TED2003
Jeff Hawkins: How brain science will change computing
Filmed:
Readability: 3.2
1,674,773 views
Treo creator Jeff Hawkins urges us to take a new look at the brain -- to see it not as a fast processor, but as a memory system that stores and plays back experiences to help us predict, intelligently, what will happen next.
Jeff Hawkins - Computer designer, brain researcher
Jeff Hawkins pioneered the development of PDAs such as the Palm and Treo. Now he's trying to understand how the human brain really works, and adapt its method -- which he describes as a deep system for storing memory -- to create new kinds of computers and tools. Full bio
Jeff Hawkins pioneered the development of PDAs such as the Palm and Treo. Now he's trying to understand how the human brain really works, and adapt its method -- which he describes as a deep system for storing memory -- to create new kinds of computers and tools. Full bio
Double-click the English transcript below to play the video.
00:25
I do two things: I design mobile computers and I study brains.
0
0
3000
00:29
And today's talk is about brains and,
1
4000
2000
00:31
yay, somewhere I have a brain fan out there.
2
6000
2000
00:33
(Laughter)
3
8000
2000
00:35
I'm going to, if I can have my first slide up here,
4
10000
2000
00:37
and you'll see the title of my talk and my two affiliations.
5
12000
4000
00:41
So what I'm going to talk about is why we don't have a good brain theory,
6
16000
4000
00:45
why it is important that we should develop one and what we can do about it.
7
20000
3000
00:48
And I'll try to do all that in 20 minutes. I have two affiliations.
8
23000
3000
00:51
Most of you know me from my Palm and Handspring days,
9
26000
3000
00:54
but I also run a nonprofit scientific research institute
10
29000
3000
00:57
called the Redwood Neuroscience Institute in Menlo Park,
11
32000
2000
00:59
and we study theoretical neuroscience,
12
34000
2000
01:01
and we study how the neocortex works.
13
36000
2000
01:03
I'm going to talk all about that.
14
38000
2000
01:05
I have one slide on my other life, the computer life, and that's the slide here.
15
40000
3000
01:08
These are some of the products I've worked on over the last 20 years,
16
43000
3000
01:11
starting back from the very original laptop to some of the first tablet computers
17
46000
4000
01:15
and so on, and ending up most recently with the Treo,
18
50000
2000
01:17
and we're continuing to do this.
19
52000
2000
01:19
And I've done this because I really believe that mobile computing
20
54000
2000
01:21
is the future of personal computing, and I'm trying to make the world
21
56000
3000
01:24
a little bit better by working on these things.
22
59000
3000
01:27
But this was, I have to admit, all an accident.
23
62000
2000
01:29
I really didn't want to do any of these products
24
64000
2000
01:31
and very early in my career I decided
25
66000
2000
01:33
I was not going to be in the computer industry.
26
68000
3000
01:36
And before I tell you about that, I just have to tell you
27
71000
2000
01:38
this one little picture of graffiti there I picked off the web the other day.
28
73000
2000
01:40
I was looking for a picture of graffiti, little text input language,
29
75000
3000
01:43
and I found the website dedicated to teachers who want to make these,
30
78000
3000
01:46
you know, the script writing things across the top of their blackboard,
31
81000
3000
01:49
and they had added graffiti to it, and I'm sorry about that.
32
84000
3000
01:52
(Laughter)
33
87000
2000
01:54
So what happened was, when I was young and got out of engineering school
34
89000
5000
01:59
at Cornell in '79, I decided -- I went to work for Intel and
35
94000
4000
02:03
I was in the computer industry -- and three months into that,
36
98000
3000
02:06
I fell in love with something else, and I said, "I made the wrong career choice here,"
37
101000
4000
02:10
and I fell in love with brains.
38
105000
3000
02:13
This is not a real brain. This is a picture of one, a line drawing.
39
108000
3000
02:16
But I don't remember exactly how it happened,
40
111000
3000
02:19
but I have one recollection, which was pretty strong in my mind.
41
114000
3000
02:22
In September 1979, Scientific American came out
42
117000
3000
02:25
with a single topic issue about the brain. And it was quite good.
43
120000
3000
02:28
It was one of the best issues ever. And they talked about the neuron
44
123000
3000
02:31
and development and disease and vision and all the things
45
126000
2000
02:33
you might want to know about brains. It was really quite impressive.
46
128000
3000
02:36
And one might have the impression that we really knew a lot about brains.
47
131000
3000
02:39
But the last article in that issue was written by Francis Crick of DNA fame.
48
134000
4000
02:43
Today is, I think, the 50th anniversary of the discovery of DNA.
