Gero Miesenboeck reengineers a brain

52,048 views ・ 2010-11-05

TED


请双击下面的英文字幕来播放视频。

翻译人员: Jenny Yang 校对人员: Yi You
00:15
I have a doppelganger.
0
15260
3000
我有个翻版。
00:18
(Laughter)
1
18260
3000
(笑声)
00:21
Dr. Gero is a brilliant
2
21260
3000
Gero 博士是动画片<<七龙珠Z-机器人传奇>>
00:24
but slightly mad scientist
3
24260
2000
里一个非常聪明但有点疯疯颠颠
00:26
in the "Dragonball Z: Android Saga."
4
26260
3000
的科学家。
00:29
If you look very carefully,
5
29260
2000
如果你仔细观察,
00:31
you see that his skull has been replaced
6
31260
3000
你会发现他的脑壳被换成了一个
00:34
with a transparent Plexiglas dome
7
34260
2000
透明的有机玻璃圆顶盖子,
00:36
so that the workings of his brain can be observed
8
36260
3000
你可以透过盖子观察他的大脑活动
00:39
and also controlled with light.
9
39260
3000
而且可以用光线去控制它。
00:42
That's exactly what I do --
10
42260
2000
这就是我的工作——
00:44
optical mind control.
11
44260
2000
光学精神控制。
00:46
(Laughter)
12
46260
2000
(笑声)
00:48
But in contrast to my evil twin
13
48260
2000
但是我跟我那邪恶的孪生兄弟不同,
00:50
who lusts after world domination,
14
50260
3000
他企图控制全世界,
00:53
my motives are not sinister.
15
53260
3000
而我没有险恶的动机。
00:56
I control the brain
16
56260
2000
我控制大脑的原因是
00:58
in order to understand how it works.
17
58260
2000
想了解它是怎样运作的。
01:00
Now wait a minute, you may say,
18
60260
2000
你也许会说,等一下,
01:02
how can you go straight to controlling the brain
19
62260
3000
你不了解大脑
01:05
without understanding it first?
20
65260
2000
怎么能控制大脑呢?
01:07
Isn't that putting the cart before the horse?
21
67260
3000
那不是本末倒置吗?
01:11
Many neuroscientists agree with this view
22
71260
3000
很多神经学科学家都同意这个观点
01:14
and think that understanding will come
23
74260
3000
认为对大脑的了解需要通过
01:17
from more detailed observation and analysis.
24
77260
3000
很多详细的观察和分析来完成。
01:20
They say, "If we could record the activity of our neurons,
25
80260
4000
他们说:“如果我们可以记录我们的神经元的活动
01:24
we would understand the brain."
26
84260
3000
我们就能了解大脑了。”
01:27
But think for a moment what that means.
27
87260
3000
但是想一想这意味着什么。
01:30
Even if we could measure
28
90260
2000
即便我们可以侧量
01:32
what every cell is doing at all times,
29
92260
2000
每个细胞每时每刻的运作,
01:34
we would still have to make sense
30
94260
2000
我们还是需要弄明白
01:36
of the recorded activity patterns,
31
96260
2000
记录下来的那些活动到底有什么规律,
01:38
and that's so difficult,
32
98260
2000
而这是很困难的。
01:40
chances are we'll understand these patterns
33
100260
2000
结果很可能是我们只能了解这些规律的一小部分
01:42
just as little as the brains that produce them.
34
102260
3000
因为我们对产生它们的大脑知之甚少。
01:45
Take a look at what brain activity might look like.
35
105260
3000
请看这个大脑活动模拟图。
01:48
In this simulation, each black dot
36
108260
2000
在这个模拟中,每个黑点
01:50
is one nerve cell.
37
110260
2000
代表一个神经细胞。
01:52
The dot is visible
38
112260
2000
每当细胞放射一条
01:54
whenever a cell fires an electrical impulse.
39
114260
2000
电脉冲时, 对应的点就会显示出来。
01:56
There's 10,000 neurons here.
40
116260
2000
这里有一万个神经细胞。
01:58
So you're looking at roughly one percent
41
118260
2000
所以你现在看到的相当于
02:00
of the brain of a cockroach.
42
120260
3000
一只蟑螂大脑的百分之一。
02:04
Your brains are about 100 million times
43
124260
3000
你们的大脑比这复杂
02:07
more complicated.
44
127260
2000
一亿倍。
02:09
Somewhere, in a pattern like this,
45
129260
2000
在这个图案中, 有一部分
02:11
is you,
46
131260
2000
代表你,
02:13
your perceptions,
47
133260
2000
代表你的感知,
02:15
your emotions, your memories,
48
135260
3000
你的情绪, 记忆,
02:18
your plans for the future.
49
138260
2000
你对将来的计划。
02:20
But we don't know where,
50
140260
2000
但是我们不知道到底哪块地方代表了这些,
02:22
since we don't know how to read the pattern.
