Chris Anderson: Help us understand what machine learning is, because that seems to be the key driver of so much of the excitement and also of the concern around artificial intelligence. How does machine learning work?
克里斯安德森:幫我們 了解一下機器學習是什麼, 因為機器學習似乎是 推動人工智慧 一些令人興奮及重要議題的關鍵動因, 機器學習如何運作?
Sebastian Thrun: So, artificial intelligence and machine learning is about 60 years old and has not had a great day in its past until recently. And the reason is that today, we have reached a scale of computing and datasets that was necessary to make machines smart. So here's how it works. If you program a computer today, say, your phone, then you hire software engineers that write a very, very long kitchen recipe, like, "If the water is too hot, turn down the temperature. If it's too cold, turn up the temperature." The recipes are not just 10 lines long. They are millions of lines long. A modern cell phone has 12 million lines of code. A browser has five million lines of code. And each bug in this recipe can cause your computer to crash. That's why a software engineer makes so much money. The new thing now is that computers can find their own rules. So instead of an expert deciphering, step by step, a rule for every contingency, what you do now is you give the computer examples and have it infer its own rules.
賽巴斯汀索朗:人工智慧和機器學習 大約有六十年歷史, 一直到近期才有輝煌的日子可言。 原因是因為現今 我們的計算能力和 資料集規模已經達到 讓機器變聰明所必要的條件。 它的運作方式是這樣的。 如果現在你要為一台電腦 寫程式,比如你的手機, 你會僱用軟體工程師, 他們會寫一份 非常非常長的廚房食譜, 比如「如果水太熱,就把溫度調低。 如果水太冷,把溫度調高。」 食譜長度並不是只有十行。 它們長達數百萬行。 一台現代手機有 1200 萬行的程式碼。 一個瀏覽器有五百萬行的程式碼。 食譜中的每一個錯誤, 都會造成你的電腦當機。 那就是為什麼軟體工程師 能賺那麼多錢。 現在的新發展是,電腦能 找到它們自己的規則。 所以不再需要找一個專家, 來針對每個情況的規則 一步一步地做理解辨識, 現在你的做法是,給電腦一些範例, 讓它推導出它自己的規則。
A really good example is AlphaGo, which recently was won by Google. Normally, in game playing, you would really write down all the rules, but in AlphaGo's case, the system looked over a million games and was able to infer its own rules and then beat the world's residing Go champion. That is exciting, because it relieves the software engineer of the need of being super smart, and pushes the burden towards the data. As I said, the inflection point where this has become really possible -- very embarrassing, my thesis was about machine learning. It was completely insignificant, don't read it, because it was 20 years ago and back then, the computers were as big as a cockroach brain. Now they are powerful enough to really emulate kind of specialized human thinking. And then the computers take advantage of the fact that they can look at much more data than people can. So I'd say AlphaGo looked at more than a million games. No human expert can ever study a million games. Google has looked at over a hundred billion web pages. No person can ever study a hundred billion web pages. So as a result, the computer can find rules that even people can't find.
最近 Google 的阿爾法圍棋贏得比賽, 就是一個很好的例子。 通常,在玩遊戲時, 你得要寫下所有的規則, 但在阿爾法圍棋的這個例子, 系統是去看了一百萬場比賽, 能推導出它自己的規則, 然後打敗世界現在的棋王。 這讓人很興奮, 因為軟體工程師能鬆口氣了, 他們不需要超聰明, 這個重任已轉到資料上。 如我所言,這件事的反轉點在於── 很慚愧,我的論文主題是機器學習, 它完全不重要,請別去讀它, 因為那是二十年前寫的, 那時,電腦和蟑螂大腦一樣大。 現在,電腦強大到能夠真正地模擬 人類的特定思想。 接著,電腦也因為 可以比人類看更多的資料 進而取得優勢, 阿爾法圍棋已經研究過 一百多萬場比賽。 沒有任何人類專家能夠 研究到一百萬場比賽。 Google 已看過了一千億個網頁。 從來沒有人有能力研究 一千億個網頁。 因此,電腦能找出一些 人類找不出來的規則。
CA: So instead of looking ahead to, "If he does that, I will do that," it's more saying, "Here is what looks like a winning pattern, here is what looks like a winning pattern."
克:換句話說,不太像是: 「如果他那樣下,我就這樣下。」 比較像是在說: 「下這裡像是獲勝的模式, 下那裡像是獲勝的模式。」
ST: Yeah. I mean, think about how you raise children. You don't spend the first 18 years giving kids a rule for every contingency and set them free and they have this big program. They stumble, fall, get up, they get slapped or spanked, and they have a positive experience, a good grade in school, and they figure it out on their own. That's happening with computers now, which makes computer programming so much easier all of a sudden. Now we don't have to think anymore. We just give them lots of data.
賽:是的,想想看 你如何養育你的孩子。 你並不會花前十八年的時間, 對每種狀況給孩子一條規則, 然後放他們自由, 他們就會做出這個大程式。 他們會摔跤,會爬起來, 他們會被賞巴掌或打屁股, 他們會有正向的經驗, 在學校有好成績, 他們會靠自己去了解這些。 現在電腦也是這樣, 突然間讓電腦寫程式就變簡單了。 我們不用再花腦筋思考了。 只要給它們大量資料即可。
CA: And so, this has been key to the spectacular improvement in power of self-driving cars. I think you gave me an example. Can you explain what's happening here?
