How we can build AI to help humans, not hurt us | Margaret Mitchell

譯者: Lilian Chiu 審譯者: NAN-KUN WU

I work on helping computers communicate about the world around us.

我的工作是在協助電腦 和我們周遭的世界交流。

There are a lot of ways to do this,

要這麼做，有很多方式，

and I like to focus on helping computers

我喜歡聚焦在協助電腦

to talk about what they see and understand.

來談它們看見什麼、了解什麼。

Given a scene like this,

給予一個這樣的情景，

a modern computer-vision algorithm

一個現代電腦視覺演算法

can tell you that there's a woman and there's a dog.

就能告訴你情景中 有一個女子和一隻狗。

It can tell you that the woman is smiling.

它能告訴你，女子在微笑。

It might even be able to tell you that the dog is incredibly cute.

它甚至可能可以告訴你， 那隻狗相當可愛。

I work on this problem

我處理這個問題時

thinking about how humans understand and process the world.

腦中想的是人類如何 了解和處理這個世界。

The thoughts, memories and stories

對人類而言，這樣的情景有可能會

that a scene like this might evoke for humans.

喚起什麼樣的思想、記憶、故事。

All the interconnections of related situations.

相關情況的所有相互連結。

Maybe you've seen a dog like this one before,

也許你以前看過像這樣的狗，

or you've spent time running on a beach like this one,

或者你曾花時間在 像這樣的海灘上跑步，

and that further evokes thoughts and memories of a past vacation,

那就會進一步喚起一個 過去假期的思想和記憶，

past times to the beach,

過去在海灘的時光，

times spent running around with other dogs.

花在和其他狗兒到處奔跑的時間。

One of my guiding principles is that by helping computers to understand

我的指導原則之一， 是要協助電腦了解

what it's like to have these experiences,

有這些經驗是什麼樣的感覺，

to understand what we share and believe and feel,

去了解我們共有什麼、 相信什麼、感受到什麼，

then we're in a great position to start evolving computer technology

那麼，我們就有很好的機會 可以開始讓電腦科技

in a way that's complementary with our own experiences.

以一種和我們自身經驗 互補的方式來演化。

So, digging more deeply into this,

所以，我更深入研究這主題，

a few years ago I began working on helping computers to generate human-like stories

幾年前，我開始努力協助 電腦產生出像人類一樣的故事，

from sequences of images.

從一連串的影像來產生。

So, one day,

有一天，

I was working with my computer to ask it what it thought about a trip to Australia.

我在問我的電腦，它對於 去澳洲旅行有什麼想法。

It took a look at the pictures, and it saw a koala.

它看了照片，看見一隻無尾熊。

It didn't know what the koala was,

它不知道無尾熊是什麼，

but it said it thought it was an interesting-looking creature.

但它說它認為這是一隻 看起來很有趣的生物。

Then I shared with it a sequence of images about a house burning down.

我和它分享了一系列的圖片， 都和房子被燒毀有關。

It took a look at the images and it said,

它看了那些圖片，說：

"This is an amazing view! This is spectacular!"

「這是好棒的景色！這好壯觀！」

It sent chills down my spine.

它讓我背脊發涼。

It saw a horrible, life-changing and life-destroying event

它看著一個會改變人生、 摧毀生命的可怕事件，

and thought it was something positive.

卻以為它是很正面的東西。

I realized that it recognized the contrast,

我了解到，它能認出對比反差、

the reds, the yellows,

紅色、黃色，

and thought it was something worth remarking on positively.

然後就認為它是 值得正面評論的東西。

And part of why it was doing this

它這麼做的部分原因，

was because most of the images I had given it

是因為我給它的大多數圖片

were positive images.

都是正面的圖片。

That's because people tend to share positive images

那是因為人在談論他們的經驗時，

when they talk about their experiences.

本來就傾向會分享正面的圖片。

When was the last time you saw a selfie at a funeral?

你上次看到在葬禮上的 自拍照是何時？

I realized that, as I worked on improving AI

我了解到，當我努力 在改善人工智慧，

task by task, dataset by dataset,

一個任務一個任務、一個資料集 一個資料集地改善，

that I was creating massive gaps,

結果我卻在「它能了解什麼」上

holes and blind spots in what it could understand.

創造出了大量的隔閡、 漏洞，以及盲點。

And while doing so,

這麼做的時候，

I was encoding all kinds of biases.