51
142260
3000
因为我们不懂如何解读这些图案。
02:25
We don't understand the code used by the brain.
52
145260
3000
我们不了解大脑所用的密码。
02:28
To make progress,
53
148260
2000
要有进展,
02:30
we need to break the code.
54
150260
2000
我们需要破解这个密码。
02:32
But how?
55
152260
2000
但怎样解码呢?
02:35
An experienced code-breaker will tell you
56
155260
2000
一个富有经验的密码破译员会告诉你
02:37
that in order to figure out what the symbols in a code mean,
57
157260
3000
要解开密码中各个符号的真正意义,
02:40
it's essential to be able to play with them,
58
160260
3000
我们必须能够使用它们,
02:43
to rearrange them at will.
59
163260
2000
以我们的意愿重新组合它们。
02:45
So in this situation too,
60
165260
2000
在这个情况下也是如此,
02:47
to decode the information
61
167260
2000
要取得蕴含在这样的图案中
02:49
contained in patterns like this,
62
169260
2000
的信息,
02:51
watching alone won't do.
63
171260
2000
关靠观察是不行的;
02:53
We need to rearrange the pattern.
64
173260
2000
我们需要重新组合这些图案。
02:55
In other words,
65
175260
2000
也就是说,
02:57
instead of recording the activity of neurons,
66
177260
2000
仅仅记录神经细胞的活动并不可取,
02:59
we need to control it.
67
179260
2000
我们需要控制它们。
03:01
It's not essential that we can control
68
181260
2000
我们不用控制
03:03
the activity of all neurons in the brain, just some.
69
183260
3000
脑部中所有神经细胞的活动,一些就足够。
03:06
The more targeted our interventions, the better.
70
186260
2000
我们对那些神经细胞活动的干预愈具定向性,效果愈佳。
03:08
And I'll show you in a moment
71
188260
2000
我稍后会给你们展示
03:10
how we can achieve the necessary precision.
72
190260
3000
我们是怎样达到所需的精确度的。
03:13
And since I'm realistic, rather than grandiose,
73
193260
3000
因为我个人比较现实而不浮夸,
03:16
I don't claim that the ability to control the function of the nervous system
74
196260
3000
我不会宣称只要具备能够操控神经系统功能的能力
03:19
will at once unravel all its mysteries.
75
199260
3000
就会一次揭开所有关于神经细胞的奥秘。
03:22
But we'll certainly learn a lot.
76
202260
3000
但是我们一定会从中学到很多。
03:27
Now, I'm by no means
77
207260
2000
现在,我决不是历史上
03:29
the first person to realize
78
209260
2000
第一个意识到工具的发明是
03:31
how powerful a tool intervention is.
79
211260
3000
多么重要的人。
03:34
The history of attempts
80
214260
2000
试图
03:36
to tinker with the function of the nervous system
81
216260
2000
去改变神经系统机能的历史
03:38
is long and illustrious.
82
218260
2000
长久卓越。
03:40
It dates back at least 200 years,
83
220260
3000
这可以追溯到至少二百年前,
03:43
to Galvani's famous experiments
84
223260
2000
十八世纪初
03:45
in the late 18th century and beyond.
85
225260
3000
伽伐尼的著名实验。
03:49
Galvani showed that a frog's legs twitched
86
229260
3000
伽伐尼证明,只要他把青蛙的腰椎神经
03:52
when he connected the lumbar nerve
87
232260
2000
接上电流
03:54
to a source of electrical current.
88
234260
3000
青蛙的脚就会抽搐。
03:57
This experiment revealed the first, and perhaps most fundamental,
89
237260
3000
这个实验揭示了第一个,也是最根本的一个,
04:00
nugget of the neural code:
90
240260
2000
关于神经编码的事实:
04:02
that information is written in the form
91
242260
2000
这些信息是以
04:04
of electrical impulses.
92
244260
3000
电子脉冲的形式编写的。
04:08
Galvani's approach
93
248260
2000
伽伐尼的
04:10
of probing the nervous system with electrodes
94
250260
2000
以电极探究神经系统的方法
04:12
has remained state-of-the-art until today,
95
252260
3000
即使在今天,仍然是最先进的,
04:15
despite a number of drawbacks.
96
255260
3000
虽然它有一些很明显的缺陷。
04:18
Sticking wires into the brain is obviously rather crude.
97
258260
3000
将电线接到脑部显然是相当残忍的。
04:21
It's hard to do in animals that run around,
98
261260
2000
这样的设计也难于在移动的动物身上实践。
04:23
and there is a physical limit
99
263260
2000
当中亦有物理上的限制
04:25
to the number of wires
100
265260
2000
只有有限数量的电线
04:27
that can be inserted simultaneously.
101
267260
3000
能够同时安插在脑部。
04:30
So around the turn of the last century,
102
270260
2000
因此,在上世纪末,
04:32
I started to think,
103
272260
2000
我开始思考,
04:34
"Wouldn't it be wonderful if one could take this logic
104
274260
3000
如果我可以把这个逻辑
04:37
and turn it upside down?"
