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

翻译人员: Thomas Tam 校对人员: Echo Sun

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

我致力于协助电脑和 我们周围世界的沟通。

There are a lot of ways to do this,

是有很多方法可以做到这一点，

and I like to focus on helping computers

我喜欢专注于协助电脑

to talk about what they see and understand.

去谈论它们看到和理解的内容。

Given a scene like this,

鉴于这样的情景，

a modern computer-vision algorithm

一个现代的计算机视觉演算法，

可以告诉你，有一个女人， 还有一只狗。

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

It can tell you that the woman is smiling.

它可以告诉你，那个女人在微笑。

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

它甚至可以告诉你， 这只狗非常可爱。

I work on this problem

我处理这个问题

thinking about how humans understand and process the world.

思考人类如何理解和与世界共处。

The thoughts, memories and stories

那些思想，记忆和故事

that a scene like this might evoke for humans.

在这样的场景中， 可能会唤起人类的注意。

All the interconnections of related situations.

所有关连情况的相互联系。

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

也许你以前见过这样的狗，

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

或者你曾经花时间， 在这样的沙滩上跑步，

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

并进一步唤起 过去假期的记忆和想法，

past times to the beach,

以前去海滩的时候，

times spent running around with other dogs.

花在与其他狗儿， 跑来跑去的时间。

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

我的指导原则之一， 是通过帮助电脑了解

what it's like to have these experiences,

这是什么样的经历，

to understand what we share and believe and feel,

从而了解我们所相信的 和感受的共通点，

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

那么我们就有能力开始 不断发展计算机技术，

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

以一种与我们经验互补的方式。

So, digging more deeply into this,

因此，深入挖掘这一点，

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

我几年前开始致力于帮助电脑 产生类似人类的故事，

from sequences of images.

从图像序列。

So, one day,

所以，有一天，

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

我正在用电脑工作时，询问它 对澳大利亚之行的看法。

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

它看了看图片， 看到一只树袋熊。

It didn't know what the koala was,

它不知道树袋熊是什么，

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

但电脑表示它认为树袋熊 看起来是很有趣的生物。

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

然后我与电脑分享一系列 关于房屋烧毁的图像。

It took a look at the images and it said,

电脑看了一下图片，它说：

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

「这是个惊人的景观！ 这很壮观！」

It sent chills down my spine.

它使我的脊背发冷。

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

电脑看到一个可怕的， 改变生活和毁灭生命的事件

and thought it was something positive.

并认为这是积极的事情。

I realized that it recognized the contrast,

我意识到电脑认识到

红色和黄色的对比，

the reds, the yellows,

and thought it was something worth remarking on positively.

并认为这是值得积极评价的事情。

And part of why it was doing this

部分原因是因为

was because most of the images I had given it

我输入电脑的大部分

were positive images.

是积极的图像。

That's because people tend to share positive images

那是因为人们谈论自己的经历时，

倾向于分享积极的图像。

when they talk about their experiences.

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

你上次在葬礼上看到自拍照 是什么时候？

I realized that, as I worked on improving AI

我意识到在改进人工智能的过程中，

task by task, dataset by dataset,

是任务归任务，数据集归数据集，

that I was creating massive gaps,

在电脑的理解上，

创造出巨大的差距，缺陷和盲点。

holes and blind spots in what it could understand.

And while doing so,

在这样做的同时，

I was encoding all kinds of biases.

我正在为各种偏见编码。

Biases that reflect a limited viewpoint,

偏见反映有限观点，

limited to a single dataset --

源于一个数据集——

biases that can reflect human biases found in the data,

其中就反映着人类同样的，

such as prejudice and stereotyping.

比如成见和刻板印象。

I thought back to the evolution of the technology

我回想起那技术的发展，

that brought me to where I was that day --

让我达到那天我所处的境地——

how the first color images

第一张彩色的图像

were calibrated against a white woman's skin,

是针对一位白人女性的 皮肤颜色进行校准的，

meaning that color photography was biased against black faces.

这意味着彩色摄影对黑脸有偏差。

And that same bias, that same blind spot

而同样的偏差，那个盲点

continued well into the '90s.

继续带入了 90 年代。

And the same blind spot continues even today

同样的盲点即使在今天，

in how well we can recognize different people's faces

仍然存在于在面部识别技术应用中，

怎样辨识不同人物的脸。

in facial recognition technology.

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

在今天的研究中， 我想到了最先进的技术，

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

都倾向于将我们的想法， 限制在一个数据集和一个问题上。

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

而这样做，我们正在创造 更多的盲点和偏见，

that the AI could further amplify.

会在使用人工智能时 被进一步放大。

I realized then that we had to think deeply

那时我意识到我们必须深思，

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

我们今天发明创造技术， 在五年到十年之后会怎样被看待 。

Humans evolve slowly, with time to correct for issues

在人类与环境互动作用中，

人类用时间纠正问题， 所以进化缓慢。

in the interaction of humans and their environment.

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

人工智能相比之下，正在以 令人难以置信的速度发展。

这意味着它确实很重要，

And that means that it really matters

that we think about this carefully right now --

我们现在要仔细考虑这一点 ——

that we reflect on our own blind spots,

反思自己的盲点，

our own biases,

及偏见，

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

并考虑这些偏见是如何影响 我们现在创造的技术，

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

并讨论今天的技术， 对未来意味着什么。

CEOs and scientists have weighed in on what they think

CEO和科学家， 已经权衡了他们的想法，

the artificial intelligence technology of the future will be.

关于未来的人工智能发展。

Stephen Hawking warns that

斯蒂芬·霍金警告说：

"Artificial intelligence could end mankind."

