ABOUT THE SPEAKER
Kwabena Boahen - Bioengineer
Kwabena Boahen wants to understand how brains work -- and to build a computer that works like the brain by reverse-engineering the nervous system. His group at Stanford is developing Neurogrid, a hardware platform that will emulate the cortex’s inner workings.

Why you should listen

Kwabena Boahen is the principal investigator at the Brains in Silicon lab at Stanford. He writes of himself:

Being a scientist at heart, I want to understand how cognition arises from neuronal properties. Being an engineer by training, I am using silicon integrated circuits to emulate the way neurons compute, linking the seemingly disparate fields of electronics and computer science with neurobiology and medicine.

My group's contributions to the field of neuromorphic engineering include a silicon retina that could be used to give the blind sight and a self-organizing chip that emulates the way the developing brain wires itself up. Our work is widely recognized, with over sixty publications, including a cover story in the May 2005 issue of Scientific American.

My current research interest is building a simulation platform that will enable the cortex's inner workings to be modeled in detail. While progress has been made linking neuronal properties to brain rhythms, the task of scaling up these models to link neuronal properties to cognition still remains. Making the supercomputer-performance required affordable is the goal of our Neurogrid project. It is at the vanguard of a profound shift in computing, away from the sequential, step-by-step Von Neumann machine towards a parallel, interconnected architecture more like the brain.

More profile about the speaker
Kwabena Boahen | Speaker | TED.com
TEDGlobal 2007

Kwabena Boahen: A computer that works like the brain

卡貝納‧博罕談論像人腦般運作的電腦

Filmed:
718,375 views

研究員卡貝納‧博罕正在利用矽晶片製作一個模仿人類大腦所擁有的超級運算能力的電腦,因為我們腦中的那些混亂的、豐富的處理機制使大腦成為一個又輕、又小的超級電腦。
- Bioengineer
Kwabena Boahen wants to understand how brains work -- and to build a computer that works like the brain by reverse-engineering the nervous system. His group at Stanford is developing Neurogrid, a hardware platform that will emulate the cortex’s inner workings. Full bio

Double-click the English transcript below to play the video.