克:所以這是自動駕駛車的能力 能夠有重大改善的關鍵。 我想你給了我一個例子。 你能否解釋一下這裡發生了什麼事?
ST: This is a drive of a self-driving car that we happened to have at Udacity and recently made into a spin-off called Voyage. We have used this thing called deep learning to train a car to drive itself, and this is driving from Mountain View, California, to San Francisco on El Camino Real on a rainy day, with bicyclists and pedestrians and 133 traffic lights. And the novel thing here is, many, many moons ago, I started the Google self-driving car team. And back in the day, I hired the world's best software engineers to find the world's best rules. This is just trained. We drive this road 20 times, we put all this data into the computer brain, and after a few hours of processing, it comes up with behavior that often surpasses human agility. So it's become really easy to program it. This is 100 percent autonomous, about 33 miles, an hour and a half.
賽:這是自動駕駛車的行車, 我們優達學城(Udacity)碰巧有, 最近變成稱為 Voyage 的副產品。 我們用所謂的「深度學習」 來訓練汽車自動駕駛, 這趟行程從加州的山景城出發 前往舊金山, 在雨天行駛 El Camino Real 路名, 路上有腳踏車騎士及行人, 途中經過 133 個交通燈號。 新奇的是, 許多個月前,我成立了 Google 自動駕駛汽車團隊, 那時,我僱用了世界上 最好的軟體工程師, 來找出世界上最好的規則。 這只是訓練出來的。 這條路我們開了二十次, 我們把所有資料放到電腦的大腦中, 經過幾小時的處理之後, 它所找出的行為, 通常都能勝過人類的機敏。 所以變得很容易為它寫程式。 這是 100% 自主的, 大約 33 英哩,一小時半。
CA: So, explain it -- on the big part of this program on the left, you're seeing basically what the computer sees as trucks and cars and those dots overtaking it and so forth.
克:解釋一下這程式左半邊的大部分, 我們可以看到電腦 所看到的卡車與汽車, 還有那些超過它的點。
ST: On the right side, you see the camera image, which is the main input here, and it's used to find lanes, other cars, traffic lights. The vehicle has a radar to do distance estimation. This is very commonly used in these kind of systems. On the left side you see a laser diagram, where you see obstacles like trees and so on depicted by the laser. But almost all the interesting work is centering on the camera image now. We're really shifting over from precision sensors like radars and lasers into very cheap, commoditized sensors. A camera costs less than eight dollars.
賽:右側的是攝影機的影像, 也就是主要的輸入, 用來找車道、其它車輛、交通號誌。 這車用個雷達來估算距離。 這是這類系統常用的方式。 左邊的是雷射圖, 可以看到雷射槍描繪出來的障礙, 如樹木等等。 但幾乎所有有趣的部份 都以攝影機影像為中心。 我們其實在從精準的感測器, 像是雷達和雷射, 轉換到極便宜的一般感測器。 一台攝影機的成本不到 $8。
CA: And that green dot on the left thing, what is that? Is that anything meaningful?
克:左邊的綠點是什麼? 是有意義的嗎?
ST: This is a look-ahead point for your adaptive cruise control, so it helps us understand how to regulate velocity based on how far the cars in front of you are.
賽:這是「向前看」的點, 供自動調整航程控制用, 它會根據前車的距離 幫助我們調整速度。
CA: And so, you've also got an example, I think, of how the actual learning part takes place. Maybe we can see that. Talk about this.
克:我想,你應該也可以舉個例說明 學習的部份實際上如何進行。 也許我們可以 邊看那畫面,邊談這個。
ST: This is an example where we posed a challenge to Udacity students to take what we call a self-driving car Nanodegree. We gave them this dataset and said "Hey, can you guys figure out how to steer this car?" And if you look at the images, it's, even for humans, quite impossible to get the steering right. And we ran a competition and said, "It's a deep learning competition, AI competition," and we gave the students 48 hours. So if you are a software house like Google or Facebook, something like this costs you at least six months of work. So we figured 48 hours is great. And within 48 hours, we got about 100 submissions from students, and the top four got it perfectly right. It drives better than I could drive on this imagery, using deep learning. And again, it's the same methodology. It's this magical thing. When you give enough data to a computer now, and give enough time to comprehend the data, it finds its own rules.
賽:這是我們挑戰 Udacity 學生的一個例子, 是取得「自駕車奈米學位」的挑戰。 我們給他們這個資料集, 說:「你們能不能想出 要如何駕駛這台車?」 如果從影像來看, 即使是人類操縱也很難駕駛好。 我們進行了一項競賽,並說: 「這是場深度學習競賽, 人工智慧競賽。」 我們給學生 48 小時。 如果你是間軟體公司, 如 Google 或臉書, 像這樣的東西會花你 至少六個月的功夫。 所以我們認為 48 小時是很棒的。 在 48 小時內,我們得到了 約一百件學生提交的結果, 前四名完全無誤。 它駕駛得比我能在 這影像上駕駛得還要好, 用的就是深度學習。 同樣的方法, 很神奇, 當你提供電腦足夠的資料, 並給它足夠時間來理解那些資料, 它就會自己找到規則。
CA: And so that has led to the development of powerful applications in all sorts of areas. You were talking to me the other day about cancer. Can I show this video?
