WEBVTT

00:00:00.875 --> 00:00:04.643
How many decisions
have been made about you today,

00:00:04.667 --> 00:00:07.268
or this week or this year,

00:00:07.292 --> 00:00:09.250
by artificial intelligence?

00:00:10.958 --> 00:00:12.643
I build AI for a living,

00:00:12.667 --> 00:00:15.684
so full disclosure -- I'm kind of a nerd.

00:00:15.708 --> 00:00:18.101
And because I'm kind of a nerd,

00:00:18.125 --> 00:00:20.476
wherever some new news story comes out

00:00:20.500 --> 00:00:23.934
about artificial intelligence
stealing all our jobs,

00:00:23.958 --> 00:00:28.143
or robots getting citizenship
of an actual country,

00:00:28.167 --> 00:00:31.309
I'm the person my friends
and followers message

00:00:31.333 --> 00:00:32.875
freaking out about the future.

00:00:33.833 --> 00:00:35.934
We see this everywhere.

00:00:35.958 --> 00:00:40.851
This media panic that
our robot overlords are taking over.

00:00:40.875 --> 00:00:42.792
We could blame Hollywood for that.

00:00:44.125 --> 00:00:48.250
But in reality, that's not the problem
we should be focusing on.

00:00:49.250 --> 00:00:52.893
There is a more pressing danger,
a bigger risk with AI,

00:00:52.917 --> 00:00:54.500
that we need to fix first.

00:00:55.417 --> 00:00:57.726
So we are back to this question:

00:00:57.750 --> 00:01:02.458
How many decisions
have been made about you today by AI?

00:01:03.792 --> 00:01:05.768
And how many of these

00:01:05.792 --> 00:01:10.292
were based on your gender,
your race or your background?

00:01:12.500 --> 00:01:15.268
Algorithms are being used all the time

00:01:15.292 --> 00:01:19.125
to make decisions about who we are
and what we want.

00:01:20.208 --> 00:01:23.851
Some of the women in this room
will know what I'm talking about

00:01:23.875 --> 00:01:27.643
if you've been made to sit through
those pregnancy test adverts on YouTube

00:01:27.667 --> 00:01:29.726
like 1,000 times.

00:01:29.750 --> 00:01:32.601
Or you've scrolled past adverts
of fertility clinics

00:01:32.625 --> 00:01:34.667
on your Facebook feed.

00:01:35.625 --> 00:01:38.018
Or in my case, Indian marriage bureaus.

00:01:38.042 --> 00:01:39.309
(Laughter)

00:01:39.333 --> 00:01:42.309
But AI isn't just being used
to make decisions

00:01:42.333 --> 00:01:44.934
about what products we want to buy

00:01:44.958 --> 00:01:47.458
or which show we want to binge watch next.

00:01:49.042 --> 00:01:54.226
I wonder how you'd feel about someone
who thought things like this:

00:01:54.250 --> 00:01:56.184
a black or Latino person

00:01:56.208 --> 00:02:00.333
is less likely than a white person
to pay off their loan on time.

00:02:01.542 --> 00:02:04.351
A person called John
makes a better programmer

00:02:04.375 --> 00:02:06.042
than a person called Mary.

00:02:07.250 --> 00:02:12.333
A black man is more likely to be
a repeat offender than a white man.

00:02:14.958 --> 00:02:16.226
You're probably thinking,

00:02:16.250 --> 00:02:20.000
"Wow, that sounds like a pretty sexist,
racist person," right?

00:02:21.000 --> 00:02:25.851
These are some real decisions
that AI has made very recently,

00:02:25.875 --> 00:02:28.809
based on the biases
it has learned from us,

00:02:28.833 --> 00:02:30.083
from the humans.

00:02:31.750 --> 00:02:36.559
AI is being used to help decide
whether or not you get that job interview;

00:02:36.583 --> 00:02:38.976
how much you pay for your car insurance;

00:02:39.000 --> 00:02:40.893
how good your credit score is;

00:02:40.917 --> 00:02:44.042
and even what rating you get
in your annual performance review.

00:02:45.083 --> 00:02:48.226
But these decisions
are all being filtered through

00:02:48.250 --> 00:02:54.125
its assumptions about our identity,
our race, our gender, our age.

00:02:56.250 --> 00:02:58.518
How is that happening?

