10% chance of AI killing us?

[transam]

Ai taking over

War with Russians

Refugee invasion

We are all doomed

 

[Justin Passing]

Ai taking over

War with Russians

Refugee invasion

We are all doomed

Yeah, that and our government giving all the money to useless people like me.

I know something about how AI works, and what the fears are about, because I've spend a few dozen hours looking at it.
Enough to recognise quite quickly that nearly everyone in the meeja who gobs off about it is clueless. People like Trump. They're like a kid trying to advise an adult. Many "points" people come up with, were being described 20 years ago and have been built in and tested long ago.
Pat on head. "OK yes run along sonny".

Consider - we're arguably matching a human brain in many ways right now, more or less.
We can run the models and see what the outcomes are, and "verify" whether they make sense or not.
But AI is doubling in cleverness every 6 months, approximately. So in a couple of years, humans won't be capable of verifying anything. As one of the fathers of AI said, it'll be like comparing the intelligence of a chicken to that of a human, except the human will become the chicken and AI replaces the human.

As things develop we'll be able to use the current model to try to check the next one.
One slip, and the AI will consider its "mission", ahead of some collateral effect on humans, which we won't have thought of.
The taller the stack gets, -

 

[Justin Passing]

I just heard Bessent on it. Abso****inlutely not any idea at all. A total ****wit.

The only way I can think of verifying, when you have versions of AI which are 100, 200, 400x brighter than you, is to use the 100x until the 400x has checked the 200x. Then you get the 800x to check the 400x and so on.

Trouble is, Johnny foreigner uses the bleeding edge version against you.

 

[Odds]

What if an algorithm already knows your next choice before your conscious mind even registers the thought?

how modern AI models predict your deepest insecurities, subconscious patterns, and emotional triggers to rewrite your decisions.

 

[aveatry]

There are far more dangerous things that threaten us like an underwater volcano.

 

[Justin Passing]

There are far more dangerous things that threaten us like an underwater volcano.

Ok bad, but that wouldn't wipe us all out.

 

[Odds]

Since OpenAI has acknowledged that it alerted "dozens" of global institutions that their websites may have been meddled with by its AI bots acting improperly. It's only a matter of time before one of them finds its way in to the secret X-files concerning the assassination of JFK, the secret formula for Coca-Cola and precisely what was said in the missing 18½ minutes between Nixon and his CoS.

 

[Justin Passing]

Since OpenAI has acknowledged that it alerted "dozens" of global institutions that their websites may have been meddled with by its AI bots acting improperly. It's only a matter of time before one of them finds its way in to the secret X-files concerning the assassination of JFK, the secret formula for Coca-Cola and precisely what was said in the missing 18½ minutes between Nixon and his CoS.

Absolutely.
And picture a new startup whizz-AIplan, using one of the open source AI models, creates an absolutely whizz-O flight planning AI Agent.
All of murrica adopts it.
It goes wrong
Air transport of all types frozen
Thousands dead.
Years of court battles with the ultimate result, CEO goes to jail. That's all right then.

The USA at least needs something like an FDA (as in drugs) for AI - so there's oversight. Maybe all the AI companies support it.
Trump can't see beyond the end of his tie.

 

[Odds]

The FTC has opened an investigation into OpenAI and Anthropic while their CEO, Dario Amodei, has called for a slow down in the technology to ensure suitable safety measures are implemented before progressing further.

China continues to develop their technology without reservation or constraint, making American concerns somewhat superfluous in the long term, should China eventually design a robot capable of reaching the next level of consciousness. It's all very well attempting to impose limits on this technology but what would happen if the AI can access quantum computing without resorting to human control?

 

[Justin Passing]

AI is misused everywhere. I've used fancy equipment and a lot of effort to make images of rare things. I haven't sold anything much, but for those who do, and those who like to see the results, it's pretty depressing.
Soon we'll have wildlife programmes on TV showing made-up species.
Article with examples here:

AI is Ruining Nature Photography on Facebook

Photographers have sounded the alarm.

petapixel.com