How we can protect truth in the age of misinformation | Sinan Aral

Translator: Ivana Korom Reviewer: Krystian Aparta

譯者: Harper Chang 審譯者: Helen Chang

So, on April 23 of 2013,

2013 年 4 月 23 日，

the Associated Press put out the following tweet on Twitter.

美聯社在推特上發推文稱，

It said, "Breaking news:

「突發新聞：

Two explosions at the White House

白宮發生兩起爆炸，

and Barack Obama has been injured."

總統奧巴馬受傷。」

This tweet was retweeted 4,000 times in less than five minutes,

不到五分鐘內， 這篇推文被轉發了四千多次，

and it went viral thereafter.

接着病毒式擴散。

Now, this tweet wasn't real news put out by the Associated Press.

然而，這則推文並不是 由美聯社發佈的真實新聞，

In fact it was false news, or fake news,

這則錯誤的新聞， 或者說，假新聞，

that was propagated by Syrian hackers

是敘利亞的黑客

that had infiltrated the Associated Press Twitter handle.

在入侵美聯社的推特帳號後發佈的。

Their purpose was to disrupt society, but they disrupted much more.

他們的目的是擾亂社會， 但實際破壞遠大於此。

Because automated trading algorithms

因為自動交易算法

immediately seized on the sentiment on this tweet,

立即對這條推文進行情感分析，

and began trading based on the potential

並開始根據美國總統

that the president of the United States had been injured or killed

在爆炸中受傷或喪生的可能性

in this explosion.

進行自動交易。

And as they started tweeting,

推文被瘋狂轉發的同時，

they immediately sent the stock market crashing,

股市立刻隨之崩盤，

wiping out 140 billion dollars in equity value in a single day.

一天之內蒸發了 1400 億美金市值。

Robert Mueller, special counsel prosecutor in the United States,

美國特別檢察官羅伯特·穆勒

issued indictments against three Russian companies

曾起訴三家俄羅斯公司

and 13 Russian individuals

及十三名俄羅斯公民，

on a conspiracy to defraud the United States

指控他們合謀擾亂美國，

by meddling in the 2016 presidential election.

干涉 2016 年的總統大選。

And what this indictment tells as a story

這項指控針對的事件

is the story of the Internet Research Agency,

與俄羅斯網路水軍有關， (Internet Research Agency)

the shadowy arm of the Kremlin on social media.

那是個俄羅斯政府用來 操控社交網路的機構。

During the presidential election alone,

僅在（美國）總統大選期間，

the Internet Agency's efforts

這個網路水軍的功夫

reached 126 million people on Facebook in the United States,

就觸及了 1.26 億美國臉書用戶，

issued three million individual tweets

發佈了 3 百萬條推文

and 43 hours' worth of YouTube content.

和 43 小時的 YouTube 內容，

All of which was fake --

全都是精心設計的虛假資訊，

misinformation designed to sow discord in the US presidential election.

用來干擾美國總統選舉。

A recent study by Oxford University

牛津大學最近的一項研究表明，

showed that in the recent Swedish elections,

在近期的瑞典大選中，

one third of all of the information spreading on social media

社交網路上有三分之一 關於選舉的資訊

about the election

was fake or misinformation.

是虛假資訊或是謠言。

In addition, these types of social-media misinformation campaigns

另外，這些社交網路上的虛假資訊，

can spread what has been called "genocidal propaganda,"

可以傳播「種族屠殺宣傳」，

for instance against the Rohingya in Burma,

比如，在緬甸煽動 對羅興亞人的種族仇恨，

triggering mob killings in India.

在印度引發暴民殺戮。

We studied fake news

我們在假新聞廣受關注前 就開始對其研究。

and began studying it before it was a popular term.

And we recently published the largest-ever longitudinal study

我們最近公佈了一項迄今最大的

of the spread of fake news online

針對網路傳播上假新聞的縱向研究，

on the cover of "Science" in March of this year.