克:所以那就導致了 強大應用程式的發展, 在各領域都有。 之前你有和我談過癌症的事。 我能播那段影片嗎?
ST: Yeah, absolutely, please. CA: This is cool.
賽:當然,請放。 克:這很酷。
ST: This is kind of an insight into what's happening in a completely different domain. This is augmenting, or competing -- it's in the eye of the beholder -- with people who are being paid 400,000 dollars a year, dermatologists, highly trained specialists. It takes more than a decade of training to be a good dermatologist. What you see here is the machine learning version of it. It's called a neural network. "Neural networks" is the technical term for these machine learning algorithms. They've been around since the 1980s. This one was invented in 1988 by a Facebook Fellow called Yann LeCun, and it propagates data stages through what you could think of as the human brain. It's not quite the same thing, but it emulates the same thing. It goes stage after stage. In the very first stage, it takes the visual input and extracts edges and rods and dots. And the next one becomes more complicated edges and shapes like little half-moons. And eventually, it's able to build really complicated concepts. Andrew Ng has been able to show that it's able to find cat faces and dog faces in vast amounts of images.
賽:這有點像是對完全不同的領域 洞察所發生的事。 在旁觀者眼裡, 這是擴增,或者可說是 與年賺 $40 萬美元的人競爭: 皮膚科醫生, 他們是受過高度訓練的專家, 要受十年以上的訓練才可能 成為好的皮膚科醫生。 這裡所看到的是它的機器學習版本。 稱為「神經網路」, 神經網路是機器學習 演算法的專有名詞, 大約出現於 1980 年代。 這個是在 1988 年由臉書的 研究專員揚勒丘恩所發明, 它傳播數據的階段 透過一種你可視為是人腦的方式。 它不是人腦,但它模仿人腦。 一個階段接著一個階段, 在第一個階段取得視覺輸入, 粹取出邊緣、細竿,和點。 下個階段就變成更複雜的邊緣 以及形狀,像是半月。 最終,它能建立出非常複雜的概念。 吳恩達就展示過, 它能夠在非常大量的影像中找出 貓和狗的臉。
What my student team at Stanford has shown is that if you train it on 129,000 images of skin conditions, including melanoma and carcinomas, you can do as good a job as the best human dermatologists. And to convince ourselves that this is the case, we captured an independent dataset that we presented to our network and to 25 board-certified Stanford-level dermatologists, and compared those. And in most cases, they were either on par or above the performance classification accuracy of human dermatologists.
我在史丹佛的學生團隊也展示過, 如果你用十二萬九千張 皮膚症狀的影像來訓練它, 包括黑色素瘤和癌, 你就能和最好的人類皮膚科醫生 做得一樣好。 為了說服我們自己確實是如此, 我們取得了一個獨立的資料集, 拿給我們的網路看, 也拿給 25 位認證過的 史丹佛水準的皮膚科醫生看, 來做比較。 在大部份狀況, 在分類正確性上, 網路的表現都和人類皮膚科醫生 並駕齊驅或更好。
CA: You were telling me an anecdote. I think about this image right here. What happened here?
克:你跟我說過一則軼事。 上面的這張影像。 這裡發生了什麼事?
ST: This was last Thursday. That's a moving piece. What we've shown before and we published in "Nature" earlier this year was this idea that we show dermatologists images and our computer program images, and count how often they're right. But all these images are past images. They've all been biopsied to make sure we had the correct classification. This one wasn't. This one was actually done at Stanford by one of our collaborators. The story goes that our collaborator, who is a world-famous dermatologist, one of the three best, apparently, looked at this mole and said, "This is not skin cancer." And then he had a second moment, where he said, "Well, let me just check with the app." So he took out his iPhone and ran our piece of software, our "pocket dermatologist," so to speak, and the iPhone said: cancer. It said melanoma. And then he was confused. And he decided, "OK, maybe I trust the iPhone a little bit more than myself," and he sent it out to the lab to get it biopsied. And it came up as an aggressive melanoma. So I think this might be the first time that we actually found, in the practice of using deep learning, an actual person whose melanoma would have gone unclassified, had it not been for deep learning.
賽:時間是上星期四, 是個進行中的故事。 我們之前展示過,且今年稍早 也刊在「Nature」期刊中, 想法是,我們讓皮膚科醫生看影像, 也讓我們的電腦程式看, 計算它們多常判斷正確。 但所有影像都是過去的影像。 都已經過切片檢查,確保分類正確。 這一張卻不是。 這張其實是在史丹佛 由我們的合作者之一做的。 故事是,我們的合作者 是世界知名的皮膚科醫生, 很顯然是最好的三位之一, 他看著這個痣,說: 「這不是皮膚癌。」 他想了一下,接著又說: 「讓我用應用程式確認一下。」 他拿出他的 iPhone, 執行我們的軟體, 可說是我們的「口袋皮膚科醫生」, 而 iPhone 說:癌症。 它說是黑色素瘤。 他很困惑。 他決定:「好,也許我應該相信 iPhone 比相信我自己多一點點。」 他把它送去實驗室做切片檢查。 結果是惡性黑色素瘤。 我想,這可能是我們第一次 真正在深度學習的實做中遇到, 如果沒有深度學習的話, 這個人的黑色素瘤就不會被發現。
CA: I mean, that's incredible.
