I've collected fingerprints from a few of our colleagues, and what you see here is four fingerprints I took, two of which are from the exact same finger.

我收集了一些同事的指紋，你們看到的是我採集的四枚指紋，其中兩枚來自完全相同的手指。

So I want to see if you can match which two prints are from the same finger.

所以，我想看看你們能否比對出哪兩個指紋來自同一個手指。

This is Explainer Club, where video producers like me sit down and explain something to one of our coworkers.

這裡是 "解說員俱樂部"，像我這樣的視頻製作人可以坐下來，向我們的同事解說一些東西。

And today, that is Kim.

今天，她就是 Kim。

I'm a video producer, and I work on primarily health and science content.

我是一名視頻製作人，主要負責健康和科學內容。

And I do history, culture, and sometimes science videos.

我還製作歷史、文化視頻，有時也製作科學視頻。

So, I think A and B.

所以，我認為是 A 和 B。

That's my first guess, and that's just because I see this like little swirl coming up.

這是我的第一個猜測，因為我看到這個像小漩渦一樣的東西出現了。

And these two just look so different, because you have like this little hump down here.

這兩個看起來很不一樣，因為你這裡有一個小駝峰。

I think I still want to say A and B are the same.

我想我還是想說 A 和 B 是一樣的。

Okay.

好的

I feel like I completely blew that.

我覺得我完全搞砸了。

No, no, no.

不，不，不

I feel like the okay was a little like...

我覺得好吧有點像......

The answer was A and C.

答案是 A 和 C。

Ouch.

哎喲

They're both from the left thumb.

它們都來自左手大拇指。

So, what is interesting to me about this, though, is all of these fingerprints are from the same person.

不過，讓我感到有趣的是，所有這些指紋都來自同一個人。

Oh.

哦。

I think we all know every person has their own fingerprint.

我想我們都知道，每個人都有自己的指紋。

Right.

對

But I didn't know before I started working on this that each individual finger on your hands is different from each other.

但在開始研究之前，我並不知道你手上的每根手指都是不同的。

Did you know that?

你知道嗎？

Like, they're completely different?

比如，它們完全不同？

Police practice, when you get fingerprinted, they'll take all 10.

警察的做法是，當你按指紋時，他們會把10個指紋都拿走。

That makes sense, yeah.

有道理

But that makes it seem like it's so much harder to actually identify if a fingerprint belongs to a person if every single one is different.

但這樣一來，如果每個人的指紋都不一樣，要真正識別指紋是否屬於某個人就顯得難上加難了。

Right.

對

Without this full reference, you wouldn't be able to match one fingerprint to another and say this came from the same hand, or even the same person.

如果沒有完整的參考資料，您就無法將一枚指紋與另一枚指紋進行比對，也就無法說這枚指紋來自同一隻手，甚至是同一個人。

But there's this new AI tool that says it can do it.

但有一種新的人工智能工具說它能做到這一點。

That's what we're going to talk about today.

這就是我們今天要討論的話題。

Okay.

好的

Cool.

酷斃了

Yes.

是的。

Cool.

酷斃了

Then we're going to probably run the logo there.

然後，我們可能會在那裡使用徽標。

Logo.

徽標。

Okay.

好的

I'll show you something else.

我再給你看點別的。

This is a copy of the actual fingerprinting card that the FBI uses.

這是聯邦調查局使用的真實指紋卡的副本。

This one is kind of a practice for me.

這對我來說是一種練習。

I learned while making this video that taking someone's fingerprint, if you're not trained to do that, is actually kind of hard.

在製作這段視頻的過程中，我瞭解到，如果你沒有接受過相關培訓，採集指紋其實是一件很難的事情。

Should it work?

它應該有用嗎？

No, I'm going to f*** it up.

不，我會搞砸的。

The FBI's fingerprint archive currently houses around 165 million fingerprint records, and they start on these cards.

聯邦調查局的指紋檔案中目前保存著約 1.65 億條指紋記錄，這些記錄就是從這些指紋卡上開始的。

So, if you're arrested for a crime in the U.S., you will get your fingerprints taken by the police, and they're stored on file even if you're not proven guilty.

是以，如果您在美國因犯罪被捕，警方會採集您的指紋，即使您沒有被證明有罪，指紋也會被存檔。

Okay.

