The Turing test: Can a computer pass for a human? - Alex Gendler

譯者: Helen Lin 審譯者: Max Chern

What is consciousness?

什麼是意識？

Can an artificial machine really think?

人造機器真的能思考嗎？

Does the mind just consist of neurons in the brain,

心智只是由腦的神經元所組成

or is there some intangible spark at its core?

或有無形的聰明才智在腦的核心嗎？

For many, these have been vital considerations

對多數人而言， 這一直是未來人工智慧的主要考慮因素。

for the future of artificial intelligence.

But British computer scientist Alan Turing decided to disregard all these questions

但英國電腦科學家 阿蘭•圖靈 決定不理會這諸多問題，

in favor of a much simpler one:

而支持更簡單的一種：

can a computer talk like a human?

電腦能和人一樣交談嗎？

This question led to an idea for measuring aritificial intelligence

這問題後來發展出 一個測量人工智慧的理念，

that would famously come to be known as the Turing test.

也就是眾所周知的「圖靈測試」。

In the 1950 paper, "Computing Machinery and Intelligence,"

在 1950 年的一篇論文《電腦機器與智能》，

Turing proposed the following game.

圖靈提出了以下的測試。

A human judge has a text conversation with unseen players

一位審判員和不被看見的 被測試者做文字對話，

and evaluates their responses.

並評估他們的回答。

To pass the test, a computer must be able to replace one of the players

要通過測試， 電腦必須能取代被測試者之一，

without substantially changing the results.

但不會明顯改變最後結果。

In other words, a computer would be considered intelligent

換言之，此電腦會被認定是有智慧 ──

if its conversation couldn't be easily distinguished from a human's.

若其對話不易與人類的區分時。

Turing predicted that by the year 2000,

圖靈預言在 2000 年之前 ，

machines with 100 megabytes of memory would be able to easily pass his test.

具備 100 MB 記憶體的電腦 將能輕易通過圖靈測試，

But he may have jumped the gun.

但他可能言之過早了。

Even though today's computers have far more memory than that,

即使目前配備更多記憶體的電腦，

few have succeeded,

也很少成功，

and those that have done well

而那些通過測試的

focused more on finding clever ways to fool judges

著重在欺騙審判員的小技巧，

than using overwhelming computing power.

而非運用強大的運算能力。

Though it was never subjected to a real test,

雖然從未接受真正測試，

the first program with some claim to success was called ELIZA.

第一個程式聲明它已成功的是 ELIZA，

With only a fairly short and simple script,

它僅以簡短的程式腳本，

it managed to mislead many people by mimicking a psychologist,

模仿心理學家，成功迷惑許多人，

encouraging them to talk more

鼓勵他們說得更多些，

and reflecting their own questions back at them.

同時也反問他們的問題。

Another early script PARRY took the opposite approach

另一個早期程式腳本 PARRY 採取相反的方式，

by imitating a paranoid schizophrenic

模仿偏狂型精神分裂症病患

who kept steering the conversation back to his own preprogrammed obsessions.

一直把話題轉回它預設的執著事物。

Their success in fooling people highlighted one weakness of the test.

它們成功愚弄人們 突顯出這測驗的弱點，

Humans regularly attribute intelligence to a whole range of things

人們通常把許多和聰明才智無關的事物 視為「智慧」。

that are not actually intelligent.

Nonetheless, annual competitions like the Loebner Prize,

儘管如此，年度競賽 例如 羅布納獎（Loebner Prize）

have made the test more formal

已使測試變得更正式了，

with judges knowing ahead of time

審判員已預先知道

that some of their conversation partners are machines.

其中有些他們的對話者是機器。

But while the quality has improved,

雖然品質已有改善，

many chatbot programmers have used similar strategies to ELIZA and PARRY.

但許多聊天機器人的程式員 是採用類似 ELIZA 和 PARRY 的方法。

1997's winner Catherine

1997 年的獲獎者 Catherine

could carry on amazingly focused and intelligent conversation,

能夠進行驚人地專注且聰明的對話，

but mostly if the judge wanted to talk about Bill Clinton.

但多半只在審判員願意談 有關 比爾•克林頓 的事時。

And the more recent winner Eugene Goostman

最近的獲獎者 尤金•古斯特曼 (Eugene Goostman)

was given the persona of a 13-year-old Ukrainian boy,

是個偽裝 13 歲烏克蘭男孩的機器人，

so judges interpreted its nonsequiturs and awkward grammar

審判員將它不合邏輯及蹩脚的文法

as language and culture barriers.

視為是語言及文化的障礙所致。

Meanwhile, other programs like Cleverbot have taken a different approach

同時，另一程式例如 Cleverbot 則採用不同的方法，

by statistically analyzing huge databases of real conversations

藉著統計分析大量真正對話的資料庫，

to determine the best responses.

以決定最佳回應。

Some also store memories of previous conversations

有些還儲存先前對話的記憶，

in order to improve over time.

以便長期改善。

But while Cleverbot's individual responses can sound incredibly human,

雖然 Cleverbot 的每次回應 聽起來極像人類，

its lack of a consistent personality

但它缺乏前後一致的個性，

and inability to deal with brand new topics

而且無法應付嶄新的話題，

are a dead giveaway.

因而明顯露出馬腳。

Who in Turing's day could have predicted that today's computers

在圖靈時代的人 怎能預料到今日的電腦

would be able to pilot spacecraft,

能駕駛太空船、

perform delicate surgeries,

執行精細手術、

and solve massive equations,

能解開龐大的方程式、

but still struggle with the most basic small talk?

卻仍難以處理最基本的小交談呢？

Human language turns out to be an amazingly complex phenomenon

人類語言 原來就是一種驚人複雜的現象，

that can't be captured by even the largest dictionary.

甚至連最大的字典也無法記錄。

Chatbots can be baffled by simple pauses, like "umm..."

一個簡單的暫停就能難倒聊天機器人， 例如「嗯……」

or questions with no correct answer.

或沒有正確解答的問題。

And a simple conversational sentence,

一個簡單的對話句子，

like, "I took the juice out of the fridge and gave it to him,

例如：「我從冰箱拿出果汁給他，

but forgot to check the date,"

但忘了檢查保存期限。」

requires a wealth of underlying knowledge and intuition to parse.

這需要很多基本理解 及直覺語法分析。

It turns out that simulating a human conversation

事實証明模擬人類對話

takes more than just increasing memory and processing power,

不是只要增加記憶體 及數據處理能力而已，

and as we get closer to Turing's goal,

當離圖靈的目標越近時，

we may have to deal with all those big questions about consciousness after all.

我們終究要去處理 所有有關「意識」的大難題。

翻譯：Helen Lin