Your Right to Repair AI Systems | Rumman Chowdhury | TED

翻译人员: Melody Hui 校对人员: suya f.

I want to tell you a story

我想给你讲一个

about artificial intelligence and farmers.

关于人工智能与农民的故事。

Now, what a strange combination, right?

这听起来是个奇怪的组合，对吧？

Two topics could not sound more different from each other.

这两个话题听起来简直完全不相干。

But did you know that modern farming actually involves a lot of technology?

但你知道现代农业 其实与科技息息相关吗？

So computer vision is used to predict crop yields.

比如，计算机视觉 用于预测作物产量。

And artificial intelligence is used to find,

人工智能用于发现、

identify and get rid of insects.

识别和消灭昆虫。

Predictive analytics helps figure out extreme weather conditions

预测分析有助于找出干旱，飓风等

like drought or hurricanes.

极端天气状况。

But this technology is also alienating to farmers.

但是这项技术 也疏远了农民。

And this all came to a head in 2017

在 2017 年， 这一切到达了顶峰。

with the tractor company John Deere when they introduced smart tractors.

拖拉机公司约翰迪尔 推出了智能拖拉机，

So before then, if a farmer's tractor broke,

在此之前，如果农民的拖拉机坏了，

they could just repair it themselves or take it to a mechanic.

他们可以自己修理或者交给机械师。

Well, the company actually made it illegal

虽然，该公司实际上 将农民私自维修设备

for farmers to fix their own equipment.

视为非法。

You had to use a licensed technician

农民必须聘请有执照的技术人员。

and farmers would have to wait for weeks

农民将不得不等上好几个星期， 直到他们的农作物腐烂，

while their crops rot and pests took over.

虫害已然占据上风的时候。

So they took matters into their own hands.

因此，他们想把事情掌握在自己手中。

Some of them learned to program,

他们中的一些人学会了编程，

and they worked with hackers to create patches to repair their own systems.

他们与黑客合作开发了 补丁来修复自己的系统。

In 2022,

在 2022 年，

at one of the largest hacker conferences in the world, DEFCON,

在世界上最大的 黑客会议之一 DEFCON 上

a hacker named Sick Codes and his team

一位名叫 Sick Codes 的黑客 和他的团队向所有人

showed everybody how to break into a John Deere tractor,

展示了如何闯入约翰迪尔拖拉机，

showing that, first of all, the technology was vulnerable,

这首先表明了，该技术很脆弱，

but also that you can and should own your own equipment.

但是，这也代表 你可以研发自己的设备了。

To be clear, this is illegal,

当然，这是非法的，

but there are people trying to change that.

但有人试图改变这种状况。

Now that movement is called the “right to repair.”

现在，该运动被称为 “维修权”。

The right to repair goes something like this.

维修权是这样的。

If you own a piece of technology,

如果你拥有一项技术，

it could be a tractor, a smart toothbrush,

它可能是拖拉机、智能牙刷、

a washing machine,

洗衣机，如果它坏了，

you should have the right to repair it if it breaks.

你应该 有权修理它。

So why am I telling you this story?

那我为什么要告诉你这个故事？

The right to repair needs to extend to artificial intelligence.

维修权需要扩展 到人工智能。

Now it seems like every week

现在，人工智能似乎每周

there is a new and mind-blowing innovation in AI.

都有一项令人兴奋的新创新。

But did you know that public confidence is actually declining?

但是你知道公众的信心 实际上在下降吗？

A recent Pew poll showed that more Americans are concerned

皮尤最近的一项民意调查 显示，美国人对这项技术的

than they are excited about the technology.

担忧多于兴奋

This is echoed throughout the world.

这在全世界都得到了共鸣。

The World Risk Poll shows

世界风险民意调查显示

that respondents from Central and South America and Africa

来自中南美洲和非洲的受访者

all said that they felt AI would lead to more harm than good for their people.

都表示，他们认为人工智能 对人民带来的弊大于利。

As a social scientist and an AI developer,

作为一名社会科学家和 人工智能开发人员，

this frustrates me.

这让我感到沮丧。

I'm a tech optimist

我是一个技术乐观主义者，

because I truly believe this technology can lead to good.

因为我真的相信 这项技术可以带来好处。

So what's the disconnect?

那么，脱节是如何发生的？

Well, I've talked to hundreds of people over the last few years.

事实上，在过去的几年里， 我和数百人谈过话。

Architects and scientists, journalists and photographers,

建筑师和科学家、 记者和摄影师、

ride-share drivers and doctors,

拼车司机和医生，

and they all say the same thing.

他们都有同样的想法。

People feel like an afterthought.

他们都有同样的后知后觉。

They all know that their data is harvested often without their permission

他们都知道，他们的数据 通常是在未经允许的情况下收集的，

to create these sophisticated systems.

