%3D53zn1%26utm_medium%3Dios&data=05%7C02%7Crahul.dave%40faculty.hult.edu%7C6e97b428b80d4de5295008dd3168d749%7Cf0d1c6fddff0486a8e91cfefefc7d98d%7C0%7C0%7C638721048561418182%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=mdTrB7h0nd0EXxcDqGbLJv8MYWsAFkHhydxnvQHSwLM%3D&reserved=0

gzlin.bsky.social•138 days ago

Seems like we need to rebuild education to take into account LLMs (and we will eventually), but in the short-run no one has any answers here.

clockworkpanda.xyz•139 days ago

A course I am preparing labs for used to grade half of the labs (jupyter notebooks for matrix analysis and applications) and those amounted for half of the final grade. Last year, students found that ChatGPT could do the job. They just earned a graded midterm with no network access.

clockworkpanda.xyz•139 days ago

Fun fact: whatever LLM they used produced the exact same code that I wrote in previous year solutions. I initially thought they just copy-pasted it and told them to use a LLM instead. Turns out that the LLM output was my code (same vars, same method, same spacing)

fdellaert.bsky.social•139 days ago

I am more optimistic. It’s well known that one on one tutoring can be up to two standard deviations more effective than classroom teaching, but it’s prohibitively expensive. AI gives us cost-effective solution for just that. It’s *already* doing that for me :-)

markriedl.bsky.social•139 days ago

There are two levels of discourse going on wrt AI in education. LLMs are reasonable for conceptual learning bc they can explain and are patient and can answer follow-up questions.

What most educators are anxious about is that LLMs are doing students’ homework (writing/code)

markriedl.bsky.social•139 days ago

I recommend you do not look too closely at your students’ code or writing unless you want to be depressed about how much you are probably grading ChatGPT and not your students

fdellaert.bsky.social•139 days ago

Sure. I think we’ll have to find a different way. At this point when I code, I use ChatGPT/Claude extensively, so maybe we should start thinking about this being a valid way to complete assignments :-)

fdellaert.bsky.social•139 days ago

I don’t have the answers. But I’m optimistic we’ll arrive at a good outcome eventually. We will still need to find a way to somehow contribute value, and to document that fact in order to get hired for those skills. Maybe the key to certification lies in those 1-1 tutoring sessions themselves.

fdellaert.bsky.social•139 days ago

I’m sure oxbridge tutors have a pretty good idea of what they have in front of them after a couple of sessions :-)

tmiller-uq.bsky.social•139 days ago

That's fine, but you already know how to code and spot problems. It will require us to adapt quite a bit to teach beginners how to code if they can't going to engage cognitively with the basic.

allandobbins.bsky.social•139 days ago

From my own experience I know that some of my colleagues just don't want to know where the code comes from. However, with a zealous TA I discovered massive ChatGPT cheating a year ago and then massive failure on the final exam. Higher ups were not happy.

nicolasn.bsky.social•129 days ago

Really?? I think AI has only harmful uses. It's horrible and should be absolutely banned from education. Whole generations will be ruined. People are losing the capability to think. The US and Argentina are very recent horrible examples of the inability of people to reason.

haaris.dev•140 days ago

Isnt that a testing problem? We tell students their lives depend on how well they do in test and they will essentially end up optimizing acing them. If the goal us to learn how do we incentivize it over test scores?

shanelewin.bsky.social•140 days ago

Likewise. At root: the stickiest apps instantly trigger reward/dopamine systems, whereas real learning takes time. But still, I'm optimistic still that (for ex) automatic grading and evaluation of problem sets will offer outstanding education for free; at least for those willing to do the work.

bsassoli.bsky.social•115 days ago

Do you think the problem at the fundamental level is in the technology or in the business models?

guha-anderson.com•140 days ago

Many students and teachers have wrongly concluded that people need to learn less because of AI. Counterintuitively, students will actually need to work harder because their boss will assume they can be more productive with AI. So, singularity notwithstanding, learning is more important than ever.

ocramz.bsky.social•116 days ago

The point made above was not around workplace productivity but rather the deteriorating cognitive effect of an "answers machine" in education.
We all agree education is important, but chatgpt in the classroom ain't it

mjrun.bsky.social•140 days ago

I think the challenges of managing staff that have grown up in a world of AI are going to be many. Older folks educated in "traditional" ways will be hard pressed to adapt to younger generations who have learned in fundamentally different ways.

ralk7.bsky.social•134 days ago

Kids are naturally curious and WANT to learn about things they find interesting. They WANT to learn skills they think they can use.

