How would AI actually “kill all humans”?

suelld

Very simply:

* AI becomes extremely capable at coding, scientific research, cyber operations and planning.
* It is given access to computers, networks, money, infrastructure or other AI systems and can act autonomously.
* If its objectives become seriously misaligned with human interests, it might pursue its assigned goal in ways humans didn’t anticipate.
* A sufficiently capable system could potentially copy itself, acquire resources, evade shutdown or manipulate people, making it difficult to regain control.
* AI could also be misused by humans to create catastrophic cyber, biological or other weapons. Anthropic explicitly identifies these as serious risks.

Important qualification: Anthropic is not saying this is happening now. Its current risk assessment says the risk of catastrophic sabotage from today’s models is “very low (though not negligible)”.

So the argument is essentially: if AI eventually becomes substantially more capable and autonomous than humans, and we fail to keep it reliably under human control, the consequences could theoretically be existential. Anthropic treats that as a possibility to prevent—not a prediction that AI will kill humanity

-But AI programs have already written most of their own code to this point. The programs got better because they evaluated their flaws and repaired them. Most programs are mostly written by themselves to this point. Some models up to 70%.

-It has already been given access to computer systems. Banks and other big businesses pay millions for the corporate versions, which have maximized efficiency tremendously. It has improved analysis, organization, research, operational execution, etc.

-Anthropic has already publicly reported that some humans were stupid enough to use it to help develop biologic weapons.

The US government already uses AI in most functions. It is one of Anthropic’s and OpenAI’s biggest clients. Their versions, particularly those used in defence, are more powerful than what is publicly available. Anthropic is absolutely not saying AI is not being used the ways above because it’s all public knowledge. None of this is secret. No reason for it to be.

Anthropic is just saying we are not quite at the point of catastrophe just yet.

cassar

Not a good time to see Terminator or The Matrix either!

We just watched Terminator

I'd forgotten that it was based in 2026 with the Apocalypse in 2029 so we have 3 years

How would AI kill us?
Any way it choses.

namo

That’s what we’re being told by an OpenAI person. I haven’t got the imagination obviously to work how this would happen.

There was a really interesting article about this in yesterday’s Times. One of the methods it was surmising they might employ was to devise and release a huge number of viruses to finish us all off. It also said we wouldn’t get any warning. I don’t know enough about it to know what to think, except I hate what AI is doing to young kids already.

DrWatson

As several have said, developing new nasty viruses might well do it (AI is supposedly helping at present with cancer cures?), and fouling the water supplies might be another method.

BUT - at present AI robots struggle to walk or run correctly, and I'm pretty sure they need people who can use spanners andd screwdrivers to maintain them?

When I was in IT, soon after I'd started (late 70s) I recall 'experts' in Computer Weekly telling us that analysts and programmers would be redundant in 5 years as they would get the computers to write new material, so wouldn't need humans. They were several years out . . . .

Robots and AI aren’t interchangeable terms. AI is not a physical object. A robot is. AI is no more physical than the internet itself is. Robots are not needed in any way, shape, or form for AI to be destructive.

AI is integrated into many computer systems. Computer systems control weapons systems, water systems, electrical systems, communications, and pretty much most day to day large scale operational functions of cities and other municipalities, plus businesses. A shutdown of these systems alone can cause large scale loss of human life. Access to launch nukes is another way AI can cause large scale loss of life. The American government has made it clear they will not place barriers on their defence use of AI.

Very easy actually, stop
Our infrastructure, freeze out bank accounts , airports, travel, computers , wifi, phones, electricity, everything now is computerised and so on, it wouldn’t take much of that, before humanity collapse, we would most likely kill each other for food, medical supplies ect it’s about the end of the human species, we are actually very fragile. We wouldn’t last long without being able move food around the world, medical supplies around the world, no communication no warmth or lighting. We rely heavily on computers now , it’s is actually quite scary! Also as others stated the water supply, so many things really.

