Del Harvey: The strangeness of scale at Twitter

譯者: Yamei Huang 審譯者: Adrienne Lin

My job at Twitter

我在推特的工作

is to ensure user trust,

就是確保使用者對推特的信任，

protect user rights and keep users safe,

以及保護使用者的權益及安全，

both from each other

不只是使用者之間

and, at times, from themselves.

有時是有關使用者本身的權益及安全。

Let's talk about what scale looks like at Twitter.

讓我們談談推特訊息的規模。

Back in January 2009,

2009 年 1 月，

we saw more than two million new tweets each day

每一天，我們在推特平台上

on the platform.

看到超過 200 萬條新訊息。

January 2014, more than 500 million.

2014 年 1 月，則有超過 5 億條訊息。

We were seeing two million tweets

六分鐘內，

in less than six minutes.

就有 200 萬條推文。

That's a 24,900-percent increase.

那是 24,900% 的成長。

Now, the vast majority of activity on Twitter

今天，絶大部分在推特上的活動

puts no one in harm's way.

不會傷害任何人。

There's no risk involved.

沒有任何風險。

My job is to root out and prevent activity that might.

我的工作則是根除 任何可能傷害他人權益的活動。

Sounds straightforward, right?

聽起來很簡單，對吧？

You might even think it'd be easy,

你或許會認為這個工作很簡單，

given that I just said the vast majority

尤其當我說，推特上絕大部分的動作

of activity on Twitter puts no one in harm's way.

並不會對任何人造成傷害。

Why spend so much time

那為什麼要花費這麼多時間

searching for potential calamities

在無害的網路活動中，

in innocuous activities?

尋找可能的危機？

Given the scale that Twitter is at,

以推特的規模來看，

a one-in-a-million chance happens

百萬分之一的機率，

500 times a day.

相當於一天會有 500 條 可能造成危害的訊息。

It's the same for other companies

這個訊息量，是其他公司

dealing at this sort of scale.

所要處理的訊息量相同

For us, edge cases,

對我們而言，那些稀少罕見，

those rare situations that are unlikely to occur,

不太可能發生的極端事件，

are more like norms.

有如家常便飯。

Say 99.999 percent of tweets

假設百分之 99.999% 的推文

pose no risk to anyone.

都不會傷害任何人，

There's no threat involved.

不涉及任何風險。

Maybe people are documenting travel landmarks

也許大家只是在記錄旅遊景點，

like Australia's Heart Reef,

像是澳洲的心形礁，

or tweeting about a concert they're attending,

或是傳些關於他們正在參加的演唱會，

or sharing pictures of cute baby animals.

或者是分享一些可愛小動物的照片。

After you take out that 99.999 percent,

當你除去那 99.999% 的機率，

that tiny percentage of tweets remaining

剩下極微小的百分比

works out to roughly

粗估下來

150,000 per month.

每月大約有十五萬條訊息。

The sheer scale of what we're dealing with

管理這麼龐大的規模，

makes for a challenge.

是個挑戰。

You know what else makes my role

你知道還有什麼

particularly challenging?

讓我的工作更具挑戰性的嗎？

People do weird things.

人會做些奇怪的事。

(Laughter)

(笑聲)

And I have to figure out what they're doing,

而我則必須搞清楚他們在做什麼，

why, and whether or not there's risk involved,

動機是什麼，還有是否有危險性，

often without much in terms of context

且通常是在沒有資料

or background.

或背景的情況下就要去搞清楚。

I'm going to show you some examples

讓我舉幾個我在推特工作時

that I've run into during my time at Twitter --

遇到的例子，

these are all real examples —

這些全都是真實的案例，

of situations that at first seemed cut and dried,

一些原先看來簡單明瞭的情況，

but the truth of the matter was something

但事情的真相

altogether different.

又是截然不同。

The details have been changed

有些細節已被更改，

to protect the innocent

是為了保護無辜的人，

and sometimes the guilty.

有時也包括罪犯。

We'll start off easy.

我們從簡單的開始。

["Yo bitch"]

[嘿 賤女人]

If you saw a Tweet that only said this,

當你在推特上看到這句話，

you might think to yourself,

你可能會認為：

"That looks like abuse."

「那是一種辱罵」。

After all, why would you want to receive the message,

畢竟，誰會希望收到這樣的訊息：

"Yo, bitch."