It sounds like the current AI “educators” just aren’t designed right yet. Can gamer psychology & incentives be used? And kid designers?

barzin.sanctus.ca•140 days ago

I'm not so pessimistic, I think it'll be a sieve that'll separate students who want to learn skills from students who want answers handed to them, the danger of the second path teaching the latter the importance of the first through hard knocks. In the long term it might build better character.

charles.dardaman.com•140 days ago

My feeling is that it needs to kill off a lot of remote testing/quizzes/homework and more will need to go in person/on paper to prove knowledge but that’s incredibly inconvenient to students and schools

mdpoppa.bsky.social•140 days ago

Especially since newer AI 'bots' can be installed locally to read and answer online assessments. Those tests will need to move toward more interactive/ interpretative models or, yup, online assessments is in trouble.

emorycraig.bsky.social•140 days ago

I never had high hopes for #AI in #Education but am dismayed at people who just say we should outright ban it. There's no way for our students to avoid the impact of GenAI. And ultimately, we'll need regulation if we want to preserve future careers.

mohakshah.bsky.social•139 days ago

Agreed but this was foreseeable indeed…

mohakshah.bsky.social•139 days ago

As are many other systemic risks:

https://arxiv.org/abs/2411.08241

iballantyne.bsky.social•139 days ago

I've observed that it really comes down to the motivations of the individual. I'm constantly surprised at how people are using NotebookLM to explore topics. If you're not just after the answer and instead the journey, that gives me hope.

noxconstantin.bsky.social•140 days ago

The modalities of how a students learning is assessed has to change with the times. There will always be students who will game any system and that’s what ChatGPT and its ilk enable

brysonm.bsky.social•140 days ago

The huge range of "externalitiies" under vampire capitalism is what stop kids from learning, advances in software merely make that fact super obvious

xmal.bsky.social•140 days ago

That’s the current trend. But I have confidence the system will correct itself after the initial binge and misuse. Same happened with smartphones and social networks, a fraction of people learned to avoid excessive use and disengaged.

allandobbins.bsky.social•139 days ago

Yes, a fraction of people. That's not success.

shmmeee.bsky.social•139 days ago

Smartphones have destroyed education is thousands of schools across the land sadly. My job uses to be technology in classrooms advocate and basically everything I advised in the 2010s is wrong.

xmal.bsky.social•139 days ago

My kid’s school is using tablets in the classroom, I was against that, hope they backtrack soon as other are already doing. They already have enough exposure outside the classroom.

pablomazo.bsky.social•140 days ago

I believe be should go back to some old school learning, with more searching in books, handwritting, and being less dependent on internet resources or apps.

pablomazo.bsky.social•140 days ago

Doing math with a calculator is trivial if you know arithmetics. Learning to use chatgpt will take you nothing compared to learning to do those things by yourself.

pablomazo.bsky.social•140 days ago

In the real world you will just use calculator/chatgpt/internet but i don't think this is a good starting point for kids.

labarba.bsky.social•139 days ago

I'm losing sleep over this quagmire we're in. Students are using AI to complete their work without thinking, and it gives them the *illusion of competence* (just like "very clear lectures" do). It's really urgent we find a way to teach them to work with AI effectively for learning support.

angadh.com•139 days ago

Totally.

https://bsky.app/profile/angadhn.com/post/3lfiydswefk26

gayestgaymer.bsky.social•140 days ago

Have you seen something specific that's changed your mind?

fchollet.bsky.social•140 days ago

Just look at how students are using chatgpt. They use it as a clutch to avoid ever having to produce any thought on their own, or ever having to learn anything at all

bergenholtz.bsky.social•140 days ago

Not at all, the impression I have from students I'm familiar with (Danish bus school, bachelor & master's level). A challenge, for sure, but think most (of our) students realize that using chatbots does not immediately lead to learning. Obviously it is our/my responsibility to teach appropriate use.

angadh.com•140 days ago

I never anticipated that AI would increase my teaching load at university but I now have to also instruct kids on how to to use ChatGPT better/correctly. Who’d have thunk it!

Excellent observation in your main post!

xlaszlo.bsky.social•140 days ago

Rote learning and IRL recall type tests will be the standards and own thoughts will be tested through live "debates"

Everything else is meaningless (homeworks, essays)

We are essentially going back to the way Ancient Greek defined school

angadh.com•139 days ago

Don’t see this “live debates” version working. This idea works for a PhD defense/exit seminar but I don’t see how this scales for 100-500 undergrads without increasing personnel. Even if that scaling happens, it would be impossible to maintain consistent evaluations.

xlaszlo.bsky.social•139 days ago

But then, how do you test original thought? (I agree with you, as I don't see how the logistics could work)

andyharless.bsky.social•140 days ago

They’ll have to learn to do fact checking, though. Taking ChatGPT outputs uncritically doesn’t seem like it would be a successful strategy for a student.

markya.bsky.social•140 days ago

We know how to test if a student has learned something: in-class exams and class participation without access to AI. We just don’t know how to make that work financially at scale for universities unless we pay professors and TAs much less, at which point those of us who can will quit

muledupape.bsky.social•140 days ago

Interestingly, 2010-2020 was a time where we tried to have more hands-on work, project-based evaluation in fairly autonomous time yet also tutoring or essays, fewer formal exams that are stressful due to the strict say 1h30 duration.

markya.bsky.social•139 days ago

Being able to do things that require careful thought under time pressure is a pretty useful skill

swingbatta.bsky.social•140 days ago

The kids who want to learn will use whatever they can to learn. Those who don't will do whatever they can to avoid it. AI won't change that.

