How to Govern AI — Even If It’s Hard to Predict | Helen Toner | TED

翻译人员: Lening Xu 校对人员: Yip Yan Yeung

When I talk to people about artificial intelligence,

当我和人们谈论人工智能时，

something I hear a lot from non-experts is “I don’t understand AI.”

从非专业人士那里听到最多的是 “我不了解人工智能”。

But when I talk to experts, a funny thing happens.

但当我和专家交谈时， 有趣的事发生了。

They say, “I don’t understand AI, and neither does anyone else.”

他们说：“我不懂人工智能， 其他人也不懂。”

This is a pretty strange state of affairs.

这是一种非常奇怪的状态。

Normally, the people building a new technology

通常，开发新技术的人 了解其由内而外的工作原理。

understand how it works inside and out.

But for AI, a technology that's radically reshaping the world around us,

但对于人工智能，这一彻底改变 我们周围世界的技术来说，

that's not so.

情况并非如此。

Experts do know plenty about how to build and run AI systems, of course.

当然，专家们确实对如何构建和运行 人工智能系统了如指掌。

But when it comes to how they work on the inside,

但是，当谈到它们的内部运行方式时，

there are serious limits to how much we know.

我们的认知有严重的局限性。

And this matters because without deeply understanding AI,

这很重要，因为如果 不深入了解人工智能，

it's really difficult for us to know what it will be able to do next,

我们真的很难知道它接下来能做什么，

or even what it can do now.

甚至很难知道它现在能做什么。

And the fact that we have such a hard time understanding

我们很难理解这项技术的进展，

what's going on with the technology and predicting where it will go next,

也很难预测下一步的发展方向，

is one of the biggest hurdles we face in figuring out how to govern AI.

这是我们在弄清楚如何治理人工智能时 面临的最大障碍之一。

But AI is already all around us,

但人工智能已经无处不在，

so we can't just sit around and wait for things to become clearer.

所以我们不能袖手旁观， 等待事情变得更清晰。

We have to forge some kind of path forward anyway.

无论如何，我们必须 开辟某种前进的道路。

I've been working on these AI policy and governance issues

我研究人工智能政策和治理问题

for about eight years,

大约有八年时间了，

First in San Francisco, now in Washington, DC.

首先是在旧金山， 现在在华盛顿特区。

Along the way, I've gotten an inside look

在此过程中，我深入了解各国政府 是如何管理这项技术的。

at how governments are working to manage this technology.

And inside the industry, I've seen a thing or two as well.

在行业内部， 我也见证了一两件事。

So I'm going to share a couple of ideas

因此，我将分享一些想法，

for what our path to governing AI could look like.

说明我们的人工智能治理之路 会是什么样子。

But first, let's talk about what actually makes AI so hard to understand

首先，让我们来谈谈究竟是什么 让人工智能如此难以理解和预测。

and predict.

One huge challenge in building artificial "intelligence"

构建人工“智能”面临的 一个巨大挑战是，

is that no one can agree on what it actually means

人们并未就“智能”的真正含义达成共识。

to be intelligent.

This is a strange place to be in when building a new tech.

这是开发新技术时 会出现的一个奇怪的情形。

When the Wright brothers started experimenting with planes,

当莱特兄弟开始试验飞机时，

they didn't know how to build one,

他们不知道如何建造飞机，

but everyone knew what it meant to fly.

但每个人都知道飞行意味着什么。

With AI on the other hand,

另一方面，对于人工智能，

different experts have completely different intuitions

不同的专家对人工智能的核心 有完全不同的直觉。

about what lies at the heart of intelligence.

Is it problem solving?

是解决问题吗？

Is it learning and adaptation,

是学习和适应吗？

are emotions,

是情感，还是身体 以某种方式参与其中？

or having a physical body somehow involved?

We genuinely don't know.

我们真的不知道。

But different answers lead to radically different expectations

但是，不同的答案会导致

人们对这项技术的发展方向 和实现速度的期望截然不同。

about where the technology is going and how fast it'll get there.