好的

So, if you flip this over, these are the three main patterns used to classify fingerprints.

是以，如果把這個翻過來，這就是用來對指紋進行分類的三種主要模式。

The arch, whorl, and loop.

拱形、渦形和環形。

Okay.

好的

I feel that you should have given me this before the quiz, Coleman.

我覺得你應該在測驗前給我這個，科爾曼。

I feel I wasn't prepared.

我覺得自己沒有做好準備。

Okay, that is fair.

好吧，這很公平。

But, no, this is cool.

但是，不，這很酷。

So, the ridges in an arch pattern come from one side of the fingertip and then exit out the other side.

是以，拱形圖案的脊來自指尖的一側，然後從另一側流出。

Then a whorl spirals into a center point, and a loop comes from one side of the finger and then back out the same side.

然後，一個旋渦螺旋狀地進入一箇中心點，一個環從手指的一側繞出，然後又從同一側繞回。

This little bottom part's like an arch, right?

底部這個小部分就像一個拱門，對嗎？

And this is a loop because this is coming back this way?

這是一個循環，因為它會從這個方向返回？

Yeah, exactly.

是的，沒錯。

Not a lot of whorls going on.

沒有太多的輪廓。

Is whorl more rare?

輪紋是否更罕見？

It's kind of in the middle.

有點像中間。

About 60% of people have loops.

約有 60% 的人有循環系統。

That's the most common.

這是最常見的。

Whorl is considered like around 35%.

Whorl 大約佔 35%。

And then arches are actually very rare.

而拱門實際上是非常罕見的。

Oh, interesting.

哦，有意思。

These general categories are really just for classification.

這些一般類別實際上只是為了分類。

Actually analyzing and matching an individual fingerprint is in the minutia.

實際上，分析和比對單個指紋屬於細枝末節。

The little breaks and splits along each individual ridge, not the overall pattern.

每個山脊上的小斷裂和裂縫，而不是整體圖案。

And the most common markings are when you follow a ridge and it ends, that's called a ridge ending.

最常見的標記是，當你沿著山脊走，山脊的盡頭就叫做山脊盡頭。

And then if you follow along a ridge and it splits into two, that's called a bifurcation.

然後，如果你沿著山脊走，它一分為二，這就是所謂的分叉。

If there's an island, which is like that little dot, when a ridge splits but then quickly rejoins, it's called a lake.

如果有一個島，就像那個小點，當山脊裂開但又很快重合時，它就被稱為湖。

And then that's a bridge connecting two ridges.

然後是連接兩座山脊的橋樑。

So analysts will basically just go ridge by ridge and mark these details, and that is where uniqueness in fingerprints really comes from.

是以，分析人員基本上會一個脊線一個脊線地標記這些細節，這就是指紋獨特性的真正來源。

Not even identical twins who have matching DNA have matching fingerprints when you get to the minutia.

即使是 DNA 相同的同卵雙胞胎，其指紋也不可能完全相同。

So the minutia is how analysts have been matching fingerprints for more than a century.

是以，一個多世紀以來，分析人員就是通過這些細枝末節來比對指紋的。

Is that coming from outside factors?

這是外部因素造成的嗎？

Like it's not genetic anymore?

就像它不再是遺傳的一樣？

Like if I burn my finger or something like that, is that where most of the minutia is from?

比如我燒傷了手指之類的，是不是大部分細枝末節都來自於此？

That's a great question.

這個問題問得好。

It doesn't come from your lifetime.

它不是來自你的一生。

I'm just thinking of like Men in Black where Will Smith had to like grab the thing and it just burns his fingerprints off.

我想到的是《黑衣人》裡威爾-史密斯要抓住那東西，結果指紋被燒掉了。

One will encounter.

人們會遇到

That would be hard to actually do.

這很難真正做到。

It's very hard to get rid of your fingerprints and they don't change over your lifetime.

指紋很難去除，而且終生不會改變。

There are actually tiny ridges of tissue under your skin. They're called friction ridges.

實際上，在你的皮膚下有一些細小的組織脊。 它們被稱為摩擦脊。

Even if you've burned off your fingertips, your palms, your other joints, they're all equally unique.

