破解密碼時，人工智能在“思考”什么？

A few weeks ago I went to the Imperial War Museum in London to watch an artificial intelligence program attempt to crack the mindbendingly complex Enigma code used by the Germans during the second world war. It did so in 12 minutes and 50 seconds. 數(shù)周前，我去了倫敦的帝國戰(zhàn)爭博物館(Imperial War Museum)，觀摩一個人工智能程序試圖破解德國人在二戰(zhàn)期間使用的復雜的英格瑪(Enigma)密碼。它用了12分50秒就破解了那套密碼。

Having already machine read some German language training data from Grimm’s Fairy Tales, the AI program crunched through billions of permutations generated by the four-rotor Enigma machine sifting combinations of letters for their “Germanness”. 在機器閱讀了來自《格林童話》(Grimm’s Fairy Tales)的一些德文培訓數(shù)據(jù)后，人工智能程序分析了由四個轉(zhuǎn)子的英格瑪機器生成的數(shù)十億個排列組合，在海量字母組合中尋找德文語句。

A challenge that had occupied some of Britain’s most brilliant mathematical minds at Bletchley Park for many months and at enormous cost was solved by a modern AI program in a few minutes for only £10. The program, developed by the data analytics company Enigma Pattern and boosted by 2,000 virtual servers, was able to check an astonishing 41m combinations a second. 曾經(jīng)讓布萊切利公園(Bletchley Park)一些英國最優(yōu)秀數(shù)學家冥思苦想好幾個月的挑戰(zhàn)，如今在幾分鐘內(nèi)就被現(xiàn)代人工智能程序解決了，而且成本只有10英鎊。由數(shù)據(jù)分析公司Enigma Pattern開發(fā)的這個程序，由2000個虛擬服務器提供支持，每秒能夠篩查驚人的4100萬個排列組合。

Simon Singh, the science writer and author of The Code Book 科學作家、《密碼故事》(The Code Book)一書的作者西蒙•辛格(Simon Singh)表示，英格瑪密碼在當時被認為是近乎不可能破譯的。他在上述演示活動中表示：“那是一套極難破譯的密碼。它曾經(jīng)長期被認為是一個不可能的挑戰(zhàn)。”

, said the Enigma code was considered all but unbreakable in its day. “It is a tough, tough code to break. It was long thought to be an impossible challenge,” he said at the event. 此次演示是一個聰明的宣傳噱頭，展示了人工智能與強大計算能力相結(jié)合之后快速增長的能力。但是，它也進一步證明了布萊切利公園的密碼破譯天才阿蘭•圖靈(Alan Turing)的早期直覺，即計算機可以在不理解的情況下展現(xiàn)出強大勝任能力。

The demonstration was a smart publicity stunt to show off the fast-growing capabilities of AI when combined with massive computing power. But it also came as further proof of the early intuition of Alan Turing, the genius codebreaker of Bletchley Park, that computers can exhibit competence without comprehension. 人工智能程序沒有意識到它正在試圖破解德國的密碼機，甚至不知道那些德文單詞的含義。相反，它將模式識別技術(shù)與蠻力計算相結(jié)合，產(chǎn)生一個基于概率的解決方案。機器智能即使在沒有完全理解問題的情況下，也能提供正確的答案。

The AI program had no conception that it was trying to crack a German cipher machine, nor even what the German words meant. Rather, it was combining pattern recognition technology and brute force computing to produce a probabilistic solution. Machine intelligence can provide the right answer even when it does not fully understand the question. 這種現(xiàn)象正蔓延至日益廣泛的人類活動范圍，從玩游戲(比如象棋和圍棋)到閱讀法律文件、解讀放射掃描結(jié)果以及駕駛汽車。計算機不理解它們正在執(zhí)行的任務，即便它們比人類更高效率地執(zhí)行了任務。

This phenomenon is spreading to an ever-widening range of human activities from playing games, such as chess and Go, to reading legal documents, interpreting radiology scans and driving cars. Computers do not comprehend what tasks they are performing even as they perform them more efficiently than humans. 計算機智能的勝任能力范圍不斷擴大，引發(fā)了許多棘手問題，癥結(jié)在于當機器幾乎能做所有事情時會發(fā)生什么?這個難題已經(jīng)催生了一大堆論述在人工智能時代身為人類將意味著什么的圖書。

