How we can teach computers to make sense of our emotions | Raphael Arar

Translator: Ivana Korom Reviewer: Joanna Pietrulewicz

翻译人员: Zihao Wang 校对人员: Peipei Xiang

I consider myself one part artist and one part designer.

我自认为既是一位艺术家， 又是一位设计师。

And I work at an artificial intelligence research lab.

我在一个研究 人工智能的实验室工作。

We're trying to create technology

我们正在尝试创造一项

that you'll want to interact with in the far future.

在遥远的将来能和人类互动的科技。

Not just six months from now, but try years and decades from now.

不是在六个月之后， 而是几年甚至几十年之后。

And we're taking a moonshot

我们正在实施一个“登月计划”，

that we'll want to be interacting with computers

我们希望能与电脑

in deeply emotional ways.

进行深层次的情感方面的互动。

So in order to do that,

为了做到这些，

the technology has to be just as much human as it is artificial.

技术不仅要智能，还要人性。

It has to get you.

它必须懂你。

You know, like that inside joke that'll have you and your best friend

就像一个朋友间的笑话， 能让你和你的朋友

on the floor, cracking up.

在地板上笑得前仰后翻；

Or that look of disappointment that you can just smell from miles away.

或者是，你远远就能 嗅到的失望气息。

I view art as the gateway to help us bridge this gap between human and machine:

我把艺术看作是帮助我们填补 人类与机器之间空白的途径：

to figure out what it means to get each other

使每个人都能相互了解对方，

so that we can train AI to get us.

使我们能训练 人工智能来“懂我们”。

See, to me, art is a way to put tangible experiences

对我来说，艺术是把有形的经历，

to intangible ideas, feelings and emotions.

转化为无形的想法、 感受、情感的方式。

And I think it's one of the most human things about us.

我认为这是人性的一个重要特征。

See, we're a complicated and complex bunch.

我们是难懂的、复杂的群体。

We have what feels like an infinite range of emotions,

我们拥有无限的情感，

and to top it off, we're all different.

而且，我们都是不同的。

We have different family backgrounds,

我们拥有不同的家庭背景，

different experiences and different psychologies.

不同的经历，不同的心理活动。

And this is what makes life really interesting.

这是为什么生活那么有趣的原因，

But this is also what makes working on intelligent technology

但这同时也是研究智能技术

extremely difficult.

最难的地方。

And right now, AI research, well,

如今，对人工智能的研究

it's a bit lopsided on the tech side.

过于偏重技术，

And that makes a lot of sense.

这也很好理解。

See, for every qualitative thing about us --

关于我们的每一个定性的特征，

you know, those parts of us that are emotional, dynamic and subjective --

比如属于我们情感的、 动态的、主观的部分——

we have to convert it to a quantitative metric:

我们要把它转化为一个量化指标：

something that can be represented with facts, figures and computer code.

能通过一些事实、图形和 电脑代码表现出来。

The issue is, there are many qualitative things

问题是，有很多定性的东西

that we just can't put our finger on.

是很难量化的。

So, think about hearing your favorite song for the first time.

想一想你第一次听到 你最喜欢的歌的时候，

What were you doing?

你在做什么？

How did you feel?

你有什么感受？

Did you get goosebumps?

你起鸡皮疙瘩了吗？

Or did you get fired up?

你有没有感到热血沸腾？

Hard to describe, right?

很难描述，对吗？

See, parts of us feel so simple,

我们一些看似很简单的感受

but under the surface, there's really a ton of complexity.

背后其实是很复杂的。

And translating that complexity to machines

而将这些复杂的东西 翻译成机器语言，

is what makes them modern-day moonshots.

这就是我们需要实现的 现代“登月计划”。

And I'm not convinced that we can answer these deeper questions

我不相信我们可以仅仅 用0和1这两个数字

with just ones and zeros alone.

来解决这些难题。

So, in the lab, I've been creating art

所以，在实验室，我通过创造艺术

as a way to help me design better experiences

来帮助我设计更好的

for bleeding-edge technology.

对尖端科技的体验。

And it's been serving as a catalyst

艺术作为一种催化剂，

to beef up the more human ways that computers can relate to us.

让电脑更加人类化， 更理解我们。

Through art, we're tacking some of the hardest questions,

通过艺术，我们在解决一些 非常困难的问题，

like what does it really mean to feel?

