Next Up for AI? Dancing Robots | Catie Cuan | TED

I have an unusual job.

I teach robots how to dance.

In 2017,

I transitioned from dancing on stages with humans

to dancing with robots.

Big robots, small robots,

single-arms and roving robots.

Robots that fly,

vacuuming robots.

A lot of robots that look like humans.

And a whole bunch that don't.

I'm pretty sure I've choreographed

and danced with more different types of robots

than anyone on Earth.

Why do I do this?

Because robots move

and choreographers understand how movement affects us.

If a robot slides politely out of a doorway to let you pass,

might make you feel seen and acknowledged.

If a robot marches quickly towards you and avoids you at the last second,

it might cause revulsion and fear.

Robots are beginning to show up in our everyday environments,

from sidewalks to offices, backyards to hospitals.

And they will be threatening and confusing to us

if we do not carefully examine how they move.

I'm also a roboticist and an engineer.

I study this connection between robot motion and emotion.

I combine my expertise in robotics and dance

in the emerging field of choreo-robotics,

the intersection of dance and robotics.

Right now, we're giving AI a robot body with physical intelligence.

AI technologies like GPT and Gemini are becoming the robot's brains,

which means that robots are getting smarter.

Much smarter.

Think of what AI has already done to text and video generation.

We are just scratching the surface on seeing what AI can do

in the physical world.

We are going to be interacting

with these new intelligent robots more and more,

interacting with them through movement.

Roboticists already spend a lot of time thinking about movement,

but they think about it in terms of efficiency,

optimizing for speed or success.

That's not enough.

We need to teach robots how to dance.

If a robot can dance,

make every movement with intention, agility, balance,

emotional expression and grace,

I believe they will also become better caregivers,

teachers and companions.

It will help us build better general-purpose robots

that we actually want to live amongst us.

If we teach robots to dance,

we expand the future of intelligent machines.

Choreographers and engineers have made dances with robots since the 1990s,

and these highly scripted performances have captured the public's imagination.

They've made us feel wonder,

made us laugh,

made us want to dance along,

and sometimes made us never want to meet a robot in real life.

Dancing with robots is no longer just about art or entertainment.

It is about transferring the lessons that one, all movement

changes emotions,

and two, contextual movement is critically important

and applying those lessons to robots in our everyday lives.

Robotics engineers are beginning to consider emotion and context

when designing intelligent machines that interact with us.

The choices we make now

are going to dictate how we live our lives with robots.

If we get this right,

robots will be more welcomed, safe, and delightful.

Humans will be more empowered and comfortable.

Before AI, programmers needed hours to script a simple

dance sequence for robot to perform.

Just like they needed hours to script the robot to open a single door.

With AI, you can teach the robot to open just a few specific doors,

and it will learn to open all of them,

even ones it hasn't seen before.

It's also true for dance.

You can teach the robot to dance with a specific person,

and it will learn how to dance

and move with many others

in many different environments and circumstances.

This is what I did at Everyday Robots,

then a robot AI moonshot at Google.

But rather than teach one robot,

I used AI to teach 15 robots how to move together as a flock.

We imagined a world where you could walk down a hallway filled with robots,

and they would part to make space for you,

like a flock of doves or a crowd of people on a city street.

Where a robot could navigate seamlessly and even beautifully through a busy,

chaotic Times Square.

The world we live in is complicated for robots to understand,

so it was impossible to script or choreograph interactions like these

step by step.

We needed to use the magic of AI

to teach the robots how to improvise and react.

This was a massive challenge that had never been done before,

and the result is an interactive performance and installation work

called Project Starling.

This took us two years of hardcore engineering to complete.

We trained an AI agent on the preferences of a choreographer,

me.

I chose the most engaging patterns and sequences.

We also taught the robots how to recognize

and respond to human gestures.

While we were at it,

we turned the robots into musical instruments.

Each joint on the robot was mapped to a different sound.

When the robot's torso moved, triggered a bass sound.

Its hand opening and closing triggered a bell sound.

The robot's actions created its own unique symphony.

This is what Project Starling looks like.

Our AI generated movement and our robot generated music.

(Music)

(Music ends)

(Applause)

I wanted it to feel like stepping out of your normal reality

and into a waking dream,

one that was so far from where we stand today.

Yet, all the same felt like a world

where your kids could grow up and feel calm,

where you could be surrounded by machines and it felt natural and safe.

People were drawn to just wander amongst these robots, you know,

some were a bit tentative at first,

but then they began to engage and explore

with curiosity and playfulness.

People kept coming back.

What we learned about gesture, navigation,

human interaction and expression can be applied to robots more generally.

Think of the following future scenario.

A robot in a senior living facility,

waiting in a crowded cafe.

You sit at your table,

you wave over to it, make a sipping gesture,

and the robot understands that you would like a glass of water.

And when it drives over to you with its glass of water,

it nods at the other people it passes

and pauses graciously to let them go by.

And when it arrives at your table, it places the glass gracefully down.

And when you thank it, it nods in acknowledgment.

Maybe it even performs a small celebratory dance in place.

We can choose to build and program robots that we welcome.

Robots that make us feel understood.

Robots that we want to spend time with.

The promise of robots that will live and work amongst us

has existed for a very long time.

From Karel Capek's 1920 play "Rossum's Universal Robots"

to The Jetsons' Rosey the Robot

to C-3PO of "Star Wars" and even Terminator,

these stories have been deeply impactful in what we imagine and build.

Yes, we can have robots that do our chores.

But we can also have robots be our dance partners.

Humans do a range of things,

ceramics to diplomacy,

singing in choirs to vertical farming.

Imagine a world where robots can do a myriad of things

that are functional or creative,

and everything in between.

General-purpose robots are coming,

and they will help us to address big demographic challenges

like aging populations and massive labor shortages.

And how they move will unlock

whether we let them into our homes

and welcome them in our workplaces.

Because we want to feel like we're still marvelously human

in this increasingly AI world.

And we can,

because we have the power to choreograph the future we want.

Thank you.

(Applause)