00:18
I got my first computer電腦 when I was a teenager青少年 growing生長 up in Accra阿克拉,
0
0
5000
我在阿克拉(迦納首都)的童年時期曾獲得一台電腦
00:23
and it was a really cool device設備.
1
5000
3000
那真的是一台非常酷的機械
00:26
You could play games遊戲 with it. You could program程序 it in BASICBASIC.
2
8000
5000
你可以用來玩遊戲,也可以用BASIC語言來寫程式
00:31
And I was fascinated入迷.
3
13000
2000
我從那時開始對它深深地著迷
00:33
So I went into the library圖書館 to figure數字 out how did this thing work.
4
15000
6000
所以我便去圖書館想找出這東西到底是如何運作的
00:39
I read about how the CPU中央處理器 is constantly經常 shuffling洗牌 data數據 back and forth向前
5
21000
5000
我讀了有關CPU(中央處理器)是如何來回傳送資料
00:44
between之間 the memory記憶, the RAM內存 and the ALUALU,
6
26000
4000
在RAM(隨機存取記憶體)和ALU(算術邏輯單元)之間
00:48
the arithmetic算術 and logic邏輯 unit單元.
7
30000
2000
也就是負責算術和邏輯運算的單元
00:50
And I thought to myself, this CPU中央處理器 really has to work like crazy
8
32000
4000
我就想到,CPU必須拼了命工作
00:54
just to keep all this data數據 moving移動 through通過 the system系統.
9
36000
4000
才能將所有資料傳送到每個系統中
00:58
But nobody沒有人 was really worried擔心 about this.
10
40000
3000
但根本沒人會去想過這個問題
01:01
When computers電腦 were first introduced介紹,
11
43000
2000
當電腦最初被發明時
01:03
they were said to be a million百萬 times faster更快 than neurons神經元.
12
45000
3000
他們聲稱傳輸速度可以比神經元細胞要快一百萬倍
01:06
People were really excited興奮. They thought they would soon不久 outstrip超過
13
48000
5000
大家對此十分興奮
01:11
the capacity容量 of the brain.
14
53000
3000
認為很快就可以開發出超過人腦容量的機械
01:14
This is a quote引用, actually其實, from Alan艾倫 Turing圖靈:
15
56000
3000
這裡有句艾倫‧圖靈所講過的話:
01:17
"In 30 years年份, it will be as easy簡單 to ask a computer電腦 a question
16
59000
4000
"三十年內,問電腦一個問題就會變的
01:21
as to ask a person."
17
63000
2000
跟問人一樣容易。"
01:23
This was in 1946. And now, in 2007, it's still not true真正.
18
65000
7000
當年是1946年,但現在已經2007年了,還沒實現
01:30
And so, the question is, why aren't we really seeing眼看
19
72000
4000
問題在於,為什麼我們看不到
01:34
this kind of power功率 in computers電腦 that we see in the brain?
20
76000
4000
電腦像人腦一樣的潛在力量
01:38
What people didn't realize實現, and I'm just beginning開始 to realize實現 right now,
21
80000
4000
一般人無法理解的,而我才正要開始探索的是
01:42
is that we pay工資 a huge巨大 price價錢 for the speed速度
22
84000
2000
我們為了追求運算速度而付出了許多代價
01:44
that we claim要求 is a big advantage優點 of these computers電腦.
23
86000
4000
而且聲稱那是電腦的一大優勢
01:48
Let's take a look at some numbers數字.
24
90000
2000
讓我們來看看一些例子
01:50
This is Blue藍色 Gene基因, the fastest最快的 computer電腦 in the world世界.
25
92000
4000
這是"藍色基因",世界上最快的電腦
01:54
It's got 120,000 processors處理器; they can basically基本上 process處理
26
96000
5000
這裡總共有十二萬顆處理器
01:59
10 quadrillion萬億 bits of information信息 per second第二.
27
101000
3000
美秒可以處理千兆以上位元運算的能力
02:02
That's 10 to the sixteenth第十六. And they consume消耗 one and a half megawatts兆瓦 of power功率.
28
104000
7000
也就是十的十六次方。 但它們必須消耗一百五十萬瓦特的電力
02:09
So that would be really great, if you could add that
29
111000
3000
那是一個極大的的數目,
02:12
to the production生產 capacity容量 in Tanzania坦桑尼亞.
30
114000
2000
如果你把這個數量加進坦尚尼亞的勞動生產力中
02:14
It would really boost促進 the economy經濟.
31
116000
2000
那將會大大地增進當地的經濟成長
02:16
Just to go back to the States狀態,
32
118000
4000
好吧,主題回到美國
02:20
if you translate翻譯 the amount of power功率 or electricity電力
33
122000
2000
如果你把那些電腦所使用的電力
02:22
this computer電腦 uses使用 to the amount of households in the States狀態,
34
124000
3000
轉換成每戶美國人家庭用電的量
02:25
you get 1,200 households in the U.S.
35
127000
4000
可以供應1200戶的家庭使用
02:29
That's how much power功率 this computer電腦 uses使用.
36
131000
2000
這可以顯示出那些電腦需吃掉多少電
02:31
Now, let's compare比較 this with the brain.
37
133000
3000
現在,讓我們跟大腦做個比較
02:34
This is a picture圖片 of, actually其實 Rory羅裡 Sayres'Sayres“ girlfriend's女友 brain.
38
136000
5000
這張影像,其實是羅瑞‧賽爾的女朋友的大腦
02:39
Rory羅裡 is a graduate畢業 student學生 at Stanford斯坦福.
39
141000
2000
羅瑞是史丹佛大學的一名研究生
02:41
He studies學習 the brain using運用 MRIMRI, and he claims索賠 that
40
143000
4000
他用核磁共振研究大腦
02:45
this is the most beautiful美麗 brain that he has ever scanned掃描.
41
147000
3000
並且說這是他看過最漂亮的大腦
02:48
(Laughter笑聲)
42
150000
2000
(笑)
02:50
So that's true真正 love, right there.
43
152000
3000
多麼動人的真愛啊,就在你眼前。
02:53
Now, how much computation計算 does the brain do?
44
155000
3000
回到正題,人腦到底可以做多少計算?
02:56
I estimate估計 10 to the 16 bits per second第二,
45
158000
2000
我估計是每秒十的十六次方位元
02:58
which哪一個 is actually其實 about very similar類似 to what Blue藍色 Gene基因 does.
46
160000
4000
就跟"藍色基因"做的很相像
03:02
So that's the question. The question is, how much --
47
164000
2000
所以問題來了,問題就是,
03:04
they are doing a similar類似 amount of processing處理, similar類似 amount of data數據 --
48
166000
3000
當他們在進行同樣的運算處理、同量的資料
03:07
the question is how much energy能源 or electricity電力 does the brain use?
49
169000
5000
到底人腦需要花掉多少的能量或是電力?
03:12
And it's actually其實 as much as your laptop筆記本電腦 computer電腦:
50
174000
3000
事實上就跟你的筆記型電腦差不多
03:15
it's just 10 watts.
51
177000
2000
