how they lived,

如何生活、

how they fought ...

如何打仗...

but how did they think?

但他們如何思考呢？

One natural idea is that the deepest aspects of human thought --

有一個很自然的方法就是， 去探索人類最深層的想法——

our ability to imagine,

我們的想像力、

to be conscious,

自覺力、

to dream --

夢想力、

have always been the same.

是否是一樣的。

Another possibility

另一種可能是，

is that the social transformations that have shaped our culture

去探索造就我們文化的社會變革，

may have also changed the structural columns of human thought.

這些變革也許就是 改變人類想法的主要因素。

We may all have different opinions about this.

對這一點，大家或許有不同的看法。

Actually, it's a long-standing philosophical debate.

實際上，這是一個存在已久的哲學辯論。

But is this question even amenable to science?

究竟這個問題是否可以 經由科學來處理？

Here I'd like to propose

我的建議是

that in the same way we can reconstruct how the ancient Greek cities looked

如同僅藉由一些磚頭， 我們得以重建希臘古都的外貌，

just based on a few bricks,

也可用同樣的方式，

that the writings of a culture are the archaeological records,

藉由一些文化作品、建築歷史、

the fossils, of human thought.

化石，來了解人類的想法。

And in fact,

而實際上，

doing some form of psychological analysis

因為做了一些

of some of the most ancient books of human culture,

人類古老文化書籍的心理分析，

Julian Jaynes came up in the '70s with a very wild and radical hypothesis:

裘利安．傑尼斯在70年代， 發表了一個相當大膽激進的假說：

that only 3,000 years ago,

他說，3000年前的人類，

humans were what today we would call schizophrenics.

是我們現在俗稱的 「精神分裂症患者」。

And he made this claim

他會如此主張的原因是，

based on the fact that the first humans described in these books

在世界各地不同的傳統及地方，

behaved consistently,

這些書籍裡面所描述的人類行為

in different traditions and in different places of the world,

似乎不約而同地都會服從

as if they were hearing and obeying voices

他們認為是從神祗那邊傳來的聲音。

that they perceived as coming from the Gods,

而如今，我們會稱之為 「幻聽」或「幻覺」。

or from the muses ...

隨著時間的洗禮，

what today we would call hallucinations.

他們開始認知到 那些聲音是他們自己創造的，

And only then, as time went on,

他們就是那些內在聲音的主人。

they began to recognize that they were the creators,

有了這樣的認知，他們學會了自省：

the owners of these inner voices.

一種反思自己想法的能力。

And with this, they gained introspection:

所以傑尼斯對「意識」的理論就是，

the ability to think about their own thoughts.

至少現今我們覺察到「意識」、

So Jaynes's theory is that consciousness,

感受到我們自己就是 人生導師的體悟

at least in the way we perceive it today,

是相當近代的文化發展。

where we feel that we are the pilots of our own existence --

這理論相當特別，

is a quite recent cultural development.

但它有一個很明顯的問題就是，

And this theory is quite spectacular,

它是建立在極少又特定的案例上。

but it has an obvious problem

所以問題是，

which is that it's built on just a few and very specific examples.

3000年來人類才建立起 自省能力的這個理論，

So the question is whether the theory

是否可以經得起「量化」 且「客觀」的考驗。

that introspection built up in human history only about 3,000 years ago

至於要如何做的問題， 也是相當簡單明瞭。

can be examined in a quantitative and objective manner.

但我的意思並非，比如， 柏拉圖有一天突然醒來說，

And the problem of how to go about this is quite obvious.

「哈囉！我是柏拉圖，

It's not like Plato woke up one day and then he wrote,

我今天，擁有完整的自省意識了」 那樣的簡單而已。

"Hello, I'm Plato,

（笑聲）

and as of today, I have a fully introspective consciousness."

而這告訴我們，我們要找出 問題的本質為何。

(Laughter)

我們必須找到從來沒有被 談論過的概念。

And this tells us actually what is the essence of the problem.

「自省」這個字，

We need to find the emergence of a concept that's never said.

在這些書本中從未出現過一次。

The word introspection does not appear a single time

所以為了解決這個問題， 我們要建立一個文字的空間。

in the books we want to analyze.

在這個大空間裡， 包含了相當多的字，

So our way to solve this is to build the space of words.