The expanding competence of computer intelligence raises many ticklish questions about what will happen when machines can do almost everything. That dilemma has already spawned a shelf load of books about what it will mean to be human in the age of AI. 但是，英格瑪故事的一個不那么為人所知的方面突顯出，人類與機器之間仍然存在著顯著差異——即使這段插曲折射出的人性并不特別光彩。

But one lesser-known aspect of the Enigma story highlights a significant difference that remains between man and machine — even if this episode does not reflect particularly well on humans. 據(jù)辛格透露，戰(zhàn)爭結(jié)束后，英國當局沒有公開他們已經(jīng)破譯了英格瑪密碼這個事實，反而維持它不可破譯的假象。這種欺騙讓他們能夠把一些繳獲的英格瑪機器賣給友好國家的政府，讓英國人能夠截獲他們最敏感的通信內(nèi)容。

According to Mr Singh, even after the war the British authorities kept secret the fact that they had cracked the Enigma code, maintaining the pretence that it was unbreakable. That deceit enabled them to sell some of the captured Enigma machines to friendly governments, allowing the British to spy on their most sensitive communications. 很難想象計算機會設計這樣一個狡猾的策略。說謊是一件復雜的事情，涉及到對意圖的掩飾、對情境和人類心理的理解，以及讓真假兩種現(xiàn)實并存。英國人的背信顯然是一種高度稀少的智力。

It is hard to imagine a computer could ever devise such a devious strategy. Lying is a complicated business involving the masking of intention, an understanding of context and human psychology, and the coexistence of two versions of reality, one true, one false. British perfidy is clearly a highly rarefied form of intelligence. 正如俄國詩人約瑟夫•布羅德斯基(Joseph Brodsky)所寫的那樣，真正的意識歷史始于我們的第一個謊言。計算機是否有朝一日會有意識地欺騙人類，仍然是一個有意思的問題。萬一它這么做了，我們這些愚笨的人類可能陷入極大的麻煩。在2015年的電影《機械姬》(Ex Machina)中，一個類人機器人接受了圖靈測試(Turing test，圖靈設計的旨在判斷機器能否思考的試驗——譯者注)，后果令人不寒而栗。

As the Russian poet Joseph Brodsky once wrote, the real history of consciousness begins with our first lie. It remains an interesting question whether a computer will ever consciously be able to deceive humans. If, and when, it ever does, we dim humans may be in serious trouble. The consequences have been chillingly explored in the 2015 film Ex Machina, in which a humanoid robot takes the Turing test. 馬克斯•泰格馬克(Max Tegmark)在其著作《生命3.0：人工智能時代的人類》(Life 3.0: Being Human in the Age of Artificial Intelligence)一書中，強調(diào)了哲學家對拉丁語中的“睿智”(智能思考能力)和“感知”(感受主觀體驗的能力)這兩個概念所做的區(qū)分。

In his book Life 3.0: Being Human in the Age of Artificial Intelligence 這位麻省理工學院(MIT)的物理學教授認為，在本世紀的某個時候，電子智能可能會超越人類智慧，挑戰(zhàn)我們自稱為地球上最聰明的物種“智人”(Homo sapiens)的主張。

, Max Tegmark highlights the distinction philosophers draw between the Latin concepts of sapience (the ability to think intelligently) and sentience (the ability to appreciate subjective experiences). 但獲得電子意識(這讓它們能夠理解意思)將是一個更為重要的事件——如果真能實現(xiàn)的話。他提出，如果人類想要延長自己的例外主義，那么也許是時候把我們重新標榜為“意識人”(Homo sentiens)了。

At some point this century, the MIT physics professor contends, electronic intelligence may well supersede human intelligence, challenging our claim to be Homo sapiens, the smartest entity on the planet. 計算機將會發(fā)現(xiàn)破解意識密碼要困難得多。

But acquiring electronic consciousness, which enables meaning, will be an even more important event, if it ever happens. If we humans wish to extend our exceptionalism, he suggests, then maybe it is the moment to rebrand ourselves as Homo sentiens.