就只有十瓦特
03:17
So what we are doing right now with computers電腦
52
179000
3000
所以我們現在要改進電腦的地方是
03:20
with the energy能源 consumed消費 by 1,200 houses房屋,
53
182000
3000
那些花掉一千兩百戶家庭用電的超級電腦所做的事
03:23
the brain is doing with the energy能源 consumed消費 by your laptop筆記本電腦.
54
185000
5000
人腦只需要消耗像筆電一樣的能量就可完成
03:28
So the question is, how is the brain able能夠 to achieve實現 this kind of efficiency效率?
55
190000
3000
問題是,人腦到底怎樣才可以達到這樣子的效率?
03:31
And let me just summarize總結. So the bottom底部 line:
56
193000
2000
先讓我做個結論,重點是
03:33
the brain processes流程 information信息 using運用 100,000 times less energy能源
57
195000
4000
與我們現有的電腦科技相比,人腦可以用十萬分之一的能量
03:37
than we do right now with this computer電腦 technology技術 that we have.
58
199000
4000
去處理同樣的訊息量
03:41
How is the brain able能夠 to do this?
59
203000
2000
大腦到底是怎麼做到的?
03:43
Let's just take a look about how the brain works作品,
60
205000
3000
我們先來了解一下大腦的運作方式
03:46
and then I'll compare比較 that with how computers電腦 work.
61
208000
4000
然後我再跟電腦的運作方式作比較
03:50
So, this clip is from the PBSPBS series系列, "The Secret秘密 Life of the Brain."
62
212000
4000
這個影片片斷是來自於PBS系列"大腦的秘密"
03:54
It shows節目 you these cells細胞 that process處理 information信息.
63
216000
3000
你可以看到細胞是如何處理資訊
03:57
They are called neurons神經元.
64
219000
1000
他們被稱作神經元
03:58
They send發送 little pulses脈衝 of electricity電力 down their processes流程 to each other,
65
220000
6000
在處理訊息時會放出微弱的電流訊號給彼此
04:04
and where they contact聯繫 each other, those little pulses脈衝
66
226000
2000
當互相接觸時
04:06
of electricity電力 can jump from one neuron神經元 to the other.
67
228000
2000
電流訊號即可從神經元移動到下個神經元
04:08
That process處理 is called a synapse突觸.
68
230000
3000
這些被稱為突觸
04:11
You've got this huge巨大 network網絡 of cells細胞 interacting互動 with each other --
69
233000
2000
你現在了解這龐大的細胞網路是如何與別人互動
04:13
about 100 million百萬 of them,
70
235000
2000
大概一億個細胞
04:15
sending發出 about 10 quadrillion萬億 of these pulses脈衝 around every一切 second第二.
71
237000
4000
每秒送出約一千萬億個脈衝
04:19
And that's basically基本上 what's going on in your brain right now as you're watching觀看 this.
72
241000
6000
這就是你現在看著這支影片時你大腦正在做的事
04:25
How does that compare比較 with the way computers電腦 work?
73
247000
2000
如果電腦運作的方式跟大腦相比呢?
04:27
In the computer電腦, you have all the data數據
74
249000
2000
對於電腦,所有資料都
04:29
going through通過 the central中央 processing處理 unit單元,
75
251000
2000
必須經過中央處理單元(CPU)
04:31
and any piece of data數據 basically基本上 has to go through通過 that bottleneck瓶頸,
76
253000
3000
而且所有的資料基本上都必須通過那瓶頸
04:34
whereas in the brain, what you have is these neurons神經元,
77
256000
4000
但在大腦中,我們擁有的是神經元
04:38
and the data數據 just really flows流動 through通過 a network網絡 of connections連接
78
260000
4000
資訊只需要流過這些神經元連結的網路
04:42
among其中 the neurons神經元. There's no bottleneck瓶頸 here.
79
264000
2000
根本不存在所謂的瓶頸
04:44
It's really a network網絡 in the literal文字 sense of the word.
80
266000
4000
這的的確確是如同字面上所說的"網路"
04:48
The net is doing the work in the brain.
81
270000
4000
這網路就在大腦裡運作著
04:52
If you just look at these two pictures圖片,
82
274000
2000
如果你看著這兩張圖片
04:54
these kind of words pop流行的 into your mind心神.
83
276000
2000
這幾個字眼就會在你腦海中浮現
04:56
This is serial串行 and it's rigid死板 -- it's like cars汽車 on a freeway高速公路,
84
278000
4000
連續、序列且死板的;就像車子在高速公路上
05:00
everything has to happen發生 in lockstep鎖步 --
85
282000
3000
每件事都必須按照先後來處理
05:03
whereas this is parallel平行 and it's fluid流體.
86
285000
2000
但這邊代表的是平行的、流暢的
05:05
Information信息 processing處理 is very dynamic動態 and adaptive自適應.
87
287000
3000
資訊處理過程十分動態並具有適應性
05:08
So I'm not the first to figure數字 this out. This is a quote引用 from Brian布賴恩 Eno伊諾:
88
290000
4000
我並不是第一個提出這個見解的人,布萊恩‧伊諾曾說過:
05:12
"the problem問題 with computers電腦 is that there is not enough足夠 Africa非洲 in them."
89
294000
4000
"電腦討人厭的地方就是它裡面沒有非洲 (指十分死板且毫無生氣可言)"
05:16
(Laughter笑聲)
90
298000
6000
(笑)
05:22
Brian布賴恩 actually其實 said this in 1995.
91
304000
3000
布萊恩在1995年時說了這句話
05:25
And nobody沒有人 was listening then,
92
307000
3000
但在當時根本沒人理他
05:28
but now people are beginning開始 to listen
93
310000
2000
但現在人們開始注意到他講的話了
05:30
because there's a pressing緊迫, technological技術性 problem問題 that we face面對.
94
312000
5000
因為我們遇到了難以解決的技術性問題
05:35
And I'll just take you through通過 that a little bit in the next下一個 few少數 slides幻燈片.
95
317000
5000
我將會在後面幾張投影片跟你稍微介紹
05:40
This is -- it's actually其實 really this remarkable卓越 convergence收斂
96
322000
4000
這事實上,是一個非常令人印象深刻的巧合
05:44
between之間 the devices設備 that we use to compute計算 in computers電腦,
97
326000
5000
對於我們在電腦中用來計算的裝置
05:49
and the devices設備 that our brains大腦 use to compute計算.
98
331000
4000
和大腦中負責計算的部份
05:53
The devices設備 that computers電腦 use are what's called a transistor晶體管.
99
335000
4000
電腦中負責計算的裝置叫電晶體
05:57
This electrode電極 here, called the gate, controls控制 the flow of current當前
100
339000
4000
這裡的電極,稱為閘極,負責控制電流的進出
06:01
