How to keep human biases out of AI
-
0:01 - 0:05How many decisions
have been made about you today, -
0:05 - 0:07or this week or this year,
-
0:07 - 0:09by artificial intelligence?
-
0:11 - 0:13I build AI for a living
-
0:13 - 0:16so, full disclosure, I'm kind of a nerd.
-
0:16 - 0:18And because I'm kind of a nerd,
-
0:18 - 0:20wherever some new news story comes out
-
0:20 - 0:24about artificial intelligence
stealing all our jobs, -
0:24 - 0:28or robots getting citizenship
of an actual country, -
0:28 - 0:31I'm the person my friends
and followers message -
0:31 - 0:33freaking out about the future.
-
0:34 - 0:36We see this everywhere.
-
0:36 - 0:41This media panic that
our robot overlords are taking over. -
0:41 - 0:43We could blame Hollywood for that.
-
0:44 - 0:48But in reality, that's not the problem
we should be focusing on. -
0:49 - 0:53There is a more pressing danger,
a bigger risk with AI, -
0:53 - 0:54that we need to fix first.
-
0:55 - 0:58So we are back to this question:
-
0:58 - 1:02How many decisions
have been made about you today by AI? -
1:04 - 1:06And how many of these
-
1:06 - 1:10were based on your gender,
your race or your background? -
1:12 - 1:15Algorithms are being used all the time
-
1:15 - 1:19to make decisions about who we are
and what we want. -
1:20 - 1:24Some of the women in this room
will know what I'm talking about -
1:24 - 1:28if you've been made to sit through
those pregnancy test adverts on YouTube -
1:28 - 1:30like 1,000 times.
-
1:30 - 1:33Or you've scrolled past adverts
of fertility clinics -
1:33 - 1:35on your Facebook feed.
-
1:36 - 1:38Or in my case, Indian marriage bureaus.
-
1:38 - 1:39(Laughter)
-
1:39 - 1:42But AI isn't just being used
to make decisions -
1:42 - 1:45about what products we want to buy
-
1:45 - 1:47or which show we want to binge watch next.
-
1:49 - 1:54I wonder how you'd feel about someone
who thought things like this: -
1:54 - 1:56"A black or Latino person
-
1:56 - 2:00is less likely than a white person
to pay off their loan on time." -
2:02 - 2:04"A person called John
makes a better programmer -
2:04 - 2:06than a person called Mary."
-
2:07 - 2:12"A black man is more likely to be
a repeat offender than a white man." -
2:15 - 2:16You're probably thinking,
-
2:16 - 2:20"Wow, that sounds like a pretty sexist,
racist person," right? -
2:21 - 2:26These are some real decisions
that AI has made very recently, -
2:26 - 2:29based on the biases
it has learned from us, -
2:29 - 2:30from the humans.
-
2:32 - 2:37AI is being used to help decide
whether or not you get that job interview; -
2:37 - 2:39how much you pay for your car insurance;
-
2:39 - 2:41how good your credit score is;
-
2:41 - 2:44and even what rating you get
in your annual performance review. -
2:45 - 2:48But these decisions
are all being filtered through -
2:48 - 2:54its assumptions about our identity,
our race, our gender, our age. -
2:56 - 2:59How is that happening?
-
2:59 - 3:02Now, imagine an AI is helping
a hiring manager -
3:02 - 3:05find the next tech leader in the company.
-
3:05 - 3:08So far, the manager
has been hiring mostly men. -
3:08 - 3:13So the AI learns men are more likely
to be programmers than women. -
3:14 - 3:16And it's a very short leap from there to:
-
3:16 - 3:18men make better programmers than women.
-
3:19 - 3:23We have reinforced
our own bias into the AI. -
3:23 - 3:27And now, it's screening out
female candidates. -
3:29 - 3:32Hang on, if a human
hiring manager did that, -
3:32 - 3:34we'd be outraged, we wouldn't allow it.
-
3:34 - 3:38This kind of gender
discrimination is not OK. -
3:38 - 3:42And yet somehow,
AI has become above the law, -
3:42 - 3:44because a machine made the decision.
-
3:46 - 3:47That's not it.
-
3:47 - 3:52We are also reinforcing our bias
in how we interact with AI. -
3:53 - 3:59How often do you use a voice assistant
like Siri, Alexa or even Cortana? -
3:59 - 4:01They all have two things in common:
-
4:02 - 4:05one, they can never get my name right,
-
4:05 - 4:07and second, they are all female.
