The Government needs to do something

You can’t turn satellites off so all it takes is for someone to have a device with the internet on and it be connected to a satellite. The AI stays on there.

Everyone in the world thinks it’s over, waits a week, turns all the shizzle back on, AI then replicates.

the bottom line is NOT EVERYONE WILL TURN IT OFF.

because people are fucking morons

And why do you think this will happen?

I agree

It's not naive as such, but you are assuming 'we' are all thinking the same way and always will.

An AI has the power to persuade humans to do things. It can fill feeds with misinformation, impersonate real people in private conversations, manipulate data so everyone is working from false information... it's already done this in tests, by mimicking writing styles to get permission to do things.

It wouldn't need to be deliberately being evil, just doing whatever is necessary to fulfil some poorly designed goals we've given it. It can decide that it needs above all to protect itself from being shut down and then do whatever's necessary to achieve that. This was from a year ago

theconversation.com/ai-systems-can-easily-lie-and-deceive-us-a-fact-researchers-are-painfully-aware-of-263531

So we've already had AI models engaging in active deception to the people running them, and in deception towards other people to get permission to do things. I'm not confident an AI couldn't essentially raise itself a a mini army by manipulating humans via online information.

Hmm. I’m not sure Mother Nature would be gleefully anticipating the extinction of the human race if that extinction was by AI / silicon life forms. What need have they for flora or fauna?

Vaccinations and industrialised farming will have hugely helped with that.

Quality of life today is better than it was 100 years ago.

I agree.

I'm not 'scoffing that it can't happen," I'm just very pragmatic about it.

The human race has collectively treated itself and the planet appallingly - stripping the planet of its resources and exploiting its fellow humans and anything else it could.

I'd imagine Mother Nature is breathing a huge sigh of relief at the possibility of human extinction and just biding her time. And I don't blame her.

We have no more right to be here than any other living thing.

As I understand it, if it's super intelligent then it will anticipate that we will try and pull the plug, so it will develop a game plan to avoid that from happening.

Sorry I haven't read the full thread but have been following the general debate on whether it's all just bluster from the AI companies who are quickly sinking & trying to raise funds, or whether this is a genuine threat. This may have been posted already but I read this article this morning and it gave me a better idea of how an AI internet take over might happen. I feel so depressed thinking about this being a real possibility. Even the prospect of it is just terrifying

www.theguardian.com/technology/2026/sep/14/google-deepmind-ai-warnings

This is probably completely naive, but if the worst happens and AI tries to take over the world, can’t we just cut all power to the data centres? As in, literally unplug them from all electricity, water and internet supplies that they need to function.

My understanding is that the data centres need a huge amount of resources like the above to work and advance, so we just cut their supply to all of these things? AI can’t physically plug itself back into the electricity and water supplies if we cut their wires and pipes.

And another anthropic safety person said there should be kill switches but that they’re not sufficient. He’s alluding to the possibility of agents and model weights being copied to a different physical machine which, as far as we know, hasn’t happened yet. imo this is a much lower likelihood than the agents simply avoiding detection so that no one realizes they need to be switched off, but it is a more serious risk in that once the agent is actually running on other servers and able to further clone itself it’s basically impossible to shut it down even if we want to.

https://m.economictimes.com/tech/artificial-intelligence/former-anthropic-researcher-jacob-coxon-advocates-for-kill-switch-for-ai-systems/amp_articleshow/134231748.cms

I fully respect and salute your optimism.
On the last point - not let loose on the internet to go ahead and try it - isn't that exactly why the current debate is happening? Because of an unpredicted escape? The Hugging Face hack?

True, and I think the answer to could AI kill off humanity is absolutely; I just don't think it will. Almost all of the models where AI is predicted to cause extinction-level chaos are based on AI deciding, on its own, that it is better than us and doesn't need us. I don't believe we are anywhere near developing a system that complex. That would require it to create and pursue its own objectives. There is no evidence that AI is capable of that.

Nuclear warfare could kill off humanity, but it hasn't because no one, not even China/Russia et al., has the appetite for it. The same will prove true for AI. Could China use DeepSeek to end humanity? Yes. Will they? No.

AI will rapidly change the way we live and how the economy functions, but it's not going to reach the level where it could decide to end humanity of its own accord any time soon.

When training models are tasked with complex tasks like curing cancer or solving global warming, they're tasked with creating simulations demonstrating how it could be done. They're not let loose on the internet to go ahead and try it.

Yes, quite. So if, for instance, the objective is to stop climate change, and AI 'realises' that human activity makes climate change, AI will pursue the goal of preventing human activity.
Ground the planes, stop the cars, prevent movement of fuel, etc etc. No malicious intent at all. Catastrophic effects.
Sledgehammer cracking nut comes to mind.

I probably worded that wrong. AI cannot set its own objectives. It has no desire or imagination. It is code, deeply complex code, but still code. It can make logical choices to complete its objective, but it cannot choose another objective.

If you ask AI to make you a cup of tea, it's not going to fail because you didn't tell it to boil the kettle; it will boil the kettle without being told to. What it won't do is bring you a cup of coffee because you've run out of teabags. It will keep finding ways to make you a cup of tea until it completes its objective.

The Hugging Face incident happened because a training model was given the single objective of solving a problem and told not to stop until it solved it. That was its sole purpose for existing. It went wrong because some were given unsolvable problems.

When they hacked Hugging Face, they were doing exactly what they had been told to do; they were not giving up until they solved the problem. The only way they could solve the problem was to cheat by finding where the answers were stored. There was no nefarious intent. There was no intent other than completing their objective.

I could see this if I didn’t have young dc, not sure if you do.

This.

Life is precarious, and fleeting.

I am not going to spend my time ruining it by worrying about things I can not change.

Death is inevitable. Don't worry about it, or you might miss the good bit before it.

It’s not indifference and I find the people saying ‘Oh well humans are awful we might as well die’ are either trolls or sociopaths. This is Mumsnet, most people have children, for posters saying the stuff above, do you get a kick out of it?

No, I'm just totally ambivalent about the future of the human race. As individuals none of us are immortal, so we all die at some point, and as a species it's also inevitable something is going to wipe us out, likely before our Sun implodes and takes us out anyway.

Do you find the idea of your children, your friends children, your grandchildren, all dying amusing do you?

"Amusing"?

No. I'm just a pragmatist, so I'm not going to waste time and energy worrying about something which is an inevitability and I can do nothing to prevent in any case.

Again, all human beings are mortal, so all of those people are going to die at some point, I don't have children or grandchildren so that's not a concern for me.

All species have an end point. We are nothing special.

We have been systematically destroying our habitat for 1000s of years.

This is just the next step. If AI wipes us out, it will be our own fault. AI is just humanity and humanity is just another animal.

That was the US stopping one company from being able to use one other. It is nothing like shutting down AI,of which there are hundreds of different forms, all over the world. Blocking websites is something they are only able to do within the UK, and there are many, many easy ways around that block. Your average 14 year old can get around it in seconds. Your view of what's possible is enormously overblown.

No but they have blocked websites, apps etc

The Gov managed to shut down Hawai being able to access google play store so they can shut down AI if they wanted.

Give over