1
00:00:05,320 --> 00:00:09,399
Speaker 1: What is special about the wrinkly outer layer of the brain,

2
00:00:09,480 --> 00:00:11,640
Speaker 1: the cortex, And what does this have to do with

3
00:00:11,680 --> 00:00:15,720
Speaker 1: the way that you come to explore and understand the world.

4
00:00:16,079 --> 00:00:17,520
Speaker 2: And by the way, why do you.

5
00:00:17,480 --> 00:00:20,760
Speaker 1: See a whole image when you open your eyes even though.

6
00:00:20,680 --> 00:00:22,799
Speaker 2: Each part of your visual.

7
00:00:22,480 --> 00:00:26,720
Speaker 1: Cortex has access to only a tiny bit of the image.

8
00:00:27,280 --> 00:00:30,040
Speaker 1: And for that matter, the brain is divided into different

9
00:00:30,080 --> 00:00:32,639
Speaker 1: areas for sight and sound and touch and so on.

10
00:00:33,159 --> 00:00:36,559
Speaker 1: And so why when you're petting a cat, why does

11
00:00:36,560 --> 00:00:40,040
Speaker 1: the cat seem unified? Why doesn't the site of the

12
00:00:40,080 --> 00:00:44,159
Speaker 1: cat seem separate from the purring and the feel of

13
00:00:44,240 --> 00:00:47,640
Speaker 1: the fur. Can we build a new model of how

14
00:00:47,640 --> 00:00:50,720
Speaker 1: the brain works and in what ways is what the

15
00:00:50,760 --> 00:00:55,640
Speaker 1: brain doing something very different than what's happening in current AI.

16
00:00:59,040 --> 00:01:02,680
Speaker 1: Welcome to Intercouse with me David Eagleman. I'm a neuroscientist

17
00:01:02,680 --> 00:01:07,080
Speaker 1: at Stanford and in these episodes we sail deeply into

18
00:01:07,080 --> 00:01:10,040
Speaker 1: our three pound universe to understand.

19
00:01:09,600 --> 00:01:12,280
Speaker 2: Why and how our lives look the way they do.

20
00:01:24,040 --> 00:01:27,720
Speaker 1: Today's episode is about a new model of the brain

21
00:01:28,240 --> 00:01:31,160
Speaker 1: developed by my friend and colleague, Jeff Hawkins, and we'll

22
00:01:31,160 --> 00:01:33,920
Speaker 1: get into an interview with him shortly but let me

23
00:01:33,959 --> 00:01:38,200
Speaker 1: preface by saying that for centuries people have stared at

24
00:01:38,200 --> 00:01:41,240
Speaker 1: the brain and tried to figure out how this thing works.

25
00:01:41,280 --> 00:01:44,360
Speaker 1: Because when you stare at it, it's just a huge

26
00:01:44,840 --> 00:01:45,960
Speaker 1: lump of cells.

27
00:01:46,160 --> 00:01:46,880
Speaker 2: You can see that.

28
00:01:46,880 --> 00:01:51,400
Speaker 1: There's a wrinkled layer on the outside. And when people

29
00:01:51,400 --> 00:01:54,480
Speaker 1: dissect that, they can see that that part is about

30
00:01:54,600 --> 00:01:58,520
Speaker 1: three millimeters thick, and it looks a little different, looks grayer.

31
00:01:58,920 --> 00:02:01,280
Speaker 1: And so that part is called the gray matter. And

32
00:02:01,360 --> 00:02:05,520
Speaker 1: we call this the cortex, which means bark, like tree bark.

33
00:02:06,000 --> 00:02:09,160
Speaker 1: And the stuff below that thin layer is called white matter.

34
00:02:09,200 --> 00:02:13,040
Speaker 1: And it looks white because the tiny data cables coming

35
00:02:13,080 --> 00:02:16,240
Speaker 1: off the cells, the axons, these are wrapped in a.

36
00:02:16,200 --> 00:02:19,120
Speaker 2: Little sheath called myelin, which makes it look white.

37
00:02:19,320 --> 00:02:22,799
Speaker 1: Okay, Now, what you immediately noticed by looking at brains

38
00:02:22,919 --> 00:02:26,960
Speaker 1: across different mammals is that all the stuff you find

39
00:02:27,120 --> 00:02:31,520
Speaker 1: under the cortex, all the sub cortical stuff, looks essentially

40
00:02:31,600 --> 00:02:34,960
Speaker 1: the same. Horses and elephants and mice. They all have

41
00:02:35,080 --> 00:02:37,799
Speaker 1: the same architecture going on that we do.

42
00:02:37,840 --> 00:02:38,440
Speaker 2: They all have.

43
00:02:38,400 --> 00:02:42,840
Speaker 1: A thalamus and hippocampus and cerebellum and so on. But

44
00:02:42,919 --> 00:02:46,560
Speaker 1: there's one thing that really distinguishes us from our cousins,

45
00:02:47,040 --> 00:02:51,200
Speaker 1: and for that we return to the gray matter, the cortex.

46
00:02:51,360 --> 00:02:54,800
Speaker 1: It's not that our cousins don't have a cortex. What

47
00:02:54,960 --> 00:03:00,760
Speaker 1: distinguishes us is the absolute enormity of our cortex. We

48
00:03:00,919 --> 00:03:05,400
Speaker 1: humans have a ton of this stuff. So take four

49
00:03:05,440 --> 00:03:08,800
Speaker 1: pieces of paper from your printer and place them next

50
00:03:08,800 --> 00:03:11,120
Speaker 1: to each other to make one really large piece.

51
00:03:11,560 --> 00:03:13,440
Speaker 2: That's how much cortex.

52
00:03:13,080 --> 00:03:16,760
Speaker 1: A human has. If you were to spread out the wrinkles. Now,

53
00:03:16,880 --> 00:03:20,040
Speaker 1: our nearest cousins, the great apes only have about one

54
00:03:20,080 --> 00:03:22,799
Speaker 1: piece of paper worth, and most mammals have a lot

55
00:03:22,880 --> 00:03:26,880
Speaker 1: less than that. So something about the story of the

56
00:03:27,080 --> 00:03:30,600
Speaker 1: runaway human success has to do with the fact that

57
00:03:30,639 --> 00:03:34,200
Speaker 1: we have way more cortex for our body size than

58
00:03:34,320 --> 00:03:38,120
Speaker 1: any other creature. And side note, I'm really talking about

59
00:03:38,200 --> 00:03:42,280
Speaker 1: what's called the neocortex or new cortex, because we also

60
00:03:42,280 --> 00:03:45,720
Speaker 1: have a little bit of paleocortex or old cortex. But

61
00:03:45,760 --> 00:03:48,320
Speaker 1: the thing that really makes us outstanding is the amount

62
00:03:48,360 --> 00:03:50,280
Speaker 1: of neo cortex that we have.

63
00:03:51,400 --> 00:03:54,040
Speaker 2: But what is this neocortex doing.

64
00:03:54,920 --> 00:03:58,280
Speaker 1: Well, if you look at any neuroscience textbook, you'll see

65
00:03:58,320 --> 00:04:01,240
Speaker 1: that this part of the brain, the cortex, is often

66
00:04:01,720 --> 00:04:05,760
Speaker 1: drawn with different colored regions like this red region over

67
00:04:05,800 --> 00:04:08,000
Speaker 1: here is devoted to vision, and this green one is

68
00:04:08,040 --> 00:04:10,760
Speaker 1: devoted to hearing, and this yellow one to touch and

69
00:04:10,800 --> 00:04:14,320
Speaker 1: so on. But something I've been obsessed with and write

70
00:04:14,360 --> 00:04:16,640
Speaker 1: about in my latest book, Live Wired, is that this

71
00:04:16,720 --> 00:04:19,000
Speaker 1: is the wrong way to think about it, because the

72
00:04:19,000 --> 00:04:22,240
Speaker 1: neocortex is remarkably flexible.

73
00:04:22,279 --> 00:04:23,599
Speaker 2: It's not a fixed map.