An example of how we're confused is how we used to talk

我们困惑的一个例子是

我们过去如何谈论 狭义人工智能与通用人工智能。

about narrow versus general AI.

For a long time, we talked in terms of two buckets.

很长一段时间以来， 我们是从两个孤立的方面来谈论它们的。

A lot of people thought we should just be dividing between narrow AI,

很多人认为，我们应该将 人工智能区分为：

狭义的人工智能， 即针对一个特定任务进行训练，

trained for one specific task,

like recommending the next YouTube video,

比如推荐下一个YouTube视频，

versus artificial general intelligence, or AGI,

另一种是人工通用智能即AGI，

that could do everything a human could do.

可以做人类能做的一切。

We thought of this distinction, narrow versus general,

我们认为这种区别， 无论是狭义还是通用，

as a core divide between what we could build in practice

是我们在实践中构建的智能 和真正智能之间的核心差异。

and what would actually be intelligent.

But then a year or two ago, along came ChatGPT.

但是一两年前， ChatGPT出现了。

If you think about it,

仔细想想，

you know, is it narrow AI, trained for one specific task?

它是针对一项特定任务 训练的狭义人工智能吗？

Or is it AGI and can do everything a human can do?

还是AGI，可以做人类能做的一切 ？

Clearly the answer is neither.

显然，两者都不是。

It's certainly general purpose.

它当然是通用的。

It can code, write poetry,

它可以编程、写诗、 分析商业问题、帮助你修车。

analyze business problems, help you fix your car.

But it's a far cry from being able to do everything

但这与你我能做的所有事情相去甚远。

as well as you or I could do it.

So it turns out this idea of generality

因此，事实证明，这种普遍性的概念

doesn't actually seem to be the right dividing line

实际上并不是智能与非智能 之间的正确分界线。

between intelligent and not.

And this kind of thing

对于目前整个人工智能领域来说， 这是一个巨大的挑战。

is a huge challenge for the whole field of AI right now.

We don't have any agreement on what we're trying to build

我们没有就我们试图构建的东西 或路线图达成任何一致。

or on what the road map looks like from here.

We don't even clearly understand the AI systems that we have today.

我们甚至没有清楚地了解 当今已经拥有的人工智能系统。

Why is that?

这是为什么？

Researchers sometimes describe deep neural networks,

研究人员有时将深度神经网络，

the main kind of AI being built today,

当今构建的主要人工智能，

as a black box.

描述为黑匣子。

But what they mean by that is not that it's inherently mysterious

但是他们的意思并不是说 它本质上是神秘的，

and we have no way of looking inside the box.

我们无法窥探盒子里面的情况。

The problem is that when we do look inside,

问题在于，当我们仔细观察内部时，

what we find are millions,

我们会发现数百万、 数十亿甚至数万亿的数字，

billions or even trillions of numbers

that get added and multiplied together in a particular way.

它们以特定的方式相加和相乘。

What makes it hard for experts to know what's going on

专家很难知道发生了什么的原因

is basically just, there are too many numbers,

基本上只是数字太多了，

and we don't yet have good ways of teasing apart what they're all doing.

而且我们还没有很好的方法 来分辨它们在做什么。

There's a little bit more to it than that, but not a lot.

还有更多的东西，但不是很多。

So how do we govern this technology

那么，我们如何管理这项 我们难以理解和预测的技术呢？

that we struggle to understand and predict?

I'm going to share two ideas.

我要分享两个想法。

One for all of us and one for policymakers.

一个给我们所有人， 一个给决策者。

First, don't be intimidated.

首先，不要被吓倒。

Either by the technology itself

要么通过技术本身，

or by the people and companies building it.

要么由开发它的人和公司来做。

On the technology,

在技术方面，

AI can be confusing, but it's not magical.

人工智能或许令人疑惑， 但它并不是空中楼阁。

There are some parts of AI systems we do already understand well,

人工智能系统的某些部分 我们已经很了解了，

and even the parts we don't understand won't be opaque forever.

