Ep156 What Do We Learn About AI by Dancing with Robots? with Catie Cuan

Why do we read so much into how a robot moves, and what does that tell us about human brains? Why did our history make us so sensitive to movement? Why do we trust graceful motion? Should we make a robot 'look' at an object it’s about to pick up, even if it doesn’t need to? Is movement the original form of animal intelligence? Join Eagleman with guest Catie Cuan, a roboticist, dancer, and choreographer. Catie’s an expert on the strange social interface between humans and machines, and she’s gotten there by dancing with robots.

Episode Transcript

Speaker 1 (00:00):
Why does the way that a robot moves change how
we feel about it? And what does that tell us
about human brains? For example, should we program a robot
to look at an object that it's about to pick up,
even if it doesn't need to look because it has
cameras all around? Why did our evolutionary history make us
so sensitive to movement? Why do we read so much

(00:21):
into it? Why do we trust graceful motion? Is movement
the original form of animal intelligence? Today we're joined by
doctor Katie Kwan, a roboticist and a professional dancer and choreographer.
Katie is an expert on this strange social interface between
robots and humans, and she's gotten there by dancing with robots.

(00:49):
Welcome to Inner Cosmos with me, David Egelman. I'm a
neuroscientist at Stanford, and in these episodes we sail deeply
into our three pound universe to understand why and how
our lives look the way they do. Here's something that

(01:18):
happens all the time in modern life, and most of
us barely notice it anymore. You walk into a grocery
store and the automatic door slides open before you touch it.
If you live in a big city, little delivery robots
roll past you on the sidewalk. If you have a
modern car, the steering wheel vibrates and pushes back against
you if you drift across the lane. These technologies are

(01:41):
all moving, by which I mean something is shifting in
space or changing velocity, or turning towards you, or pausing
or accelerating, or signaling intention. Now, I know that I'm
guilty of thinking about the digital revolution mostly as zeros
and ones, But it's more than that, And as we're
now entering the dawn of robotics, it's about to be

(02:04):
much more than that. We are going to be surrounded
with moving computers, and movement is actually something much closer
to what your brain understands. Long before language and writing
and agriculture, animal brains evolve to answer a very old question,
what the heck is moving out there? And why? Nervous

(02:25):
systems became exquisitely tuned to reading emotion because survival depended
on it. So your brain, like all animal brains, is
constantly inferring intention from movement. You look at tiny changes
in speed or hesitation, or smoothness, or rhythm or posture.

Speaker 2 (02:47):
These are all.

Speaker 1 (02:47):
Things that matter to your brain For example, you can
tell when somebody is angry from the way that they
close the cabinet. You can tell when somebody is afraid
from their body postures they're walking across a parking lot
at night. You can tell when somebody loves you from
the way they move toward you after a long absence.

(03:09):
So we are movement reading machines, and now, for the
first time in history, we're entering a world where huge
numbers of non living things move autonomously around us. And
that's of course totally new, because for millions of years,
everything that moved on its own belonged to nature. But
now the movement around us includes machines that we program.

(03:34):
We have cars driving themselves. We have drones buzzing over
our heads. We have warehouse robots navigating aisles. We have
humanoid robots sloping across factory floors. Whenever I'm in the
restaurant near my house, there's a robotic bus buoy. It's
essentially like a big cylinder that carries the food back
and forth from the kitchen. And many people have pancaked

(03:57):
sized robots that vacuum the house. And when you watch
a moving robot, your brain has to decide very quickly,
what is this thing? What does it intend? Is it safe?
Is it aware of me? Now here's a strange psychological
thing that we're going to talk about today. A machine
can perform the exact same task in two different ways,

(04:18):
and one version feels reassuring and the other version feels
unsettling because our nervous systems assign meaning to movement. So
we're about to spend this next century surrounded by moving machines,
which raises this new set of questions. What makes a
robot feel graceful? Why do we instinctively search for what

(04:41):
the eyes are doing and what that tells us about intention?
Why do some movements make us feel uneasy? Could robots
eventually feel socially legible to us the way that animals
and humans do? And what can dance teach us about robotics?
Now that might seem like a weird question, but dancers
and choreographers spend their lives studying something in fine detail

(05:06):
that engineers rarely, if ever, think about, which is the
emotional meaning of motion, The arc of an arm, the
timing of a pause, the smoothness of a transition, the
emotional payload that gets carried by velocity and rhythm, and
today's guest lives precisely at that intersection. Katie Kwan is

(05:29):
a roboticist and a dancer and a choreographer. She's an
adjunct professor at Stanford and an entrepreneur working at the
frontier of human robot interaction. She earned her PhD in robotics,
specializing in what's known as imitation learning, which is teaching
robots to learn tasks from human movement. She's performed professionally

(05:50):
in hundreds of dance performances involving humans and robotics moving
together on stage. Here's Katie Kwan. So, Katie, you say
that how a robot moves changes how we feel about it.
Why is movements so psychologically powerful?

Speaker 3 (06:10):
Humans have moved together in large groups as long as
we've been humans. Every early human society shows evidence of
dancing and collective movement.

Speaker 2 (06:21):
Because if you.

Speaker 3 (06:22):
Can dance, you can hunt very large animals, you can
win wars against other warring tribes, you can celebrate successes
and create strong intersocial bonding. So movement is this very
core evolutionary quality of what it means to be a person.

Speaker 1 (06:41):
Let me understand that. Why if you can dance can
you do these other things because.

Speaker 3 (06:46):
You can coordinate large numbers of humans together, I see
with their movement, and this involves things like being able
to mirror other people's movement, being spatially aware of where
they are, calling verbals that are associated with timing, and
movement that has intention and is not only abstract. And

(07:07):
so if you can coordinate dancing together in very large groups,
you can be tremendously effective at these other things that
require a movement coordination. There is a great headline and
Scientific American in twenty seventeen which said does dancing just
feel good? Or did it help early humans survive? The
takeaway is that it's both. The other flip side of

(07:30):
why movement and a robots movement is super important to
us is because how we perceive the world has so
much to do with what in the world is moving
and why because we used to get chased down and
eaten by large predators, so we have this sharp acuity
as to what movement means. For example, almost everything that

(07:52):
we've ever been exposed to that moves autonomously is a
part of nature. That's things like trees moving in the wind,
squirrels running up and down the field of a park
watching other people walking versus running. Almost all the movement
that's autonomous that we've been exposed to is a part
of nature that's completely changing now because robots move autonomously,

(08:17):
but we don't have hundreds of thousands of years of
evolution to help anticipate whether the movement that they're doing
is safe or scary. And that's where how we choreograph
a robot and how it moves is going to be
table stakes when it comes to people accepting and feeling
safe around these tools.

Speaker 1 (08:34):
And so if the same robot accomplishes a task with
two different ways of moving, what does that cause in
our brains.

Speaker 3 (08:44):
Let's say that you and I are sitting here having
a completely normal conversation and then I somewhat eerily grab
my water bottle while I'm still making eye contact with you.

Speaker 2 (08:53):
How did that make you feel? Just now?

Speaker 1 (08:55):
Right? So for the listeners who sort of reached out
kind of like a robot and grabbed it while breaking
her gaze, so, yeah, it's weird. It was weird to me. Really,
it was like a zombie movie. I guess. Normally we
attribute attention and intention to where your eyes are and
so somehow you were just reaching out as though you
already knew the bottle was there, and picked it up

(09:15):
like a robot.

Speaker 3 (09:16):
Right, And let's say, I'm now, for everyone who's not
watching the video but listening, I'm going to put the
bottle back on the table.