74
00:04:24,160 --> 00:04:27,440
Speaker 1: If you are born blind, the part of your cortex

75
00:04:27,520 --> 00:04:30,000
Speaker 1: that we would have thought of as visual cortex gets

76
00:04:30,040 --> 00:04:33,720
Speaker 1: taken over by hearing and touch and so on. Now

77
00:04:33,839 --> 00:04:35,400
Speaker 1: let me just be really clear what I mean by

78
00:04:35,480 --> 00:04:39,679
Speaker 1: taking over. The neurons there are the same The cortex

79
00:04:39,760 --> 00:04:42,479
Speaker 1: looks exactly the same from the outside, but the function

80
00:04:42,600 --> 00:04:46,760
Speaker 1: of those particular neurons is now not visual. They have

81
00:04:46,839 --> 00:04:50,120
Speaker 1: nothing to do with visual information anymore. Now that same

82
00:04:50,240 --> 00:04:55,200
Speaker 1: neuron instead of firing when it detects a moving object,

83
00:04:55,600 --> 00:04:58,880
Speaker 1: now it responds to a touch on your.

84
00:04:58,720 --> 00:05:03,479
Speaker 2: Toe, or hearing a B flat note or whatever. So

85
00:05:03,600 --> 00:05:05,000
Speaker 2: the little labels that.

86
00:05:04,920 --> 00:05:08,240
Speaker 1: We draw onto the brain, these maps that we impose,

87
00:05:08,960 --> 00:05:09,400
Speaker 1: these are.

88
00:05:09,320 --> 00:05:11,240
Speaker 2: Actually massively flexible.

89
00:05:11,480 --> 00:05:13,320
Speaker 1: And as you may know, I gave a talk at

90
00:05:13,480 --> 00:05:15,520
Speaker 1: TED about this a while ago, where I showed that

91
00:05:15,880 --> 00:05:19,680
Speaker 1: you can feed in new kinds of information, let's say

92
00:05:19,680 --> 00:05:22,440
Speaker 1: through the ears or the skin, and the brain will

93
00:05:22,480 --> 00:05:25,320
Speaker 1: figure out how to deal with that data. It will

94
00:05:25,760 --> 00:05:30,200
Speaker 1: flexibly devote part of its cortical real estate to that.

95
00:05:30,839 --> 00:05:34,360
Speaker 1: And this line of thinking led some scientists, like Vernon

96
00:05:34,360 --> 00:05:37,920
Speaker 1: mount Castle some decades ago to realize that the cells

97
00:05:37,920 --> 00:05:39,839
Speaker 1: of the cortex are.

98
00:05:39,839 --> 00:05:41,120
Speaker 2: A one trick pony.

99
00:05:41,800 --> 00:05:46,840
Speaker 1: No neuron is inherently a visual neuron or a neuron

100
00:05:46,880 --> 00:05:50,120
Speaker 1: devoted to hearing or touch or smell or taste or

101
00:05:50,160 --> 00:05:54,039
Speaker 1: memory or whatever. All parts of the cortex are perfectly

102
00:05:54,120 --> 00:05:58,120
Speaker 1: capable and willing to take on any job. So that

103
00:05:58,240 --> 00:06:02,200
Speaker 1: suggests they're all running some sort of basic algorithm. And

104
00:06:02,200 --> 00:06:05,400
Speaker 1: it doesn't matter what kind of data you feed in.

105
00:06:05,440 --> 00:06:08,479
Speaker 1: Different parts of the cortex will say cool, I'll build

106
00:06:08,560 --> 00:06:11,479
Speaker 1: a representation of that data. I don't care if it

107
00:06:11,480 --> 00:06:15,520
Speaker 1: comes from photons or air compression waves or temperature or whatever.

108
00:06:15,960 --> 00:06:19,520
Speaker 1: I'm on the job here to build an understanding of

109
00:06:19,560 --> 00:06:24,680
Speaker 1: whatever is coming in locally. Now, it's not individual neurons

110
00:06:24,680 --> 00:06:28,560
Speaker 1: that are building models, but instead groups of many tens

111
00:06:28,680 --> 00:06:33,760
Speaker 1: of thousands of neurons arranged in a six layered cylinder.

112
00:06:33,920 --> 00:06:37,080
Speaker 1: So think about this like you're a geologist and you

113
00:06:37,200 --> 00:06:40,480
Speaker 1: drilled out a cylinder of rock and you saw six

114
00:06:40,600 --> 00:06:43,120
Speaker 1: layers in it, six sedimentary layers.

115
00:06:43,480 --> 00:06:45,359
Speaker 2: That's what the neocortex looks like.

116
00:06:45,520 --> 00:06:49,680
Speaker 1: Six layers. And it's built out of these columns which

117
00:06:49,720 --> 00:06:52,839
Speaker 1: have the same types of neurons with the same connection

118
00:06:53,000 --> 00:06:56,760
Speaker 1: patterns in each column. And so think about the cortex

119
00:06:57,160 --> 00:07:00,440
Speaker 1: as being made of lots of these columns, like taking

120
00:07:00,880 --> 00:07:03,600
Speaker 1: hundreds of thousands of grains of rice and standing them

121
00:07:03,680 --> 00:07:05,760
Speaker 1: up on their end and packing them all next to

122
00:07:05,800 --> 00:07:09,560
Speaker 1: each other. Now, people have known about cortical columns for

123
00:07:09,600 --> 00:07:13,600
Speaker 1: many decades since Vernon Mountcastle first discovered these in nineteen

124
00:07:13,680 --> 00:07:17,880
Speaker 1: fifty seven. But recently someone has pulled together several different

125
00:07:17,960 --> 00:07:22,640
Speaker 1: threads to propose how this could underlie what the cortex

126
00:07:22,760 --> 00:07:26,800
Speaker 1: is all about. And that's someone is Jeff Hawkins and

127
00:07:26,840 --> 00:07:30,040
Speaker 1: his team. And so I met with Jeff in my studio. Now,

128
00:07:30,200 --> 00:07:33,000
Speaker 1: Jeff is one of my favorite people because he does

129
00:07:33,600 --> 00:07:37,000
Speaker 1: theoretical neuroscience. He really tries to figure out the big

130
00:07:37,080 --> 00:07:40,360
Speaker 1: picture of what the brain is doing. Now, Jeff has

131
00:07:40,400 --> 00:07:43,160
Speaker 1: a very interesting history, so I'll just mention that in

132
00:07:43,200 --> 00:07:46,400
Speaker 1: the nineteen eighties he was a graduate student at Berkeley,

133
00:07:46,400 --> 00:07:50,119
Speaker 1: where he proposed a PhD thesis on a new theory

134
00:07:50,160 --> 00:07:53,560
Speaker 1: of the cortex, but his proposal was rejected, and so

135
00:07:53,640 --> 00:07:58,000
Speaker 1: he ended up pursuing his vision for mobile computing instead,

136
00:07:58,080 --> 00:08:01,640
Speaker 1: and in nineteen ninety two he launch the company Palm,

137
00:08:01,840 --> 00:08:05,160
Speaker 1: which made the Palm Pilot. If you remember that, this

138
00:08:05,320 --> 00:08:08,720
Speaker 1: was this little handheld device and you could write on

139
00:08:08,760 --> 00:08:11,880
Speaker 1: it with a stylus and it would translate your handwriting

140
00:08:11,880 --> 00:08:14,520
Speaker 1: into text. And you can use this for your address

141
00:08:14,520 --> 00:08:17,119
Speaker 1: book and your calendar and your contacts and note taking.

142
00:08:17,440 --> 00:08:21,400
Speaker 1: This was the first entrant into the world of portable computing,

143
00:08:21,480 --> 00:08:21,720
Speaker 1: and it.

144
00:08:21,680 --> 00:08:22,679
Speaker 2: Really changed the world.

145
00:08:23,480 --> 00:08:27,320
Speaker 1: Anyhow, A decade later, Jeff returned to his original love,

146
00:08:27,360 --> 00:08:31,440
Speaker 1: which was theoretical neuroscience, trying to figure out what's going

147
00:08:31,520 --> 00:08:34,440
Speaker 1: on with the brain, and he wrote a book in

148
00:08:34,440 --> 00:08:37,800
Speaker 1: two thousand and four called on Intelligence, which was very

149
00:08:37,880 --> 00:08:40,640
Speaker 1: influential on me and lots of other thinkers I know.