105
277260
2000
翻个个儿,那该多好啊。
04:39
So instead of inserting a wire
106
279260
2000
不把电线
04:41
into one spot of the brain,
107
281260
3000
接到大脑的某一点,
04:44
re-engineer the brain itself
108
284260
2000
而是改造大脑本身
04:46
so that some of its neural elements
109
286260
3000
使得某些神经系统的组件
04:49
become responsive to diffusely broadcast signals
110
289260
3000
可以感应到漫射讯号,
04:52
such as a flash of light.
111
292260
3000
比如说一道闪光。
04:55
Such an approach would literally, in a flash of light,
112
295260
3000
这样的方式能够,在电光一闪之间,
04:58
overcome many of the obstacles to discovery.
113
298260
3000
跨越很多科学发现的障碍。
05:01
First, it's clearly a non-invasive,
114
301260
3000
首先,这显然是个无创伤性的,
05:04
wireless form of communication.
115
304260
3000
无线的交流形式。
05:07
And second, just as in a radio broadcast,
116
307260
2000
其次,就像无线电广播一样,
05:09
you can communicate with many receivers at once.
117
309260
3000
你可以跟很多讯号接收器同时沟通。
05:12
You don't need to know where these receivers are,
118
312260
3000
你并不需要知道那些接收器所在的地点。
05:15
and it doesn't matter if these receivers move --
119
315260
2000
再者,讯号接收器移动也没关系——
05:17
just think of the stereo in your car.
120
317260
3000
就好象你车里的收音机。
05:20
It gets even better,
121
320260
3000
我们的情况更加理想,
05:23
for it turns out that we can fabricate the receivers
122
323260
3000
因为我们可以用DNA中所编码的材料
05:26
out of materials that are encoded in DNA.
123
326260
3000
来制造这些接收器。
05:29
So each nerve cell
124
329260
2000
每个具有合适的基因构造
05:31
with the right genetic makeup
125
331260
2000
的神经细胞
05:33
will spontaneously produce a receiver
126
333260
3000
都会自然产生出接收器
05:36
that allows us to control its function.
127
336260
3000
让我们能控制它的功能。
05:39
I hope you'll appreciate
128
339260
2000
我希望你们能够欣赏
05:41
the beautiful simplicity
129
341260
2000
这个概念中
05:43
of this concept.
130
343260
2000
美丽的简单朴素。
05:45
There's no high-tech gizmos here,
131
345260
2000
这里面没有安置高科技的小玩意儿
05:47
just biology revealed through biology.
132
347260
3000
只有用生物学揭示出的生物学。
05:51
Now let's take a look at these miraculous receivers up close.
133
351260
3000
现在,让我们来近距离看一下这些不可思异的接收器。
05:54
As we zoom in on one of these purple neurons,
134
354260
3000
当我们把镜头拉近到一个紫色的神经细胞,
05:57
we see that its outer membrane
135
357260
2000
我们可以看到它的细胞外膜
05:59
is studded with microscopic pores.
136
359260
2000
布满着微细的气孔。
06:01
Pores like these conduct electrical current
137
361260
2000
这些气孔能够让电流通过
06:03
and are responsible
138
363260
2000
以及负责
06:05
for all the communication in the nervous system.
139
365260
2000
为神经系统传递信息。
06:07
But these pores here are special.
140
367260
2000
但是你看到的是一些特别的气孔。
06:09
They are coupled to light receptors
141
369260
2000
它们结合了光感应受体
06:11
similar to the ones in your eyes.
142
371260
3000
这些光感应受体跟你眼睛里的很相似。
06:14
Whenever a flash of light hits the receptor,
143
374260
2000
只要一束闪光射在这些光感应受体上,
06:16
the pore opens, an electrical current is switched on,
144
376260
3000
那些气孔便会打开,电流亦会随之而接通,
06:19
and the neuron fires electrical impulses.
145
379260
3000
那神经细胞便会发出电子脉冲。
06:22
Because the light-activated pore is encoded in DNA,
146
382260
3000
因为那些光驱动气孔是编写在DNA里的,
06:25
we can achieve incredible precision.
147
385260
3000
我们可以达到无比的精确性。
06:28
This is because,
148
388260
2000
这是因为
06:30
although each cell in our bodies
149
390260
2000
尽管我们身体里的每一个细胞
06:32
contains the same set of genes,
150
392260
2000
都有同一组基因,
06:34
different mixes of genes get turned on and off
151
394260
2000
但是在不同的细胞里
06:36
in different cells.
152
396260
2000
不同组合的基因会被开启或关闭。
06:38
You can exploit this to make sure
153
398260
2000
你可以利用这个原理来确保
06:40
that only some neurons
154
400260
2000
只有部分的神经细胞
06:42
contain our light-activated pore and others don't.
