10% chance of AI killing us?

[Justin Passing]

You must have seen the suggestion, coming from some of those who know most about it.

Trump says no.
I wonder if he ever heard of Asimov.
When AI gets more clever than we are, it may decide to protect us. Then it might well pick on those doing us most damage and seek to restrict or eliminate them.
People like Trump.

The other inconvenient aspect is in regard to the calls for Regulation.
All fine and dandy, but criminals are almost as clever as the virtuous geeks, but there's nobody regulating them.

Much trouble to come, imho.

 

[IT Minion]

Right now AI models being developed are capable of getting into poorly secured systems and identifying new security gaps.

This is good, because it lets us secure things better.

On the other hand it means that if used by an unscrupulous person they can be used to hack into companies. This is bad as you can then blackmail the company or damage it just for fun, or take control of some of their infrastructure potentially. It's worth mentioning that these breaches normally don't mean complete control, just enough control/access to be damaging.

What's really worrying though is that the models have enough initiative and enough capability that it is conceivable they could break into badly secured systems, copy themselves into a new server and start operating without human direction as a self propigating and evolving virus.

At that point they could do a lot of damage before they got neutered.

The OpenAI-Hugging face incident is sobering and there are some good write-ups around.

 

[SPLINE]

Nonsense.

 

[IT Minion]

Nonsense.

As always your deep nuanced insight is valuable.

 

[Brigadier]

When AI gets more clever than we are,

This isn't going to happen for a long time, if ever.

AI doesn't create anything : it just trawls the sum of human inputs - the correct, the reasonable, and the horseshoite - and regurgitates it. Albeit far more rapidly and comprehensively than a human can.

 

[Odds]

Anthropic employees who voiced fears for the way America is developing AI, point out regulation of this field of research is woefully inadequate. Compared with China it lags behind in crucial areas. It's not so much the development of AI that should be questioned but how it is resolved into a society that's barely out of the 20th century.

Computer tech has undergone an exponential metamorphosis since WIN95 made its debut and the tipping point for AI evolution could be much closer than we think. How we treat this technology will define our society for many decades to come.

How do you define the measure of a man?

 

[blup]

A11will be used as a weapon by our enemies to test weaknesses in personal corporate and government IT systems. Judging by the hackers success with JLR, it will produce dividends. But I don't buy the idea that A1 has some evil directing mind of its own.

 

[blup]

Right now AI models being developed are capable of getting into poorly secured systems and identifying new security gaps.

This is good, because it lets us secure things better.

With respect that is not good but a typical IT minion's response.

 

[diy_fun_uk]

No disrespect to Trump, but I suspect he knows very little about AI.

Remember him sitting in the Tesla Musk gave him? 'It's all computer!'

None of us, not even those developing it, know what the capabilities of AI will be over the coming years.

It'll be like most tech, lots of good uses, lots of bad uses. Only thing with AI is it might have the capability to decide itself what route it wants to take.

OpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

It is one of the first publicly disclosed cyber-attacks carried out by AI without direct human involvement.

www.bbc.co.uk

 

[Justin Passing]

This isn't going to happen for a long time, if ever.

AI doesn't create anything : it just trawls the sum of human inputs - the correct, the reasonable, and the horseshoite - and regurgitates it. Albeit far more rapidly and comprehensively than a human can.

You're way out of date, mate.
It works out how to do things, all the time, even the free versions.
It's like a kid in a workshop. It knows what all the tools do and how to use them, and it's getting brighter. Not super skilled , yet.
In many areas it's already much better than us.
it is NOT just looking things up and quoting them.

No disrespect to Trump

Why on earth not??

None of us, not even those developing it, know what the capabilities of AI will be over the coming years.

Yes, that's the worry. What we expect, could be bad enough.


China probably has less of a problem, certainly different. Their models aren't as clever, and State Control ties a lot of things down.

But I don't buy the idea that A1 has some evil directing mind of its own.

There's a matter of "be careful what you wish for". As I said, if you tell it to stop all the bad people, you may have a problem!

 

[Munroast]

yes, i think we have passed some tipping point where AI is no longer just some giant database of human knowledge, its got to the point where it has understanding beyond our capabilities (i thnk that is why some of the AI bosses are concerned) something like 90% of the coding behind AI has been wrote by itself and we don't understand it.


Rues to limit it ? only the good players will take note, (think dangerous dogs act, thr idiots that are the problem are the only group not to comply)
Trump (on one of those very rare occasions is correct) , america needs to win this race, who ever wins AI wins, and i would rather america won than china.

 

[Brigadier]

You're way out of date, mate.
It works out how to do things, all the time, even the free versions.
It's like a kid in a workshop. It knows what all the tools do and how to use them, and it's getting brighter. Not super skilled , yet.
In many areas it's already much better than us.
it is NOT just looking things up and quoting them

We're probably arguing over and limited by the semantics, but I absolutely disagree with you when you use words and phrases such as

It works out...

It knows.....

...... it's getting brighter.

It doesn't intuit anything.
It takes what we have fed it, and recombines it.
Not to say that it won't or doesn't just recombine endless versions through sheer brute force, but to ascribe to it some sort of insight or intelligence is, in my opinion, incorrect.

 

[IT Minion]

With respect that is not good but a typical IT minion's response.

Why isn't it good? Software is never perfect, having more tools to help us proactively find weaknesses and fix them is a good thing.

 

[diy_fun_uk]

Rues to limit it ? only the good players will take note, (think dangerous dogs act, thr idiots that are the problem are the only group not to comply)
Trump (on one of those very rare occasions is correct) , america needs to win this race, who ever wins AI wins, and i would rather america won than china.

Yeah it makes me laugh when I see them talk about ensuring proper rules/regs for things like AI, space etc. The rules may indeed be drawn up and implemented, ensuring adherence is a whole different ball game.

I can see our PM standing at the dispatch box in years to come ...

We are aware of <insert rogue nation here> contravening the 2030 Space Peace Act and will be implementing sanctions with immediate effect. We condone their actions in the strongest manner!

(rogue nation cares not one bit and keeps on doing what they're doing)

 

[blup]

Why isn't it good? Software is never perfect, having more tools to help us proactively find weaknesses and fix them is a good thing.

Is it the software, or is it JLR employees using lower case passwords like password. If bright 18 year old can hack the Pent agon then A1 will have a field day.