150
00:08:40,920 --> 00:08:43,360
Speaker 1: So I was very excited when Jeff recently came out

151
00:08:43,400 --> 00:08:47,319
Speaker 1: with his next book that represents his last decade and

152
00:08:47,400 --> 00:08:51,040
Speaker 1: a half of research. It's called One Thousand Brains, a

153
00:08:51,080 --> 00:08:54,559
Speaker 1: New Theory of Intelligence, and it describes his framework for

154
00:08:54,640 --> 00:08:57,880
Speaker 1: thinking about the brain. So, without further ado, let's dive

155
00:08:57,920 --> 00:09:06,560
Speaker 1: into a very cool new model of the brain. Okay, Jeff,

156
00:09:06,600 --> 00:09:09,079
Speaker 1: So you are a theoretician. You think about the brain

157
00:09:09,080 --> 00:09:11,840
Speaker 1: from a high level. We're in this era now of

158
00:09:12,000 --> 00:09:14,400
Speaker 1: AI where AI is doing all kinds of things that

159
00:09:14,440 --> 00:09:17,200
Speaker 1: are amazing and no one expected. But you see the

160
00:09:17,280 --> 00:09:19,920
Speaker 1: brain as being very different from what is going on

161
00:09:19,960 --> 00:09:22,520
Speaker 1: with let's say, large language models. So tell us about that.

162
00:09:22,840 --> 00:09:26,480
Speaker 2: That's absolutely true. You know, the current AI wave is

163
00:09:26,480 --> 00:09:29,200
Speaker 2: really amazing, but those models don't work at all like

164
00:09:29,280 --> 00:09:32,160
Speaker 2: the brain. And I think you could start with one

165
00:09:32,360 --> 00:09:37,720
Speaker 2: really fundamental difference. Brains work through movement. We move our

166
00:09:37,760 --> 00:09:40,640
Speaker 2: bodies through the world. We move our hands over objects

167
00:09:40,679 --> 00:09:42,520
Speaker 2: to touch and learn what they are. We move our

168
00:09:42,520 --> 00:09:45,600
Speaker 2: eyes constantly, so the inputs of the brain are constantly changing,

169
00:09:45,640 --> 00:09:48,800
Speaker 2: but mostly because we're moving through the world. And the

170
00:09:48,880 --> 00:09:51,520
Speaker 2: term for that is a centory motor system. And the

171
00:09:51,559 --> 00:09:54,560
Speaker 2: brain can't understand its inputs unless it knows how it's

172
00:09:54,559 --> 00:09:57,280
Speaker 2: moving through the world. So we learn by exploring, by

173
00:09:57,679 --> 00:10:01,880
Speaker 2: moving different places, picking things up, touching, so on, and

174
00:10:01,920 --> 00:10:05,120
Speaker 2: that's all Animals that move in the world learn this way.

175
00:10:05,320 --> 00:10:07,760
Speaker 2: So this idea that the brain is the central motor

176
00:10:07,800 --> 00:10:10,320
Speaker 2: system has been known back in the late eighteen hundreds,

177
00:10:10,480 --> 00:10:14,400
Speaker 2: but it's pretty much ignored by everybody. But it leads

178
00:10:14,440 --> 00:10:18,080
Speaker 2: to a very fundamental different way of how we acquire

179
00:10:18,120 --> 00:10:21,520
Speaker 2: knowledge and how knowledge is represented in the brain. Whereas

180
00:10:22,080 --> 00:10:25,000
Speaker 2: today's AI is most of it's built on well deep

181
00:10:25,080 --> 00:10:28,280
Speaker 2: learning of transformer technologies, which are essentially we feed data

182
00:10:28,320 --> 00:10:31,360
Speaker 2: to it. We don't it doesn't explore it, and we

183
00:10:31,400 --> 00:10:34,280
Speaker 2: feed to large language models. We just feed a language.

184
00:10:34,400 --> 00:10:37,280
Speaker 2: So there's no inherent knowledge about what these words mean,

185
00:10:37,400 --> 00:10:40,720
Speaker 2: only what these words mean in the context of other words. Right,

186
00:10:40,760 --> 00:10:43,320
Speaker 2: But you and I can pick up a cat and

187
00:10:43,400 --> 00:10:45,640
Speaker 2: touch it and feel it and know this warmth, and

188
00:10:45,640 --> 00:10:48,360
Speaker 2: we understand how its body's moved because no one has

189
00:10:48,440 --> 00:10:52,240
Speaker 2: to tell us that. We just experience it directly. So

190
00:10:52,880 --> 00:10:56,880
Speaker 2: this is a huge gap between brains. Pretty much all

191
00:10:56,920 --> 00:11:00,080
Speaker 2: brains work by century motor learning and almost all oh

192
00:11:00,559 --> 00:11:03,920
Speaker 2: it doesn't. And you can just peel the layers apart

193
00:11:03,960 --> 00:11:05,920
Speaker 2: and see what the differences are, and it makes a

194
00:11:06,040 --> 00:11:08,679
Speaker 2: huge difference. So I'm not a fan. I'm a fan

195
00:11:08,760 --> 00:11:10,920
Speaker 2: of AI today, but I don't think it's the future

196
00:11:10,960 --> 00:11:12,960
Speaker 2: of AI. I don't think it's going to get you

197
00:11:13,080 --> 00:11:15,480
Speaker 2: to what people really want or truly intelligent machines.

198
00:11:16,840 --> 00:11:19,000
Speaker 1: Okay, terrific, And we'll dive into that more in a

199
00:11:19,040 --> 00:11:19,480
Speaker 1: little bit.

200
00:11:19,559 --> 00:11:19,760
Speaker 2: Now.

201
00:11:19,960 --> 00:11:23,559
Speaker 1: When we look at let's say the human brain, there's

202
00:11:24,000 --> 00:11:26,760
Speaker 1: lots of areas that we can point to. There's the

203
00:11:26,920 --> 00:11:31,199
Speaker 1: cortext and wrinkly outer bit, there's all these subcortical areas.

204
00:11:31,440 --> 00:11:35,640
Speaker 1: When you think about intelligence and the stuff that we're

205
00:11:35,640 --> 00:11:37,280
Speaker 1: going to talk about today, what is the part that

206
00:11:37,320 --> 00:11:38,040
Speaker 1: you concentrate on.

207
00:11:38,160 --> 00:11:41,319
Speaker 2: Well, we concentrate first and foremost in the New York cortext,

208
00:11:41,360 --> 00:11:43,560
Speaker 2: which is about seventy five percent of the volume of

209
00:11:43,600 --> 00:11:45,120
Speaker 2: your brain. I mean it's what you see, as you said,

210
00:11:45,360 --> 00:11:46,760
Speaker 2: if you can take a scope, and that's what you

211
00:11:46,800 --> 00:11:50,240
Speaker 2: see in the New York cortext. And so it's a pretty

212
00:11:50,280 --> 00:11:53,600
Speaker 2: dominant part of what we think of intelligence. You can't

213
00:11:53,640 --> 00:11:56,120
Speaker 2: consider it completely on its own. I mean, it's connected

214
00:11:56,160 --> 00:11:59,000
Speaker 2: to all these other things. And so we also study

215
00:11:59,080 --> 00:12:02,160
Speaker 2: those other things and in service to the new or cortex.

216
00:12:02,320 --> 00:12:05,040
Speaker 2: So we study the thalmbists, and we study the cerebellopment,

217
00:12:05,080 --> 00:12:07,320
Speaker 2: We study the basic just because you have to know

218
00:12:07,360 --> 00:12:12,640
Speaker 2: how the cortex workship these other things. But our primarily

219
00:12:12,679 --> 00:12:15,400
Speaker 2: our goal and many neurosign this goal is to understand

220
00:12:15,440 --> 00:12:18,360
Speaker 2: the New York cortex, because that's what mammals have. We've

221
00:12:18,360 --> 00:12:21,280
Speaker 2: got a big one. You know, everything we think, most

222
00:12:21,320 --> 00:12:23,560
Speaker 2: of what we think about being intelligent, about our ability

223
00:12:23,600 --> 00:12:25,960
Speaker 2: to understand the world and generate language and see and

224
00:12:26,040 --> 00:12:28,080
Speaker 2: hear and and so on, is the New York cortext

225
00:12:28,160 --> 00:12:31,640
Speaker 2: not one hundred percent, but most of it. Unfortunately. Also,

226
00:12:31,800 --> 00:12:33,520
Speaker 2: not only is it the biggest structure, but it's a

227
00:12:33,679 --> 00:12:36,800
Speaker 2: very very regular structure. So you can look at this thing,

228
00:12:36,800 --> 00:12:38,760
Speaker 2: the New or Courts is like a sheet of cells.

