If we take a look back at the history

智慧，是什麽？

of how intelligence has been viewed,

如果我們回顧歷史

one seminal example has been

對智慧的定義，

Edsger Dijkstra's famous quote that

有一個基本的例子是，

"the question of whether a machine can think

艾茲赫爾·戴克斯特拉說過的一句話： (註：著名電腦科學家)

is about as interesting

“關於機械是否能思考的問題

as the question of whether a submarine

就有如在問

can swim."

潛水艇是否能游泳

Now, Edsger Dijkstra, when he wrote this,

一樣有意思。”

intended it as a criticism

當艾茲赫爾·戴克斯特拉寫下這句話，

of the early pioneers of computer science,

是在質疑

like Alan Turing.

早期的電腦科學先驅，

However, if you take a look back

譬如艾倫·圖靈。

and think about what have been

然而，如果你回顧

the most empowering innovations

並思考，

that enabled us to build

是什麼重大的創新

artificial machines that swim

使我們能夠製造出

and artificial machines that [fly],

會游泳和會飛的

you find that it was only through understanding

人造機器，

the underlying physical mechanisms

你就會發現，

of swimming and flight

唯有透過了解

that we were able to build these machines.

游泳和飛翔的基本物理機制，

And so, several years ago,

我們才能製造出這些機器。

I undertook a program to try to understand

因此，幾年前，

the fundamental physical mechanisms

我著手進行一個計劃，

underlying intelligence.

試圖去了解什麼是

Let's take a step back.

智慧的基本物理機制。

Let's first begin with a thought experiment.

先讓我們退一步，

Pretend that you're an alien race

先從一個發想實驗開始。

that doesn't know anything about Earth biology

假設你是一個外星人，

or Earth neuroscience or Earth intelligence,

對地球的生物完全不了解，

but you have amazing telescopes

也不了解地球的神經學和生物智慧，

and you're able to watch the Earth,

但你有很棒的望遠鏡，

and you have amazingly long lives,

可以直接看到地球，

so you're able to watch the Earth

而且你有很長很長的壽命，

over millions, even billions of years.

所以你有好幾百萬年甚至好幾十億年的時間

And you observe a really strange effect.

來觀察地球。

You observe that, over the course of the millennia,

你發現一個很怪異的事情。

Earth is continually bombarded with asteroids

你發現，在千禧年這個過程中，

up until a point,

地球不斷地遭到小行星的撞擊，

and that at some point,

直到某一天，

corresponding roughly to our year, 2000 AD,

在某一個時刻，

asteroids that are on

大約就是我們現在的西元兩千年左右，

a collision course with the Earth

小行星原本運行在

that otherwise would have collided

會撞擊到地球的軌道上，

mysteriously get deflected

但是那個軌道

or they detonate before they can hit the Earth.

神奇地偏移了，

Now of course, as earthlings,

或者小行星在撞到地球前爆炸了。

we know the reason would be

當然，身為地球人，

that we're trying to save ourselves.

我們知道那是因為

We're trying to prevent an impact.

我們試著拯救人類，

But if you're an alien race

試著避免撞擊發生。

who doesn't know any of this,

但如果你是外星人，

doesn't have any concept of Earth intelligence,

不知道這些，

you'd be forced to put together

對地球上的智慧沒有任何概念，

a physical theory that explains how,

那麼你只好勉強拼湊出一個

up until a certain point in time,

物理理論來解釋，

asteroids that would demolish the surface of a planet

直到某一個時刻，

mysteriously stop doing that.

應該毀滅地表一切的小行星

And so I claim that this is the same question

神奇地不再發生。

as understanding the physical nature of intelligence.

而我認為這跟要了解

So in this program that I undertook several years ago,

智慧的物理機制是一樣的問題。

I looked at a variety of different threads

因此，在這項我幾年前開始進行的計劃中，

across science, across a variety of disciplines,

我研究各式各樣的想法，

that were pointing, I think,

橫跨科學以及不同領域，

towards a single, underlying mechanism

我認為，

for intelligence.