00:02:58.542 --> 00:03:02.059
Now, imagine an AI is helping
a hiring manager

00:03:02.083 --> 00:03:04.934
find the next tech leader in the company.

00:03:04.958 --> 00:03:08.059
So far, the manager
has been hiring mostly men.

00:03:08.083 --> 00:03:12.833
So the AI learns men are more likely
to be programmers than women.

00:03:13.542 --> 00:03:16.434
And it's a very short leap from there to:

00:03:16.458 --> 00:03:18.500
men make better programmers than women.

00:03:19.417 --> 00:03:23.143
We have reinforced
our own bias into the AI.

00:03:23.167 --> 00:03:26.792
And now, it's screening out
female candidates.

00:03:28.917 --> 00:03:31.934
Hang on, if a human
hiring manager did that,

00:03:31.958 --> 00:03:34.309
we'd be outraged, we wouldn't allow it.

00:03:34.333 --> 00:03:37.809
This kind of gender
discrimination is not OK.

00:03:37.833 --> 00:03:42.351
And yet somehow,
AI has become above the law,

00:03:42.375 --> 00:03:44.458
because a machine made the decision.

00:03:45.833 --> 00:03:47.351
That's not it.

00:03:47.375 --> 00:03:52.250
We are also reinforcing our bias
in how we interact with AI.

00:03:52.917 --> 00:03:58.893
How often do you use a voice assistant
like Siri, Alexa or even Cortana?

00:03:58.917 --> 00:04:01.476
They all have two things in common:

00:04:01.500 --> 00:04:04.601
one, they can never get my name right,

00:04:04.625 --> 00:04:07.292
and second, they are all female.

00:04:08.417 --> 00:04:11.184
They are designed to be
our obedient servants,

00:04:11.208 --> 00:04:14.458
turning your lights on and off,
ordering your shopping.

00:04:15.125 --> 00:04:18.434
You get male AIs too,
but they tend to be more high-powered,

00:04:18.458 --> 00:04:21.517
like IBM Watson,
making business decisions,

00:04:21.541 --> 00:04:25.333
Salesforce Einstein
or ROSS, the robot lawyer.

00:04:26.208 --> 00:04:30.268
So poor robots, even they suffer
from sexism in the workplace.

00:04:30.292 --> 00:04:31.417
(Laughter)

00:04:32.542 --> 00:04:35.393
Think about how these two things combine

00:04:35.417 --> 00:04:40.726
and affect a kid growing up
in today's world around AI.

00:04:40.750 --> 00:04:43.684
So they're doing some research
for a school project

00:04:43.708 --> 00:04:46.726
and they Google images of CEO.

00:04:46.750 --> 00:04:49.643
The algorithm shows them
results of mostly men.

00:04:49.667 --> 00:04:52.226
And now, they Google personal assistant.

00:04:52.250 --> 00:04:55.684
As you can guess,
it shows them mostly females.

00:04:55.708 --> 00:04:59.309
And then they want to put on some music,
and maybe order some food,

00:04:59.333 --> 00:05:05.917
and now, they are barking orders
at an obedient female voice assistant.

00:05:07.542 --> 00:05:12.851
Some of our brightest minds
are creating this technology today.

00:05:12.875 --> 00:05:17.059
Technology that they could have created
in any way they wanted.

00:05:17.083 --> 00:05:22.768
And yet, they have chosen to create it
in the style of 1950s "Mad Man" secretary.

00:05:22.792 --> 00:05:24.292
Yay!

00:05:24.958 --> 00:05:26.268
But OK, don't worry,

00:05:26.292 --> 00:05:28.351
this is not going to end
with me telling you

00:05:28.375 --> 00:05:31.852
that we are all heading towards
sexist, racist machines running the world.

00:05:32.792 --> 00:05:38.583
The good news about AI
is that it is entirely within our control.

00:05:39.333 --> 00:05:43.333
We get to teach the right values,
the right ethics to AI.

00:05:44.167 --> 00:05:46.351
So there are three things we can do.

00:05:46.375 --> 00:05:49.726
One, we can be aware of our own biases

00:05:49.750 --> 00:05:52.476
and the bias in machines around us.

00:05:52.500 --> 00:05:57.018
Two, we can make sure that diverse teams
are building this technology.

00:05:57.042 --> 00:06:01.958
And three, we have to give it
diverse experiences to learn from.