「嘿，賤女人。」

Now, I try to stay relatively hip

現在，我試著跟上趨勢

to the latest trends and memes,

及最新流行用語，

so I knew that "yo, bitch"

所以我知道「嘿，賤女人」

was also often a common greeting between friends,

也常被用作朋友間的招呼用語

as well as being a popular "Breaking Bad" reference.

是來自於《絕命毒師》的說法。

I will admit that I did not expect

我得承認我沒有想到

to encounter a fourth use case.

這句話會有第四種用法。

It turns out it is also used on Twitter

原來在推特上，扮成狗的人

when people are role-playing as dogs.

也會用這個詞。

(Laughter)

（笑聲）

And in fact, in that case,

事實上，在這個情況下，

it's not only not abusive,

不止沒有辱罵的意味，

it's technically just an accurate greeting.

嚴格說來，這是一個準確的問候用語。

(Laughter)

（笑聲）

So okay, determining whether or not

所以，一條沒有來龍去脈的訊息

something is abusive without context,

要去判定這個訊息是否有辱罵的意味，

definitely hard.

絕對是很困難的。

Let's look at spam.

我們來看看垃圾訊息。

Here's an example of an account engaged

這是使用者傳送垃圾訊息

in classic spammer behavior,

的典型例子，

sending the exact same message

一直不斷地傳送相同的訊息

to thousands of people.

給上千個人。

While this is a mockup I put together using my account,

這是我用自己帳號作出的模擬範例，

we see accounts doing this all the time.

我們總可以看到使用者傳送這樣的訊息。

Seems pretty straightforward.

看起來相當簡單明瞭。

We should just automatically suspend accounts

我們應該自動封鎖

engaging in this kind of behavior.

涉及這種行為的帳號。

Turns out there's some exceptions to that rule.

結果總有些例外。

Turns out that that message could also be a notification

這些訊息，也有可能是通知

you signed up for that the International Space Station is passing overhead

你登記參加國際太空站經過你上空的活動，

because you wanted to go outside

你希望收到通知，即時走到戶外

and see if you could see it.

可以親自目睹。

You're not going to get that chance

你絶不會因為

if we mistakenly suspend the account

誤認為這是垃圾訊息

thinking it's spam.

而停用這個帳號的情況發生。

Okay. Let's make the stakes higher.

好。讓我們再把風險的層級提高。

Back to my account,

再來看我的帳號，

again exhibiting classic behavior.

在推特上展示特定的行為。

This time it's sending the same message and link.

這次是在持特上傳送相同的訊息和連結

This is often indicative of something called phishing,

這通常是一種網路釣魚，

somebody trying to steal another person's account information

有人試著去引導他人到另一個網站

by directing them to another website.

然後盜用他的帳號

That's pretty clearly not a good thing.

很明顯這不是一件好事。

We want to, and do, suspend accounts

我們要，而且必須去阻止

engaging in that kind of behavior.

可疑的帳號去做這樣的行為。

So why are the stakes higher for this?

但是，為何這麼做風險更高？

Well, this could also be a bystander at a rally

這像是遊行人潮當中的旁觀者

who managed to record a video

拿著攝影機，對著

of a police officer beating a non-violent protester

警察動手打一個 無暴力行為的抗議者攝影，

who's trying to let the world know what's happening.

好讓全世界的人知道此事。

We don't want to gamble

我們不想冒這個險

on potentially silencing that crucial speech

把有可能很重要的訊息

by classifying it as spam and suspending it.

歸類為垃圾訊息，然後停用帳號。

That means we evaluate hundreds of parameters

那意味著，當我們在觀察使用者行為時

when looking at account behaviors,

我們憑估成千上百個因素，

and even then, we can still get it wrong

即使是這麼做了，百密仍有一疏，

and have to reevaluate.

必須再重新評估這些訊息。

Now, given the sorts of challenges I'm up against,

現在，面臨各式各樣的挑戰，

it's crucial that I not only predict

重要的是，不但要去預測可能發生的事，

but also design protections for the unexpected.

而且要對可能發生的事， 設計一套因應的保護措施。

And that's not just an issue for me,

這不僅事關我和推特，

or for Twitter, it's an issue for you.

這也關係到你。

It's an issue for anybody who's building or creating

關係到任何想創造美好事物，

something that you think is going to be amazing

and will let people do awesome things.