155
402260
3000
具有我们设计的光驱动气孔,而其他的没有。
06:45
So in this cartoon, the bluish white cell
156
405260
2000
在这个动画中,蓝白色的细胞
06:47
in the upper-left corner
157
407260
2000
在左上角的那些
06:49
does not respond to light
158
409260
2000
并不会对光有反应
06:51
because it lacks the light-activated pore.
159
411260
3000
因为它们缺少了光驱动气孔。
06:54
The approach works so well
160
414260
2000
这个方法很管用
06:56
that we can write purely artificial messages
161
416260
2000
我们可以把完全人工的讯息
06:58
directly to the brain.
162
418260
2000
直接编写入大脑。
07:00
In this example, each electrical impulse,
163
420260
2000
在这个例子中,每一个电子脉冲,
07:02
each deflection on the trace,
164
422260
3000
轨迹的每一次偏斜,
07:05
is caused by a brief pulse of light.
165
425260
3000
都是由一束短暂的光脉冲造成的。
07:08
And the approach, of course, also works
166
428260
2000
同时,这个方法亦适用
07:10
in moving, behaving animals.
167
430260
3000
于移动中,表现良好的动物身上。
07:13
This is the first ever such experiment,
168
433260
2000
这是同类实验中的第一个,
07:15
sort of the optical equivalent of Galvani's.
169
435260
3000
可以说是光学版本的伽伐尼实验。
07:18
It was done six or seven years ago
170
438260
2000
这是在六七年前
07:20
by my then graduate student, Susana Lima.
171
440260
3000
由我的研究生苏珊娜.利马完成的。
07:23
Susana had engineered the fruit fly on the left
172
443260
3000
苏珊娜改变了左边那只果蝇的基因
07:26
so that just two out of the 200,000 cells in its brain
173
446260
4000
使它大脑中二十万神经细胞中的两个
07:30
expressed the light-activated pore.
174
450260
3000
出现光驱动气孔。
07:33
You're familiar with these cells
175
453260
2000
你应该相当熟悉这些神经细胞
07:35
because they are the ones that frustrate you
176
455260
2000
因为它们就是那些在你尝试拍打果蝇的时候
07:37
when you try to swat the fly.
177
457260
2000
让你倍感沮丧的神经细胞。
07:39
They trained the escape reflex that makes the fly jump into the air
178
459260
3000
它们造就了果蝇的逃跑反射,使果蝇
07:42
and fly away whenever you move your hand in position.
179
462260
3000
在你的手要靠近的时候飞到空中逃走。
07:46
And you can see here that the flash of light has exactly the same effect.
180
466260
3000
你可以看到,一阵闪光跟拍打的动作有着同样的效果。
07:49
The animal jumps, it spreads its wings, it vibrates them,
181
469260
3000
果蝇跳起,伸展翅膀,震动它们,
07:52
but it can't actually take off
182
472260
2000
但它们不能真的飞离地面,
07:54
because the fly is sandwiched between two glass plates.
183
474260
3000
因为这只果蝇被夹在两片玻璃板中间。
07:58
Now to make sure that this was no reaction of the fly
184
478260
2000
要保证果蝇如此反应
08:00
to a flash it could see,
185
480260
3000
并非因为它看到了那道闪光,
08:03
Susana did a simple
186
483260
2000
苏珊娜做了一个简单
08:05
but brutally effective experiment.
187
485260
2000
但直截了当而有效的实验。
08:07
She cut the heads off of her flies.
188
487260
3000
她把果蝇的头切掉。
08:11
These headless bodies can live for about a day,
189
491260
3000
这些无头的身体可以生存约一天。
08:14
but they don't do much.
190
494260
2000
但它们并不会有太多活动。
08:16
They just stand around
191
496260
3000
它们会站着
08:19
and groom excessively.
192
499260
3000
并不停的替自己梳理。
08:22
So it seems that the only trait that survives decapitation is vanity.
193
502260
3000
如此看来,断头之后,好像只有虚荣心这一个人格特征能够保存下来。
08:25
(Laughter)
194
505260
3000
(笑声)
08:30
Anyway, as you'll see in a moment,
195
510260
2000
无论如何,一会后你将会看到,
08:32
Susana was able to turn on the flight motor
196
512260
3000
苏珊娜能够启动果蝇的逃走运动神经
08:35
of what's the equivalent of the spinal cord of these flies
197
515260
3000
这相等于果蝇的脊椎
08:38
and get some of the headless bodies
198
518260
2000
并令一些无头的身体
08:40
to actually take off and fly away.
199
520260
3000
离地飞走。
08:47
They didn't get very far, obviously.
200
527260
2000
很明显,它们飞不了很远。
08:50
Since we took these first steps,
201
530260
2000
自从我们走出了第一步之后,
08:52
the field of optogenetics has exploded.
202
532260
3000
光基因技术这个领域百花齐放。
08:55
And there are now hundreds of labs
203
535260
2000
现在,已经有数以百计的实验室
08:57
using these approaches.
