ABOUT THE SPEAKER
Henry Markram - Neuroscientist
Henry Markram is director of Blue Brain, a supercomputing project that can model components of the mammalian brain to precise cellular detail -- and simulate their activity in 3D. Soon he'll simulate a whole rat brain in real time.

Why you should listen

In the microscopic, yet-uncharted circuitry of the cortex, Henry Markram is perhaps the most ambitious -- and our most promising -- frontiersman. Backed by the extraordinary power of the IBM Blue Gene supercomputing architecture, which can perform hundreds of trillions of calculations per second, he's using complex models to precisely simulate the neocortical column (and its tens of millions of neural connections) in 3D.

Though the aim of Blue Brain research is mainly biomedical, it has been edging up on some deep, contentious philosophical questions about the mind -- "Can a robot think?" and "Can consciousness be reduced to mechanical components?" -- the consequence of which Markram is well aware: Asked by Seed Magazine what a simulation of a full brain might do, he answered, "Everything. I mean everything" -- with a grin.

Now, with a successful proof-of-concept for simulation in hand (the project's first phase was completed in 2007), Markram is looking toward a future where brains might be modeled even down to the molecular and genetic level. Computing power marching rightward and up along the graph of Moore's Law, Markram is sure to be at the forefront as answers to the mysteries of cognition emerge.

More profile about the speaker
Henry Markram | Speaker | TED.com
TEDGlobal 2009

Henry Markram: A brain in a supercomputer

Filmed:
1,469,354 views

Henry Markram says the mysteries of the mind can be solved -- soon. Mental illness, memory, perception: they're made of neurons and electric signals, and he plans to find them with a supercomputer that models all the brain's 100,000,000,000,000 synapses.
- Neuroscientist
Henry Markram is director of Blue Brain, a supercomputing project that can model components of the mammalian brain to precise cellular detail -- and simulate their activity in 3D. Soon he'll simulate a whole rat brain in real time. Full bio

