How I'm fighting bias in algorithms | Joy Buolamwini

譯者: Suzie Tang 審譯者: Helen Chang

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

你好 我叫玖伊 是個寫媒體程式的詩人

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

我的使命是

終止一個隱形力量的崛起

a force that I called "the coded gaze,"

我稱這種力量為「數碼凝視」

my term for algorithmic bias.

是我替偏差演算法取的名稱

Algorithmic bias, like human bias, results in unfairness.

偏差的演算法跟人的偏見一樣

會導致不公平的結果

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

然而演算法更像病毒

它傳播的偏見

大量而迅速

at a rapid pace.

Algorithmic bias can also lead to exclusionary experiences

演算法偏差讓人 體驗到什麼叫做被排擠

and discriminatory practices.

也會導致差別對待

Let me show you what I mean.

讓我告訴你我的意思

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

嗨 相機 我有一張臉

Can you see my face?

你能看見我的臉嗎？

No-glasses face?

不戴眼鏡呢？

You can see her face.

你看得見她啊

What about my face?

那麼我的臉呢？

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

戴上面具 你看得見戴上面具嗎？

Joy Buolamwini: So how did this happen?

到底是怎麽回事？

Why am I sitting in front of a computer

我為什麽要坐在電腦前

in a white mask,

戴著白色面具

trying to be detected by a cheap webcam?

好讓這台廉價的攝影機能看得見我

Well, when I'm not fighting the coded gaze

如果我沒有忙著對抗數碼凝視

as a poet of code,

當個媒體程式詩人

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

我就是麻省理工媒體實驗室的研究生

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

我在那裡從事一些稀奇古怪的計劃

including the Aspire Mirror,

包括照妖鏡

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

照妖鏡計劃

讓我能把數位面具投射在自己臉上

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

早上起來如果我需要強大的力量

I could put on a lion.

我就投上一個獅子面具

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

如果我缺乏鬥志

我就放一段名人名言

So I used generic facial recognition software

因為我使用一般的臉部辨識軟體

to build the system,

來測試這個系統

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

結果竟然發現

電腦無法偵測到我

除非我戴上白色面具

很不幸我之前就碰過這種問題

Unfortunately, I've run into this issue before.

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

先前我在喬治亞理工學院

攻讀電腦科學學士學位時

I used to work on social robots,

我研究社交機器人

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

其中的一個實驗

就是和機器人玩躲貓貓

a simple turn-taking game

這個簡單的互動遊戲

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

讓對手先遮住臉再放開

同時要說 peek-a-boo

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

問題是如果看不到對方

遊戲就玩不下去了

and my robot couldn't see me.

我的機器人就是看不到我

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

最後我只好借我室友的臉來完成

submitted the assignment,

做完實驗時我想

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

總有一天會有別人解決這個問題

Not too long after,

不久之後

I was in Hong Kong for an entrepreneurship competition.

我去香港參加一個

業界舉辦的競技比賽

The organizers decided to take participants

主辦單位先帶每位參賽者

on a tour of local start-ups.

去參觀當地的新創市場

One of the start-ups had a social robot,

其中一項就是社交機器人

and they decided to do a demo.

當他們用社交機器人展示成果時

The demo worked on everybody until it got to me,

社交機器人對每個參賽者都有反應

and you can probably guess it.

直到遇到了我

接下來的情形你應該能想像

It couldn't detect my face.

社交機器人怎樣都偵測不到我的臉

I asked the developers what was going on,

我問軟體開發人員是怎麼一回事

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

才驚覺當年通用的

人臉辨識軟體

Halfway around the world,

竟然飄洋過海到了香港

I learned that algorithmic bias can travel as quickly

偏差的演算邏輯快速散播

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

只要從網路下載幾個檔案就搞定了

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

為什麼機器人就是看不見我的臉？

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

得先知道我們如何賦予機器視力

Computer vision uses machine learning techniques

電腦使用機器學習的技術

to do facial recognition.

來辨識人臉

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

你必須用許多實作測試來訓練他們

這是人臉這是人臉這是人臉

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

這不是人臉

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

一而再再而三你就能教機器人

辨識其他的人臉

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

但是如果實作測試不夠多樣化

any face that deviates too much from the established norm

當出現的人臉

與既定規範相去太遠時

will be harder to detect,

電腦就很難判斷了

which is what was happening to me.

我的親身經驗就是這樣

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

但別慌張 有好消息

Training sets don't just materialize out of nowhere.

實作測試並不是無中生有

We actually can create them.

事實上我們能夠建的

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

我們可以有一套更周詳的測試樣本

that reflect a richer portrait of humanity.

涵蓋人種的多樣性

Now you've seen in my examples

我的實驗說明了

how social robots

社交機器人

was how I found out about exclusion with algorithmic bias.

產生排他現象

因為偏差的演算邏輯

But algorithmic bias can also lead to discriminatory practices.

偏差的演算邏輯

也可能讓偏見成為一種習慣

Across the US,

美國各地的警方

police departments are starting to use facial recognition software

正開始使用這套人臉辨識軟體

in their crime-fighting arsenal.

