Why AI Needs a “Nutrition Label” | Kasia Chmielinski | TED

譯者: 麗玲 辛

Now, I haven't met most of you or really any of you,

我雖然還沒認識在座的多數人， 甚至可以說根本沒見過面，

but I feel a really good vibe in the room.

但我感覺這裡的氣氛很好。

(Laughter)

（笑聲）

And so I think I'd like to treat you all to a meal.

所以我想請大家吃頓飯。

What do you think?

你們覺得如何？

Yes? Great, so many new friends.

好嗎？太好了，這麼多新朋友。

So we're going to go to this cafe,

我們要去這家咖啡館，

they serve sandwiches.

他們那裡有賣三明治，

And the sandwiches are really delicious.

而且非常好吃。

But I have to tell you that sometimes they make people really, really sick.

但我得跟你說，有時候這些三明治 會讓人非常非常不舒服。

(Laughter)

（笑聲）

And we don't know why.

我們也不知道為什麼，

Because the cafe won't tell us how they make the sandwich,

因為咖啡館不告訴我們 他們是怎麼做三明治的，

they won't tell us about the ingredients.

也不告訴我們用的食材是什麼。

And then the authorities have no way to fix the problem.

這樣一來，相關部門也 無法解決這個問題。

But the offer still stands.

不過，我的邀請仍然有效。

So who wants to get a sandwich?

誰想吃三明治？

(Laughter)

（笑聲）

Some brave souls, we can talk after.

有些勇敢的人，我們可以事後再聊。

But for the rest of you, I understand.

至於其他人，我能理解，

You don't have enough information

因為你們沒有足夠的資訊

to make good choices about your safety

來做出自身安全的好選擇，

or even fix the issue.

甚至也無法解決這個問題。

Now, before I further the anxiety here, I'm not actually trying to make you sick,

我不想繼續加深大家的焦慮，

我也不想讓你們不舒服，

but this is an analogy to how we're currently making algorithmic systems,

這只是個比喻，來說明 我們目前製作演算系統的方式，

also known as artificial intelligence or AI.

也就是所謂的人工智慧 (AI)。

Now, for those who haven't thought about the relationship

至於從沒想過 AI 和三明治有什麼關係的人，

between AI and sandwiches, don't worry about it,

別擔心，

I'm here for you, I'm going to explain.

我會解釋給你聽。

You see, AI systems, they provide benefit to society.

AI 系統對社會有益，

They feed us,

它們餵養我們，

but they're also inconsistently making us sick.

但它們也不時讓我們生病。

And we don't have access to the ingredients that go into the AI.

我們無法得知 AI 的原料，

And so we can't actually address the issues.

所以也無法真正解決問題。

We also can't stop eating AI

我們也無法像不吃有疑慮的 三明治那樣停止使用 AI，

like we can just stop eating a shady sandwich

because it's everywhere,

因為它無處不在，

and we often don't even know that we're encountering a system

而且我們甚至常常不知道 自己在使用演算系統。

that's algorithmically based.

So today, I'm going to tell you about some of the AI trends that I see.

今天，我會告訴你們 一些我所看到的 AI 趨勢。

I'm going to draw on my experience building these systems

我會根據過去二十年來 我建立這些系統的經驗，

over the last two decades to tell you about the tools

介紹我和其他人所開發的工具，

that I and others have built to look into these AI ingredients.

用來檢視這些 AI 的成分。

And finally, I'm going to leave you with three principles

最後，我會提供三個原則，

這些原則能讓我們和那些建立 AI 的公司有更健康的關係。

that I think will give us a healthier relationship

to the companies that build artificial intelligence.

I'm going to start with the question, how did we get here?

我會從一個問題開始： 我們怎麼走到今天的情況？

AI is not new.

AI 並不是新事物。

We have been living alongside AI for two decades.

我們已經與 AI 共處了二十年。

Every time that you apply for something online,

每當你在線上申請某些東西、

you open a bank account or you go through passport control,

開設銀行帳戶或通過護照查驗，

you're encountering an algorithmic system.

你都會接觸到某個演算系統。

We've also been living with the negative repercussions of AI for 20 years,

過去二十年來，我們也一直 在承受 AI 帶來的負面影響，

and this is how it makes us sick.

這就是它讓我們生病的原因。

These systems get deployed on broad populations,

這些系統被部署在廣大人群中，

and then certain subsets end up getting negatively disparately impacted,

然後某些群體會因此 受到負面的差異性影響，

usually on the basis of race or gender or other characteristics.

