The Unrelenting Weirdness of Interacting With Robots

[Narrator] Finally, the robots we've been promised

for so long have arrived.

More and more, intelligent machines are entering our lives.

Be they companion robots or self-driving cars.

The question is,

are we actually ready to get along with them?

Welcome to the strange new world of human robot interaction.

Now, I'm not saying the machines are going to get surly and

hellbent on our destruction.

I'm saying that interacting with increasingly sophisticated

robots is going to lead to some fascinating conundrums.

This is UC Berkeley roboticist, Anca Dragan.

She's tackling the looming problems of robot interaction.

When you put robots into people's homes,

of course you have the huge challenges that people's

homes are unstructured unlike factory floors and that,

the functional side of things,

they're really delicate issues around there.

[Narrator] A companion robot like Curry here,

has both navigate the chaos of the home and the people

moving around it.

That means not only rolling about delicately,

but setting the right expectations about its abilities.

For instance, Curry doesn't speak human, which is smart.

(beeping)

Because studies have shown,

if humans can converse well with a robot,

he'll assume that it's also physically capable

of handling tasks that humans do.

This is just a hypothesis but that's probably because

we tend to want to amorphi robots a lot.

So, the moment we see some evidence

of human-like intelligence,

we generalize that to different other avenues where we

expect robots to have human-like capabilities as well.

[Narrator] Getting along with robots also means

perfecting even the subtlest interactions.

Take self-driving cars.

We're probably going to want them to drive a lot like we do.

Imagine that you're looking across the road.

Ideally, drivers would gently slow to a stop to signal

that they are in fact, going to stop.

(screech)

A self-driving car though,

could easily wait until the last minute

to stop if it wanted.

Just imagine being a pedestrian,

seeing a car come towards you at 50 miles an hour

promising that you can go.

Maybe you wouldn't because you wouldn't necessarily trust

that it would actually stop.

And compare that with a car that makes sure to slow down.

Not because it has to,

because it could just stop at the last moment,

but because then you actually understand that,

okay, it turns out that this car will stop for me and so

I'm actually comfortable going.

[Narrator] Where this gets more complicated is the fact

that human drivers also communicate with hand signals.

For instance, to wave each other through.

Or, other hand signals too.

A robot car can't do that.

So how will it still fully communicate with other drivers?

It's certainly an interesting problem.

Regardless, self-driving cars and other robots will

have to be customizable if we expect them to not

drive us crazy.

You can't have robots rely on a

universal model of a person.

That's just ridiculous.

They need to actually customize their understanding of

people based on the individual that

they're actually interacting with.

And it won't happen in a split second but,

through the interaction,

they get more and more observations

of what kind of driver are you.

How much help do you want around the house?

How much should I be assisting you?

[Narrator] So get ready to share a future

with the machines.

It'll be fun, I promise.

And also awkward.

Hey, Curry, I love you.

(chime)

No.