就像，感受到底是什么意思？

Or how do we engage and know how to be present with each other?

我们如何真正参与或投入其中？

And how does intuition affect the way that we interact?

我们的直觉怎样影响 我们互动的方式？

So, take for example human emotion.

以人类的情感为例，

Right now, computers can make sense of our most basic ones,

如今，电脑能够明白 我们的基本情感，

like joy, sadness, anger, fear and disgust,

比如开心、伤心、 生气、恐惧、厌恶，

by converting those characteristics to math.

把这些特征转化为数学。

But what about the more complex emotions?

那较复杂的情感呢？

You know, those emotions

比如那些很难

that we have a hard time describing to each other?

用文字向对方描述的情感，

Like nostalgia.

比如，怀旧。

So, to explore this, I created a piece of art, an experience,

所以，为了探索这个问题， 我创造了一件艺术品，一种体验，

that asked people to share a memory,

要求人们分享他们的记忆，

and I teamed up with some data scientists

我和一些数据科学家组成一个团队，

to figure out how to take an emotion that's so highly subjective

去研究高度主观的情感是怎样的，

and convert it into something mathematically precise.

如何将它们精确地转化为数学。

So, we created what we call a nostalgia score

我们创造了一个叫怀旧分数的东西，

这是这个装置的核心。

and it's the heart of this installation.

To do that, the installation asks you to share a story,

这个装置会要求你分享一则故事，

the computer then analyzes it for its simpler emotions,

电脑会分析它的一些简单的情感，

it checks for your tendency to use past-tense wording

它会检测你使用 过去时态的词语的偏好，

and also looks for words that we tend to associate with nostalgia,

还会寻找与怀旧有关的词语，

like "home," "childhood" and "the past."

比如“家”、“童年”和“过去”。

It then creates a nostalgia score

它最后会给出一个怀旧分数，

to indicate how nostalgic your story is.

代表着你的故事的怀旧程度，

And that score is the driving force behind these light-based sculptures

这个分数会 让这个灯箱的颜色发生变化，

that serve as physical embodiments of your contribution.

代表着你的贡献。

And the higher the score, the rosier the hue.

分数越高，色调越偏向玫瑰红色，

You know, like looking at the world through rose-colored glasses.

就像是通过玫红色的眼镜看世界。

So, when you see your score

当你看到你的分数，

and the physical representation of it,

以及它的外部反映时，

sometimes you'd agree and sometimes you wouldn't.

有时你会赞同，有时不赞同。

It's as if it really understood how that experience made you feel.

有的时候就好像它真的明白 故事里的你当时的感受，

But other times it gets tripped up

但有的时候它也会出错，

and has you thinking it doesn't understand you at all.

会让你觉得它一点也不懂你。

But the piece really serves to show

但这个装置能够说明，

that if we have a hard time explaining the emotions that we have to each other,

如果连我们都很难表述清楚的情感，

how can we teach a computer to make sense of them?

我们该如何教电脑明白呢？

So, even the more objective parts about being human are hard to describe.

甚至很多关于 人性的客观方面也很难描述。

Like, conversation.

比如，对话。

Have you ever really tried to break down the steps?

你曾经尝试过分解谈话的步骤吗？

So think about sitting with your friend at a coffee shop

试着想象一下， 你和你的朋友坐在咖啡馆，

and just having small talk.

进行简单的交谈，

How do you know when to take a turn?

你怎么知道轮到你说话了？

How do you know when to shift topics?

你怎么知道什么时候该转换话题？

And how do you even know what topics to discuss?

你怎么知道要讨论些什么？

See, most of us don't really think about it,

大多数人都不会想这些问题，

because it's almost second nature.

因为这对我们来说是很自然的。

And when we get to know someone, we learn more about what makes them tick,

当我们认识一个人的时候， 我们会对他们越来越了解，

and then we learn what topics we can discuss.

然后我们会知道能聊些什么话题。

But when it comes to teaching AI systems how to interact with people,

但是，当你教人工智能 怎样与人类互动时，

we have to teach them step by step what to do.

我们需要一步一步教它们该如何做。

And right now, it feels clunky.

而现在，这个过程还感觉很笨拙。

If you've ever tried to talk with Alexa, Siri or Google Assistant,

如果你曾尝试和 Alexa, Siri 或谷歌助手聊天，

you can tell that it or they can still sound cold.

