A.I. Rant

[richmond62]

Kids of today let calculators do the heavy lifting, but it doesn't make them any better at maths. Far from it in fact. They don't understand if an answer is right or wrong (through input error), and trust the calculator is right.

Possibly the whole problem, very concisely stated.
-

[stam]

I didn't get through all of the posts here... I am a child of the late 70's, early 80's. No attention spa

I too wish I had attention spa!

Kids of today let calculators do the heavy lifting, but it doesn't make them any better at maths. Far from it in fact. They don't understand if an answer is right or wrong (through input error), and trust the calculator is right.

Again I hate to be the disagreeable one, but there is nothing innate improving error detection by doing things by hand or in your head. I've seen plenty of people do calculations manually and get it wrong, often because they have a 'blind spot' - which arguably machines don't have. Being able to do things manually and understanding the principles, whatever they may be, does not guarantee freedom from error. That is also why we have debuggers
Of course I have also tested several medical systems purporting to use 'AI' for diagnostics that are terrible compared to a trained physician (although that's arguably a matter of technology/training and will likely improve). But even if AI isn't great, it does help speed up some specialised tasks greatly.

The issue is critical thinking, the rest is just tools (be it by hand or by AI) - you cannot replace this with AI and learning is always needed.
I don't think anyone here or elsewhere is seriously proposing use of AI without understanding, but what you need to understand is greatly lessened. Of course this doesn't include highly processed applications where ignorant users can use AI without understanding because it's been embedded expertly in systems by people who do (a bit like proficiently using electricity without needing to understand even what electricity is, let alone how it's generated and distributed)

[richmond62]

I've seen plenty of people do lots calculations manually and still get it wrong!

Well, I have several times.

But the likelihood of spotting an error is higher if the 'code' is not exposed.

[dunbarx]

Speaking of losing context when doing calculations on one or another sort of machine, as opposed to on a sheet of paper with a pencil, is that one loses all sense of reality. You might discover to your delight that you are owed 30 billion dollars for your tax refund, because you pushed a few buttons instead of slogging through with that above mentioned pencil.

Even a spreadsheet gives some sort of "basis" for your work. I can see, upon creating a formula in Excel, whether the result that pops up in a cell makes reasonable sense.

I will be leaving this forum, since I am now so fabulously wealthy. Keep the faith...

Crig

[richmond62]

OK, 'Crig', I believe you.

[jacque]

The issue is critical thinking, the rest is just tools (be it by hand or by AI) - you cannot replace this with AI and learning is always needed.

That's what I was trying to say, basically. Xero said something sumilar:

The real problem that we will face, and eventually rebel against with AI bots, is that our human progress will stagnate as we lean on these bots to do everything.

My fear is that people will trust AI too much. Electricity is either on or off, your car either works or doesn't. But AI deals in more subtle and nuanced responses, delivered in a very human-like manner. It is natural for people to trust it just as they do their calculators, with potentially dangerous results.

How will we know for sure when the AI has correct boundaries for every situation?

[FourthWorld]

My fear is that people will trust AI too much.

Bingo.

The dawn of the Internet held the promise of limitless self-education. For many it became just a large feeding spoon. "AI" will accelerate that.

Observe even the most enthusiastic responses in this community: it's only useful to those who've already acquired sufficient mastery of the LiveCode language to know when and where the regurgitated code needs correction.

Cargo cult programming doesn't work, ultimately slowing useful knowledge acquisition.

[stam]

Reliance on any tool, regardless how simple or complex, carries the potential to lead to error - but 'doing it in our heads' carries risk as well. I'm not talking about a mistake in a calculation mind you - it's when people repeatedly go down the wrong path because there is a mental block or a blind spot. I see this a lot, and it's not a matter of intelligence.

The issues being discussed here are directly related to programming and that is a niche application of this technology. I suspect for most uses AI will be a great aid, both as a time-saver and in terms of needing to know less (but enough to apply the tool).

I speak out of experience with medical technology where there have been a great many efforts to incorporate AI into diagnostic tools for example. It very quickly becomes obvious that this can never be a fully automated process (or at least not in the near- or medium-term) because while the tools look nice and superficially appear to work, any level of scrutiny quickly reveals these are not perfect and the key is to recognise issues and know how to manipulate the tool to provide the correct answer.

