WEBVTT

00:00:06.361 --> 00:00:08.084
Marita Cheng: When I was growing up,

00:00:08.085 --> 00:00:11.595
I had a family friend
who became blind in his 20s.

00:00:12.438 --> 00:00:15.602
When we went out as a family,
he would say to me,

00:00:15.603 --> 00:00:19.881
"Rita hold my hand, hold my arm,
and tell me what you see."

00:00:19.882 --> 00:00:23.682
So I'd say, "There's 
some flowers here to the left.

00:00:23.683 --> 00:00:25.681
There's a gate here to the right.

00:00:25.682 --> 00:00:28.212
There's a mountain over in the distance."

00:00:28.213 --> 00:00:31.493
And he would say,
"What color are those flowers?

00:00:32.403 --> 00:00:34.753
Can I use my hand,
reach out, and touch them?

00:00:34.754 --> 00:00:36.322
Could you lead my hand to that?"

00:00:36.323 --> 00:00:39.622
I'd say, "Oh, they're pink, they're blue."

00:00:39.623 --> 00:00:42.833
And he'd say, "Tell me more,
tell me more about what you can see.

00:00:42.834 --> 00:00:44.062
Share it with me."

00:00:44.063 --> 00:00:48.592
About eight months ago,
Alberto and I decided to create an app

00:00:48.593 --> 00:00:52.663
to enable blind people
to recognize their surroundings.

00:00:52.667 --> 00:00:55.959
We used something called
convolutional neural networks,

00:00:55.972 --> 00:01:00.283
which is a computer system
that's been trained on millions of images.

00:01:00.284 --> 00:01:04.242
It learns the features of a dog.
It learns what a flower looks like.

00:01:04.244 --> 00:01:07.964
It learns a fork, a knife,
everyday objects.

00:01:09.667 --> 00:01:14.418
Using this system,
we created something called "Aipoly"

00:01:14.454 --> 00:01:17.734
that recognizes
over 1,000 everyday objects.

00:01:17.735 --> 00:01:21.598
So, a blind person just needs
to walk around with their phone,

00:01:21.599 --> 00:01:26.924
and put it over various objects,
and it will say the name of the object.

00:01:27.444 --> 00:01:32.773
Using voice over, the phone can relay
the word on the screen

00:01:32.774 --> 00:01:34.113
to that blind person,

00:01:34.114 --> 00:01:36.316
so they know exactly
what's in front of them.

00:01:36.317 --> 00:01:38.193
(Applause)

00:01:46.103 --> 00:01:49.103
Since we released
our application in January,

00:01:49.104 --> 00:01:52.634
we've had over a 100,000 downloads
around the world.

00:01:53.536 --> 00:01:58.444
The app has been so popular
we've translated it into seven languages.

00:01:59.167 --> 00:02:02.376
Alberto Rizzoli: After experiencing
the technology the first time,

00:02:02.377 --> 00:02:04.764
our users kept asking us for more.

00:02:05.555 --> 00:02:09.384
We asked them

00:02:09.386 --> 00:02:14.225
to think of our technology
as a superpower for a moment,

00:02:14.226 --> 00:02:19.425
something they could effortlessly
evoke at any time,

00:02:19.426 --> 00:02:22.346
and gain understanding
of what was in front of them.

00:02:23.080 --> 00:02:25.705
And surprisingly,

00:02:25.706 --> 00:02:29.545
nobody really wants
X-ray vision or telescopic goggles,

00:02:29.546 --> 00:02:33.166
but what everyone wants
is more information.

00:02:33.167 --> 00:02:34.495
It's not surprising

00:02:34.496 --> 00:02:40.134
because 60% of the information
that we perceive comes through sight.

00:02:40.135 --> 00:02:41.846
It is the main tool that we use

00:02:41.847 --> 00:02:45.994
to understand our surroundings
and often, to make decisions.