Speaker 2 (09:24):
How did that feel?

Speaker 1 (09:25):
Okay? Now she picked it back up. Looking at it, it
looked much more normal, to the extent that I didn't
even notice you were doing in a sense.

Speaker 3 (09:31):
Sure, So robots have sensors that are different than ours. Right,
they have time of flight sensors, they have light r
they have cameras that are in all different kinds of positions,
and it gives them a type of spatial awareness which
is very different than our spatial awareness.

Speaker 2 (09:44):
So robot might be able.

Speaker 3 (09:46):
To grab a water bottle without quote unquote looking at
it with its gaze because it probably doesn't have eyes
the same way that we do. That feels very creepy
for us as people, because over the course of your
entire life, you've seen millions, billions, I don't know, trillions
of interactions with other people where you've learned to anticipate

(10:08):
what their movement means, and that if I move my
gaze towards something that means that my attention is on
it and I'm likely to take an action, whereas a
robot doesn't need to do that because the way its
body is constructed is fundamentally different than a human body.
And for someone especially who's not that exposed to robots
or doesn't spend a ton of time with them, maybe

(10:29):
you're an eight year old in school, maybe you're an
eighty eight year old and a caretaking facility. If your
first exposure with a robot, the designers and programmers and
creators of that robot are not socially attuned to these
types of very minor cues, then you are going to
scare and alienate a whole lot of people because you're
not thinking about these intangibles which actually affect the human

(10:52):
perception of the machine.

Speaker 1 (10:53):
You know, this makes me rethink something. So I saw
a terminator. I think it was too where he comes
back and these sort of help the kid Schwarzenegger, this
terminated robot, and one thing that struck me is he
you know, he's always turning around and looking at what's
going on. And I thought, gosh, if you're a robot
from the future, you don't need to have two little
cameras on the front to look around. You can have
eyes everywhere, and you don't need to turn your stupid

(11:16):
head all the time. But what you're saying really changes
my view on this suddenly, because there is something useful
for us about a robot that moves the way that
we do, if only because we're used to that? Is
that right?

Speaker 2 (11:32):
Absolutely right?

Speaker 3 (11:33):
If you look around I live in San Francisco, there's
Weamo's driving everywhere. It's amazing because they have view and
gaze all they have a three sixty vision of understanding.
They have sensors all around the exteria there, and still
when people go to cross the street, they look into
the driver's seat to see if there's someone there to
validate yes, indeed, I'm safe across the road, because that's

(11:55):
what we're so accustomed to. Weimo has a nice user
signaling mechanism that they've worked around with this, where they
have a small screen that flips on top of the
autonomous car that indicates to the pedestrian.

Speaker 2 (12:08):
Yes, I see you.

Speaker 3 (12:10):
But it's so essential that when you're interacting with a machine,
it's not only that you see the machine, but you
want to know that it also sees and is legible
to you, and that relationship is I think fundamentally under explored.
Researchers have looked at it fairly closely, and a lot
of different university settings and social roboticists, you know, people

(12:33):
who are making machines that are specifically designed for human
intention think about this very deeply. But I think we
have a massive new class of robots that are cropping
up right now, tremendous amount of investment going into the
space where I don't hear leaders from many of those
companies highlighting the importance of making their new class of

(12:54):
robots clear and legible to the humans that they're going
to be around.

Speaker 1 (12:58):
Okay, this is great, So let's go into your origin
story and then we'll come back to these issues. So
you are very special because you are both a dancer
and a roboticist. So tell us your path to get here.

Speaker 3 (13:11):
That's very generous thing to say, thanks, David. I think, well,
so I've as long as I can remember, I've always
been in love with numbers, and I always love to move.
My family's cuban. My dad says, if you don't know
how to dance, you're probably not a Cuban, and so
my whole family grew up doing social dance, and I
trained in dance most of my early life and also

(13:33):
was very athletic. I played volleyball, I did track and field,
and I always had this balance where I loved my
math and science classes and I always love to express
and move in choreograph. Specifically, I love to choreograph, and I.

Speaker 2 (13:46):
Had some exposure to the tech world.

Speaker 3 (13:47):
I got to intern at a couple of big tech
companies when I was in college, and then my first
job out of college, I did private equity consulting at
one of the big firms. Realized very quickly that wasn't
for me. I used to sit in my first job.
I was twenty one in my suit on the fourteenth
floor of a building in Times Square and looked down
into the Broadway dance studios. Literally I could see straight

(14:09):
down into them from my desk and thought, that's my
childhood dream, that's my other life. If I don't pursue
this now, I'll never do it. Wound up leaving that job.
I became a professional dancer, so I got to dance
at them at Topolitan Opera Ballet at Lincoln Center in
New York. I had my own dance company. I had
this very rich artistic life, and because of my background
and technology and business, I also was always leading early

(14:33):
stage tech companies of some variety, and so I had
this very bifurcated existence between being a professional artist and
being in the technology world. I knew on some level
I wanted to combine those things. And around that same time,
my dad got really sick and spent quite a bit
of time in the hospital. English is his third language,
he's an immigrant, and he was surrounded by all these

(14:56):
giant machines and they were beeping and making noise. And
if you've spent an extended period in hospitals, depending on
what facility you're in, there's whole parts of that experience
that are really emotionally taxing. And I think to see
my dad's experience, especially in relationship to these big machines,
was an obvious lightning bolt moment for me of I

(15:18):
should take my skills as an artist, that artists know
how to oscillate and change emotions and apply that to
the design of these new technologies and tools. And so
my life took another right turn because then I decided
to go get a PhD in robotics, and so I
did my master's and my PhD at Stanford, and imitation
learning was my specialization in my thesis. So it's a

(15:39):
sub field of machine learning that's specifically about robots and
learning tasks from people. Did that at Stanford and then Google.
Spent quite a bit of time at Google while I
was doing my PhD, and still spend quite a bit
of time at Stanford where I'm an adjunct professor there now.

Speaker 2 (15:55):
So I've gotten to.

Speaker 3 (15:57):
Have this semi circuitous But for me, it always seemed
very aligned with this exploration of numbers and expression and
really about making nascent technologies more human centered, whimsical, and
widely accessible.

Speaker 1 (16:14):
And so one of the big projects that you did
was dancing with robots, So tell us about that.

Speaker 3 (16:21):
I've choreographed and danced I think with more different types
of robots than anyone on earth.

Speaker 2 (16:28):
If someone else had let me know.

Speaker 3 (16:31):
Because we should exchange notes since it's a cuckoo experience.
But yeah, I've used robots in quite a few artistic
and research projects, everything from the giant abb industrial robots
that solder welds on cars all the way down to
small rumbas that drive around, and I've made a bunch
of my own robots for different projects and for me,

(16:53):
the first time I ever danced with a robot, which
was during a residency I had at the University of
Illinois Urbana champagn with Amy Levis, who was the professor
there at the time, was one of the most incredible
experiences of my life because I felt myself imbue my
physical agency into this machine and then interact with it

(17:15):
so I could choreograph it to do something and then
also experience it in real time. Instead of I'm going
to make you do something, now you do it. It
wound up becoming a bi directional interactive experience.

Speaker 1 (17:26):
As in you reacted to it m hmm exactly And
did it have sensors so it could sense what you
were doing or it was pre programmed.

Speaker 3 (17:35):
All the robots I've worked with have some form of sensing,
which means that their behavior can be nondeterministic. I mean,
even if you have sensors, you can still be doing
a pre choreographed motion. But in the first robot I
ever got to use in that setting was actually called
a Baxter robot was made by a company called Rethink
Robotics is Rodney Brooks' longtime roboticist, made that sort of bilateral,

(17:59):
two armed, symmetric, vaguely humanoid robot that was supposed to
be used in collaborative settings, and it also seemed very safe,
but also risky because you can't anticipate what it's going.