這些都指向智慧的一個單一

In cosmology, for example,

基本機制。

there have been a variety of different threads of evidence

以宇宙論為例，

that our universe appears to be finely tuned

有各種不同的證據顯示

for the development of intelligence,

我們所在的宇宙是被精心調整到

and, in particular, for the development

適合發展出智慧的，

of universal states

尤其是發展出一個

that maximize the diversity of possible futures.

普遍性的狀態

In game play, for example, in Go --

能使未來的可能性上做最大化。

everyone remembers in 1997

以圍棋為例，

when IBM's Deep Blue beat Garry Kasparov at chess --

大家都記得1997年

fewer people are aware

IBM 的深藍電腦打敗棋王卡斯帕羅夫，

that in the past 10 years or so,

但只有少數人知道

the game of Go,

在過去的十年，

arguably a much more challenging game

圍棋，

because it has a much higher branching factor,

被視為是非常具挑戰性的遊戲，

has also started to succumb

因為它有更多的分歧因素，

to computer game players

同時也開始讓

for the same reason:

電腦玩家臣服，

the best techniques right now for computers playing Go

這些都是同樣的理由：

are techniques that try to maximize future options

現在讓電腦下棋最好的技巧

during game play.

就是將下棋過程可能發生的事件數

Finally, in robotic motion planning,

最大化。

there have been a variety of recent techniques

最後，在機器人的行動規劃中，

that have tried to take advantage

最近的各種技術

of abilities of robots to maximize

都是試圖讓機器人

future freedom of action

在未來能自由行動的可能性

in order to accomplish complex tasks.

做最大化，

And so, taking all of these different threads

以完成某些複雜的任務。

and putting them together,

所以，用這些不同的想法，

I asked, starting several years ago,

把它們拼湊在一起，

is there an underlying mechanism for intelligence

在幾年前我開始問，

that we can factor out

有沒有一個關於智慧的基本機制

of all of these different threads?

是我們可以從這些不同的想法中

Is there a single equation for intelligence?

分解出來的？

And the answer, I believe, is yes. ["F = T ∇ Sτ"]

有沒有一個屬於智慧的方程式？

What you're seeing is probably

我相信答案是，有的。 ["F = T ∇ Sτ"]

the closest equivalent to an E = mc²

你現在看到的

for intelligence that I've seen.

或許是我看過最接近 E = mc²

So what you're seeing here

的屬於智慧的方程式。

is a statement of correspondence

你所看到的

that intelligence is a force, F,

是相對應的詮釋，

that acts so as to maximize future freedom of action.

智慧是一種力量，F

It acts to maximize future freedom of action,

它的作用是最大化行動的自由度。

or keep options open,

它的作用會最大化行動的自由度

with some strength T,

或是一直保有開放的選擇，

with the diversity of possible accessible futures, S,

配合某一強度 T，

up to some future time horizon, tau.

和可能發生的未來多樣性，S

In short, intelligence doesn't like to get trapped.

直到未來的某一個時間點，t。

Intelligence tries to maximize future freedom of action

簡單地說，智慧不喜歡被約束住。

and keep options open.

智慧希望最大化未來行動的自由度，

And so, given this one equation,

保持開放的選項。

it's natural to ask, so what can you do with this?

所以，有了這一個方程式，

How predictive is it?

很自然地就會問，你能用它做甚麼？

Does it predict human-level intelligence?

它的預測能力如何？

Does it predict artificial intelligence?

它能否預測人類的智慧？

So I'm going to show you now a video

它能否預測人工智慧？

that will, I think, demonstrate

現在我要給各位看一段影片，

some of the amazing applications

我認為可以說明

of just this single equation.

一些令人驚訝的應用，

(Video) Narrator: Recent research in cosmology

而且都只來自這一個方程式。

has suggested that universes that produce

(影片) 旁白：宇宙學最近的研究

more disorder, or "entropy," over their lifetimes

推論宇宙會產生愈來愈多的

should tend to have more favorable conditions

失序，或是熵 (entropy)，

for the existence of intelligent beings such as ourselves.

