The People Who Built AI Are Now Looking for the Off Switch

One of the co-founders of Anthropic, the company behind the AI system Claude, has suggested that powerful artificial intelligence may eventually need a compulsory “kill switch.” Not a polite little log out button. Not “Would you like to save your changes?” An actual emergency mechanism that can shut the whole thing down if AI becomes dangerous or starts behaving in ways nobody can control.

That is reassuring.

It is rather like the inventor of the chainsaw announcing, after selling several million of them, that perhaps everyone should also be issued with a very large sticking plaster.

The people developing AI have spent years telling us how magnificent it will be. It will cure diseases, solve climate change, write computer code, drive our cars, run businesses and possibly fold a fitted sheet. 

I'm a child of the 60's. Didn't these idiots every see 'The Terminator"? And it's believable. Oh not the going back from the future to try to fix things; but that AI will some day turn on us...

Now some of those people that wanted us to use it, are warning that it could become uncontrollable, improve itself beyond our understanding and pose a serious threat to humanity.

Suddenly, the sales presentation has taken a turn.

I use AI regularly. It helps me write, organise ideas, create pictures and answer questions I would otherwise spend three hours Googling before ending up watching a video about a woman renovating a caravan in Finland. I like AI. It saves me a fortune! Whereas once I believed the crappy add for sticking plasters that are going to fix my bent figures or my toe hump; now I ask AI (who 9 out of 10 times says it's crap).

I also like my toaster, and I would become suspicious if the toaster started rewriting its own software and locking me out of the kitchen.

The proposed switch would apparently need to be independently verified, which makes sense. We probably shouldn’t leave it entirely to the companies racing one another to build the most powerful AI on Earth. That would be like asking the greyhounds to supervise the rabbit.

Of course, creating a kill switch is the easy part. Deciding who gets to press it is another matter. Governments? Technology companies? The United Nations? Elon Musk after a difficult afternoon? And what happens if the AI has already read the instruction manual and quietly disconnected the button?

Perhaps we are worrying unnecessarily. AI may remain an extremely useful tool that helps humanity rather than destroys it. At present, its main crimes seem to be inventing references, giving people six fingers and confidently providing recipes containing ingredients that don’t exist.

Still, I’m in favour of the switch.

Make it enormous. Paint it bright red. Put it behind glass and write BREAK IN CASE THE COMPUTER DECIDES HUMANS ARE INEFFICIENT across the front.

And perhaps don’t connect it to Wi-Fi.