Speaker 1 (18:11):
To do in what way, as in it it'll make
movements that you didn't see coming.

Speaker 3 (18:16):
Well, I think all robots, even if they're very reliable,
they're complex systems, so they do things all the time
that surprise you. Maybe they wind up in a configuration
that they weren't supposed to be in and it stops moving,
or maybe it's been running for longer than it should
have been and it gets too hot in times out
or in that case with that robot, the backs to

(18:37):
robot because of where it was in the lab, there
was sort of static charge everywhere, and I'd been shuffling
my plastic sneakers against the flooring and then at some
point the computer actually zapped because of all the slot
collect we had to you know, no one could have
anticipated that that would have happened. Of here we are
wearing sneakers and now it's creating problems for the robot.

Speaker 2 (18:57):
But I enjoy that. Actually, I like.

Speaker 3 (18:59):
Working with complex systems that are not one hundred percent
predictable all the time because you need to be on
your toes.

Speaker 1 (19:06):
So for somebody who hasn't seen dancing with robots, give
a sense of what that looks like.

Speaker 3 (19:13):
Robots move in ways that are very different than people.
The most obvious example is the drone. This is something
that flies, so it moves off the floor and has
a completely different dimension than humans do. We can jump,
but we don't stay in the air for very long.
I went to the San Francisco Valley this week and
I was amazed at.

Speaker 2 (19:28):
How high they can jump.

Speaker 3 (19:29):
But they're not hanging in the air right, So so
robots move in ways that are very different than how
humans move, and for me, as a choreographer, a maker,
and now an entrepreneur, it opens an enormous canvas of
things to play with. It's almost like if you or
a person who exclusively played acoustic guitar and then you

(19:51):
discovered that you could also play the electric guitar, and
you could use a loop pedal and you could.

Speaker 2 (19:57):
Broaden the range of or.

Speaker 3 (20:00):
For you, David, you know you think about, I'm working
as an author, but I'm also now I can speak
to the things that I'm writing, and I can put
them in new mediums and modalities, which are films and
not only in fiction books. And so it all of
a sudden it opened does up this new canvas of
ways that you can work. And so for someone who's
never seen a performance between a human and a robot,

(20:21):
you're depending on the kind of machine that gets used.
You'll see types of motions that are a little bit
foreign or alien to you, and that can convey different
types of stories or different ways of being. I often
use robots in my pieces as a character, but it
doesn't look like a human looks as a character, And

(20:41):
so what does it mean for those things to still
convey emotion and story when they don't look like people.

Speaker 1 (20:47):
One thing I've been wondering about with robot movement generally
is the uncanny valley where for anyone who does know
this term, it's that you know, if a robot looks
more and more human, gets closer and closer looking human,
at some point just before it gets there, there's this
sort of uncanny thing where it looks kind of almost human,
but it's not quite right. But tell us. Usually we

(21:09):
think about that in terms of looks, but tell us
about that in terms of movement. And when it comes
to dancing with robots, does that matter or is being
in the uncanny valley kind of an interesting place?

Speaker 3 (21:21):
Have you ever seen a robot that you, for you,
has put you squarely in this uncanny space?

Speaker 1 (21:26):
Yeah, which one Sophia, And these various robots that are
almost human but they're kind of jerking around a little
bit and their faces move around in kind of strange ways.

Speaker 3 (21:37):
Right, So I've seen a few of these robots that
for me put me in the uncanny valley. And the
first question that always comes up for me is why
you know this is a creepy space to be in.
I'm not sure I want to be in this space.
And I also it doesn't draw me towards the machine.
It makes me feel more alienated. I think where that

(21:58):
shows up in terms of movement. So movement's amazing because
it happens over a time horizon. It's the difference between
a picture and a film. So it's taking a snapshot,
which is a pose, versus seeing a movement actually unfold
over time, which is your actual movement sequence. And I
think when we design these kinds of trajectories, the beginning,

(22:19):
the middle, and the end are all important. How it
started or why it started, what the movement looks like throughout,
where it resolves and ends, and then where it goes next.
You could even argue that you're in a choreography for
your entire life. From the moment that you are born
until the moment the lights go out, you are moving continuously.

(22:39):
You are in one long choreographic sequence. I think where
this can sometimes get creepy from an uncanny Valley standpoint
is if those transitions, so why I started a movement
is unusual, unpredictable, strange, where I am in the middle
of that movement, and then how it gets resolved. And
I guess the best exact sample as I could give

(23:00):
you is like if I was sitting here right now
and then all of a sudden it did some sort
of big jerker shape, that would be a discontinuity. You
wouldn't understand where it came from. If you saw that
kind of thing on a stage, you would ascribe meaning
to that continuity because you would say, oh, here were
all these dancers moving in soft graceful light patterns, and

(23:22):
then all of a sudden one of them jerked around.

Speaker 2 (23:24):
What does that mean? You know?

Speaker 3 (23:27):
So, I think with robots we need to be thinking
about sequence, which is the beginning of the middle of
the end. We need to be thinking about discontinuities. And then,
of course it's always relational. So if a robot has
been moving near you doing things that you can follow
and understand, and then stops doing that for some reason,

(23:50):
you're going to ascribe meaning to it.

Speaker 1 (23:52):
So is that useful when you choreographed to have robots
do things that are a little bit inhuman I think that.

Speaker 3 (23:58):
Robots should move like for me, the expressive potential of
them not being the same as people, you know, having
motors or linear actuators that are different than ours are
I enjoy that, you know, I like the It's almost
if you were a painter and you said you like
the fact that painting with a brush is different than

(24:19):
painting with your fingers. You know, it's it's experientially different.
You want to highlight with those changes. It doesn't mean
that painting with your fingers is worse or painting with
your brushes are worse or better. You couldn't qualify. It's
they just provide different artistic expressive potential.

Speaker 2 (24:36):
You know.

Speaker 3 (24:37):
I think as a choreographer, I'm always interested in how
can I use these devices to say something about the
world that I wouldn't be able to do without them.
So it's not marketing on behalf of a robot. It
is saying something about what it is to be human

(24:57):
in this moment in twenty twenty six that I couldn't
say otherwise.

Speaker 1 (25:12):
As a choreographer. What do you see in motion that
an engineer would ignore or just not notice?

Speaker 3 (25:20):
Engineers mostly think about how can I have a robot
efficiently perform a task. What that typically means is I
want to minimize the amount of energy that this robot takes,
which often is very closely linked to I want to
minimize the time it takes to do that. For example,
if you have a robot in a factory that needs

(25:42):
to pick objects out of a bin, you need to
meet a certain threshold of how many picks you perform
per hour, and you want to reach a certain threshold
of accuracy so you pick the right object when that
command is required. So I think most of my colleague
engineers are thinking very much about efficiency and the time, energy,

(26:02):
accuracy domain, et cetera. When I'm thinking about choreography with
robots that need to be around people, Yes, it's important
if you had a robot that was near you in
a restaurant and was taking a glass off your table,
for example, for it to grab the glass and not
to grab your purse, or for it to grab a
plate and not to grab your phone. Accuracy is important.

(26:24):
Of course, we want it to be fast and efficient
enough that provides value to the restaurant or whoever the
buyer would be in that case, and you are at
a restaurant, you're having a good experience. You want to
enjoy the food. You chances are there's good lighting in there.
You probably picked a nice shirt to be able to.

