0:00:00.881,0:00:04.635
How many decisions[br]have been made about you today,

0:00:04.659,0:00:07.278
or this week or this year,

0:00:07.302,0:00:09.238
by artificial intelligence?

0:00:10.943,0:00:12.641
I build AI for a living,

0:00:12.665,0:00:15.236
so full disclosure -- I'm kind of a nerd.

0:00:15.760,0:00:18.093
And because I'm kind of a nerd,

0:00:18.117,0:00:20.471
wherever some new news story comes out

0:00:20.495,0:00:23.918
about artificial intelligence[br]stealing all our jobs,

0:00:23.942,0:00:28.172
or robots getting citizenship[br]of an actual country,

0:00:28.196,0:00:31.292
I'm the person my friends[br]and followers message

0:00:31.316,0:00:33.173
freaking out about the future.

0:00:33.839,0:00:35.474
We see this everywhere.

0:00:35.957,0:00:40.395
This media panic that[br]our robot overlords are taking over.

0:00:40.871,0:00:43.137
We could blame Hollywood for that.

0:00:44.109,0:00:48.267
But in reality, that's not the problem[br]we should be focusing on.

0:00:49.259,0:00:52.910
There is a more pressing danger,[br]a bigger risk with AI,

0:00:52.934,0:00:54.667
that we need to fix first.

0:00:55.394,0:00:57.744
So we are back to this question:

0:00:57.768,0:01:02.466
How many decisions have been made[br]about you today by AI?

0:01:03.800,0:01:05.752
And how many of these

0:01:05.776,0:01:10.275
were based on your gender,[br]your race or your background?

0:01:12.513,0:01:15.247
Algorithms are being used all the time

0:01:15.271,0:01:19.275
to make decisions about who we are[br]and what we want.

0:01:20.204,0:01:23.847
Some of the women in this room[br]will know what I'm talking about

0:01:23.871,0:01:27.609
if you've been made to sit through[br]those pregnancy test adverts on You Tube

0:01:27.633,0:01:29.289
like 1,000 times.

0:01:29.736,0:01:32.611
Or you've scrolled past adverts[br]of fertility clinics

0:01:32.635,0:01:34.974
on your Facebook feed.

0:01:35.641,0:01:38.030
Or in my case, Indian marriage bureaus.

0:01:38.054,0:01:39.307
(Laughter)

0:01:39.331,0:01:42.323
But AI isn't just being used[br]to make decisions

0:01:42.347,0:01:44.950
about what products we want to buy

0:01:44.974,0:01:47.774
or which show we want to binge watch next.

0:01:49.030,0:01:53.567
I wonder how you'd feel about someone[br]who thought things like this:

0:01:54.266,0:01:56.170
a black or Latino person

0:01:56.194,0:02:00.291
is less likely than a white person[br]to pay off their loan on time.

0:02:01.528,0:02:04.366
A person called John[br]makes a better programmer

0:02:04.390,0:02:06.055
than a person called Marry.

0:02:07.230,0:02:12.341
A black man is more likely to be[br]a repeat offender than a white man.

0:02:14.950,0:02:16.144
You're probably thinking,

0:02:16.168,0:02:19.989
"Wow, that sounds like a pretty sexist,[br]racist person," right?

0:02:20.990,0:02:25.859
These are some real decisions[br]that AI has made very recently,

0:02:25.883,0:02:28.812
based on the biases[br]it has learned from us,

0:02:28.836,0:02:29.986
from the humans.

0:02:31.741,0:02:36.556
AI is being used to help decide[br]whether or not you get that job interview;

0:02:36.580,0:02:39.000
how much you pay for your car insurance;

0:02:39.024,0:02:40.889
how good your credit score is;

0:02:40.913,0:02:44.611
and even what rating you get[br]in your annual performance review.

0:02:45.006,0:02:48.219
But these decisions[br]are all being filtered through

0:02:48.243,0:02:54.138
it's assumptions about our identity,[br]our race, our gender, our age.

0:02:56.269,0:02:57.736
How is that happening?