通常是基於種族、性別或其他特徵。

We need to be able to understand the ingredients to these systems

我們需要了解這些系統的成分，

so that we can address the issues.

才能解決問題。

So what are the ingredients to an AI system?

那麼，AI 系統的成分是什麼呢？

Well, data fuels the AI.

數據是 AI 的燃料。

The AI is going to look like the data that you gave it.

AI 的表現會反映你給它的數據。

So for example,

例如，

if I want to make a risk-assessment system for diabetes,

如果我想製作一個 糖尿病風險評估系統，

my training data set might be adults in a certain region.

我的訓練數據集 可能來自某個地區的成年人。

And so I'll build that system,

這樣，我建立的這個系統， 對這些成年人來說就很有用。

it'll work really well for those adults in that region.

但它對其他地區的成年人 或者小孩可能根本沒幫助。

But it does not work for adults in other regions

or maybe at all for children.

So you can imagine if we deploy this for all those populations,

你可以想像，如果我們把這個系統 部署到所有群體中，

會有很多人受到傷害。

there are going to be a lot of people who are harmed.

我們需要在使用數據前， 了解它的質量。

We need to be able to understand the quality of the data before we use it.

But I'm sorry to tell you that we currently live

但很遺憾地告訴你，

我們目前生活在 我名為數據的狂野西部時代。

in what I call the Wild West of data.

It's really hard to assess quality of data before you use it.

很難在使用數據前評估它的質量。

There are no global standards for data quality assessment,

沒有全球性的數據質量評估標準，

and there are very few data regulations around how you can use data

而且關於如何使用數據 和能使用哪些數據的法規也很少。

and what types of data you can use.

This is kind of like in the food safety realm.

這就像食品安全領域一樣，

If we couldn't understand where the ingredients were sourced,

如果我們無法了解原料的來源，

we also had no idea whether they were safe for us to consume.

也無法知道它們是否可安全食用。

We also tend to stitch data together,

我們還經常把數據拼湊在一起，

and every time we stitch this data together,

每次我們拼湊數據時，

無論是從網路上抓取、生成還是獲取，

which we might find on the internet, scrape, we might generate it,

we could source it.

We lose information about the quality of the data.

都會失去關於數據質量的資訊。

And the folks who are building the models

而建立模型的人

並不是那些找到數據的人，

are not the ones that found the data.

So there's further information that's lost.

所以資訊進一步丟失。

Now, I've been asking myself a lot of questions

我一直在問自己，

about how can we understand the data quality before we use it.

如何在使用數據前，了解數據質量。

And this emerges from two decades of building these kinds of systems.

這是源於我二十年來 建立這些系統的經驗。

The way I was trained to build systems is similar to how people do it today.

我之前受訓建立系統的方式 和今天的人們相似，

You build for the middle of the distribution.

你為常態分佈的中間群體建立系統，

That's your normal user.

那是你的常態用戶。

So for me, a lot of my training data sets

就我而言，我的很多訓練數據集

would include information about people from the Western world who speak English,

包括來自西方世界、說英語、

who have certain normative characteristics.

具有某些標準特徵的人的訊息。

And it took me an embarrassingly long amount of time

我花了很長時間才意識到

to realize that I was not my own user.

我自己並不是我設定的用戶。

So I identify as non-binary, as mixed race,

我自認為是非二元性別、混血種族，

I wear a hearing aid

戴助聽器，

and I just wasn't represented in the data sets that I was using.

而我在使用的數據集中 看不到我這樣的人。

And so I was building systems that literally didn't work for me.

所以我建立的系統根本不適合我。

And for example, I once built a system that repeatedly told me

例如，我曾經建立過一個系統，

反覆告訴我，我是個東歐白人女性。

that I was a white Eastern-European lady.

This did a real number on my identity.

這對我的身份認同造成了很大的傷害。

(Laughter)

（笑聲）

But perhaps even more worrying,

但或許更令人擔憂的是，

this was a system to be deployed in health care,

這個系統要部署在醫療領域，

where your background can determine things like risk scores for diseases.

依個人背景決定某些疾病的風險評分。

And so I started to wonder,

所以我開始思考，

can I build tools and work with others to do this

我能否建立工具，並與他人合作，

so that I can look inside of a dataset before I use it?

在使用數據前，檢查數據集？

In 2018, I was part of a fellowship at Harvard and MIT,

2018 年，我參加了哈佛和 麻省理工學院的研究員計畫，

and I, with some colleagues, decided to try to address this problem.

我和一些同事決定嘗試解決這個問題。

And so we launched the Data Nutrition Project,

我們推出了「數據營養計畫」，

which is a research group and also a nonprofit

這是一個研究團體， 也是個非營利組織，

that builds nutrition labels for datasets.