Speaker 2 (26:43):
Wear at a dinner.

Speaker 3 (26:45):
You don't want something that screeches and scars you or
interrupts you in the middle of conversation. And that's where
I think choreography has a tremendous role to play in
terms of taking that movement from not only being accurate
or efficient or subcases sacrificing some of that accuracy efficiency
timing in order to be safer, legible, and more empowering.

Speaker 2 (27:06):
For people to be around.

Speaker 1 (27:08):
So what's an example of the kind of way that
you might sacrifice sufficiency to make let's say a waiter
robot you feel more comfortable.

Speaker 3 (27:17):
I think you know the path that you take in
order to get to a table, right, So it might
be that the shortest path requires you to bypass a
bunch of people who are in the middle of conversation,
cut through the middle of them. Or it might be
that the fastest path for you to be able to
grab that dish is to move over someone's arm for example.

(27:37):
Those things are rude, I mean, they're not polite. You
want a robot in a restaurant that takes the path.
Maybe that's a little bit longer, but doesn't stop to
people who are flirting at the table from being continuing
to be able.

Speaker 2 (27:49):
To talk to each other.

Speaker 3 (27:49):
So I think it's paying attention to social cues and
being able to abide by them that would warrant reducing
your efficiency.

Speaker 1 (27:59):
Evit, move around to the other side and gently take
the thing out and so on. Got it. So efficiency
is not always the acme that we're shooting for. It's
other kinds of social cues that we get from movement.

Speaker 2 (28:12):
Sure, which we can call those things cost functions.

Speaker 3 (28:14):
So we have ways of describing this to robots algorithmically
and mathematically and saying, hey, there's a cost associated with
when you're reaching in between two people who are sitting
next to each other to grab something off of a table,
you should be going around. So I think there are
very seamless ways for us to be able to do
this if they are things that we value.

Speaker 1 (28:33):
Yeah. Cool, And this comes back to that second question
that I asked you about. If the same robot performs
a movement in two very different ways, how does that
strike us very differently in terms of our brains. So
in the case of grabbing over the people versus reaching
around in a nice way, we have all kinds of
social interpretation on that. Yeah.

Speaker 3 (28:54):
Plus, I mean, the saturation of robots in society right
now is really low. So these types of first exposures,
if they go well or poorly, they affect people's impressions
of these new tools. And you'll notice there's a whole
new genre of robots that do spill detection and grocery
stores and pharmacies or big box retailers like Target or Walmart.
I'm not sure if all of them have adapted these,

(29:15):
but I've seen them in a variety of different locations
and so funny because they are these very tall columnal
robots that can move in any direction. But what the
company has done is they've stuck googly eyes on them
because people need a place to look. They need to
know where do I direct my gaze? And so even
if you, as a robot designer, don't put this on

(29:36):
the robot yourself, chances are humans will come up with
workarounds to make them more palatable.

Speaker 1 (29:42):
Now you've done movement analysis on your robots, and the
question is how do you translate movement into code?

Speaker 3 (29:50):
Wow, so many different ways we can generate trajectories. So
we can have big long lists of numbers that are
positions in space. So it's for every single motor that's
on a robot, we're going to give it value in
radiance of where we want it to go, and then
it'll go there and it winds up creating a shape
at the end. We can we call do kinesthetic teaching,

(30:12):
which you mean you could put a robot. We call
it in compliant mode, so all it's doing is resistant
gravity and somebody comes up grabs it presses record moves
the robot around with their arms or sometimes with a
teleoperation device that's simply a device that somehow mirrors the
movement of the robot in some capacity, and you can

(30:34):
record movements that way. You can also record movements from
motion captures. So I could be standing in a dance
studio wearing one of those motion capture suits like they
wear in the movies, and move around a bunch and
then decide if there were specific primitives from what I
was doing that I wanted to. So there are so
many different ways that you can choreograph and move a machine.

(30:55):
And I keep using this word choreograph. I think you
could call it animate. You could, I'd say design motion
for or plan you know motion planning depends on how
you cut it, but yes, that would be a part
of it too. So there's so many different ways to
make robots move.

Speaker 1 (31:11):
When a robot moves gracefully, what is it that we
are perceiving.

Speaker 3 (31:18):
You're seeing velocities that fall within a specific band that
roughly mirror the kinds of velocities that we see from
other things in the natural world.

Speaker 2 (31:30):
You're also seeing fairly low jerk.

Speaker 3 (31:33):
This is kind of the when a robots starts moving,
it's whether or not it creates these strong discontinuities between
its sets of motions. I think the other thing that
people don't think about enough, which in my company and
all my performance works we're very attuned to, is actually sound.
We're very sensitive to the combination of movement and sound,

(31:56):
and we're sensitive also to the discontinuities between those things.
If you have a movement that looks very leisurely, where
it's smooth, it's continuous, but you have hardcore screeching, intense
high volume that can be very jarring and uncanny for people.

(32:16):
And similarly, if you're seeing a robot that's super jerky
or creating joint configurations and moving around in patterns which
seem strange, and it's being done to this soft ambling.
So people are very sensitive to that as well, and
have in a product setting to be super clear, not

(32:38):
just on a stage. They actually are very sensitive to
those things in products that ship in the real world.
And robots make a lot of movement because they have motors,
a lot of them have fans typically to keep them cool,
and so we need to be sensitive to the sound
as well.

Speaker 1 (32:53):
Would two people from two different cultures interpret the movement
of a robot differently, or as is it some fundamental
evolutionary issue that we would all interpret the movements same.

Speaker 3 (33:04):
Way, the same way that we have gestures, which mean
different things in different cultures, so gesticulating in general has
different sets of meanings.

Speaker 2 (33:13):
I think movement overall is.

Speaker 3 (33:15):
Perceived differently by people who come from different cultures, and
I think there are universal similarities because of this evolutionary
prior that we have to being very sensitive to.

Speaker 2 (33:26):
Things that are moving fast towards us and don't appear
to be slowing down.

Speaker 3 (33:30):
Regardless of what culture you're in, I think you would
infer that as being threatening or scary, or movement happening
out of the corner of your eye that seems jerky
and discontinuitous, like you would have some sensitivity to that regardless.
So I think gesture one thousand percent varies by culture.
I think there are some attributes of motion which we

(33:52):
can all agree on in terms of how we react.

Speaker 1 (33:55):
So what's more unsettling? Do you think a beautiful, realistic
looking robot that moves in a jerky manner or a
very mechanical, weird robot that moves smoothly like a human.

Speaker 2 (34:07):
I think it's the first one.

Speaker 3 (34:08):
I think when something looks similar to humans, are expectations
for what it can do go through the roof. Rodney
Brooks talks about this a lot longtime MIT professor, founder
of i Robot now robust Ai.

Speaker 2 (34:22):
He says, the closer you.

Speaker 3 (34:24):
Make something to a human, the higher people's expectations are
for it, for what it can do and accomplish. And
so I think the former case, where you have something
that looks very close to a human, but it moves
in ways they're super unsettling, that's hard for people. I
think it also as a builder and a maker, we
have this huge design canvas of what's possible. We don't

(34:47):
necessarily want to constrain ourselves all the time. I mean humans,
we watch Toy Story, we see you Slinky's being animated
on the screen, or mister potato Head. You know these
things that are abstract up but still give a convey
emotion and feeling. And so I think as roboticists we
can also be playing in different corners of the design space.

Speaker 1 (35:11):
And when you think about intelligence, do you think that
that is just a cognitive problem that we could solve
with lllms or do you think that embodiment actually having
a body is part of what we mean by intelligence.