克:那很了不起。
(Applause)
(掌聲)
It feels like there'd be an instant demand for an app like this right now, that you might freak out a lot of people. Are you thinking of doing this, making an app that allows self-checking?
感覺現在對於像這樣的應用程式, 有很迫切的需求, 你可能會嚇壞很多人。 你有想過要這麼做嗎? 做個自我檢測的應用程式?
ST: So my in-box is flooded about cancer apps, with heartbreaking stories of people. I mean, some people have had 10, 15, 20 melanomas removed, and are scared that one might be overlooked, like this one, and also, about, I don't know, flying cars and speaker inquiries these days, I guess. My take is, we need more testing. I want to be very careful. It's very easy to give a flashy result and impress a TED audience. It's much harder to put something out that's ethical. And if people were to use the app and choose not to consult the assistance of a doctor because we get it wrong, I would feel really bad about it. So we're currently doing clinical tests, and if these clinical tests commence and our data holds up, we might be able at some point to take this kind of technology and take it out of the Stanford clinic and bring it to the entire world, places where Stanford doctors never, ever set foot.
賽:我的收件匣被關於癌症 應用程式的信件給淹滿了, 信上都是人們的心碎故事。 有些人已經移除了 10、15、20 個黑色素瘤, 很害怕會漏掉任何一個,就像這個, 還有些內容是,我不知道, 飛天車、演說邀請,我猜是吧。 我的反應是,我們需要更多測試。 我想要非常小心。 很容易就可以丟出亮眼的結果 來讓 TED 觀眾印象深刻。 要端出合乎道德的東西就難很多。 如果人們要用這個應用程式, 且選擇不去諮詢醫生的協助, 而我們弄錯的話, 我就會感覺非常糟。 所以我們目前在做臨床實驗, 如果這些實驗開始之後, 我們的資料站得住腳, 在某個時點,我們或許可以把這技術 拿到史丹佛臨床課之外, 把它帶給全世界, 帶到史丹佛的醫生 從來沒有去過的地方。
CA: And do I hear this right, that it seemed like what you were saying, because you are working with this army of Udacity students, that in a way, you're applying a different form of machine learning than might take place in a company, which is you're combining machine learning with a form of crowd wisdom. Are you saying that sometimes you think that could actually outperform what a company can do, even a vast company?
克:我有沒有聽正確, 聽起來像是你在說 因為你在和這支 Udacity 學生大軍合作, 以某種方式,你們在應用 一種不同形式的機器學習, 可能會發生在公司中的形式, 也就是你們將機器學習 與一種群眾智慧結合。 你是不是在說, 有時你認為那能夠勝過 公司所能做到的,甚至大型公司?
ST: I believe there's now instances that blow my mind, and I'm still trying to understand. What Chris is referring to is these competitions that we run. We turn them around in 48 hours, and we've been able to build a self-driving car that can drive from Mountain View to San Francisco on surface streets. It's not quite on par with Google after seven years of Google work, but it's getting there. And it took us only two engineers and three months to do this. And the reason is, we have an army of students who participate in competitions. We're not the only ones who use crowdsourcing. Uber and Didi use crowdsource for driving. Airbnb uses crowdsourcing for hotels. There's now many examples where people do bug-finding crowdsourcing or protein folding, of all things, in crowdsourcing. But we've been able to build this car in three months, so I am actually rethinking how we organize corporations.
賽:我相信現在有一些 讓我很興奮的例子, 我還在試著了解。 克里斯指的是 我們的競賽才進行了大約 四十八小時就打開來用; 而我們建的自駕車 能從山景城開上馬路去到舊金山; 它尚未趕上 Google 投入七年心血的成果, 但是就快追上了。 我們的研發只花了兩個工程師 用了三個月就做到這樣, 原因是,我們有一支學生大軍, 參與競賽的那些學生。 我們並非唯一使用「群眾外包」的人, Uber 和 Didi 用群眾外包做駕駛, Airbnb 用群眾外包做飯店。 現在有許多例子是 群眾外包除錯工作 或群眾外包蛋白質摺疊等。 但我們能在三個月內建造這台車, 所以我其實在重新思考, 我們要如何組織企業。
We have a staff of 9,000 people who are never hired, that I never fire. They show up to work and I don't even know. Then they submit to me maybe 9,000 answers. I'm not obliged to use any of those. I end up -- I pay only the winners, so I'm actually very cheapskate here, which is maybe not the best thing to do. But they consider it part of their education, too, which is nice. But these students have been able to produce amazing deep learning results. So yeah, the synthesis of great people and great machine learning is amazing.