0:02:58.538,0:03:02.102
Now, imagine an AI is helping[br]a hiring manager

0:03:02.126,0:03:04.934
find the next tech leader in the company.

0:03:04.958,0:03:08.077
So far, the manager[br]has been hiring mostly men.

0:03:08.101,0:03:12.729
So the AI learns men are more likely[br]to be programmers than women.

0:03:13.526,0:03:16.454
And it's a very short leap from there to:

0:03:16.478,0:03:18.884
men make better programmers than women.

0:03:19.336,0:03:22.784
We have reinforced[br]our own bias into the AI.

0:03:23.165,0:03:26.791
And now, it's screening out[br]female candidates.

0:03:28.903,0:03:31.926
Hang on, if a human[br]hiring manager did that,

0:03:31.950,0:03:34.307
we'd be outraged, we wouldn't allow it.

0:03:34.331,0:03:37.398
This kind of gender[br]discrimination is not OK.

0:03:37.815,0:03:42.338
And yet somehow,[br]AI has become above the law,

0:03:42.362,0:03:44.489
because a machine made the decision.

0:03:45.838,0:03:46.988
That's not it.

0:03:47.377,0:03:52.053
We are also reinforcing our bias[br]in how we interact with AI.

0:03:52.902,0:03:58.059
How often do you use a voice assistant[br]like Siri, Alexa or even Cortana?

0:03:58.850,0:04:01.445
They all have two things in common:

0:04:01.469,0:04:04.596
one, they can never get my name right,

0:04:04.620,0:04:07.309
and second, they are all female.

0:04:08.419,0:04:11.188
They are designed to be[br]our obedient servants,

0:04:11.212,0:04:14.238
turning your lights on and off,[br]ordering your shopping.

0:04:15.117,0:04:18.426
You get male AIs too,[br]but they tend to be more high-powered,

0:04:18.450,0:04:21.526
like IBM Watson,[br]making business decisions,

0:04:21.550,0:04:25.558
Salesforce Einstein[br]or ROSS, the robot lawyer.

0:04:26.228,0:04:30.267
So poor robots, even they suffer[br]from sexism in the workplace.

0:04:30.291,0:04:31.441
(Laughter)

0:04:32.530,0:04:35.378
Think about how these two things combine

0:04:35.402,0:04:39.984
and affect a kid growing up[br]in today's world around AI.

0:04:40.744,0:04:43.664
So they're doing some research[br]for a school project

0:04:43.688,0:04:46.728
and they google images of CEO.

0:04:46.752,0:04:49.649
The algorithm shows them[br]results of mostly men.

0:04:49.673,0:04:52.221
And now, they google personal assistant.

0:04:52.245,0:04:55.275
As you can guess,[br]it shows them mostly females.

0:04:55.704,0:04:59.323
And then they want to put on some music,[br]and maybe order some food,

0:04:59.347,0:05:05.545
and now, they are barking orders[br]at an obedient female voice assistant.

0:05:07.554,0:05:12.279
Some of our brightest minds[br]are creating this technology today.

0:05:12.848,0:05:16.541
Technology that they could have created[br]in any way they wanted.

0:05:17.085,0:05:22.752
And yet, they have chosen to create it[br]in the style of 1950s "Mad Man" secretary.

0:05:22.776,0:05:23.926
Yay!

0:05:24.958,0:05:26.252
But OK, don't worry,

0:05:26.276,0:05:28.331
this is not going to end[br]with me telling you

0:05:28.355,0:05:31.847
that we are all heading towards[br]sexist, racist machines running the world.

0:05:32.783,0:05:38.457
The good news about AI[br]is that it is entirely within our control.

0:05:39.346,0:05:43.053
We get to teach the right values,[br]the right ethics to AI.

0:05:44.172,0:05:46.354
So there are three things we can do.

0:05:46.378,0:05:49.735
One, we can be aware of our own biases

0:05:49.759,0:05:52.093
and the bias in machines around us.

0:05:52.498,0:05:57.037
Two, we can make sure that our[br]diverse teams building this technology.

0:05:57.061,0:06:01.566
And three, we have to give it[br]diverse experiences to learn from.