229
00:12:38,880 --> 00:12:41,600
Speaker 2: It's you know, it's like a size of a large

230
00:12:41,600 --> 00:12:44,439
Speaker 2: dinner napkin and only a few millimeters thick, and it

231
00:12:44,480 --> 00:12:48,520
Speaker 2: gets wrinkly because it's stuck it in your head that

232
00:12:48,960 --> 00:12:53,239
Speaker 2: everywhere you look on it, it looks remarkably complicated and remarkably

233
00:12:53,280 --> 00:12:56,240
Speaker 2: the same. So the areas are doing vision, look like

234
00:12:56,520 --> 00:12:59,880
Speaker 2: the areas are doing languages, look like the areas are

235
00:12:59,880 --> 00:13:02,520
Speaker 2: doing touch. Look the areas that are doing everything. Really

236
00:13:02,960 --> 00:13:05,839
Speaker 2: and so there's been long speculated that there's sort of

237
00:13:05,880 --> 00:13:10,480
Speaker 2: a common algorithmic principle that's applying to everything everything we do,

238
00:13:10,520 --> 00:13:12,679
Speaker 2: all of our sensory inputs, all of our thinking, all

239
00:13:12,720 --> 00:13:16,160
Speaker 2: the language, it's hard to believe, but the evidence is overwhelming,

240
00:13:16,800 --> 00:13:19,560
Speaker 2: and so our research has really been to understand what

241
00:13:19,720 --> 00:13:23,679
Speaker 2: is that that algorithm, corical algorithm, often referred to as

242
00:13:23,679 --> 00:13:27,640
Speaker 2: a cortical column. You know, this repeated structure that seems

243
00:13:27,640 --> 00:13:29,760
Speaker 2: to underline vision and hearing and touch and thought and

244
00:13:29,800 --> 00:13:33,760
Speaker 2: everything we do. And that's that's just just an appealing

245
00:13:33,800 --> 00:13:35,920
Speaker 2: thing to try to understand. And we've cracked it. We've

246
00:13:35,960 --> 00:13:38,160
Speaker 2: actually we actually cracked it. We understand what's going on.

247
00:13:38,280 --> 00:13:38,880
Speaker 2: That's awesome.

248
00:13:38,920 --> 00:13:41,640
Speaker 1: Okay, so a couple of things, right, So the way

249
00:13:41,679 --> 00:13:43,959
Speaker 1: I sometimes phrase this to people is that if I

250
00:13:44,000 --> 00:13:46,720
Speaker 1: had a magical microscope and could show you a part

251
00:13:46,720 --> 00:13:48,360
Speaker 1: of the brain and you can see all the activity

252
00:13:48,480 --> 00:13:51,680
Speaker 1: running around in the cortext there, could you tell me

253
00:13:51,800 --> 00:13:54,599
Speaker 1: is that visual cortex are auditory or somatosensory? And the

254
00:13:54,640 --> 00:13:56,400
Speaker 1: answer is you couldn't tell me, and I couldn't tell you.

255
00:13:56,400 --> 00:13:59,120
Speaker 2: Guys, Right, it all looks the same, and there's a

256
00:13:59,280 --> 00:14:01,240
Speaker 2: and there's as you know, oh, there's these experiments people

257
00:14:01,240 --> 00:14:02,840
Speaker 2: have done where well, first of all, if you have

258
00:14:02,920 --> 00:14:05,319
Speaker 2: trauma to one part of the cortext other parts will

259
00:14:05,320 --> 00:14:08,800
Speaker 2: pick up the same function. You can also people re

260
00:14:08,960 --> 00:14:11,280
Speaker 2: routed sensory and puts the different parts of the cortext

261
00:14:11,559 --> 00:14:13,280
Speaker 2: in animals and they seem the work.

262
00:14:13,559 --> 00:14:17,480
Speaker 1: So, for example, you have visual information instead of going

263
00:14:17,480 --> 00:14:19,920
Speaker 1: to the back of the visual cortex, that gets rerouted

264
00:14:19,960 --> 00:14:24,520
Speaker 1: to the auditory cortex, and that auditory cortex becomes visual cortext.

265
00:14:24,640 --> 00:14:30,080
Speaker 2: Right, It's incredibly powerful and flexible system. And mammals, you know,

266
00:14:30,160 --> 00:14:32,800
Speaker 2: all we have we have a set of sensors, quite

267
00:14:32,800 --> 00:14:34,440
Speaker 2: a few action more than most people think because the

268
00:14:34,480 --> 00:14:37,800
Speaker 2: skin there's a lot of different sensors. But other animals

269
00:14:37,880 --> 00:14:40,680
Speaker 2: have different sensors and they have cortext too, And so

270
00:14:41,920 --> 00:14:45,760
Speaker 2: there's seems to be this universal algorithm that can be applied.

271
00:14:46,440 --> 00:14:50,000
Speaker 2: And now we know it's a century motor algorithm. It

272
00:14:50,000 --> 00:14:52,840
Speaker 2: can be applied, and and we've spent decades trying to

273
00:14:52,840 --> 00:14:55,480
Speaker 2: figure this out and we've cracked it. Oh that's awesome.

274
00:14:55,760 --> 00:14:57,520
Speaker 1: Just before you tell us about that, So tell us

275
00:14:57,520 --> 00:14:58,880
Speaker 1: what a cortical column is.

276
00:14:59,000 --> 00:15:01,240
Speaker 2: Okay, So imagine we talked about. The near cortex is

277
00:15:01,240 --> 00:15:04,800
Speaker 2: a sheet of cells like three milimeters stick. A cortical

278
00:15:04,920 --> 00:15:07,600
Speaker 2: column is a little section of that going through the

279
00:15:07,640 --> 00:15:11,480
Speaker 2: three free milimeters. It's convariating with from a third of

280
00:15:11,480 --> 00:15:14,040
Speaker 2: a millimeters to a millimeters in diameter. It's it's it's

281
00:15:14,040 --> 00:15:15,920
Speaker 2: a it's not something you would see. It's not like

282
00:15:15,960 --> 00:15:18,520
Speaker 2: it's sitting there to be plucked out. But we know

283
00:15:18,560 --> 00:15:21,920
Speaker 2: they exist, and so within that, let's let's say it's

284
00:15:21,960 --> 00:15:25,600
Speaker 2: a three millimeters tall and a half millimeters wide cylinder

285
00:15:25,640 --> 00:15:30,320
Speaker 2: that goes across the cortex that contains all the neural

286
00:15:30,400 --> 00:15:33,560
Speaker 2: machinery that you would see anywhere in the cortex and

287
00:15:33,560 --> 00:15:36,880
Speaker 2: in each cortical column, because they look like a little

288
00:15:36,880 --> 00:15:38,520
Speaker 2: grain of rice in some sense. Why you can imagine

289
00:15:38,560 --> 00:15:40,440
Speaker 2: what's a little brains of ice stacked next to each other.

290
00:15:41,160 --> 00:15:44,640
Speaker 2: Each cornico column gets input from some well in parts

291
00:15:44,640 --> 00:15:47,320
Speaker 2: of the bend they get from some pats of sensory input,

292
00:15:47,400 --> 00:15:49,520
Speaker 2: so from pats to the retina, from pats to the

293
00:15:49,560 --> 00:15:53,600
Speaker 2: cochlea of patrio skin. Other parts get information from parts

294
00:15:53,640 --> 00:15:56,560
Speaker 2: of the near cortex. So the cortex is connected to cortex,

295
00:15:57,200 --> 00:15:58,960
Speaker 2: but each one is looking at a small If you

296
00:15:58,960 --> 00:16:01,640
Speaker 2: think about the primary entry regions of the cortex, which

297
00:16:01,640 --> 00:16:06,240
Speaker 2: are quite large, they're getting input from a small sensory area, right,

298
00:16:06,320 --> 00:16:09,400
Speaker 2: And so people used to think that, well, if this

299
00:16:09,560 --> 00:16:11,480
Speaker 2: bil colm is only getting input from a small part

300
00:16:11,480 --> 00:16:13,840
Speaker 2: of the rent. Now, it can't really doing very much right,

301
00:16:13,880 --> 00:16:15,920
Speaker 2: It can't be very smart. All you could do is

302
00:16:15,960 --> 00:16:18,880
Speaker 2: process a little piece of information there, and therefore maybe

303
00:16:18,920 --> 00:16:20,760
Speaker 2: it's going to detect an edge or something like that,

304
00:16:20,800 --> 00:16:22,560
Speaker 2: and there's a lot of evidence for that. But we

305
00:16:22,600 --> 00:16:27,560
Speaker 2: now know what happens is that the cortical comms they

306
00:16:27,600 --> 00:16:30,800
Speaker 2: get input from over time, from different parts of the world.

