Which of these three faces do you think is the oldest?

這三張人臉中，你覺得哪一位看起來年紀最大？

A, B or C?

A、B 還是 C？

The answer is actually none of them, because none of these people were ever born.

答案是沒有任何一個，因為圖中的人從未出生於這世上。

These faces were all generated by artificial intelligence.

這些臉都是透過人工智慧技術產生的。

Now, thinking about artificial intelligence, look carefully at these two paintings.

想到人工智慧的話，來仔細看一下這兩幅畫。

Which of these does your mind think is created by AI?

你們認為哪一幅是由 AI 創作出來的？

Is it painting A or painting B?

是畫像 A 還是 B？

Give up?

要放棄了嗎？

The clue is in the signature.

答案就藏在簽名裡。

The AI seems a bit busted on this one.

AI 的這幅簽名看起來有些許破綻。

But someone loved it as it went at an auction for $432,500.

但還是有人喜歡這幅畫，因為這幅畫以 432,500 美元成功拍賣。

To explain how these images are generated, let's first look at these images of fruit.

為了要解釋這些圖片是如何產生的，先給你們看看以下水果的圖片。

By now, you know that one isn't real, but is it the apples or the oranges?

看到現在，你們應該知道有一張不是真的，那到底是蘋果還是橘子？

If you said apples, you're right!

如果你的回答是蘋果，那你就答對了！

Essentially, a system is fed hundreds of images to learn from.

基本上，系統已經放了上百張影像去讓它進行學習。

It's then split into two systems.

然後再把它分成兩個系統。

System 1's job is to create a convincing image of an apple,

系統一的工作是要創造一張具說服力的蘋果圖像，而系統二是要分辨哪一張蘋果的圖像是假的。

and system 2 is designed to figure out which images of apples are fake.

This is known as a generative adversarial network.

我們稱之為生成對抗網絡。

If system 1 fails at tricking, it continues to evolve its apple designing skills,

如果系統一欺騙失敗的話，將繼續發展其蘋果設計技能；而如果系統二無法辨識是偽造的圖像的話，則需提高其捕捉偽造品的能力。

and if system 2 fails at identifying the fake, it must also improve its ability to catch a fake.

This ultimately creates a feedback loop that generates extremely believable images.

系統最終會不斷循環地產出回饋，從而造出令人以假亂真的圖像。

Now, let's try to relax and listen to some music while you try to deduce which is composed by a computer.

現在來放鬆聽點音樂，來試試看能不能推測哪一首歌曲是由電腦創作的。

Turns out that composition A is generated by a machine learning system called Emily Howell,

結果是作品 A 是由名為 Emily Howell 的機器學習系統產出，這系統於 90 年代設計，並從音樂庫中學習去創作原創曲。

which was created in the 90s and learned from a library of music to create original pieces.

And composition B?

那作品 B 呢？

That's by Johann Sebastian Bach.

是出自巴赫。

A larger library of music will give the machine learning networks more data to learn from, and the same is true for video.

音樂庫愈大，機器的學習網路就可以獲得愈多數據去學習，而製作影片也是一模一樣。

Because we upload such huge amounts of video content online,

由於我們上載大量影片到網路上，神經網路開始創造出令人信服的影片。

neural networks have begun to create convincing fake videos.

Here's a video of the 44th President of the United States, Barack Obama.

以下的影片是美國第 44 屆總統奧巴馬。

People not trapped by our past, but able to remake ourselves as we choose.

人們不應受過去所困，但應依我們的選擇改造自己。

But today, our immigration system is broken, and everybody knows it.

但時至今天，我們的移民系統有漏洞，而這大家也都知道。

And here's another.

這是另外一支影片。

It's been less than a week since the deadliest mass shooting in American history.

美國史上最多人致命的大型槍殺案已經發生了不到一周。

And foremost in all of our minds has been the loss and the grief felt by the people of Orlando.

在所有人心中所想的是奧蘭多人民的損失和悲痛。

Which of these two are real?

到底哪一支影片才是真的？

Video A or B?

影片 A 還是 B？

Known as a deep fake video, video B is generated using AI.

影片 B 是利用 AI 的技術製做的，稱之為深度偽造 (Deepfake) 的影片。

With over 14 hours of footage from the president's weekly addresses, neural network AI is able to model the shape of Obama's mouth,

總統每週演講超過 14 小時，而神經網路 AI 有了這些影像便能夠對歐巴馬的嘴巴進行建模，再使用歐巴馬以前演說的音頻。

and uses audio from one of Obama's previous addresses.

But can AI generate convincing synthetic audio?

不過，AI 可以製做出令人信以為真的合成音頻嗎？

Listen to this phone call.

來聽聽看以下的電話內容。

How can I help you?

有什麼可以協助您的嗎？

Hello?

喂？

Hello, what's up, man?

喂，先生，有什麼事嗎？

Hey, um, I wanted to know what are your hours for today?

我想知道你們今天的工作時間？

10 a.m. to 6:30 p.m.

早上 10 點至下午 6 點半。

Okay.

好的。

Got it.

了解。

Thank you for your time.

謝謝您寶貴的時間。

No problem sir.

不會的。

Bye.

掰掰。

Can you tell which of the callers is human?

你們可以告訴我哪一個講電話的人是人類？

Voice B, the voice you hear asking about the hours the business was open, was fake.

聲音 B，詢問工作時間的那把聲音是偽造的。

We're relying more and more on technology today, which is making our lives easier.

我們現在變得愈來愈依賴科技，因為科技讓我們的生活過得更便利。

Much like SimpliSafe, the sleek new home security system who sponsored today's video.

就像時尚嶄新的家庭安全系統的 SimpliSafe，他們贊助了今天這支影片。

We bought a house recently and finally realized what it feels like to go anywhere and be like, okay legit we're probably gonna be robbed right now, and the stats are in.

我們最近買了房子，終於體會到那種無論去到哪，都有一種我們現在都有可能會被搶劫，現在來講一下數據。

America has a robbery every 15 seconds, most of which happen while you're at work.

在美國，每 15 秒就有一宗盜竊案，而大多數都是在你上班的時侯發生。

So join our new zen life by going to simplisafe dot com slash asap and ease your mind.

一起加入輕鬆人生，就要到 simplisafe.com/asap。

Importantly for us in our new pad is that it looks cute.

重要的是放在我們的新家看起來就很搭！

The design is awesome, it's intuitive to set up and use, you barely notice the sensors are there,

這個設計真的是很讚，整個在設置和使用上都非常的簡單，難以發現感測器的存在。而且我們還很喜歡有一點很酷的是，出門沒關門會提醒你，因為像 Gregg 就常常發生這種事。

and we like that it does cool things like remind you if you left the door open because Gregg literally does that all the time.

He leaves the door wide open when he leaves.

他常常出門的時候，大門永遠都會是敞開。

In the event of an actual break-in, it can immediately contact the police.

萬一真的有人闖入，會馬上聯絡警方。

We literally just set up our own at home, and it's just one of those things where you get like peace of mind knowing that it's there for you.

我們只是剛弄好我們的房子，其中一個可以讓你安心的地方。

And it can even tell the difference between a pet and an intruder!

這東西還可以分辨家裡寵物和入侵者！

So go to simplisafe dot com slash asap to get yours today and make sure you subscribe for more videos every Thursday.

今天就到 simplisafe.com/asap 去購買，還有確認你有訂閱頻道才可以收看每星期四上傳的影片。