00:06:02.875 --> 00:06:06.184
I can talk about the first two
from personal experience.

00:06:06.208 --> 00:06:07.643
When you work in technology

00:06:07.667 --> 00:06:11.059
and you don't look like
a Mark Zuckerberg or Elon Musk,

00:06:11.083 --> 00:06:14.833
your life is a little bit difficult,
your ability gets questioned.

00:06:15.875 --> 00:06:17.268
Here's just one example.

00:06:17.292 --> 00:06:21.018
Like most developers,
I often join online tech forums

00:06:21.042 --> 00:06:24.268
and share my knowledge to help others.

00:06:24.292 --> 00:06:25.601
And I've found,

00:06:25.625 --> 00:06:29.601
when I log on as myself,
with my own photo, my own name,

00:06:29.625 --> 00:06:34.226
I tend to get questions
or comments like this:

00:06:34.250 --> 00:06:37.250
"What makes you think
you're qualified to talk about AI?"

00:06:38.458 --> 00:06:41.934
"What makes you think
you know about machine learning?"

00:06:41.958 --> 00:06:45.393
So, as you do, I made a new profile,

00:06:45.417 --> 00:06:50.268
and this time, instead of my own picture,
I chose a cat with a jet pack on it.

00:06:50.292 --> 00:06:52.750
And I chose a name
that did not reveal my gender.

00:06:53.917 --> 00:06:56.643
You can probably guess
where this is going, right?

00:06:56.667 --> 00:07:03.059
So, this time, I didn't get any of those
patronizing comments about my ability

00:07:03.083 --> 00:07:06.417
and I was able to actually
get some work done.

00:07:07.500 --> 00:07:09.351
And it sucks, guys.

00:07:09.375 --> 00:07:11.851
I've been building robots since I was 15,

00:07:11.875 --> 00:07:14.143
I have a few degrees in computer science,

00:07:14.167 --> 00:07:16.601
and yet, I had to hide my gender

00:07:16.625 --> 00:07:18.875
in order for my work
to be taken seriously.

00:07:19.875 --> 00:07:21.768
So, what's going on here?

00:07:21.792 --> 00:07:25.000
Are men just better
at technology than women?

00:07:25.917 --> 00:07:27.476
Another study found

00:07:27.500 --> 00:07:32.434
that when women coders on one platform
hid their gender, like myself,

00:07:32.458 --> 00:07:35.708
their code was accepted
four percent more than men.

00:07:36.542 --> 00:07:39.458
So this is not about the talent.

00:07:39.958 --> 00:07:42.851
This is about an elitism in AI

00:07:42.875 --> 00:07:45.667
that says a programmer
needs to look like a certain person.

00:07:47.375 --> 00:07:50.476
What we really need to do
to make AI better

00:07:50.500 --> 00:07:53.542
is bring people
from all kinds of backgrounds.

00:07:54.542 --> 00:07:57.101
We need people who can
write and tell stories

00:07:57.125 --> 00:07:59.292
to help us create personalities of AI.

00:08:00.208 --> 00:08:02.250
We need people who can solve problems.

00:08:03.125 --> 00:08:06.893
We need people
who face different challenges

00:08:06.917 --> 00:08:12.268
and we need people who can tell us
what are the real issues that need fixing,

00:08:12.292 --> 00:08:15.333
and help us find ways
that technology can actually fix it.

00:08:17.833 --> 00:08:21.559
Because, when people
from diverse backgrounds come together,

00:08:21.583 --> 00:08:23.726
when we build things in the right way,

00:08:23.750 --> 00:08:25.792
the possibilities are limitless.

00:08:26.750 --> 00:08:30.059
And that's what I want to end
by talking to you about.

00:08:30.083 --> 00:08:34.308
Less racist robots, less machines
that are going to take our jobs

00:08:34.332 --> 00:08:37.457
and more about what technology
can actually achieve.

00:08:38.292 --> 00:08:41.726
So, yes, some of the energy
in the world of AI,

00:08:41.750 --> 00:08:43.143
in the world of technology

00:08:43.167 --> 00:08:47.434
is going to be about
what ads you see on your stream.

00:08:47.458 --> 00:08:52.667
But a lot of it is going towards
making the world so much better.

00:08:53.500 --> 00:08:57.268
Think about a pregnant woman
in the Democratic Republic of Congo,

00:08:57.292 --> 00:09:01.476
who has to walk 17 hours
to her nearest rural prenatal clinic

00:09:01.500 --> 00:09:03.351
to get a checkup.

