Why I draw with robots

What happens when humans and robots make art together? In this awe-inspiring talk, artist Sougwen Chung shows how she "taught" her artistic style to a machine -- and shares the results of their collaboration after making an unexpected discovery: robots make mistakes, too. "Part of the beauty of human and machine systems is their inherent, shared fallibility," she says.

Many of us here use technology
in our day-to-day.

And some of us rely
on technology to do our jobs.

For a while, I thought of machines
and the technologies that drive them

as perfect tools that could make my work
more efficient and more productive.

But with the rise of automation
across so many different industries,

it led me to wonder:

If machines are starting
to be able to do the work

traditionally done by humans,

what will become of the human hand?

How does our desire for perfection,
precision and automation

affect our ability to be creative?

In my work as an artist and researcher,
I explore AI and robotics

to develop new processes
for human creativity.

For the past few years,

I've made work alongside machines,
data and emerging technologies.

It's part of a lifelong fascination

about the dynamics
of individuals and systems

and all the messiness that that entails.

It's how I'm exploring questions about
where AI ends and we begin

and where I'm developing processes

that investigate potential
sensory mixes of the future.

I think it's where philosophy
and technology intersect.

Doing this work
has taught me a few things.

It's taught me how embracing imperfection

can actually teach us
something about ourselves.

It's taught me that exploring art

can actually help shape
the technology that shapes us.

And it's taught me
that combining AI and robotics

with traditional forms of creativity --
visual arts in my case --

can help us think a little bit more deeply

about what is human
and what is the machine.

And it's led me to the realization

that collaboration is the key
to creating the space for both

as we move forward.

It all started with a simple
experiment with machines,

called "Drawing Operations
Unit: Generation 1."

I call the machine "D.O.U.G." for short.

Before I built D.O.U.G,

I didn't know anything
about building robots.

I took some open-source
robotic arm designs,

I hacked together a system
where the robot would match my gestures

and follow [them] in real time.

The premise was simple:

I would lead, and it would follow.

I would draw a line,
and it would mimic my line.

So back in 2015, there we were,
drawing for the first time,

in front of a small audience
in New York City.

The process was pretty sparse --

no lights, no sounds,
nothing to hide behind.

Just my palms sweating
and the robot's new servos heating up.

(Laughs) Clearly, we were
not built for this.

But something interesting happened,
something I didn't anticipate.

See, D.O.U.G., in its primitive form,
wasn't tracking my line perfectly.

While in the simulation
that happened onscreen

it was pixel-perfect,

in physical reality,
it was a different story.

It would slip and slide
and punctuate and falter,

and I would be forced to respond.

There was nothing pristine about it.

And yet, somehow, the mistakes
made the work more interesting.

The machine was interpreting
my line but not perfectly.

And I was forced to respond.

We were adapting
to each other in real time.

And seeing this taught me a few things.

It showed me that our mistakes
actually made the work more interesting.

And I realized that, you know,
through the imperfection of the machine,

our imperfections became
what was beautiful about the interaction.

And I was excited,
because it led me to the realization

that maybe part of the beauty
of human and machine systems

is their shared inherent fallibility.

For the second generation of D.O.U.G.,

I knew I wanted to explore this idea.

But instead of an accident produced
by pushing a robotic arm to its limits,

I wanted to design a system
that would respond to my drawings

in ways that I didn't expect.

So, I used a visual algorithm
to extract visual information

from decades of my digital
and analog drawings.

I trained a neural net on these drawings

in order to generate
recurring patterns in the work

that were then fed through custom software
back into the machine.

I painstakingly collected
as many of my drawings as I could find --

finished works, unfinished experiments
and random sketches --

and tagged them for the AI system.

And since I'm an artist,
I've been making work for over 20 years.

Collecting that many drawings took months,

it was a whole thing.

And here's the thing
about training AI systems:

it's actually a lot of hard work.

A lot of work goes on behind the scenes.

But in doing the work,
I realized a little bit more

about how the architecture
of an AI is constructed.

And I realized it's not just made
of models and classifiers

for the neural network.

But it's a fundamentally
malleable and shapable system,

one in which the human hand
is always present.

It's far from the omnipotent AI
we've been told to believe in.

So I collected these drawings
for the neural net.

And we realized something
that wasn't previously possible.

My robot D.O.U.G. became
a real-time interactive reflection

of the work I'd done
through the course of my life.

The data was personal,
but the results were powerful.

And I got really excited,

because I started thinking maybe
machines don't need to be just tools,

but they can function
as nonhuman collaborators.

And even more than that,

I thought maybe
the future of human creativity

isn't in what it makes

but how it comes together
to explore new ways of making.

So if D.O.U.G._1 was the muscle,

and D.O.U.G._2 was the brain,

then I like to think
of D.O.U.G._3 as the family.

I knew I wanted to explore this idea
of human-nonhuman collaboration at scale.

So over the past few months,

I worked with my team
to develop 20 custom robots

that could work with me as a collective.

They would work as a group,

and together, we would collaborate
with all of New York City.

I was really inspired
by Stanford researcher Fei-Fei Li,

who said, "if we want to teach
machines how to think,

we need to first teach them how to see."

It made me think of the past decade
of my life in New York,

and how I'd been all watched over by these
surveillance cameras around the city.

And I thought it would be
really interesting

if I could use them
to teach my robots to see.

So with this project,

I thought about the gaze of the machine,

and I began to think about vision
as multidimensional,

as views from somewhere.

We collected video

from publicly available
camera feeds on the internet

of people walking on the sidewalks,

cars and taxis on the road,

all kinds of urban movement.

We trained a vision algorithm
on those feeds

based on a technique
called "optical flow,"

to analyze the collective density,

direction, dwell and velocity states
of urban movement.

Our system extracted those states
from the feeds as positional data

and became pads for my
robotic units to draw on.

Instead of a collaboration of one-to-one,

we made a collaboration of many-to-many.

By combining the vision of human
and machine in the city,

we reimagined what
a landscape painting could be.

Throughout all of my
experiments with D.O.U.G.,

no two performances
have ever been the same.

And through collaboration,

we create something that neither of us
could have done alone:

we explore the boundaries
of our creativity,

human and nonhuman working in parallel.

I think this is just the beginning.

This year, I've launched Scilicet,

my new lab exploring human
and interhuman collaboration.

We're really interested
in the feedback loop

between individual, artificial
and ecological systems.

We're connecting human and machine output

to biometrics and other kinds
of environmental data.

We're inviting anyone who's interested
in the future of work, systems

and interhuman collaboration

to explore with us.

We know it's not just technologists
that have to do this work

and that we all have a role to play.

We believe that by teaching machines

how to do the work
traditionally done by humans,

we can explore and evolve our criteria

of what's made possible by the human hand.

And part of that journey
is embracing the imperfections

and recognizing the fallibility
of both human and machine,

in order to expand the potential of both.

Today, I'm still in pursuit
of finding the beauty

in human and nonhuman creativity.

In the future, I have no idea
what that will look like,

but I'm pretty curious to find out.

Thank you.

(Applause)