204
537260
2000
在使用这些方法。
08:59
And we've come a long way
205
539260
2000
从伽伐尼及苏珊娜的第一步
09:01
since Galvani's and Susana's first successes
206
541260
3000
令动物抽搐或跳动的成功到现在
09:04
in making animals twitch or jump.
207
544260
2000
我们取得了长足的进步。
09:06
We can now actually interfere with their psychology
208
546260
3000
现在我们能够
09:09
in rather profound ways,
209
549260
2000
彻底地干扰它们的心理。
09:11
as I'll show you in my last example,
210
551260
2000
在我最后的例子中,我将会为你展示,
09:13
which is directed at a familiar question.
211
553260
3000
一个老生常谈的问题。
09:16
Life is a string of choices
212
556260
3000
生命就是一连串的决择
09:19
creating a constant pressure to decide what to do next.
213
559260
3000
这就造成了一个常存的压力,迫使我们决定下一步的行动。
09:23
We cope with this pressure by having brains,
214
563260
3000
我们以大脑来应对这个压力。
09:26
and within our brains, decision-making centers
215
566260
3000
我们的大脑是我们的决策中心
09:29
that I've called here the "Actor."
216
569260
3000
我把它叫做“执行者”。
09:33
The Actor implements a policy that takes into account
217
573260
3000
这个“执行者”执行一个政策
09:36
the state of the environment
218
576260
2000
这政策会考虑到周遭环境的因素
09:38
and the context in which we operate.
219
578260
3000
以及我们生活的背景。
09:41
Our actions change the environment, or context,
220
581260
3000
我们的行为改变环境,或背景,
09:44
and these changes are then fed back into the decision loop.
221
584260
3000
而这些改变会反馈到我们的决策回路中。
09:48
Now to put some neurobiological meat
222
588260
3000
现在要把一些神经生物学的东西
09:51
on this abstract model,
223
591260
2000
添加到这个模型中,
09:53
we constructed a simple one-dimensional world
224
593260
2000
我们建造了一个简单的一维空间
09:55
for our favorite subject, fruit flies.
225
595260
3000
把我们最爱的实验对象,果蝇,放进去。
09:58
Each chamber in these two vertical stacks
226
598260
2000
两架玻璃管中的每一个小室
10:00
contains one fly.
227
600260
2000
都放有一只果蝇。
10:02
The left and the right halves of the chamber
228
602260
3000
小室的左右两边
10:05
are filled with two different odors,
229
605260
2000
弥漫着两种不同的气味,
10:07
and a security camera watches
230
607260
2000
闭路电视会一直观察着
10:09
as the flies pace up and down between them.
231
609260
3000
看着果蝇来回踱步。
10:12
Here's some such CCTV footage.
232
612260
2000
这里有一些闭路电视的片段。
10:14
Whenever a fly reaches the midpoint of the chamber
233
614260
3000
当一只果蝇到达小室的中间
10:17
where the two odor streams meet,
234
617260
2000
两种气味会交汇的地方时,
10:19
it has to make a decision.
235
619260
2000
果蝇必须作出决定。
10:21
It has to decide whether to turn around
236
621260
2000
它要决定是回头
10:23
and stay in the same odor,
237
623260
2000
留在同一种气味中,
10:25
or whether to cross the midline
238
625260
2000
还是跨过中线
10:27
and try something new.
239
627260
2000
去尝试新事物。
10:29
These decisions are clearly a reflection
240
629260
3000
这些决策显然是
10:32
of the Actor's policy.
241
632260
3000
根据“执行者”的政策来执行的。
10:36
Now for an intelligent being like our fly,
242
636260
3000
对我们的果蝇,这样有智慧的生物来说,
10:39
this policy is not written in stone
243
639260
3000
这个政策并非刻在石头上一成不变的,
10:42
but rather changes as the animal learns from experience.
244
642260
3000
它会根据自身的经验而转变。
10:45
We can incorporate such an element
245
645260
2000
我们可以将这样一个
10:47
of adaptive intelligence into our model
246
647260
3000
适应性智能的元素加进我们的模型里
10:50
by assuming that the fly's brain
247
650260
2000
我们要假设果蝇的大脑里
10:52
contains not only an Actor,
248
652260
2000
不只有一个“执行者”,
10:54
but a different group of cells,
249
654260
2000
而是也拥有多组不同的细胞,
10:56
a "Critic," that provides a running commentary
250
656260
3000
包括一个“批评家”,不断地
10:59
on the Actor's choices.
251
659260
2000
给“执行者”的决定提出意见。
11:01
You can think of this nagging inner voice
252
661260
3000
你可以把这个唠叨的内在声音
11:04
as sort of the brain's equivalent
253
664260
2000
当成是大脑里的
11:06
of the Catholic Church,
254
666260
2000
天主教教堂,
11:08
if you're an Austrian like me,
255
668260
3000
如果你跟我一样是奥地利人的话,
11:11
or the super-ego, if you're Freudian,
256
671260
3000
或者可以把它当成佛洛依德学说中所说的“超我”,
11:14
or your mother, if you're Jewish.