要建立數據集的營養標示。

So similar to food nutrition labels,

就像食品營養標示一樣，

the idea here is that you can look inside of a data set before you use it.

這個概念是讓你在使用數據集之前

能夠檢視它的內部，了解其成分，

You can understand the ingredients,

看看是否適合你的用途。

see whether it's healthy for the things that you want to do.

這是一個卡通化的標示版本。

Now this is a cartoonified version of the label.

The top part tells you about the completion of the label itself.

頂部告訴你標示本身的完成情況。

And underneath that you have information about the data,

在下方，你可以看到關於數據的訊息，

the description, the keywords, the tags,

包括描述、關鍵字、標籤，

and importantly, on the right hand side,

以及很重要的，在右側，說明 如何使用和不應使用這些數據。

how you should and should not use the data.

If you could scroll on this cartoon,

如果你捲動這個卡通標示，

you would see information about risks and mitigation strategies

你會看到關於風險 和緩解策略的各種訊息。

across a number of vectors.

And we launched this with two audiences in mind.

我們推出這個標示是針對兩類受眾。

The first audience are folks who are building AI.

第一類是建立 AI 的人，

So they’re choosing datasets.

他們是選擇數據集的人，

We want to help them make a better choice.

我們希望幫助他們做出更好的選擇。

The second audience are folks who are building datasets.

第二類是建立數據集的人。

And it turns out

事實證明，

that when you tell someone they have to put a label on something,

當你告訴某人他們必須 在某物上貼標示時，

他們會在製作之前考慮成分。

they think about the ingredients beforehand.

The analogy here might be,

這就像如果我想做無麩質三明治，

if I want to make a sandwich and say that it’s gluten-free,

I have to think about all the components as I make the sandwich,

我必須在製作過程中考慮所有的成分，

the bread and the ingredients, the sauces.

麵包和其他材料、餡料，

I can't just put it on a sandwich and put it in front of you

不能只是做好後、放到你面前，

and tell you it's gluten-free.

就說它是無麩質的。

We're really proud of the work that we've done.

我們對我們所做的工作感到非常自豪。

We launched this as a design and then a prototype

我們首先推出了設計，然後是原型，

and ultimately a tool for others to make their own labels.

最終成為一項工具， 讓其他人製作自己的標示。

And we've worked with experts at places like Microsoft Research,

我們與微軟研究院、

the United Nations and professors globally

聯合國及全球的教授等專家合作，

to integrate the label and the methodology

把這個標示和方法

into their work flows and into their curricula.

整合到他們的工作流程和課程中。

But we know it only goes so far.

但我們知道這還遠遠不夠。

And that's because it's actually really hard to get a label

因為要在每一個數據集上貼標示 實際上非常困難。

on every single dataset.

And this comes down to the question

這就歸結到這個問題： 為什麼要先在數據集上貼標示？

of why would you put a label on a dataset to begin with?

Well, the first reason is not rocket science.

第一個原因並不複雜，

It's that you have to.

就是因為必須這麼做。

And this is, quite frankly, why food nutrition labels exist.

老實說，這也是為什麼 食品營養標示存在的原因。

It's because if they didn't put them on the boxes, it would be illegal.

因為如果他們不在盒子上 貼標示，這就是違法的。

However, we don't really have AI regulation.

然而，我們並沒有 真正的 AI 法規，

We don't have much regulation around the use of data.

我們對數據使用的法規也很少。

Now there is some on the horizon.

不過，有些法規即將出現。

For example, the EU AI Act just passed this week.

例如，本週歐盟 剛通過了 AI 法案，

And although there are no requirements around making the training data available,

雖然沒有要求提供訓練數據，

they do have provisions for creating transparency labeling

但他們確實有關於創建 標示透明度的規定，

like the dataset nutrition label, data sheets, data statements.

比如數據集營養標示、 數據表和數據聲明。

There are many in the space.

在這個領域有很多這樣的做法。

We think this is a really good first step.

我們認為這是很好的一步。

The second reason that you might have a label on a dataset

第二個該在數據集上貼標示的原因，

is because it is a best practice or a cultural norm.

這是一種最佳實踐或文化規範。

這裡的例子可能是

The example here might be how we're starting to see

more and more food packaging and menus at restaurants

我們越來越常看到 食品包裝和餐廳菜單上

include information about whether there's gluten.

有關於是否含有麩質的訊息。

This is not required by law,

這不是法律要求的，

although if you do say it, it had better be true.