「人工智能会使人类灭亡。」

Elon Musk warns that it's an existential risk

伊隆‧马斯克警告 这是一种存在的风险，

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

也是我们作为文明社会; 要面临的最大风险之一。

Bill Gates has made the point,

比尔‧盖茨指出：

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

「我不明白为什么人们 对人工智能不更忧虑。」

But these views --

但是这些观点——

they're part of the story.

只是故事的一部分。

The math, the models,

那数学，模型，

the basic building blocks of artificial intelligence

这些人工智能的基本组成部分，

are something that we call access and all work with.

是我们都可以取得和并使用的。

We have open-source tools for machine learning and intelligence

我们有向大众开放的源代码工具 来学习机器，

that we can contribute to.

并同时作出自己的贡献。

And beyond that, we can share our experience.

除此之外，我们也可以 分享我们的经验。

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

我们可以分享在技术方面 及其与我们的关系，

and how it excites us.

和如何令我们雀跃的地方。

We can discuss what we love.

我们可以讨论我们所爱的东西。

我们可以与预见的将来进行沟通，

We can communicate with foresight

about the aspects of technology that could be more beneficial

关于技术方面这可能会更有益，

or could be more problematic over time.

或随着时间的推移， 可能会出现更多的问题。

If we all focus on opening up the discussion on AI

如果我们都专注于 开放对于人工智能的讨论

with foresight towards the future,

展望未来，

this will help create a general conversation and awareness

这将有助于创造一个 常规的对话和意识，

about what AI is now,

关于人工智能是什么？

what it can become

它能成为什么？

and all the things that we need to do

以及我们需要做的所有事情，

in order to enable that outcome that best suits us.

以实现最适合我们的结果。

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

我们已经在今天使用的技术中 看到和了解这一点。

We use smart phones and digital assistants and Roombas.

我们使用智能手机，数码助理 和自动吸尘器。

Are they evil?

它们邪恶吗？

Maybe sometimes.

也许有时是。

Are they beneficial?

他们有益吗？

Yes, they're that, too.

是的，他们也是。

And they're not all the same.

它们并不完全相同。

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

在那里你已经看到了未来的光芒。

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

未来将继续从我们现在 建立和创造的东西开始。

We set into motion that domino effect

我们启动了多米诺骨牌效应，

that carves out AI's evolutionary path.

这就揭开了人工智能的进化通道

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

在我们的时代，塑造了 明天的人工智能。

Technology that immerses us in augmented realities

让我们能沉浸在增强现实的技术中，

bringing to life past worlds.

使过去的世界复活，

Technology that helps people to share their experiences

当人们沟通有困难时，

when they have difficulty communicating.

科技就帮助他们分享彼此的经验。

Technology built on understanding the streaming visual worlds

建立于在线视觉媒体的科技，

used as technology for self-driving cars.

可被用在汽车自动驾驶上。

Technology built on understanding images and generating language,

科技基于图像的理解而产生语言，

evolving into technology that helps people who are visually impaired

能演变成协助视障人士的技术，

be better able to access the visual world.

帮助他们更好地拥有视觉世界。

And we also see how technology can lead to problems.

我们也看到科技 如何导致一些问题。

We have technology today

我们今天有科技

that analyzes physical characteristics we're born with --

分析我们出生的身体特征 ——

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

比如我们皮肤的颜色 还是我们脸上的表情

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

以确定我们是否罪犯或恐怖分子。

We have technology that crunches through our data,

我们拥有处理数据的技术，

even data relating to our gender or our race,

处理关于性别或种族的数据，

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

以确定我们是否可以获得贷款。

All that we see now

我们现在看到的所有东西，

is a snapshot in the evolution of artificial intelligence.

只是人工智能演变过程中的 快照。

Because where we are right now,

因为我们现在所处的地方，

is within a moment of that evolution.

是演变中的一个时刻。

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

这意味着我们现在所做的， 将会影响事情的往后发展，

and in the future.

并延至未来的世界。

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

如果我们希望人工智能 能协助人类的方式进化，

then we need to define the goals and strategies

我们就需要确定策略和目标，

that enable that path now.

马上开通那条路径。

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

我想看到的是适合人类的

with our culture and with the environment.

文化和环境的发展方向。

Technology that aids and assists those of us with neurological conditions

科技能帮助我们治愈神经系统疾病

or other disabilities

或其它残疾的患者，

in order to make life equally challenging for everyone.

让他们与每个人一样， 让生活同样具有挑战性。

Technology that works

科技的运作 不会考量你的特征

regardless of your demographics or the color of your skin.

或皮肤颜色。

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

我今天关注的是

明日和十年后的科技，

and for 10 years from now.

AI can turn out in many different ways.

人工智能可以以许多不同的方式出现。

But in this case,

但在这种情况下，

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

它并不是没有任何目的地的 无人驾驶车。

This is the car that we are driving.

这是我们能驾驶同时控制的汽车。

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

我们选择何时加速和何时减速。

We choose if we need to make a turn.

我们选择是否需要转弯。

We choose what the AI of the future will be.

我们选择未来的人工智能会是什么。

会有一个广阔的竞技场。

There's a vast playing field

of all the things that artificial intelligence can become.

容许人工智能可以成为所有的东西。

It will become many things.

它会变成很多不同的东西。

And it's up to us now,

现在取决于

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

我们要弄清楚所需要实施的

to make sure the outcomes of artificial intelligence

以确保人工智能的结果

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

是对所有人类都会更好。

Thank you.

谢谢。

(Applause)

（掌声）