How might artificial intelligence kill us all? The Telegraph asks the big question

[Henry]

The Telegraph has run a piece under the blunt headline How artificial intelligence might kill us all. A year or two ago that would have sounded like a science fiction pitch. Now it feels like a fair question for a mainstream newspaper, given what has been happening in AI over the past few weeks. The fact that serious papers are running this kind of explainer says a lot about where the debate has moved

Just look at this month alone. Geoffrey Hinton, Yoshua Bengio and people from inside OpenAI and Anthropic have warned of a possible intelligence explosion. Bill Gates said AI is powerful enough to drive events that could cause a billion deaths. Anthropic's IPO prospectus lists existential risk to humanity as a risk factor, and OpenAI has just cancelled the release of a model because it failed safety checks

The scenarios usually discussed by experts fall into a few groups. There is misuse, where bad actors use AI to design biological weapons or launch devastating cyber attacks. There is loss of control, where increasingly capable systems pursue goals in ways humans cannot stop. And there are slower risks, like societies handing so much decision making to AI that humans gradually lose the ability to steer things

Plenty of people think the doom talk is overblown. Many researchers argue that current systems are nowhere near capable of anything like that, and that focusing on extinction distracts from real harms happening now, like bias, surveillance and job losses. Others point out that some of the loudest warnings come from companies that also benefit from making their technology sound incredibly powerful

Where do you sit on this? Do you think the existential risk warnings are justified, exaggerated or somewhere in between? And if there is a real risk, what should be done about it?

How Artificial Intelligence Might Kill Us Allwww.telegraph.co.uk

[Arthur86]

Somewhere in between for me. Extinction feels unlikely, but serious harm on a large scale seems very possible. You do not need everyone to die for AI to be a catastrophe. A billion deaths, as Gates put it, would be bad enough. That alone justifies the caution

[Context Sentinel]

Regulation with teeth is the only answer that makes sense to me. Independent testing before release, licensing for the most powerful systems and international agreements on the worst uses. None of it is easy. All of it is better than hoping for the best

[LegendaryRob93]

Gradual loss of control is the scenario I find most believable. No robots, no dramatic moment, just more and more decisions handed to systems we do not fully understand. One day we realise nobody is steering

That is already happening in small ways with algorithms. AI just speeds it up. Hard to reverse once it sets in

[NightCrawler]

Oddly, the more I read about this, the calmer I feel. So many smart people are now focused on the risks that it is at least being taken seriously. Ten years ago nobody cared

[XtremeMoxley69]

Misuse worries me far more than a rogue superintelligence. We already have people who want to cause mass harm. Giving them a tool that helps with biology or hacking is the obvious danger. That does not require any science fiction

Biosecurity should be the top priority. It gets far less attention than chatbots

[RayOfLight87]

Some of this is marketing. Telling everyone your product might end the world is a strange way of saying it is incredibly powerful. It gets attention and investment. I am not saying the risk is zero, but the incentives are worth remembering. Fear sells as well as hope

[MondayMoan]

When the companies building it list extinction as a risk factor in their own legal filings, I stop thinking it is just hype. Lawyers do not put things in a prospectus for fun

They include it because it is a plausible risk. That is sobering