connecting the notorious privacy incident

我想告訴各位一則故事

involving Adam and Eve,

是著名的亞當和夏娃隱私事件

and the remarkable shift in the boundaries

是著名的亞當和夏娃隱私事件

between public and private which has occurred

以及這10年來, 在公眾與個人之間分界的重大改變

in the past 10 years.

以及這10年來, 在公眾與個人之間分界的重大改變

You know the incident.

以及這10年來, 在公眾與個人之間分界的重大改變

Adam and Eve one day in the Garden of Eden

你知道這起事件

realize they are naked.

有一天亞當和夏娃在伊甸園

They freak out.

發現他們都沒穿衣服

And the rest is history.

他們嚇壞了

Nowadays, Adam and Eve

其餘的部分你們都知道了

would probably act differently.

換做是現在的話, 亞當和夏娃

[@Adam Last nite was a blast! loved dat apple LOL]

可能會有不同的反應

[@Eve yep.. babe, know what happened to my pants tho?]

(twitter)@亞當 昨晚的表現真是精采! 愛死那顆蘋果了

We do reveal so much more information

(twitter)@夏娃 寶貝你知道我的褲子怎麼了嗎?

about ourselves online than ever before,

在網路上, 我們都比從前透露出更多關於自己的訊息

and so much information about us

在網路上, 我們都比從前透露出更多關於自己的訊息

is being collected by organizations.

而這些有關我們的訊息

Now there is much to gain and benefit

正被許多政府機構收集起來

from this massive analysis of personal information,

現在可以從分析個人資訊或是巨量資料中得到許多利益

or big data,

現在可以從分析個人資訊或是巨量資料中得到許多利益

but there are also complex tradeoffs that come

現在可以從分析個人資訊或是巨量資料中得到許多利益

from giving away our privacy.

但是在捨棄隱私權的同時, 也伴隨著複雜的得失交換

And my story is about these tradeoffs.

但是在捨棄隱私權的同時, 也伴隨著複雜的得失交換

We start with an observation which, in my mind,

我要講的故事是有關這些得失交換

has become clearer and clearer in the past few years,

我們先從觀察開始

that any personal information

在我看來, 過去幾年中, 這個情況已經變得越來越明確

can become sensitive information.

任何個人資訊

Back in the year 2000, about 100 billion photos

都能變成敏感的訊息

were shot worldwide,

回朔到西元2000年的時候,

but only a minuscule proportion of them

全世界上所有人約拍出1000億張照片

were actually uploaded online.

但是只有極少數的照片

In 2010, only on Facebook, in a single month,

被上傳到網路上

2.5 billion photos were uploaded,

在2010年 光是一個月, Facebook用戶就上傳25億張照片

most of them identified.

在2010年時 光是一個月, Facebook用戶就上傳25億張照片

In the same span of time,

而多數照片上的人都可以被辨識出來

computers' ability to recognize people in photos

在這段時間裡

improved by three orders of magnitude.

辨認照片內人物的電腦運算能力

What happens when you combine

也加快了1000倍

these technologies together:

如果將這些技術結合起來

increasing availability of facial data;

會發生什麼事?

improving facial recognizing ability by computers;

取得了更多的臉部資料

but also cloud computing,

改善了電腦臉部辨識的能力

which gives anyone in this theater

還有雲端運算

the kind of computational power

這會給與現場任何一個人

which a few years ago was only the domain

一種運算能力

of three-letter agencies;

而這種運算能力在幾年前只專屬於

and ubiquitous computing,

那些政府機構

which allows my phone, which is not a supercomputer,

這種普及的運算能力

to connect to the Internet

能讓我的普通手機

and do there hundreds of thousands

連上網際網路

of face metrics in a few seconds?

在幾秒之內進行數十萬次的人臉辨識

Well, we conjecture that the result

在幾秒之內進行數十萬次的人臉辨識

of this combination of technologies

我們推測這些

will be a radical change in our very notions

技術結合的結果

of privacy and anonymity.

會顛覆我們

To test that, we did an experiment

對於隱私權與匿名性最初的想法

on Carnegie Mellon University campus.

為了進行測試 我們做了一項實驗

We asked students who were walking by

在卡內基美隆大學的校園裡

to participate in a study,

我們找路過的學生

and we took a shot with a webcam,

來參與這項研究

and we asked them to fill out a survey on a laptop.