來建立警方的打擊犯罪系統

Georgetown Law published a report

喬治城大學法律中心的報告指出

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

每兩個美國成年人就有一個人

也就是一億一千七百萬筆臉部資料

have their faces in facial recognition networks.

在美國警方這套系統裡

Police departments can currently look at these networks unregulated,

警方這套系統既缺乏規範

using algorithms that have not been audited for accuracy.

也缺乏正確合法的演算邏輯

Yet we know facial recognition is not fail proof,

你要知道人臉辨識並非萬無一失

and labeling faces consistently remains a challenge.

要一貫正確地標註人臉 往往不是那麼容易

You might have seen this on Facebook.

或許你在臉書上看過

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

朋友和我常覺得很好笑

看見有人標註朋友卻標錯了

mislabeled in our photos.

But misidentifying a suspected criminal is no laughing matter,

如果標錯的是犯人的臉呢

那就讓人笑不出來了

nor is breaching civil liberties.

侵害公民自由也同樣讓人笑不出來

Machine learning is being used for facial recognition,

不僅辨識人臉倚賴機器學習的技術

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

許多領域其實都要用到機器學習

In her book, "Weapons of Math Destruction,"

《大數據的傲慢與偏見》 這本書的作者

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

數據科學家凱西 歐尼爾

談到新 WMD 勢力的崛起

widespread, mysterious and destructive algorithms

WMD 是廣泛 神秘和具破壞性的算法

that are increasingly being used to make decisions

演算法漸漸取代我們做決定

that impact more aspects of our lives.

影響我們生活的更多層面

So who gets hired or fired?

例如誰升了官？誰丟了飯碗？

Do you get that loan? Do you get insurance?

你借到錢了嗎？你買保險了嗎？

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

你進入心目中理想的大學了嗎？

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

我們花同樣多的錢在同樣的平台上

purchased on the same platform?

買到同樣的產品嗎？

Law enforcement is also starting to use machine learning

警方也開始使用機器學習

for predictive policing.

來防範犯罪

Some judges use machine-generated risk scores to determine

法官根據電腦顯示的危險因子數據

how long an individual is going to spend in prison.

來決定一個人要在監獄待幾年

So we really have to think about these decisions.

我們得仔細想想這些判定

Are they fair?

它們真的公平嗎？

And we've seen that algorithmic bias

我們親眼看見偏差的演算邏輯

doesn't necessarily always lead to fair outcomes.

未必做出正確的判斷

So what can we do about it?

我們該怎麽辦呢？

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

我們要先確定程式碼是否具多樣性

and employ inclusive coding practices.

以及寫程式的過程是否周詳

It really starts with people.

事實上全都始於人

So who codes matters.

程式是誰寫的有關係

Are we creating full-spectrum teams with diverse individuals

寫程式的團隊是否由 多元的個體組成呢？

who can check each other's blind spots?

這樣才能互補並找出彼此的盲點

On the technical side, how we code matters.

從技術面而言 我們如何寫程式很重要

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

我們是否對公平這項要素

在系統開發階段就考量到呢？

And finally, why we code matters.

最後 我們為什麼寫程式也重要

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

我們使用計算創造工具 開啟了巨額財富之門

We now have the opportunity to unlock even greater equality

我們現在有機會實現更大的平等

if we make social change a priority

如果我們將社會變革作為優先事項

and not an afterthought.

而不是事後的想法

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

這裡有改革程式的三元素

Who codes matters,

程式是誰寫的重要

how we code matters

如何寫程式重要

and why we code matters.

以及為何寫程式重要

So to go towards incoding, we can start thinking about

要成功改革程式

我們可以先從建立能夠 找出偏差的分析平台開始

building platforms that can identify bias

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

作法是收集人們的親身經歷 像是我剛才分享的經歷

but also auditing existing software.

也檢視現存的軟體

We can also start to create more inclusive training sets.

我們可以著手建立 更具包容性的測試樣本

Imagine a "Selfies for Inclusion" campaign

想像「包容的自拍」活動

where you and I can help developers test and create

我們可以幫助開發人員測試和創建

more inclusive training sets.

更具包容性的測試樣本

And we can also start thinking more conscientiously

我們也要更自省

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

我們發展的科技帶給社會的衝擊

To get the incoding movement started,

為了著手程式改革

I've launched the Algorithmic Justice League,

我發起了「演算邏輯正義聯盟」

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

只要你贊同公平

就可以加入打擊數碼凝視的行列

On codedgaze.com, you can report bias,

只要上 codedgaze.com 網路

可以舉報你發現的偏差演算邏輯

request audits, become a tester

可以申請測試

可以成為受測者

and join the ongoing conversation,

也可以加入論壇

#codedgaze.

只要搜尋 #codedgaze

So I invite you to join me

我在此邀請大家加入我的行列

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

創造一個技術適用於 我們所有人的世界

not just some of us,

而不是只適用於某些人

a world where we value inclusion and center social change.

一個重視包容性 和以社會變革為中心的世界

Thank you.

謝謝

(Applause)

（掌聲）

But I have one question:

我還有一個問題

Will you join me in the fight?

你要和我並肩作戰嗎？

(Laughter)

（笑聲）

(Applause)

（掌聲）