你可以感觉得到， 它们仍听上去冷冰冰的。

And have you ever gotten annoyed

你是否曾经因为它们

when they didn't understand what you were saying

不明白你说什么而变得恼怒，

and you had to rephrase what you wanted 20 times just to play a song?

比如为了让它们放一首歌， 你得说上20次？

Alright, to the credit of the designers, realistic communication is really hard.

不过我们也要理解设计师，毕竟 让机器学会真实的沟通是非常难的。

And there's a whole branch of sociology,

有一个社会学的分支，

called conversation analysis,

叫做会话分析，

that tries to make blueprints for different types of conversation.

它尝试做不同对话类型的蓝图，

Types like customer service or counseling, teaching and others.

例如像客户服务、心理咨询、 教授课程等等的会话类型。

I've been collaborating with a conversation analyst at the lab

我已经在和会话分析学家 在实验室展开合作，

to try to help our AI systems hold more human-sounding conversations.

尝试帮助我们的人工智能系统 进行更多的人性化的对话。

This way, when you have an interaction with a chatbot on your phone

这样，当你和手机 聊天机器人进行互动时，

or a voice-based system in the car,

或者和车载语音系统互动时，

it sounds a little more human and less cold and disjointed.

这种声音就听上去更人性， 不那么冷淡和缺乏逻辑。

So I created a piece of art

我创造的这个艺术品，

that tries to highlight the robotic, clunky interaction

重点突显了机械化的、 笨拙的互动方式，

to help us understand, as designers,

以帮助我们这些设计师明白，

why it doesn't sound human yet and, well, what we can do about it.

为什么它听上去不像人类， 我们该如何解决这个问题。

The piece is called Bot to Bot

这个艺术品叫 Bot to Bot，

and it puts one conversational system against another

它将一个会话系统 搭建在另一个会话系统之上，

and then exposes it to the general public.

然后展示给公众。

And what ends up happening is that you get something

最终会发生的就是，

that tries to mimic human conversation,

它尝试模仿人类的对话，

but falls short.

但是却明显有不足之处。

Sometimes it works and sometimes it gets into these, well,

有的时候它还可以，而有的时候

会陷入误解的循环。

loops of misunderstanding.

So even though the machine-to-machine conversation can make sense,

虽然机器与机器的对话 从语法、用意上

grammatically and colloquially,

能让人明白，

但是你还是能感觉到 这个对话的冰冷和机械化。

it can still end up feeling cold and robotic.

And despite checking all the boxes, the dialogue lacks soul

尽管对话的其他要素都具备， 但却缺少了灵魂，

and those one-off quirks that make each of us who we are.

缺少了那些使我们 之所以为人类的特质。

So while it might be grammatically correct

尽管它的语法也许正确，

and uses all the right hashtags and emojis,

用对了所有的话题标签和符号表情，

it can end up sounding mechanical and, well, a little creepy.

但最终听起来还是有些呆板， 还有点儿吓人。

And we call this the uncanny valley.

我们把这称为恐怖谷，

You know, that creepiness factor of tech

这种科技的恐怖之处在于，

where it's close to human but just slightly off.

它无比接近人类，却又缺了点什么。

And the piece will start being

这件艺术品能开始用于

one way that we test for the humanness of a conversation

测试交流的人性化，

and the parts that get lost in translation.

以及被误解的部分。

So there are other things that get lost in translation, too,

还有其他一些事情 也容易被电脑误解，

like human intuition.

比如，人类的直觉。

Right now, computers are gaining more autonomy.

如今，电脑拥有更多的自主权，

They can take care of things for us,

能为我们管理一些东西，

like change the temperature of our houses based on our preferences

比如根据我们的偏好 调整房子的温度，

and even help us drive on the freeway.

甚至帮助我们在高速公路驾驶。

But there are things that you and I do in person

但是一些我和你会做的事，

that are really difficult to translate to AI.

是非常难翻译给人工智能的。

So think about the last time that you saw an old classmate or coworker.

想一想你上一次 见到一位老同学或老同事时，

Did you give them a hug or go in for a handshake?

你跟他们拥抱还是握手了呢？

You probably didn't think twice

你可能想都没想，

because you've had so many built up experiences

因为你有过许多这样的经历，

that had you do one or the other.