from the source資源 to the drain排水 -- these two electrodes電極.
101
343000
3000
從源極到洩極
06:04
And that current當前, electrical電動 current當前,
102
346000
2000
然後電流呢
06:06
is carried攜帶的 by electrons電子, just like in your house and so on.
103
348000
6000
就像家裡用的那些
06:12
And what you have here is, when you actually其實 turn on the gate,
104
354000
5000
當你把閘極打開,這裡發生什麼事呢
06:17
you get an increase增加 in the amount of current當前, and you get a steady穩定 flow of current當前.
105
359000
4000
電流通過的量將會瞬間增加,然後得到一穩定的電流
06:21
And when you turn off the gate, there's no current當前 flowing流動 through通過 the device設備.
106
363000
4000
當關上閘極時,將沒有電流通過
06:25
Your computer電腦 uses使用 this presence存在 of current當前 to represent代表 a one,
107
367000
5000
電腦就是利用電流通過代表一
06:30
and the absence缺席 of current當前 to represent代表 a zero.
108
372000
4000
沒有電流通過時代表零
06:34
Now, what's happening事件 is that as transistors晶體管 are getting得到 smaller and smaller and smaller,
109
376000
6000
現在,如果當電晶體變得越來越小的時候會發生什麼事?
06:40
they no longer behave表現 like this.
110
382000
2000
他們就不再呈現這樣的行為
06:42
In fact事實, they are starting開始 to behave表現 like the device設備 that neurons神經元 use to compute計算,
111
384000
5000
事實上,將變得類似神經元傳送訊息一樣的方法
06:47
which哪一個 is called an ion離子 channel渠道.
112
389000
2000
我們稱之為離子通道
06:49
And this is a little protein蛋白 molecule分子.
113
391000
2000
這是一個小小的蛋白質分子
06:51
I mean, neurons神經元 have thousands數千 of these.
114
393000
4000
意思是,神經元有幾千個這種分子
06:55
And it sits坐鎮 in the membrane of the cell細胞 and it's got a pore in it.
115
397000
4000
它位於細胞膜上並且位於中央有條通道
06:59
And these are individual個人 potassium ions離子
116
401000
3000
在細胞中的鉀離子
07:02
that are flowing流動 through通過 that pore.
117
404000
2000
就可以穿過這條通道
07:04
Now, this pore can open打開 and close.
118
406000
2000
而且這條通道可關可開
07:06
But, when it's open打開, because these ions離子 have to line up
119
408000
5000
但當通道開啟時,離子必須排成一列
07:11
and flow through通過, one at a time, you get a kind of sporadic零星的, not steady穩定 --
120
413000
5000
一次只能通過一個,所以變成零星發生的並非持續穩定的
07:16
it's a sporadic零星的 flow of current當前.
121
418000
3000
呈現的是斷斷續續的電流
07:19
And even when you close the pore -- which哪一個 neurons神經元 can do,
122
421000
3000
而且當你關閉通道的時候,神經元可以這樣做
07:22
they can open打開 and close these pores毛孔 to generate生成 electrical電動 activity活動 --
123
424000
5000
他們可以藉由開關離子通道來產生電流
07:27
even when it's closed關閉, because these ions離子 are so small,
124
429000
3000
當它關閉的時候,因為離子體積很小
07:30
they can actually其實 sneak潛行 through通過, a few少數 can sneak潛行 through通過 at a time.
125
432000
3000
所以他們其實可以偶爾偷偷從通道溜走
07:33
So, what you have is that when the pore is open打開,
126
435000
3000
所以變成當通道開啟時,
07:36
you get some current當前 sometimes有時.
127
438000
2000
你可以得到電流通過
07:38
These are your ones那些, but you've got a few少數 zeros thrown拋出 in.
128
440000
3000
但裡面偶爾會有些"零電流"藏在裡面
07:41
And when it's closed關閉, you have a zero,
129
443000
4000
當通道關閉時,基本上是沒有電流通過的
07:45
but you have a few少數 ones那些 thrown拋出 in.
130
447000
3000
但你偶爾會收到一些電流訊號流過,懂嗎?
07:48
Now, this is starting開始 to happen發生 in transistors晶體管.
131
450000
3000
現在這就是我們希望讓電晶體產生同樣的效果
07:51
And the reason原因 why that's happening事件 is that, right now, in 2007 --
132
453000
5000
原因是直到現在,2007年
07:56
the technology技術 that we are using運用 -- a transistor晶體管 is big enough足夠
133
458000
4000
我們的科技才足以讓電晶體中的通道
08:00
that several一些 electrons電子 can flow through通過 the channel渠道 simultaneously同時, side by side.
134
462000
5000
大到能同時讓電子並列地通過
08:05
In fact事實, there's about 12 electrons電子 can all be flowing流動 this way.
135
467000
4000
事實上,大概可以允許十二個電子同時流過通道
08:09
And that means手段 that a transistor晶體管 corresponds對應
136
471000
2000
那樣代表著一個電晶體
08:11
to about 12 ion離子 channels渠道 in parallel平行.
137
473000
3000
相當於12個平行的離子通道
08:14
Now, in a few少數 years年份 time, by 2015, we will shrink收縮 transistors晶體管 so much.
138
476000
5000
在未來幾年,也許在2015之前,我們打算將電晶體縮到更小
08:19
This is what Intel英特爾 does to keep adding加入 more cores核心 onto the chip芯片.
139
481000
5000
這就是英特爾打算進行的,目的是加更多核心到一個晶片上
08:24
Or your memory記憶 sticks that you have now can carry攜帶 one gigabyte技嘉
140
486000
3000
或是記憶體上,這樣你就可以擁有1GB容量
08:27
of stuff東東 on them -- before, it was 256.
141
489000
2000
記得以前只有256MB呢
08:29
Transistors晶體管 are getting得到 smaller to allow允許 this to happen發生,
142
491000
3000
電晶體體積越來越小導致上述的這些可能成真
08:32
and technology技術 has really benefitted受益 from that.
143
494000
3000
而且也為科技帶來不少利益
08:35
But what's happening事件 now is that in 2015, the transistor晶體管 is going to become成為 so small,
144
497000
5000
現在我們打算做的是,在2015年,電晶體縮小到
08:40
that it corresponds對應 to only one electron電子 at a time
145
502000
3000
相當一次只能讓一個電子
08:43
can flow through通過 that channel渠道,
146
505000
2000
流過它的通道
08:45
and that corresponds對應 to a single ion離子 channel渠道.
147
507000
2000
那就代表一個獨立的離子通道
08:47
And you start開始 having the same相同 kind of traffic交通 jams果醬 that you have in the ion離子 channel渠道.