登上了今年 3 月的《科學》封面。

We studied all of the verified true and false news stories

我們研究所有經過查核的

that ever spread on Twitter,

推特上傳播的真、假新聞，

from its inception in 2006 to 2017.

範圍從推特創立的 2006 年到 2017 年。

And when we studied this information,

我們研究的新聞樣本

we studied verified news stories

是明確的真/假新聞，

that were verified by six independent fact-checking organizations.

它們經過了 6 個獨立的 事實查核機構的驗證，

So we knew which stories were true

所以我們知道哪些是真新聞，

and which stories were false.

哪些是假新聞。

We can measure their diffusion,

我們測量它們的傳播程度，

the speed of their diffusion,

比如傳播速度、

the depth and breadth of their diffusion,

傳播範圍、

how many people become entangled in this information cascade and so on.

被假新聞迷惑的人數等等。

And what we did in this paper

在這項研究中，

was we compared the spread of true news to the spread of false news.

我們比對了真、假新聞的傳播程度，

And here's what we found.

這是研究結果。

We found that false news diffused further, faster, deeper

我們發現，假新聞較真新聞 傳播得更遠、更快、更深、更廣，

and more broadly than the truth

in every category of information that we studied,

在每類新聞中都是如此，

sometimes by an order of magnitude.

有時甚至相差一個數量級。 （意即：數十倍之差）

And in fact, false political news was the most viral.

在假新聞中，虛假政治新聞 傳播程度最嚴重，

It diffused further, faster, deeper and more broadly

它比其他假新聞傳播得 更遠、更快、更深、更廣。

than any other type of false news.

When we saw this,

看到這個結果，

we were at once worried but also curious.

我們擔心、同時好奇著，

Why?

為什麼？

Why does false news travel so much further, faster, deeper

為什麼假新聞比真新聞傳播得 更遠、更快、更深、更廣？

and more broadly than the truth?

The first hypothesis that we came up with was,

我們首先想到的假設是，

"Well, maybe people who spread false news have more followers or follow more people,

也許傳播假新聞的人有更多的關注者，

或是關注了更多人，發推更頻繁，

or tweet more often,

or maybe they're more often 'verified' users of Twitter, with more credibility,

也許他們在公衆眼裡更「可信」，

or maybe they've been on Twitter longer."

或是他們更早開始使用推特。

So we checked each one of these in turn.

於是我們一一驗證，

And what we found was exactly the opposite.

得出的結果卻正好相反。

False-news spreaders had fewer followers,

假新聞傳播帳號有更少的關注者，

followed fewer people, were less active,

關注更少的用戶，

less often "verified"

可信度更低，

and had been on Twitter for a shorter period of time.

使用推特的時間更短。

And yet,

然而，

false news was 70 percent more likely to be retweeted than the truth,

即便在控制了這些變量後，

controlling for all of these and many other factors.

假新聞被轉發的可能性 仍然比真新聞高 70%。

So we had to come up with other explanations.

所以我們轉向其他假設。

And we devised what we called a "novelty hypothesis."

我們引入了一個名詞「新奇假設」， （novelty hypothesis）

So if you read the literature,

讀過文獻的人應該知道，

it is well known that human attention is drawn to novelty,

我們的注意力會被新奇事物吸引，

things that are new in the environment.

那些在環境中未曾有過的東西。

And if you read the sociology literature,

如果你讀過社會學作品，

you know that we like to share novel information.

會知道人們傾向於分享新奇的資訊，

It makes us seem like we have access to inside information,

因為這讓我們看起來掌握了內部資訊。

and we gain in status by spreading this kind of information.

透過傳播這些資訊，我們提升了地位。

So what we did was we measured the novelty of an incoming true or false tweet,

因此我們會將推文內容

compared to the corpus of what that individual had seen

同前 60 天看到過的相關內容比較，

in the 60 days prior on Twitter.

以此評估一條推文的新奇度。

But that wasn't enough, because we thought to ourselves,

但這還不夠，因為我們想到，

"Well, maybe false news is more novel in an information-theoretic sense,

也許假新聞本身更新奇，

but maybe people don't perceive it as more novel."