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

00:18
Our mission is to build
0
0
3000
00:21
a detailed, realistic
1
3000
2000
00:23
computer model of the human brain.
2
5000
2000
00:25
And we've done, in the past four years,
3
7000
3000
00:28
a proof of concept
4
10000
2000
00:30
on a small part of the rodent brain,
5
12000
3000
00:33
and with this proof of concept we are now scaling the project up
6
15000
3000
00:36
to reach the human brain.
7
18000
3000
00:39
Why are we doing this?
8
21000
2000
00:41
There are three important reasons.
9
23000
2000
00:43
The first is, it's essential for us to understand the human brain
10
25000
4000
00:47
if we do want to get along in society,
11
29000
2000
00:49
and I think that it is a key step in evolution.
12
31000
4000
00:53
The second reason is,
13
35000
2000
00:55
we cannot keep doing animal experimentation forever,
14
37000
6000
01:01
and we have to embody all our data and all our knowledge
15
43000
4000
01:05
into a working model.
16
47000
3000
01:08
It's like a Noah's Ark. It's like an archive.
17
50000
4000
01:12
And the third reason is that there are two billion people on the planet
18
54000
3000
01:15
that are affected by mental disorder,
19
57000
4000
01:19
and the drugs that are used today
20
61000
2000
01:21
are largely empirical.
21
63000
2000
01:23
I think that we can come up with very concrete solutions on
22
65000
3000
01:26
how to treat disorders.
23
68000
3000
01:29
Now, even at this stage,
24
71000
3000
01:32
we can use the brain model
25
74000
2000
01:34
to explore some fundamental questions
26
76000
3000
01:37
about how the brain works.
27
79000
2000
01:39
And here, at TED, for the first time,
28
81000
2000
01:41
I'd like to share with you how we're addressing
29
83000
2000
01:43
one theory -- there are many theories --
30
85000
3000
01:46
one theory of how the brain works.
31
88000
4000
01:50
So, this theory is that the brain
32
92000
4000
01:54
creates, builds, a version of the universe,
33
96000
6000
02:00
and projects this version of the universe,
34
102000
3000
02:03
like a bubble, all around us.
35
105000
4000
02:07
Now, this is of course a topic of philosophical debate for centuries.
36
109000
4000
02:11
But, for the first time, we can actually address this,
37
113000
3000
02:14
with brain simulation,
38
116000
2000
02:16
and ask very systematic and rigorous questions,
39
118000
4000
02:20
whether this theory could possibly be true.
40
122000
4000
02:24
The reason why the moon is huge on the horizon
41
126000
3000
02:27
is simply because our perceptual bubble
42
129000
3000
02:30
does not stretch out 380,000 kilometers.
43
132000
4000
02:34
It runs out of space.
44
136000
2000
02:36
And so what we do is we compare the buildings
45
138000
4000
02:40
within our perceptual bubble,
46
142000
2000
02:42
and we make a decision.
47
144000
2000
02:44
We make a decision it's that big,
48
146000
2000
02:46
even though it's not that big.
49
148000
2000
02:48
And what that illustrates
50
150000
2000
02:50
is that decisions are the key things
51
152000
2000
02:52
that support our perceptual bubble. It keeps it alive.
52
154000
5000
02:57
Without decisions you cannot see, you cannot think,
53
159000
2000
02:59
you cannot feel.
54
161000
2000
03:01
And you may think that anesthetics work
55
163000
2000
03:03
by sending you into some deep sleep,
56
165000
3000
03:06
or by blocking your receptors so that you don't feel pain,
57
168000
3000
03:09
but in fact most anesthetics don't work that way.
58
171000
3000
03:12
What they do is they introduce a noise
59
174000
3000
03:15
into the brain so that the neurons cannot understand each other.
60
177000
3000
03:18
They are confused,
61
180000
2000
03:20
and you cannot make a decision.
62
182000
3000
03:23
So, while you're trying to make up your mind
63
185000
3000
03:26
what the doctor, the surgeon, is doing
64
188000
2000
03:28
while he's hacking away at your body, he's long gone.
65
190000
2000
03:30
He's at home having tea.
66
192000
2000
03:32
(Laughter)
67
194000
2000
03:34
So, when you walk up to a door and you open it,
68
196000
3000
03:37
what you compulsively have to do to perceive
69
199000
3000
03:40
is to make decisions,
70
202000
2000
03:42
thousands of decisions about the size of the room,
71
204000
3000
03:45
the walls, the height, the objects in this room.
72
207000
3000
03:48
99 percent of what you see
73
210000
3000
03:51
is not what comes in through the eyes.
74
213000
4000
03:55
It is what you infer about that room.
75
217000
4000
03:59
So I can say, with some certainty,
76
221000
4000
04:03
"I think, therefore I am."
77
225000
3000
04:06
But I cannot say, "You think, therefore you are,"
78
228000
4000
04:10
because "you" are within my perceptual bubble.
79
232000
5000
04:15
Now, we can speculate and philosophize this,
80
237000
3000
04:18
but we don't actually have to for the next hundred years.
81
240000
3000
04:21
We can ask a very concrete question.
82
243000
2000
04:23
"Can the brain build such a perception?"
83
245000
4000
04:27
Is it capable of doing it?
84
249000
2000
04:29
Does it have the substance to do it?
85
251000
2000
04:31
And that's what I'm going to describe to you today.
86
253000
3000
04:34
So, it took the universe 11 billion years to build the brain.
87
256000
4000
04:38
It had to improve it a little bit.
88
260000
2000
04:40
It had to add to the frontal part, so that you would have instincts,
89
262000
3000
04:43
because they had to cope on land.
90
265000
3000
04:46
But the real big step was the neocortex.
91
268000
4000
04:50
It's a new brain. You needed it.
92
272000
2000
04:52
The mammals needed it
93
274000
2000
04:54
because they had to cope with parenthood,
94
276000
4000
04:58
social interactions,
95
280000
2000
05:00
complex cognitive functions.
96
282000
3000
05:03
So, you can think of the neocortex
97
285000
2000
05:05
actually as the ultimate solution today,
98
287000
5000
05:10
of the universe as we know it.
99
292000
3000
05:13
It's the pinnacle, it's the final product
100
295000
2000
05:15
that the universe has produced.
101
297000
4000
05:19
It was so successful in evolution
102
301000
2000
05:21
that from mouse to man it expanded
103
303000
2000
05:23
about a thousandfold in terms of the numbers of neurons,
104
305000
3000
05:26
to produce this almost frightening
105
308000
3000
05:29
organ, structure.
106
311000
3000
05:32
