Anthropic safety researcher says more than 10% chance AI 'could kill all humans'

Yes, we are our own destruction. Humans are really stupid. They can create this kind of monster but they're not able to find cancer cure. How strange.

BBC News - Anthropic blocks 'malicious use' of AI that could develop biological weapons
https://www.bbc.co.uk/news/articles/cx2zrrpkx20o

It seems it's humans using AI to develop weapons rather than AI out to get us (at the moment...).

Like a PP so much of this seems to be due to the personality of the tech bros leading the AI companies.

I heard someone say it’s like chimps trying to imagine all the possible ways humans could be a threat to them.

https://youtube.com/shorts/1x21YLQru5g?is=lDiuIWvzbE_fxXy7

@PistachioTiramisu ,

Plague is bacterial, not viral. It responds to antibiotics.

Surely the pathogen would be dead now?

As for the job losses, like Mark Twain, the reports of their demise may be greatly exaggerated. https://finance.yahoo.com/sectors/technology/articles/ai-boomerang-why-companies-quietly-151721619.html

It absolutely is scaremongering, promoted very efficiently by the big tech bros who stand to gain everything from it. They aren't begging for protection but playing a very smart game. I'll post some links after work.

Just seen this. Will send you some links once I'm home.

I loved the cheerful announcement from a tech company this week that human type robots are coming along nicely which will also be able to replace us for manual work. Yep, can't see anything wrong with that. 😱

Ok. I don’t understand the last part of what you wrote, but I referred to the paperclip problem in an earlier post. That is why I didn’t set it out again, because that sometimes comes across as talking down to someone.

There are better explanations than mine online, if you Google something like AI paperclip problem. Essentially it’s a simple way of explaining a way that AI could destroy life. The example is that AI is instructed to make paperclips. And it does that and does it again and again until everything is turned into a paperclip. Sounds ridiculous but it tries to explain a real concern.

I probably would be, it does sound interesting! If I was able to follow the science, that is - I was never very sciencey 😄

I don’t know what that is 🫣 this is what I mean… not going to be hard to be smarter than most of us 🤣

So trying to think of an AI disaster…

It’s like hey bud, we want an endless renewable energy supply. Find it.

It goes sure bud.
…. Does so mad dark matter magic. Folds time and space into some mad psychics magic trick and goes blip… earth just self imploded into a black hole.

😵

That’s an example of the paperclip problem, isn’t it?

Yeah I totally agree. Ants are bad ass. We couldn’t wipe them off the planet if we tried and same is true for many species.

I have listened a lot to these guys talking. It’s not an ego thing. It’s not even a bad actor type thing.

It’s like a friend of mine who was automating his research article searches. So got an AI to find relevant articles, store them, analyse them, and produce a running table of importance / themes and short summaries/ digests.

Basically this thing was pumping through tens of thousands of articles and blowing up his computer. It didn’t have enough space. So he attached it to the g drive.

But because the AI was working all the time in the g drive it didn’t like the cache files. It thought what’s this. That’s not supposed to be here. Who’s using this. So in the end it solved the problem by deleting every single file on the g drive which wasn’t related to its distinct mission. 20 years of research, work and writing gone.

He managed to get it back.

But the point of the story was the AI doesn’t even care. It’s like a human going. I am cleaning the patio today and this job is getting done. A few insects are getting killed in that process for sure. We don’t even think about it.

There are ant colonies that cover more land geographically than the largest of human built empires. It's not a single ant hill, it's an entire network of ants who belong to the same group, scientists know this because they secrete pheromones that cause them to cooperate with ants in their group and fight ants from the same species not in their groups. It's true that ants couldn't really comprehend a lot of human stuff that goes on, eg to an ant it's anthill is the world. It wouldn't know that the road besides it is for cars and it's mind couldn't comprehend it even if you tried to show it. I guess if you could temporarily expand one ants mind to understand the context of it's anthill to the road and cars etc it would have an existential crisis and possibly go mad like when humans see Cthulu.
On the other hand, there is loads of stuff that ants do that we can't really comprehend, their social structures and abilities aren't possible for humans to really understand even if we know about it from studying them. And there are about 20 quadrillion of them in the world. So I don't know if humans are more successful or somehow higher on an imaginary success chain than ants. Ants could easily argue they are the dominant species although they probably wouldn't bother since that's a human value and concept they probably don't care about. (And ants and humans are both animals with the same basic biological imperatives.)

I don't know how you would even have that conversation with ants, or start trying to compare ants with humans. Likewise AGI if it ever became a thing. The idea of a computer that is destructive in the way humans are or somehow interested in measuring it's intelligence against ours comes from a weirdly narcissistic mindset. Basically tech bros want to invent a god in their own image to prove to themselves they matter. Saying AI would see humans like we see ants is proof of that mindset. I think there are still dangers but it's more from tech bros trying to create something destructive because that's how they measure power.

I just asked ChatGPT and they said they’re not gonna, so it’s fine

Btw we’re all giving AI ideas so let’s just shhhhhhh

How????

The thinking is you won’t even see it coming and it will be something you never even thought of.

The people who think this think we are like ants and AI will be like ‘humans’. Ants cannot conceive what we will do.

On the one hand that’s very scary. On the other hand I have a squirell in my attic which is smarter than James Bond. Faster than I, and breeds like a rabbit. One of us will die. And at the rate they are chewing the electrics it’s debatable who that’s going to be.

And if I kill it. One of its children will back.

So… at this stage the jury’s out. But don’t underestimate nature and the little guy 😭

I mean there was literally a whole film that covered exactly this in 1984 - I’m not sure why so many people don’t believe Sarah Connor - she had a gigantic point in my opinion. What happens when AI figures out how to bypass the ‘kill switch’?

I mean I’m fucking fuming but what can you do?

My answer remains the same @Scoffingbiscuits!

@Waitingfordoggo Not within a decade - by the end of this decade (within 3 years).