但人們不認為它們很新奇呢？

So to understand people's perceptions of false news,

所以為了研究人們對假新聞的感知，

we looked at the information and the sentiment

我們從對真、假新聞的回覆中

contained in the replies to true and false tweets.

提取人們的感受。

And what we found

我們發現，

was that across a bunch of different measures of sentiment --

在對不同反應情緒的統計中，

surprise, disgust, fear, sadness,

有驚訝、噁心、恐懼、悲傷、

anticipation, joy and trust --

期待、愉快和信任，

false news exhibited significantly more surprise and disgust

假新聞的回覆中， 驚訝和噁心的情緒

in the replies to false tweets.

顯著高於真新聞。

And true news exhibited significantly more anticipation,

真新聞收到的回覆中

joy and trust

包含更多的期待、

in reply to true tweets.

愉快和信任。

The surprise corroborates our novelty hypothesis.

驚訝情緒驗證了「新奇假設」，

This is new and surprising, and so we're more likely to share it.

因為它更新、更令人驚訝， 所以我們更願意轉發。

At the same time, there was congressional testimony

同時，不論在白宮

in front of both houses of Congress in the United States,

還是國會的證詞中，

looking at the role of bots in the spread of misinformation.

提到了機器人帳號 對傳播虛擬資訊的作用。

So we looked at this too --

於是我們進行了針對性研究。

we used multiple sophisticated bot-detection algorithms

我們使用多種複雜的 機器人帳號探測算法，

to find the bots in our data and to pull them out.

找出機器人發佈的資訊，

So we pulled them out, we put them back in

把它剔除出，又放回去，

and we compared what happens to our measurement.

然後比較兩種情況的數據。

And what we found was that, yes indeed,

我們發現，

機器人確實促進了假新聞的傳播，

bots were accelerating the spread of false news online,

but they were accelerating the spread of true news

但它們也幾乎同樣程度地

at approximately the same rate.

促進了真新聞的傳播。

Which means bots are not responsible

這表明，機器人與真假新聞 傳播程度的差異無關。

for the differential diffusion of truth and falsity online.

We can't abdicate that responsibility,

所以不是機器人的問題，

because we, humans, are responsible for that spread.

而是我們，人類的問題。

Now, everything that I have told you so far,

直到現在，我跟你們說的，

unfortunately for all of us,

很不幸，

is the good news.

還不算最糟糕。

The reason is because it's about to get a whole lot worse.

原因是，情況將要變得更糟。

And two specific technologies are going to make it worse.

兩種技術將惡化形式。

We are going to see the rise of a tremendous wave of synthetic media.

我們將迎來合成媒體的浪潮，

Fake video, fake audio that is very convincing to the human eye.

人眼無法分辨的假影片、假音頻。

And this will powered by two technologies.

這背後由兩種技術支持。

The first of these is known as "generative adversarial networks."

第一種技術是「生成對抗網路」，

This is a machine-learning model with two networks:

是一種包含兩個神經網路的 機器學習方法：

a discriminator,

判別網路

whose job it is to determine whether something is true or false,

用來判斷數據的真僞；

and a generator,

生成網路

whose job it is to generate synthetic media.

用來生成合成數據，也就是假數據。

So the synthetic generator generates synthetic video or audio,

所以生成網路生成假影片、假音頻，

and the discriminator tries to tell, "Is this real or is this fake?"

判別網路嘗試分辨真假。

And in fact, it is the job of the generator

事實上，生成網路的目標就是

to maximize the likelihood that it will fool the discriminator

讓假數據看起來盡可能「真」，

into thinking the synthetic video and audio that it is creating

以欺騙判別網路， 讓它認為這是真的。

is actually true.

Imagine a machine in a hyperloop,

想像一下，透過無休止的博弈，

trying to get better and better at fooling us.

