How I'm fighting bias in algorithms | Joy Buolamwini

翻译人员: Cong Zhu 校对人员: Aviva Nassimi

Hello, I'm Joy, a poet of code,

大家好 我是乔伊 一位写代码的诗人

on a mission to stop an unseen force that's rising,

我正努力阻止一股 逐渐凸显的无形力量

a force that I called "the coded gaze,"

一种我称为 代码的凝视 的力量

my term for algorithmic bias.

这是我用来定义算法偏见的术语

Algorithmic bias, like human bias, results in unfairness.

正如人类之间的偏见 算法偏见也会导致不公平

However, algorithms, like viruses, can spread bias on a massive scale

然而算法就像病毒一样 会以飞快的速度大范围地

at a rapid pace.

扩散偏见

Algorithmic bias can also lead to exclusionary experiences

算法也将会导致排他的经历和

and discriminatory practices.

歧视性的做法

Let me show you what I mean.

给大家举个例子

(Video) Joy Buolamwini: Hi, camera. I've got a face.

（录像）乔伊·博拉维尼： 嘿 摄像头 我来了

Can you see my face?

你可以看到我的脸吗

No-glasses face?

没有戴眼镜的脸呢

You can see her face.

你可以看到她的脸

What about my face?

那么我的脸呢

I've got a mask. Can you see my mask?

我戴上了一个面罩 你可以看到我的面罩吗

Joy Buolamwini: So how did this happen?

乔伊·博拉维尼： 这是怎么回事呢

Why am I sitting in front of a computer

为什么我坐在一台电脑前

in a white mask,

戴着一个白色的面罩

trying to be detected by a cheap webcam?

尝试着被一个廉价的 网络摄像头检测到

Well, when I'm not fighting the coded gaze

当我的身份不是写代码的诗人

as a poet of code,

与 代码的凝视 较劲的时候

I'm a graduate student at the MIT Media Lab,

我是MIT媒体实验室的 一位硕士生

and there I have the opportunity to work on all sorts of whimsical projects,

在那里我有机会参与 各种不同的项目

including the Aspire Mirror,

包括激励镜子

a project I did so I could project digital masks onto my reflection.

一个可以将数字面罩 投射在我的映像上的项目

So in the morning, if I wanted to feel powerful,

在早上的时候 如果我想充满力量

I could put on a lion.

我可以放上一个狮子的图像

If I wanted to be uplifted, I might have a quote.

如果我想要感到积极向上 我也许就会放上一句格言

So I used generic facial recognition software

我使用通用的人脸识别软件

to build the system,

来搭建系统

but found it was really hard to test it unless I wore a white mask.

但是我发现除非我戴上白色的面罩 否则测试很难成功

Unfortunately, I've run into this issue before.

遗憾的是 我以前 也曾遇到过这种问题

When I was an undergraduate at Georgia Tech studying computer science,

当我在佐治亚理工学院 读计算机科学专业本科的时候

I used to work on social robots,

我曾经在一个 社交机器人上进行实验

and one of my tasks was to get a robot to play peek-a-boo,

我的任务之一是 让机器人玩躲猫猫

a simple turn-taking game

一个简单的轮换游戏

where partners cover their face and then uncover it saying, "Peek-a-boo!"

在游戏中玩伴盖住他们的脸 然后掀开说“躲猫猫！“

The problem is, peek-a-boo doesn't really work if I can't see you,

问题是躲猫猫在我不能 看见你的时候不起作用

and my robot couldn't see me.

而我的机器人看不见我

But I borrowed my roommate's face to get the project done,

我只好借了我室友的脸 去完成这个项目

submitted the assignment,

递交了作业

and figured, you know what, somebody else will solve this problem.

寻思着总会有人 来解决这个问题的把

Not too long after,

不久之后

I was in Hong Kong for an entrepreneurship competition.

我在香港参加一次创业比赛

The organizers decided to take participants

组织者决定将各位参与者

on a tour of local start-ups.

带到当地的初创企业参观

One of the start-ups had a social robot,

其中一个创业公司 有一个社交机器人

and they decided to do a demo.