0:06:02.864,0:06:06.181
I can talk about the first two[br]from personal experience.

0:06:06.205,0:06:07.659
When you work in technology

0:06:07.683,0:06:11.070
and you don't look like[br]a Mark Zuckerberg or Elon Musk,

0:06:11.094,0:06:14.537
your life is a little bit difficult,[br]your ability gets questioned.

0:06:15.910,0:06:17.263
Here's just one example.

0:06:17.287,0:06:21.038
Like most developers,[br]I often join online tech forums

0:06:21.062,0:06:23.595
and share my knowledge to help others.

0:06:24.299,0:06:25.617
And I've found,

0:06:25.641,0:06:29.585
when I log on as myself,[br]with my own photo, my own name,

0:06:29.609,0:06:34.212
I tend to get questions[br]or comments like this:

0:06:34.236,0:06:37.562
"What makes you think[br]you're qualified to talk about AI?"

0:06:38.443,0:06:41.514
"What makes you think[br]you know about machine learning?"

0:06:41.974,0:06:45.379
So, as you do, I made a new profile,

0:06:45.403,0:06:49.576
and this time, instead of my own picture,[br]I chose a cat with a jet pack on it.

0:06:50.292,0:06:53.152
And I chose a name[br]that did not reveal my gender.

0:06:53.927,0:06:56.633
You can probably guess[br]where this is going, right?

0:06:56.657,0:07:03.077
So, this time, I didn't get any of those[br]patronizing comments about my ability

0:07:03.101,0:07:06.172
and I was able to actually[br]get some work done.

0:07:07.482,0:07:08.966
And it sucks, guys.

0:07:09.371,0:07:11.871
I've been building robots since I was 15,

0:07:11.895,0:07:14.156
I have a few degrees in computer science,

0:07:14.180,0:07:16.593
and yet, I had to hide my gender

0:07:16.617,0:07:19.312
in order for my work[br]to be taken seriously.

0:07:19.887,0:07:21.752
So, what's going on here?

0:07:21.776,0:07:25.212
Are men just better[br]at technology than women?

0:07:25.887,0:07:27.490
Another study found

0:07:27.514,0:07:32.414
that when women coders on one platform[br]hid their gender, like myself,

0:07:32.438,0:07:35.680
their code was accepted[br]four percent more than men.

0:07:36.557,0:07:39.064
So this is not about the talent.

0:07:39.973,0:07:42.847
This is about an elitism in AI

0:07:42.871,0:07:46.162
that says a programmer[br]needs to look like a certain person.

0:07:47.363,0:07:50.490
What we really need to do[br]to make AI better

0:07:50.514,0:07:53.647
is bring people from[br]all kinds of backgrounds.

0:07:54.537,0:07:57.109
We need people who can[br]write and tell stories

0:07:57.133,0:07:59.666
to help us create personalities of AI.

0:08:00.220,0:08:02.753
We need people who can solve problems.

0:08:03.109,0:08:06.886
We need people who face[br]different challenges

0:08:06.910,0:08:12.275
and we need people who can tell us[br]what are the real issues that need fixing,

0:08:12.299,0:08:15.735
and help us find ways[br]that technology can actually fix it.

0:08:17.830,0:08:21.576
Because, when people[br]from diverse backgrounds come together,

0:08:21.600,0:08:23.726
when we build things in the right way,

0:08:23.750,0:08:25.884
the possibilities are limitless.

0:08:26.720,0:08:29.548
And that's what I want to end[br]by talking to you about.

0:08:30.101,0:08:34.322
Less racist robots, less machines[br]that are going to take our jobs,

0:08:34.346,0:08:37.712
and more about what technology[br]can actually achieve.

0:08:38.284,0:08:41.714
So, yes, some of the energy[br]in the world of AI,

0:08:41.738,0:08:43.140
in the world of technology

0:08:43.164,0:08:46.844
is going to be about[br]what ads you see on your stream.

0:08:47.458,0:08:52.271
But a lot of it is going towards[br]making the world so much better.