用這種方式，可以量測出

This is a huge space that contains all words

兩個字彼此之間的 關聯性程度。

in such a way that the distance between any two of them

舉個例子，

is indicative of how closely related they are.

你會想，「狗」、「貓」 應該是比較有關聯性的，

So for instance,

但「葡萄柚」和「對數」 就沒甚麼關聯了。

you want the words "dog" and "cat" to be very close together,

而在這個空間裡的任何兩個字， 都必須是可以被量測出來的。

but the words "grapefruit" and "logarithm" to be very far away.

而我們有很多方式 可以建立起這些字的空間架構，

And this has to be true for any two words within the space.

方法一，是只要請教專家就行了，

And there are different ways that we can construct the space of words.

有點類似查字典。

One is just asking the experts,

另一個可行的方法是，

a bit like we do with dictionaries.

當兩個字出現關聯性時， 去追蹤它們的預設狀況，

Another possibility

它們可能會出現在同一句、

is following the simple assumption that when two words are related,

同一段落、

they tend to appear in the same sentences,

或同一文件中，

in the same paragraphs,

多於「偶然」地出現。

in the same documents,

在這個簡單的前提下，

more often than would be expected just by pure chance.

這個單純且帶有運算技巧 的方法必須好用，

And this simple hypothesis,

而這個複雜且高維度的空間，

this simple method,

事後證明，相當有效。

with some computational tricks

向各位介紹一下，它多有效，

that have to do with the fact

我們分析了一些經常用到的字，

that this is a very complex and high-dimensional space,

首先你可以看到，

turns out to be quite effective.

這些詞彙會自動地歸納成 語義相近的相鄰群組，

And just to give you a flavor of how well this works,

所以你可看到，水果跟身體部位，

this is the result we get when we analyze this for some familiar words.

電腦與科學字彙等等。

And you can see first

演算法也可以把我們要 整理的概念分門別類出來。

that words automatically organize into semantic neighborhoods.

舉個例子，

So you get the fruits, the body parts,

你可以看到，科學的字彙 被拆解成兩個子類，

the computer parts, the scientific terms and so on.

分別是太空與物理的詞彙。

The algorithm also identifies that we organize concepts in a hierarchy.

然後你會發現一件好玩的事，

So for instance,

舉個例子，「天文學」這個詞彙，

you can see that the scientific terms break down into two subcategories

它應該擺的位置

of the astronomic and the physics terms.

與它現在的位置

And then there are very fine things.

好像不太搭嘎，

For instance, the word astronomy,

它現在介於真實科學與

which seems a bit bizarre where it is,

天文學之間，偏向科學的位置，

is actually exactly where it should be,

而它自己卻是一個天文學的字彙。

between what it is,

我們可以持續尋找 其它類似的情況。

an actual science,

實際上，如果你盯著這些字一陣子，

and between what it describes,

然後隨機搭配連結一下這些字，

the astronomical terms.

你會覺得好像自己在吟詩。

And we could go on and on with this.

那是因為，在某種程度上，

Actually, if you stare at this for a while,

在這些空間字彙裡漫遊， 就像是在腦海中吟詩一樣。

and you just build random trajectories,

最後，

you will see that it actually feels a bit like doing poetry.

演算法也能辨識出人類的直覺字彙，

And this is because, in a way,

並歸納到內省的相鄰字彙中。

walking in this space is like walking in the mind.

舉個例子，

And the last thing

像是自我、內疚、理由、情緒

is that this algorithm also identifies what are our intuitions,

與內省相關的字彙非常接近，

of which words should lead in the neighborhood of introspection.

但其它的字，

So for instance,

像是，紅色、足球、蠟燭、香蕉

words such as "self," "guilt," "reason," "emotion,"

就差很遠了。

are very close to "introspection,"

所以一旦我們建立起 這樣的詞彙空間，

but other words,

有關於內省的歷史，

such as "red," "football," "candle," "banana,"

有關與任何概念的歷史，

are just very far away.

以前被認為是抽象或是有點模糊的字彙，

And so once we've built the space,

都可以變成紮紮實實

the question of the history of introspection,

可以被量化的科學。

or of the history of any concept

而我們要做的就是， 拿起這些書，

which before could seem abstract and somehow vague,

把它們數位化，

becomes concrete --

然後把這些字，像子彈一樣

becomes amenable to quantitative science.