And it has not stopped its evolutionary path.
107
314000
3000
05:35
In fact, the neocortex in the human brain
108
317000
2000
05:37
is evolving at an enormous speed.
109
319000
3000
05:40
If you zoom into the surface of the neocortex,
110
322000
2000
05:42
you discover that it's made up of little modules,
111
324000
3000
05:45
G5 processors, like in a computer.
112
327000
2000
05:47
But there are about a million of them.
113
329000
3000
05:50
They were so successful in evolution
114
332000
2000
05:52
that what we did was to duplicate them
115
334000
2000
05:54
over and over and add more and more of them to the brain
116
336000
2000
05:56
until we ran out of space in the skull.
117
338000
3000
05:59
And the brain started to fold in on itself,
118
341000
2000
06:01
and that's why the neocortex is so highly convoluted.
119
343000
3000
06:04
We're just packing in columns,
120
346000
2000
06:06
so that we'd have more neocortical columns
121
348000
3000
06:09
to perform more complex functions.
122
351000
3000
06:12
So you can think of the neocortex actually as
123
354000
2000
06:14
a massive grand piano,
124
356000
2000
06:16
a million-key grand piano.
125
358000
3000
06:19
Each of these neocortical columns
126
361000
2000
06:21
would produce a note.
127
363000
2000
06:23
You stimulate it; it produces a symphony.
128
365000
3000
06:26
But it's not just a symphony of perception.
129
368000
3000
06:29
It's a symphony of your universe, your reality.
130
371000
3000
06:32
Now, of course it takes years to learn how
131
374000
3000
06:35
to master a grand piano with a million keys.
132
377000
3000
06:38
That's why you have to send your kids to good schools,
133
380000
2000
06:40
hopefully eventually to Oxford.
134
382000
2000
06:42
But it's not only education.
135
384000
3000
06:45
It's also genetics.
136
387000
2000
06:47
You may be born lucky,
137
389000
2000
06:49
where you know how to master your neocortical column,
138
391000
4000
06:53
and you can play a fantastic symphony.
139
395000
2000
06:55
In fact, there is a new theory of autism
140
397000
3000
06:58
called the "intense world" theory,
141
400000
2000
07:00
which suggests that the neocortical columns are super-columns.
142
402000
4000
07:04
They are highly reactive, and they are super-plastic,
143
406000
4000
07:08
and so the autists are probably capable of
144
410000
3000
07:11
building and learning a symphony
145
413000
2000
07:13
which is unthinkable for us.
146
415000
2000
07:15
But you can also understand
147
417000
2000
07:17
that if you have a disease
148
419000
2000
07:19
within one of these columns,
149
421000
2000
07:21
the note is going to be off.
150
423000
2000
07:23
The perception, the symphony that you create
151
425000
2000
07:25
is going to be corrupted,
152
427000
2000
07:27
and you will have symptoms of disease.
153
429000
3000
07:30
So, the Holy Grail for neuroscience
154
432000
4000
07:34
is really to understand the design of the neocoritical column --
155
436000
4000
07:38
and it's not just for neuroscience;
156
440000
2000
07:40
it's perhaps to understand perception, to understand reality,
157
442000
3000
07:43
and perhaps to even also understand physical reality.
158
445000
4000
07:47
So, what we did was, for the past 15 years,
159
449000
3000
07:50
was to dissect out the neocortex, systematically.
160
452000
4000
07:54
It's a bit like going and cataloging a piece of the rainforest.
161
456000
4000
07:58
How many trees does it have?
162
460000
2000
08:00
What shapes are the trees?
163
462000
2000
08:02
How many of each type of tree do you have? Where are they positioned?
164
464000
3000
08:05
But it's a bit more than cataloging because you actually have to
165
467000
2000
08:07
describe and discover all the rules of communication,
166
469000
4000
08:11
the rules of connectivity,
167
473000
2000
08:13
because the neurons don't just like to connect with any neuron.
168
475000
3000
08:16
They choose very carefully who they connect with.
169
478000
3000
08:19
It's also more than cataloging
170
481000
3000
08:22
because you actually have to build three-dimensional
171
484000
2000
08:24
digital models of them.
172
486000
2000
08:26
And we did that for tens of thousands of neurons,
173
488000
2000
08:28
built digital models of all the different types
174
490000
3000
08:31
of neurons we came across.
175
493000
2000
08:33
And once you have that, you can actually
176
495000
2000
08:35
begin to build the neocortical column.
177
497000
4000
08:39
And here we're coiling them up.
178
501000
3000
08:42
But as you do this, what you see
179
504000
3000
08:45
is that the branches intersect
180
507000
2000
08:47
actually in millions of locations,
181
509000
3000
08:50
and at each of these intersections
182
512000
3000
08:53
they can form a synapse.
183
515000
2000
08:55
And a synapse is a chemical location
184
517000
2000
08:57
where they communicate with each other.
185
519000
3000
09:00
And these synapses together
186
522000
2000
09:02
form the network
187
524000
2000
09:04
or the circuit of the brain.
188
526000
3000
09:07
Now, the circuit, you could also think of as
189
529000
4000
09:11
the fabric of the brain.
190
533000
2000
09:13
And when you think of the fabric of the brain,
191
535000
3000
09:16
the structure, how is it built? What is the pattern of the carpet?
192
538000
4000
09:20
You realize that this poses
193
542000
2000
09:22
a fundamental challenge to any theory of the brain,
194
544000
4000
09:26
and especially to a theory that says
195
548000
2000
09:28
that there is some reality that emerges
196
550000
2000
09:30
out of this carpet, out of this particular carpet
197
552000
3000
09:33
with a particular pattern.
198
555000
2000
09:35
The reason is because the most important design secret of the brain
199
557000
3000
09:38
is diversity.
200
560000
2000
09:40
Every neuron is different.
201
562000
2000
09:42
It's the same in the forest. Every pine tree is different.
202
564000
2000
09:44
You may have many different types of trees,
203
566000
2000
09:46
but every pine tree is different. And in the brain it's the same.
204
568000
3000
09:49
So there is no neuron in my brain that is the same as another,
205
571000
3000
09:52
and there is no neuron in my brain that is the same as in yours.
206
574000
3000
09:55
And your neurons are not going to be oriented and positioned
207
577000
3000
09:58
in exactly the same way.
208
580000
2000
10:00