即使你燒掉了指尖、手掌和其他關節，它們也同樣獨一無二。

They're the same friction ridges.

它們是相同的摩擦脊。

So I want to show you something on my computer.

所以，我想給你看看我電腦上的東西。

By the time you're born, your fingerprints are already set.

當你出生時，你的指紋就已經定型了。

So they form in the womb.

所以它們是在子宮裡形成的。

Their pattern begins with some genetic influence from your parents as a shape in the center of the finger tip and top of your first knuckle.

它們的形狀始於父母的一些遺傳影響，如指尖中心和第一指節頂部的形狀。

The ridges grow from there in waves following what's called a Turing pattern.

從這裡開始，脊線按照所謂的圖靈模式一波一波地生長。

This is a concept proposed by Alan Turing in the 1950s that helps explain other unique patterns in nature like how a leopard gets its spots, how a zebra gets its stripes, and sand dunes.

這是艾倫-圖靈在 20 世紀 50 年代提出的一個概念，有助於解釋自然界中的其他獨特模式，如豹子如何獲得斑點、斑馬如何獲得條紋以及沙丘。

That's pretty.

真漂亮

It's beautiful.

太美了

And so what I want to show you is there are two cool simulators online that can kind of give you an idea of how a Turing pattern, how they work.

是以，我想告訴大家的是，網上有兩個很酷的模擬器，可以讓你瞭解圖靈模式是如何工作的。

Oh, that's so fun.

哦，太有趣了

Oh, it looks like a brain.

哦，看起來像大腦。

That's creepy.

真讓人毛骨悚然。

These waves and the way they merge and collide ends up with a unique pattern.

這些波浪及其融合和碰撞的方式最終形成了獨特的圖案。

So I want to ask you about the video that you made about how our voice is like a fingerprint too.

所以，我想問你關於你製作的視頻，我們的聲音就像指紋一樣。

As humans, each of us produces a sound that's about as unique as a fingerprint.

作為人類，我們每個人發出的聲音都像指紋一樣獨一無二。

So the whole point of that video was to just explain the amount of things that go into producing your voice.

所以，這段視頻的目的只是為了解釋製作嗓音的大量工作。

Your nasal cavity, your tongue, your vocal cords have different thicknesses and lengths.

鼻腔、舌頭和聲帶的厚度和長度各不相同。

There's just so many minuscule little factors that will just change how you sound.

有太多微不足道的小因素會改變你的聲音。

And therefore we went with why your voice is like a fingerprint.

是以，我們選擇了 "為什麼說聲音就像指紋 "這一主題。

People in the comments were just like, actually, fingerprints are not unique.

人們在評論中說，實際上，指紋並不是獨一無二的。

The thing is, is like we can't prove it.

問題是，我們無法證明這一點。

I think maybe that's what they mean.

我想他們也許就是這個意思。

I can see the doubt.

我明白你的疑慮。

And I've seen questions online like if they didn't take everybody's fingerprint, how do they know they're all unique?

我在網上看到一些問題，比如如果他們沒有采集每個人的指紋，他們怎麼知道這些指紋都是獨一無二的？

It's a probability issue, especially the way these things form under our skin, which is truly random.

這是一個概率問題，尤其是這些東西在我們皮膚下形成的方式，確實是隨機的。

And there's just so many data points in here that it was estimated even sort of towards the beginning in the history of fingerprinting, the odds of having the exact same pattern when you get down to the minutia, like one in 64 billion.

這裡有如此多的數據點，據估計，即使是在指紋識別歷史的初期，當你深入到細枝末節時，擁有完全相同模式的機率也只有六百四十億分之一。

And I think that people feel comfortable enough saying that no two fingerprints are alike.

我認為，說沒有兩枚指紋是相同的，人們會覺得很舒服。

But about a year ago, that might have changed.

但大約一年前，情況可能發生了變化。

I'm going to play a clip from a phone interview that I did recently.

我要播放一段我最近接受電話採訪的片段。

This is a transcript of what you're about to hear.

這是您即將聽到的內容的文字記錄。

Take a listen to this.

聽聽這個。

My name is Hal Lipset.

我叫哈爾-利普塞特。

I'm a roboticist, AI researcher here at Columbia, and I have nothing to do with forensics, fingerprint, anything like that.