我們有從未被僱用的九千名員工, 我也從未開除他們, 我不知道他們什麼時候工作。 後來他們提交大約九千份答案給我。 我沒有義務要用任何一個答案。 最後我只付錢給贏家, 所以在這裡我算是個小氣鬼, 這不見得是最好的做法。 但他們認為這是他們 教育的一部份,這樣想很好。 但這些學生能夠產出非常 了不起的深度學習結果。 所以,厲害的人結合 偉大的機器學習是很驚人的。
CA: I mean, Gary Kasparov said on the first day [of TED2017] that the winners of chess, surprisingly, turned out to be two amateur chess players with three mediocre-ish, mediocre-to-good, computer programs, that could outperform one grand master with one great chess player, like it was all part of the process. And it almost seems like you're talking about a much richer version of that same idea.
克:加里卡斯帕洛夫 在(TED 2017)第一天說, 很意外的,棋賽的贏家 是兩位業餘的棋手, 用三個平庸、中上的電腦程式 就勝過了一個大師 和一個很棒的棋手, 就像這過程的一部份, 幾乎和你談的想法同樣, 而是更豐富的版本。
ST: Yeah, I mean, as you followed the fantastic panels yesterday morning, two sessions about AI, robotic overlords and the human response, many, many great things were said. But one of the concerns is that we sometimes confuse what's actually been done with AI with this kind of overlord threat, where your AI develops consciousness, right? The last thing I want is for my AI to have consciousness. I don't want to come into my kitchen and have the refrigerator fall in love with the dishwasher and tell me, because I wasn't nice enough, my food is now warm. I wouldn't buy these products, and I don't want them. But the truth is, for me, AI has always been an augmentation of people. It's been an augmentation of us, to make us stronger. And I think Kasparov was exactly correct. It's been the combination of human smarts and machine smarts that make us stronger. The theme of machines making us stronger is as old as machines are. The agricultural revolution took place because it made steam engines and farming equipment that couldn't farm by itself, that never replaced us; it made us stronger. And I believe this new wave of AI will make us much, much stronger as a human race.
賽:是的,昨天早上的小組討論很棒, 兩場關於人工智慧的討論, 機器超載和人類回應, 說到很多很棒的內容。 但是讓人擔心的事情之一 是有時我們混淆了 人工智慧實際做的事 和機器超載的威脅, 也就是人工智慧發展出意識,對吧? 我最不想要人工智慧有意識。 我可不想進到廚房, 發現冰箱愛上了洗碗機, 然後告訴我,因為我不夠好, 我的食物現在溫的。 我不會買這些產品, 我也不想要它們。 但,事實是,對我來說, 人工智慧一直都是人的擴增。 它一直是我們的擴增, 讓我們更強大。 我認為卡斯帕洛夫完全正確。 一直都是人類的聰明 結合機器的聰明, 才讓我們更強。 機器讓我們更強的主題, 就像機器本身一樣老。 發生農業革命是因為 做出了蒸汽引擎以及耕作設備, 它們不會自己耕作或取代我們, 而是會讓我們更強。 而我相信,這波新的人工智慧風潮 會讓我們人類更強大許多。
CA: We'll come on to that a bit more, but just to continue with the scary part of this for some people, like, what feels like it gets scary for people is when you have a computer that can, one, rewrite its own code, so, it can create multiple copies of itself, try a bunch of different code versions, possibly even at random, and then check them out and see if a goal is achieved and improved. So, say the goal is to do better on an intelligence test. You know, a computer that's moderately good at that, you could try a million versions of that. You might find one that was better, and then, you know, repeat. And so the concern is that you get some sort of runaway effect where everything is fine on Thursday evening, and you come back into the lab on Friday morning, and because of the speed of computers and so forth, things have gone crazy, and suddenly --
克:我們等等會再談那個話題, 但先繼續聊這個 對一些人來說很駭人的部份, 對人們來說,會覺得害怕的是 你讓電腦能重寫它自己的程式, 它就能複製多個自己, 嘗試各種不同版本的程式, 甚至可能是隨機嘗試, 然後再確認看看 目標是否有達成或改善。 所以,比如,目標是要在一項 智力測驗中得到更好的成績。 一台電腦只要還算擅長, 就能嘗試一百萬個版本, 可能會找到一版比較理想, 重覆做下去。 擔心的是,你會有某種失控效應, 在星期四晚上一切都很好, 你星期五早上回到實驗室, 因為電腦的速度等等, 一切就天翻地覆,突然間──
ST: I would say this is a possibility, but it's a very remote possibility. So let me just translate what I heard you say. In the AlphaGo case, we had exactly this thing: the computer would play the game against itself and then learn new rules. And what machine learning is is a rewriting of the rules. It's the rewriting of code. But I think there was absolutely no concern that AlphaGo would take over the world. It can't even play chess.
賽:我會說,這有可能, 卻是非常遙遠的可能。 所以讓我翻譯一下我剛聽你說的。 阿爾法圍棋的例子就有這樣的狀況: 電腦會自己對抗自己來下棋, 接著學習新規則。 機器學習就是重寫規則。 就是重寫程式。 但我認為完全不用擔心 阿爾法圍棋會佔領世界。 它不會下西洋棋。
CA: No, no, no, but now, these are all very single-domain things. But it's possible to imagine. I mean, we just saw a computer that seemed nearly capable of passing a university entrance test, that can kind of -- it can't read and understand in the sense that we can, but it can certainly absorb all the text and maybe see increased patterns of meaning. Isn't there a chance that, as this broadens out, there could be a different kind of runaway effect?