Speaker 3 (35:25):
I just finished reading Michael Polland's new book, A World Appears.

Speaker 2 (35:28):
Have you read this?

Speaker 3 (35:29):
I have? Okay, what a beautiful book. Yeah, he's such
a fabulous writer. He writes about embodiment in the book
with this question of can something truly be intelligent if
it doesn't have a body? One critique I have for
Michael Pollin if you're listening about this book, is he
gestures and speaks towards people of the literary tradition like

(35:53):
Henry James or Virginia Woolf and how they write about
this experience of being conscient, and that they fold language
into these interesting configurations to convey this continuous, abstract stream
of conscious thought. I'm thinking the whole time you need

(36:13):
to have interviewed some dancers for this book, because Hopemore,
who is a choreographer in San Francisco, has this statement
She've put on a lot of T shirts for a
long time, which.

Speaker 2 (36:23):
Said, the body is the brain.

Speaker 3 (36:26):
There is so much to the experience of consciousness and
of being intelligent that has to do with living and
having a body in the real world. And dancers understand
this intuitively because when you look at the training that
goes into shaping and sculpting the muscles, the body, the awareness,

(36:49):
the balance, that's all fantastic. And it's the intangibles of
creating emotional connection while you're performing on stage that you
derived so over the course of performing hundreds thousands of times,
that you get from exchanging and dancing with people in
a studio, and that so much of that intelligence about

(37:12):
being relational and connecting is based in your body and
these body based somatic kinesthetic memories. And I'm thinking, wow,
I love that book, you know, top to bottom, and
how can we describe what's the kind of intelligence that
you get from being that somatically centered for so long?

Speaker 2 (37:35):
So I think you know, AI.

Speaker 3 (37:37):
Has been hugely impactful for all of us on screens
contained into these two D light boxes, in our phones
and on our laptops. Now we're moving into this new
era where AI is coming off of the screen. It's
manifesting in the physical world. Has a physical body, whether
that's a big industrial robot arm that's powered by all

(37:57):
the latest we call them in robotics via so vision
language action models as opposed to LLMS LLMS well how
that whole train sits together LMS Large language models VLMs
Vision language models. So these are models that can look
at pictures and images and describe them in natural language.
And then on top of that, in robotics, we have

(38:18):
what we call vla's vision language action models, So it's
teaching robots how to then take action on the basis
of that knowledge. And in our case, what we're building
at my company we call them VLI is so vision
language interaction models. These are models that are specifically geared
towards teaching autonomous physical agents to interact with people. Your

(38:40):
original question was about how intelligence is going to sit
in the body. I think, for sure, a huge part
of being a human and our experience of our consciousness
is based in our body. Everything from what you ate
affects how you feel to whether you've stretched enough I
think can also affect you know, how open you are

(39:01):
to different sets of ideas, and so we can't separate them.
I mean, we also don't know how consciousness is even
dispersed in the body. I mean, you know this more
than anyone. Whether it's can't only be in the brain,
It kind of sits in all these other places. So
do you think that we can have a true intelligence
that doesn't have some form of embodiment.

Speaker 1 (39:21):
I wonder this all the time. What's clear is that
the three pounds of brain tissue that's where most of
the action is. And the reason we know that is
because you can get your heart replaced and an artificial
heart put in, and you can get a kidney removed,
and you can get your limbs amputated, and you're still
the same person. But if you damage just a tiny
part of brain tissue, it changes who you are. So
we know that that's the densest representation of who you are.

(39:44):
But the way that I think about it is that
the brain is like the city, let's say Manhattan, and
then all the rest of this is sort of the
outlying areas, and there's still a lot of communication between
the main city and the rural areas. There's lots of
stuff happening back and forth. So you know, if something
were to happen to Manhattan, like an earthquake or something,

(40:05):
you got real troubles there. But there's so much signaling
going back and forth between the countryside, and it may
well be that consciences requires all that. I'll just mention
there's a company that's doing cryogenics. They will freeze you
after you die. So you pay a certain amount of
money and you can get your brain frozen. And the
idea is that sometime in the future civilizations will figure

(40:26):
out how to thaw this out. But if you pay
more money, you can get your whole body frozen. And
people have asked me, what's my opinion on whether I
think it's worth it to pay the more money, and
I suspect it's probably worth it to pay to get
the whole body there so that it's still the same you.

Speaker 2 (40:42):
I'm curious as well.

Speaker 3 (40:43):
I read an article from Bruce Hood, who's at the
University of Bristol, who said that the majority of the
neurons in the brain are actually in the cerebellum. It's
the back of the brain that controls your movement, so
that part of the reason we might have a brain
at all is to support our motion. We have something
like eighty percent in this year, and then only twenty

(41:04):
percent control these higher order thought processes of language and
math and things like that. Is that right, well sort of.

Speaker 1 (41:11):
So it is the case that the cerebralm has an
enormous number of neurons in it. And what that indicates
is the absolute importance of movement in the world. The
fact that you only have x number of neurons that
are devoted towards the higher function stuff doesn't say anything
to its relative importance, just its recency. And so what

(41:32):
that tells us is that once we got the movement down,
and especially once we became bipedal and started using our
hands for tool use and so on, then we really
took off as a species because we had the capacity
to ause tools and b as our cortex the wrinkly
outer bit as that exploded in size. Then we became
able to think about possible futures and evaluate hypotheses before

(41:56):
acting on them, stuff like that. So you're absolutely right
that fundamentally we're all about moving this body. And then
secondarily we came up with all this other great stuff
that allowed us take off.

Speaker 2 (42:09):
That makes me think of two things.

Speaker 3 (42:10):
One is this sea snail example that I often given
some of my talks, that swims all around the ocean
and then at some point in its life it plants
itself on a rock and eats its own brain because
it doesn't have to move anymore. OK, yeah, exactly, I
don't have to move. I can reprieve myself of this
big honking brain. That is a fascinating But the second
thing that made me think of as I was recently

(42:32):
exposed to some ideas that Aristotle had, which is, we
shouldn't be called homo sapiens. There are a lot of
things in the world that are wise Homo sapiens, wise man,
you know, intelligent man. There are lots of things that
have what we should be called is homotechne because the
thing that defines our species is our relationship to our
tools and the fact that we have been able to

(42:53):
manipulate our environment to work together in society because we
are such strong wielder of tools. And I think robots
are the next instantiation of that. We're going to have
billions of robots in our lifetimes. I one thousand percent
believe in that it's going to happen super fast, and

(43:14):
they will be extensions of all of the tools that
we currently have in place, powered by many of these
amazing advances, whether it's in light ar or edge compute
or large language models. The difference is that they have bodies,
they are fully embodied and they move around in ways
without our assistance. Think of how many other tools are

(43:37):
entirely reliant upon human intervention, either using them in real time,
whether it's a chainsaw or setting and forgetting like a camera.
But these robots, they're autonomous, embodied moving tools, and that
represents a complete paradigm shift in how we live in
our environment and how we relate to other entities that

(43:59):
are in it.

Speaker 1 (44:00):
It's right, Well, you know, we're all thinking about AI
agents nowadays and how they run off and do things,
but we will have this physically where you'll bump into
robots on the sidewalk who are on a task for
their master and going and doing things and coming back
and reporting in later. Yeah, and so and so. The

(44:21):
part that you think about the most is about how
do these things move in such a way that we
feel comfortable with them, that we appreciate them and see
intelligence in them and so on. So when you imagine
a world forty years from now when robots are just
running all around the place, what is it that you

(44:43):
want to see? What does that world look like?