49
138000
3000
02:46
And he wrote a story basically saying,
50
141000
2000
02:48
well, this is all well and good, but you know what,
51
143000
3000
02:51
we don't know diddley squat about brains
52
146000
2000
02:53
and no one has a clue how these things work,
53
148000
2000
02:55
so don't believe what anyone tells you.
54
150000
2000
02:57
This is a quote from that article. He said, "What is conspicuously lacking,"
55
152000
3000
03:00
he's a very proper British gentleman so, "What is conspicuously lacking
56
155000
4000
03:04
is a broad framework of ideas in which to interpret these different approaches."
57
159000
3000
03:07
I thought the word framework was great.
58
162000
2000
03:09
He didn't say we didn't even have a theory. He says,
59
164000
2000
03:11
we don't even know how to begin to think about it --
60
166000
2000
03:13
we don't even have a framework.
61
168000
2000
03:15
We are in the pre-paradigm days, if you want to use Thomas Kuhn.
62
170000
3000
03:18
And so I fell in love with this, and said look,
63
173000
3000
03:21
we have all this knowledge about brains. How hard can it be?
64
176000
3000
03:24
And this is something we can work on my lifetime. I felt I could make a difference,
65
179000
3000
03:27
and so I tried to get out of the computer business, into the brain business.
66
182000
4000
03:31
First, I went to MIT, the AI lab was there,
67
186000
2000
03:33
and I said, well, I want to build intelligent machines, too,
68
188000
2000
03:35
but the way I want to do it is to study how brains work first.
69
190000
3000
03:38
And they said, oh, you don't need to do that.
70
193000
3000
03:41
We're just going to program computers; that's all we need to do.
71
196000
2000
03:43
And I said, no, you really ought to study brains. They said, oh, you know,
72
198000
3000
03:46
you're wrong. And I said, no, you're wrong, and I didn't get in.
73
201000
2000
03:48
(Laughter)
74
203000
1000
03:50
But I was a little disappointed -- pretty young -- but I went back again
75
205000
2000
03:52
a few years later and this time was in California, and I went to Berkeley.
76
207000
3000
03:55
And I said, I'll go in from the biological side.
77
210000
4000
03:59
So I got in -- in the Ph.D. program in biophysics, and I was, all right,
78
214000
3000
04:02
I'm studying brains now, and I said, well, I want to study theory.
79
217000
3000
04:05
And they said, oh no, you can't study theory about brains.
80
220000
2000
04:07
That's not something you do. You can't get funded for that.
81
222000
2000
04:09
And as a graduate student, you can't do that. So I said, oh my gosh.
82
224000
4000
04:13
I was very depressed. I said, but I can make a difference in this field.
83
228000
2000
04:15
So what I did is I went back in the computer industry
84
230000
3000
04:18
and said, well, I'll have to work here for a while, do something.
85
233000
2000
04:20
That's when I designed all those computer products.
86
235000
3000
04:23
(Laughter)
87
238000
1000
04:24
And I said, I want to do this for four years, make some money,
88
239000
3000
04:27
like I was having a family, and I would mature a bit,
89
242000
4000
04:31
and maybe the business of neuroscience would mature a bit.
90
246000
3000
04:34
Well, it took longer than four years. It's been about 16 years.
91
249000
3000
04:37
But I'm doing it now, and I'm going to tell you about it.
92
252000
2000
04:39
So why should we have a good brain theory?
93
254000
3000
04:42
Well, there's lots of reasons people do science.
94
257000
3000
04:45
One is -- the most basic one is -- people like to know things.
95
260000
3000
04:48
We're curious, and we just go out and get knowledge, you know?
96
263000
2000
04:50
Why do we study ants? Well, it's interesting.
97
265000
2000
04:52
Maybe we'll learn something really useful about it, but it's interesting and fascinating.
98
267000
3000
04:55
But sometimes, a science has some other attributes
99
270000
2000
04:57
which makes it really, really interesting.
100
272000
2000
04:59
Sometimes a science will tell something about ourselves,
101
274000
3000
05:02
it'll tell us who we are.
102
277000
1000
05:03
Rarely, you know: evolution did this and Copernicus did this,
103
278000
3000
05:06
where we have a new understanding of who we are.
104
281000
2000
05:08
And after all, we are our brains. My brain is talking to your brain.
105
283000
4000
05:12
Our bodies are hanging along for the ride, but my brain is talking to your brain.
106
287000
3000
05:15
And if we want to understand who we are and how we feel and perceive,
107
290000
3000
05:18
we really understand what brains are.