克:不,不,現在這些 都還是非常單一領域的東西。 但有可能去想像, 我是指,我們剛看到幾乎有能力 通過大學入學測驗的電腦, 就像它無法用 我們的方式去閱讀及了解, 但它絕對可以吸收所有的文字, 也許能看到越來越多有意義的模式。 有沒有可能,當拓展更廣時, 會是不同種類的失控效應?
ST: That's where I draw the line, honestly. And the chance exists -- I don't want to downplay it -- but I think it's remote, and it's not the thing that's on my mind these days, because I think the big revolution is something else. Everything successful in AI to the present date has been extremely specialized, and it's been thriving on a single idea, which is massive amounts of data. The reason AlphaGo works so well is because of massive numbers of Go plays, and AlphaGo can't drive a car or fly a plane. The Google self-driving car or the Udacity self-driving car thrives on massive amounts of data, and it can't do anything else. It can't even control a motorcycle. It's a very specific, domain-specific function, and the same is true for our cancer app. There has been almost no progress on this thing called "general AI," where you go to an AI and say, "Hey, invent for me special relativity or string theory." It's totally in the infancy.
賽:老實說,我會把底線設在那裡。 存在這可能性,我不想低估它, 但我認為它很遙遠, 現在我腦中不會去想這個, 因為我認為大革命是另一回事。 目前為止,人工智慧所有的成功, 都是極度專門化的, 一直以來,它能興盛全靠一個辦法: 大量的資料。 阿爾法圍棋能如此成功 是因為下過大量的圍棋棋譜, 阿爾法圍棋無法開車或開飛機。 Google 的自動駕駛汽車或 Udacity 的自動駕駛汽車 能成功是因為有大量的資料, 它們無法做其他事, 甚至無法開摩托車。 它是非常明確、專門領域的功能, 我們的癌症應用程式也是如此。 所謂的「一般性人工智慧」幾無進展, 就是你可以叫它: 「嘿,為我發明 狹義相對論或弦理論」的那種 完全還在嬰兒期。
The reason I want to emphasize this, I see the concerns, and I want to acknowledge them. But if I were to think about one thing, I would ask myself the question, "What if we can take anything repetitive and make ourselves 100 times as efficient?" It so turns out, 300 years ago, we all worked in agriculture and did farming and did repetitive things. Today, 75 percent of us work in offices and do repetitive things. We've become spreadsheet monkeys. And not just low-end labor. We've become dermatologists doing repetitive things, lawyers doing repetitive things. I think we are at the brink of being able to take an AI, look over our shoulders, and they make us maybe 10 or 50 times as effective in these repetitive things. That's what is on my mind.
我想要強調這點的理由 是我知道人們擔心,我聽見了。 但如果要我思考一件事,我會自問: 「如果我們能夠把任何重覆事物的 效率提高一百倍,會如何?」 事實證明,三百年前我們都從事農業, 耕種,做重覆性的事。 現今,我們有 75% 的人 在辦公室工作, 做重覆性的事。 我們已變成了試算表猴子。 不只是低階勞工, 我們的皮膚科醫生 已經開始做重覆性工作, 律師也做重覆性工作。 我認為我們正處於 能夠採用 AI 的邊緣, 保持警覺, 可以提高我們執行 重複性工作的效率十或五十倍。 我在想的是這個。
CA: That sounds super exciting. The process of getting there seems a little terrifying to some people, because once a computer can do this repetitive thing much better than the dermatologist or than the driver, especially, is the thing that's talked about so much now, suddenly millions of jobs go, and, you know, the country's in revolution before we ever get to the more glorious aspects of what's possible.
克:那聽起來非常讓人興奮。 對於一些人來說,要達成 那樣的過程似乎有點嚇人, 因為一旦電腦能做重覆性的事, 且做得比皮膚科醫生好, 尤其做得比司機好, 這是現在熱門的話題, 突然間,幾百萬個工作就沒了, 你知道的,這個國家正在革命之中, 我們都還來不及去做到 可能達成的輝煌面。
ST: Yeah, and that's an issue, and it's a big issue, and it was pointed out yesterday morning by several guest speakers. Now, prior to me showing up onstage, I confessed I'm a positive, optimistic person, so let me give you an optimistic pitch, which is, think of yourself back 300 years ago. Europe just survived 140 years of continuous war, none of you could read or write, there were no jobs that you hold today, like investment banker or software engineer or TV anchor. We would all be in the fields and farming. Now here comes little Sebastian with a little steam engine in his pocket, saying, "Hey guys, look at this. It's going to make you 100 times as strong, so you can do something else." And then back in the day, there was no real stage, but Chris and I hang out with the cows in the stable, and he says, "I'm really concerned about it, because I milk my cow every day, and what if the machine does this for me?"
