Helen Toner

And they're starting to learn they need to turn down the dial of how much should the AI try to create answers humans like, because it's very easy for them to end up saying things like, oh, you're worried that the government is listening to you through the walls.

You're right.

You should be worried.

Or you're concerned about your family.

You're right.

You should be worried.

Instead of doing what a therapist or a friend would say, which is, hey, hang on.

That sounds kind of crazy.

They're not trained to do that right now, or they haven't been so far.

I would say it's pretty experimental right now.

What I'm seeing is a lot of the bigger kind of name brands are being pretty cautious about whether that's a good idea.

But certainly brands that don't have so much to lose are trying out

You know, can I put a chatbot into something, you know, a fuzzy bear and say that it's educational?

Try and sell it as this will help your kid practice math or something like that.

But the AI system inside is often, it's very similar to these other chatbots.

It doesn't have great guardrails.

It's very hard for the makers of the toys to be confident that they're not going to go off the rails if the child kind of takes them in a direction that wasn't expected.

And I think, I mean, I think that...

Yes.

I mean, I think what is a big part of what is scary here is how fast the companies feel like they have to move.