今年3月,谷歌下屬公司DeepMind開發(fā)的人工智能軟件AlphaGo在圍棋比賽中大勝世界圍棋冠軍李世石,在世界范圍內(nèi)引起了軒然大波。AlphaGo的勝利代表了人工智能飛躍式的進(jìn)步,人們紛紛開始擔(dān)憂機(jī)器人對人類產(chǎn)生威脅的時代即將到來……人工智能究竟是什么東西,居然可能會有超越人類的智慧?先隨下文一起來了解一下吧!
If you touch a hot metal object, you will yank1) your hand away immediately. When this happens to you the first time, the sequence of events and the result (the burning of your hand) gets stored in your brain. This is what we call an experience. When you see a hot metal object next time, you will not touch it. You will use the knowledge of your previous experience and decide to not repeat it again.
This process of learning, comparing a previous experience, making a decision and acting upon it is the key to human intelligence. We can make more and more complicated decisions by learning from our past experiences.
Ever since machines were invented, scientists have dreamt of making them learn and perform intelligent tasks—like humans.
Artificial intelligence (AI) is a branch of science which is into2) making machines think like humans. These machines, or computers, can store large amounts of information and process them accurately and at an amazing speed. What they lack is an ability to learn and make "intelligent decisions".
What do we need to make an intelligent machine? A memory or a space where experiences or information can be stored, and a method of applying these experiences to new ones, comparing experiences to come to logical conclusions, like holding the hot object with a glove on. That would be an intelligent machine.
Take your iron3) for example. The electric iron understands that its temperature is beyond what is required and automatically switches itself off4). We could say that the electric iron is intelligent as it can react to a particular state (the iron being hot), make a decision based on it and switch itself off. However, since the iron has not learned this through experience, it is not a truly intelligent machine.
Scientists are creating new software programs which try to recreate the process of human learning in a computer, in an attempt to make them "think". These programs try to copy the functioning of the brain. One such program is called a neural5) network.
摸到發(fā)燙的金屬物體,你會立馬猛地抽回手來。當(dāng)你第一次碰到這種情況時,整個事件的經(jīng)過和結(jié)果(手被燙)便會被存入你的大腦,這就是我們所說的“經(jīng)歷”。當(dāng)你下次看到發(fā)燙的金屬物時,你就不會伸手去摸它了。你會運(yùn)用先前的經(jīng)歷所帶來的知識,決定不去重蹈覆轍。
這一“認(rèn)識了解—對比過往經(jīng)歷—做出決策—付諸行動”的過程正是人類智能形成的關(guān)鍵。我們能夠從以往的各種經(jīng)歷中汲取教訓(xùn),做出愈加復(fù)雜的決策。
自從機(jī)器問世以來,科學(xué)家們就一直夢想著能讓它們學(xué)習(xí)并開展智力活動——就像人類一樣。
人工智能(簡稱AI)是一個科學(xué)分支,致力于制造可以像人類一樣思考的機(jī)器。這些機(jī)器或計算機(jī)可以儲存大量的信息,并能夠以驚人的速度準(zhǔn)確地對這些信息進(jìn)行加工處理。它們所缺乏的是學(xué)習(xí)并做出“明智決定”的能力。
那制造一臺智能機(jī)器都需要什么條件呢?一個可以存放經(jīng)歷或信息的存儲器或者空間,以及一種可將這些經(jīng)歷應(yīng)用到新環(huán)境的方法,即比較各種經(jīng)歷以得出合理結(jié)論(比如,戴上手套去拿高溫物體)。滿足這些條件的機(jī)器就是智能機(jī)器。
以電熨斗為例,電熨斗感知到自身溫度超出需求時就會自動斷電。我們可以說,電熨斗具有智能,因為它可以針對特定情況(熨斗過熱)有所反應(yīng),并據(jù)此做出決定,自動斷電。不過,由于電熨斗并不是通過自身經(jīng)歷學(xué)會的這項本領(lǐng),所以它還算不上是真正的智能機(jī)器。
科學(xué)家們正在開發(fā)一些新的軟件程序,試圖讓它們進(jìn)行“思考”。這些程序力圖在計算機(jī)上重建人類的學(xué)習(xí)過程,試圖復(fù)制大腦的運(yùn)轉(zhuǎn)機(jī)制,其中一個程序叫做“神經(jīng)網(wǎng)絡(luò)”。
Our brain is composed of6) billions of densely packed cells called neurons. Each neuron is like a tiny individual switch in a net of billions of such neurons.