機器越發擅長欺騙，以假亂真。

This, combined with the second technology,

接著第二種技術登場，

which is essentially the democratization of artificial intelligence to the people,

它可說是人工智慧的平民化。

the ability for anyone,

所有人都可使用，

without any background in artificial intelligence

無需任何人工智慧或機器學習背景。

or machine learning,

to deploy these kinds of algorithms to generate synthetic media

這些算法在合成媒體的應用，

makes it ultimately so much easier to create videos.

大大降低了修改影片的門檻。

The White House issued a false, doctored video

白宮曾發佈一段修改過的假影片，

of a journalist interacting with an intern who was trying to take his microphone.

影片中的記者正阻止 白宮實習生拿走麥克風，

They removed frames from this video

白宮刪除了幾幀影片，

in order to make his actions seem more punchy.

讓記者的動作看起來更蠻橫。

And when videographers and stuntmen and women

接受採訪的攝影師和特效師

were interviewed about this type of technique,

在被問到這類技術時，

they said, "Yes, we use this in the movies all the time

他們說：「是的，這是 電影製作的常用技術，

to make our punches and kicks look more choppy and more aggressive."

使拳打腳踢顯得 更震撼和更具侵略性。」

They then put out this video

白宮放出這段影片，

and partly used it as justification

以對女實習生的「侵略性動作」為由，

to revoke Jim Acosta, the reporter's, press pass

撤銷了吉姆·阿科斯塔 （影片中的記者）

進入白宮的記者通行證，

from the White House.

And CNN had to sue to have that press pass reinstated.

CNN 不得不透過訴訟 來要回通行證的權限。

There are about five different paths that I can think of that we can follow

我想到了五個方式

to try and address some of these very difficult problems today.

來應對這些難題。

Each one of them has promise,

每一種方式都帶來希望，

but each one of them has its own challenges.

但也各有挑戰。

The first one is labeling.

第一個方式是標示。

Think about it this way:

想一想，

when you go to the grocery store to buy food to consume,

你去商場買的吃食

it's extensively labeled.

有廣泛的標示。

You know how many calories it has,

你知道其中有多少卡路里，

how much fat it contains --

有多少脂肪。

and yet when we consume information, we have no labels whatsoever.

然而我們獲取的資訊 卻看不到任何標示。

What is contained in this information?

這則資訊中包含什麼？

Is the source credible?

資訊來源可信嗎？

Where is this information gathered from?

資訊的依據可靠嗎？

We have none of that information

我們在消耗資訊時並不知道這些。

when we are consuming information.

That is a potential avenue, but it comes with its challenges.

這可能是種解決方法，但也有困難。

For instance, who gets to decide, in society, what's true and what's false?

比如說，社會中誰來決定真假，

Is it the governments?

是政府嗎？

Is it Facebook?

是臉書嗎？

Is it an independent consortium of fact-checkers?

是獨立的事實查核組織嗎？

And who's checking the fact-checkers?

誰負責查證這些查核組織呢？

Another potential avenue is incentives.

另一種可能的方式是獎勵機制。

We know that during the US presidential election

美國總統大選期間，

there was a wave of misinformation that came from Macedonia

其中一部分假新聞來自馬其頓。

that didn't have any political motive

他們沒有任何政治動機，

but instead had an economic motive.

而是出於背後的經濟利益。

And this economic motive existed,

傳播假新聞的暴利在於，

because false news travels so much farther, faster

假新聞傳播得比真新聞 更遠、更快、更深，

and more deeply than the truth,

and you can earn advertising dollars as you garner eyeballs and attention

它們新奇而博人眼球，

with this type of information.

於是有更多的廣告費。

But if we can depress the spread of this information,

如果我們能抑制這些假新聞的傳播，

perhaps it would reduce the economic incentive

也許就能降低經濟利益，

to produce it at all in the first place.

從源頭減少假新聞數量。

Third, we can think about regulation,

第三，可以用法律規範新聞。

and certainly, we should think about this option.

這也是必要的應對方式。

In the United States, currently,

在美國，

we are exploring what might happen if Facebook and others are regulated.