他们决定进行一个项目演示

The demo worked on everybody until it got to me,

这个项目演示对除我之外的 每个人都有效果

and you can probably guess it.

你恐怕可以猜到

It couldn't detect my face.

它不能检测到我的脸

I asked the developers what was going on,

我问开发师到底发生了什么

and it turned out we had used the same generic facial recognition software.

结果是我们使用了同一款 通用面部识别软件

Halfway around the world,

在地球的另一边

I learned that algorithmic bias can travel as quickly

我意识到算法偏见 传播得如此之快

as it takes to download some files off of the internet.

只需要从互联网上 下载一些文件

So what's going on? Why isn't my face being detected?

那么到底发生了什么 为什么我的脸没有被检测到

Well, we have to look at how we give machines sight.

我们需要了解我们 如何教会机器识别

Computer vision uses machine learning techniques

计算机视觉使用机器学习技术

to do facial recognition.

来进行面部识别

So how this works is, you create a training set with examples of faces.

所以你要用一系列脸的样本 创建一个训练体系

This is a face. This is a face. This is not a face.

这是一张脸 这是一张脸 而这不是一张脸

And over time, you can teach a computer how to recognize other faces.

慢慢地你可以教电脑 如何识别其它的脸

However, if the training sets aren't really that diverse,

然而如果这个训练集 不是那么的多样化

any face that deviates too much from the established norm

那些与已建立的标准 偏差较多的脸

will be harder to detect,

将会难以被检测到

which is what was happening to me.

而这正是我遭遇的问题

But don't worry -- there's some good news.

不过别担心 我们还有好消息

Training sets don't just materialize out of nowhere.

训练集并不是凭空产生的

We actually can create them.

实际上我们可以创造它们

So there's an opportunity to create full-spectrum training sets

现在就有机会去创造 全波段光谱的训练集

that reflect a richer portrait of humanity.

可以反映更加饱满的人类面貌

Now you've seen in my examples

现在你看到了在我的例子中

how social robots

社交机器人

was how I found out about exclusion with algorithmic bias.

使我发现了算法偏见的排他性

But algorithmic bias can also lead to discriminatory practices.

不过算法偏见还会导致 各种歧视性的做法

Across the US,

美国境内的警察局

police departments are starting to use facial recognition software

在打击犯罪的过程中

in their crime-fighting arsenal.

开始使用面部识别软件

Georgetown Law published a report

乔治敦大学法学院 发表了一个报告

showing that one in two adults in the US -- that's 117 million people --

表明在全美两个成年人中就有一个 也就是近1.2亿的人口

have their faces in facial recognition networks.

他们的面部信息 被储存在了面部识别网络中

Police departments can currently look at these networks unregulated,

警察局如今可以访问 这些未被规范的

using algorithms that have not been audited for accuracy.

使用着未审核准确性的 算法的面部识别网络

Yet we know facial recognition is not fail proof,

然而我们知道面部识别 并非万无一失

and labeling faces consistently remains a challenge.

而持续地给面部标签 还是很有挑战性的

You might have seen this on Facebook.

你也许在Facebook上见过这个

My friends and I laugh all the time when we see other people

当我和我的朋友看到其他人 在我们的照片上被错误标注时

mislabeled in our photos.

都会捧腹大笑

But misidentifying a suspected criminal is no laughing matter,

但是误认一个犯罪嫌疑人 可不是闹着玩儿的

nor is breaching civil liberties.

对公民自由的侵犯也不容忽视

Machine learning is being used for facial recognition,

机器学习正被用于面部识别

but it's also extending beyond the realm of computer vision.

但也延伸到了计算机视觉领域之外

In her book, "Weapons of Math Destruction,"

在数据科学家凯西·欧奈尔在她 《数学杀伤性武器》一书中

data scientist Cathy O'Neil talks about the rising new WMDs --

叙述了逐渐严重的 新型大规模杀伤性武器

widespread, mysterious and destructive algorithms

即 广泛应用而又神秘的 具有破坏性的算法

that are increasingly being used to make decisions

正在被越来越多地 运用于决策制定上

that impact more aspects of our lives.

