IRENE ALVARADO: Hi, everyone.

My name is Irene, and I work at a place

called the Creative Lab, a team inside of Google.

And some of us are interested in creating

what we call experiments to showcase

and make more accessible some of the machine learning research

that's coming out of Google.

And a lot of our work goes to this site

called the Experiments with Google site.

Now, before I talk about some of the products on the site,

let me just say that we're really

inspired by pioneering AI researcher Seymour Papert, who

wrote a lot about learning theories

in humans and essentially kind of how

to make learning not suck.

So this is one of his great quotes.

"Every maker of video games knows something

that the makers of curriculum don't seem to understand.

You'll never see a video being advertised as being easy.

Kids who do not like school will tell you

it's not because it's too hard.

It's because it's boring."

So if there are some parents in the room,

you might be agreeing with this statement.

So I'll show you some projects that

were inspired by this thinking that learning should

be engaging, made in collaboration

with the TensorFlow.js team and many other research

teams at Google.

So this is the first one.

It's called Teachable Machine.

And essentially it's a KNN classifier that

runs entirely in the browser.

And it lets you train three classes of images

that trigger different kinds of inputs, like GIFs and sound.

So I don't have time to demo it, but I'll

show you what happens after you train a model with the tool.

So can I get the video?

[VIDEO PLAYBACK]

[SPOOKY ORGAN MUSIC]

[BIRDS CHIRPING]

[SPOOKY ORGAN MUSIC]

[BIRDS CHIRPING]

[SPOOKY ORGAN MUSIC]

[BIRDS CHIRPING]

[SPOOKY ORGAN MUSIC]

[END PLAYBACK]

See it choosing between two classes.

Yeah, so, hopefully, you get how it works.

Alex Chen, the creator, he trained a class

to recognize the bird origami and another class

to recognize the spooky person origami.

OK, back to the slides.

Thank you.

So we released the experiment online.

All the inference and training is happening in the browser.

And we also released the open source--

we open sourced the boilerplate code that

went along with the experiment.

And what happened next was that we were really kind of taken

aback by all the stories of teachers around the world,

like this one, who started using Teachable Machine to introduce

ML into the classroom.

Here's another example of kids learning

about smart cities and kind of training

the computer to recognize handmade stop signs.

This was really amazing.

And finally, we heard from another renowned and pioneering

researcher, Hal Abelson, who teaches at MIT, that he

had been using Teachable Machine to introduce ML

to policymakers.

And for a lot of them, it was the first time

that they had ever trained a model.

So needless to say, we're really happy that although simple

in nature, Teachable Machine ended up

being a really good tool for educators and people

that were new to machine learning.

So here's another example.

This one's called Move Mirror.

And the concept is really simple.

You strike a pose in front of a webcam,

and you get an image with a matching pose.

And again, this is all happening on the web.

So here's another example of, actually, people using it

in the form of an installation.

People do really funny moves.

And again, this is happening on a phone,

but on the phone's browser.

And so the story for this one was

that in order to make the experiment really accessible,

we had to take the tech to the web,

so that we wouldn't require users

to have a complicated tech setup or to use IR cameras or depth

sensors, which can be expensive.

So PoseNet was born.

To our knowledge, it's the first pose estimation

model for the web.

And it's open source.

It runs locally in your browser.

And it uses good ol' RGB webcams.

So again, we were really taken aback

by all the creative projects that we saw popping up online.

Just to give you a sense, the one on the left

is a musical interface.

The one in the middle is a ping pong game

that you can use with your head.

I really want to play that one.

And the one on the right is a kind of performative motion

capture animation.

But we also started hearing from people

in the accessibility world that they were using PoseNet.

So we decided to partner with a bunch of groups

that work at the intersection of disability and technology,

like the NYU Ability Project, and musicians, artists, makers

in the accessibility world.

And out of that collaboration came a set of creative tools

that we're calling Creatability.

And a lot of them use PoseNet for users

who have motor impairments to be able to interface

with a computer with their whole bodies

instead of through a keyboard and a mouse.

So again, I don't have time to demo these.

But just give you a sense, the one on the bottom left

is a visualization tool made by a musician named

Jay Zimmerman, who's deaf, and the one on the top right

is an accessible musical instrument

made by a group called Open Up Music.

And we just took their designs and kind of moved it

to the web.

So again, all of the components that

made this project are accessible and they've been open sourced.

So just a step back for a second,

if we were to think about what made these projects successful

or at least useful for other people,

we can see that they were all interactive and accessible

through the browser.

So it really lowered the barrier of entry for a lot of people.

They all had an open-source component,

so that people could kind of look under the hood,

see what's happening, modify them, play with them.

And then, finally, they're all free,

because the processing is happening locally

in the browser with TensorFlow.js.

And that gave us privacy, so that we

didn't have to send images of people's bodies

and faces to any servers.

So again, all the projects that I went through kind of quickly,

they're on the Experiments.withGoogle.com

site.

And even though these were created in-house,

we actually feature work by more than 1,700 developers

from around the world.

So if any of this resonates with you,

this is really an open invitation for you

to submit your work.

And I hope to have showed that you never

know who you might inspire or who

might take your work and kind of innovate on top of it

and use in really creative ways.

Thank you.

[APPLAUSE]

[MUSIC PLAYING]