我是在把各種偏見做編碼。

Biases that reflect a limited viewpoint,

這些偏見反映出受限的觀點，

limited to a single dataset --

受限於單一資料集──

biases that can reflect human biases found in the data,

這些偏見能反應出 在資料中的人類偏見，

such as prejudice and stereotyping.

比如偏袒以及刻板印象。

I thought back to the evolution of the technology

我回頭去想一路帶我走到

that brought me to where I was that day --

那個時點的科技演化──

how the first color images

第一批彩色影像如何

were calibrated against a white woman's skin,

根據一個白種女子的皮膚來做校準，

meaning that color photography was biased against black faces.

這表示，彩色照片對於 黑皮膚臉孔是有偏見的。

And that same bias, that same blind spot

同樣的偏見，同樣的盲點，

continued well into the '90s.

持續湧入了九○年代。

And the same blind spot continues even today

而同樣的盲點甚至持續到現今，

in how well we can recognize different people's faces

出現在我們對於不同人的 臉部辨識能力中，

in facial recognition technology.

在人臉辨識技術中。

I though about the state of the art in research today,

我思考了現今在研究上發展水平，

where we tend to limit our thinking to one dataset and one problem.

我們傾向會把我們的思路限制 在一個資料集或一個問題上。

And that in doing so, we were creating more blind spots and biases

這麼做時，我們就會 創造出更多盲點和偏見，

that the AI could further amplify.

它們可能會被人工智慧給放大。

I realized then that we had to think deeply

那時，我了解到，我們必須要

about how the technology we work on today looks in five years, in 10 years.

深入思考我們現今努力發展的科技， 在五年、十年後會是什麼樣子。

Humans evolve slowly, with time to correct for issues

人類演進很慢，有時間可以去修正

in the interaction of humans and their environment.

在人類互動以及其環境中的議題。

In contrast, artificial intelligence is evolving at an incredibly fast rate.

相對的，人工智慧的 演進速度非常快。

And that means that it really matters

那就意味著，很重要的是

that we think about this carefully right now --

我們現在要如何仔細思考這件事──

that we reflect on our own blind spots,

我們要反省我們自己的盲點，

our own biases,

我們自己的偏見，

and think about how that's informing the technology we're creating

並想想它們帶給我們所 創造出的科技什麼樣的資訊，

and discuss what the technology of today will mean for tomorrow.

並討論現今的科技在 將來代表的是什麼意涵。

CEOs and scientists have weighed in on what they think

對於未來的人工智慧 應該是什麼樣子，

the artificial intelligence technology of the future will be.

執行長和科學家的意見 是很有份量的。

Stephen Hawking warns that

史蒂芬霍金警告過：

"Artificial intelligence could end mankind."

「人工智慧可能終結人類。」

Elon Musk warns that it's an existential risk

伊隆馬斯克警告過， 它是個生存風險，

and one of the greatest risks that we face as a civilization.

也是我們人類文明所 面臨最大的風險之一。

Bill Gates has made the point,

比爾蓋茲有個論點：

"I don't understand why people aren't more concerned."

「我不了解為什麼 人們不更關心一點。」

But these views --

但，這些看法──

they're part of the story.

它們是故事的一部分。

The math, the models,

數學、模型，

the basic building blocks of artificial intelligence

人工智慧的基礎材料

are something that we call access and all work with.

是我們所有人都能夠取得並使用的。

We have open-source tools for machine learning and intelligence

我們有機器學習和智慧用的 開放原始碼工具，

that we can contribute to.

我們都能對其做出貢獻。

And beyond that, we can share our experience.

在那之外，我們可以 分享我們的經驗。

We can share our experiences with technology and how it concerns us

分享關於科技、它如何影響我們、

and how it excites us.

它如何讓我們興奮的經驗。

We can discuss what we love.

我們可以討論我們所愛的。

We can communicate with foresight

我們能帶著遠見來交流，

about the aspects of technology that could be more beneficial

談談關於科技有哪些面向，

隨著時間發展可能可以更有助益， 或可能產生問題。

or could be more problematic over time.

If we all focus on opening up the discussion on AI

若我們都能把焦點放在

開放地帶著對未來的遠見 來討論人工智慧，

with foresight towards the future,

this will help create a general conversation and awareness

這就能創造出一般性的談話和意識，

about what AI is now,

關於人工智慧現在是什麼樣子、

what it can become

它未來可以變成什麼樣子，

and all the things that we need to do

以及所有我們需要做的事，

in order to enable that outcome that best suits us.

以產生出最適合我們的結果。

We already see and know this in the technology that we use today.