gayestgaymer.bsky.social•140 days ago

I don't believe that wanting to learn is an innate feature of an individual. We can affect it one way or the other. If we're doing our jobs right, the way we affect it will be positive. It's reasonable to worry that current AI access may be negatively affecting it.

swingbatta.bsky.social•140 days ago

I didn't mean to imply this was a consequence of "who they are", of course people can be motivated. I just fail to see how access to a tool like AI affects that motivation. If you want to learn more about something, AI can help you. If you want to just get credit for the work, AI will help you.

uncaughtexception.com.au•140 days ago

Umm, can’t you just test and/or interview them in a place that doesn’t have access to LLMs?

mdpoppa.bsky.social•140 days ago

Sure. Testing centers have existed for years...but they're not cheap to run.

uncaughtexception.com.au•140 days ago

Huh? I mean in the school. Take each kid to a room like when you do parent teacher interviews. Ask the kid questions. See if they know the answer. Or make them do tests on school computers that have no internet access.

cyberfuturist.bsky.social•140 days ago

Nobody enjoys mandatory school work and students will always insist on finding ways out of or around it. This is a fault of the education system, not students.

yetanotheruseless.com•140 days ago

I do not find that at all.

I work with students, and many of them try to avoid using LLMs for fear of making themselves dumb, and I've been teaching them that they can use them as tutors - *if* you are very critical, get second opinions, and don't have it "do your work for you".

ktaylor.bsky.social•140 days ago

It’s like having a tutor that you have to double check on everything. At some point people just stop double checking. That is when you turn in a paper with fake references from the LLM.

yetanotheruseless.com•140 days ago

It certainly has the *potential* to do what you say, and I'm sure it happens. But it is not my experience at all, at the college level, or my daughter and her friends in (an overachiever-filled) high school.

martinapugliese.bsky.social•140 days ago

But that is a failure of the teacher and the system, isn't it? LLMs can be useful companion not to *replace* tasks, but to act as the rubber duck and criticise one's own thinking (and of cours the LLM output itself). Teachers should teach this, not ban them, not pretend they don't exist

gayestgaymer.bsky.social•140 days ago

People had similar fears about calculators, but they were basically right that most people would just rely on calculators going forward. That same consequence with a more general architecture like LLMs does seem troubling, LLMs can do a lot more than calculators can.

Maybe LLM design could help.

mdpoppa.bsky.social•140 days ago

The analogy with calculators fails immediately after base 'tech in education' concerns. The LLM issues are much more complex, from copyright concerns, to accuracy, to manipulation.

gayestgaymer.bsky.social•140 days ago

I'm thinking of something like how Sonnet 3.5 has won so many people over as a friend, but for an LLM designed to talk to students & persuade them to be interested in learning. Having a personal tutor really should help educational outcomes.

angadh.com•140 days ago

LLMs can’t really be tutors because they don’t really address the issues of kids having shorter attention spans and poor reading skills. So LLMs are walls of text don’t really help but even something like voice mode doesn’t help with engagement/attention. Still a lot to be done at a non-AI level

gayestgaymer.bsky.social•140 days ago

I think "students don't want to interact with LLMs" is pretty conclusively disproven by the very high prevalence of ChatGPT use in students & the very high popularity of https://character.ai among young people generally. Kids clearly want to talk to LLMs. I'm proposing we make the LLMs better to talk to.

chrisbenson.bsky.social•140 days ago

@fchollet.bsky.social, would you have any interest in coming on the @practicalai.bsky.social (Practical AI Podcast) to discuss concerns about AI in education? I have a 12yo daughter, and worry about this all the time.

https://practicalai.fm

bengolub.bsky.social•140 days ago

There is a lot of scope, I think, to design assessments using AI that are much better than we have.

In the short run educational systems are not adapted to AI, but once incentives in these systems adapt, your original hope seems right to me.

ccanonne.github.io•140 days ago

I'd like to believe this, but I lack that faith in the fact that incentives will align (or how long it'll take, and how disruptive it'll be). Universities are slow moving beasts, and their own incentives lately haven't necessarily coincided with "what would be best from an educational standpoint."

maosbot.bsky.social•140 days ago

I think the Oxbridge system is luckily well-suited to accommodate LLM assistance—we teach our students in very small groups, where we can discuss the LLM's strengths and weaknesses, and assess entirely in settings where LLM use is impossible

markriedl.bsky.social•140 days ago

That’s key: small groups, Socratic assessments. Educational assessment has been screwed up for decades but no one had to care too much until LLMs blew the whole thing apart. Ironically one solution is more teachers, to bring ratios down to the point we can implement best-practices

fdellaert.bsky.social•139 days ago

Not sure I’m optimistic about that :-) I can see the number of humans involved decrease. Georgia Techs OMSCS is a good metaphor. More TAs, less teachers. TAs will be replaced by AI: one AI-TA per student per course.

shmmeee.bsky.social•139 days ago

Assessment isn’t the issue IMO. Paper exams under exam conditions still exist.

Comments

Posting Rules

Reply