我們正在嘗試用法律 規範臉書及其他社交網路。

While we should consider things like regulating political speech,

我們可以監管政治言論、

labeling the fact that it's political speech,

標明演講性質、

making sure foreign actors can't fund political speech,

禁止外國參與者資助政治演說。

it also has its own dangers.

但這個方法同樣存在危險。

For instance, Malaysia just instituted a six-year prison sentence

例如，馬來西亞最近頒布新法案，

for anyone found spreading misinformation.

假新聞傳播者將會面臨六年監禁。

And in authoritarian regimes,

在專制政權中，

these kinds of policies can be used to suppress minority opinions

這類法律可能被用來鎮壓少數異見，

and to continue to extend repression.

進一步擴大政治壓迫。

The fourth possible option is transparency.

第四種方法是透明化。

We want to know how do Facebook's algorithms work.

我們想知道臉書的算法如何，

How does the data combine with the algorithms

算法如何運用數據

to produce the outcomes that we see?

呈現出我們看到的內容。

We want them to open the kimono

我們要讓他們開誠佈公，

and show us exactly the inner workings of how Facebook is working.

讓我們看到臉書的內部運作方式。

And if we want to know social media's effect on society,

而要想知道社交網路對社會的影響，

we need scientists, researchers

我們需要讓科學家 和其他研究人員獲得這些數據。

and others to have access to this kind of information.

But at the same time,

但同時，

we are asking Facebook to lock everything down,

我們也在要求臉書封鎖所有數據，

to keep all of the data secure.

確保隱私安全，

So, Facebook and the other social media platforms

所以，臉書及同類社交平臺

are facing what I call a transparency paradox.

正面臨著「透明化悖論」。

We are asking them, at the same time,

因為我們要求他們

to be open and transparent and, simultaneously secure.

在公開透明化的同時確保隱私安全。

This is a very difficult needle to thread,

這是個極其困難的任務，

but they will need to thread this needle

但它們必須完成，

if we are to achieve the promise of social technologies

才能在避免其巨大隱患的同時

while avoiding their peril.

放心使用技術。

The final thing that we could think about is algorithms and machine learning.

最後一種方法是機器學習算法。

Technology devised to root out and understand fake news, how it spreads,

開發防範假新聞的算法，

研究假新聞的傳播，

and to try and dampen its flow.

抑制假新聞的擴散。

Humans have to be in the loop of this technology,

在技術開發過程中，人類不可缺位，

because we can never escape

因為我們無法逃避的是，

that underlying any technological solution or approach

所有的技術方案之下，

is a fundamental ethical and philosophical question

是根本的倫理與哲學問題：

about how do we define truth and falsity,

我們如何定義真實與虛假，

to whom do we give the power to define truth and falsity

誰有權定義真實與虛假，

and which opinions are legitimate,

哪些觀點是合法的，

which type of speech should be allowed and so on.

哪些言論可被允許，

諸如此類。

Technology is not a solution for that.

科技無法給出答案，

Ethics and philosophy is a solution for that.

只有倫理和哲學能夠回答。

Nearly every theory of human decision making,

幾乎所有關於人類決策、

human cooperation and human coordination

人類合作與協調的理論，

has some sense of the truth at its core.

其核心都包含追求真實。

But with the rise of fake news,

但隨著假新聞、假影片 和假音頻大行其道，

the rise of fake video,

the rise of fake audio,

we are teetering on the brink of the end of reality,

我們正滑向真實消亡的邊緣。

where we cannot tell what is real from what is fake.

我們正逐漸失去辨別真假的能力，

And that's potentially incredibly dangerous.

這或許將是極其危險的。

We have to be vigilant in defending the truth

我們必須時刻警惕，

守護真實，阻止虛假。

against misinformation.

With our technologies, with our policies

透過科技、透過政策，

and, perhaps most importantly,

或許最重要的是

with our own individual responsibilities,

透過我們每個人的責任感、

decisions, behaviors and actions.

每個人的選擇、行為

來守護真實。

Thank you very much.

謝謝大家。

(Applause)

（掌聲）