257
674260
2000
如果你是犹太人,或者可以把它当成你的“母亲大人”。
11:16
(Laughter)
258
676260
4000
(笑声)
11:20
Now obviously,
259
680260
2000
现在,明显的
11:22
the Critic is a key ingredient
260
682260
3000
这个“批评家”在我们的智力系统中
11:25
in what makes us intelligent.
261
685260
2000
是一个重要的组成部分
11:27
So we set out to identify
262
687260
2000
所以,我们希望可以认出
11:29
the cells in the fly's brain
263
689260
2000
这些在果蝇的大脑中
11:31
that played the role of the Critic.
264
691260
2000
扮演着“批评家”角色的细胞。
11:33
And the logic of our experiment was simple.
265
693260
3000
我们实验的逻辑很简单。
11:36
We thought if we could use our optical remote control
266
696260
3000
我们的假设是:如果我们能够用我们的光学遥控器
11:39
to activate the cells of the Critic,
267
699260
3000
来驱动那些“批评家”细胞,
11:42
we should be able, artificially, to nag the Actor
268
702260
3000
那么我们应该可以,人工地,不断烦扰“执行者”
11:45
into changing its policy.
269
705260
2000
使它改变它的政策。
11:47
In other words,
270
707260
2000
换句话说,
11:49
the fly should learn from mistakes
271
709260
2000
这飞蝇应该能够从错误中学习
11:51
that it thought it had made
272
711260
2000
它们会认为自己犯了错误,
11:53
but, in reality, it had not made.
273
713260
3000
即使它们其实并没有。
11:56
So we bred flies
274
716260
2000
我们培植了一些果蝇
11:58
whose brains were more or less randomly peppered
275
718260
3000
它们的脑部被多多少少地随机安置了
12:01
with cells that were light addressable.
276
721260
2000
一些可以用光驱动的细胞。
12:03
And then we took these flies
277
723260
2000
我们拿出这些果蝇
12:05
and allowed them to make choices.
278
725260
2000
给予它们做决策的机会。
12:07
And whenever they made one of the two choices,
279
727260
2000
每当它们作出两选一的决定,
12:09
chose one odor,
280
729260
2000
选到一种气味时,
12:11
in this case the blue one over the orange one,
281
731260
2000
在这个具体案例中,它选了蓝色而非橙色的那种,
12:13
we switched on the lights.
282
733260
2000
我们就亮灯。
12:15
If the Critic was among the optically activated cells,
283
735260
3000
如果“批评家”在光驱动细胞中,
12:18
the result of this intervention
284
738260
2000
这种干扰的结果会使
12:20
should be a change in policy.
285
740260
3000
果蝇改变它的政策。
12:23
The fly should learn to avoid
286
743260
2000
果蝇就应该学会去避开
12:25
the optically reinforced odor.
287
745260
2000
那种受光学加强的气味。
12:27
Here's what happened in two instances:
288
747260
3000
这里有两个不同案例中的情况。
12:30
We're comparing two strains of flies,
289
750260
3000
我们在比较两个不同品种的果蝇,
12:33
each of them having
290
753260
2000
它们的脑部分别有
12:35
about 100 light-addressable cells in their brains,
291
755260
2000
大概一百个可光驱动的细胞,
12:37
shown here in green on the left and on the right.
292
757260
3000
在这里以绿色来显示。
12:40
What's common among these groups of cells
293
760260
3000
两组细胞的相同之处
12:43
is that they all produce the neurotransmitter dopamine.
294
763260
3000
在于它们都会制造神经递质多巴胺。
12:46
But the identities of the individual
295
766260
2000
但是每个制造多巴胺的神经细胞
12:48
dopamine-producing neurons
296
768260
2000
的身份
12:50
are clearly largely different on the left and on the right.
297
770260
3000
在左右两边显然是不同的。
12:53
Optically activating
298
773260
2000
以光学来驱动
12:55
these hundred or so cells
299
775260
2000
大概一百多个细胞
12:57
into two strains of flies
300
777260
2000
在两个不同品种的果蝇里,
12:59
has dramatically different consequences.
301
779260
2000
会有极其不同的后果。
13:01
If you look first at the behavior
302
781260
2000
如果你先看看
13:03
of the fly on the right,
303
783260
2000
右边果蝇的行为,
13:05
you can see that whenever it reaches the midpoint of the chamber
304
785260
3000
你可以看到,当它到达小室的中心点
13:08
where the two odors meet,
305
788260
2000
两种气味的交汇的地方时,
13:10
it marches straight through, as it did before.
306
790260
3000
它会一直走过去,就像从前一样。
13:13
Its behavior is completely unchanged.
307
793260
2000
它的行为完全没有受光学驱动细胞的影响。
13:15
But the behavior of the fly on the left is very different.