-
4:08 - 4:11They are designed to be
our obedient servants, -
4:11 - 4:14turning your lights on and off,
ordering your shopping. -
4:15 - 4:18You get male AIs too,
but they tend to be more high-powered, -
4:18 - 4:22like IBM Watson,
making business decisions, -
4:22 - 4:25Salesforce Einstein
or ROSS, the robot lawyer. -
4:26 - 4:30So poor robots, even they suffer
from sexism in the workplace. -
4:30 - 4:31(Laughter)
-
4:33 - 4:35Think about how these two things combine
-
4:35 - 4:41and affect a kid growing up
in today's world around AI. -
4:41 - 4:44So they're doing some research
for a school project -
4:44 - 4:47and they Google images of CEO.
-
4:47 - 4:50The algorithm shows them
results of mostly men. -
4:50 - 4:52And now, they Google personal assistant.
-
4:52 - 4:56As you can guess,
it shows them mostly females. -
4:56 - 4:59And then they want to put on some music,
and maybe order some food, -
4:59 - 5:06and now, they are barking orders
at an obedient female voice assistant. -
5:08 - 5:13Some of our brightest minds
are creating this technology today. -
5:13 - 5:17Technology that they could have created
in any way they wanted. -
5:17 - 5:23And yet, they have chosen to create it
in the style of 1950s "Mad Man" secretary. -
5:23 - 5:24Yay!
-
5:25 - 5:26But OK, don't worry,
-
5:26 - 5:28this is not going to end
with me telling you -
5:28 - 5:32that we are all heading towards
sexist, racist machines running the world. -
5:33 - 5:39The good news about AI
is that it is entirely within our control. -
5:39 - 5:43We get to teach the right values,
the right ethics to AI. -
5:44 - 5:46So there are three things we can do.
-
5:46 - 5:50One, we can be aware of our own biases
-
5:50 - 5:52and the bias in machines around us.
-
5:52 - 5:57Two, we can make sure that diverse teams
are building this technology. -
5:57 - 6:02And three, we have to give it
diverse experiences to learn from. -
6:03 - 6:06I can talk about the first two
from personal experience. -
6:06 - 6:08When you work in technology
-
6:08 - 6:11and you don't look like
a Mark Zuckerberg or Elon Musk, -
6:11 - 6:15your life is a little bit difficult,
your ability gets questioned. -
6:16 - 6:17Here's just one example.
-
6:17 - 6:21Like most developers,
I often join online tech forums -
6:21 - 6:24and share my knowledge to help others.
-
6:24 - 6:26And I've found,
-
6:26 - 6:30when I log on as myself,
with my own photo, my own name, -
6:30 - 6:34I tend to get questions
or comments like this: -
6:34 - 6:37"What makes you think
you're qualified to talk about AI?" -
6:38 - 6:42"What makes you think
you know about machine learning?" -
6:42 - 6:45So, as you do, I made a new profile,
-
6:45 - 6:50and this time, instead of my own picture,
I chose a cat with a jet pack on it. -
6:50 - 6:53And I chose a name
that did not reveal my gender. -
6:54 - 6:57You can probably guess
where this is going, right? -
6:57 - 7:03So, this time, I didn't get any of those
patronizing comments about my ability -
7:03 - 7:06and I was able to actually
get some work done. -
7:08 - 7:09And it sucks, guys.
-
7:09 - 7:12I've been building robots since I was 15,
-
7:12 - 7:14I have a few degrees in computer science,
-
7:14 - 7:17and yet, I had to hide my gender
-
7:17 - 7:19in order for my work
to be taken seriously. -
7:20 - 7:22So, what's going on here?
-
7:22 - 7:25Are men just better
at technology than women? -
7:26 - 7:27Another study found
-
7:28 - 7:32that when women coders on one platform
hid their gender, like myself, -
7:32 - 7:36their code was accepted
four percent more than men. -
7:37 - 7:39So this is not about the talent.
-
7:40 - 7:43This is about an elitism in AI
-
7:43 - 7:46that says a programmer
needs to look like a certain person. -
7:47 - 7:50What we really need to do
to make AI better -
7:50 - 7:54is bring people
from all kinds of backgrounds. -
7:55 - 7:57We need people who can
write and tell stories -
7:57 - 7:59to help us create personalities of AI.
-
8:00 - 8:02We need people who can solve problems.
-
8:03 - 8:07We need people
who face different challenges -
8:07 - 8:12and we need people who can tell us
what are the real issues that need fixing -
8:12 - 8:15and help us find ways
that technology can actually fix it. -
8:18 - 8:22Because, when people
from diverse backgrounds come together, -
8:22 - 8:24when we build things in the right way,
-
8:24 - 8:26the possibilities are limitless.