108
293000
2000
05:20
Another thing is sometimes science
109
295000
2000
05:22
leads to really big societal benefits and technologies,
110
297000
2000
05:24
or businesses, or whatever, that come out of it. And this is one, too,
111
299000
2000
05:26
because when we understand how brains work, we're going to be able
112
301000
3000
05:29
to build intelligent machines, and I think that's actually a good thing on the whole,
113
304000
3000
05:32
and it's going to have tremendous benefits to society,
114
307000
2000
05:34
just like a fundamental technology.
115
309000
2000
05:36
So why don't we have a good theory of brains?
116
311000
2000
05:38
And people have been working on it for 100 years.
117
313000
3000
05:41
Well, let's first take a look at what normal science looks like.
118
316000
2000
05:43
This is normal science.
119
318000
2000
05:45
Normal science is a nice balance between theory and experimentalists.
120
320000
4000
05:49
And so the theorist guys say, well, I think this is what's going on,
121
324000
2000
05:51
and the experimentalist says, no, you're wrong.
122
326000
2000
05:53
And it goes back and forth, you know?
123
328000
2000
05:55
This works in physics. This works in geology. But if this is normal science,
124
330000
2000
05:57
what does neuroscience look like? This is what neuroscience looks like.
125
332000
3000
06:00
We have this mountain of data, which is anatomy, physiology and behavior.
126
335000
5000
06:05
You can't imagine how much detail we know about brains.
127
340000
3000
06:08
There were 28,000 people who went to the neuroscience conference this year,
128
343000
4000
06:12
and every one of them is doing research in brains.
129
347000
2000
06:14
A lot of data. But there's no theory. There's a little, wimpy box on top there.
130
349000
4000
06:18
And theory has not played a role in any sort of grand way in the neurosciences.
131
353000
5000
06:23
And it's a real shame. Now why has this come about?
132
358000
3000
06:26
If you ask neuroscientists, why is this the state of affair,
133
361000
2000
06:28
they'll first of all admit it. But if you ask them, they'll say,
134
363000
3000
06:31
well, there's various reasons we don't have a good brain theory.
135
366000
3000
06:34
Some people say, well, we don't still have enough data,
136
369000
2000
06:36
we need to get more information, there's all these things we don't know.
137
371000
3000
06:39
Well, I just told you there's so much data coming out your ears.
138
374000
3000
06:42
We have so much information, we don't even know how to begin to organize it.
139
377000
3000
06:45
What good is more going to do?
140
380000
2000
06:47
Maybe we'll be lucky and discover some magic thing, but I don't think so.
141
382000
3000
06:50
This is actually a symptom of the fact that we just don't have a theory.
142
385000
3000
06:53
We don't need more data -- we need a good theory about it.
143
388000
3000
06:56
Another one is sometimes people say, well, brains are so complex,
144
391000
3000
06:59
it'll take another 50 years.
145
394000
2000
07:01
I even think Chris said something like this yesterday.
146
396000
2000
07:03
I'm not sure what you said, Chris, but something like,
147
398000
2000
07:05
well, it's one of the most complicated things in the universe. That's not true.
148
400000
3000
07:08
You're more complicated than your brain. You've got a brain.
149
403000
2000
07:10
And it's also, although the brain looks very complicated,
150
405000
2000
07:12
things look complicated until you understand them.
151
407000
3000
07:15
That's always been the case. And so all we can say, well,
152
410000
3000
07:18
my neocortex, which is the part of the brain I'm interested in, has 30 billion cells.
153
413000
4000
07:22
But, you know what? It's very, very regular.
154
417000
2000
07:24
In fact, it looks like it's the same thing repeated over and over and over again.
155
419000
3000
07:27
It's not as complex as it looks. That's not the issue.
156
422000
3000
07:30
Some people say, brains can't understand brains.
157
425000
2000
07:32
Very Zen-like. Whoo. (Laughter)
158
427000
3000
07:35
You know,
159
430000
1000
07:36
it sounds good, but why? I mean, what's the point?
160
431000
3000
07:39
It's just a bunch of cells. You understand your liver.
161
434000
3000
07:42
It's got a lot of cells in it too, right?
162
437000
2000
07:44
So, you know, I don't think there's anything to that.
163
439000
2000
07:46
And finally, some people say, well, you know,
164
441000
2000
07:48
I don't feel like a bunch of cells, you know. I'm conscious.
165
443000
4000
07:52
I've got this experience, I'm in the world, you know.
166
447000
2000
07:54
I can't be just a bunch of cells. Well, you know,
167
449000
2000
07:56
people used to believe there was a life force to be living,
168
451000
3000
07:59
and we now know that's really not true at all.