308
795260
3000
但是左边的果蝇,情况就大不相同了
13:18
Whenever it comes up to the midpoint,
309
798260
3000
每当它来到中心点,
13:21
it pauses,
310
801260
2000
它会停下,
13:23
it carefully scans the odor interface
311
803260
2000
谨慎地检查气味的接合点,
13:25
as if it was sniffing out its environment,
312
805260
2000
就好像在打探周遭的环境,
13:27
and then it turns around.
313
807260
2000
然后它会回头。
13:29
This means that the policy that the Actor implements
314
809260
3000
这证明“执行者”实施的政策中
13:32
now includes an instruction to avoid the odor
315
812260
2000
现在包括了躲开那种气味的指令
13:34
that's in the right half of the chamber.
316
814260
3000
该种气味是从小室右边散发过来的。
13:37
This means that the Critic
317
817260
2000
这说明“批评家”
13:39
must have spoken in that animal,
318
819260
2000
一定对果蝇提过意见了,
13:41
and that the Critic must be contained
319
821260
2000
这说明在左边的果蝇中,“批评家”
13:43
among the dopamine-producing neurons on the left,
320
823260
3000
是那些能够制造多巴胺的神经细胞中的一个,
13:46
but not among the dopamine producing neurons on the right.
321
826260
3000
但在右边的果蝇中,“批评家”并不在那些能够制造多巴胺的神经细胞中。
13:49
Through many such experiments,
322
829260
3000
通过很多这样的实验,
13:52
we were able to narrow down
323
832260
2000
我们成功地把
13:54
the identity of the Critic
324
834260
2000
“批评家”的身分
13:56
to just 12 cells.
325
836260
2000
缩小到了十二个细胞。
13:58
These 12 cells, as shown here in green,
326
838260
3000
这十二个细胞,在这里以绿色显示,
14:01
send the output to a brain structure
327
841260
2000
它们对一个
14:03
called the "mushroom body,"
328
843260
2000
叫做“蘑菇体”的脑结构发出讯号,
14:05
which is shown here in gray.
329
845260
2000
在这里以灰色呈现出来。
14:07
We know from our formal model
330
847260
2000
我们从抽象模型中知道
14:09
that the brain structure
331
849260
2000
那个接收“评论员”批评的脑结构
14:11
at the receiving end of the Critic's commentary is the Actor.
332
851260
3000
就是“执行者”。
14:14
So this anatomy suggests
333
854260
2000
这个分析指出
14:16
that the mushroom bodies have something to do
334
856260
3000
蘑菇体和
14:19
with action choice.
335
859260
2000
行动的决策有着一定的关系。
14:21
Based on everything we know about the mushroom bodies,
336
861260
2000
基于我们对蘑菇体的了解,
14:23
this makes perfect sense.
337
863260
2000
这是完全能够想像的。
14:25
In fact, it makes so much sense
338
865260
2000
事实上,这非常合理,
14:27
that we can construct an electronic toy circuit
339
867260
3000
我们能够建构一个电子玩具电路
14:30
that simulates the behavior of the fly.
340
870260
3000
用以模拟果蝇的行为。
14:33
In this electronic toy circuit,
341
873260
3000
这个电子玩具电路中,
14:36
the mushroom body neurons are symbolized
342
876260
2000
蘑菇体神经细胞以
14:38
by the vertical bank of blue LEDs
343
878260
3000
在电路中心的竖排蓝色发光二极管
14:41
in the center of the board.
344
881260
3000
来表示。
14:44
These LED's are wired to sensors
345
884260
2000
这些发光二极管上接有感应器
14:46
that detect the presence of odorous molecules in the air.
346
886260
3000
用以探测空气中的气味分子。
14:50
Each odor activates a different combination of sensors,
347
890260
3000
不同的气味都会激活不同组合的感应器,
14:53
which in turn activates
348
893260
2000
而它们再驱动
14:55
a different odor detector in the mushroom body.
349
895260
3000
蘑菇体中另一个气味检测器。
14:58
So the pilot in the cockpit of the fly,
350
898260
2000
所以在果蝇驾驶舱中的飞行员,
15:00
the Actor,
351
900260
2000
“执行者”,
15:02
can tell which odor is present
352
902260
2000
要知道哪一种气味存在
15:04
simply by looking at which of the blue LEDs lights up.
353
904260
4000
只要看看哪颗蓝色发光二极管亮起来就行了。
15:09
What the Actor does with this information
354
909260
2000
“执行者”得到这个讯息之后的行为
15:11
depends on its policy,
355
911260
2000
取决于它的政策,
15:13
which is stored in the strengths of the connection,
356
913260
2000
这些政策都是根据
15:15
between the odor detectors
357
915260
2000
气味检测器
15:17
and the motors
358
917260
2000
与运动神经之间关联的强度来储存的
15:19
that power the fly's evasive actions.