應該更容易擁有有利的環境，

But what if that tentative cosmological connection

讓智慧存在。

between entropy and intelligence

但如果把這個宇宙學待驗證的

hints at a deeper relationship?

亂度和智慧的關係

What if intelligent behavior doesn't just correlate

再進一步加深會怎樣？

with the production of long-term entropy,

如果智慧和長期亂度的增加

but actually emerges directly from it?

不只是有正相關性，

To find out, we developed a software engine

而且是從中發展出來的呢？

called Entropica, designed to maximize

為了解答這問題，我們開發了一個軟體

the production of long-term entropy

叫做 "Entropica"，

of any system that it finds itself in.

可以把任何系統中

Amazingly, Entropica was able to pass

熵的長期成長最大化。

multiple animal intelligence tests, play human games,

令人驚訝的是，Entropica 能夠通過

and even earn money trading stocks,

多項動物智慧測試，玩人類的遊戲，

all without being instructed to do so.

甚至從股票交易中賺到錢，

Here are some examples of Entropica in action.

而且事前完全不用去教導它。

Just like a human standing upright without falling over,

這裡有幾個 Entropica 的實例。

here we see Entropica

像人可以直立站著不會跌倒，

automatically balancing a pole using a cart.

我們可以看到，

This behavior is remarkable in part

Entropica使用一台車來自動平衡桿子。

because we never gave Entropica a goal.

這個表現在某方面很了不起，

It simply decided on its own to balance the pole.

因為我們從來沒有為Entropica設定一個目標。

This balancing ability will have appliactions

由它自己決定要去平衡這個桿子。

for humanoid robotics

這個平衡的能力可以應用在

and human assistive technologies.

機器人上，

Just as some animals can use objects

以及人類行動輔助技術。

in their environments as tools

就像有些動物

to reach into narrow spaces,

會使用週遭的物品當作工具，

here we see that Entropica,

以便能伸及到窄小的地方，

again on its own initiative,

我們可以再次看到 Entropica

was able to move a large disk representing an animal

由它自己決定，

around so as to cause a small disk,

可以移動代表動物的大圓圈，

representing a tool, to reach into a confined space

讓代表工具的小圓圈

holding a third disk

進入一個有第三個圓圈的

and release the third disk from its initially fixed position.

狹小空間，

This tool use ability will have applications

然後把第三個圓圈從裡面擠出來。

for smart manufacturing and agriculture.

這個使用工具的能力可以應用在

In addition, just as some other animals

智慧製造和農業上。

are able to cooperate by pulling opposite ends of a rope

另外，就像其它動物

at the same time to release food,

會同時合力拉下繩索的兩端，

here we see that Entropica is able to accomplish

讓食物掉出來，

a model version of that task.

我們看到 Entropica 可以完成

This cooperative ability has interesting implications

模組化後的同樣任務。

for economic planning and a variety of other fields.

這個合作的能力可以應用在

Entropica is broadly applicable

經濟規劃和其它各樣的領域。

to a variety of domains.

Entropica 可以廣泛的應用在

For example, here we see it successfully

各樣的領域。

playing a game of pong against itself,

例如，我們可以看到它

illustrating its potential for gaming.

成功地和自己玩 "乓" (Pong)，

Here we see Entropica orchestrating

代表它能玩遊戲的潛力。

new connections on a social network

我們看到 Entropica 精心地

where friends are constantly falling out of touch

建立起社群的新連結，

and successfully keeping the network well connected.

當朋友們不時地失去聯繫，

This same network orchestration ability

它會成功地維持這個網絡。

also has applications in health care,

這樣的網絡連結能力

energy, and intelligence.

同樣可以應用在醫療照顧，

Here we see Entropica directing the paths

能源和智慧發展上。

of a fleet of ships,

這裡我們看到 Entropica

successfully discovering and utilizing the Panama Canal

為海洋中的船隊指引路徑，

to globally extend its reach from the Atlantic

成功地發現並使用巴拿馬運河，

to the Pacific.

使它的足跡遍及全球每個角落，從大西洋

By the same token, Entropica

到太平洋。

is broadly applicable to problems

同樣的，Entropica

in autonomous defense, logistics and transportation.