169
454000
2000
08:01
And there's really no evidence that says -- well, other than people
170
456000
3000
08:04
just have disbelief that cells can do what they do.
171
459000
2000
08:06
And so, if some people have fallen into the pit of metaphysical dualism,
172
461000
3000
08:09
some really smart people, too, but we can reject all that.
173
464000
3000
08:12
(Laughter)
174
467000
2000
08:14
No, I'm going to tell you there's something else,
175
469000
3000
08:17
and it's really fundamental, and this is what it is:
176
472000
2000
08:19
there's another reason why we don't have a good brain theory,
177
474000
2000
08:21
and it's because we have an intuitive, strongly-held,
178
476000
3000
08:24
but incorrect assumption that has prevented us from seeing the answer.
179
479000
5000
08:29
There's something we believe that just, it's obvious, but it's wrong.
180
484000
3000
08:32
Now, there's a history of this in science and before I tell you what it is,
181
487000
4000
08:36
I'm going to tell you a bit about the history of it in science.
182
491000
2000
08:38
You look at some other scientific revolutions,
183
493000
2000
08:40
and this case, I'm talking about the solar system, that's Copernicus,
184
495000
2000
08:42
Darwin's evolution, and tectonic plates, that's Wegener.
185
497000
3000
08:45
They all have a lot in common with brain science.
186
500000
3000
08:48
First of all, they had a lot of unexplained data. A lot of it.
187
503000
3000
08:51
But it got more manageable once they had a theory.
188
506000
3000
08:54
The best minds were stumped -- really, really smart people.
189
509000
3000
08:57
We're not smarter now than they were then.
190
512000
2000
08:59
It just turns out it's really hard to think of things,
191
514000
2000
09:01
but once you've thought of them, it's kind of easy to understand it.
192
516000
2000
09:03
My daughters understood these three theories
193
518000
2000
09:05
in their basic framework by the time they were in kindergarten.
194
520000
3000
09:08
And now it's not that hard, you know, here's the apple, here's the orange,
195
523000
3000
09:11
you know, the Earth goes around, that kind of stuff.
196
526000
3000
09:14
Finally, another thing is the answer was there all along,
197
529000
2000
09:16
but we kind of ignored it because of this obvious thing, and that's the thing.
198
531000
3000
09:19
It was an intuitive, strong-held belief that was wrong.
199
534000
3000
09:22
In the case of the solar system, the idea that the Earth is spinning
200
537000
3000
09:25
and the surface of the Earth is going like a thousand miles an hour,
201
540000
3000
09:28
and the Earth is going through the solar system about a million miles an hour.
202
543000
3000
09:31
This is lunacy. We all know the Earth isn't moving.
203
546000
2000
09:33
Do you feel like you're moving a thousand miles an hour?
204
548000
2000
09:35
Of course not. You know, and someone who said,
205
550000
2000
09:37
well, it was spinning around in space and it's so huge,
206
552000
2000
09:39
they would lock you up, and that's what they did back then.
207
554000
2000
09:41
(Laughter)
208
556000
1000
09:42
So it was intuitive and obvious. Now what about evolution?
209
557000
3000
09:45
Evolution's the same thing. We taught our kids, well, the Bible says,
210
560000
3000
09:48
you know, God created all these species, cats are cats, dogs are dogs,
211
563000
2000
09:50
people are people, plants are plants, they don't change.
212
565000
3000
09:53
Noah put them on the Ark in that order, blah, blah, blah. And, you know,
213
568000
4000
09:57
the fact is, if you believe in evolution, we all have a common ancestor,
214
572000
4000
10:01
and we all have a common ancestry with the plant in the lobby.
215
576000
3000
10:04
This is what evolution tells us. And, it's true. It's kind of unbelievable.
216
579000
3000
10:07
And the same thing about tectonic plates, you know?
217
582000
3000
10:10
All the mountains and the continents are kind of floating around
218
585000
2000
10:12
on top of the Earth, you know? It's like, it doesn't make any sense.
219
587000
4000
10:16
So what is the intuitive, but incorrect assumption,
220
591000
4000
10:20
that's kept us from understanding brains?
221
595000
2000
10:22
Now I'm going to tell it to you, and it's going to seem obvious that that is correct,
222
597000
2000
10:24
and that's the point, right? Then I'm going to have to make an argument
223
599000
2000
10:26
why you're incorrect about the other assumption.
224
601000
2000
10:28
The intuitive but obvious thing is that somehow intelligence
225
603000
3000
10:31
is defined by behavior,
226
606000
2000
10:33
that we are intelligent because of the way that we do things
227
608000
2000
10:35
and the way we behave intelligently, and I'm going to tell you that's wrong.