307
00:16:30,880 --> 00:16:33,160
Speaker 2: So the eyes are moving like three times a second,

308
00:16:33,360 --> 00:16:35,320
Speaker 2: and so that cortical comms may be looking at three

309
00:16:35,360 --> 00:16:39,640
Speaker 2: different things every second, and it can integrate how the

310
00:16:39,720 --> 00:16:42,240
Speaker 2: sensor is moving, how your eyes are moving, with what

311
00:16:42,320 --> 00:16:45,400
Speaker 2: it's sensing to build models that are much larger than

312
00:16:45,400 --> 00:16:47,840
Speaker 2: it can sense. And the same way that you could

313
00:16:47,840 --> 00:16:51,280
Speaker 2: take your finger in a dark room and say, okay, David,

314
00:16:51,320 --> 00:16:53,280
Speaker 2: I want you to learn this new object. Let's call it.

315
00:16:53,320 --> 00:16:55,880
Speaker 2: You know, a coffee cup. You never touch it, and

316
00:16:55,880 --> 00:16:57,640
Speaker 2: so you could do is you touch the coffee cup

317
00:16:57,680 --> 00:16:59,600
Speaker 2: and you move your finger along and around, and as

318
00:16:59,640 --> 00:17:01,840
Speaker 2: you do, you build a three dimensional model of the cup.

319
00:17:01,920 --> 00:17:04,639
Speaker 2: Even though you're only getting input from one fingertip, the

320
00:17:04,720 --> 00:17:06,640
Speaker 2: eyes are doing the same thing. It's surprising you don't

321
00:17:06,680 --> 00:17:10,200
Speaker 2: realize this so every quarter do Colm, when we understand

322
00:17:10,240 --> 00:17:13,880
Speaker 2: now is doing this sort of processing movement, information and

323
00:17:13,960 --> 00:17:17,200
Speaker 2: sense for information, building what we call structure or three

324
00:17:17,280 --> 00:17:19,400
Speaker 2: D models of things in the world. So it's quite

325
00:17:19,440 --> 00:17:22,560
Speaker 2: different than even those neurosciences think about it, and there's

326
00:17:22,600 --> 00:17:24,080
Speaker 2: a lot of reasons we can talk about how it

327
00:17:24,160 --> 00:17:25,840
Speaker 2: was missed for all these years.

328
00:17:25,960 --> 00:17:28,200
Speaker 1: So in the court, you have a century six layers

329
00:17:28,240 --> 00:17:32,679
Speaker 1: of cells, and a column is all six layers, is

330
00:17:32,720 --> 00:17:36,639
Speaker 1: all six layers. It's going up and down. It's like

331
00:17:36,880 --> 00:17:38,440
Speaker 1: think of it like layers of a cake. And the

332
00:17:38,520 --> 00:17:41,480
Speaker 1: column is you're taking a straw and shoving it through

333
00:17:41,520 --> 00:17:43,000
Speaker 1: the top, and so you've got.

334
00:17:42,760 --> 00:17:46,320
Speaker 2: This, Okay, got a straw cake.

335
00:17:46,560 --> 00:17:49,280
Speaker 1: Okay, great, And so the idea is if you're looking

336
00:17:49,280 --> 00:17:54,120
Speaker 1: at some column in you know, in primary visual cortex. Yeah,

337
00:17:54,240 --> 00:17:57,600
Speaker 1: your point, Jeff was that, you know, it's it's like

338
00:17:57,680 --> 00:17:59,960
Speaker 1: looking at the world through a straw.

339
00:18:00,119 --> 00:18:02,439
Speaker 2: It only sees a little tiny piece of the world.

340
00:18:02,480 --> 00:18:05,520
Speaker 1: But because the eyes are moving arout, because you're exploring

341
00:18:05,560 --> 00:18:09,560
Speaker 1: the world, this is actually getting lots of parts of information.

342
00:18:09,600 --> 00:18:11,920
Speaker 1: It's exploring the world in the same way that your

343
00:18:11,920 --> 00:18:13,000
Speaker 1: finger typically.

344
00:18:12,640 --> 00:18:15,160
Speaker 2: Right, and it has to integrate information over time, that's

345
00:18:15,160 --> 00:18:17,040
Speaker 2: the key, right, And you can literally do this. You

346
00:18:17,040 --> 00:18:19,720
Speaker 2: can look at the world through a straw, right, and

347
00:18:19,720 --> 00:18:21,600
Speaker 2: and you can say, oh, what am I looking at? Well,

348
00:18:21,600 --> 00:18:24,320
Speaker 2: you can't tell them. You start moving the straw and

349
00:18:24,359 --> 00:18:26,600
Speaker 2: then you can start and you can also learn objects

350
00:18:26,640 --> 00:18:29,200
Speaker 2: that way. So literally you can learn by looking through

351
00:18:29,200 --> 00:18:31,640
Speaker 2: a straw, which is what sort of what one column

352
00:18:31,680 --> 00:18:32,920
Speaker 2: is doing? Got it?

353
00:18:33,040 --> 00:18:37,119
Speaker 1: And in your model there are thousands of such columns

354
00:18:37,320 --> 00:18:42,359
Speaker 1: and each one of these is learning a model of

355
00:18:42,400 --> 00:18:44,800
Speaker 1: the world as it's going. So tell us about right, right.

356
00:18:44,920 --> 00:18:48,720
Speaker 2: So I think this idea that there's all these columns

357
00:18:48,800 --> 00:18:50,960
Speaker 2: is not a new idea and that they have this

358
00:18:51,040 --> 00:18:54,320
Speaker 2: fundamental argithm. But what we were I think the first

359
00:18:54,320 --> 00:18:56,439
Speaker 2: people to kind of figure out what it is and

360
00:18:56,480 --> 00:18:59,119
Speaker 2: what it's doing. So the trick of this thing is

361
00:18:59,280 --> 00:19:01,680
Speaker 2: it's trick it here. You know, when you look out

362
00:19:01,680 --> 00:19:04,439
Speaker 2: at the world, you have a sense anybody you have

363
00:19:04,480 --> 00:19:06,200
Speaker 2: a sense where things are. I have a sense where

364
00:19:06,240 --> 00:19:07,760
Speaker 2: you are relative to me. I have a sense where

365
00:19:07,800 --> 00:19:09,240
Speaker 2: this microphone is relative to me. I know where my

366
00:19:09,240 --> 00:19:12,399
Speaker 2: hand is relative to this cop I Now there turns

367
00:19:12,400 --> 00:19:14,880
Speaker 2: out that you have any kind of sense of location

368
00:19:15,040 --> 00:19:17,840
Speaker 2: in space, you have to have neurons representing it. There's

369
00:19:17,880 --> 00:19:19,920
Speaker 2: nothing goes on in the brain if there aren't neurons

370
00:19:19,960 --> 00:19:22,960
Speaker 2: firing doing it. Turns out most of the machinery in

371
00:19:23,000 --> 00:19:25,600
Speaker 2: the New York cortex is keeping track of where things

372
00:19:25,640 --> 00:19:28,720
Speaker 2: are relative to other things. So those six layers, all

373
00:19:28,720 --> 00:19:32,640
Speaker 2: those cells, at least half of that circuitry is tracking

374
00:19:33,160 --> 00:19:35,679
Speaker 2: where the sensory input is coming from in the world.

375
00:19:36,000 --> 00:19:38,120
Speaker 2: So if I move my finger over this coffee cup,

376
00:19:38,760 --> 00:19:41,560
Speaker 2: the part that's getting information from the sensory like I'm

377
00:19:41,600 --> 00:19:44,399
Speaker 2: sensing an edge, for your example, as I move my finger,

378
00:19:44,560 --> 00:19:47,000
Speaker 2: it has to keep track of where my finger is,

379
00:19:47,040 --> 00:19:50,080
Speaker 2: a location of it and its orientation relative this cup.