可以廣泛地應用在

Finally, here we see Entropica

自主防衛和物流運輸上。

spontaneously discovering and executing

最後，我們看到 Entropica

a buy-low, sell-high strategy

自己發現並且執行

on a simulated range traded stock,

"低買高賣"的策略，

successfully growing assets under management

在一個區間交易的股票模擬市場中，

exponentially.

成功地將管理資產規模

This risk management ability

指數性成長。

will have broad applications in finance

這樣的風險管理能力

and insurance.

可以應用在財務

Alex Wissner-Gross: So what you've just seen

和保險上。

is that a variety of signature human intelligent

艾力克斯·威斯奈-格羅斯： 以上你們所看到的

cognitive behaviors

是一個代表人類智慧的

such as tool use and walking upright

認知行為能力，

and social cooperation

像是工具的使用、直立行走、

all follow from a single equation,

以及群體合作，

which drives a system

全部都遵行一個方程式，

to maximize its future freedom of action.

這個方程式驅使一個系統

Now, there's a profound irony here.

可以最大化未來行動的自由。

Going back to the beginning

然而，有一個很大的諷刺是，

of the usage of the term robot,

回顧最初

the play "RUR,"

使用”機器人”這個名詞時，

there was always a concept

在舞台劇《羅梭的萬能工人》(R.U.R,) 中，

that if we developed machine intelligence,

一直有一個概念：

there would be a cybernetic revolt.

如果我們發展了人工智慧，

The machines would rise up against us.

機器人將會起義反抗，

One major consequence of this work

對抗我們人類。

is that maybe all of these decades,

我們這個研究主要的結論之一是，

we've had the whole concept of cybernetic revolt

或許在過去這幾十年來，

in reverse.

我們在逆向思考"機器人反抗”

It's not that machines first become intelligent

這個概念。

and then megalomaniacal

並不是機器先變聰明，

and try to take over the world.

然後自大，

It's quite the opposite,

然後才企圖統治全世界，

that the urge to take control

而是應該反過來看，

of all possible futures

想要控制所有未來可能性

is a more fundamental principle

的慾望，

than that of intelligence,

比控制智慧

that general intelligence may in fact emerge

是更加基本的原則，

directly from this sort of control-grabbing,

一般的智慧或許是

rather than vice versa.

直接從操控中產生的，

Another important consequence is goal seeking.

並非反過來。

I'm often asked, how does the ability to seek goals

另一個重要的結論是尋找目標。

follow from this sort of framework?

我經常被問到，尋找目標的能力

And the answer is, the ability to seek goals

是如何從這個架構中產生的？

will follow directly from this

答案是，尋找目標的能力

in the following sense:

會直接來自於

just like you would travel through a tunnel,

以下這個想法：

a bottleneck in your future path space,

就像你行經一個隧道，

in order to achieve many other

一個在你未來道路上的瓶頸，

diverse objectives later on,

是為了到達許多

or just like you would invest

在未來的不同目的地，

in a financial security,

或者，就像你在證券上的

reducing your short-term liquidity

投資，

in order to increase your wealth over the long term,

降低短期的流動性，

goal seeking emerges directly

是為了增加長期的財富，

from a long-term drive

而尋找目標是來自於

to increase future freedom of action.

一個長期的趨動力

Finally, Richard Feynman, famous physicist,

用來增加未來的行動自由。

once wrote that if human civilization were destroyed

最後，知名的物理學家理察費曼曾說，

and you could pass only a single concept

如果人類文明要被毀滅了，

on to our descendants

而你只能留下一個概念

to help them rebuild civilization,

給後世的子孫，

that concept should be

以便協助他們重建文明，

that all matter around us

那麼這個概念應該是：

is made out of tiny elements

所有我們週遭的物質

that attract each other when they're far apart

是是由微小的元素組成，

but repel each other when they're close together.

當它們相隔很遠時會互相吸引，

My equivalent of that statement

但靠近時會互相排斥。

to pass on to descendants

而我同樣要