148
509000
4000
有時候在通道內就會產生了像交通阻塞那樣的情況
08:51
The current當前 will turn on and off at random隨機,
149
513000
3000
電流將不規則地斷斷續續、若有若無
08:54
even when it's supposed應該 to be on.
150
516000
2000
就算它本來應該是開放要讓電流通過的
08:56
And that means手段 your computer電腦 is going to get
151
518000
2000
這樣代表你們的電腦將會收到
08:58
its ones那些 and zeros mixed up, and that's going to crash緊急 your machine.
152
520000
4000
混合著一和零的訊號,這將會讓電腦當機
09:02
So, we are at the stage階段 where we
153
524000
4000
所以這就是我們現階段面臨的問題
09:06
don't really know how to compute計算 with these kinds of devices設備.
154
528000
3000
我們並不知道該如何用這種方法來進行計算
09:09
And the only kind of thing -- the only thing we know right now
155
531000
3000
我們現在唯一知道的是
09:12
that can compute計算 with these kinds of devices設備 are the brain.
156
534000
3000
能用這種機制進行計算的,就只有人腦而已。
09:15
OK, so a computer電腦 picks精選 a specific具體 item項目 of data數據 from memory記憶,
157
537000
4000
好,所以電腦從記憶體中挑取某些資料
09:19
it sends發送 it into the processor處理器 or the ALUALU,
158
541000
3000
送到中央處理器或是算術邏輯單元
09:22
and then it puts看跌期權 the result結果 back into memory記憶.
159
544000
2000
然後再將結果送回記憶體
09:24
That's the red path路徑 that's highlighted突出.
160
546000
2000
這就是圖上用紅線標示的途徑
09:26
The way brains大腦 work, I told you all, you have got all these neurons神經元.
161
548000
4000
大腦運作的方式是,用這些神經元
09:30
And the way they represent代表 information信息 is
162
552000
2000
他們呈現這些訊息的方法是
09:32
they break打破 up that data數據 into little pieces
163
554000
2000
將訊息打散成許多碎片
09:34
that are represented代表 by pulses脈衝 and different不同 neurons神經元.
164
556000
3000
分別以脈衝和不同的神經元負責
09:37
So you have all these pieces of data數據
165
559000
2000
所以這些訊息片斷
09:39
distributed分散式 throughout始終 the network網絡.
166
561000
2000
透過神經的網路分散各地
09:41
And then the way that you process處理 that data數據 to get a result結果
167
563000
3000
這些資料再被經過處理並產生結果的方法
09:44
is that you translate翻譯 this pattern模式 of activity活動 into a new pattern模式 of activity活動,
168
566000
4000
就是轉譯原本的行為模式進入另一種行為模式
09:48
just by it flowing流動 through通過 the network網絡.
169
570000
3000
而且只靠通過這個網路就可達成目的
09:51
So you set up these connections連接
170
573000
2000
所以我們現在畫面上可以看到這些連結
09:53
such這樣 that the input輸入 pattern模式 just flows流動
171
575000
3000
讓輸入的行為模式只要通過
09:56
and generates生成 the output產量 pattern模式.
172
578000
2000
就可以產生輸出的新模式
09:58
What you see here is that there's these redundant connections連接.
173
580000
4000
你在這裡可以看到多重的連結分支
10:02
So if this piece of data數據 or this piece of the data數據 gets得到 clobbered重挫,
174
584000
4000
所以如果這個綠色的片斷或是另個綠色的片斷遺失了
10:06
it doesn't show顯示 up over here, these two pieces can activate啟用 the missing失踪 part部分
175
588000
5000
它沒有在另一端出現,那麼另兩個片斷就可以補完遺失的部份
10:11
with these redundant connections連接.
176
593000
2000
透過這些多重的分支
10:13
So even when you go to these crappy蹩腳的 devices設備
177
595000
2000
就算設備品質本身不良或有瑕疵
10:15
where sometimes有時 you want a one and you get a zero, and it doesn't show顯示 up,
178
597000
3000
意思是你想要"一"卻收到"零"的時候
10:18
there's redundancy冗餘 in the network網絡
179
600000
2000
如果這裡有多重的網路分支
10:20
that can actually其實 recover恢復 the missing失踪 information信息.
180
602000
3000
他們就可以恢復遺失的部份資料
10:23
It makes品牌 the brain inherently本質 robust強大的.
181
605000
3000
這就是大腦天生強健的秘密
10:26
What you have here is a system系統 where you store商店 data數據 locally本地.
182
608000
3000
這邊的系統使用在局部儲存資料的方式
10:29
And it's brittle, because each of these steps腳步 has to be flawless完美無瑕,
183
611000
4000
這種方法很脆弱,因為每一步不能發生任何差錯
10:33
otherwise除此以外 you lose失去 that data數據, whereas in the brain, you have a system系統
184
615000
3000
要不然資料就會遺失,但在大腦中
10:36
that stores商店 data數據 in a distributed分散式 way, and it's robust強大的.
185
618000
4000
資料是以分散式的方式儲存著,十分強韌
10:40
What I want to basically基本上 talk about is my dream夢想,
186
622000
4000
我現在想談論的是有關我的夢想
10:44
which哪一個 is to build建立 a computer電腦 that works作品 like the brain.
187
626000
3000
也就是建造一台能像大腦般運作的電腦
10:47
This is something that we've我們已經 been working加工 on for the last couple一對 of years年份.
188
629000
4000
這是我們過去幾年來一直在持續進行的計劃
10:51
And I'm going to show顯示 you a system系統 that we designed設計
189
633000
3000
我現在要介紹給你看我們所設計的系統
10:54
to model模型 the retina視網膜,
190
636000
3000
用以模擬視網膜
10:57
which哪一個 is a piece of brain that lines the inside of your eyeball眼球.
191
639000
5000
就是在眼球內連接大腦的細胞
11:02
We didn't do this by actually其實 writing寫作 code, like you do in a computer電腦.
192
644000
6000
我們進行這個計劃不用一般編寫程式碼的方式
11:08
In fact事實, the processing處理 that happens發生
193
650000
3000
事實上,那是因為
11:11
in that little piece of brain is very similar類似
194
653000
2000
腦中的處理訊息的過程
11:13
to the kind of processing處理 that computers電腦
195
655000
1000
很像電腦當要
11:14
do when they stream video視頻 over the Internet互聯網.
196
656000
4000
透過網路傳送影片的程序
11:18
They want to compress壓縮 the information信息 --
197
660000
1000
他們必須將大量的資料進行壓縮
11:19
they just want to send發送 the changes變化, what's new in the image圖片, and so on --