380
00:19:50,160 --> 00:19:53,719
Speaker 2: Is quite complicated, but that's what it has to do

381
00:19:53,800 --> 00:19:55,639
Speaker 2: to build the models. And now we know how it

382
00:19:55,720 --> 00:19:59,000
Speaker 2: does it. There's all this evidence for it. So the

383
00:19:59,080 --> 00:20:01,280
Speaker 2: brain is just trying to track of where all of

384
00:20:01,280 --> 00:20:03,479
Speaker 2: its inputs are in the world, all relative other things.

385
00:20:03,520 --> 00:20:05,800
Speaker 2: Then it builds up these three dimensional models of the world.

386
00:20:06,000 --> 00:20:08,120
Speaker 2: So tell us about how it does that then, right,

387
00:20:08,200 --> 00:20:10,520
Speaker 2: so you can think about when you're in high school,

388
00:20:10,520 --> 00:20:13,360
Speaker 2: you learned about Cartesian coordinates, x, y, and z coordinates, right,

389
00:20:13,800 --> 00:20:15,359
Speaker 2: and so if I wanted to say where is something?

390
00:20:15,400 --> 00:20:17,399
Speaker 2: Where are your relative to me? I might say, okay,

391
00:20:17,440 --> 00:20:19,359
Speaker 2: your nose the origin, and I could say it's some

392
00:20:19,680 --> 00:20:21,879
Speaker 2: distance from here, and you know X, Y and Z.

393
00:20:22,520 --> 00:20:26,080
Speaker 2: Well you have to have something like that. But brains

394
00:20:26,119 --> 00:20:27,840
Speaker 2: don't do it that way. They do it another way.

395
00:20:27,880 --> 00:20:31,320
Speaker 2: And this was some very clever research in the last

396
00:20:31,359 --> 00:20:35,280
Speaker 2: twenty years that people discovered in the antarilo cortex and hippocampus.

397
00:20:35,280 --> 00:20:37,639
Speaker 2: These cells called grid cells and play cells, which actually

398
00:20:38,000 --> 00:20:42,280
Speaker 2: operate as reference frames. They are a way of neurons

399
00:20:42,320 --> 00:20:46,800
Speaker 2: to represent locations and they work differently than X, Y

400
00:20:46,840 --> 00:20:48,880
Speaker 2: and Z, so there's no origin. It's kind of really

401
00:20:48,920 --> 00:20:52,639
Speaker 2: clever how they work. The nature has discovered a different

402
00:20:52,640 --> 00:20:54,560
Speaker 2: way of doing this, so yeah, make sure you tell

403
00:20:54,640 --> 00:20:56,240
Speaker 2: us a little bit about that. Well, okay, but these

404
00:20:56,280 --> 00:20:58,160
Speaker 2: these are well known thing. It's just like grid cells,

405
00:20:58,160 --> 00:21:01,400
Speaker 2: which entronoic cord six. What they do is they these cells,

406
00:21:01,680 --> 00:21:04,000
Speaker 2: if you take a set of them, individual cells could

407
00:21:04,119 --> 00:21:06,240
Speaker 2: are not unique, and any d real sell me said,

408
00:21:06,280 --> 00:21:08,639
Speaker 2: I fired different locations in space, but if you take

409
00:21:08,640 --> 00:21:10,680
Speaker 2: a set of them, they're unique, and so you can

410
00:21:10,960 --> 00:21:14,600
Speaker 2: encode a unique location in space. And the key thing

411
00:21:14,640 --> 00:21:18,680
Speaker 2: about them is these cells automatically update as you move.

412
00:21:18,960 --> 00:21:20,960
Speaker 2: So the original grid cells are where your body is

413
00:21:20,960 --> 00:21:25,080
Speaker 2: in a room, and as you move, it's called past integration.

414
00:21:25,160 --> 00:21:27,800
Speaker 2: It says, okay, you're moving at this dis direction at

415
00:21:27,800 --> 00:21:30,720
Speaker 2: this speed, so we'll just automatically update these neurons. Is

416
00:21:30,720 --> 00:21:33,160
Speaker 2: if we know where you are right and so it's

417
00:21:33,400 --> 00:21:35,919
Speaker 2: it's what sales used to do dead reckoning. You just say, oh,

418
00:21:36,920 --> 00:21:38,960
Speaker 2: you know, I could I'm heading north for an hour

419
00:21:39,080 --> 00:21:41,479
Speaker 2: or three knots there for all these three miles in

420
00:21:41,520 --> 00:21:45,200
Speaker 2: this direction. So we know that these cells exist. They've

421
00:21:45,200 --> 00:21:47,359
Speaker 2: been well studied, people with Nobel Prize for these things.

422
00:21:47,880 --> 00:21:52,720
Speaker 2: So we speculated that the same neural mechanisms, these grid

423
00:21:52,760 --> 00:21:55,560
Speaker 2: cells and equivalents would be in the cortex at every

424
00:21:55,560 --> 00:21:59,560
Speaker 2: corticle home and sure enough they're finding that now. So

425
00:22:00,040 --> 00:22:02,320
Speaker 2: all kinds of research now they're finding in humans and

426
00:22:02,359 --> 00:22:05,320
Speaker 2: other animals that there are grid cell like structures in

427
00:22:05,359 --> 00:22:07,159
Speaker 2: cortical column And so what does that tell you? It

428
00:22:07,160 --> 00:22:09,520
Speaker 2: tells me that that's the mechanism by which the brain

429
00:22:09,640 --> 00:22:12,160
Speaker 2: uses for reference frames. And so literally, when you build

430
00:22:12,200 --> 00:22:13,840
Speaker 2: a model of something in the world, like a model

431
00:22:13,840 --> 00:22:16,240
Speaker 2: of a cup or a model of anything, it's essentially

432
00:22:16,320 --> 00:22:18,679
Speaker 2: what you're doing. You're just saying, here's the sensation, and

433
00:22:18,720 --> 00:22:22,000
Speaker 2: here it's location. Here's another sensation a different location. Here's

434
00:22:22,000 --> 00:22:24,479
Speaker 2: another sensation at a different location. You add all these together

435
00:22:24,560 --> 00:22:26,880
Speaker 2: and you get a three dimensional model. You can say,

436
00:22:26,960 --> 00:22:30,520
Speaker 2: this thing consists of these features in these locations relative

437
00:22:30,520 --> 00:22:34,160
Speaker 2: to each other. And so literally, in our head we

438
00:22:34,240 --> 00:22:37,159
Speaker 2: build models of the world that are three dimensional analogs

439
00:22:37,200 --> 00:22:41,400
Speaker 2: of the physical things we interact with. And that's why

440
00:22:41,440 --> 00:22:43,760
Speaker 2: you appears three dimensional to me. You know you're not

441
00:22:43,800 --> 00:22:46,040
Speaker 2: an image, You're a three dimensional structure because I have

442
00:22:46,080 --> 00:22:48,080
Speaker 2: a three dimensional model of humans, and I have a

443
00:22:48,080 --> 00:22:49,240
Speaker 2: special model for you, David.

444
00:22:51,040 --> 00:23:09,440
Speaker 1: Okay, great, okay. So you've got these columns in the cortex.

445
00:23:09,520 --> 00:23:12,280
Speaker 1: They're building three dimensional models or keeping track of where

446
00:23:12,320 --> 00:23:14,760
Speaker 1: your fingertips are, where your eyes are. So we've got

447
00:23:14,760 --> 00:23:18,720
Speaker 1: these different windows into the brain. You've got these data

448
00:23:18,760 --> 00:23:21,760
Speaker 1: cables coming in carrying spikes. It's all spikes, but some

449
00:23:21,800 --> 00:23:24,080
Speaker 1: of them carrying visual intraces to monitory is some touch.

450
00:23:25,320 --> 00:23:27,400
Speaker 2: Every brain doesn't know that, by the way, exactly right.

451
00:23:28,640 --> 00:23:32,840
Speaker 2: It's all the spikes exactly right. And so for any

452
00:23:32,920 --> 00:23:37,720
Speaker 2: particular column, it might only be getting a subset of

453
00:23:37,760 --> 00:23:40,720
Speaker 2: those tell us about that, right, right? Well, any well,

454
00:23:40,760 --> 00:23:41,960
Speaker 2: I'm not shore going to be a subset.

455
00:23:42,119 --> 00:23:44,239
Speaker 1: But what I mean is if I if I am

456
00:23:44,240 --> 00:23:45,800
Speaker 1: a cortical column that happens to be sitting in the

457
00:23:45,880 --> 00:23:48,200
Speaker 1: visual cortex, that I happen to be getting visual information,

458
00:23:48,320 --> 00:23:50,600
Speaker 1: but I'm not getting auditorial So.