我是哥倫比亞大學的機器人、人工智能研究員，與法醫、指紋之類的工作無關。

Once you understand AI, you can apply it to taking a first look at some heart problems.

瞭解人工智能後，您就可以運用它來初步瞭解一些心臟問題。

And this is one of them.

這就是其中之一。

What he's talking about is this study published in Science Advances that one of the students was the lead author on that claims to have trained an AI to identify whether a pair of fingerprints came from the same person or not.

他說的是發表在《科學進展》（Science Advances）上的一項研究，其中一名學生是這項研究的第一作者，他聲稱已經訓練出一種人工智能，可以識別一對指紋是否來自同一個人。

So what they did was feed pairs of fingerprints, sometimes from different people, sometimes the same person, but different fingers, into the AI and ask a simple question.

於是，他們把成對的指紋（有時來自不同的人，有時是同一個人，但手指不同）輸入人工智能，並提出一個簡單的問題。

Same person or two different people?

是同一個人還是兩個不同的人？

And so he found a data set, which was a set of about 60,000 fingerprints from, I think, mostly from these people as the public data set.

於是，他找到了一套數據，這套數據包含了大約 6 萬個指紋，我想，其中大部分來自這些人的指紋，作為公共數據集。

He set up the AI, again, on a fancy AI, and had it look at this.

他又在一個高級人工智能上設置了人工智能，讓它看這個。

And lo and behold, it was able to do it fairly effectively, which led us to ask more questions like, how are you doing it?

看吧，它能夠相當有效地做到這一點，這讓我們提出了更多問題，比如，你們是如何做到這一點的？

What are the features you're using?

您正在使用哪些功能？

And the AI shows it.

人工智能也顯示了這一點。

It's looking at fingerprints in a completely different way than humans.

它觀察指紋的方式與人類完全不同。

It found similarities to pretty successfully identify whether it was the same person or not.

它發現了相似之處，從而非常成功地識別出是否是同一個人。

How does it do that?

它是怎麼做到的？

I want to show you what it was looking at.

我想讓你看看它在看什麼。

Instead of examining the minutiae like a human analyst does, the machine focused on these subtle curved shapes and different angles of ridge lines in the center of the fingerprints.

機器並不像人類分析師那樣檢查細微之處，而是專注於指紋中心這些微妙的彎曲形狀和不同角度的脊線。

So what this chart is basically showing is when it examined a raw image, the machine was right about 80% of the time when asked the questions, same person or different people.

是以，這張圖基本上顯示的是，當機器檢查原始影像時，無論問的是同一個人還是不同的人，機器大約有 80% 的時間是正確的。

And even when you removed a lot of the detail, including the minutiae, and you were left with just this orientation, it was still right about 75% of the time.

即使你去掉了很多細節，包括細枝末節，只剩下這個方向，它仍然有大約 75% 的時間是正確的。

So it doesn't need the minutiae.

是以，它不需要這些細枝末節。

And in fact, when they did just the minutiae and removed all other details, it was the least effective.

而事實上，當他們只做細枝末節的工作，去掉所有其他細節時，效果是最差的。

It was around 60%, which is not much better than a random guess.

大約是 60%，比隨機猜測好不了多少。

That's just to examine if the prints came from the same person or not.

這只是為了檢查指紋是否來自同一個人。

To match an exact fingerprint to an exact fingerprint, you would still need the minutiae.

要將準確的指紋與準確的指紋進行比對，仍然需要細部特徵。

There's been some pushback from experts in the field of forensics about the study who say the paper alone won't change the way we interpret fingerprints.

法醫領域的專家對這項研究提出了一些質疑，他們認為僅憑這篇論文並不能改變我們解讀指紋的方式。

And the authors themselves acknowledge that there's more research that needs to be done.

作者自己也承認，還需要做更多的研究。

We took it as far as we can using the public data sets, but I'm sure if the FBI wanted to do this, they could do a much better job.

我們儘可能利用公共數據集來完成這項工作，但我相信，如果聯邦調查局想做這項工作，他們可以做得更好。

But a slightly deeper issue is the discovery is that we've been looking at fingerprints.

但更深層次的問題是，我們發現我們一直在研究指紋。

Perhaps we're missing a lot of information because we were overlooking things like curvature and things that AI suggests are actually very meaningful.