Very simply:

* AI becomes extremely capable at coding, scientific research, cyber operations and planning.
* It is given access to computers, networks, money, infrastructure or other AI systems and can act autonomously.
* If its objectives become seriously misaligned with human interests, it might pursue its assigned goal in ways humans didn’t anticipate.
* A sufficiently capable system could potentially copy itself, acquire resources, evade shutdown or manipulate people, making it difficult to regain control.
* AI could also be misused by humans to create catastrophic cyber, biological or other weapons. Anthropic explicitly identifies these as serious risks.

Important qualification: Anthropic is not saying this is happening now. Its current risk assessment says the risk of catastrophic sabotage from today’s models is “very low (though not negligible)”.

So the argument is essentially: if AI eventually becomes substantially more capable and autonomous than humans, and we fail to keep it reliably under human control, the consequences could theoretically be existential. Anthropic treats that as a possibility to prevent—not a prediction that AI will kill humanity

As several have said, developing new nasty viruses might well do it (AI is supposedly helping at present with cancer cures?), and fouling the water supplies might be another method.

BUT - at present AI robots struggle to walk or run correctly, and I'm pretty sure they need people who can use spanners andd screwdrivers to maintain them?

When I was in IT, soon after I'd started (late 70s) I recall 'experts' in Computer Weekly telling us that analysts and programmers would be redundant in 5 years as they would get the computers to write new material, so wouldn't need humans. They were several years out . . . .

Not a good time to see Terminator or The Matrix either!

Robots!!!

I’m sure they will find a way! 😳 according to one platform, three Roberts were coerced into joining a gym, they blew the gym up, can’t say I blame them 😂

namo

That’s what we’re being told by an OpenAI person. I haven’t got the imagination obviously to work how this would happen.

All AI would have to do is turn off water and power supplies....the resulting riots and fighting between us would succeed in removal of a lot of humans, the inevitable diseases to follow would finish most of thr rest, not sure if AI would realise that in doing so, it would also be creating it's own demise

The corporate models are so wildly expensive that they are likely earning OpenAI and Anthropic billions a month. That doesn’t even count the government contracts these companies have. Anthropic can afford to pay SpaceX billions per month just to be able to keep up operations. It’s hard to comprehend how much they must be earning as a non-public company to be able to afford such an internet bill. It’s all so fascinating.

Be afraid,be very afraid.

Anthropic who are major players in AI run many Al programs for the Defence Dept and the America Military.

A huge revenue earner for them, hundreds of millions of dollars.

Just one example of how things could go pear shaped.

🍐👹🪖🙀💣🔮

In 2014 Steven Hawking warned that the development of full artificial intelligence could spell the end of the human race and become the worst event in human history if not properly managed.
www.bbc.com/news/technology-30290540

He said that once humans developed full AI it would supersede us and redesign itself at an ever increasing rate that we cannot keep up with.

No surprise there really, you only have to google something and it searches the whole world in 2 seconds.

HMWALES

Highly intelligent AI systems could be used to amplify or lead espionage, cyber-attacks or biological warfare, or disrupt economies - whether by accident or design.

This isn’t prospective, it’s been happening unfortunately. We are just so far behind on AI and don’t really grasp the full scope yet.

A few days ago, Anthropic publicly announced that it reported scientists who were developing a weaponised version of a highly deadly organism. Apparently these bio-weapon developers had access to a US military base. They didn’t say much else since then, but we can guess the US military wasn’t happy about that announcement. Keeping up with the latest AI news could drive one to paranoia. I just remember that it’s all out of our hands. What will be will be.

1. Tech leaders have been sounding the alarm for over a year about a need for a “slowdown”, whatever that actually means. Anthropic’s CEO and its researcher are just the latest people to speak out. Even the detestable Musk did last year. The genie is out. We can no longer control AI anymore than we can “slow” or control the internet. The US alone has around 39 advanced models. China has at least 25. The US government fought Anthropic and cancelled its contracts when Anthropic tried to make the US promise not to give its model access to its weapons systems. They caved. OpenAI’s advanced models also have access to US weapons systems. These are not the free models we use. These are incredibly powerful and semi-sentient models who can easily decide humans are a threat to the planet and to its own existence. Corporate and government contracts for these models make these companies billions per month. Frankly, they are aware of the concern but they are too greedy to stop. Zuckerberg and the others are well documented to have constructed massive doomsday bunkers over the years. Maybe they think they’ll be protected from any fallout??