198
661000
4000
只針對那些有改變的影像進行傳送
11:23
and that is how your eyeball眼球
199
665000
3000
而這就是如何你的眼睛
11:26
is able能夠 to squeeze all that information信息 down to your optic視神經 nerve神經,
200
668000
3000
具有壓縮所有訊息到你的視神經
11:29
to send發送 to the rest休息 of the brain.
201
671000
2000
並送到大腦的其他部份
11:31
Instead代替 of doing this in software軟件, or doing those kinds of algorithms算法,
202
673000
3000
而非用軟體處理這個模擬,或用演算法進行
11:34
we went and talked to neurobiologists神經生物學家
203
676000
3000
我們和一位神經生物學家進行訪談過
11:37
who have actually其實 reverse相反 engineered工程 that piece of brain that's called the retina視網膜.
204
679000
4000
他曾經對視網膜進行反向工程
11:41
And they figured想通 out all the different不同 cells細胞,
205
683000
2000
分析出所有不同的細胞
11:43
and they figured想通 out the network網絡, and we just took that network網絡
206
685000
3000
和解構出神經網路,我們就參考該網路
11:46
and we used it as the blueprint藍圖 for the design設計 of a silicon chip芯片.
207
688000
4000
做為藍圖並設計了一顆矽晶片
11:50
So now the neurons神經元 are represented代表 by little nodes節點 or circuits電路 on the chip芯片,
208
692000
6000
我們在晶片上用節點和電路來代表神經元
11:56
and the connections連接 among其中 the neurons神經元 are represented代表, actually其實 modeled仿照 by transistors晶體管.
209
698000
5000
並且神經元之間用電晶體來作為連接
12:01
And these transistors晶體管 are behaving行為 essentially實質上
210
703000
2000
當然這些電晶體必須要正常運作
12:03
just like ion離子 channels渠道 behave表現 in the brain.
211
705000
3000
就像大腦中的離子通道一樣
12:06
It will give you the same相同 kind of robust強大的 architecture建築 that I described描述.
212
708000
5000
我等會將會介紹給你我剛描述的那個穩固的架構模式
12:11
Here is actually其實 what our artificial人造 eye looks容貌 like.
213
713000
4000
做出的人工眼睛長的像這樣
12:15
The retina視網膜 chip芯片 that we designed設計 sits坐鎮 behind背後 this lens鏡片 here.
214
717000
5000
我們設計的視網膜晶片設置在這個透鏡後
12:20
And the chip芯片 -- I'm going to show顯示 you a video視頻
215
722000
2000
還有晶片,我將會播放一段影片
12:22
that the silicon retina視網膜 put out of its output產量
216
724000
3000
顯示這個矽製的視網膜輸出的結果
12:25
when it was looking at Kareem卡里姆 Zaghloul扎格盧勒,
217
727000
3000
當它看著卡林姆‧沙酷
12:28
who's誰是 the student學生 who designed設計 this chip芯片.
218
730000
2000
也就是設計了這整個晶片的學生
12:30
Let me explain說明 what you're going to see, OK,
219
732000
2000
讓我解釋你等會將看到什麼,好嗎?
12:32
because it's putting out different不同 kinds of information信息,
220
734000
3000
因為它輸出很多不同的訊息
12:35
it's not as straightforward直截了當 as a camera相機.
221
737000
2000
所以這並不像照相機一樣簡單明瞭
12:37
The retina視網膜 chip芯片 extracts提取物 four different不同 kinds of information信息.
222
739000
3000
這個視網膜晶片可以解析出四種資訊
12:40
It extracts提取物 regions地區 with dark黑暗 contrast對比,
223
742000
3000
它可以解析出較暗的區域
12:43
which哪一個 will show顯示 up on the video視頻 as red.
224
745000
3000
並在影片中用紅色表示
12:46
And it extracts提取物 regions地區 with white白色 or light contrast對比,
225
748000
4000
和解析出白色或較亮的區域
12:50
which哪一個 will show顯示 up on the video視頻 as green綠色.
226
752000
2000
用綠色標記在影片中
12:52
This is Kareem's賈巴爾的 dark黑暗 eyes眼睛
227
754000
2000
這是卡林姆的深褐色眼睛
12:54
and that's the white白色 background背景 that you see here.
228
756000
3000
你這裡可以看到的是白色的部份
12:57
And then it also extracts提取物 movement運動.
229
759000
2000
它同時可以解析出動作
12:59
When Kareem卡里姆 moves移動 his head to the right,
230
761000
2000
當卡林姆將頭往右移
13:01
you will see this blue藍色 activity活動 there;
231
763000
2000
你可以見到藍色字的這裡
13:03
it represents代表 regions地區 where the contrast對比 is increasing增加 in the image圖片,
232
765000
3000
表現出影像中的對比度增加了
13:06
that's where it's going from dark黑暗 to light.
233
768000
3000
從暗轉向亮度漸增
13:09
And you also see this yellow黃色 activity活動,
234
771000
2000
你也可以看到黃色字的這裡
13:11
which哪一個 represents代表 regions地區 where contrast對比 is decreasing減少;
235
773000
4000
代表對比度下降了
13:15
it's going from light to dark黑暗.
236
777000
2000
影像從亮漸漸變暗
13:17
And these four types類型 of information信息 --
237
779000
3000
這四種訊息
13:20
your optic視神經 nerve神經 has about a million百萬 fibers纖維 in it,
238
782000
4000
在你的視神經內有大概一百萬個纖維
13:24
and 900,000 of those fibers纖維
239
786000
3000
而其中九十萬個這種纖維
13:27
send發送 these four types類型 of information信息.
240
789000
2000
會送出上述的這四種訊號
13:29
So we are really duplicating複製 the kind of signals信號 that you have on the optic視神經 nerve神經.
241
791000
4000
所以實際上我們正在仿傚視神經內的這幾種訊號
13:33
What you notice注意 here is that these snapshots快照
242
795000
3000
你現在看到的這幾張從視網膜晶片輸出的影像
13:36
taken採取 from the output產量 of the retina視網膜 chip芯片 are very sparse, right?
243
798000
4000
事實上都非常粗糙稀疏
13:40
It doesn't light up green綠色 everywhere到處 in the background背景,
244
802000
2000
幾乎很少有綠點整片出現在影像中
13:42
only on the edges邊緣, and then in the hair頭髮, and so on.
245
804000
3000
只有少數零星幾個出現在邊緣,諸如此類
13:45
And this is the same相同 thing you see
246
807000
1000
這跟人類要壓縮影片準備傳送是同樣的原理
13:46
when people compress壓縮 video視頻 to send發送: they want to make it very sparse,
247
808000
4000
他們打算將影像弄的非常分散
13:50
because that file文件 is smaller. And this is what the retina視網膜 is doing,