459
00:23:50,560 --> 00:23:52,720
Speaker 2: There's a real One of the first things we had

460
00:23:52,720 --> 00:23:55,040
Speaker 2: to address with this series is why does the world

461
00:23:55,080 --> 00:23:59,280
Speaker 2: appear unified? Right? I don't feel like you know, I

462
00:23:59,480 --> 00:24:01,760
Speaker 2: I don't feel like, oh, I'm touching something with my

463
00:24:01,880 --> 00:24:03,680
Speaker 2: hands and I'm looking at something else in my eyes.

464
00:24:03,720 --> 00:24:05,639
Speaker 2: It's all one thing. There's this cup, right, and I

465
00:24:05,640 --> 00:24:07,439
Speaker 2: feel the warmth of it, and I know it. I mean,

466
00:24:07,440 --> 00:24:09,239
Speaker 2: it's one thing. It's and yet we have all these

467
00:24:09,280 --> 00:24:11,639
Speaker 2: different models. So it turns out you have models of

468
00:24:11,640 --> 00:24:14,560
Speaker 2: cups and that are taxile models. They're based on how

469
00:24:14,800 --> 00:24:16,760
Speaker 2: how it feels. You have models of how it looks.

470
00:24:17,520 --> 00:24:19,199
Speaker 2: You might even have a model how it sounds like

471
00:24:19,240 --> 00:24:21,800
Speaker 2: this particular ceramic cup. I have an expectation what it

472
00:24:21,920 --> 00:24:23,760
Speaker 2: sound like. I put on this account here my different

473
00:24:24,400 --> 00:24:29,080
Speaker 2: ceramic counter. And yet these models are they're all independent,

474
00:24:29,119 --> 00:24:32,120
Speaker 2: but they're not completely in pandit so there's these long

475
00:24:32,200 --> 00:24:34,960
Speaker 2: range connections in the cortex. They go from all different

476
00:24:34,960 --> 00:24:36,159
Speaker 2: sides to the left side of the brain and the right

477
00:24:36,200 --> 00:24:37,960
Speaker 2: side of the brain and all over the place. There's lots

478
00:24:38,000 --> 00:24:41,119
Speaker 2: of different types. What they're essentially doing is they're voting.

479
00:24:41,600 --> 00:24:44,200
Speaker 2: They're all saying, like one my finger says, I think

480
00:24:44,200 --> 00:24:46,240
Speaker 2: I'm touching something that feels like a cup, and I

481
00:24:46,280 --> 00:24:48,360
Speaker 2: may not be certain. Another thing, I have something too

482
00:24:48,440 --> 00:24:50,320
Speaker 2: that's I'm not really certain, and the nice thing and

483
00:24:50,359 --> 00:24:52,399
Speaker 2: they very quickly are reaching the set. The only thing

484
00:24:52,400 --> 00:24:55,119
Speaker 2: that makes sense for all our input is we're all

485
00:24:55,160 --> 00:24:57,720
Speaker 2: looking at the same object. And so there's like across

486
00:24:57,840 --> 00:25:02,399
Speaker 2: these long range connections, it'll into a percept that's what

487
00:25:02,440 --> 00:25:05,960
Speaker 2: you perceive. You don't actually normally perceive the individual sensations

488
00:25:05,960 --> 00:25:07,680
Speaker 2: from your eye or your fingers. You just say, I'm

489
00:25:07,680 --> 00:25:11,159
Speaker 2: holding this cup in my hand, and it's one percept.

490
00:25:11,280 --> 00:25:13,800
Speaker 2: And so it's these long raine connections and how these

491
00:25:13,840 --> 00:25:16,960
Speaker 2: columns vote all the time. This is why I can

492
00:25:17,000 --> 00:25:21,399
Speaker 2: flash an image in front of your eye and say, okay,

493
00:25:21,440 --> 00:25:23,800
Speaker 2: well each column is looking at part of that image.

494
00:25:23,960 --> 00:25:27,040
Speaker 2: Who decides what the whole image right? And by the way,

495
00:25:27,080 --> 00:25:28,600
Speaker 2: I don't even have time to move my eyes. Once

496
00:25:28,600 --> 00:25:30,399
Speaker 2: I've run into objects, I don't have to move my

497
00:25:30,440 --> 00:25:32,760
Speaker 2: eyes to recognize them. Man, what we call a flash inference.

498
00:25:33,800 --> 00:25:35,600
Speaker 2: The reason is because each part of the court to

499
00:25:35,720 --> 00:25:38,719
Speaker 2: visual cortext has a hypothesis about what it might be seeing,

500
00:25:39,000 --> 00:25:40,920
Speaker 2: and they vote, and the only thing that makes sense

501
00:25:40,960 --> 00:25:43,520
Speaker 2: is the final thing they agree upon. So I have

502
00:25:43,600 --> 00:25:46,000
Speaker 2: to learn by moving my eyes, by tending to different

503
00:25:46,000 --> 00:25:48,800
Speaker 2: things and my fingers. But I don't always have to

504
00:25:48,840 --> 00:25:51,359
Speaker 2: infer or recognize things by movement. I don't always have to.

505
00:25:51,480 --> 00:25:53,320
Speaker 2: I can just flash an image in front of you

506
00:25:53,520 --> 00:25:54,800
Speaker 2: and you say, I know what that is, and you

507
00:25:54,840 --> 00:25:57,119
Speaker 2: don't have time to move rise. This folded a lot

508
00:25:57,119 --> 00:25:59,199
Speaker 2: of vision researchers for many years because they assume that

509
00:25:59,320 --> 00:26:02,480
Speaker 2: the movement was necessary because I can flash an image

510
00:26:02,480 --> 00:26:04,520
Speaker 2: in front of you. But you can't learn that way.

511
00:26:04,720 --> 00:26:08,160
Speaker 2: You have to learn by attending to different things, quite right,

512
00:26:08,600 --> 00:26:10,480
Speaker 2: Just so it's clear to the audience.

513
00:26:10,520 --> 00:26:13,840
Speaker 1: So this issue about voting, it's not that they're all

514
00:26:13,840 --> 00:26:16,920
Speaker 1: submitting their votes to some central agency. It's that they're

515
00:26:16,920 --> 00:26:22,120
Speaker 1: all talking with one another simultaneously, simultaneously. And something about

516
00:26:22,119 --> 00:26:25,040
Speaker 1: the spike patterns holds into shape.

517
00:26:24,760 --> 00:26:27,199
Speaker 2: Right, right, well, we know exactly how this occurs. We

518
00:26:27,240 --> 00:26:30,960
Speaker 2: have models of it, and we've simulated and matches of neuroscience.

519
00:26:31,840 --> 00:26:33,320
Speaker 2: It's a little it takes a little while for people

520
00:26:33,359 --> 00:26:34,560
Speaker 2: to get the sense of it. You're right, there's no

521
00:26:34,720 --> 00:26:38,919
Speaker 2: central voting tally. It's like it's and I don't have

522
00:26:39,040 --> 00:26:40,399
Speaker 2: all the commns, don't have to talk to all the

523
00:26:40,400 --> 00:26:42,040
Speaker 2: other comms. It turns out they only have to stalk

524
00:26:42,040 --> 00:26:43,800
Speaker 2: to a few other commns as long as everyone talks

525
00:26:43,840 --> 00:26:45,800
Speaker 2: to somebody and the whole thing is connected, so they

526
00:26:45,800 --> 00:26:49,760
Speaker 2: don't have to like in Zilian connections. But it's it's

527
00:26:49,800 --> 00:26:53,199
Speaker 2: more like you have a neuro you know, magic neurons

528
00:26:53,200 --> 00:26:57,520
Speaker 2: are spiking, and in I have I have five thousand

529
00:26:57,520 --> 00:27:00,639
Speaker 2: neurons that representing what I'm seeing. That's not that man actually,

530
00:27:00,640 --> 00:27:03,400
Speaker 2: so five thousand neurons and in the brain. We're getting

531
00:27:03,400 --> 00:27:07,159
Speaker 2: a little technical here. Activations are typically sparse, meaning of

532
00:27:07,200 --> 00:27:10,919
Speaker 2: those five thousand cells, maybe only two percent or one

533
00:27:10,960 --> 00:27:12,960
Speaker 2: hundred are active at any point in time. The others

534
00:27:12,960 --> 00:27:18,600
Speaker 2: are silent. So I'm representing something by saying there's one

535
00:27:18,640 --> 00:27:21,359
Speaker 2: hundred neurons active out of five thousand. Now, if I

536
00:27:21,600 --> 00:27:25,560
Speaker 2: wasn't certain, I might say, oh, well, let's do this.