Speaker 3 (44:45):
Demographics are shifting worldwide really quickly right now. I think
China is one point four billion people by the end
of this century, they'll be seven hundred million. At the
US our birth rates.

Speaker 1 (44:57):
You're saying it's going to happen.

Speaker 3 (44:59):
Oh yeah, the US our birth rate is one point
six the last time I checked. I mean, we have
historically been fine in terms of population growth because we
had a lot of immigration. That's changing a little bit.
If you look at what are the drivers of economic growth,
it's two things. It's population growth or it's innovation. When

(45:19):
we observe worldwide that population growth is declining in many
places except for Africa. Actually, Africa is the only continent
where there's going to be net population growth in this century.
So if you look all around you and you see
this decline in population growth, the other economic driver that
you have is innovation. Okay, well, what does innovation mean

(45:41):
in twenty twenty six embodiment ai agents, et cetera. If
we want to keep growing economic output the way that
we have historically, we need to see those efficiency gains
in other places. And the demand for specific kinds of
labor are still very high, for example, in the US,
and ten years will have more people over the age

(46:02):
of sixty five than under the age of eighteen for
the first time in the country's history history of the
United States. Okay, if you think about that, that means
you have fewer young people who are not only working
and providing labor into the market, but also paying into
some of these systems like Social Security and Medicare, and so,
you know, I won't discuss that now, but I think
we have these enormous, unavoidable demographic shifts that are coming up,

(46:27):
and we will have demand for certain kinds of labor,
whether that's caretaking for elderly people who want to stay
at home for longer, things like data center management and upkeep,
even cleaning solar panels. You know, I heard that as
an amazing application for robots recently, because it turns out
there's a lot of solar panels and when they get dirty,
they're less efficient, and you know, it's not a great

(46:49):
job to be standing out in the hot, hot desert
moving a cleaning baton up and down, and so there's
so many opportunities for robots to be helpful and useful
in our lives. My strong argument is that for humans
to accept those solutions, these robots need to engage and
interact with them in ways that are safe, legible, and

(47:10):
empowering period. Otherwise you and I are going to be
sitting here in forty years. Hopefully we won't both be
criogenetically frozen at that point, but like, you and I
are going to be sitting here in forty years and
being like, why are we surrounded by all these horrible,
ugly robots and machines? And it is too late for
us to change that, because when a new technology like robotics,

(47:32):
which I'm arguing is new ish in this new wave
because of how much they're powered by all these new
state of the art models. When a new technology shows up,
that adoption curve. The near term choices you make are
so impactful because then they get embedded and they get entrenched,
and we've seen it in everything from cell phone technology

(47:53):
and it just becomes the default, de facto way of being.
And I think we have a huge opportunity right now
and we want to land it. We want to land
this robot plane correctly because it is going to affect
how we relate and engage and interact with these machines
in the future.

Speaker 1 (48:20):
So I want to come back to two quick points. One,
is this a technical point that I'm curious about. So
everybody's heard about large language models. These are based on
a transformer model. And what it's doing is it's taking in, Okay,
there's a bunch of tokens. Let's think of them as
words that have come in. What's the next word that
I should put here? Given this huge context window? Is

(48:42):
it the case that these VLAs you mentioned in the
vlis are doing the same thing where they're taking in
a bunch of context and saying, hey, what's the next
token here? But the context now is hey, let's say
I'm moving my arm and I've gone through this motion,
this motion, this motion. Where am I like to move next?
Is it the same structure?

Speaker 2 (49:02):
It really varies.

Speaker 3 (49:04):
I mean lots of different companies have different approaches to
this problem. When I was at Google, it was largely
a transformer architecture. So the big seminal paper that blew
this up was in December of twenty twenty two at
a conference called CORAL, the Conference on Robot Learning. And
it was a paper that came out of Google, and
it was called do as I can, not.

Speaker 2 (49:24):
As I say.

Speaker 3 (49:25):
So they called it say cant and paper was the
name of this paper, And what was amazing about it
was they took this LM, so they had a transformer
that was also helping with breaking down long horizon planning.
So if I tell you, David to go clean a bathroom,
you understand that that means you need to get a mop,

(49:46):
you need to find a bucket, you need to get
a sponge, you need to get some cleaning solution. Then
sequentially you need to navigate to where the bathroom is.
Then once you're in the bathroom, you've got to put
down your tools and prior to a lot of this
transformer work, and you know the big blow up in LLLMS,
we used to do a lot of that long horizon
planning semi manually, right, so you could take a pretty

(50:08):
abstract task and then figure out how to break it
down into all these subtasks. Now, go ask Claud. Claud
is very good at saying, Hey, Claud, I'm a robot.
I need to clean up this bookshelf. What are the
different micro tasks that I would do? And Cloud can
give you a nice long list of what you would
need to do to clean up the bathroom, by the
bookshelf or anything anything else. Part of the major innovation

(50:29):
in this sacan paper was it then took those sets
of instructions and actually mapped them on to actions that
the robot could do. So the robot can have a
set of we call them parameterized actions. It's things like
go to might be an example. A robot can go
to a specific position in space. Right, you can give it,

(50:50):
you know, an XYZ point and an orientation, and you
say you're going to go to this point in space. Well,
you stick together this long list of actions that you
know you need in order to clean a bathroom. You
stick together your perception of the environment, so where different
objects are located at what point in space, and a
frame that's relative to where you are as the robot.

(51:11):
And then you tack on the calling of all these
different parameterized commands and boom, you have a robot that
will do things end to end in the real world.
And so part of where that core innovation came from
was actually stringing together these sets of prior knowledge that
then turned into a very valuable output at the end,

(51:33):
which is that you could give a robot a language command,
and one of the ones that they give in the paper,
or it might have been in the demo, because I
was at when they did the live demo in New
Zealand and December twenty twenty two, was grab the spicy
chips for me, you know, and there's six things on
the counter and it figures out which one of those
is sure enough, a bag of spicy Helipinia chips, and

(51:54):
it grabs that one instead of grabbing the pencil or
the fork.

Speaker 1 (51:57):
I want to come back to you a quick question.
After you years of training as a dancer, when you
watch somebody dancing on stage, do you sort of feel
yourself moving with them? That's my experience when I watch
a dancer on stage. I'm not even a dancer, and
so if so, my question is, what is it when
you're watching a robot doing moves that a human couldn't do.

Speaker 3 (52:18):
I completely agree about the experience that you just described
when you're watching a dancer, because I went to go
see Alvin Alei American Dance Theater in Berkeley a few
weeks ago with my with my mom, and I have
a whoop, so I wear this thing on my wrist
that you know you probably a lot of the people
listening know what.

Speaker 2 (52:33):
A woop is. But during that show, I was not
in low stress mode.

Speaker 3 (52:39):
I was wiggling and shaking up and down between medium
and high meet because I was having the experience that
I was doing these dances, especially some of that choreography,
especially in the last piece, which is a masterful work.
It's called Revelations by Alvin Alee. I've actually learned some
of that choreography, so I felt like I was doing
the performance. When I watch robots do that, it's certainly

(53:01):
not as let's say, I'm not in the high whoop zone.
But I think when I watch a robot move, especially
in the pieces that I've made historically, there's something that's
mesmerizing and hypnotic about it. It feels like I'm on
this long continuum of humans and our tools from fire

(53:23):
and wheels, printing press all the way up until now,
and I we've made this kooky world for ourselves. I mean,
we used to live in caves with fireplaces, and we
would stick our hands on the wall and do paintings.
And now we live in these boxes that we make
and with overhead lighting and electrical outlets and all these
attri It's kind of wild how much human existence has

(53:45):
changed in a relatively short period and what that has
to do with me watching a robot moves. It seems
like this continuous reshaping of what it is to be
a person and to have experiences in the world.