用于创建这些复杂的系统。

They know that these systems are determining their life opportunities.

他们知道这些系统 正在改变他们的生活。

They also know that nobody ever bothered to ask them

他们也知道从来没有人过问他们的意见

how the system should be built,

来如何构建系统，

and they certainly have no idea where to go if something goes wrong.

他们也无从得知如果 系统出现问题该去哪里解决。

We may not own AI systems,

我们可能不拥有人工智能系统，

but they are slowly dominating our lives.

但它们正在慢慢主导我们的生活。

We need a better feedback loop

我们需要建立一个良好的反馈循环

between the people who are making these systems,

在系统研发者

and the people who are best determined to tell us

与大众之间构建一个桥梁

how these AI systems should interact in their world.

来帮助AI系统更好地融入大众的生活

One step towards this is a process called red teaming.

而“红队合作” ，是我们 迈向这个目标的一步。

Now, red teaming is a practice that was started in the military,

红队合作是一种源于军队的做法，

and it's used in cybersecurity.

现在，它被用于网络安全。

In a traditional red-teaming exercise,

在传统的红队活动中，

external experts are brought in to break into a system,

外部专家被邀请进入一个系统，

sort of like what Sick Codes did with tractors, but legal.

有点像 Sick Codes 对拖拉机所做的那样，但合法。

So red teaming acts as a way of testing your defenses

因此，红队合作是测试 你的防御能力的一种方式，

and when you can figure out where something will go wrong,

当你弄清楚 哪里会出现问题时，

you can figure out how to fix it.

你就能想出解决问题的方法。

But when AI systems go rogue,

但是，当人工智能系统变成盗贼时。

it's more than just a hacker breaking in.

不仅仅是黑客的入侵。

The model could malfunction or misrepresent reality.

该模型可能会出现故障或歪曲现实。

So, for example, not too long ago,

因此，举例来说，不久前，

we saw an AI system attempting diversity

我们看到一个人工智能系统试图

by showing historically inaccurate photos.

通过显示历史上不准确 的照片来实现多样性。

Anybody with a basic understanding of Western history

任何对西方历史有基本了解的人

could have told you that neither the Founding Fathers

都可以告诉你无论是 开国元勋还是纳粹时代的士兵，

nor Nazi-era soldiers would have been Black.

都不会是黑人。

In that case, who qualifies as an expert?

在这种情况下， 谁有资格成为专家？

You.

你。

I'm working with thousands of people all around the world

我正在与世界各地成千上万的人

on large and small red-teaming exercises,

一起进行各规模的红队练习，

and through them we found and fixed mistakes in AI models.

通过不断的练习，我们发现 并修复了AI 模型中的缺陷。

We also work with some of the biggest tech companies in the world:

我们还与世界上一些最大的 科技公司合作：

OpenAI, Meta, Anthropic, Google.

OpenAI、Meta、Anthropic、谷歌。

And through this, we've made models work better for more people.

通过这种方式，

我们使模型更好地 适用于更多的人。

Here's a bit of what we've learned.

以下是我们学到的一些东西。

We partnered with the Royal Society in London to do a scientific,

我们与伦敦皇家学会合作，与疾病科家

mis- and disinformation event with disease scientists.

一起举办了一次 关于科学与虚假信息的活动。

What these scientists found

这些科学家发现，

is that AI models actually had a lot of protections

人工智能模型实际上具有很多

against COVID misinformation.

针对 COVID 错误信息的保护措施。

But for other diseases like measles, mumps and the flu,

但同样保护措施并不适用于

the same protections didn't apply.

麻疹、流行性 腮腺炎和流感等其他疾病，

We reported these changes,

我们报告了这些变化，

they’re fixed and now we are all better protected

它们得到了改进， 而我们都得到了更好的保护，

against scientific mis- and disinformation.

让我们免受科学错误 和虚假信息的侵害。

We did a really similar exercise with architects at Autodesk University,

我们对 Autodesk 大学的建筑师 做了一个非常相似的练习，

and we asked them a simple question:

我们问了他们一个简单的问题：

Will AI put them out of a job?

人工智能会让他们失业吗？

Or more specifically,

或者更具体地说，

could they imagine a modern AI system

他们能否想象一个 能够设计现代艺术博物馆的

that would be able to design the specs of a modern art museum?

现代人工智能系统？

The answer, resoundingly, was no.

答案很明显，是否定的。

Here's why, architects do more than just draw buildings.

这就是建筑师所做的 不仅仅是画建筑物的原因。

They have to understand physics and material science.

他们必须了解物理学 和材料科学。

They have to know building codes,

他们必须了解建筑规范，

and they have to do that

而且他们必须制作

while making something that evokes emotion.