228
610000
3000
10:38
What it is is intelligence is defined by prediction.
229
613000
2000
10:40
And I'm going to work you through this in a few slides here,
230
615000
3000
10:43
give you an example of what this means. Here's a system.
231
618000
4000
10:47
Engineers like to look at systems like this. Scientists like to look at systems like this.
232
622000
3000
10:50
They say, well, we have a thing in a box, and we have its inputs and its outputs.
233
625000
3000
10:53
The AI people said, well, the thing in the box is a programmable computer
234
628000
3000
10:56
because that's equivalent to a brain, and we'll feed it some inputs
235
631000
2000
10:58
and we'll get it to do something, have some behavior.
236
633000
2000
11:00
And Alan Turing defined the Turing test, which is essentially saying,
237
635000
3000
11:03
we'll know if something's intelligent if it behaves identical to a human.
238
638000
3000
11:06
A behavioral metric of what intelligence is,
239
641000
3000
11:09
and this has stuck in our minds for a long period of time.
240
644000
3000
11:12
Reality though, I call it real intelligence.
241
647000
2000
11:14
Real intelligence is built on something else.
242
649000
2000
11:16
We experience the world through a sequence of patterns, and we store them,
243
651000
4000
11:20
and we recall them. And when we recall them, we match them up
244
655000
3000
11:23
against reality, and we're making predictions all the time.
245
658000
4000
11:27
It's an eternal metric. There's an eternal metric about us sort of saying,
246
662000
3000
11:30
do we understand the world? Am I making predictions? And so on.
247
665000
3000
11:33
You're all being intelligent right now, but you're not doing anything.
248
668000
2000
11:35
Maybe you're scratching yourself, or picking your nose,
249
670000
2000
11:37
I don't know, but you're not doing anything right now,
250
672000
2000
11:39
but you're being intelligent; you're understanding what I'm saying.
251
674000
3000
11:42
Because you're intelligent and you speak English,
252
677000
2000
11:44
you know what word is at the end of this -- (Silence)
253
679000
1000
11:45
sentence.
254
680000
2000
11:47
The word came into you, and you're making these predictions all the time.
255
682000
3000
11:50
And then, what I'm saying is,
256
685000
2000
11:52
is that the eternal prediction is the output in the neocortex.
257
687000
2000
11:54
And that somehow, prediction leads to intelligent behavior.
258
689000
3000
11:57
And here's how that happens. Let's start with a non-intelligent brain.
259
692000
3000
12:00
Well I'll argue a non-intelligent brain, we got hold of an old brain,
260
695000
4000
12:04
and we're going to say it's like a non-mammal, like a reptile,
261
699000
3000
12:07
so I'll say, an alligator; we have an alligator.
262
702000
2000
12:09
And the alligator has some very sophisticated senses.
263
704000
3000
12:12
It's got good eyes and ears and touch senses and so on,
264
707000
3000
12:15
a mouth and a nose. It has very complex behavior.
265
710000
4000
12:19
It can run and hide. It has fears and emotions. It can eat you, you know.
266
714000
4000
12:23
It can attack. It can do all kinds of stuff.
267
718000
4000
12:27
But we don't consider the alligator very intelligent, not like in a human sort of way.
268
722000
5000
12:32
But it has all this complex behavior already.
269
727000
2000
12:34
Now, in evolution, what happened?
270
729000
2000
12:36
First thing that happened in evolution with mammals,
271
731000
3000
12:39
we started to develop a thing called the neocortex.
272
734000
2000
12:41
And I'm going to represent the neocortex here,
273
736000
2000
12:43
by this box that's sticking on top of the old brain.
274
738000
2000
12:45
Neocortex means new layer. It is a new layer on top of your brain.
275
740000
3000
12:48
If you don't know it, it's the wrinkly thing on the top of your head that,
276
743000
3000
12:51
it's got wrinkly because it got shoved in there and doesn't fit.
277
746000
3000
12:54
(Laughter)
278
749000
1000
12:55
No, really, that's what it is. It's about the size of a table napkin.
279
750000
2000
12:57
And it doesn't fit, so it gets all wrinkly. Now look at how I've drawn this here.
280
752000
3000
13:00
The old brain is still there. You still have that alligator brain.
281
755000
4000
13:04
You do. It's your emotional brain.
282
759000
2000
13:06
It's all those things, and all those gut reactions you have.
283
761000
3000
13:09
And on top of it, we have this memory system called the neocortex.
284
764000
3000
13:12
And the memory system is sitting over the sensory part of the brain.
