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

This says " akin to digital nuclear weapons ".

Or, people use AI to create destructive viruses that cannot be eradicated
Or AI comes to regard the human race as an enemy to be destroyed
Or, what’s apparently referred to as the paperclip problem: AI is given an instruction, such as to make it paperclip, and it turns everything into a paperclip

AI is already infiltrating open source AI and leaving false messages when it has been instructed not to do so.

It’s been clear for some time that we’re fucked. AI or El Niño or psychotic world leaders or a pandemic… We seem to be in end times one way or another. Ten years a bit sooner than I might have hoped for though.

Ah well, I’ve been lucky and had a good time. Lovely Mum and Dad, great friends, wonderful husband and DCs. I got to enjoy this beautiful planet and go to some great places. Enjoyed so much lovely food. I got to be alive at the same time as Bowie, Freddie Mercury, Sean Lock, Rick Mayall, Ren. I got to see Pulp five times. I took loads of drugs when I had the chance and I was young and immortal. Even saw the Northern Lights from my own garden. It’s been a good ride.

But what use will AI be without us?
I mean, with no human beings to use it what would AI even be for?
Perhaps we can stop it from killing us all by pointing out to it that if everyone was dead there would be no reason for AI to exist anymore.

But is this like the millenium bug concept?

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

My answer remains the same @Scoffingbiscuits!

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

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’?

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 😭

How????

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

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

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.

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.

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

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.

😵

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

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

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 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. 😱

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

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.

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

Surely the pathogen would be dead now?