0:00:01.041,0:00:04.175
Hello, I'm Joy, a poet of code,

0:00:04.199,0:00:09.192
on a mission to stop[br]an unseen force that's rising,

0:00:09.216,0:00:12.072
a force that I called "the coded gaze,"

0:00:12.096,0:00:15.405
my term for algorithmic bias.

0:00:15.429,0:00:19.729
Algorithmic bias, like human bias,[br]results in unfairness.

0:00:19.753,0:00:25.775
However, algorithms, like viruses,[br]can spread bias on a massive scale

0:00:25.799,0:00:27.381
at a rapid pace.

0:00:27.943,0:00:32.330
Algorithmic bias can also lead[br]to exclusionary experiences

0:00:32.354,0:00:34.482
and discriminatory practices.

0:00:34.506,0:00:36.567
Let me show you what I mean.

0:00:38.250,0:00:40.490
(Video) Joy Boulamwini: Camera.[br]I've got a face.

0:00:40.514,0:00:42.387
Can you see my face?

0:00:42.411,0:00:45.524
No-glasses face.

0:00:45.548,0:00:48.241
You can see her face.

0:00:48.265,0:00:50.495
What about my face?

0:00:53.255,0:00:54.406
(Laughter)

0:00:54.430,0:00:58.180
I've got a mask. Can you see my mask?

0:00:59.014,0:01:01.379
Joy Boulamwini: So how did this happen?

0:01:01.403,0:01:04.544
Why am I sitting in front of a computer

0:01:04.568,0:01:05.992
in a white mask,

0:01:06.016,0:01:09.666
trying to be detected by a cheap webcam?

0:01:09.690,0:01:11.981
Well, when I'm not fighting the coded gaze

0:01:12.005,0:01:13.525
as a poet of code,

0:01:13.549,0:01:16.821
I'm a graduate student[br]at the MIT Media Lab,

0:01:16.845,0:01:21.762
and there I have the opportunity to work[br]on all sorts of whimsical projects,

0:01:21.786,0:01:23.813
including the Aspire Mirror,

0:01:23.837,0:01:28.971
a project I did so I could project[br]digital masks onto my reflection.

0:01:28.995,0:01:31.345
So in the morning, if I wanted[br]to feel powerful,

0:01:31.369,0:01:32.803
I could put on a lion.

0:01:32.827,0:01:36.323
If I wanted to be uplifted,[br]I might have a quote.

0:01:36.347,0:01:39.336
So I used generic[br]facial recognition software

0:01:39.360,0:01:40.711
to build the system,

0:01:40.735,0:01:45.838
but found it was really hard to test it[br]unless I wore a white mask.

0:01:46.822,0:01:51.168
Unfortunately, I've run[br]into this issue before.

0:01:51.192,0:01:55.495
When I was an undergraduate[br]at Georgia Tech studying computer science,

0:01:55.519,0:01:57.574
I used to work on social robots,

0:01:57.598,0:02:01.375
and one of my tasks was to get a robot[br]to play peek-a-boo,

0:02:01.399,0:02:03.082
a simple turn-taking game

0:02:03.106,0:02:07.427
where partners cover their face[br]and then uncover it saying, "Peek-a-boo!"

0:02:07.451,0:02:11.880
The problem is, peek-a-boo[br]doesn't really work if I can't see you,

0:02:11.904,0:02:14.403
and my robot couldn't see me.

0:02:14.427,0:02:18.377
But I borrowed my roommate's face[br]to get the project done,

0:02:18.401,0:02:19.781
submitted the assignment,

0:02:19.805,0:02:23.558
and figured, you know what,[br]somebody else will solve this problem.

0:02:24.209,0:02:26.212
Not too longer after,

0:02:26.236,0:02:30.395
I was in Hong Kong[br]for an entrepreneurship competition.

0:02:30.879,0:02:33.573
The organizers decided[br]to take participants

0:02:33.597,0:02:35.969
on a tour of local start-ups.

0:02:35.993,0:02:38.708
One of the start-ups had a social robot,

0:02:38.732,0:02:40.644
and they decided to do a demo.

0:02:40.668,0:02:43.648
The demo worked on everybody[br]until it got to me,

0:02:43.672,0:02:45.595
and you can probably guess it.