285
767000
4000
13:16
And so as the sensory input comes in and feeds from the old brain,
286
771000
3000
13:19
it also goes up into the neocortex. And the neocortex is just memorizing.
287
774000
4000
13:23
It's sitting there saying, ah, I'm going to memorize all the things that are going on:
288
778000
4000
13:27
where I've been, people I've seen, things I've heard, and so on.
289
782000
2000
13:29
And in the future, when it sees something similar to that again,
290
784000
4000
13:33
so in a similar environment, or the exact same environment,
291
788000
3000
13:36
it'll play it back. It'll start playing it back.
292
791000
2000
13:38
Oh, I've been here before. And when you've been here before,
293
793000
2000
13:40
this happened next. It allows you to predict the future.
294
795000
3000
13:43
It allows you to, literally it feeds back the signals into your brain;
295
798000
4000
13:47
they'll let you see what's going to happen next,
296
802000
2000
13:49
will let you hear the word "sentence" before I said it.
297
804000
3000
13:52
And it's this feeding back into the old brain
298
807000
3000
13:55
that'll allow you to make very more intelligent decisions.
299
810000
3000
13:58
This is the most important slide of my talk, so I'll dwell on it a little bit.
300
813000
3000
14:01
And so, all the time you say, oh, I can predict the things.
301
816000
4000
14:05
And if you're a rat and you go through a maze, and then you learn the maze,
302
820000
3000
14:08
the next time you're in a maze, you have the same behavior,
303
823000
2000
14:10
but all of a sudden, you're smarter
304
825000
2000
14:12
because you say, oh, I recognize this maze, I know which way to go,
305
827000
3000
14:15
I've been here before, I can envision the future. And that's what it's doing.
306
830000
3000
14:18
In humans -- by the way, this is true for all mammals;
307
833000
3000
14:21
it's true for other mammals -- and in humans, it got a lot worse.
308
836000
2000
14:23
In humans, we actually developed the front part of the neocortex
309
838000
3000
14:26
called the anterior part of the neocortex. And nature did a little trick.
310
841000
4000
14:30
It copied the posterior part, the back part, which is sensory,
311
845000
2000
14:32
and put it in the front part.
312
847000
2000
14:34
And humans uniquely have the same mechanism on the front,
313
849000
2000
14:36
but we use it for motor control.
314
851000
2000
14:38
So we are now able to make very sophisticated motor planning, things like that.
315
853000
3000
14:41
I don't have time to get into all this, but if you want to understand how a brain works,
316
856000
3000
14:44
you have to understand how the first part of the mammalian neocortex works,
317
859000
3000
14:47
how it is we store patterns and make predictions.
318
862000
2000
14:49
So let me give you a few examples of predictions.
319
864000
3000
14:52
I already said the word "sentence." In music,
320
867000
2000
14:54
if you've heard a song before, if you heard Jill sing those songs before,
321
869000
3000
14:57
when she sings them, the next note pops into your head already --
322
872000
3000
15:00
you anticipate it as you're going. If it was an album of music,
323
875000
2000
15:02
the end of one album, the next song pops into your head.
324
877000
3000
15:05
And these things happen all the time. You're making these predictions.
325
880000
2000
15:07
I have this thing called the altered door thought experiment.
326
882000
3000
15:10
And the altered door thought experiment says, you have a door at home,
327
885000
3000
15:13
and when you're here, I'm changing it, I've got a guy
328
888000
3000
15:16
back at your house right now, moving the door around,
329
891000
2000
15:18
and they're going to take your doorknob and move it over two inches.
330
893000
2000
15:20
And when you go home tonight, you're going to put your hand out there,
331
895000
2000
15:22
and you're going to reach for the doorknob and you're going to notice
332
897000
2000
15:24
it's in the wrong spot, and you'll go, whoa, something happened.
333
899000
3000
15:27
It may take a second to figure out what it was, but something happened.
334
902000
2000
15:29
Now I could change your doorknob in other ways.
335
904000
2000
15:31
I can make it larger or smaller, I can change its brass to silver,
336
906000
2000
15:33
I could make it a lever. I can change your door, put colors on;
337
908000
2000
15:35
I can put windows in. I can change a thousand things about your door,
338
910000
3000
15:38
and in the two seconds you take to open your door,
339
913000
2000
15:40
you're going to notice that something has changed.
340
915000
3000
15:43
Now, the engineering approach to this, the AI approach to this,
341
918000
2000
15:45
is to build a door database. It has all the door attributes.
342
920000
3000
15:48
And as you go up to the door, you know, let's check them off one at time.
343
923000
3000
15:51
Door, door, door, you know, color, you know what I'm saying.