射到這些字彙空間裡面，

All that we have to do is take the books,

然後我們問電腦， 這些字彙所行經的軌跡

we digitize them,

花了多少的時間 才達到內省概念的字彙中。

and we take this stream of words as a trajectory

有了這些數據，

and project them into the space,

我們就可以分析古希臘傳統中，

and then we ask whether this trajectory spends significant time

有關於內省的歷史，

circling closely to the concept of introspection.

因為有著最完整的文字記錄。

And with this,

所以，我們先把這些書，

we could analyze the history of introspection

按照時間排列，

in the ancient Greek tradition,

然後把這些字

for which we have the best available written record.

投射到字彙空間裡面，

So what we did is we took all the books --

然後我們問電腦，這些字 與內省有多少的相關性，

we just ordered them by time --

再把它們平均起來。

for each book we take the words

然後，我們不斷地問電腦問題，

and we project them to the space,

這些書就會

and then we ask for each word how close it is to introspection,

越來越接近內省的概念。

and we just average that.

而這正是古希臘所發生的事。

And then we ask whether, as time goes on and on,

各位可以看到在 荷馬時代最古老的書籍，

these books get closer, and closer and closer

與內省的相關性只有一點點。

to the concept of introspection.

但約在西元前400年左右，

And this is exactly what happens in the ancient Greek tradition.

快速成長了五倍，

So you can see that for the oldest books in the Homeric tradition,

這些書與內省的概念

there is a small increase with books getting closer to introspection.

越來越接近。

But about four centuries before Christ,

最棒的是，

this starts ramping up very rapidly to an almost five-fold increase

我們可以問電腦，

of books getting closer, and closer and closer

在不同的、獨立的傳統文化中， 是否也有一樣的現象。

to the concept of introspection.

所以，我們用同樣的方法， 分析了傳統猶太基督教的書籍，

And one of the nice things about this

也得到了類似的趨勢。

is that now we can ask

在最古老的舊約聖經中， 你可以看到它緩慢地增加，

whether this is also true in a different, independent tradition.

之後在新約聖經中， 它快速地增長，

So we just ran this same analysis on the Judeo-Christian tradition,

大約西元400年，聖奧古斯丁的《懺悔錄》 內省達到了最高峰。

and we got virtually the same pattern.

這個方法相當重要，

Again, you see a small increase for the oldest books in the Old Testament,

因為聖奧古斯丁已經被多位學者、

and then it increases much more rapidly

心理學家、歷史學家公認為

in the new books of the New Testament.

是內省的創始人之一。

And then we get the peak of introspection

有些人認為他是現代心理學之父。

in "The Confessions of Saint Augustine,"

所以，我們演算法的優點

about four centuries after Christ.

不僅可以量化、

And this was very important,

而且客觀、

because Saint Augustine had been recognized by scholars,

當然速度也相當快——

philologists, historians,

幾秒就可以跑完——

as one of the founders of introspection.

並捕捉到若使用傳統方法 必須費長時間調查

Actually, some believe him to be the father of modern psychology.

才能抓到的一些重點。

So our algorithm,

這也是科學美好的地方，

which has the virtue of being quantitative,

它可以可以解讀、歸納這想法，

of being objective,

然後廣泛應用在許多不同的領域上。

and of course of being extremely fast --

或許，最具挑戰性的問題是，

it just runs in a fraction of a second --

我們用電腦來分析過去的 自我意識發展的方法，

can capture some of the most important conclusions

是不是亦可以告訴我們 自我意識的未來趨向呢？

of this long tradition of investigation.

更精確地說，

And this is in a way one of the beauties of science,

我們現在說的話，

which is that now this idea can be translated

是否可以告訴我們

and generalized to a whole lot of different domains.

接下來的幾天、幾個月或幾年後， 我們的心智會達到什樣的情況。

So in the same way that we asked about the past of human consciousness,

同樣的方式，我們現在很多人 都使用穿戴式偵測器，

maybe the most challenging question we can pose to ourselves

可以偵測我們的心跳、

is whether this can tell us something about the future of our own consciousness.

呼吸、

To put it more precisely,

基因，

whether the words we say today

讓我們可以預防疾病的發生，

can tell us something of where our minds will be in a few days,

我們是否已可以藉由 偵測分析我們所說的話、

in a few months

推的文、郵寄的信、寫的文字，

or a few years from now.

