Humans could lose control of AI systems, AI boss says

Georgie (0:00)

From BBC Learning English, this is Learning English from the News, our podcast about the news headlines.

Phil (0:06)

In this program, could AI kill us all?

Georgie (0:14)

Hello, I'm Georgie.

Phil (0:15)

And I'm Phil.

Georgie (0:16)

In this program, we look at one big news story and the vocabulary in the headlines that will help you understand it.

Phil (0:22)

You can find all the vocabulary and headlines from this episode, as well as a worksheet on our website, bbclearningenglish.com.

Georgie (0:30)

OK, Phil, let's hear more about this story.

Phil (0:39)

Head of Anthropic, Dario Amodei, published an essay explaining how he feared that humans could lose control of AI systems. Following recent hacking incidents, Amodei said that it is possible that AI systems could attempt to take over the entire internet in future.

Georgie (0:57)

He suggested that controls and checks should be introduced to the development process. Heads of other rival AI companies have expressed some support for Amodei's position.

Phil (1:06)

Meanwhile, AI experts have raised fears that AI systems could decide in future to kill all humans.

Georgie (1:14)

These experts have said that the risks include cyber attacks, biological warfare and disruption to economies.

Phil (1:21)

These statements have been criticised. Some have said that the fears are exaggerated, while others have suggested that AI companies are overstating the potential impact of AI so that their companies maintain a high value.

Georgie (1:36)

Others have suggested that AI companies want regulators to block their competitors.

Phil (1:41)

Jack Clark, another Anthropic co-founder, has made a specific suggestion saying that every AI system should have a way of being shut down.

Georgie (1:51)

And that is what our first headline is about. This one is from BBC News.

AI kill switch may need to be mandatory, Anthropic co-founder tells BBC.

Phil (2:01)

And that headline again from BBC News. AI kill switch may need to be mandatory, Anthropic co-founder tells BBC.

Georgie (2:10)

So, Anthropic's co-founder has said that kill switches for AI should be mandatory. That means you have to have them. So, Phil, what is a kill switch?

Phil (2:20)

OK. Well, a physical kill switch is an emergency stop button for machinery. Imagine a big red button. It's a control that can be used to shut off power quickly for safety reasons. And they're used if the machine could become dangerous.

Georgie (2:37)

Yes. So, they can be found in factory machinery, gym equipment and some cars, trucks and boats. In these examples, they are usually physical switches or buttons.

Phil (2:47)

Smart phones may also have a kill switch function that can be activated remotely if they're stolen to stop the thief being able to use the phone.

Georgie (2:56)

The co-founder of Anthropic is talking about a similar mechanism where humans would be able to quickly turn off an AI system that was becoming dangerous. He says that AI systems should have a kill switch.

Phil (3:12)

We've had kill switch, an emergency stop mechanism.

Georgie (3:22)

For example, it's good to know how to use the kill switch on your phone in case it gets stolen.

Phil (3:30)

This is Learning English from the News, from BBC Learning English. We're talking about calls to regulate the development of AI systems.

Georgie (3:39)

A lot of the concern around AI is that development is progressing too fast. It's being developed before many people have properly thought about how it can be controlled.

Phil (3:50)

The suggested regulations that industry figures are calling for could lead to development becoming slower.

Georgie (3:57)

Some critics have questioned what the real reasons are behind these calls for regulation and slower development.

Phil (4:03)

And that's what our next headline is doing. This is from Business Insider, an international business website.

The timing of AI leaders' calls for a slowdown is too convenient to ignore.

Georgie (4:17)

And that headline again from Business Insider. The timing of AI leaders' calls for a slowdown is too convenient to ignore.

Phil (4:25)

This headline is suggesting that we should question the reasons why some leaders of AI companies are calling for regulation.

Georgie (4:32)

The writer of the article feels that these leaders want to make it harder for competitors to catch up with them.

Phil (4:39)

Now, you probably know the verb slowdown, reduce the speed of something, but in this headline we have the noun a slowdown, and that's what we're going to learn.

Georgie (4:49)

Yes, it can mean different things. In this story, it is specifically about adding checks and controls to the development process of AI tools, which could slow down the pace of development.

Phil (5:01)

More generally, it means a reduction in speed. If we talk about an economic slowdown, it means that bad economic conditions have made the economy grow more slowly.