Whenever a particular piece of information, like someone's telephone number reaches your brain, it creates a pattern of on and off switches using these neurons.
Let's use an example to understand this phenomenon. We put up garlands7) of electric lights to decorate our houses during Diwali8). These lights create various patterns and designs, one switch creates a series of circles, another switch a pattern of flowers and so on.
A neural network is like these garlands of lights: A particular input creates a particular pattern. Each nerve cell or neuron in our brain acts like a light bulb. It creates a particular pattern on receiving an input.
When we memorize someone's telephone number, we actually create a pattern in our brain. And when we try to remember the same number, we simply try to recreate that pattern, unlike the lights which need to be switched on or off every time that pattern needs to be created.
A neural network is a copy of the brain's functioning inside a computer, using a software program. It can be taught to recognize patterns.
In fact, when it is trained, it can classify and identify patterns in a large amount of information. It can do all this at very high speeds and sometimes faster than humans.
This throws open innumerable possibilities. Imagine computers, which can look at the past weather and climate data, match them with current conditions and tell us where it is going to rain and how much.
我們的大腦由數(shù)億個密密麻麻的細(xì)胞構(gòu)成,這些細(xì)胞稱為神經(jīng)元。數(shù)億個這樣的神經(jīng)元構(gòu)成了一個神經(jīng)元網(wǎng)絡(luò),每個神經(jīng)元就像是網(wǎng)絡(luò)上一個小小的獨(dú)立開關(guān)。
每當(dāng)一條特定的信息,比如某人的電話號碼,進(jìn)入你的大腦,大腦就會啟動神經(jīng)元,生成一種開關(guān)模式。
我們可以打個比方來理解這一現(xiàn)象。過排燈節(jié)時,我們都會用一串串彩燈來裝點(diǎn)屋子。這些彩燈可以制造出不同的樣式和圖案,打開一個開關(guān)形成一系列圓圈,打開另一個開關(guān)變成花朵的圖案等等。
神經(jīng)網(wǎng)絡(luò)就好比這些彩燈串:一條特定的輸入信息會生成一個特定的模式。我們大腦里的每個神經(jīng)細(xì)胞或者說神經(jīng)元就像一個燈泡,收到一條輸入信息就生成一種特定的模式。
我們在記一個人的電話號碼時,實(shí)際上是在大腦里生成一種模式。當(dāng)我們試圖記起相同號碼的時候,我們只需努力再現(xiàn)這種模式即可,不用像燈泡那樣每次都要打開或關(guān)閉開關(guān)才能形成這種模式。
神經(jīng)網(wǎng)絡(luò)是在計算機(jī)內(nèi)部對大腦功能的復(fù)制版,通過使用一個軟件程序來實(shí)現(xiàn)。我們可以教它識別各種模式。
事實(shí)上,經(jīng)過訓(xùn)練的神經(jīng)網(wǎng)絡(luò)可以從海量的信息里對各種模式進(jìn)行歸類和識別,而且完成這些任務(wù)的速度很快,有時甚至快過人類。
這將產(chǎn)生無限可能。試想一下:電腦可以參照以往的天氣和氣候數(shù)據(jù),根據(jù)當(dāng)前的氣象條件進(jìn)行匹配,然后告訴我們哪里會降雨、雨量多少。
In 1950, famous mathematician Alan Turing devised a method of testing a computer's intelligence. A person is kept inside a closed cell9) and asked to speak to a hidden human being and a computer.
The person, who is also called the interrogator10) (one who questions), does not have any clue about who is the human being and who the machine. His task is to find out which of the two candidates is the computer, and which is the human by asking them questions. If the interrogator is unable to decide within a certain time, the machine is considered intelligent.
1950年,著名數(shù)學(xué)家艾倫·圖靈設(shè)計出一種測試計算機(jī)智能的方法。一個人被關(guān)進(jìn)一個封閉的房間,并被要求去跟隱藏的一個人和一臺計算機(jī)對話。
這個人,又稱訊問者(提問的人),完全不知道對方誰是人,誰是計算機(jī)。他的任務(wù)是通過對這兩者提問,辨出兩者哪個是計算機(jī),哪個是人。如果訊問者在一定時間內(nèi)無法做出判定,那么就可認(rèn)為那臺計算機(jī)具有智能。