-
8:27 - 8:30And that's what I want to end
by talking to you about. -
8:30 - 8:34Less racist robots, less machines
that are going to take our jobs -- -
8:34 - 8:37and more about what technology
can actually achieve. -
8:38 - 8:42So, yes, some of the energy
in the world of AI, -
8:42 - 8:43in the world of technology
-
8:43 - 8:47is going to be about
what ads you see on your stream. -
8:47 - 8:53But a lot of it is going towards
making the world so much better. -
8:54 - 8:57Think about a pregnant woman
in the Democratic Republic of Congo, -
8:57 - 9:01who has to walk 17 hours
to her nearest rural prenatal clinic -
9:02 - 9:03to get a checkup.
-
9:03 - 9:06What if she could get diagnosis
on her phone, instead? -
9:08 - 9:10Or think about what AI could do
-
9:10 - 9:12for those one in three women
in South Africa -
9:12 - 9:14who face domestic violence.
-
9:15 - 9:18If it wasn't safe to talk out loud,
-
9:18 - 9:20they could get an AI service
to raise alarm, -
9:20 - 9:23get financial and legal advice.
-
9:24 - 9:29These are all real examples of projects
that people, including myself, -
9:29 - 9:32are working on right now, using AI.
-
9:34 - 9:37So, I'm sure in the next couple of days
there will be yet another news story -
9:37 - 9:40about the existential risk,
-
9:40 - 9:42robots taking over
and coming for your jobs. -
9:42 - 9:43(Laughter)
-
9:43 - 9:46And when something like that happens,
-
9:46 - 9:49I know I'll get the same messages
worrying about the future. -
9:49 - 9:53But I feel incredibly positive
about this technology. -
9:55 - 10:01This is our chance to remake the world
into a much more equal place. -
10:02 - 10:06But to do that, we need to build it
the right way from the get go. -
10:08 - 10:13We need people of different genders,
races, sexualities and backgrounds. -
10:14 - 10:17We need women to be the makers
-
10:17 - 10:20and not just the machines
who do the makers' bidding. -
10:22 - 10:26We need to think very carefully
what we teach machines, -
10:26 - 10:27what data we give them,
-
10:27 - 10:30so they don't just repeat
our own past mistakes. -
10:32 - 10:36So I hope I leave you
thinking about two things. -
10:37 - 10:41First, I hope you leave
thinking about bias today. -
10:41 - 10:44And that the next time
you scroll past an advert -
10:44 - 10:47that assumes you are interested
in fertility clinics -
10:47 - 10:50or online betting websites,
-
10:50 - 10:52that you think and remember
-
10:52 - 10:57that the same technology is assuming
that a black man will reoffend. -
10:58 - 11:02Or that a woman is more likely
to be a personal assistant than a CEO. -
11:03 - 11:07And I hope that reminds you
that we need to do something about it. -
11:09 - 11:11And second,
-
11:11 - 11:13I hope you think about the fact
-
11:13 - 11:15that you don't need to look a certain way
-
11:15 - 11:19or have a certain background
in engineering or technology -
11:19 - 11:20to create AI,
-
11:20 - 11:23which is going to be
a phenomenal force for our future. -
11:24 - 11:26You don't need to look
like a Mark Zuckerberg, -
11:26 - 11:28you can look like me.
-
11:29 - 11:32And it is up to all of us in this room
-
11:32 - 11:35to convince the governments
and the corporations -
11:35 - 11:38to build AI technology for everyone,
-
11:38 - 11:40including the edge cases.
-
11:40 - 11:42And for us all to get education
-
11:42 - 11:45about this phenomenal
technology in the future. -
11:46 - 11:48Because if we do that,
-
11:48 - 11:53then we've only just scratched the surface
of what we can achieve with AI. -
11:53 - 11:54Thank you.
-
11:54 - 11:57(Applause)
- Title:
- How to keep human biases out of AI
- Speaker:
- Kriti Sharma
- Description:
-
AI algorithms make important decisions about you all the time -- like how much you should pay for car insurance or whether or not you get that job interview. But what happens when these machines are built with human biases coded into their systems? Technologist Kriti Sharma explores how the lack of diversity in tech is creeping into our AI, offering three ways we can start making more ethical algorithms.
- Video Language:
- English
- Team:
- closed TED
- Project:
- TEDTalks
- Duration:
- 12:10
Oliver Friedman edited English subtitles for How to keep human bias out of AI | ||
Oliver Friedman approved English subtitles for How to keep human bias out of AI | ||
Oliver Friedman edited English subtitles for How to keep human bias out of AI | ||
Joanna Pietrulewicz accepted English subtitles for How to keep human bias out of AI | ||
Joanna Pietrulewicz edited English subtitles for How to keep human bias out of AI | ||
Joanna Pietrulewicz edited English subtitles for How to keep human bias out of AI | ||
Ivana Korom edited English subtitles for How to keep human bias out of AI | ||
Ivana Korom edited English subtitles for How to keep human bias out of AI |