0:02:45.619,0:02:48.584
It couldn't detect my face.

0:02:48.608,0:02:51.119
I asked the developers what was going on,

0:02:51.143,0:02:56.676
and it turned out we had used the same[br]generic facial recognition software.

0:02:56.700,0:02:58.350
Halfway around the world,

0:02:58.374,0:03:02.226
I learned that algorithmic bias[br]can travel as quickly

0:03:02.250,0:03:05.420
as it takes to download[br]some files off of the internet.

0:03:06.285,0:03:09.361
So what's going on?[br]Why isn't my face being detected?

0:03:09.385,0:03:12.741
Well, we have to look at[br]how we give machines sight.

0:03:12.765,0:03:16.174
Computer vision uses[br]machine learning techniques

0:03:16.198,0:03:18.078
to do facial recognition.

0:03:18.102,0:03:21.999
So how this works is, you create[br]a training set with examples of faces.

0:03:22.023,0:03:24.841
This is a face. This is a face.[br]This is not a face.

0:03:24.865,0:03:29.384
And over time, you can teach a computer[br]how to recognize other faces.

0:03:29.408,0:03:33.397
However, if the training sets[br]aren't really that diverse,

0:03:33.421,0:03:36.770
any face that deviates too much[br]from the established norm

0:03:36.794,0:03:38.443
will be harder to detect,

0:03:38.467,0:03:40.430
which is what was happening to me.

0:03:40.454,0:03:42.836
But don't worry -- there's some good news.

0:03:42.860,0:03:45.631
Training sets don't just[br]materialize out of nowhere.

0:03:45.655,0:03:47.443
We actually can create them.

0:03:47.467,0:03:51.643
So there's an opportunity to create[br]full-spectrum training sets

0:03:51.667,0:03:55.491
that reflect a richer[br]portrait of humanity.

0:03:55.515,0:03:57.736
Now you've seen in my examples

0:03:57.760,0:03:59.528
how social robots

0:03:59.552,0:04:04.163
was how I found out about exclusion[br]with algorithmic bias.

0:04:04.187,0:04:09.002
But algorithmic bias can also lead[br]to discriminatory practices.

0:04:09.977,0:04:11.430
Across the US,

0:04:11.454,0:04:15.652
police departments are starting to use[br]facial recognition software

0:04:15.676,0:04:18.135
in their crime-fighting arsenal.

0:04:18.159,0:04:20.172
Georgetown Law published a report

0:04:20.196,0:04:26.959
showing that one in two adults[br]in the US -- that's 117 million people --

0:04:26.983,0:04:30.517
have their faces[br]in facial recognition networks.

0:04:30.541,0:04:35.093
Police departments can currently look[br]at these networks unregulated,

0:04:35.117,0:04:39.403
using algorithms that have not[br]been audited for accuracy.

0:04:39.427,0:04:43.291
Yet we know facial recognition[br]is not fail proof,

0:04:43.315,0:04:47.494
and labeling faces consistently[br]remains a challenge.

0:04:47.518,0:04:49.280
You might have seen this on Facebook.

0:04:49.304,0:04:52.292
My friends and I laugh all the time[br]when we see other people

0:04:52.316,0:04:54.774
mislabeled in our photos.

0:04:54.798,0:05:00.389
But misidentifying a suspected criminal[br]is no laughing matter,

0:05:00.413,0:05:03.240
nor is breaching civil liberties.

0:05:03.264,0:05:06.469
Machine learning is being used[br]for facial recognition,

0:05:06.493,0:05:10.998
but it's also extending beyond the realm[br]of computer vision.

0:05:11.806,0:05:15.822
In her book, "Weapons[br]of Math Destruction,"

0:05:15.846,0:05:22.527
data scientist Cathy O'Neil[br]talks about the rising new WMDs --

0:05:22.551,0:05:26.904
widespread, mysterious[br]and destructive algorithms

0:05:26.928,0:05:29.892
that are increasingly being used[br]to make decisions

0:05:29.916,0:05:33.093
that impact more aspects of our lives.

0:05:33.117,0:05:34.987
So who gets hired or fired?

0:05:35.011,0:05:37.123
Do you get that loan?[br]Do you get insurance?