而这些决策影响着 我们生活的方方面面

So who gets hired or fired?

谁被录用 又有谁被解雇

Do you get that loan? Do you get insurance?

你得到了贷款吗 你买到了保险吗

Are you admitted into the college you wanted to get into?

你被心目中的理想大学录取了吗

Do you and I pay the same price for the same product

在同一平台上的同一件产品

purchased on the same platform?

你和我是否支付同样的价格

Law enforcement is also starting to use machine learning

为了实现警情预测 执法机构也开始

for predictive policing.

使用起机器学习

Some judges use machine-generated risk scores to determine

一些法官使用机器生成的 危险评分来决定

how long an individual is going to spend in prison.

囚犯要在监狱里呆多久

So we really have to think about these decisions.

我们真的应该 仔细思考这些决定

Are they fair?

它们公平吗

And we've seen that algorithmic bias

我们已经清楚了 算法偏见

doesn't necessarily always lead to fair outcomes.

不一定总能带来公平的结果

So what can we do about it?

那我们应该怎么做呢

Well, we can start thinking about how we create more inclusive code

我们可以开始思考如何 创造更具有包容性的代码

and employ inclusive coding practices.

并且运用有包容性的编程实践

It really starts with people.

这真的要从人开始

So who codes matters.

由谁来编程很重要

Are we creating full-spectrum teams with diverse individuals

我们组建的全光谱团队中 是否包括各种各样的个体

who can check each other's blind spots?

他们可以弥补彼此的盲区吗

On the technical side, how we code matters.

在技术层面上 我们如何编程很重要

Are we factoring in fairness as we're developing systems?

我们在研发系统的同时 有没有也考虑到公平的因素

And finally, why we code matters.

最后一点 我们为什么编程也很重要

We've used tools of computational creation to unlock immense wealth.

我们用计算机创建的工具 创造了巨大的财富

We now have the opportunity to unlock even greater equality

现在我们有机会去 创造进一步的平等

if we make social change a priority

我们应该优先考虑社会变革

and not an afterthought.

而不是想着事后优化

And so these are the three tenets that will make up the "incoding" movement.

所以这三个宗旨 将构成“译码”运动

Who codes matters,

由谁来编程很重要

how we code matters

我们如何编程很重要

and why we code matters.

以及我们为什么编程很重要

So to go towards incoding, we can start thinking about

所以就译码来说 我们可以开始考虑

building platforms that can identify bias

建立一个我们可以辨识偏见的平台

by collecting people's experiences like the ones I shared,

通过收集人们与我类似的经历

but also auditing existing software.

不过也要审查现有的软件

We can also start to create more inclusive training sets.

我们也可以创造一些 更有包容性的训练集

Imagine a "Selfies for Inclusion" campaign

想象一个为了包容性的自拍运动

where you and I can help developers test and create

在那里 你和我可以帮助 程序员测试以及创造

more inclusive training sets.

更具包容性的训练集

And we can also start thinking more conscientiously

我们还可以开始更认真地思考

about the social impact of the technology that we're developing.

关于正在发展的科技 造成的社会影响

To get the incoding movement started,

为了开启译码运动

I've launched the Algorithmic Justice League,

我发起了算法正义联盟

where anyone who cares about fairness can help fight the coded gaze.

在那里任何关心公平的人 可以出力来对抗 代码的凝视

On codedgaze.com, you can report bias,

在codedgaze.com网站 你可以举报偏见

request audits, become a tester

请求审核 成为测试者

and join the ongoing conversation,

以及加入正在进行的谈话

#codedgaze.

标签就是 代码的凝视

So I invite you to join me

我在此邀请各位加入我

in creating a world where technology works for all of us,

去创造一个让科技为我们 所有人服务的世界

not just some of us,

而不是只服务于部分人

a world where we value inclusion and center social change.

一个我们珍惜包容和 聚焦社会变革的世界

Thank you.

谢谢

(Applause)

（掌声）

But I have one question:

不过我还有一个问题

Will you join me in the fight?

你会与我并肩战斗吗

(Laughter)

（笑声）

(Applause)

（掌声）