00:09:03.375 --> 00:09:06.292
What if she could get diagnosis
on her phone, instead?

00:09:07.750 --> 00:09:09.559
Or think about what AI could do

00:09:09.583 --> 00:09:12.309
for those one in three women
in South Africa

00:09:12.333 --> 00:09:14.458
who face domestic violence.

00:09:15.083 --> 00:09:17.809
If it wasn't safe to talk out loud,

00:09:17.833 --> 00:09:20.309
they could get an AI service
to raise alarm,

00:09:20.333 --> 00:09:22.792
get financial and legal advice.

00:09:23.958 --> 00:09:28.976
These are all real examples of projects
that people, including myself,

00:09:29.000 --> 00:09:31.500
are working on right now, using AI.

00:09:33.542 --> 00:09:37.143
So, I'm sure in the next couple of days
there will be yet another news story

00:09:37.167 --> 00:09:39.851
about the existential risk,

00:09:39.875 --> 00:09:42.309
robots taking over
and coming for your jobs.

00:09:42.333 --> 00:09:43.351
(Laughter)

00:09:43.375 --> 00:09:45.684
And when something like that happens,

00:09:45.708 --> 00:09:49.309
I know I'll get the same messages
worrying about the future.

00:09:49.333 --> 00:09:53.000
But I feel incredibly positive
about this technology.

00:09:55.458 --> 00:10:01.417
This is our chance to remake the world
into a much more equal place.

00:10:02.458 --> 00:10:06.458
But to do that, we need to build it
the right way from the get go.

00:10:07.667 --> 00:10:12.750
We need people of different genders,
races, sexualities and backgrounds.

00:10:14.458 --> 00:10:16.934
We need women to be the makers

00:10:16.958 --> 00:10:19.958
and not just the machines
who do the makers' bidding.

00:10:21.875 --> 00:10:25.643
We need to think very carefully
what we teach machines,

00:10:25.667 --> 00:10:27.309
what data we give them,

00:10:27.333 --> 00:10:30.458
so they don't just repeat
our own past mistakes.

00:10:32.125 --> 00:10:35.667
So I hope I leave you
thinking about two things.

00:10:36.542 --> 00:10:41.101
First, I hope you leave
thinking about bias today.

00:10:41.125 --> 00:10:44.309
And that the next time
you scroll past an advert

00:10:44.333 --> 00:10:47.143
that assumes you are interested
in fertility clinics

00:10:47.167 --> 00:10:50.018
or online betting websites,

00:10:50.042 --> 00:10:52.059
that you think and remember

00:10:52.083 --> 00:10:56.708
that the same technology is assuming
that a black man will reoffend.

00:10:57.833 --> 00:11:02.000
Or that a woman is more likely
to be a personal assistant than a CEO.

00:11:02.958 --> 00:11:06.667
And I hope that reminds you
that we need to do something about it.

00:11:08.917 --> 00:11:10.768
And second,

00:11:10.792 --> 00:11:12.684
I hope you think about the fact

00:11:12.708 --> 00:11:14.684
that you don't need to look a certain way

00:11:14.708 --> 00:11:18.559
or have a certain background
in engineering or technology

00:11:18.583 --> 00:11:19.851
to create AI,

00:11:19.875 --> 00:11:22.750
which is going to be
a phenomenal force for our future.

00:11:24.166 --> 00:11:26.309
You don't need to look
like a Mark Zuckerberg,

00:11:26.333 --> 00:11:27.583
you can look like me.

00:11:29.250 --> 00:11:32.143
And it is up to all of us in this room

00:11:32.167 --> 00:11:34.893
to convince the governments
and the corporations

00:11:34.917 --> 00:11:37.809
to build AI technology for everyone,

00:11:37.833 --> 00:11:40.226
including the edge cases.

00:11:40.250 --> 00:11:42.309
And for us all to get education

00:11:42.333 --> 00:11:44.708
about this phenomenal
technology in the future.

00:11:46.167 --> 00:11:48.184
Because if we do that,

00:11:48.208 --> 00:11:53.101
then we've only just scratched the surface
of what we can achieve with AI.

00:11:53.125 --> 00:11:54.393
Thank you.

00:11:54.417 --> 00:11:57.125
(Applause)