248
812000
3000
因為這樣可以讓檔案容量大幅縮小,而這正是視網膜在進行的事
13:53
and it's doing it just with the circuitry電路, and how this network網絡 of neurons神經元
249
815000
4000
我們用電路裝置來模擬其行為,
13:57
that are interacting互動 in there, which哪一個 we've我們已經 captured捕獲 on the chip芯片.
250
819000
3000
並用晶片補捉神經元網路的行為模式
14:00
But the point that I want to make -- I'll show顯示 you up here.
251
822000
3000
但我想強調的是,
14:03
So this image圖片 here is going to look like these ones那些,
252
825000
3000
最上面這個影像跟其他的並沒有什麼相異之處
14:06
but here I'll show顯示 you that we can reconstruct重建 the image圖片,
253
828000
2000
可是我們能重建這個影像
14:08
so, you know, you can almost幾乎 recognize認識 Kareem卡里姆 in that top最佳 part部分 there.
254
830000
5000
所以,就你所知,你幾乎可以從上面這個圖來辨認卡林姆
14:13
And so, here you go.
255
835000
2000
請看。
14:24
Yes, so that's the idea理念.
256
846000
3000
這就是我們最主要的概念
14:27
When you stand still, you just see the light and dark黑暗 contrasts對比.
257
849000
2000
當人靜止不動的時候,你只能見到黑和白的對比
14:29
But when it's moving移動 back and forth向前,
258
851000
2000
但當人前後移動的時候,
14:31
the retina視網膜 picks精選 up these changes變化.
259
853000
3000
視網膜可以接收到這些改變
14:34
And that's why, you know, when you're sitting坐在 here
260
856000
1000
這也就是為什麼,以你的經驗,當你坐在那
14:35
and something happens發生 in your background背景,
261
857000
2000
有事情在你背後發生時
14:37
you merely僅僅 move移動 your eyes眼睛 to it.
262
859000
2000
你很少會把眼神轉移過去
14:39
There are these cells細胞 that detect檢測 change更改
263
861000
2000
細胞們就可以偵測到那些改變
14:41
and you move移動 your attention注意 to it.
264
863000
2000
然後才讓你去注意到它
14:43
So those are very important重要 for catching somebody
265
865000
2000
所以這對於發現是誰
14:45
who's誰是 trying to sneak潛行 up on you.
266
867000
2000
從背後偷偷走近你是十分重要的
14:47
Let me just end結束 by saying that this is what happens發生
267
869000
3000
讓我說句話作結:
14:50
when you put Africa非洲 in a piano鋼琴, OK.
268
872000
3000
這就是當你把一個"非洲"裝進鋼琴裡的情況
14:53
This is a steel drum here that has been modified改性,
269
875000
3000
這是一個被改裝過的鋼鼓
14:56
and that's what happens發生 when you put Africa非洲 in a piano鋼琴.
270
878000
3000
而這就是把"非洲"放到一架鋼琴裡的情況
14:59
And what I would like us to do is put Africa非洲 in the computer電腦,
271
881000
4000
而我想要做的就是,把"非洲"裝入電腦裡
15:03
and come up with a new kind of computer電腦
272
885000
2000
並且研發出一種全新的電腦
15:05
that will generate生成 thought, imagination想像力, be creative創作的 and things like that.
273
887000
3000
它可以產生想法、幻想、創意等那些東西
15:08
Thank you.
274
890000
2000
謝謝你們。
15:10
(Applause掌聲)
275
892000
2000
(掌聲)
15:12
Chris克里斯 Anderson安德森: Question for you, Kwabena誇貝納.
276
894000
2000
基斯安德森:問你一個問題,卡貝納
15:14
Do you put together一起 in your mind心神 the work you're doing,
277
896000
4000
你覺得你現在作的工作
15:18
the future未來 of Africa非洲, this conference會議 --
278
900000
3000
和非洲的未來、這次的大會
15:21
what connections連接 can we make, if any, between之間 them?
279
903000
3000
在這三點之間,有何關聯?
15:24
Kwabena誇貝納 BoahenBoahen: Yes, like I said at the beginning開始,
280
906000
2000
卡貝納‧博罕:是的,就如我剛在開頭講過的
15:26
I got my first computer電腦 when I was a teenager青少年, growing生長 up in Accra阿克拉.
281
908000
4000
我在阿克拉的童年時期曾獲得一台電腦
15:30
And I had this gut腸道 reaction反應 that this was the wrong錯誤 way to do it.
282
912000
4000
我的直覺告訴我這種方法是錯誤的
15:34
It was very brute畜生 force; it was very inelegant不雅.
283
916000
3000
因為這樣非常不理性且一點也不優雅
15:37
I don't think that I would've會一直 had that reaction反應,
284
919000
2000
我認為我不會對電腦產生興趣
15:39
if I'd grown長大的 up reading all this science科學 fiction小說,
285
921000
3000
如果我從小就讀著科幻小說長大
15:42
hearing聽力 about RDRD2D2, whatever隨你 it was called, and just -- you know,
286
924000
4000
聽著有關星際大戰機器人RD2D2,不管怎麼稱呼
15:46
buying購買 into this hype炒作 about computers電腦.
287
928000
1000
以及認同有關電腦的誇大炒作消息
15:47
I was coming未來 at it from a different不同 perspective透視,
288
929000
2000
我是從另一個不同的視角來接觸電腦的
15:49
where I was bringing使 that different不同 perspective透視
289
931000
2000
我正是帶著這種不同的觀點
15:51
to bear on the problem問題.
290
933000
2000
來解決這些問題
15:53
And I think a lot of people in Africa非洲 have this different不同 perspective透視,
291
935000
3000
並且我相信很多非洲人有這種不同的觀點
15:56
and I think that's going to impact碰撞 technology技術.
292
938000
2000
我認為那將會衝擊現有的科技
15:58
And that's going to impact碰撞 how it's going to evolve發展.
293
940000
2000
並且會衝擊技術演化的方向
16:00
And I think you're going to be able能夠 to see, use that infusion注入,
294
942000
2000
我想你們應該能了解,利用那種新思維
16:02
to come up with new things,
295
944000
2000
來激發出創新的點子
16:04
because you're coming未來 from a different不同 perspective透視.
296
946000
3000
因為你立足於另一種全然不同的觀點
16:07
I think we can contribute有助於. We can dream夢想 like everybody每個人 else其他.
297
949000
4000
我覺得我們也可以產生貢獻,或是像其他人般做夢
16:11
CACA: Thanks謝謝 Kwabena誇貝納, that was really interesting有趣.
298
953000
2000
基斯安德森:謝謝卡貝納,這真是非常有趣
16:13
Thank you.
299
955000
1000
謝謝你們
16:14
(Applause掌聲)
300
956000
2000
(掌聲)
Translated by Pei-Jan Hung
Reviewed by Shelley Krishna R. TSANG

