Del Harvey: The strangeness of scale at Twitter

翻译人员: Xiaoou Chen 校对人员: Keke Gu

My job at Twitter

我在推特的工作

is to ensure user trust,

就是去确保用户的信赖，

protect user rights and keep users safe,

保护用户之间的

both from each other

以及他们自身的

and, at times, from themselves.

权利和安全。

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

让我们讨论一下在推特，比例是什么样的。

Back in January 2009,

在2009年1月，

we saw more than two million new tweets each day

每天，在推特上我们可以看见

on the platform.

超过两百万条推特更新。

January 2014, more than 500 million.

2014年1月有超过五亿条。

We were seeing two million tweets

我们那时在六分钟之内

in less than six minutes.

就可以看见两百万条。

That's a 24,900-percent increase.

那是一个24,900％的增长。

Now, the vast majority of activity on Twitter

现在，推特上绝大多数的活动

puts no one in harm's way.

都没有伤害到任何人。

There's no risk involved.

不涉及任何风险。

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

我的工作就是铲除并防止这类事情的发生。

Sounds straightforward, right?

听起来简单明了，对吧？

You might even think it'd be easy,

你可能认为这件事很容易，

given that I just said the vast majority

因为我刚说过绝大多数

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

在推特上的行为都是无害的。

Why spend so much time

为什么花这么多时间

searching for potential calamities

在无害的行为中

in innocuous activities?

搜寻潜在的危机呢？

Given the scale that Twitter is at,

考虑推特的规模，

a one-in-a-million chance happens

百万分之一几率的可能，

500 times a day.

一天会发生五百次。

It's the same for other companies

对于其它公司来说，

dealing at this sort of scale.

他们面临的这个比例是一样的。

For us, edge cases,

对于我们，边缘案例

those rare situations that are unlikely to occur,

那些不常有，也不大可能发生的情况

are more like norms.

更像是家常便饭。

Say 99.999 percent of tweets

假设99.999%的推特

pose no risk to anyone.

对任何人无害。

There's no threat involved.

不涉及任何威胁。

Maybe people are documenting travel landmarks

人们可能在记录旅游胜地，

like Australia's Heart Reef,

比如澳大利亚心型礁，

or tweeting about a concert they're attending,

或者推文他们正在参加的演唱会，

or sharing pictures of cute baby animals.

或者分享可爱动物的图片。

After you take out that 99.999 percent,

在你剔除那99.999%之后，

that tiny percentage of tweets remaining

剩下的那丁点推文

works out to roughly

被计算出

150,000 per month.

每月约有15万条。

The sheer scale of what we're dealing with

我们所应付的这个庞大规模

makes for a challenge.

是一个挑战。

You know what else makes my role

你知道还有什么让我的职位

particularly challenging?

特别具有挑战性？

People do weird things.

人们做奇怪的事情。

(Laughter)

（笑声）

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

我必须弄明白他们在做什么，

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

为什么，以及涉及危险与否，

often without much in terms of context

而这通常是在我没有掌握

or background.

来龙去脉的情况下。

I'm going to show you some examples

我将要展示给你们几个例子，

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

是我在推特工作中遇到的---

these are all real examples —

这些都是真实的例子-

of situations that at first seemed cut and dried,

这些情况乍看似乎直接了当，

but the truth of the matter was something

但事情的真相

altogether different.

是截然不同的。

The details have been changed

例子的细节有所改动

to protect the innocent

是为了去保护那些无辜者

and sometimes the guilty.

有时也包括有过的那方。

We'll start off easy.

让我们从简单的开始。

["Yo bitch"]

【“呦，bitch”】（bitch有母狗，婊子，娘们等意思）

If you saw a Tweet that only said this,

如果你看到一条推文只有这一句话，

you might think to yourself,

你可能认为

"That looks like abuse."

”那看起来像是在谩骂。“

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

毕竟，你为什么会想收到这条信息呢，

"Yo, bitch."