00:02:45.995 --> 00:02:50.356
If you're blind you must rely
on other senses like touch or hearing,

00:02:50.357 --> 00:02:53.465
and you miss out
on the lightning-fast identification

00:02:53.466 --> 00:02:58.717
that our brain and eyes do every second
of every day, if you're a sighted person.

00:02:59.077 --> 00:03:02.154
We went

00:03:02.155 --> 00:03:05.024
to the Santa Clara Valley Blind Center,

00:03:05.025 --> 00:03:07.606
and we tried to build this superpower.

00:03:07.607 --> 00:03:10.577
We tried to see what kind
of information people wanted,

00:03:10.578 --> 00:03:13.565
and it's simple things
like whether a dish is clean or not,

00:03:13.566 --> 00:03:17.655
whether you can cross the street,
what product am I looking at?

00:03:17.656 --> 00:03:19.396
Things that can lead to a decision,

00:03:19.397 --> 00:03:23.797
from a simple gaze, to an understanding
of the situation in front of you.

00:03:24.557 --> 00:03:28.887
We asked what form factor
people preferred, and we built it.

00:03:29.707 --> 00:03:31.485
So we put together

00:03:31.486 --> 00:03:37.147
some jawbone conductive headphones,
a pair of sunglasses, and a tiny camera,

00:03:38.177 --> 00:03:42.046
and we asked our friends
to think of a common situation

00:03:42.047 --> 00:03:45.056
in which they had to make
many small decisions,

00:03:45.057 --> 00:03:47.857
and we told them
we will be giving them the prototype,

00:03:47.858 --> 00:03:50.166
and taking them
in the middle of that situation.

00:03:50.167 --> 00:03:52.003
Let's see how it went. (Video starts)

00:03:52.004 --> 00:03:53.351
[We asked blind individuals

00:03:53.352 --> 00:03:56.089
[what are the hardest things to do
when visually impaired]

00:03:56.090 --> 00:03:58.656
I mean it takes me forever
to go grocery-shopping.

00:03:58.657 --> 00:04:02.316
Even with someone helping me
that has known me for years.

00:04:02.317 --> 00:04:05.827
I'll say, "What's in that cabinet?"

00:04:07.097 --> 00:04:10.397
Or develop a system,
right to left, top to bottom.

00:04:10.398 --> 00:04:13.058
[So we took them grocery shopping
with our technology]

00:04:13.358 --> 00:04:14.728
Computer: Oranges.

00:04:16.048 --> 00:04:18.618
Man: This is great. I'm really liking it.

00:04:19.567 --> 00:04:23.237
Apples, grapes, carrots.

00:04:23.238 --> 00:04:25.626
I'm looking, I'm looking.

00:04:25.627 --> 00:04:27.576
Computer: Lilies.
Man: Lilies.

00:04:27.577 --> 00:04:31.317
Computer: Bouquet.
Man: A bouquet, ahh!

00:04:31.318 --> 00:04:33.211
Computer: Roses, flowers.

00:04:33.212 --> 00:04:35.177
Man: Can I take these home?

00:04:35.178 --> 00:04:36.778
This is great.

00:04:36.792 --> 00:04:38.710
Computer: Roses.
Man: Roses.

00:04:38.711 --> 00:04:41.367
Computer: Bouquet, tulips.
Man: Tulips.

00:04:41.368 --> 00:04:44.019
Computer: Pineapple.
Woman: It's a pineapple.

00:04:45.089 --> 00:04:46.830
Computer: Mango.
Woman: Mango.

00:04:46.831 --> 00:04:48.389
Computer: M&Ms.
Woman: M&Ms.

00:04:48.375 --> 00:04:50.750
Computer: Tic Tac.
Woman: It just said, "Tic Tac."

00:04:50.759 --> 00:04:52.377
Computer: Tic Tac.
Woman: Tic Tac.

00:04:52.378 --> 00:04:54.309
Computer: Paper note, calendar.