▲Back to top

ABOUT THE SPEAKER
Kwabena Boahen - Bioengineer
Kwabena Boahen wants to understand how brains work -- and to build a computer that works like the brain by reverse-engineering the nervous system. His group at Stanford is developing Neurogrid, a hardware platform that will emulate the cortex’s inner workings.

Why you should listen

Kwabena Boahen is the principal investigator at the Brains in Silicon lab at Stanford. He writes of himself:

Being a scientist at heart, I want to understand how cognition arises from neuronal properties. Being an engineer by training, I am using silicon integrated circuits to emulate the way neurons compute, linking the seemingly disparate fields of electronics and computer science with neurobiology and medicine.

My group's contributions to the field of neuromorphic engineering include a silicon retina that could be used to give the blind sight and a self-organizing chip that emulates the way the developing brain wires itself up. Our work is widely recognized, with over sixty publications, including a cover story in the May 2005 issue of Scientific American.

My current research interest is building a simulation platform that will enable the cortex's inner workings to be modeled in detail. While progress has been made linking neuronal properties to brain rhythms, the task of scaling up these models to link neuronal properties to cognition still remains. Making the supercomputer-performance required affordable is the goal of our Neurogrid project. It is at the vanguard of a profound shift in computing, away from the sequential, step-by-step Von Neumann machine towards a parallel, interconnected architecture more like the brain.

More profile about the speaker
Kwabena Boahen | Speaker | TED.com