不過如果你說沒有， 那最好是真的如此。

And the reason that people are adding this to their menus

而人們把這些訊息加到他們的菜單

and their food packaging

和食品包裝上，

is because there's an increased awareness of the sensitivity

是因為人們對這種過敏或病症的敏感性

and kind of the seriousness of that kind of an allergy or condition.

和嚴重性有了更多的認識。

So we're also seeing some movement in this area.

在這領域，我們也看到了一些改變。

Folks who are building datasets are starting to put nutrition labels,

建立數據集的人開始在他們的數據集上

加上營養標示、數據表。

data sheets on their datasets.

And people who are using data are starting to request the information.

而使用數據的人 也開始要求這些資訊。

This is really heartening.

這真的讓人很振奮。

你可能會問，「Kasia， 那你為什麼還要來這裡？

And you might say, "Kasia, why are you up here?

一切看起來都很順利， 似乎越來越好了。」

Everything seems to be going well, seems to be getting better."

In some ways it is.

在某些方面，是的。

But I'm also here to tell you that our relationship to data

但我也要告訴你， 我們和數據的關係在惡化。

is getting worse.

Now the last few years have seen a supercharged interest

過去幾年，我們看到了 對收集數據集的興趣激增。

in gathering datasets.

Companies are scraping the web.

有公司在抓取網頁，

They're transcribing millions of hours of YouTube videos into text.

他們將數百萬小時的 YouTube 影片轉錄成文本。

By some estimates, they'll run out of information on the internet by 2026.

據一些估計，到 2026 年， 網際網路上的資訊將被抓取完。

They're even considering buying publishing houses

他們甚至考慮購買出版社，

so they can get access to printed text and books.

以獲取印刷文本和書籍的存取權。

So why are they gathering this information?

他們為什麼要收集這些資訊呢？

Well, they need more and more information

他們需要越來越多的資訊

to train a new technique called generative AI.

來訓練一種新技術， 叫做生成式 AI。

I want to tell you about the size of these datasets.

我想讓你了解這些數據集的規模。

If you look at GPT-3, which is a model that launched in 2020,

如果你看一下 2020 年發布的 GPT-3，

the training dataset included 300 billion words, or parts of words.

訓練數據集包括 三千億個單詞或字詞片段。

Now for context, the English language contains less than a million words.

作為參照，英語包含 不到一百萬個單詞。

Just three years later, DBRX was launched,

僅僅三年後，DBRX 發布， （開源大型語言模型）

which was trained on eight trillion words.

它的訓練數據集包括八萬億個單詞。

So 300 billion to eight trillion in three years.

所以三年內從三千億增加到八萬億。

And the datasets are getting bigger.

數據集越來越大。

Now with each successive model launch,

隨著每個新的模型發布，

the datasets are actually less and less transparent.

數據集變得越來越不透明。

And even we have access to the information,

即使我們有存取權，

it's so big, it's so hard to look inside without any kind of transparency tooling.

它也太大，很難在沒有任何 透明工具的情況下檢視內部。

And the generative AI itself is also causing some worries.

生成式 AI 本身 也引起了一些擔憂。

你可能通過 ChatGPT 接觸過這種技術。

And you've probably encountered this technique through ChatGPT.

I don't need to know what you do on the internet,

我不需要知道你在網上做了什麼，

那是你和網路之間的事，

但跟我一樣，你可能知道，

that's between you and the internet,

but you probably know, just like I do,

使用 ChatGPT 和其他 生成式 AI 技術創建訊息

how easy it is to create information using ChatGPT

and other generative AI technologies

並將其發布到網上是多麼容易。

and to put that out onto the web.

And so we're looking at a situation

所以我們正面臨著

大量的演算法生成的資訊， 但我們不會知道它們是演算法生成的，

in which we're going to encounter lots of information

that's algorithmically generated but we won't know it

也無法判斷其真實性。

and we won't know whether it's true.

這增加了 AI 潛在風險 和危害的規模。

And this increases the scale of the potential risks and harms from AI.

Not only that, I'm sorry,

不僅如此，抱歉，

再者，這些模型本身

but the models themselves are getting controlled

也受到越來越少數的 美國科技公司的控制。

by a smaller and smaller number of private actors in US tech firms.

這是去年，2023 年發布的模型。

So this is the models that were launched last year, in 2023.

你可以看到大多數是粉紅色的， 意味著它們來自工業界。

And you can see most of them are pink, meaning they came out of industry.