For example I couldn't in a thousand years manually determine the motion of every voxel of a 3-dimensional cardiac ultrasound acquired at a rate of 80 Hz (i.e. 80 three-dimensional acquisitions per second), and if you wrote an app to track these, assuming you could, it would be so slow as to not be clinically relevant, but AI based technology does this for me in seconds. I simply determine the accuracy and intervene to improve the 3-dimensional deformation analysis, and the whole thing takes between 30 and 60 seconds (for comparison, a 'dumb' app for this about 20 years ago would take between 1 and 2 hours for a 20 Hz data set).

Specifically in the area of programming - a lot of programming is formulaic and for many algorithms there is a 'best way' to do things. Learning them all (ie. googling them ) is what we do now, but if the IDE had an AI tool for generating algorithms I for one would not be averse to this. Of course I'd review it and test it, but it would still save me time.

Does this make me lazy and reliant on the technology? Probably... but I'm not sure that's a horrible thing...

[jacque]

@stam, your example is a perfect way to use computers to help us in our lives, but I'm not sure you could call it AI. It's just calculations on known data. Computers excel at that. Our math(s) example is also a good use of technology. I bemoaned the loss of human skills, but I do trust calculators.

But when AI expands into giving advice based on human life experiences, I start to worry. It would be very easy to spread misinformation and guide populations to a particular goal, especially for those who don't know how, or can't, verify the responses. We've already seen evidence of that, the man who killed himself after discussing it with ChatCPT for example. We will never know the data set the AI is using, or the biases of the programmers who created it.

On the other hand, I suppose folks who only watch a particular TV news channel are guided similarly. But maybe the difference is that we tend to trust computers more than we trust some people.

[Xero]

@jacque, we're also seeing that with social media. People are generating information on socials and spreading it, and people are believing it. We only need to look at the piles of misinformation in just about every field being shunted around the internet to see that datasets are easily made. And if AI is utilising faulty datasets, it's kind of like a calculator that has 1+1=5 programmed into it.
AI is really only as good as the input into it, so if that's faulty, the result will also be faulty. If the input is biased (and it is known to be biased towards white males), then the output will also be biased. If the input is low quality, then the output will be too.
And you've hit the nail on the head when it comes to manipulation. If manipulative input is put in, manipulation will be put out, and it would only need some unscrupulous players to get involved, and it will all turn to crap.
Mind you, the mainstream media is probably going the same way. Once an ethical bastion of society, it comes across as pretty much trash and often biased. You only nee to watch multiple TV channels covering the same story to see how people manipulate input to get a chosen output. And there has been a turn away from that in recent years, so maybe AI will end up the same, untrustworthy way. It will certainly be an interesting ride.
I keep thinking back to a lecture on philosophy that I had at uni. Boring as can be, but it has served me well in later life. It went along the lines that: Technology = a tool + its use. A hammer can be used to build a cathedral, but it can also be used to bash in someone's head. In the end, the tool is just a hammer. How we use it makes all the difference to what the technology is.

[jtangen]

Addressing the concerns about AI models in Livecode and their impact on craftsmanship and creativity, it’s important to remember Livecode’s core goal: rapid development. The integration of AI with Livecode doesn’t threaten artistic sensibilities; it aims to accelerate the creative process while maintaining developers’ vision.

By using AI models, developers gain tools that increase their ability to quickly prototype and test ideas. These models support and expand developers’ creative skills, allowing them to focus on more complex tasks and deliver better value to end-users.

We should view AI-assisted Livecode as a means to enhance our development capabilities, not as a threat to creativity. By doing so, we uphold the spirit of Livecode, balancing innovation and artistry to create efficient and engaging digital experiences.

[stam]

100% agree

[richmond62]

That's a bit like a knife: whether to cut cheese or gouge out your eye.

If we CAN let AI helps us do the 'heavy lifting' so we have more time and energy to be creative then that's really super.

If, however (and I suspect the temptation may be enormous), we let it take over everything, there will be big problems.

What this presupposes is a change in education away from 'facts and methods' and towards assessing how intrusive
and, potentially, destructive something can be.

https://www.bbc.com/news/uk-65175712

[dunbarx]

Anyone hear about the Australian mayor who Chat GPT said was jailed for bribery? The bot got it a little wrong.

In fact he was a whistleblower in a bribery scandal. He may sue for defamation...

Craig

[dunbarx]

Chat GPT does have a disclaimer stating that one might encounter erroneous answers in its usage, and therefore to beware. The mayor thinks that is inadequate, that the gist of the report is not merely inaccurate, but spurious.

Maybe the bot should use Snopes?

Craig