Speaker 1 (54:02):
Love it. One of the classes I teach is Stanford
is Literature in the Brain. And one of the things
that's fascinating about literature is that it can be a
testing ground for things, in the case of a lot
of science fiction, a testing ground for human robot interactions.
My question to you is, to what degree you're in
what sense is dance a testing ground for the future

(54:24):
of human robot interactions?

Speaker 2 (54:25):
What a fabulous question.

Speaker 3 (54:26):
I wrote a paper when I was an artist in
residence at the University of Illinois or Banish Champaign alongside
Amy and Sean. We actually came up with this tactic
we called something called Curtain. We said it was an
experimental test bed for human robot interaction. What we did
was we had designed these sets of diotic interactions so

(54:48):
we would have a person stand up and move around
and follow the agent that was in front of it.
This was a part of a larger performance work and
research work called time to Compile. So, for example, you
have a small humanoid robot, it's moving around and you
have a person that's standing across from it moving. So
we designed these different diotic interactions diet meaning to people,

(55:12):
yes exactly, and that the human who was participating was
always told follow the thing in front of you. So
we did this with a humanoid robot. We did this
with a representation of a humanoid on a screen. We
did this with a body in virtual reality. We did
this with a big, big robot, and then we also
did this with a person. So you stand across and

(55:32):
you follow the person in front of you in time
to compile. The reveal to the audience was that you
are it's a little hard to explain, but you're actually
always following the person who came before you because their
movement data is being captured and that drives the agent
in the following room. So it was this commentary on

(55:54):
the hidden human network and how we're actually always all
interacting and talking to other people. Because when I speak
into a microphone, I'm having an interaction with a person
who designed this somewhere in someplace, and the gentlemen who
came and moved it in the room, and you know,
all of these things are just these abstract representations of
human human hands, so in time to compile, we created.

(56:16):
But what we did was we ran that performance and
that installation as a research study, and so we asked
people what were your impressions of robots before they did this,
and then what were their impressions afterwards. We also ran
a similar research study where we showed a whole bunch

(56:36):
of different objects on a stage, including people, including robots.
We had people rank them in terms of their complexity,
also their intelligence, their capability. Then we did a performance
with all of these things and we had the audience
rate those things again. So they said, oh wow, you know,
the humans are much more expressive than I thought they

(56:57):
were before, and the robots were much less. So we
actually found ourselves using performances and art installations as interventions
for scientific experiments. And I think what's fabulous too about
that is you can play. I mean, you can try
so many different performances and installations, and you know, it's

(57:17):
hard to be able to replicate that across board. But
I think we called it an experimental test bed for
human robot interaction because we were giving people interactions with
robots and then measuring if you're in this heightened artistically
aware theatrical experience with a machine. How does it change

(57:40):
your expectations for what it is?

Speaker 1 (57:42):
You know, there's something I've been wondering about recently, which
is obviously, as a society we've blown past the Turing test,
which is can you tell whether you're talking to human
or AI? And you know at this point the Turing
test has been passed. But it turns out it's actually
really useful sometimes to know whether you're talking to a
bot or a human, because, for example, with political debate,

(58:03):
it turns out, if I'm talking to another human, I'm
really reluctant. There's lots of studies on this. Humans are
really reluctant to change their opinion, but if they know
they're talking to a bot, they're much more open to
the idea of hearing facts and thinking about different ways
they might change their opinion. So when you think about
the future of dance with human and robotics, is there

(58:25):
is some advantage to understanding the difference and seeing, oh,
now I'm interacting with a robot.

Speaker 3 (58:32):
So I taught a class to Stanford last quarter. It
is called Computer Science three thirty four Robots and Arts.
We had students from mostly computer science but designed civil engineering.
A few from the Business School theater, Arts and Performance studies,
and they made philosophical political social statements about the world
using AI and robots in the class and one of them,

(58:54):
one group, did a project where they it was.

Speaker 2 (58:58):
Almost like a research study as an installation.

Speaker 3 (59:01):
What they presented in their final report was they had
a group of people come in and they show them
to artwork side by side, and they said which one
do you like better? And you would vote with a survey,
and then they would show you another two sets of artworks.
In this case it's actually two poems, and you vote.
And then the third was they would show you some dance,
some movement. And what they found is that humans far

(59:24):
and away prefer things that have been made by people.
And after they reveal to everyone, you know, this is
which painting was a human versus an ai, This is
which poem was a human versus an AI. This is
Zeny and Keshav who were in my class. They find
that people will actually switch so they say, I don't.

Speaker 2 (59:45):
Like that painting as much anymore. I think I like
this one. They like the human one more.

Speaker 3 (59:49):
Yeah, And they their project was reflecting a series of
a couple of other papers that we had read in
the class as well, which is, let's say that we're
going to hang out on Saturday, and I'm like, look, David,
we have your choice of two museums that we can
go to. We can go to this one, which has
all the artwork in this museum is made by people,
or we can go to this one, which has all
the artwork in the museum is made by ais. People again,

(01:00:13):
far and away prefer to go to the museum that's
been made by people, or with art that's been made
by people. I think it's not just because of whatever
the art is that's in there, but think about why
you go to a museum. We want to be able
to tell your friends about it. Oh, I had this
great time hanging out with Katie. We went to this
cool museum. You want to be able to relate to
the history, largely the human history of whatever it's been

(01:00:35):
painted about. Maybe it was a specific, you know, religious
happening that was documented in a beautiful art work, and
that's important to you, And so people want to engage
in this human made work for reasons aside from I
looked at it and I had a sharp, immediate emotional
yes or no towards it?

Speaker 2 (01:00:54):
And I think how that.

Speaker 3 (01:00:56):
Relates to dance and if we have robots in the
future that are all moving gracefully and perfectly, is it's
the exact same phenomenon? Do you want to be able
to learn about this dancer? What was their training like?
What were some of the influences that affected how they move?
Have I seen them tour at this place before? Do

(01:01:16):
I get to meet them and have a relationship with
this person? Can I How is their dance emblematic of
these other trends that other people are curious about, maybe
on TikTok or And I think it's all of the
mush around it that gives the richness.

Speaker 2 (01:01:37):
But does it matter to you that a human or
an AI makes something?

Speaker 1 (01:01:41):
Absolutely? In fact, I've been describing this as the effort phenomenon,
which is that given you know, something produced by AI
and something produced by human. Because I'm a writer, I
think about writing all the time, we just don't care.
Even if it were the exact same set of words,
we would really value the thing written by William And
if Claude put out exactly the same words, we just

(01:02:03):
we wouldn't value it. And I think This has to
do with our sense of the effort that goes into it,
because there are these old psychological studies about Let's say
I'm showing you two pieces of artwork. One of them
looks like it was really easy to do, and the
other one looks like it took a really long time.
People always like the one that took a long time.
And I actually think there's an interesting analogy here with diamonds.

(01:02:25):
People will pay a lot more for a diamond, which
is just a matrix of carbon atoms shoved together if
it came from the Earth versus if it was a
synthetic diamond made in the lab. It's the exact same
matrix of carbon atoms, but Mother Nature took billions of
years of effort for this one, whereas this one just
took five days in the laboratory. So we really do

(01:02:46):
care about that. Yeah, I agree with you.