賽:是啊,那是個課題,大課題, 昨天早上有幾位嘉賓指出這一點。 在我上台之前, 我坦白說我是個正面、樂觀的人, 讓我為各位定個樂觀的調, 就是,試想你回到三百年前, 歐洲剛結束了持續 140 年的戰爭, 你們都不會讀或寫, 沒有你們現在做的工作, 比如投資銀行家、 軟體工程師、電視台主播, 我們都在田野裡耕種。 現在,來了一個小賽巴斯汀, 口袋中有個小蒸氣引擎, 說:「嘿,各位,看看這個。 它會讓你強大一百倍, 這樣你們就可以做其它事了。」 在那個年代,沒有真正的舞台, 但克里斯和我在畜舍中 和乳牛在一起, 他說:「我真的很擔心這事, 我每天給乳牛擠奶, 如果讓機器來為我擠,會如何?
The reason why I mention this is, we're always good in acknowledging past progress and the benefit of it, like our iPhones or our planes or electricity or medical supply. We all love to live to 80, which was impossible 300 years ago. But we kind of don't apply the same rules to the future. So if I look at my own job as a CEO, I would say 90 percent of my work is repetitive, I don't enjoy it, I spend about four hours per day on stupid, repetitive email. And I'm burning to have something that helps me get rid of this. Why? Because I believe all of us are insanely creative; I think the TED community more than anybody else. But even blue-collar workers; I think you can go to your hotel maid and have a drink with him or her, and an hour later, you find a creative idea. What this will empower is to turn this creativity into action. Like, what if you could build Google in a day? What if you could sit over beer and invent the next Snapchat, whatever it is, and tomorrow morning it's up and running?
我提到這一點的原因是, 我們向來都很擅長認可 過去的進展和它帶來的益處, 就像我們的 iPhone、 飛機、電力、醫材。 我們都喜歡活到八十歲, 這在三百年前是不可能的。 但我們似乎不太會用 同樣的規則看未來。 如果我看我自己的工作,執行長, 我會說,我 90% 的工作 是重覆性的, 我並不享受做那些, 我每天要花大約四小時在 愚蠢、重覆性的電子郵件上。 我極渴望有什麼能協助我擺脫這些。 為什麼? 因為我相信我們所有人 都極度有創意; 我認為,比起其他人, TED 社區更是如此。 但即使藍領階級勞工;我認為 你可以去找你的飯店服務員, 和他或她喝杯飲料, 一小時後,你會找到一個創意想法。 人工智慧能賦予人能力, 將創意轉化為行動。 比如,如果你能在一天就 建造出 Google,會如何? 如果你能坐著喝啤酒,就發明出 下一個 Snapchat,會如何? 不論發明的是什麼, 明早它就可以開始運作,會如何?
And that is not science fiction. What's going to happen is, we are already in history. We've unleashed this amazing creativity by de-slaving us from farming and later, of course, from factory work and have invented so many things. It's going to be even better, in my opinion. And there's going to be great side effects. One of the side effects will be that things like food and medical supply and education and shelter and transportation will all become much more affordable to all of us, not just the rich people.
那不是科幻小說。 將會發生的事是, 我們已經在歷史中。 我們已經釋放出了這了不起的創意, 讓我們脫離耕種的奴役, 當然,之後又脫離了 工廠工作的奴役, 且發明出了這麼多東西。 依我所見,將來還會更好。 將來會有很大的副作用。 其中一項副作用會是, 很多東西,比如食物、 醫材、教育、庇護所 以及交通, 都會變為大家更負擔得起, 不只是有錢人的專利。
CA: Hmm. So when Martin Ford argued, you know, that this time it's different because the intelligence that we've used in the past to find new ways to be will be matched at the same pace by computers taking over those things, what I hear you saying is that, not completely, because of human creativity. Do you think that that's fundamentally different from the kind of creativity that computers can do?
克:嗯。 所以,當馬丁福特主張, 這次會有所不同, 因為我們在過去用來 找出新方式的智慧, 將會以同樣的速度 被接手那些事的電腦給比過, 我聽到你說並不完全如此, 因為人類有創意。 你認為那和電腦能做的那種創意 在根本上是不同的嗎?
ST: So, that's my firm belief as an AI person -- that I haven't seen any real progress on creativity and out-of-the-box thinking. What I see right now -- and this is really important for people to realize, because the word "artificial intelligence" is so threatening, and then we have Steve Spielberg tossing a movie in, where all of a sudden the computer is our overlord, but it's really a technology. It's a technology that helps us do repetitive things. And the progress has been entirely on the repetitive end. It's been in legal document discovery. It's been contract drafting. It's been screening X-rays of your chest. And these things are so specialized, I don't see the big threat of humanity. In fact, we as people -- I mean, let's face it: we've become superhuman. We've made us superhuman. We can swim across the Atlantic in 11 hours. We can take a device out of our pocket and shout all the way to Australia, and in real time, have that person shouting back to us. That's physically not possible. We're breaking the rules of physics. When this is said and done, we're going to remember everything we've ever said and seen, you'll remember every person, which is good for me in my early stages of Alzheimer's. Sorry, what was I saying? I forgot.