以及想要讓他人也一起 做美好事物的推特使用者。

So what do I do?

所以我要怎麼做呢？

I pause and I think,

我一再思考這問題

how could all of this

這些事情

go horribly wrong?

到底怎麼會出錯？

I visualize catastrophe.

我想像發生災難的情形。

And that's hard. There's a sort of

這很困難，

inherent cognitive dissonance in doing that,

因為這麼做， 有點像是內在認知不協調，

like when you're writing your wedding vows

就像是寫結婚誓言時，

at the same time as your prenuptial agreement.

同時也寫婚前協議書。

(Laughter)

（笑聲）

But you still have to do it,

但還是必須要去做，

particularly if you're marrying 500 million tweets per day.

特別是每天要處理 5 億條推文。

What do I mean by "visualize catastrophe?"

我所說的「想像災難」是什麼意思呢？

I try to think of how something as

我試著去想像，

benign and innocuous as a picture of a cat

像是一張無害的貓咪照片

could lead to death,

為何可能導致死亡，

and what to do to prevent that.

以及如何避免這種事情發生。

Which happens to be my next example.

正是接下來我要說的例子。

This is my cat, Eli.

這隻是我的貓，叫伊萊。

We wanted to give users the ability

我們盡可能讓推特的使用者

to add photos to their tweets.

在推特上傳送圖片，

A picture is worth a thousand words.

一張圖勝過千言萬語，

You only get 140 characters.

而一次推文只能傳送 140 個字。

You add a photo to your tweet,

你在推文加入圖片，

look at how much more content you've got now.

你會發現推文的內容更加豐富。

There's all sorts of great things you can do

藉由推特加入圖片的功能，

by adding a photo to a tweet.

你可以做各種麼美妙的事。

My job isn't to think of those.

我的工作不是去想這些事情。

It's to think of what could go wrong.

而是去想事情可能會出什麼差錯。

How could this picture

這張圖片

lead to my death?

如何導致我死亡？

Well, here's one possibility.

有一個可能性。

There's more in that picture than just a cat.

這張圖的資訊不只是一隻貓。

There's geodata.

還有地理資訊在裡頭。

When you take a picture with your smartphone

當你以智慧型手機

or digital camera,

或數位相機拍照，

there's a lot of additional information

會有許多額外的資訊

saved along in that image.

儲存在照片裡。

In fact, this image also contains

事實上，這張照片還包含

the equivalent of this,

相當於這個的資訊，

more specifically, this.

更具體地說是這個。

Sure, it's not likely that someone's going to try

當然，不太可能有人嘗試

to track me down and do me harm

根據這張貓照片的相關資訊

based upon image data associated

追蹤我以及傷害我。

with a picture I took of my cat,

but I start by assuming the worst will happen.

但我一開始就要假設 最壞的情況一定會發生，

That's why, when we launched photos on Twitter,

這就是為什麼我們 在開放上傳照片到推特時，

we made the decision to strip that geodata out.

就決定把照片裡的地理資訊全刪掉。

(Applause)

（掌聲）

If I start by assuming the worst

如果一開始，我就假設 可能發生最壞的情況，

and work backwards,

然後再往前倒推，

I can make sure that the protections we build

我可以確定我們建立的保護制度，

work for both expected

可以應付意料中

and unexpected use cases.

以及意料外的事件。

Given that I spend my days and nights

我日夜地

imagining the worst that could happen,

想像發生最壞情況的情形，

it wouldn't be surprising if my worldview was gloomy.

如果因此造成我憂鬱的世界觀， 也不會令人感到意外。

(Laughter)

（笶聲）

It's not.

其實並非如此。

The vast majority of interactions I see --

我看到的絶大部份互動，

and I see a lot, believe me -- are positive,

我看了很多，相信我， 它們都是正面的。

people reaching out to help

人們伸出援手互相幫忙，

or to connect or share information with each other.

彼此互相連絡或分享資訊。

It's just that for those of us dealing with scale,

只是我們要處理龐大的資訊量，

for those of us tasked with keeping people safe,

承擔保護使用者安全的責任，

we have to assume the worst will happen,

所以必須假設將發生最壞的情況，

because for us, a one-in-a-million chance

對我們來說，百萬分之一的可能性

is pretty good odds.

是相當高的機率。

Thank you.

謝謝。

(Applause)

（掌聲）