Speaker 3 (01:02:48):
I think Terrence Tao, is a very famous mathematician, has
a statement about this, which is, if you were given
two choices, you can climb Mount Everest, so you can
train for X period of time, you can get your equipment,
you can find the right you can climb Mount Everest
and get to the peak, or choice number two. I've

(01:03:09):
got my helicopter right here. You can hop in. We
can helicopter that. I promise you it's safe. We're going
to helicopter to the top of Mount Everest. Hop out,
take the selfie.

Speaker 2 (01:03:20):
Come back down.

Speaker 3 (01:03:22):
Which of those choices are you making? I asked that
question in my class because I said, for a lot
of the students, part of making projects that have richness
to them again is this effort piece. It's a demonstration
of agency of choice making. And I asked them, you
know who's taking the helicopter. I go, look, if you

(01:03:42):
take the helicopter, probably your parents are going to be
a little bit happier since your safety is more guaranteed.
You still get the selfie, so you get to be
up there and experience what it was like. You can
still tell your friends how cool you are because you
were at You know, there's some benefits to taking the helicopter,
and I think almost nobody raised their hands right. They
all would rather climb because you put in your best effort.

(01:04:06):
Or part of why we glorify specific amazing artworks or
why we go to the ballet is because if everything
was easy, we would all do it. Part of what
gives the meaning to certain things is that it acquired
effort and specialization, curiosity, physical exertion in many cases to

(01:04:28):
get there, and that's important to us. I think for me,
if I were to snap my fingers and say am
I taking the helicopter? Am I climbing Mount Everest? Well,
it's a hard choice. But do your best work or
don't do it at all.

Speaker 1 (01:04:42):
I totally agree. You know, I've got a bunch of
I've seen this happening a lot lately where someone says, hey,
I've written a book. It's half written by me, half
written by chatchpt and they put that on the cover
of the book and they think it's sort of cool
and futuristic that they're doing that. And I immediately think,
I don't want to read that book just only it
is probably unfair because it might be a perfectly good book,

(01:05:03):
but it's just it feels like the requisite effort was
not put in.

Speaker 3 (01:05:06):
I think it's also when we talked about intelligence and
consciousness sort of. I think the intelligence that I have
now at this age is different than the one that
I had at twenty four, and it's different than the
one I had at sixteen and it's because the experiences
that you have and the effort. And I remember Alison Gopnik,
who's an amazing professor at UC Berkeley, said this thing

(01:05:27):
when she came to visit at Google, where she told
us how grad school is annealing. It's a process of
annealing your brain. And anneeling, which I had to look up,
is when you heat up a metal a little bit
so that you can bend it, and then you wait
and you let it cool off, and then you heat
it up a little bit more and you bend it
even farther. And it's so that every single one of

(01:05:49):
those marches up the staircase demonstrates to you that you
can keep going higher and higher because you landed those
initial steps.

Speaker 2 (01:05:56):
And my PhD.

Speaker 3 (01:05:59):
I actually didn't have an undergraduate degree in engineering, and
so it was my first foray into a lot of
that material that I had been exposed to more in
coating boot camps or I even took some remedial classes
online before I started grad school. And for me, it
really felt that way, this process of a kneeling, of
heating up a little bit, bending, heating up a little

(01:06:19):
bit more, bending more, and that by the time I
was finished with that degree, I was like, wow, I
am not the same person with the same set of intelligence,
and I earned that capacity. And I love that you
have an appreciation for that, because it's not that we
should be doing that because we like pain or we
like suffering, but it's because it means something about the

(01:06:42):
new intelligence that's been acquired. You can say something different
having earned those small bends along the way.

Speaker 1 (01:06:50):
Okay, So in wrapping up, imagine the future, like, really
picture ourselves twenty years now. First of all, what are
robots looking like at that point? Are they humanoid? Are
they others shapes like R two D two.

Speaker 3 (01:07:02):
We're going to have many different shapes and instantiations of
robots in the future. We will have humanoids for sure.
Some of them will be walking, some of them will
be on wheels. We will have quadruped robots. These are
robots that look a little bit abstractly like dogs, so
they have a centered torso and then four legs. We'll
have robots that have more or fewer legs than that.

(01:07:25):
We might have tripod robots. I've seen an amazing robot
that moves around a bit on a globe, so it's
almost like it's floating on top of a globe that's
wheeling around. We'll have disc and puck like ones, We'll
have ones that fly. We're going to see so many
different form factors of these machines in the future. We'll
even see robots that are soft, that are made out

(01:07:47):
of softer materials like fabric or plastic bags, and they're
inflated with air and they move around because the pressure
in different parts of the robot changes. And so I
think it's really the sky's the limit of what some
of these new form factors are going to look like.
And the key I think in many cases is that

(01:08:10):
they not only are useful, but they got to bring
some beauty and onto the world. We have unlimited imagination.
We've built all kinds of crazy stuff as people. Let's
make robots that look interesting but also do things that
are a little unexpected. I love the idea of what

(01:08:33):
if I had a robot that could go out in
my garden and pick me five or six fresh flowers
every morning and put them in a little cup on
my shelf. Sounds like a great robot to be I
would love that robot. What if I had a robot
that could go around my house, not only rearranging and reorganizing,

(01:08:53):
but also surprise me a little bit. Maybe it puts
a book in a new place and says, oh, Katie,
you talked about this author a few times. I want
to put this next to your bag by the door
in case you decide to take it with you. What
if I had a robot that could tell me compliments
and jokes and they could be use for fun facts.
I mean, there's it's an unlimited number.

Speaker 2 (01:09:14):
Of things that we could use these tools for in
the future.

Speaker 3 (01:09:18):
And I think the embodiment, the whole robot becomes an
interface because it's three dimensional. And the big difference between
a robot and a computer is that a robot moves.
Robot moves and has a body. And so what are
all the new ways that we can create relationships and

(01:09:39):
derive utility and also have fun in the world. Because
we don't have an I mean, unless we're all cryogenically
frozen to come back to that, we don't have unlimited time.
And I want the tools that we use to inspire
us to be useful and also to make us feel

(01:10:01):
more optimistic.

Speaker 1 (01:10:07):
That's Katie Kwan. One of the things that I keep
thinking about after this conversation is how much of human
life depends on movement that we almost never consciously notice.
Somebody leans forward slightly during a conversation, or your friend
matches his walking pace to yours, or a stranger steps
aside on a crowded sidewalk, or a parent reaches out

(01:10:31):
towards a child. There are a million tiny choreographies that
collectively form this invisible social fabric of our civilization, and
your brain tracks all of it automatically. Movement is almost
never just a mechanical thing to us. We experience it
as intention. And this makes sense, of course, through an

(01:10:51):
evolutionary lens, because movement was one of the earliest forms
of communication. You needed to infer whether something approaching you
is dangerous or a cooperative or aggressive or playful or whatever.
And this all becomes massively relevant now that AI is
leaving the screen. The challenge is that machines are plugging

(01:11:12):
into an ancient perceptual system that evolved long before machines existed.
And this is why choreographers are going to be involved
in this next step of invention, because choreography is the
study of how movement creates meaning. So what I love
about Katie's perspective is that she understands robotics not just

(01:11:34):
as an engineering problem, but as a humanistic problem. The
future of robotics is going to depend partly on batteries
and actuators of machine learning, but also partly on empathy
and aesthetics and social fluency. It's going to depend on
whether machines can come to understand the unwritten choreography that

(01:11:56):
humans have spent millions of years developing, which with each other.
Go to eagleman dot com slash podcast for more information
and to find further reading. Join the weekly discussions on
my substack, and check out and subscribe to Inner Cosmos
on YouTube for videos of each episode and to leave

(01:12:16):
comments until next time. I'm David Eagleman, and this is
Inner Cosmos.