“呦，婊子。”

Now, I try to stay relatively hip

现在，我试图与流行用语的

to the latest trends and memes,

最新的释义保持同步，

so I knew that "yo, bitch"

所以我知道“呦，婊子”

was also often a common greeting between friends,

有时候也是朋友之间常见的问候方式，

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

同时也是美剧《绝命毒师》中一个流行说法。

I will admit that I did not expect

我要承认，我没有想到

to encounter a fourth use case.

我会遇到这个词的第四种用法。

It turns out it is also used on Twitter

在推特上

when people are role-playing as dogs.

人们角色扮演狗的时候，也用这个词。

(Laughter)

（笑声）

And in fact, in that case,

所以，在那种情况下，

it's not only not abusive,

这不仅不是谩骂，

it's technically just an accurate greeting.

严格的来说，那就是一个准确的问候。

(Laughter)

（笑声）

So okay, determining whether or not

所以判断一些没有来龙去脉的东西

something is abusive without context,

是否出于恶意

definitely hard.

确实困难。

Let's look at spam.

让我们来看一下垃圾邮件。

Here's an example of an account engaged

这是一个参与传播

in classic spammer behavior,

常见垃圾邮件的账户，

sending the exact same message

它向数以千计的人

to thousands of people.

发送相同的信息。

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

虽然这是我用我的账号模仿的，

we see accounts doing this all the time.

但我们总可以看到有账户在传播这样的垃圾信息。

Seems pretty straightforward.

看起来非常直白简单。

We should just automatically suspend accounts

我们应该就自动暂停

engaging in this kind of behavior.

参与这种行为的账号。

Turns out there's some exceptions to that rule.

但结果中总有些例外情况。

Turns out that that message could also be a notification

那些信息也可能是公告提醒，

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

比如你想目睹国际空间站略过你上空的情形

because you wanted to go outside

而登记了这个信息。

and see if you could see it.

希望可以收到提醒，尝试目睹它。

You're not going to get that chance

如果我们错误地认为这是垃圾信息，

if we mistakenly suspend the account

并封了那个账号，

thinking it's spam.

你将失去目睹国际空间站略过上空的机会。

Okay. Let's make the stakes higher.

让我们把赌注加高一些。

Back to my account,

再来看我的帐号，

again exhibiting classic behavior.

还是展现常见的行为。

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

这一次是发同样的信息和链接。

This is often indicative of something called phishing,

这通常意味着钓鱼式攻击，（注：一种网络诈骗的手段）

somebody trying to steal another person's account information

有人通过将一个人导向另一个网站

by directing them to another website.

去盗取其账户信息。

That's pretty clearly not a good thing.

很明显那不是什么好事。

We want to, and do, suspend accounts

我们想，也确实封了

engaging in that kind of behavior.

从事那种行为的账户。

So why are the stakes higher for this?

但为什么对这种行为的赌注更高呢？

Well, this could also be a bystander at a rally

这也可能是一个身处集会中的旁观者

who managed to record a video

录下了一段关于

of a police officer beating a non-violent protester

警察殴打一个无辜抗议者的视频

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

他想让全世界知道发生了什么。

We don't want to gamble

我们不想

on potentially silencing that crucial speech

在把那个关键演说通过分类为垃圾并暂停账号而可能导致的后果

by classifying it as spam and suspending it.

上做冒险。

That means we evaluate hundreds of parameters

那也就意味着，当我们观察账户行为的时候

when looking at account behaviors,

我们评估成百上千的因素

and even then, we can still get it wrong

即使这样，我们仍然会犯错，

and have to reevaluate.

并需要重新评价。

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

在这些挑战面前

it's crucial that I not only predict

关键在于我不仅要预测

but also design protections for the unexpected.