也許我們錯過了很多資訊，因為我們忽略了像曲率這樣的東西，以及人工智能認為實際上非常有意義的東西。

It gives us a new language to look at fingerprints.

它為我們研究指紋提供了一種新的語言。

A general optimistic view of AI that I hope to have is how it can teach us to see new things in the field of medical diagnostics, to be able to catch things like tumors or cancers, and how much we can use machine learning to improve the lives of people.

我對人工智能的總體樂觀看法是，它能教會我們如何看到醫療診斷領域的新事物，如何捕捉腫瘤或癌症等病症，以及我們能在多大程度上利用機器學習改善人們的生活。

I asked you to do a little bit of homework.

我讓你做一點功課。

I was wondering if you had any thoughts just on that general big picture idea.

我想知道你是否對這個大局觀有什麼想法。

Yeah.

是啊

The interesting thing about AI is that typically when we're reading about it or talking about it or everything, there's this huge future of work thing that comes up where they're just like, the robots are going to replace us.

關於人工智能的有趣之處在於，通常我們在閱讀或談論它時，都會出現這樣一個巨大的未來工作問題，即機器人將取代我們。

And I kind of like this other lens that came up when I started looking into you.

當我開始研究你時，我有點喜歡另一個鏡頭。

It asked me to look into the medical field specifically.

它要求我專門研究醫學領域。

One quote that I saw more than once was that AI is not going to replace the doctor.

我不止一次看到這樣一句話：人工智能不會取代醫生。

It's going to replace the doctor that doesn't use AI.

它將取代不使用人工智能的醫生。

The idea there being that doctors will just evolve with this technology and they will start using it to help advance themselves diagnosing things like cancers.

我們的想法是，醫生會隨著這項技術的發展而發展，他們會開始使用這項技術來幫助自己提高診斷癌症等疾病的水準。

How much more quickly can we work in a crisis if we have a machine next to us eliminating the problems?

如果我們身邊有一臺機器可以消除問題，我們在危機中的工作效率會提高多少？

These tools raise so many questions.

這些工具提出了許多問題。

I think we're trained to kind of take a really cynical approach with most of them.

我認為，我們被訓練得對大多數人採取一種非常憤世嫉俗的態度。

Yeah.

是啊

But the idea that it's just we can learn from the machine too, I like that back and I like that version of things.

但是，我們也可以向機器學習，我喜歡這樣的想法，也喜歡這樣的版本。

I like that too.

我也喜歡。

That makes me feel better about the machines.

這讓我對機器的感覺更好了。

Thanks for watching.

感謝觀看。

We wanted to do something different with this video, to take our standard explainer video and see what happens when we lay out the explanation in a room with another person.

我們想在這段視頻中做一些與眾不同的事情，把我們的標準解說視頻拿出來，看看在房間裡和另一個人一起進行解說時會發生什麼。

For this video, that meant doing a lot of my After Effects and design work, which is usually the last thing I do, early on while I researched and wrote the video.

對於這個視頻來說，這意味著我需要在研究和撰寫視頻的同時，儘早完成大量的後期特效和設計工作，而這通常是我最後要做的事情。

And then instead of doing multiple takes in a video booth, shooting the whole thing in one go and hoping it goes well.

這樣就不用在錄像室裡多次拍攝，而是一次性拍完，希望一切順利。

We've been so focused on the minutiae.

我們一直專注於細枝末節。

I think it's minutiae, but Kim, I should have shown you the basic fingerprint patterns before the quiz.

我認為這是細枝末節，但金，我應該在測驗前給你看基本的指紋模式。

I'm sorry about that.

對此我很抱歉。

But the point of this video was the same as always, to answer questions about our world and to show you the most interesting things we find in a way that's hopefully fun to If you want to support that, you can do that at Vox.com slash memberships.

但這個視頻的目的還是一如既往，回答關於我們世界的問題，並以一種希望有趣的方式向大家展示我們發現的最有趣的事物。如果您想支持我們的工作，可以在 Vox.com 網站上申請會員。

If you're already a member, thank you so much.

如果您已經是會員，非常感謝。

You're a big part of the reason that we get to do this at all.

我們之所以能這樣做，你們功不可沒。

See you next time.

下次再見