即使是我们不了解的部分 也不会永远不透明。

An area of research known as “AI interpretability”

一个被称为 “人工智能可解释性”的研究领域，

has made quite a lot of progress in the last few years

在过去的几年中，

in making sense of what all those billions of numbers are doing.

在理解这数十亿数字的作用方面 取得了长足的进展。

One team of researchers, for example,

例如，一组研究人员

found a way to identify different parts of a neural network

找到了一种识别神经网络 不同部分的方法，

that they could dial up or dial down

他们可以向上或向下调节这些部分，

to make the AI's answers happier or angrier,

从而使人工智能的答案 更快乐或更愤怒，

more honest,

更诚实或更狡诈等等。

more Machiavellian, and so on.

If we can push forward this kind of research further,

如果我们能进一步推进这种研究，

then five or 10 years from now,

那么五到十年后，

we might have a much clearer understanding of what's going on

我们可能会更清楚地了解

所谓的黑匣子里正在发生的事情。

inside the so-called black box.

And when it comes to those building the technology,

当涉及到技术开发人员时，

technologists sometimes act as though

技术人员有时会表现得

if you're not elbows deep in the technical details,

好像你对技术细节不甚了解，

then you're not entitled to an opinion on what we should do with it.

那么你就无权就我们应该 如何使用它发表意见。

Expertise has its place, of course,

当然，专业知识有其一席之地，

but history shows us how important it is

但历史向我们表明，

that the people affected by a new technology

受新技术影响的人们，

get to play a role in shaping how we use it.

在塑造我们的使用方式方面 发挥的作用是多么重要。

Like the factory workers in the 20th century who fought for factory safety,

就像 20 世纪为工厂安全 而战的工厂工人，

or the disability advocates

或者确保万维网 无障碍使用的残疾人倡导者。

who made sure the world wide web was accessible.

You don't have to be a scientist or engineer to have a voice.

你不必是科学家 或工程师就能有发言权。

(Applause)

（掌声）

Second, we need to focus on adaptability, not certainty.

其次，我们需要 关注适应性，而不是确定性。

A lot of conversations about how to make policy for AI

关于如何制定 人工智能政策的许多对话

get bogged down in fights between, on the one side,

都陷入了僵局，

一方面，人们说：“我们现在必须 非常严格地监管人工智能，

people saying, "We have to regulate AI really hard right now

because it's so risky."

因为它风险很大。”

And on the other side, people saying,

另一方面，人们说：

“But regulation will kill innovation, and those risks are made up anyway.”

“但是监管会扼杀创新， 而这些风险无论如何都是虚构的。”

But the way I see it,

但是从我的角度来看，

it’s not just a choice between slamming on the brakes

这不仅仅是刹车 或加油之间的选择。

or hitting the gas.

If you're driving down a road with unexpected twists and turns,

如果你开车行驶在一条 有着意想不到弯道的路上，

then two things that will help you a lot

那么有两件事对你有很大帮助，

are having a clear view out the windshield

那就是挡风玻璃的清晰视野

and an excellent steering system.

和出色的转向系统。

In AI, this means having a clear picture of where the technology is

在人工智能中，这意味着要清楚地 了解技术的进展和发展方向，

and where it's going,

and having plans in place for what to do in different scenarios.

并为在不同场景中 该做什么制定计划。

Concretely, this means things like investing in our ability to measure

具体而言，这意味着向我们 评估人工智能系统功能的能力投资。

what AI systems can do.

This sounds nerdy, but it really matters.

这听起来很书呆子，但确实很重要。

Right now, if we want to figure out

现在，如果我们想弄清楚

whether an AI can do something concerning,

人工智能是否会做 一些令人担忧的事情，

like hack critical infrastructure

比如入侵关键基础设施

or persuade someone to change their political beliefs,

或说服某人改变他们的政治信仰，

our methods of measuring that are rudimentary.

我们的评估方法还很原始。

We need better.

我们需要更先进的。

We should also be requiring AI companies,

我们还应该要求人工智能公司，

especially the companies building the most advanced AI systems,

尤其是构建 最先进的人工智能系统的公司，

to share information about what they're building,

共享有关他们正在构建什么、

what their systems can do

他们的系统可以做什么

and how they're managing risks.

