Kill switch

in case you weren't scared before that, extinction.

Out front now, Congressman Ted Lieu, a member of House Democratic leadership who serves on the Artificial Intelligence Caucus and has introduced the bipartisan AI Kill Switch Act.

So Congressman Lieu, I appreciate your time.

There's a lot I wanna ask you, but I wanna start with the, AI kill switch act, okay? My understanding, reading from the act, is it would require developers of the most powerful AI systems, this is what you write, to maintain the technical capability to throttle, suspend, or shut them down.

Now, I'm, you know, thinking about a world where we're hearing about hugging face and swarms of agents in different civilizations that don't even know their agents and, you know, makes me think how in the world could there be a kill switch? How exactly would it work?

Thank you, Erin, for your question.

Two, Senator Bernie Sanders and Representative Greg Kasa have announced a US bill to ban superintelligence, which Control AI consulted on.

Three, Lord Clement Jones introduced in the UK legislature an amendment drafted by Control AI that would establish an emergency AI kill switch for the government.

Heading.

The first bill to ban superintelligent AI introduced in any legislature.

Sanders and Kasa announce the first US bill to ban superintelligent AI.

Last week, Senator Bernie Sanders and Representative Greg Kasa announced the first US bill to ban superintelligence.

This announcement comes after Representatives Nathaniel Moran and Ted Lieu introduced the AI Kill Switch Act in July, which we were happy to endorse.

As reported by The Washington Post, Control AI consulted on the Sanders-Kasa framework.

Again, no pedals and no wheels

Apparently, the House of Lords has been calling that they want an AI kill switch in the UK as well.

Many of the characteristics that drive a person to become a leader and drive a person to become successful, um, are those same traits that make you more likely to burn out.

Consumer group, Which?, was able to book themselves in to a fake listing for 10 Downing Street on booking.com. Do you think we rent out Downing Street for a couple of days?

... um, we talked last week about the fact that, uh, your system might be turned off without, without you, a buy or leave to you.

Mm-hmm.

This kill switch that the US senators were talking about last week, apparently lords, the House of Lords has been calling that they want an AI kill switch in the UK as well.

Now I can't help-

If we're talking about the need for a kill switch already, you know, we're only, what, four years in to generative AI, agenda AI, like the hype.

And Jess, it feels like longer.

It does.

This week, we're trying something a bit different, or in the season, I should say.

Every other week, as always, we will bring you an interview with an IT leader But in the weeks between those, Penny and I are going to be talking about the news in tech and how it affects the UK tech market.

So, Penny, this week, we've seen the AI kill switch.

That's kind of the big thing in news right now.

Yeah, so on Tuesday, The Cybersecurity and Resilience Bill was debated.

Why would you want another operating system if OpenAI's capabilities can do this? This becomes the operating system.

The launch came wrapped in safety language, staged access, restrictions, a kill switch.

What is underneath all of that?

And I think what's more interesting underneath all of the security theater is architectural decisions.

It proliferates there.

It will just sit there.

So you pull the kill switch in your data center, and then it's still there somewhere else.

It's poisoning the data or it's kind of doing all these things again.

And I love how the media spins it.

Oh, the federal law about Kill Switch means it could be watching you.

No, they admittedly are.

OnStar went in 30 years ago, been listening to you ever since.

You have a dynamically balancing robot.

Someone hits a E-stop, what happens? Like, it's just gonna fall over, right? So what do we do? And, you know, there's that ISO standard, uh, being developed, so there's fantastic people working on how to make the robots safe.

Ju-just so, yeah, just to, to, for, for those who don't know, to, uh, just to g- like drill down a little bit, so dynamically stable means that when you cut power to it, it's just gonna collapse.

Yeah.

What we are always interested in as the Robotics Ethics and Society team is talking directly to users and, like, a different kind of expert.

So not to the robotics experts, but to workers who have experience working in factories, in, um, warehouses, uh, experience working with different kinds of robots and other machinery but have, like, a really good handle on, like, what would work best for their space.

And so we can present them with different options for what the robot should do when you hit the E-stop, and they have, like, very strong opinions on what would work and what wouldn't.

So that's the kind of data that we're collecting.

Administrators have to be able to set and audit confidence thresholds that dictate response levels.

There has to be a master kill switch that stops pending autonomous actions at the higher tiers within seconds, and there has to be an undo function that reverses commands the agents have already triggered.

Stop on the undo function.

I hoped you would.

A kill switch means you expect a need to stop it.

An undo function means you expect it to have already done the wrong thing.

That's the distinction, and it's a far harder engineering problem.

And the line it all rests on is short.