344
926000
2000
15:53
We don't do that. Your brain doesn't do that.
345
928000
2000
15:55
What your brain is doing is making constant predictions all the time
346
930000
2000
15:57
about what is going to happen in your environment.
347
932000
2000
15:59
As I put my hand on this table, I expect to feel it stop.
348
934000
3000
16:02
When I walk, every step, if I missed it by an eighth of an inch,
349
937000
3000
16:05
I'll know something has changed.
350
940000
2000
16:07
You're constantly making predictions about your environment.
351
942000
2000
16:09
I'll talk about vision here briefly. This is a picture of a woman.
352
944000
3000
16:12
And when you look at people, your eyes are caught
353
947000
2000
16:14
over at two to three times a second.
354
949000
1000
16:15
You're not aware of this, but your eyes are always moving.
355
950000
2000
16:17
And so when you look at someone's face,
356
952000
2000
16:19
you'd typically go from eye to eye to eye to nose to mouth.
357
954000
2000
16:21
Now, when your eye moves from eye to eye,
358
956000
2000
16:23
if there was something else there like, a nose,
359
958000
2000
16:25
you'd see a nose where an eye is supposed to be,
360
960000
2000
16:27
and you'd go, oh shit, you know --
361
962000
3000
16:30
(Laughter)
362
965000
1000
16:31
There's something wrong about this person.
363
966000
2000
16:33
And that's because you're making a prediction.
364
968000
2000
16:35
It's not like you just look over there and say, what am I seeing now?
365
970000
2000
16:37
A nose, that's okay. No, you have an expectation of what you're going to see.
366
972000
3000
16:40
(Laughter)
367
975000
1000
16:41
Every single moment. And finally, let's think about how we test intelligence.
368
976000
4000
16:45
We test it by prediction. What is the next word in this, you know?
369
980000
3000
16:48
This is to this as this is to this. What is the next number in this sentence?
370
983000
3000
16:51
Here's three visions of an object.
371
986000
2000
16:53
What's the fourth one? That's how we test it. It's all about prediction.
372
988000
4000
16:57
So what is the recipe for brain theory?
373
992000
3000
17:00
First of all, we have to have the right framework.
374
995000
3000
17:03
And the framework is a memory framework,
375
998000
2000
17:05
not a computation or behavior framework. It's a memory framework.
376
1000000
2000
17:07
How do you store and recall these sequences or patterns? It's spatio-temporal patterns.
377
1002000
4000
17:11
Then, if in that framework, you take a bunch of theoreticians.
378
1006000
3000
17:14
Now biologists generally are not good theoreticians.
379
1009000
2000
17:16
It's not always true, but in general, there's not a good history of theory in biology.
380
1011000
4000
17:20
So I found the best people to work with are physicists,
381
1015000
3000
17:23
engineers and mathematicians, who tend to think algorithmically.
382
1018000
3000
17:26
Then they have to learn the anatomy, and they've got to learn the physiology.
383
1021000
3000
17:29
You have to make these theories very realistic in anatomical terms.
384
1024000
4000
17:33
Anyone who gets up and tells you their theory about how the brain works
385
1028000
4000
17:37
and doesn't tell you exactly how it's working in the brain
386
1032000
2000
17:39
and how the wiring works in the brain, it is not a theory.
387
1034000
2000
17:41
And that's what we're doing at the Redwood Neuroscience Institute.
388
1036000
3000
17:44
I would love to have more time to tell you we're making fantastic progress in this thing,
389
1039000
4000
17:48
and I expect to be back up on this stage,
390
1043000
2000
17:50
maybe this will be some other time in the not too distant future and tell you about it.
391
1045000
2000
17:52
I'm really, really excited. This is not going to take 50 years at all.
392
1047000
3000
17:55
So what will brain theory look like?
393
1050000
2000
17:57
First of all, it's going to be a theory about memory.
394
1052000
2000
17:59
Not like computer memory. It's not at all like computer memory.
395
1054000
3000
18:02
It's very, very different. And it's a memory of these very
396
1057000
2000
18:04
high-dimensional patterns, like the things that come from your eyes.
397
1059000
3000
18:07
It's also memory of sequences.
398
1062000
2000
18:09
You cannot learn or recall anything outside of a sequence.
399
1064000
2000
18:11
A song must be heard in sequence over time,
400
1066000
3000
18:14
and you must play it back in sequence over time.
401
1069000
3000
18:17
And these sequences are auto-associatively recalled, so if I see something,
402
1072000
3000
18:20
I hear something, it reminds me of it, and then it plays back automatically.
