Your Right to Repair AI Systems | Rumman Chowdhury | TED

譯者: C Leung 審譯者: 麗玲 辛

I want to tell you a story

我想告訴你一個

about artificial intelligence and farmers.

關於人工智慧和農民的故事。

Now, what a strange combination, right?

多麼奇怪的組合，對吧？

Two topics could not sound more different from each other.

兩個主題聽起來風馬牛不相及，

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

但你知道現代農業 其實涉及很多科技嗎？

So computer vision is used to predict crop yields.

電腦視覺用於預測農產量、

And artificial intelligence is used to find,

人工智慧用於發現、 識別和消滅昆蟲，

identify and get rid of insects.

Predictive analytics helps figure out extreme weather conditions

預測分析有助於找出 乾旱或颶風等

like drought or hurricanes.

極端天氣情況。

But this technology is also alienating to farmers.

但科技也會讓農民感到格格不入。

And this all came to a head in 2017

在 2017 年，拖拉機公司 約翰·迪爾 (John Deere)

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

推出智慧拖拉機時， 一切都到達緊要關頭。

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

在以前，如果農民的拖拉機壞了，

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

他們可以自行或帶給技工修理。

Well, the company actually made it illegal

而該公司實際上

for farmers to fix their own equipment.

將農民自行修理設備視為非法。

You had to use a licensed technician

農民必須找有執照的技術人員維修，

and farmers would have to wait for weeks

即使當農作物腐爛、害蟲侵佔，

while their crops rot and pests took over.

他們也得等上數星期。

So they took matters into their own hands.

所以他們親力親為。

Some of them learned to program,

有些人學會了編程，

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

並且與駭客合作， 建立修補程序來修復自己的系統。

In 2022,

2022 年，

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

在世上最大的駭客會議 DEFCON 上，

a hacker named Sick Codes and his team

名為 Sick Codes 的 駭客和他的團隊

showed everybody how to break into a John Deere tractor,

向大家展示了如何入侵 約翰迪爾拖拉機，

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

首先指出，該技術易受攻擊，

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

同時也表明，你可以、 也應該擁有自己的設備。

To be clear, this is illegal,

明確地說，這是非法的，

but there are people trying to change that.

但有些人想改變這點。

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

這個行動被稱為「維修權」。

The right to repair goes something like this.

維修權是這樣：

If you own a piece of technology,

如果你擁有一件科技產品，

it could be a tractor, a smart toothbrush,

可能是拖拉機 、智慧牙刷 、

a washing machine,

洗衣機，

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

萬一損壞，你應有權去維修它。

So why am I telling you this story?

那我為何要告訴你這故事？

The right to repair needs to extend to artificial intelligence.

維修權需要擴展到人工智慧。

Now it seems like every week

現在，似乎每週

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

都有令人振奮的 AI 創新。

But did you know that public confidence is actually declining?

但你知道公眾信心其實正下降嗎？

A recent Pew poll showed that more Americans are concerned

最近的皮尤民意調查 (Pew) 顯示，

美國人對該技術的擔心多於興奮。

than they are excited about the technology.

This is echoed throughout the world.

世界各地都有同感。

The World Risk Poll shows

世界風險調查顯示，

that respondents from Central and South America and Africa

來自中南美洲和非洲的受訪者，

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

大都表示認為 AI 對人們是弊大於利。

As a social scientist and an AI developer,

作為社會科學家和 AI 開發者，

this frustrates me.

這令我沮喪。

I'm a tech optimist

我是技術樂觀主義者，

because I truly believe this technology can lead to good.

因為我真誠相信 這種技術可以帶來益處。

So what's the disconnect?

那當中障礙是什麼？

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

我在過去幾年裡和數百人談過，

Architects and scientists, journalists and photographers,

建築師和科學家 、 記者和攝影師、

ride-share drivers and doctors,

共乘車司機和醫生，

and they all say the same thing.

他們都有同樣想法。

People feel like an afterthought.

他們覺得自己後知後覺。

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

他們都清楚自己的數據 通常在未經許可下被收集，

to create these sophisticated systems.