我們拿視訊攝影機拍照

While they were filling out the survey,

請他們用筆電填寫問卷調查

we uploaded their shot to a cloud-computing cluster,

他們在填寫問卷的時候

and we started using a facial recognizer

上傳他們的照片到一個雲端運算群組

to match that shot to a database

使用一個臉部辨識系統

of some hundreds of thousands of images

將這組照片拿去與

which we had downloaded from Facebook profiles.

一個約有數十萬張圖像的資料庫比對

By the time the subject reached the last page

這些圖像是我們從Facebook的個人簡介下載下來的

on the survey, the page had been dynamically updated

等到受測者填寫到問卷最後一頁的時候

with the 10 best matching photos

畫面會更新成辨識器找出的10張最相符的照片

which the recognizer had found,

畫面會更新成辨識器找出的10張最相符的照片

and we asked the subjects to indicate

畫面會更新成辨識器找出的10張最相符的照片

whether he or she found themselves in the photo.

我們要求受測者指出

Do you see the subject?

是否有在這些照片中找到他們自己

Well, the computer did, and in fact did so

你有看到這名受測者嗎?

for one out of three subjects.

是的, 電腦有找到

So essentially, we can start from an anonymous face,

三人之中就有一人被找到

offline or online, and we can use facial recognition

基本上 我們能夠從一張不知名的臉開始,

to give a name to that anonymous face

不管是離線或在線 我們都能利用臉部辨識

thanks to social media data.

讓一張不知名的臉找到它的名字

But a few years back, we did something else.

這都是拜社群媒體資料庫所賜

We started from social media data,

但是幾年前 我們又做其他的事情

we combined it statistically with data

我們從社群媒體開始著手-

from U.S. government social security,

我們將它與美國社會安全局的資料做結合

and we ended up predicting social security numbers,

我們將它與美國社會安全局的資料做結合

which in the United States

我們可以猜出個人的社會安全號碼

are extremely sensitive information.

這在美國是一項極度敏感的個人資訊

Do you see where I'm going with this?

這在美國是一項極度敏感的個人資訊

So if you combine the two studies together,

你知道我在講什麼嗎?

then the question becomes,

所以如果你們將兩種研究結果加在一起,

can you start from a face and,

那這個問題就會變成

using facial recognition, find a name

你能從一張臉開始

and publicly available information

利用臉部辨識技術 找到他的名字

about that name and that person,

找到這個人的公開資訊

and from that publicly available information

找到這個人的公開資訊

infer non-publicly available information,

再從公開資訊

much more sensitive ones

推測出那些更加敏感的非公開資訊

which you link back to the face?

推測出那些更加敏感的非公開資訊

And the answer is, yes, we can, and we did.

然後你再回想起這張臉嗎?

Of course, the accuracy keeps getting worse.

答案是可以的, 而且我們也做到了

[27% of subjects' first 5 SSN digits identified (with 4 attempts)]

當然, 準確度還不是很好

But in fact, we even decided to develop an iPhone app

在四次嘗試中, 可以辨識出27%受測者的社會安全號碼前五碼

which uses the phone's internal camera

但事實上 我們甚至決定做一個 iPhone app

to take a shot of a subject

利用手機內建像機

and then upload it to a cloud

幫受測者拍一張照片

and then do what I just described to you in real time:

然後上傳至雲端網路

looking for a match, finding public information,

接下來馬上就像我對大家描述的一樣

trying to infer sensitive information,

即時找出相符的臉, 找出公開資訊

and then sending back to the phone

試著推斷敏感的私人資訊

so that it is overlaid on the face of the subject,

然後傳回手機

an example of augmented reality,

這些資訊會顯示在受測者的臉部照片旁

probably a creepy example of augmented reality.

這是一個擴增實境的例子

In fact, we didn't develop the app to make it available,

也許是一個會令人毛骨悚然的擴增實境案例

just as a proof of concept.

事實上我們並沒有讓這個app上市

In fact, take these technologies

只是做為一種觀念的證明

and push them to their logical extreme.

事實上, 利用這些科技到極致的時候

Imagine a future in which strangers around you

事實上, 利用這些科技到極致的時候

will look at you through their Google Glasses

想像一下未來, 你身旁的陌生人

or, one day, their contact lenses,

能透過Google眼鏡來看你

and use seven or eight data points about you

或者有一天 隱形眼鏡也能做到同樣的事情

to infer anything else

使用七或八個有關於你的資訊

which may be known about you.