And you may have more or less neurons.
209
582000
2000
10:02
So it's very unlikely
210
584000
2000
10:04
that you got the same fabric, the same circuitry.
211
586000
4000
10:08
So, how could we possibly create a reality
212
590000
2000
10:10
that we can even understand each other?
213
592000
3000
10:13
Well, we don't have to speculate.
214
595000
2000
10:15
We can look at all 10 million synapses now.
215
597000
3000
10:18
We can look at the fabric. And we can change neurons.
216
600000
3000
10:21
We can use different neurons with different variations.
217
603000
2000
10:23
We can position them in different places,
218
605000
2000
10:25
orient them in different places.
219
607000
2000
10:27
We can use less or more of them.
220
609000
2000
10:29
And when we do that
221
611000
2000
10:31
what we discovered is that the circuitry does change.
222
613000
3000
10:34
But the pattern of how the circuitry is designed does not.
223
616000
7000
10:41
So, the fabric of the brain,
224
623000
2000
10:43
even though your brain may be smaller, bigger,
225
625000
2000
10:45
it may have different types of neurons,
226
627000
3000
10:48
different morphologies of neurons,
227
630000
2000
10:50
we actually do share
228
632000
3000
10:53
the same fabric.
229
635000
2000
10:55
And we think this is species-specific,
230
637000
2000
10:57
which means that that could explain
231
639000
2000
10:59
why we can't communicate across species.
232
641000
2000
11:01
So, let's switch it on. But to do it, what you have to do
233
643000
3000
11:04
is you have to make this come alive.
234
646000
2000
11:06
We make it come alive
235
648000
2000
11:08
with equations, a lot of mathematics.
236
650000
2000
11:10
And, in fact, the equations that make neurons into electrical generators
237
652000
4000
11:14
were discovered by two Cambridge Nobel Laureates.
238
656000
3000
11:17
So, we have the mathematics to make neurons come alive.
239
659000
3000
11:20
We also have the mathematics to describe
240
662000
2000
11:22
how neurons collect information,
241
664000
3000
11:25
and how they create a little lightning bolt
242
667000
3000
11:28
to communicate with each other.
243
670000
2000
11:30
And when they get to the synapse,
244
672000
2000
11:32
what they do is they effectively,
245
674000
2000
11:34
literally, shock the synapse.
246
676000
3000
11:37
It's like electrical shock
247
679000
2000
11:39
that releases the chemicals from these synapses.
248
681000
3000
11:42
And we've got the mathematics to describe this process.
249
684000
3000
11:45
So we can describe the communication between the neurons.
250
687000
4000
11:49
There literally are only a handful
251
691000
3000
11:52
of equations that you need to simulate
252
694000
2000
11:54
the activity of the neocortex.
253
696000
2000
11:56
But what you do need is a very big computer.
254
698000
3000
11:59
And in fact you need one laptop
255
701000
2000
12:01
to do all the calculations just for one neuron.
256
703000
3000
12:04
So you need 10,000 laptops.
257
706000
2000
12:06
So where do you go? You go to IBM,
258
708000
2000
12:08
and you get a supercomputer, because they know how to take
259
710000
2000
12:10
10,000 laptops and put it into the size of a refrigerator.
260
712000
4000
12:14
So now we have this Blue Gene supercomputer.
261
716000
3000
12:17
We can load up all the neurons,
262
719000
2000
12:19
each one on to its processor,
263
721000
2000
12:21
and fire it up, and see what happens.
264
723000
4000
12:25
Take the magic carpet for a ride.
265
727000
3000
12:28
Here we activate it. And this gives the first glimpse
266
730000
3000
12:31
of what is happening in your brain
267
733000
2000
12:33
when there is a stimulation.
268
735000
2000
12:35
It's the first view.
269
737000
2000
12:37
Now, when you look at that the first time, you think,
270
739000
2000
12:39
"My god. How is reality coming out of that?"
271
741000
5000
12:44
But, in fact, you can start,
272
746000
3000
12:47
even though we haven't trained this neocortical column
273
749000
4000
12:51
to create a specific reality.
274
753000
2000
12:53
But we can ask, "Where is the rose?"
275
755000
4000
12:57
We can ask, "Where is it inside,
276
759000
2000
12:59
if we stimulate it with a picture?"
277
761000
3000
13:02
Where is it inside the neocortex?
278
764000
2000
13:04
Ultimately it's got to be there if we stimulated it with it.
279
766000
4000
13:08
So, the way that we can look at that
280
770000
2000
13:10
is to ignore the neurons, ignore the synapses,
281
772000
3000
13:13
and look just at the raw electrical activity.
282
775000
2000
13:15
Because that is what it's creating.
283
777000
2000
13:17
It's creating electrical patterns.
284
779000
2000
13:19
So when we did this,
285
781000
2000
13:21
we indeed, for the first time,
286
783000
2000
13:23
saw these ghost-like structures:
287
785000
3000
13:26
electrical objects appearing
288
788000
3000
13:29
within the neocortical column.
289
791000
3000
13:32
And it's these electrical objects
290
794000
3000
13:35
that are holding all the information about
291
797000
3000
13:38
whatever stimulated it.
292
800000
3000
13:41
And then when we zoomed into this,
293
803000
2000
13:43
it's like a veritable universe.
294
805000
4000
13:47
So the next step
295
809000
2000
13:49
is just to take these brain coordinates
296
811000
4000
13:53
and to project them into perceptual space.
297
815000
4000
13:57
And if you do that,
298
819000
2000
13:59
you will be able to step inside
299
821000
2000
14:01
the reality that is created
300
823000
2000
14:03
by this machine,
301
825000
2000
14:05
by this piece of the brain.
302
827000
3000
14:08
So, in summary,
303
830000
2000
14:10
I think that the universe may have --
304
832000
2000
14:12
it's possible --
305
834000
2000
14:14
evolved a brain to see itself,
306
836000
3000
14:17
which may be a first step in becoming aware of itself.
307
839000
5000
14:22
There is a lot more to do to test these theories,
308
844000
2000
14:24
and to test any other theories.
309
846000
3000
14:27
But I hope that you are at least partly convinced
310
849000
3000
14:30
that it is not impossible to build a brain.
311
852000
3000
14:33
We can do it within 10 years,
312
855000
2000
14:35
and if we do succeed,
313
857000
2000
14:37
we will send to TED, in 10 years,
314
859000
2000
14:39
a hologram to talk to you. Thank you.
315
861000
3000
14:42
(Applause)
316
864000
6000