如果你仔細看，隨著時間， 越來越多來自工業界，

And if you look at this over time, more and more are coming out of industry

而來自包括學術界和政府在內的 其他部門越來越少，

and fewer and fewer are coming out of all the other sectors combined,

including academia and government,

而這些部門推出的技術 通常更容易受到審查。

where technology is often launched in a way

that's more easy to be scrutinized.

So if we go back to our cafe analogy,

如果我們回到咖啡館的比喻，

這就像少數私營公司

this is like you have a small number of private actors

擁有所有的成分，

who own all the ingredients,

他們在全球製作所有的三明治，

they make all the sandwiches globally,

而且沒有太多法規限制。

and there's not a lot of regulation.

這時你可能感到害怕，

And so at this point you're probably scared

甚至有點不舒服。

and maybe feeling a little uncomfortable.

這很諷刺，因為幾分鐘前 我請你吃三明治，你說好。

Which is ironic because a few minutes ago, I was going to get you all sandwiches

and you said yes.

這就是為什麼你不應該 接受陌生人的食物。

This is why you should not accept food from strangers.

但如果我不抱有希望， 我也不會站在這裡。

But I wouldn't be up here if I weren't also optimistic.

那是因為我認為

And that's because I think we have momentum

我們在法規和文化變革方面有動力。

behind the regulation and the culture changes.

尤其是如果我們遵循三個基本原則

Especially if we align ourselves with three basic principles

來與數據公司互動。

about how corporations should engage with data.

第一個原則是收集數據的公司 應該告訴我們，他們在收集什麼。

The first principle is that companies that gather data should tell us

what they're gathering.

這樣我們才能提出問題， 例如，這是版權材料嗎？

This would allow us to ask questions like, is it copyrighted material?

這些資訊是私人的嗎？

Is that information private?

Could you please stop?

你能停止嗎？

這也開啟了數據的科學研究。

It also opens up the data to scientific inquiry.

第二個原則是收集數據的公司 應該在做任何事情之前，

The second principle is that companies that are gathering our data should tell us

what they're going to do with it before they do anything with it.

告訴我們他們打算做什麼。

要求公司告訴我們他們的計畫，

And by requiring that companies tell us their plan,

這意味著他們必須有計畫，

this means that they have to have a plan,

這將是很好的一步。

which would be a great first step.

這也可能讓數據收集最小化，

It also probably would lead to the minimization of data capture,

因為如果他們不知道要做什麼，

because they wouldn't be able to capture data

他們就無法收集數據。

if they didn't know what they were already going to do with it.

最後，第三個原則是

And finally, principle three,

companies that build AI should tell us about the data

建立 AI 的公司應該告訴我們

他們用來訓練 AI 的數據。

that they use to train the AI.

這就是數據集營養標示

And this is where dataset nutrition labels

和其他透明度標示的功能。

and other transparency labeling comes into play.

You know, in the case where the data itself won't be made available,

你知道，在大多數情況下， 數據本身不會被公開，

which is most of the time, probably,

所以標示對我們來說十分重要， 讓我們能夠檢視成分，

the labeling is critical for us to be able to investigate the ingredients

並開始尋找解決方案。

and start to find solutions.

最後，我想告訴你一個好消息，

So I want to leave you with the good news,

那就是數據營養計畫和其他項目

and that is that the data nutrition projects and other projects

只是全球 AI 問責運動的一小部分。

are just a small part of a global movement

towards AI accountability.

數據集營養標示 和其他項目只是第一步。

Dataset Nutrition Label and other projects are just a first step.

法規即將到來，

Regulation's on the horizon,

文化規範正在改變，

the cultural norms are shifting,

尤其是如果我們遵循這三個基本原則，

especially if we align with these three basic principles

即公司應該告訴我們他們在收集什麼，

that companies should tell us what they're gathering,

在做任何事情之前告訴我們 他們打算做什麼，

tell us what they're going to do with it before they do anything with it,

並解釋他們用來 建立 AI 系統的數據。

and that companies that are building AI

should explain the data that they're using to build the system.

我們必須要求這些建立 AI 的公司 為他們所構建的 AI 負責，

We need to hold these organizations accountable

for the AI that they're building

追問他們，就像我們 會問食品公司的一樣，

by asking them, just like we do with the food industry,

內容是什麼，怎麼製作的？

what's inside and how did you make it?

只有這樣，我們才能在問題發生前

Only then can we mitigate the issues before they occur,

而不是發生後，減輕問題的危害。

as opposed to after they occur.

這樣才能創造一個整合算法 的網際網路，對所有人都更有助益。

And in doing so, create an integrated algorithmic internet

that is healthier for everyone.

謝謝大家。

Thank you.

（掌聲）

(Applause)