去推測其他

What will this future without secrets look like?

可能與你相關的事

And should we care?

沒有秘密的未來會是什麼樣子?

We may like to believe

我們應該關心這個嗎?

that the future with so much wealth of data

我們可能比較願意去相信

would be a future with no more biases,

一個有這麼多數據資料的未來

but in fact, having so much information

會是一個沒有偏差的未來

doesn't mean that we will make decisions

但是, 事實上, 擁有這麼多資訊

which are more objective.

不表示我們能夠做出

In another experiment, we presented to our subjects

更客觀的決定

information about a potential job candidate.

在另一個實驗裡 我們把求職者的資訊給受測者看

We included in this information some references

在另一個實驗裡 我們把求職者的資訊給受測者看

to some funny, absolutely legal,

我們的資料含括

but perhaps slightly embarrassing information

關於一些有趣, 絕對合法

that the subject had posted online.

但也許稍微有點尷尬的訊息

Now interestingly, among our subjects,

這些都是受測者張貼在網路上的資訊

some had posted comparable information,

有趣的是 我們實驗的對象中

and some had not.

有些人也發表了類似的訊息

Which group do you think

但有些人則沒有

was more likely to judge harshly our subject?

你認為哪一組人

Paradoxically, it was the group

比較可能嚴厲批評我們的受測者?

who had posted similar information,

答案出乎意料的是

an example of moral dissonance.

那些發表類似訊息的人

Now you may be thinking,

這也是種道德觀念不一致的例子

this does not apply to me,

現在你可能正在想

because I have nothing to hide.

這對我來說沒用

But in fact, privacy is not about

因為我沒有什麼要藏的東西

having something negative to hide.

但事實上 隱私不只是

Imagine that you are the H.R. director

有什麼不好的東西要藏起來而已

of a certain organization, and you receive résumés,

想像你是某個組織的人事主管

and you decide to find more information about the candidates.

你收到應徵者寄來的履歷

Therefore, you Google their names

你決定要找出更多該名應徵者的訊息

and in a certain universe,

因此 你就在google上搜尋他們的名字

you find this information.

在特定時空

Or in a parallel universe, you find this information.

你可以找到這筆資訊

Do you think that you would be equally likely

或是在平行時空 你找到這筆資訊

to call either candidate for an interview?

你認為你會同樣的

If you think so, then you are not

打電話通知應徵者前來面試嗎?

like the U.S. employers who are, in fact,

如果你這麼認為,

part of our experiment, meaning we did exactly that.

那你就不像美國雇主

We created Facebook profiles, manipulating traits,

事實上, 他們也在我們的實驗當中

then we started sending out résumés to companies in the U.S.,

我們創造了一些Facebook個人簡介,

and we detected, we monitored,

然後寄送履歷到美國各家公司

whether they were searching for our candidates,

然後我們偵查、監控

and whether they were acting on the information

看是否他們正在上網搜尋我們的應徵者

they found on social media. And they were.

並且依照這些社群媒體上找到的資訊做事.

Discrimination was happening through social media

他們真的這麼作.

for equally skilled candidates.

透過社群媒體, 對技能相當的應徵者們來說

Now marketers like us to believe

也會發生不公平待遇的事情

that all information about us will always

現在行銷的人想讓我們相信

be used in a manner which is in our favor.

所有關於我們的個人資訊都會

But think again. Why should that be always the case?

用在對我們有利的面向

In a movie which came out a few years ago,

但是再想想, 真的會這樣嗎?

"Minority Report," a famous scene

幾年前上映的一部電影

had Tom Cruise walk in a mall

「關鍵報告」裡一個著名的場景

and holographic personalized advertising

就是湯姆克‧魯斯走進一間賣場

would appear around him.

有一個個人化的雷射投影廣告

Now, that movie is set in 2054,

出現在他旁邊

about 40 years from now,

那部電影的時空背景設定於2054年

and as exciting as that technology looks,

從現在算起 大約是40年之後

it already vastly underestimates

那種技術看起來很精彩

the amount of information that organizations

它已經大大低估

can gather about you, and how they can use it

各組織能夠匯集起有關你個人的資料量

to influence you in a way that you will not even detect.

與他們是如何運用這些資料

So as an example, this is another experiment

以某一個你無法查覺的方式, 對你造成影響

actually we are running, not yet completed.

