AI fails

[C R]

I still think that slide rules are clever.

And sufficiently accurate for most uses thanks to the power of the vernier.

 

[briantrumpet]

And sufficiently accurate for most uses thanks to the power of the vernier.

That was the thing that most amazed me when I learnt to use them at school – how many significant places were possible if read carefully.

 

[C R]

That was the thing that most amazed me when I learnt to use them at school – how many significant places were possible if read carefully.

People forget that most of the space race happened before desk/pocket calculators were available.

 

[briantrumpet]

People forget that most of the space race happened before desk/pocket calculators were available.

...and that Concorde was built using slide rules and a mixture of metric/imperial measurements.

As a counterpoint, I had lunch with a younger friend, and having been in coding for 10 years is now in the IT section of the Met Office, and was reporting on the amazing advances in AI coding he's seeing at breakneck speed... though he added that the Met Office aren't just 'trusting' it to be right, so have rigorous systems to check the output.

 

[PurplePenguin]

I still use logarithms often.

That wasn't the question.

 

[briantrumpet]

That wasn't the question.

Sure, some devices go out of date, but cognitive development doesn't. No one uses Latin in daily communication any more, but it's a helpful stepping stone if you want to be a linguist. And yes, sure, you can be a linguist without doing Latin, but you'll probably be missing some of the scaffolding in your understanding if you've skipped that step.

 

[Psamathe]

I still have a set. Also several slide rules incl. a compact circular one.

Edit: And my vernier guage is "old school" and I can still read it (though sometimes I need reading glasses as well as contact lenses and hold it with the light at the correct angle. Not gone to one of those "digital" things. But never really appreciated the reasons the scales work.

 

[PurplePenguin]

Sure, some devices go out of date, but cognitive development doesn't. No one uses Latin in daily communication any more, but it's a helpful stepping stone if you want to be a linguist. And yes, sure, you can be a linguist without doing Latin, but you'll probably be missing some of the scaffolding in your understanding if you've skipped that step.

When a calculator was invented, kids stopped using log books, but still learnt maths. The focus changed a bit, but maths was still very much learnt. I don't see how AI needs to change that.

By way of example, I have "vibe" coded a spelling game. It covers all the weekly spellings for an entire year for one of my kids. It is an easy game, saves me loads of time and means he learns them much more quickly and accurately. Isn't that a good thing?

Separately, when I think some reading comprehension is nonsense, I don't need to phone a friend, I can just argue with AI and its deep language knowledge.

I don't see how any of this makes people worse at maths or English.

 

[briantrumpet]

When a calculator was invented, kids stopped using log books, but still learnt maths. The focus changed a bit, but maths was still very much learnt. I don't see how AI needs to change that.

By way of example, I have "vibe" coded a spelling game. It covers all the weekly spellings for an entire year for one of my kids. It is an easy game, saves me loads of time and means he learns them much more quickly and accurately. Isn't that a good thing?

Separately, when I think some reading comprehension is nonsense, I don't need to phone a friend, I can just argue with AI and its deep language knowledge.

I don't see how any of this makes people worse at maths or English.

You're arguing something different. Sure, AI can (if used well) be useful for certain tasks. But we *know* that it cognitively impairs habitual users of it when they use it to sidestep the hard work of understanding stuff (conceptual understanding, plus understanding through reading), and the point that the people pushing it hardest don't care if children can't do maths or read properly (by which I mean – and I assume he means – comprehend properly from the written word) should send shivers down the spines of all educators.

I think it's a justified concern that the AI hawks would think it a win (for them) if human cognition was degraded to the point where we are forced to subscribe to AI businesses to do all the hard stuff, because no-one understands how AI got its answers, so can't realistically challenge it.

There's a gulf between what it can do well and what the AI businesses will try and hook us into. Opioid pain relief is very effective, but I think it's fair to say that the drug companies and medical profession didn't cover themselves in glory in the inappropriate promotion of their use. Profits trump ethics, and the same will be true of AI, if we allow it.

And that's the context of flashing red lights for an AI boss not worrying about maths or reading ability in humans. He wouldn't GAF.

 

[Psamathe]

You're arguing something different. Sure, AI can (if used well) be useful for certain tasks. But we *know* that it cognitively impairs habitual users of it when they use it to sidestep the hard work of understanding stuff (conceptual understanding, plus understanding through reading), and the point that the people pushing it hardest don't care if children can't do maths or read properly (by which I mean – and I assume he means – comprehend properly from the written word) should send shivers down the spines of all educators.
...