以及如何管理风险的信息。

And they should have to let in external AI auditors to scrutinize their work

而且，他们应该让外部 人工智能审计师来审查他们的工作，

so that the companies aren't just grading their own homework.

这样公司就不会只是在 给自己的作业打分。

(Applause)

（掌声）

A final example of what this can look like

最后一个例子

is setting up incident reporting mechanisms,

是建立事件报告机制，

so that when things do go wrong in the real world,

这样，当现实世界出现问题时，

we have a way to collect data on what happened

我们就有办法收集 有关所发生的事情

and how we can fix it next time.

以及下次如何修复的数据。

Just like the data we collect on plane crashes and cyber attacks.

就像我们收集的飞机失事 和网络攻击数据一样。

None of these ideas are mine,

这些想法并不是我个人的，

and some of them are already starting to be implemented in places like Brussels,

其中一些已经开始在布鲁塞尔、伦敦、 甚至华盛顿等地付诸实施。

London, even Washington.

But the reason I'm highlighting these ideas,

但我之所以强调这些想法，

measurement, disclosure, incident reporting,

评估、披露、事件报告，

is that they help us navigate progress in AI

是因为它们为我们提供了更清晰的视角，

by giving us a clearer view out the windshield.

帮助我们在人工智能领域取得进展。

If AI is progressing fast in dangerous directions,

如果人工智能朝着 危险的方向快速发展，

these policies will help us see that.

这些政策将帮助我们看到这一点。

And if everything is going smoothly, they'll show us that too,

如果一切顺利， 它们也会向我们展示这一点，

and we can respond accordingly.

我们可以做出相应的回应。

What I want to leave you with

我想告诉你们的是，

is that it's both true that there's a ton of uncertainty

人工智能领域确实 存在大量的不确定性和分歧。

and disagreement in the field of AI.

And that companies are already building and deploying AI

这些公司以影响我们所有人的方式

all over the place anyway in ways that affect all of us.

在各地构建和部署人工智能。

Left to their own devices,

任由他们自己决定。

it looks like AI companies might go in a similar direction

看来人工智能公司可能会

to social media companies,

朝着与社交媒体公司相似的方向发展，

spending most of their resources on building web apps

将大部分资源花 在构建网络应用程序

and for users' attention.

和吸引用户注意力上。

And by default, it looks like the enormous power of more advanced AI systems

默认情况下，看起来像是 更先进的人工智能系统的巨大力量，

might stay concentrated in the hands of a small number of companies,

可能会集中在少数公司 甚至少数个人手中。

or even a small number of individuals.

But AI's potential goes so far beyond that.

但是人工智能的潜力远不止于此。

AI already lets us leap over language barriers

人工智能已经让我们跨越语言障碍

and predict protein structures.

并预测蛋白质结构。

More advanced systems could unlock clean, limitless fusion energy

更先进的系统可以释放清洁、 无限的聚变能量

or revolutionize how we grow food

或彻底改变我们生产食物或 其他 1,000 种东西的方式。

or 1,000 other things.

And we each have a voice in what happens.

我们每个人对所发生的事都有发言权。

We're not just data sources,

我们不仅是数据源，

we are users,

我们是用户，

we're workers,

我们是员工，

we're citizens.

我们是公民。

So as tempting as it might be,

因此，尽管它可能很诱人，

we can't wait for clarity or expert consensus

但我们迫不及待地想得到澄清 或专家共识

to figure out what we want to happen with AI.

来弄清楚我们 想用人工智能实现什么。

AI is already happening to us.

人工智能已然降临到我们身上，

What we can do is put policies in place

我们能做的是制定政策， 让我们尽可能清楚地了解

to give us as clear a picture as we can get

of how the technology is changing,

技术是如何变化的，

and then we can get in the arena and push for futures we actually want.

然后我们就可以参与其中， 推动我们真正想要的未来。

Thank you.

谢谢。

(Applause)

（掌声）