用來建立這些複雜的系統。

They know that these systems are determining their life opportunities.

他們知道這些系統 正決定他們的命運。

They also know that nobody ever bothered to ask them

他們也知道， 沒有人會費心問他們

how the system should be built,

系統該如何建構，

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

他們當然不知道如果問題出現時， 應該去找誰負責。

We may not own AI systems,

我們也許不能擁有 AI 系統，

but they are slowly dominating our lives.

但它們正慢慢地主導我們的生活。

We need a better feedback loop

我們需要更好的反饋迴路，

between the people who are making these systems,

在系統製造者

and the people who are best determined to tell us

及普羅大眾之間架設渠道，

how these AI systems should interact in their world.

去使 AI 系統更好地融入人們生活。

One step towards this is a process called red teaming.

邁向這目標的一步， 是靠稱為紅隊演練的過程。

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

紅隊演練是從軍事而來的一種做法，

and it's used in cybersecurity.

應用於網絡安全。

In a traditional red-teaming exercise,

在傳統的紅隊演練過程中，

external experts are brought in to break into a system,

外部專家被招募去入侵系統，

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

如同 Sick Codes 對拖拉機 所做的事， 但是合法。

So red teaming acts as a way of testing your defenses

因此，紅隊演練可以作為 測試防禦的方法，

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

當你能找出問題所在時，

you can figure out how to fix it.

你就能找出解決方案。

But when AI systems go rogue,

但當 AI 系統變得異常，

it's more than just a hacker breaking in.

它不只是駭客入侵般簡單。

The model could malfunction or misrepresent reality.

該模型可能發生故障或歪曲現實。

So, for example, not too long ago,

例如不久之前，我們看到 AI 系統

we saw an AI system attempting diversity

使用不準確的歷史照片 來嘗試展現多元。

by showing historically inaccurate photos.

Anybody with a basic understanding of Western history

任何對西方歷史 有基本了解的人都知道，

could have told you that neither the Founding Fathers

美國開國元勛或納粹時代的 士兵都不會是黑人。

nor Nazi-era soldiers would have been Black.

In that case, who qualifies as an expert?

在這種情況下，誰有資格成為專家？

You.

你。

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

我正與世界各地成千上萬的人合作，

on large and small red-teaming exercises,

進行各種大小的紅隊演練，

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

通過這些演練，我們發現 並修正了 AI 模型的錯誤。

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

我們還與世界上 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.

通過這些演練，我們使模型 更適合給更多人使用。

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

以下是我們所學到的。

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

我們與倫敦皇家學會合作，

和疾病科學家舉辦一場關於 科學、錯誤和虛假信息的活動。

mis- and disinformation event with disease scientists.

What these scientists found

這些科學家發現到，

is that AI models actually had a lot of protections

AI 模型實際上具有

很多針對新冠肺炎 錯誤信息的保護措施。

against COVID misinformation.

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

但對於麻疹，腮腺炎和 流感等其他疾病，

the same protections didn't apply.

同樣的保護措施並不適用。

We reported these changes,

我們舉報該有這些改變，

它們已被修復， 現在我們都得到更好的保護，

they’re fixed and now we are all better protected

against scientific mis- and disinformation.

免受科學錯誤和虛假信息的侵害。

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

我們與 Autodesk 大學的建築師 進行了一項非常類似的練習，

and we asked them a simple question:

我們問了個簡單問題：

Will AI put them out of a job?

AI 會令他們失業嗎？

Or more specifically,

更具體說，

could they imagine a modern AI system

他們能否想像一個能夠設計

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

現代藝術博物館規格的 現代 AI 系統？

The answer, resoundingly, was no.

答案清脆利落，是否定的。

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

原因是建築師不單單是繪製建築物，

They have to understand physics and material science.

他們必須了解物理和材料科學。

They have to know building codes,

他們必須清楚建築規範，

and they have to do that

並且同時，還要能 喚起情感，引發大眾共鳴。

while making something that evokes emotion.