I feel the process of forming a view, coming to a non straightforward decision is a mental exercise. You consider the different aspects, compare importance of different aspects that maybe don't readily compare. Balance multiple factors (eg cost, time, reliability, impacts on others, etc) and doing that is an important learning process and later once outcome known you can look back at the different aspects and learn. And keep doing that and each time helps better decision making or understanding.

Or ask ChatGPT and get the answer and learn nothing, don't develop the skills to look and compare different aspects, etc.

 

[PurplePenguin]

You're arguing something different. Sure, AI can (if used well) be useful for certain tasks. But we *know* that it cognitively impairs habitual users of it when they use it to sidestep the hard work of understanding stuff (conceptual understanding, plus understanding through reading), and the point that the people pushing it hardest don't care if children can't do maths or read properly (by which I mean – and I assume he means – comprehend properly from the written word) should send shivers down the spines of all educators.

I think it's a justified concern that the AI hawks would think it a win (for them) if human cognition was degraded to the point where we are forced to subscribe to AI businesses to do all the hard stuff, because no-one understands how AI got its answers, so can't realistically challenge it.

There's a gulf between what it can do well and what the AI businesses will try and hook us into. Opioid pain relief is very effective, but I think it's fair to say that the drug companies and medical profession didn't cover themselves in glory in the inappropriate promotion of their use. Profits trump ethics, and the same will be true of AI, if we allow it.

And that's the context of flashing red lights for an AI boss not worrying about maths or reading ability in humans. He wouldn't GAF.

I just feel like your argument is much the same as how calculators were going to destroy maths learning, because no one would learn times tables. The education system may need to change e.g. more exams and less course work, but I think a way will be found.

 

[C R]

When a calculator was invented, kids stopped using log books, but still learnt maths. The focus changed a bit, but maths was still very much learnt. I don't see how AI needs to change that.

By way of example, I have "vibe" coded a spelling game. It covers all the weekly spellings for an entire year for one of my kids. It is an easy game, saves me loads of time and means he learns them much more quickly and accurately. Isn't that a good thing?

Separately, when I think some reading comprehension is nonsense, I don't need to phone a friend, I can just argue with AI and its deep language knowledge.

I don't see how any of this makes people worse at maths or English.

I was in high school in the mid 80s. Scientific calculators were commonplace by then, but we still learnt how to use logarithm and trigonometry tables. It was a good thing to learn, not the tables in themselves, but the practice of things like approximating for a value not on the table.

There's a significant difference between the Introduction of calculators, which meant you didn't have to use the tables, but you still had to reason the problem, and LLM agents that can spit out a solution without the user having a clue.

 

[briantrumpet]

I just feel like your argument is much the same as how calculators were going to destroy maths learning, because no one would learn times tables. The education system may need to change e.g. more exams and less course work, but I think a way will be found.

This isn't about exam results and the method of assessment. It's about an industry that doesn't care if humans can understand stuff, and they want to make a profit out of humans not understanding stuff.

I'll recount again the time we had a session in maths with adding machines like this:

Before we were allowed to use them, the most important instruction was to get an idea in your head about the sort of number that you expected, because it was quite easy to turn the handle on the wrong unit (1's, 10's etc).

That applies a billion times to AI, but how do you know "what kind of thing should come out" if you've no idea how it's worked out what's come out?

In other words, the maths lesson wasn't really about maths, but about how much trust you put in machines, and that you need to understand how to mitigate the risk of the machines getting it wrong.

tl;dr - the AI industry has a financial incentive for humans to lose cognitive skills. They might give us a few nice (and useful) toys along the way.

 

[Psamathe]

When a calculator was invented, kids stopped using log books, but still learnt maths.

Tesco might disagree. Few years back I used to shop there and it was a frequent thing where a bit of mental arithmetic and you'd spot buying eg 3x100g pots was cheaper than the "special offer" 1x250g pot - you'd get more product as well as pay less money.

They clearly relied on people not being able to do the trivial mental arithmetic and not carrying their calculators around.

 

[briantrumpet]

Not sure if this is an AI failure, but putting the same info three times into a character-limited micro-blogging space seems daft.