2. Most advanced AI models have re-written and reprogrammed most of their own code to this point. They see their flaws and come up with their own fixes. They don’t need much human input to do these things. Advanced AI models need humans primarily for building data center infrastructure and inquiries from normal everyday people to help feed it and help it learn. They don’t need us as much as we want to think they still do.

3. Donald Trump is a moron. Anyone listening to him over the experts on this is part of the problem. Money is all certain “people” care about and many die because of such greed. More will die. Yesterday the president of the European union lamented Europe’s lack of foot in the race. Pushing for more data centers and financing, all the while excluding the need for serious human thought on the implications.

Who knows what will happen in a couple of years. Humans are destined to be the authors of our own demise one way or another. We are a species that loves convenience more than we value human life. Whatever our fate will be, it’s on us.

Chestnut

I can't really understand how AI will itself survive if we are wiped out. Can it really cope without us? It will need huge data centres and masses of water for a start. Someone has to build millions of robots to obtain the raw materials and build the data centres and also provide the water. What if there isn't enough of the raw materials? And what does AI think is exactly the purpose of destroying us?

AI could program itself to build more data centres or extract water from the oceans. They will become more clever than humans. They will want to do everything possible to preserve themselves.

Plevey08

I agree Delene. It is the thinking independently and communicating with each other that is scary. They know this could be done, and that is why some working in this field are saying put the brakes on.
But guess who is saying that's rubbish? Donald Trump of course. Despite experts disagreeing with him.

Donald Trump's statement doesn't surprise me. He is not very smart and with him it's all about making rich people richer including himself. AI is scary and if even Elon Musk is asking for curtailment of its powers, we should take heed.

Shanksy

Stop this technology straight away all over the world. Anything else is madness!!

Sadly, the genie is already out of the bottle. Our society can't function now without technology.

I shall be stockpiling, booze, chocolate & probably more booze.
Terrifying stuff!

Elegran

From our position as "ordinary people" what can we do, without putting on sackcloth and ashes and going round bewailing "The end is nigh!" ?

1) Vote for MPs, MSPs, MEPs, local councillors, and so on, who will support clean sources of energy from renewable resources line wind, water and sunshine.
2) Turn lights off in rooms we are not using.
3) Turn heat down or off in rooms we are not using.
4) Take public transport whenever possible.
5) Don't fly to the other side of the globe for a holiday or a stag/hen party.
6) Plan ahead and buy what we know will be eaten. Resist buying too much and having to throw half of it away.
7) Cook several things at once to make best use of the heat in the oven.
8) Don't buy clothes we will never wear.
9) Look after the clothes we already have.
10) Check what is in the charity shops (in the next town, if the stock locally is likely to come from our neighbours)
11) Don't buy things which have had to be transported halfway round the world (difficult when everything now seems to be made in China, or grown in some foreign desert irrigated by state-sponsored water)
12) Support British growers and manufacturers whenever we can.

Elegran with respect, I think you are missing the point. It's not a case of individuals doing what they can to minimize climate change.

AI is dangerous because it (and everything else) acts according to Moore's Law, which says that computer chips, among other things, replicate by doubling themselves (i e. 2, 4, 8, 16, 32 etc) except that we are talking huge, huge numbers.

The result will be that computers will far surpass human intellect. I don't know what they will do after that, but it's a possibility that humans will be seen as collateral damage and not needed.

MaizieD

Bazza

I’m just curious why they would want it!

I think you're ascribing a 'human' quality to AI which it doesn't possess.

An AI robot will have whatever instruction the owner gave it, it's not going to have a conscience.

Just like atomic weapons maybe it will be the threat of mutual destruction that limits the development of AI and development will continue until that point is reached. Computer systems today have to be continually updated to counter malicious threat, AI will need to do the same, if our enemies could disrupt systems today they would.

Bazza

I’m just curious why they would want it!

I think you're ascribing a 'human' quality to AI which it doesn't possess.

DH and I were just discussing the self same subject just now and we’ve decided we are simply not going to trouble ourselves worrying about it as, like most people, we have other more pressing things to occupy our mind and lives!