What the architects wanted was an AI system

建築師想要的是

與他們互動的 AI 系統， 可以給他們反饋，

that interacted with them, that would give them feedback,

maybe proactively offer design recommendations.

或主動提供設計建議，

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

而現今的 AI 系統仍未達標。

But those are technical problems.

但這些是技術問題。

People building AI are incredibly smart,

建立 AI 的人非常聰明，

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

也許他們可以在幾年內 解決所有問題。

But that wasn't their biggest concern.

但這並非是建築師的最大擔憂。

Their biggest concern was trust.

他們最大的擔憂是信任。

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

若建築物出現問題， 建築師就要承擔責任。

They could lose their license,

他們可能會失去執照，

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

可能被罰款， 甚至有牢獄之災。

And failures can happen in a million different ways.

失敗能以百萬種不同方式發生。

For example, exit doors that open the wrong way,

例如，逃生門的打開方向出錯，

leading to people being crushed in an evacuation crisis,

導致人們在緊急疏散中被擠壓，

or broken glass raining down onto pedestrians in the street

或因為強風打破窗戶，

玻璃碎落在街道上的行人身上。

because the wind blows too hard and shatters windows.

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

那為何建築師要信任一個 AI 系統，

押上自己的工作和人身自由，

with their literal freedom,

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

但當發現系統錯誤時 ， 卻不能親自修復它？

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

所以我們需要弄清楚這些問題， 我會告訴你原因。

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

下一波人工智慧系統， 稱為代理型 AI ，

is a true tipping point

是個真正轉折點，

between whether or not we retain human agency,

要麼我們保留人類自主性，

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

要麼 AI 系統替我們作出決策。

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

想想一個像個人助理般的 AI 代理人。

So, for example, a medical agent might determine

例如，醫療代理可能會判斷

whether or not your family needs doctor's appointments,

你的家人是否需要醫生預約、

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

補充處方藥物、 或在緊急情況下

send medical records to the hospital.

將醫療記錄發送到醫院。

But AI agents can't and won't exist

但除非我們有真正的維修權，

unless we have a true right to repair.

否則 AI 代理無法，也不會存在。

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

除非可以執行一些基本診斷，

否則哪有家長會將孩子的健康 交託給 AI 系統？

unless you could run some basic diagnostics?

What professional would trust an AI system with job decisions,

除非可以按照初級員工的方式 重新培訓 AI，

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

否則哪有專業人員 會信任 AI 系統進行工作決策？

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

維修權可能看起來像這樣。

You could have a diagnostics board

有一個診斷板，

where you run basic tests that you design,

你可以在其中執行 自己設計的基本測試，

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

如果出現問題，則可以向該公司報告，

並在問題修復後得到回覆。

and hear back when it's fixed.

Or you could work with third parties like ethical hackers

或者你可以與第三方合作， 例如道德駭客，

who make patches for systems like we do today.

為系統製作修補程序， 跟我們所做一樣。

You can download them and use them to improve your system

你可以下載並使用它們， 按你想要的方式改進系統。

the way you want it to be improved.

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

或者你可以像這些堅毅的農民一樣，

學習編程和微調自己的系統。

and fine-tune your own systems.

We won't achieve the promised benefits of artificial intelligence

除非我們弄清楚 如何將人們帶入開發過程，

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

否則我們將無法實現 人工智慧所承諾的益處。

I've dedicated my career to responsible AI,

我投入「負責任 AI」 的事業，

and in that field we ask the question,

在這個領域我們提出問題，

what can companies build to ensure that people trust AI?

「企業能建立什麼 來確保人們信任 AI？」

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

現在，通過這些紅隊演練 及跟你們的交談，

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

我發現我們一直問錯問題了。

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

我們應該問的是：

「我們能建立哪些工具， 讓人們可以使用 AI 而獲益？」

so people can make AI beneficial for them?

Technologists can't do it alone.

技術人員無法獨自做到；

We can only do it with you.

我們只能與你一起做到。

Thank you.

謝謝。

(Applause)

（掌聲）