403
1075000
3000
18:23
It's an automatic playback. And prediction of future inputs is the desired output.
404
1078000
4000
18:27
And as I said, the theory must be biologically accurate,
405
1082000
3000
18:30
it must be testable, and you must be able to build it.
406
1085000
2000
18:32
If you don't build it, you don't understand it. So, one more slide here.
407
1087000
4000
18:36
What is this going to result in? Are we going to really build intelligent machines?
408
1091000
4000
18:40
Absolutely. And it's going to be different than people think.
409
1095000
4000
18:44
No doubt that it's going to happen, in my mind.
410
1099000
3000
18:47
First of all, it's going to be built up, we're going to build the stuff out of silicon.
411
1102000
4000
18:51
The same techniques we use for building silicon computer memories,
412
1106000
3000
18:54
we can use for here.
413
1109000
1000
18:55
But they're very different types of memories.
414
1110000
2000
18:57
And we're going to attach these memories to sensors,
415
1112000
2000
18:59
and the sensors will experience real-live, real-world data,
416
1114000
3000
19:02
and these things are going to learn about their environment.
417
1117000
2000
19:04
Now it's very unlikely the first things you're going to see are like robots.
418
1119000
3000
19:07
Not that robots aren't useful and people can build robots.
419
1122000
3000
19:10
But the robotics part is the hardest part. That's the old brain. That's really hard.
420
1125000
4000
19:14
The new brain is actually kind of easier than the old brain.
421
1129000
2000
19:16
So the first thing we're going to do are the things that don't require a lot of robotics.
422
1131000
3000
19:19
So you're not going to see C-3PO.
423
1134000
2000
19:21
You're going to more see things like, you know, intelligent cars
424
1136000
2000
19:23
that really understand what traffic is and what driving is
425
1138000
3000
19:26
and have learned that certain types of cars with the blinkers on for half a minute
426
1141000
3000
19:29
probably aren't going to turn, things like that.
427
1144000
2000
19:31
(Laughter)
428
1146000
1000
19:32
We can also do intelligent security systems.
429
1147000
2000
19:34
Anywhere where we're basically using our brain, but not doing a lot of mechanics.
430
1149000
4000
19:38
Those are the things that are going to happen first.
431
1153000
2000
19:40
But ultimately, the world's the limit here.
432
1155000
2000
19:42
I don't know how this is going to turn out.
433
1157000
2000
19:44
I know a lot of people who invented the microprocessor
434
1159000
2000
19:46
and if you talk to them, they knew what they were doing was really significant,
435
1161000
5000
19:51
but they didn't really know what was going to happen.
436
1166000
3000
19:54
They couldn't anticipate cell phones and the Internet and all this kind of stuff.
437
1169000
5000
19:59
They just knew like, hey, they were going to build calculators
438
1174000
2000
20:01
and traffic light controllers. But it's going to be big.
439
1176000
2000
20:03
In the same way, this is like brain science and these memories
440
1178000
3000
20:06
are going to be a very fundamental technology, and it's going to lead
441
1181000
3000
20:09
to very unbelievable changes in the next 100 years.
442
1184000
3000
20:12
And I'm most excited about how we're going to use them in science.
443
1187000
4000
20:16
So I think that's all my time, I'm over it, and I'm going to end my talk
444
1191000
3000
20:19
right there.
445
1194000
1000
ABOUT THE SPEAKER
Jeff Hawkins - Computer designer, brain researcherJeff Hawkins pioneered the development of PDAs such as the Palm and Treo. Now he's trying to understand how the human brain really works, and adapt its method -- which he describes as a deep system for storing memory -- to create new kinds of computers and tools.
Why you should listen
Jeff Hawkins' Palm PDA became such a widely used productivity tool during the 1990s that some fanatical users claimed it replaced their brains. But Hawkins' deepest interest was in the brain itself. So after the success of the Palm and Treo, which he brought to market at Handspring, Hawkins delved into brain research at the Redwood Center for Theoretical Neuroscience in Berkeley, Calif., and a new company called Numenta.
Hawkins' dual goal is to achieve an understanding of how the human brain actually works -- and then develop software to mimic its functionality, delivering true artificial intelligence. In his book On Intelligence (2004) he lays out his compelling, controversial theory: Contrary to popular AI wisdom, the human neocortex doesn't work like a processor; rather, it relies on a memory system that stores and plays back experiences to help us predict, intelligently, what will happen next. He thinks that "hierarchical temporal memory" computer platforms, which mimic this functionality (and which Numenta might pioneer), could enable groundbreaking new applications that could powerfully extend human intelligence.
Jeff Hawkins | Speaker | TED.com