Comment by FloorEgg | Hacker News Reader

Comment by FloorEgg | Hacker News Reader

Wouldn't it matter a lot how AI is being used?

I would assume if children are allowed to use AI without rails as a shortcut it will undermine their learning, and it's used for feedback and as a patient tutor it would accelerate their learning?

It seems like the problem is that they don't have the science and tooling to use it constructively at scale, so the desperate solution is to ban it outright until a scalable constructive approach is understood?

The article doesn't explain any of this directly...

It's frustrating to me when bold statements are directed at "AI" holistically and vaguely, completely ignoring any nuance.

There is a massive gap between letting elementary students free reign use chatGPT 3.5 (hallucinations and all) to do whatever, vs using a very guard-railed pedagogically optimized app powered by a SOTA model to support students in a specific way that accelerates good outcomes.

Most respectful interpretation is that the leaders know this and have a plan to figure it out, but for some reason it's not making it's way into this article. Is the absence representative of the truth of the situation, or some editors choice to pile on to a holistic anti-ai narrative?

> it's used for feedback and as a patient tutor it would accelerate their learning?

We have mounting evidence AI hurts learning and cognition in many circumstances. I have not yet seen similar-quality evidence for it helping.

Given that balance, restricting AI in education in the general population (while studying how to best deploy it) seems prudent. Especially given the Norwegian approach, which gradually introduces AI as kids get older.

AI makes me worse at programming but helps me learn things much more quickly than reading textbooks. Both can be true.

> but helps me learn things much more quickly than reading textbooks

Have you tested this against an external metric of competence? The research seems to show that AI is great at making you feel you know something. But I think the studies looking at language learning found those using AI extensively tested below peers using traditional methods.

I don't care what the studies say. It's an incredibly good tutor.

AI helps me fill gaps in my knowledge quickly rather than hunting around for hours for exactly the right chapter which kind-of-but-not-quite explains the concept I am hunting for.

> It's an incredibly good tutor

It feels like a good tutor. If you aren’t externally benchmarking your comprehension, you really can’t say.

> helps me fill gaps in my knowledge quickly rather than hunting around for hours for exactly the right chapter

Have you considered that learning to phrase your questions is part of indexing and thus learning a subject?

I’m not saying AI can’t help with that search process. But we have no evidence it helps and lots of evidence it hurts, and everyone with anecdotes to the contrary seems to be going off vibes around how much they learned without any external reference.

You don't always need the same level of deep knowledge on everything you do. A lot of things in software development requires some basic level understanding of some obscure API you would never use again. LLM definitely speed up that part.

> A lot of things in software development requires some basic level understanding of some obscure API you would never use again

This is glorified look-up. AI is great at this! Learning through look-up doesn’t work.

Elsewhere in this thread is someone arguing they learned linear algebra through AI. I’d love to be surprised by their acing an exam. I’m thoroughly doubtful they could get through one. AI is trained to be a sycophant, not to teach. Maybe one day we will solve this. I have seen zero evidence we have that today.

Yes, but (I assume) that you are old enough to have already gone through elementary school and perhaps further and learned and internalized a model for learning and retaining information.

I think that is what is at risk.

Be careful. Speed of learning isn't necessarily the goal. Durability is another metric. I learn more quickly with LLMs, too, but it's certainly questionable if that learning is as durable or deep as learning through struggling with a book.

The students of lowly-rated profs had better 10-year outcomes than those with highly-rated profs according to a study that I think came out of the Naval Institute a decade or two ago. "No shine without friction."

We need more data. Certainly turning students loose with AI stunts them. There's probably some happy medium. But where kids need the most practice with fundamentals when they're young, a blanket ban for now seems sensible. And it also seems like a good plan to introduce it when they get older. I suspect we'll learn a lot from this Norwegian experiment.

Oddly enough, I never was able to learn things from textbooks, except in the context of a traditional classroom lecture course. I've also met maybe one or two people in my life who were able to learn the subjects of my college majors -- math and physics -- at any level from a textbook alone.

Do you delegate all opinions to only those which have studies?

> Do you delegate all opinions to only those which have studies?

For policy decisions around something like education standards, something for which we have an established status quo (which works in Norway)? Yes. I don’t think waiting for evidence to act is imprudent in that situation.

I'm actually not really criticizing the decision so much the article and communications around it. If student learning outcomes are crashing and they desperately want to turn them around I understand why they would take dramatic action.

Giving students uncontrolled access to generic LLMs probably would hurt outcomes. Research process is slow (IRB and all that) so they are dealing with data from years ago (models that confident hallucinated a lot more than current SOTA) so if thats what they are basing it on its reasonable.

My frustration isn't with the decision (hey all teachers - no more chatGPT in the classroom). My frustration is with the reporting / nuance of "until we can research this better and figure out how to harness AI to improve outcomes and not undermine them".

> the reporting / nuance of "until we can research this better and figure out how to harness AI to improve outcomes and not undermine them"

It’s balancing the irrationally exuberant narrative of the tech bros and AI pushers. You have to stop the bleeding before you can dress the wound to promote healing.

I don't think one balances a lack of nuance with more lack of nuance.

One avoids nuance for clicks or to propagate a narrative, sew division, distract, etc.

Again. As I said in both my comments, I'm not criticizing the ban, I'm criticizing the absence of any communication regarding a plan for researching potentially constructive uses. As a reader, I can't tell if the Norwegian leaders have no plan, or if they didn't communicate a plan, or if they did and Reuters chose not to include it in the article.

Not everything has to be a culture war. When we are talking about our children's future it would be cool to do so pragmatically.

Eh, if the politician thinks this is the clarity of language necessary to send the message, I think that’s fine. Studying this and that can come later. It’s not like anyone is banning that research.

And I guess that's a big part of my frustration. I don't know what the politician actually said. I don't see any link to the/an official statement in the article.

I'm just an old man shaking fist at clouds.

I'm sitting on a mountain of evidence (n=44,000) that used in a very specific way and context AI accelerates and improves lasting learning outcomes. Th3 data is new, but the science that explains it actually goes back decades, predating AI - it's based on pedagogy from texts such as How People Learn (NRC).

My data also shows that students using AI the wrong way perform way worse - the performance gap is widening between students who want to learn and struggle (and use AI to optimize struggle) and students who want instant gratification and use AI for shortcuts.

So I know that if they truly slammed the door on this potential then they threw the baby out with the bath water.

But I don't know the truth because Reuters doesn't report the truth, and that's what tips me from concerned to frustrated. But I guess by complaining about modern journalism standards in a thread about banning AI I'm breaking HN guidelines. Time for me to log off...

> sitting on a mountain of evidence (n=44,000) that used in a very specific way and context AI accelerates and improves lasting learning outcomes

Can you point to it?

I think soemthing ignored in the discussion is that there are "good" students who can make use of AI to further their learning versus "bad" students who use AI as a crutch that prevents them from learning well

Acknowledging this divide exists would come down to individual parenting and assessment of a student's use of the tools, rather than a blanket ban of AI tools which punish the "good" students

It's not just being ignored, I was down voted and argued with for trying to make this point.

Reply on news.ycombinator.com