359
919260
2000
这驱动了果蝇的逃亡行为。
15:22
If the connection is weak, the motors will stay off
360
922260
2000
如果关联性弱,运动神经会保持关上
15:24
and the fly will continue straight on its course.
361
924260
3000
那只果蝇会继续前进。
15:27
If the connection is strong, the motors will turn on
362
927260
3000
如果关联性强,运动神经就会启动
15:30
and the fly will initiate a turn.
363
930260
3000
那只果蝇就会作一个转身。
15:33
Now consider a situation
364
933260
2000
现在试想这样一个情况
15:35
in which the motors stay off,
365
935260
2000
就是当运动神经保持关上时,
15:37
the fly continues on its path
366
937260
3000
那只果蝇继续前行
15:40
and it suffers some painful consequence
367
940260
2000
它就会遭受一些痛苦的后果
15:42
such as getting zapped.
368
942260
2000
例如遭到电击。
15:44
In a situation like this,
369
944260
2000
在这样的情况下,
15:46
we would expect the Critic to speak up
370
946260
2000
我们可以预期“批评家”会发表意见
15:48
and to tell the Actor
371
948260
2000
并告诉“执行者”
15:50
to change its policy.
372
950260
2000
要它改变它的政策。
15:52
We have created such a situation, artificially,
373
952260
3000
我们人工地制造了这样的一个情境
15:55
by turning on the critic with a flash of light.
374
955260
3000
以一束光来启动“批评家”。
15:58
That caused a strengthening of the connections
375
958260
3000
这样就可以加强
16:01
between the currently active odor detector
376
961260
3000
正在起作用的气味检测细胞
16:04
and the motors.
377
964260
2000
与运动神经之间的关联。
16:06
So the next time
378
966260
2000
所以,下一次
16:08
the fly finds itself facing the same odor again,
379
968260
3000
当果蝇再次面对同样的气味时,
16:11
the connection is strong enough to turn on the motors
380
971260
3000
关联性将会有足够的强度去启动运动神经
16:14
and to trigger an evasive maneuver.
381
974260
3000
并引发果蝇进行回避。
16:19
I don't know about you,
382
979260
3000
我不知道你的想法,
16:22
but I find it exhilarating to see
383
982260
3000
但我觉得这样的实验非常令人兴奋
16:25
how vague psychological notions
384
985260
3000
虚无的心理学概念
16:28
evaporate and give rise
385
988260
2000
不见了,却引出了
16:30
to a physical, mechanistic understanding of the mind,
386
990260
3000
对思维在物理学上,机能上的理解,
16:33
even if it's the mind of the fly.
387
993260
3000
虽然这只是果蝇的思维。
16:36
This is one piece of good news.
388
996260
3000
这是一个好消息。
16:39
The other piece of good news,
389
999260
2000
另一个好消息,
16:41
for a scientist at least,
390
1001260
2000
至少对于一个科学家来说,
16:43
is that much remains to be discovered.
391
1003260
3000
就是世界上还有很多尚待发掘的东西。
16:46
In the experiments I told you about,
392
1006260
2000
在我讲诉的这个实验里,
16:48
we have lifted the identity of the Critic,
393
1008260
3000
我们发掘出了“批评家”的真正身分,
16:51
but we still have no idea
394
1011260
2000
但是我们还不知道
16:53
how the Critic does its job.
395
1013260
2000
“批评家”是怎样完成它的工作的。
16:55
Come to think of it, knowing when you're wrong
396
1015260
2000
试想一下,在没有老师,或者你的母亲大人告诉你的情况下
16:57
without a teacher, or your mother, telling you,
397
1017260
3000
要知道你自己犯了错,
17:00
is a very hard problem.
398
1020260
2000
是一件非常不容易的事情。
17:02
There are some ideas in computer science
399
1022260
2000
电脑科学领域
17:04
and in artificial intelligence
400
1024260
2000
以及人工智能领域中的一些想法
17:06
as to how this might be done,
401
1026260
2000
就是试图去解释这件事情是如何发生的,
17:08
but we still haven't solved
402
1028260
2000
但是到现在我们还无法解释
17:10
a single example
403
1030260
2000
任何一个实际例子中
17:12
of how intelligent behavior
404
1032260
3000
智能行为是如何
17:15
springs from the physical interactions
405
1035260
2000
在生命体的物理交互中
17:17
in living matter.
406
1037260
2000
产生的。
17:19
I think we'll get there in the not too distant future.
407
1039260
3000
我想在不久的将来, 我们会找到答案的。
17:22
Thank you.
408
1042260
2000
谢谢。
17:24
(Applause)
409
1044260
4000
(掌声)
关于本网站

这个网站将向你介绍对学习英语有用的YouTube视频。你将看到来自世界各地的一流教师教授的英语课程。双击每个视频页面上显示的英文字幕,即可从那里播放视频。字幕会随着视频的播放而同步滚动。如果你有任何意见或要求,请使用此联系表与我们联系。

https://forms.gle/WvT1wiN1qDtmnspy7