要么拥抱要么握手。

And as an artist, I feel that access to one's intuition,

作为一名艺术家，我认为 了解一个人的直觉，

your unconscious knowing,

你的潜意识的知觉，

is what helps us create amazing things.

能够帮助我们创造 令人惊叹的东西。

Big ideas, from that abstract, nonlinear place in our consciousness

大的点子，我们潜意识中 抽象的，非线性的东西

that is the culmination of all of our experiences.

是我们所有经历的总和。

And if we want computers to relate to us and help amplify our creative abilities,

如果我们想让电脑 帮我们提升创造力，

I feel that we'll need to start thinking about how to make computers be intuitive.

我认为我们需要思考 如何才能让电脑有直觉，

So I wanted to explore how something like human intuition

所以，我想探究 如何将像人类直觉的东西

could be directly translated to artificial intelligence.

直接地翻译给人工智能。

于是我创造了一台通过现实空间

And I created a piece that explores computer-based intuition

in a physical space.

探究电脑直觉的机器。

The piece is called Wayfinding,

它叫 Wayfinding，

and it's set up as a symbolic compass that has four kinetic sculptures.

它有4个动态装置，像一个指南针。

Each one represents a direction,

每一个装置代表着一个方向，

north, east, south and west.

北、东、南、西。

And there are sensors set up on the top of each sculpture

装在每个装置顶端的传感器，

that capture how far away you are from them.

能够捕获你离它们的距离有多远。

And the data that gets collected

接着数据会被采集，

ends up changing the way that sculptures move

最终装置就会移动，

and the direction of the compass.

从而改变指南针的方向。

The thing is, the piece doesn't work like the automatic door sensor

不过不像自动门的传感器那样——

that just opens when you walk in front of it.

你走到它前面的时候，门就会打开，

See, your contribution is only a part of its collection of lived experiences.

你的行为只是它 搜集的体验的一部分，

And all of those experiences affect the way that it moves.

所有的体验都会影响它的移动。

So when you walk in front of it,

所以当你在它前面走动时，

it starts to use all of the data

它开始用所有之前

that it's captured throughout its exhibition history --

捕获的数据——

or its intuition --

或它的直觉——

to mechanically respond to you based on what it's learned from others.

基于它从其他人那里学习到的， 对你做出机械的响应。

And what ends up happening is that as participants

最终，作为参与者，

we start to learn the level of detail that we need

我们意识到我们需要怎样的细节

in order to manage expectations

才能同时管理

from both humans and machines.

人类和机器的预期。

We can almost see our intuition being played out on the computer,

我们几乎可以看到我们的直觉 在电脑中被展示出来，

picturing all of that data being processed in our mind's eye.

想象所有的数据 被我们的心灵之眼所处理。

My hope is that this type of art

我希望这种艺术方式，

will help us think differently about intuition

能帮助我们从不同角度思考直觉，

and how to apply that to AI in the future.

以及将来如何 将它运用到人工智能中去。

So these are just a few examples of how I'm using art to feed into my work

这些都是我在自己的 人工智能设计和研究的工作中

as a designer and researcher of artificial intelligence.

如何利用艺术的例子。

And I see it as a crucial way to move innovation forward.

我觉得这是一个 推动创新的重要方式。

Because right now, there are a lot of extremes when it comes to AI.

因为现在说到人工智能， 两极分化的态度很严重。

Popular movies show it as this destructive force

比如一些流行电影 将其描绘成毁灭性的力量，

while commercials are showing it as a savior

而一些广告则 把它们描绘为救世主——

to solve some of the world's most complex problems.

能解决一些世界上 极端复杂的问题。

But regardless of where you stand,

但是不管你站在哪一边，

it's hard to deny that we're living in a world

我们都无法否认，我们正生活在一个

越来越数字化的世界中。

that's becoming more and more digital by the second.

Our lives revolve around our devices, smart appliances and more.

我们的生活被设备、智能家居等充斥，

And I don't think this will let up any time soon.

而我不觉得这种状况会停止。

So, I'm trying to embed more humanness from the start.

我想在一开始就植入更多的人性，

And I have a hunch that bringing art into an AI research process

而我有预感，将艺术 带入人工智能研究

is a way to do just that.

就是其中一个方法。

Thank you.

谢谢。

(Applause)

（掌声）