而且防御不可预测的事情。

And that's not just an issue for me,

那不单是我的问题，

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

或者推特的问题，这也是你们的问题。

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

对于任何在创建美好事物，

something that you think is going to be amazing

或者为他人造福的人来说

and will let people do awesome things.

都是一个问题。

So what do I do?

那么我能做些什么呢？

I pause and I think,

我停下并思考，

how could all of this

这些事能如何

go horribly wrong?

变得很糟糕的呢？

I visualize catastrophe.

我想象灾难。

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

但是那很难。好像有一种

inherent cognitive dissonance in doing that,

与生俱来的认知失调在作怪，

like when you're writing your wedding vows

就像你同时写结婚誓言

at the same time as your prenuptial agreement.

和婚前协议一样。

(Laughter)

（笑声）

But you still have to do it,

但你还是得去做，

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

特别是当你一天得处理5亿条推文时。

What do I mean by "visualize catastrophe?"

我说的“想象灾难”是什么意思呢？

I try to think of how something as

我试想，像猫的照片一样

benign and innocuous as a picture of a cat

温和并无害的东西

could lead to death,

能如何导致死亡，

and what to do to prevent that.

并想办法去阻止其发生。

Which happens to be my next example.

这也是我的下一个例子。

This is my cat, Eli.

这是我的猫，伊莱。

We wanted to give users the ability

我们想要给予用户

to add photos to their tweets.

将图片加到他们推文的能力。

A picture is worth a thousand words.

一张图片胜过千言万语。

You only get 140 characters.

（一次推文）你只能输入140个字。

You add a photo to your tweet,

当你在推文里加一张图片时，

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

看看现在你发表的内容有多丰富。

There's all sorts of great things you can do

通过在推文里添加图片，

by adding a photo to a tweet.

你可以做各种各样神奇的事。

My job isn't to think of those.

我的工作不是去想那些事情，

It's to think of what could go wrong.

而是去想会发生什么问题。

How could this picture

这张图片能如何

lead to my death?

导致我的死亡呢？

Well, here's one possibility.

有一个可能性。

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

除了一只猫以外，这个图片里还有其它信息。

There's geodata.

那里有地理信息。

When you take a picture with your smartphone

当你用你的智能手机或数码相机

or digital camera,

照一张照片时，

there's a lot of additional information

很多额外的信息

saved along in that image.

也会随着照片保留下来。

In fact, this image also contains

事实上，这张照图片还

the equivalent of this,

指明了这个，

more specifically, this.

更加具体些，是这个。

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

没错，不大可能有人准备

to track me down and do me harm

根据我的猫的照片中数据

based upon image data associated

追踪我

with a picture I took of my cat,

并伤害我，

but I start by assuming the worst will happen.

但是我开始假设最坏的事情会发生。

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

这也是为什么当我们推出加载图片功能的时候，

we made the decision to strip that geodata out.

我们决定消除那些地理数据。

(Applause)

（掌声）

If I start by assuming the worst

如果我从假设最坏的事开始，

and work backwards,

并反向推理，

I can make sure that the protections we build

我可以确保我们所设置的保护

work for both expected

对于可预知与

and unexpected use cases.

不可预知的事件同时有效。

Given that I spend my days and nights

假设我日夜

imagining the worst that could happen,

想象可能发生的最坏的事情，

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

我的世界观有些阴郁并不令人惊奇。

(Laughter)

（笑声）

It's not.

但这并不是事实。

The vast majority of interactions I see --

我看到的绝大多数的（推特）互动

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

是积极的，相信我，我看过很多，

people reaching out to help

人们伸出援助之手，

or to connect or share information with each other.

或者相互连接或分享信息。

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

因为我们要应付安全风险，

for those of us tasked with keeping people safe,

我们承担着保证大众安全的责任，

we have to assume the worst will happen,

我们必须假设最坏的事情会发生，

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

因为对于我们来说百万分之一的几率

is pretty good odds.

是一个非常大的可能性。

Thank you.

谢谢

(Applause)

（掌声）