我們已經在現今我們所使用的 科技中看見這一點了。

We use smart phones and digital assistants and Roombas.

我們用智慧手機、數位助理， 以及掃地機器人。

Are they evil?

它們邪惡嗎？

Maybe sometimes.

也許有時候。

Are they beneficial?

它們有助益嗎？

Yes, they're that, too.

是的，這也是事實。

And they're not all the same.

且它們並非全都一樣的。

And there you already see a light shining on what the future holds.

你們已經看到未來 可能性的一絲光芒。

The future continues on from what we build and create right now.

未來延續的基礎，是我們 現在所建立和創造的。

We set into motion that domino effect

我們開始了骨牌效應，

that carves out AI's evolutionary path.

刻劃出了人工智慧的演進路徑。

In our time right now, we shape the AI of tomorrow.

我們在現在這時代 形塑未來的人工智慧。

Technology that immerses us in augmented realities

讓我們能沉浸入擴增實境中的科技，

bringing to life past worlds.

讓過去的世界又活了過來。

Technology that helps people to share their experiences

協助人們在溝通困難時

when they have difficulty communicating.

還能分享經驗的科技。

Technology built on understanding the streaming visual worlds

立基在了解串流 視覺世界之上的科技，

used as technology for self-driving cars.

被用來當作自動駕駛汽車的科技。

Technology built on understanding images and generating language,

立基在了解圖像和產生語言的科技，

evolving into technology that helps people who are visually impaired

演進成協助視覺損傷者的科技，

be better able to access the visual world.

讓他們更能進入視覺的世界。

And we also see how technology can lead to problems.

我們也看到了科技如何導致問題。

We have technology today

現今，我們有科技

that analyzes physical characteristics we're born with --

能夠分析我們天生的身體特徵──

such as the color of our skin or the look of our face --

比如膚色或面部的外觀──

in order to determine whether or not we might be criminals or terrorists.

可以用來判斷我們是否 有可能是罪犯或恐怖份子。

We have technology that crunches through our data,

我們有科技能夠分析我們的資料，

even data relating to our gender or our race,

甚至和我們的性別 或種族相關的資料，

in order to determine whether or not we might get a loan.

來決定我們的貸款是否能被核准。

All that we see now

我們現在所看見的一切，

is a snapshot in the evolution of artificial intelligence.

都是人工智慧演進的約略寫照。

Because where we are right now,

因為我們現在所處的位置，

is within a moment of that evolution.

是在那演進的一個時刻當中。

That means that what we do now will affect what happens down the line

那表示，我們現在 所做的，會影響到後續

and in the future.

未來發生的事。

If we want AI to evolve in a way that helps humans,

如果我們想讓人工智慧的 演進方式是對人類有助益的，

then we need to define the goals and strategies

那麼我們現在就得要 定義目標和策略，

that enable that path now.

來讓那條路成為可能。

What I'd like to see is something that fits well with humans,

我想要看見的東西是要能夠 和人類、我們的文化，

with our culture and with the environment.

及我們的環境能非常符合的東西。

Technology that aids and assists those of us with neurological conditions

這種科技要能夠幫助和協助 有神經系統疾病

or other disabilities

或其他殘疾者的人，

in order to make life equally challenging for everyone.

讓人生對於每個人的 挑戰程度是平等的。

Technology that works

這種科技的運作

regardless of your demographics or the color of your skin.

不會考量你的 人口統計資料或你的膚色。

And so today, what I focus on is the technology for tomorrow

所以，現今我著重的是明日的科技

and for 10 years from now.

和十年後的科技。

AI can turn out in many different ways.

產生人工智慧的方式相當多。

But in this case,

但在這個情況中，

it isn't a self-driving car without any destination.

它並不是沒有目的地的 自動駕駛汽車。

This is the car that we are driving.

它是我們在開的汽車。

We choose when to speed up and when to slow down.

我們選擇何時要加速何時要減速。

We choose if we need to make a turn.

我們選擇是否要轉彎。

We choose what the AI of the future will be.

我們選擇將來的 人工智慧會是哪一種。

There's a vast playing field

人工智慧能夠

of all the things that artificial intelligence can become.

變成各式各樣的東西。

It will become many things.

它會變成許多東西。

And it's up to us now,

現在，決定權在我們，

in order to figure out what we need to put in place

我們要想清楚我們得要準備什麼，

to make sure the outcomes of artificial intelligence

來確保人工智慧的結果

are the ones that will be better for all of us.

會是對所有人都更好的結果。

Thank you.

謝謝。

(Applause)

（掌聲）