▲Back to top

ABOUT THE SPEAKER
Henry Markram - Neuroscientist
Henry Markram is director of Blue Brain, a supercomputing project that can model components of the mammalian brain to precise cellular detail -- and simulate their activity in 3D. Soon he'll simulate a whole rat brain in real time.

Why you should listen

In the microscopic, yet-uncharted circuitry of the cortex, Henry Markram is perhaps the most ambitious -- and our most promising -- frontiersman. Backed by the extraordinary power of the IBM Blue Gene supercomputing architecture, which can perform hundreds of trillions of calculations per second, he's using complex models to precisely simulate the neocortical column (and its tens of millions of neural connections) in 3D.

Though the aim of Blue Brain research is mainly biomedical, it has been edging up on some deep, contentious philosophical questions about the mind -- "Can a robot think?" and "Can consciousness be reduced to mechanical components?" -- the consequence of which Markram is well aware: Asked by Seed Magazine what a simulation of a full brain might do, he answered, "Everything. I mean everything" -- with a grin.

Now, with a successful proof-of-concept for simulation in hand (the project's first phase was completed in 2007), Markram is looking toward a future where brains might be modeled even down to the molecular and genetic level. Computing power marching rightward and up along the graph of Moore's Law, Markram is sure to be at the forefront as answers to the mysteries of cognition emerge.

More profile about the speaker
Henry Markram | Speaker | TED.com