00:04:54.310 --> 00:04:55.719
Woman: Calendar, you got it.

00:04:56.329 --> 00:04:59.318
Wow, I didn't know what that was at all.

00:04:59.319 --> 00:05:01.288
Computer: Pretzels.
Woman: Pretzels.

00:05:01.289 --> 00:05:03.609
Computer: Pretzels.
Woman: It said, "Pretzels."

00:05:03.610 --> 00:05:05.409
Computer: Lipton tea.
Woman 2: Lipton?

00:05:05.410 --> 00:05:06.589
Tea?

00:05:06.590 --> 00:05:08.548
Computer: Lipton teabags.

00:05:08.549 --> 00:05:12.209
Woman 2: It's like I'm seeing it,
but I'm not, it's seeing it for me.

00:05:12.210 --> 00:05:14.971
Computer: Coffee mate.
Woman 2: Mate; coffee mate.

00:05:16.281 --> 00:05:18.739
It didn't say "coffee,"
but it kept saying "mate."

00:05:18.740 --> 00:05:21.369
Computer: Mate, mate.

00:05:22.190 --> 00:05:24.731
Man 2: I put on the glasses,

00:05:24.732 --> 00:05:27.869
and right away, it told me
there was an apple, there were oranges,

00:05:27.870 --> 00:05:32.561
and there was this, and there was that,
and it's like, "This is great!"

00:05:33.420 --> 00:05:34.799
Instant love.

00:05:34.800 --> 00:05:35.966
(Video ends)

00:05:35.967 --> 00:05:37.301
(Applause)

00:05:45.251 --> 00:05:48.759
AR: That little pair of glasses
connected to their phones

00:05:48.760 --> 00:05:52.589
could identify four
to 5,000 objects in real time.

00:05:52.590 --> 00:05:56.080
That's about the capacity
of a five-year-old child.

00:05:57.440 --> 00:06:03.519
A simple accessory
can now expand a person's perception

00:06:03.520 --> 00:06:05.900
to thousands of new possibilities.

00:06:06.630 --> 00:06:11.456
This is the power of marrying
artificial and human intelligence,

00:06:12.316 --> 00:06:14.666
and the potential
is still vastly untapped.

00:06:14.667 --> 00:06:18.209
This isn't going to be a revolution
just because GPUs are getting faster,

00:06:18.210 --> 00:06:20.550
or the research is getting more open,

00:06:20.551 --> 00:06:24.250
but because the barriers of entry
to impacting millions of lives

00:06:24.251 --> 00:06:27.472
for artificial intelligence
are getting lower and lower.

00:06:28.142 --> 00:06:31.110
The Paralympic games
are starting in a few weeks,

00:06:31.111 --> 00:06:36.731
an event where sheer force of will,
training, and technology

00:06:36.732 --> 00:06:40.122
turn people with a disability
into super humans.

00:06:40.532 --> 00:06:46.961
and so, too, will all ability to think,
perceive, make decisions, and learn

00:06:46.962 --> 00:06:48.951
increase exponentially.

00:06:48.952 --> 00:06:52.472
You will be building
the tools to make this happen.

00:06:54.032 --> 00:06:57.291
So tomorrow, with your morning coffee,

00:06:57.292 --> 00:07:00.652
take 40 minutes
and try out a tutorial on deep learning.

00:07:01.212 --> 00:07:03.232
Build yourself a small superpower.

00:07:03.233 --> 00:07:05.902
All it takes is your laptop,
and a bunch of data,

00:07:05.903 --> 00:07:07.763
like your holiday pictures.

00:07:08.063 --> 00:07:12.593
Superpower engineer -
that's a great dream job.

00:07:14.073 --> 00:07:18.283
The good news is that the world
needs many, many more of them

00:07:18.632 --> 00:07:22.303
so we can't wait to see
what you will be building next.

00:07:22.603 --> 00:07:23.721
Thank you.

00:07:23.722 --> 00:07:25.114
(Applause)