0:08:53.486,0:08:57.279
Think about a pregnant woman[br]in the Democratic Republic of Congo,

0:08:57.303,0:09:01.482
who has to walk 17 hours[br]to her nearest rural prenatal clinic

0:09:01.506,0:09:02.656
to get a checkup.

0:09:03.355,0:09:06.712
What if she could get diagnosis[br]on her phone, instead?

0:09:07.783,0:09:09.577
Or think about what AI could do

0:09:09.601,0:09:12.300
for those one in three women[br]in South Africa

0:09:12.324,0:09:14.124
who face domestic violence.

0:09:15.101,0:09:17.798
If it wasn't safe to talk out loud,

0:09:17.822,0:09:20.314
they could get an AI service[br]to raise alarm,

0:09:20.338,0:09:22.405
get financial and legal advice.

0:09:23.950,0:09:28.974
These are all real examples of projects[br]that people, including myself,

0:09:28.998,0:09:31.505
are working on right now, using AI.

0:09:33.553,0:09:37.140
So, I'm sure in the next couple of days[br]there will be yet another news story

0:09:37.164,0:09:39.846
about the existential risk,

0:09:39.870,0:09:42.292
robots taking over[br]and coming for your jobs.

0:09:42.316,0:09:43.339
(Laughter)

0:09:43.363,0:09:45.696
And when something like that happens,

0:09:45.720,0:09:49.052
I know I'll get the same messages[br]worrying about the future.

0:09:49.353,0:09:53.019
But I feel incredibly positive[br]about this technology.

0:09:55.456,0:10:01.111
This is our chance to remake the world[br]into a much more equal place.

0:10:02.461,0:10:06.455
But to do that, we need to build it[br]the right way from the get go.

0:10:07.651,0:10:12.546
We need people of different genders,[br]races, sexualities and backgrounds.

0:10:14.451,0:10:16.943
We need women to be the makers

0:10:16.967,0:10:20.500
and not just the machines[br]who do the makers' bidding.

0:10:21.895,0:10:25.625
We need to think very carefully[br]what we teach machines,

0:10:25.649,0:10:27.300
what data we give them,

0:10:27.324,0:10:30.591
so they don't just repeat[br]our own past mistakes.

0:10:32.118,0:10:35.663
So I hope I leave you[br]thinking about two things.

0:10:36.567,0:10:40.614
First, I hope you leave[br]thinking about bias today.

0:10:41.114,0:10:44.289
And that the next time[br]you scroll past an advert

0:10:44.313,0:10:47.127
that assumes you are interested[br]in fertility clinics

0:10:47.151,0:10:49.998
or online betting websites,

0:10:50.022,0:10:52.045
that you think and remember

0:10:52.069,0:10:56.399
that the same technology is assuming[br]that a black man will reoffend.

0:10:57.831,0:11:02.204
Or that a woman is more likely[br]to be a personal assistant than a CEO.

0:11:02.956,0:11:06.661
And I hope that reminds you[br]that we need to do something about it.

0:11:08.908,0:11:10.773
And second,

0:11:10.797,0:11:12.688
I hope you think about the fact

0:11:12.712,0:11:16.133
that you don't need to look a certain way[br]or have a certain background

0:11:16.157,0:11:19.649
in engineering or technology to create AI,

0:11:19.673,0:11:22.731
which is going to be a phenomenal[br]force for our future.

0:11:24.157,0:11:26.315
You don't need to look[br]like a Mark Zuckerberg,

0:11:26.339,0:11:27.739
you can look like me.

0:11:29.266,0:11:32.163
And it is up to all of us in this room

0:11:32.187,0:11:34.879
to convince the governments[br]and the corporations

0:11:34.903,0:11:37.806
to build AI technology for everyone,

0:11:37.830,0:11:39.559
including the edge cases.

0:11:40.225,0:11:42.297
And for us, all to get education

0:11:42.321,0:11:45.017
about this phenomenal[br]technology in the future.

0:11:46.130,0:11:48.193
Because if we do that,

0:11:48.217,0:11:52.633
then we've only just scratched the surface[br]of what we can achieve with AI.

0:11:53.143,0:11:54.301
Thank you.

0:11:54.325,0:11:57.128
(Applause)