還有一個例子 這是另一項實驗

Imagine that an organization has access

是我們正在進行中的實驗, 還沒有完成

to your list of Facebook friends,

想像一個組織能夠進入

and through some kind of algorithm

你的Facebook好友清單

they can detect the two friends that you like the most.

透過某種運算規則

And then they create, in real time,

他們可以偵測到你最喜歡的兩個好友

a facial composite of these two friends.

然後他們就能即時創造出

Now studies prior to ours have shown that people

由這兩個好友所組成的臉部合成照

don't recognize any longer even themselves

在我們之前有研究已經顯示

in facial composites, but they react

人們在看臉部合成照, 連他們自己都認不出來

to those composites in a positive manner.

人們在看臉部合成照, 連他們自己都認不出來

So next time you are looking for a certain product,

但是他們對那些合成照有正面評價

and there is an ad suggesting you to buy it,

所以下次你在找某項產品

it will not be just a standard spokesperson.

此時有一個建議購買的廣告

It will be one of your friends,

廣告將不會是一個固定的代言人

and you will not even know that this is happening.

他很可能是你其中一位朋友

Now the problem is that

你甚至不知道這種事正發生在你的生活中

the current policy mechanisms we have

現在問題就是

to protect ourselves from the abuses of personal information

目前政策機制是我們必須

are like bringing a knife to a gunfight.

保護我們自己免於個人資料遭到濫用

One of these mechanisms is transparency,

這就像是以卵擊石

telling people what you are going to do with their data.

其中一項機制就是資訊透明化

And in principle, that's a very good thing.

你必須告訴人們你想拿他們資料做什麼

It's necessary, but it is not sufficient.

原則上 這是一件非常好的事情

Transparency can be misdirected.

這是應該的, 但是這麼做還不夠

You can tell people what you are going to do,

資訊透明化的方向可能會被誤導

and then you still nudge them to disclose

你可以告訴大家你想做什麼

arbitrary amounts of personal information.

然後你促使他人揭露

So in yet another experiment, this one with students,

或多或少的個人資訊

we asked them to provide information

在另一項實驗中, 實驗對象是學生

about their campus behavior,

我們要求他們提供個人資訊

including pretty sensitive questions, such as this one.

關於他們在學校裡做的事

[Have you ever cheated in an exam?]

包括一些相當敏感的問題 就像這一個

Now to one group of subjects, we told them,

在考試的時候 你有作弊過嗎?

"Only other students will see your answers."

對其中一組受測者, 我們告訴他們

To another group of subjects, we told them,

只有其他的同學會看到你的答案

"Students and faculty will see your answers."

對另一組受測者 我們告訴他們

Transparency. Notification. And sure enough, this worked,

所有學生和教職員都會看到你的答案

in the sense that the first group of subjects

透明化 告知. 這真的有用.

were much more likely to disclose than the second.

第一組受測者

It makes sense, right?

比第二組受測者更有可能公佈事實

But then we added the misdirection.

合理吧?

We repeated the experiment with the same two groups,

但是之後我們加入誤導手段

this time adding a delay

我們對相同兩組學生重覆進行實驗

between the time we told subjects

這次在不同的時間告訴受測者我們是如何使用他們的資料

how we would use their data

這次在不同的時間告訴受測者我們是如何使用他們的資料

and the time we actually started answering the questions.

這次在不同的時間告訴受測者我們是如何使用他們的資料

How long a delay do you think we had to add

現在我們知道了

in order to nullify the inhibitory effect

你認為我們必須要延遲多久時間

of knowing that faculty would see your answers?

為使這種抑制效應無效

Ten minutes?

而這種效應就是知道教職員會看見你的答案?

Five minutes?

10分鐘?

One minute?

5分鐘?

How about 15 seconds?

1分鐘?

Fifteen seconds were sufficient to have the two groups

15秒怎樣?

disclose the same amount of information,

15秒就足夠讓兩組人

as if the second group now no longer cares

透露出相同資訊量

for faculty reading their answers.

就好像第二組人現在不再關心教職員會看他們的答案

Now I have to admit that this talk so far

就好像第二組人現在不再關心教職員會看他們的答案

may sound exceedingly gloomy,

現在我必須承認目前為止我說的這些話

but that is not my point.

可能聽起來非常沉悶

In fact, I want to share with you the fact that

但我要說的不是這個

there are alternatives.

事實上 我想與大家分享的是

The way we are doing things now is not the only way