賽:身為人工智慧人,我堅定相信 我尚未看到任何真正創意上的進展, 也沒有創造性思維。 我現在看到的── 人們很需要了解這一點, 因為「人工智慧」這詞 深具威脅性的, 史帝芬史匹柏拍了一部電影, 電影中電腦突然成了我們的主人, 但它其實只是一項技術, 協助我們做重覆工作的技術, 而進展完全在重覆性方面: 在法律文件探索上有進展, 在合約起草上有進展, 在判讀胸腔 X 光上有進展。 這些工作都很專門化, 我看不出對人類有什麼大威脅。 事實上,我們人類── 我們得承認,我們已經變成超人。 我們已經把自己變成超人。 我們可以在 11 小時泳渡大西洋。 我們能從口袋中拿出一個裝置 然後對著遙遠的澳洲大吼, 而且對方還會即時吼回來。 在物理上是不可能的, 我們打破了物理的規則。 說到底, 我們會記得曾經說過和看過的一切, 你們將會記得每個人, 對在阿滋海默前期的我是件好事。 抱歉,我剛在說什麼?我忘了。
CA: (Laughs)
克:(笑聲)
ST: We will probably have an IQ of 1,000 or more. There will be no more spelling classes for our kids, because there's no spelling issue anymore. There's no math issue anymore. And I think what really will happen is that we can be super creative. And we are. We are creative. That's our secret weapon.
賽:我們將來可能會有 超過 1,000 的智商。 我們的孩子將不用再學習拼字, 因為將不再有拼字問題。 將不再有數學問題。 我認為會發生的是, 我們會超級有創意。 我們是有創意的, 那是我們的秘密武器。
CA: So the jobs that are getting lost, in a way, even though it's going to be painful, humans are capable of more than those jobs. This is the dream. The dream is that humans can rise to just a new level of empowerment and discovery. That's the dream.
克:所以正在消失中的工作, 在某個層面上,即使會很痛苦, 人類的能力是超過這些工作的。 這是個夢想。 夢想是人類可以崛起 爬升到賦能與探索的新層級。 這是個夢想。
ST: And think about this: if you look at the history of humanity, that might be whatever -- 60-100,000 years old, give or take -- almost everything that you cherish in terms of invention, of technology, of things we've built, has been invented in the last 150 years. If you toss in the book and the wheel, it's a little bit older. Or the axe. But your phone, your sneakers, these chairs, modern manufacturing, penicillin -- the things we cherish. Now, that to me means the next 150 years will find more things. In fact, the pace of invention has gone up, not gone down, in my opinion. I believe only one percent of interesting things have been invented yet. Right? We haven't cured cancer. We don't have flying cars -- yet. Hopefully, I'll change this. That used to be an example people laughed about. (Laughs) It's funny, isn't it? Working secretly on flying cars. We don't live twice as long yet. OK? We don't have this magic implant in our brain that gives us the information we want. And you might be appalled by it, but I promise you, once you have it, you'll love it. I hope you will. It's a bit scary, I know.
賽:想想這一點: 如果你去看人類的歷史, 也許 6~10 萬年前左右, 幾乎你所珍惜的一切, 發明、科技、我們建造的東西, 都是在最近的 150 年間發明的。 如果你把書和輪子放進來, 那就古老一些。 或是斧頭。 但你的手機、你的運動鞋、 這些椅子、現代工業、盤尼西林── 我們珍視的東西。 對我來說,那意味著, 接下來的 150 年會發現更多東西。 事實上,依我所見,發明的速度 已經變快了,不是變慢。 我相信,我們才只發明了 1% 有趣的東西出來。對吧? 我們還沒有治癒癌症。 我們沒有飛天車,還沒有。 希望我能改變這一點。 以前那是個會讓人發笑的例子。 (笑聲) 很有趣,是吧? 暗地裡致力發明飛天車。 我們的壽命還沒到兩倍長。是嗎? 我們在大腦中還沒有神奇的植入物 能給予我們想要的資訊。 你可能會覺得它很駭人, 但我保證,一旦你有了它 就會愛上它。 我希望你會。 它有點可怕,我知道。
There are so many things we haven't invented yet that I think we'll invent. We have no gravity shields. We can't beam ourselves from one location to another. That sounds ridiculous, but about 200 years ago, experts were of the opinion that flight wouldn't exist, even 120 years ago, and if you moved faster than you could run, you would instantly die. So who says we are correct today that you can't beam a person from here to Mars?
還有好多我認為我們應該 發明的東西還沒被發明出來。 我們沒有重力保護罩。 我們無法把自己從一地 用光束傳送到另一地。 那聽起來很荒謬, 但大約 200 年前, 專家認為飛機不會存在, 甚至 120 年前。 還有認為如果你移動速度 比你的跑步速度快, 你就會馬上死掉。 所以現在誰敢肯定說 我們不能把一個人用光束 從這裡傳送到火星?
CA: Sebastian, thank you so much for your incredibly inspiring vision and your brilliance. Thank you, Sebastian Thrun. That was fantastic. (Applause)
克:賽巴斯汀,非常謝謝你 分享啟發靈感的遠景和你的智慧。 謝謝你,賽巴斯汀索朗。 賽:很棒的經驗。(掌聲)