0:05:37.147,0:05:40.650
Are you admitted into the college[br]you wanted to get into?

0:05:40.674,0:05:44.183
Do you and I pay the same price[br]for the same product

0:05:44.207,0:05:46.649
purchased on the same platform?

0:05:46.673,0:05:50.432
Law enforcement is also starting[br]to use machine learning

0:05:50.456,0:05:52.745
for predictive policing.

0:05:52.769,0:05:56.263
Some judges use machine-generated[br]risk scores to determine

0:05:56.287,0:06:00.689
how long an individual[br]is going to spend in prison.

0:06:00.713,0:06:03.167
So we really have to think[br]about these decisions.

0:06:03.191,0:06:04.373
Are they fair?

0:06:04.397,0:06:07.287
And we've seen that algorithmic bias

0:06:07.311,0:06:10.685
doesn't necessarily always[br]lead to fair outcomes.

0:06:10.709,0:06:12.673
So what can we do about it?

0:06:12.697,0:06:16.377
Well, we can start thinking about[br]how we create more inclusive code

0:06:16.401,0:06:19.391
and employ inclusive coding practices.

0:06:19.415,0:06:21.724
It really starts with people.

0:06:22.248,0:06:24.209
So who codes matters.

0:06:24.233,0:06:28.352
Are we creating full-spectrum teams[br]with diverse individuals

0:06:28.376,0:06:30.787
who can check each other's blind spots?

0:06:30.811,0:06:34.356
On the technical side,[br]how we code matters.

0:06:34.380,0:06:38.031
Are we factoring in fairness[br]as we're developing systems?

0:06:38.055,0:06:40.968
And finally, why we code matters.

0:06:41.325,0:06:46.408
We've used tools of computational creation[br]to unlock immense wealth.

0:06:46.432,0:06:50.879
We now have the opportunity[br]to unlock even greater equality

0:06:50.903,0:06:53.833
if we make social change a priority

0:06:53.857,0:06:56.027
and not an afterthought.

0:06:56.548,0:07:01.070
And so these are the three tenets[br]that will make up the "incoding" movement.

0:07:01.094,0:07:02.746
Who codes matters,

0:07:02.770,0:07:04.313
how we code matters

0:07:04.337,0:07:06.360
and why we code matters.

0:07:06.384,0:07:09.483
So to go towards encoding,[br]we can start thinking about

0:07:09.507,0:07:12.671
building platforms that can identify bias

0:07:12.695,0:07:15.773
by collecting people's experiences[br]like the ones I shared,

0:07:15.797,0:07:18.867
but also auditing existing software.

0:07:18.891,0:07:22.656
We can also start to create[br]more inclusive training sets.

0:07:22.680,0:07:25.483
Imagine a "Selfies for Inclusion" campaign

0:07:25.507,0:07:29.162
where you and I can help[br]developers test and create

0:07:29.186,0:07:31.279
more inclusive training sets.

0:07:31.842,0:07:34.670
And we can also start thinking[br]more conscientiously

0:07:34.694,0:07:40.085
about the social impact[br]of the technology that we're developing.

0:07:40.109,0:07:42.502
To get the incoding movement started,

0:07:42.526,0:07:45.373
I've launched the Algorithmic[br]Justice League,

0:07:45.397,0:07:51.269
where anyone who cares about fairness[br]can help fight the coded gaze.

0:07:51.293,0:07:54.589
On codedgaze.com, you can report bias,

0:07:54.613,0:07:57.058
request audits, become a tester

0:07:57.082,0:07:59.853
and join the ongoing conversation,

0:07:59.877,0:08:02.164
#codedgaze.

0:08:03.282,0:08:05.769
So I invite you to join me

0:08:05.793,0:08:09.512
in creating a world where technology[br]works for all of us,

0:08:09.536,0:08:11.433
not just some of us,

0:08:11.457,0:08:16.045
a world where we value inclusion[br]and center social change.

0:08:16.069,0:08:17.244
Thank you.

0:08:17.268,0:08:21.539
(Applause)

0:08:23.413,0:08:26.267
But I have one question:

0:08:26.291,0:08:28.350
Will you join me in the fight?

0:08:28.374,0:08:29.659
(Laughter)

0:08:29.683,0:08:33.370
(Applause)