引起大众共鸣的东西。

What the architects wanted was an AI system

建筑师他们想要的是一个 可以与他们互动的人工智能系统，

that interacted with them, that would give them feedback,

可以向他们提供反馈，

maybe proactively offer design recommendations.

也许可以主动提供设计建议。

And today's AI systems, not quite there yet.

而当今的人工智能系统， 还没有完全到来。

But those are technical problems.

但这些都是技术问题。

People building AI are incredibly smart,

构建人工智能的人非常聪明，

and maybe they could solve all that in a few years.

也许他们可以在几年内解决 所有这些问题。

But that wasn't their biggest concern.

但这并不是他们最关心的问题。

Their biggest concern was trust.

他们最担心的是信任。

Now architects are liable if something goes wrong with their buildings.

现在，如果建筑物出现问题， 建筑师要承担责任。

They could lose their license,

他们可能会失去执照，

they could be fined, they could even go to prison.

可能会被处以罚款， 甚至可能入狱。

And failures can happen in a million different ways.

失败可能 以一百万种不同的方式发生。

For example, exit doors that open the wrong way,

例如，出口门 开错了方向，

leading to people being crushed in an evacuation crisis,

导致人们在疏散危机 中被压伤，

or broken glass raining down onto pedestrians in the street

或者

because the wind blows too hard and shatters windows.

由于风吹得太厉害，窗户被打碎， 碎玻璃落到街上的行人身上。

So why would an architect trust an AI system with their job,

那么，建筑师要如何信任人工智能 来完成自己的工作

with their literal freedom,

他们会失去自主权

if they couldn't go in and fix a mistake if they found it?

无法干预或修正 AI 的在工作上失误

So we need to figure out these problems today, and I'll tell you why.

因此，我们今天 需要弄清楚这些问题，

The next wave of artificial intelligence systems, called agentic AI,

下一波人工智能系统， 即代理人工智能，

is a true tipping point

是一个真正的转折点。

between whether or not we retain human agency,

要么保留人类能动性，

or whether or not AI systems make our decisions for us.

要么人工智能系统来 完全为我们做出决策。

Imagine an AI agent as kind of like a personal assistant.

想象一下 AI 代理有点像私人助理。

So, for example, a medical agent might determine

因此，举例来说， 医疗代理人可能会决定你的家人

whether or not your family needs doctor's appointments,

是否需要预约医生，

it might refill prescription medications, or in case of an emergency,

它可能会补充处方药， 或者在紧急情况下向医院

send medical records to the hospital.

发送病历。

But AI agents can't and won't exist

但是，人工智能代理 无法真正地诞生

unless we have a true right to repair.

如果我们没有真正的修复权

What parent would trust their child's health to an AI system

没有家长会将孩子的健康委托 给人工智能系统

unless you could run some basic diagnostics?

除非你能进行一些基本的判断

What professional would trust an AI system with job decisions,

哪位专业人士会相信人工智能 能做出专业的决策

unless you could retrain it the way you might a junior employee?

除非你能对人工智能系统 进行专业培训？

Now, a right to repair might look something like this.

现在，维修权 可能看起来像这样。

You could have a diagnostics board

你可以有一个诊断板

where you run basic tests that you design,

来运行你设计的基本测试，

and if something's wrong, you could report it to the company

如果出现问题， 你可以向公司报告，

and hear back when it's fixed.

并在修复后收到回复。

Or you could work with third parties like ethical hackers

或者你可以与第三方合作， 比如道德黑客，

who make patches for systems like we do today.

他们像我们今天一样 为系统制作补丁。

You can download them and use them to improve your system

你可以下载它们并使用它们

the way you want it to be improved.

来按照你想要的方式改进系统。

Or you could be like these intrepid farmers and learn to program

或者你可以像这些干劲十足的 农民一样，学会编程

and fine-tune your own systems.

和微调自己的系统。

We won't achieve the promised benefits of artificial intelligence

我们需要想出如何让人们 参与开发过程，

unless we figure out how to bring people into the development process.

否则我们将无法实现 人工智能所承诺的好处。

I've dedicated my career to responsible AI,

我的职业生涯一直 致力于负责任的人工智能，

and in that field we ask the question,

在这个领域我们问一个问题， 企业可以建立

what can companies build to ensure that people trust AI?

什么来确保人们信任人工智能？

Now, through these red-teaming exercises, and by talking to you,

现在，通过这些红队练习， 通过与你们的交谈，

I've come to realize that we've been asking the wrong question all along.

我意识到我们一直在问错误的问题。

What we should have been asking is what tools can we build

我们应该问的是，我们可以开发

so people can make AI beneficial for them?

哪些工具，让人们能够让 人工智能为他们带来好处？

Technologists can't do it alone.

技术人员无法独自完成这项工作。

We can only do it with you.

我们只能和你一起做。

Thank you.

谢谢。

(Applause)

（掌声）