來提前告訴我們，我們的心智 可能要發生問題了？

And in the same way many of us are now wearing sensors

我跟我的兄弟，吉列爾莫．切基，

that detect our heart rate,

扛起了這項任務。

our respiration,

我們紀錄分析了 34 位年輕人的談話。

our genes,

他們過去曾經是罹患 精神分裂症的高風險族群。

on the hopes that this may help us prevent diseases,

我們測量了他們第一天的談話，

we can ask whether monitoring and analyzing the words we speak,

然後問電腦，從他們的話中， 是否可以預測出，

we tweet, we email, we write,

未來三年內，

can tell us ahead of time whether something may go wrong with our minds.

他們會不會精神錯亂。

And with Guillermo Cecchi,

但我們大失所望，

who has been my brother in this adventure,

一次又一次的失敗。

we took on this task.

因為沒有足夠的語義資訊

And we did so by analyzing the recorded speech of 34 young people

來預測未來的心智發展。

who were at a high risk of developing schizophrenia.

它在分辨精神病患及控制組上

And so what we did is, we measured speech at day one,

已經有足夠的能力，

and then we asked whether the properties of the speech could predict,

因為這有點像我們之前 做古文字的分析，

within a window of almost three years,

但沒辦法預測未來 精神錯亂的發病。

the future development of psychosis.

後來我們了解到，

But despite our hopes,

也許最重要的關鍵 不是他們說了甚麼，

we got failure after failure.

而是他們怎麼說。

There was just not enough information in semantics

更精確地說，

to predict the future organization of the mind.

不是他們說的「話」落在哪個 語義相近的群組裡，

It was good enough

而是他們說話的「方式」 是否會在這幾個語義相近的群組裡

to distinguish between a group of schizophrenics and a control group,

快速地跳來跳去。

a bit like we had done for the ancient texts,

所以我們想出了一個

but not to predict the future onset of psychosis.

叫做「語義連貫性」的量測方法，

But then we realized

本質上就是測量談話的持續性

that maybe the most important thing was not so much what they were saying,

是否會落在同一個 語義主題或類別上。

but how they were saying it.

結果顯示，剛剛的 34 位年輕人，

More specifically,

透過這個語義連貫性演算法，

it was not in which semantic neighborhoods the words were,

預測誰會精神錯亂的正確率 達到百分之百。

but how far and fast they jumped

目前臨床上所有測量方式 都無法達到、或接近這個數字。

from one semantic neighborhood to the other one.

在我做這項研究的時候， 清楚地記得一件事，

And so we came up with this measure,

當時我坐在電腦前面，

which we termed semantic coherence,

看到之前我回到布宜諾斯艾利斯的第一個學生 ——保羅，傳了一堆信息給我，

which essentially measures the persistence of speech within one semantic topic,

當時他住在紐約。

within one semantic category.

我發現訊息不太對勁——

And it turned out to be that for this group of 34 people,

雖然我講不出個所以然來， 因為他寫得不清不楚——

the algorithm based on semantic coherence could predict,

但我有一個強烈的直覺， 一定是出事了。

with 100 percent accuracy,

所以，我打電話給保羅，

who developed psychosis and who will not.

沒錯，他當時感覺不太舒服。

And this was something that could not be achieved --

用這樣一個單純的辨認方式，

not even close --

從他的字裡行間，

with all the other existing clinical measures.

我可以隱約感受到他的感覺，

And I remember vividly, while I was working on this,

並在第一時間有效地幫助他。

I was sitting at my computer

今天我要告訴各位的是，

and I saw a bunch of tweets by Polo --

我們已經越來越能理解

Polo had been my first student back in Buenos Aires,

如何把我們共有的直覺， 轉換成演算法。

and at the time he was living in New York.

經由這樣做，

And there was something in this tweets --

未來我們也許可以看到一種 全然不同的心理健康模式，

I could not tell exactly what because nothing was said explicitly --

而且是基於一種客觀、量化的方式

but I got this strong hunch,

來自動分析出我們所寫的字、

this strong intuition, that something was going wrong.

我們所說的話。

So I picked up the phone, and I called Polo,

謝謝。

and in fact he was not feeling well.

（掌聲）

And this simple fact,

that reading in between the lines,

I could sense, through words, his feelings,

was a simple, but very effective way to help.

What I tell you today