537
00:27:25,960 --> 00:27:27,560
Speaker 2: I'm going to say it could be object day, it

538
00:27:27,560 --> 00:27:29,239
Speaker 2: could be object being, could be object C. And I'm

539
00:27:29,240 --> 00:27:31,000
Speaker 2: gonna activate them all the same time. So now I

540
00:27:31,040 --> 00:27:34,959
Speaker 2: have three hundred neurons out of five hundred they're simultaneously active.

541
00:27:35,280 --> 00:27:39,080
Speaker 2: Now that might seem confusing, but it isn't. No trouble

542
00:27:39,119 --> 00:27:42,159
Speaker 2: is this and everybody's doing the same thing. They're all

543
00:27:42,280 --> 00:27:46,160
Speaker 2: doing multiple hy positives, and it very quickly says you're

544
00:27:46,200 --> 00:27:49,600
Speaker 2: supporting this positive and you're supporting this po. It happens simultaneously.

545
00:27:49,680 --> 00:27:52,000
Speaker 2: No one has to go through so early. There's no

546
00:27:52,160 --> 00:27:54,199
Speaker 2: like counting the vote. So let's try this like positive

547
00:27:54,200 --> 00:27:56,440
Speaker 2: in this it all settles very very quickly.

548
00:27:57,840 --> 00:28:00,199
Speaker 1: It's kind of cool thing if you thought about what

549
00:28:00,359 --> 00:28:05,159
Speaker 1: happens when you settle on a hypothesis and then you switch.

550
00:28:05,200 --> 00:28:07,560
Speaker 1: For example, looking at the Neckar cube, this cube made

551
00:28:07,560 --> 00:28:08,760
Speaker 1: out of it, yes, twelve lines.

552
00:28:09,040 --> 00:28:09,480
Speaker 2: What you know?

553
00:28:09,680 --> 00:28:11,240
Speaker 1: You see it one way, then you see it the

554
00:28:11,280 --> 00:28:14,239
Speaker 1: other way? What is it that allows it to switch? All?

555
00:28:14,320 --> 00:28:17,520
Speaker 2: Right? Now, a Necker cube is a two dimensional image, right,

556
00:28:17,560 --> 00:28:21,440
Speaker 2: it's a two dimensional image of a three dimensional wireframe

557
00:28:21,520 --> 00:28:24,679
Speaker 2: cube or something like that. Right, And so it's not

558
00:28:24,680 --> 00:28:28,480
Speaker 2: three dimensional. It's really two dimensional, but your brain wants

559
00:28:28,480 --> 00:28:31,280
Speaker 2: to make it three dimensional, right because it doesn't know

560
00:28:31,359 --> 00:28:34,680
Speaker 2: two dimensional things that look like that, And so everything

561
00:28:34,720 --> 00:28:36,880
Speaker 2: we try to do fits into our models, right, Right.

562
00:28:36,960 --> 00:28:38,680
Speaker 2: We don't say, oh, that's a two dimensional image. It

563
00:28:38,720 --> 00:28:40,200
Speaker 2: can't be a cube. No, you says, oh, no, that's

564
00:28:40,200 --> 00:28:41,640
Speaker 2: got to be a cube, because I know cubes. I

565
00:28:41,640 --> 00:28:43,719
Speaker 2: don't think it looks like that's not a cube. So

566
00:28:43,840 --> 00:28:46,440
Speaker 2: it wants to settle on a hypothesis. Is like, okay,

567
00:28:46,480 --> 00:28:48,400
Speaker 2: well this corner is in front of that corner, and

568
00:28:48,400 --> 00:28:49,920
Speaker 2: this corner is behind that corner, the corner to the

569
00:28:50,000 --> 00:28:51,760
Speaker 2: left of that corner. It just has to do that

570
00:28:51,800 --> 00:28:53,800
Speaker 2: to fit its models. That's right.

571
00:28:53,960 --> 00:28:56,600
Speaker 1: But why doesn't it land on a hypothesis and stick there.

572
00:28:56,680 --> 00:28:59,960
Speaker 2: Well, I don't really know, but there's other people hypothesis

573
00:29:00,200 --> 00:29:03,640
Speaker 2: about this is that the evidence goes both ways, right,

574
00:29:03,640 --> 00:29:07,200
Speaker 2: there's multiple hypositive and so neurons have a way of

575
00:29:07,200 --> 00:29:09,600
Speaker 2: getting tired about what they're doing. After a while, they say,

576
00:29:09,640 --> 00:29:12,480
Speaker 2: you know, literally, they have a way of they say,

577
00:29:12,480 --> 00:29:14,600
Speaker 2: you know, I'm not going to keep finding on this forever.

578
00:29:14,720 --> 00:29:16,640
Speaker 2: You know, things are changing the world. We don't just

579
00:29:16,680 --> 00:29:21,520
Speaker 2: get stuck. So there's various speculated mechanisms BEHUWD neurons, and

580
00:29:21,560 --> 00:29:24,560
Speaker 2: it's been observed. We'll sort of, you know, say okay,

581
00:29:24,600 --> 00:29:25,960
Speaker 2: I'll be active a little while and then I'm going

582
00:29:26,040 --> 00:29:28,400
Speaker 2: to stop U. Lets someone else try something, right.

583
00:29:28,400 --> 00:29:29,920
Speaker 1: Yeah, it, But what it means is that the other

584
00:29:30,000 --> 00:29:33,160
Speaker 1: hypothesis has to be kept alive somewhere somehow.

585
00:29:33,200 --> 00:29:35,720
Speaker 2: Well it maybe not. Maybe just like I have this

586
00:29:35,840 --> 00:29:38,560
Speaker 2: hypothesis that locked in on it, and now I'm going

587
00:29:38,640 --> 00:29:41,760
Speaker 2: to say that's no longer possible. Just go back to

588
00:29:41,800 --> 00:29:44,760
Speaker 2: square one. What is possible? You know? So it's not

589
00:29:44,920 --> 00:29:48,360
Speaker 2: like I have these two images in my head conceptual

590
00:29:48,600 --> 00:29:51,120
Speaker 2: or perceptually. You don't feel that way, right, You only

591
00:29:51,160 --> 00:29:53,080
Speaker 2: one or the other? Do you lock in the one?

592
00:29:53,320 --> 00:29:57,160
Speaker 2: The other is forgotten? But then if i'd say disabled

593
00:29:57,200 --> 00:29:59,320
Speaker 2: the first hypothesis, We're not gonna allow that be anymore.

594
00:29:59,520 --> 00:30:02,280
Speaker 2: Then it's okay, what's possible. This one's possible, I'll switch

595
00:30:02,320 --> 00:30:05,400
Speaker 2: to that one. It's not like they're both active. One's

596
00:30:05,440 --> 00:30:07,880
Speaker 2: active and then it gets tired and then the other, well,

597
00:30:08,120 --> 00:30:08,520
Speaker 2: I work.

598
00:30:09,480 --> 00:30:11,920
Speaker 1: So coming back to the main thing, one part that

599
00:30:11,960 --> 00:30:14,560
Speaker 1: I want to return to is just this issue that

600
00:30:15,080 --> 00:30:20,080
Speaker 1: a particular column might only be receiving touch information. Another

601
00:30:20,080 --> 00:30:22,960
Speaker 1: call might be receiving only auditory information, and so on.

602
00:30:23,080 --> 00:30:26,520
Speaker 2: Well, they build independent models, right they. I could have

603
00:30:26,560 --> 00:30:28,520
Speaker 2: a tactle model of an object, a visual model of

604
00:30:28,520 --> 00:30:30,880
Speaker 2: an object, right, they're not the same. The visual model

605
00:30:30,840 --> 00:30:32,760
Speaker 2: of the object will have color, perhaps the tact will

606
00:30:32,760 --> 00:30:37,320
Speaker 2: have temperature and texture and things like that. So they're

607
00:30:37,360 --> 00:30:40,280
Speaker 2: different models. But because they can vote, you have a

608
00:30:40,320 --> 00:30:42,400
Speaker 2: single percept of it. Yeah, okay.

609
00:30:42,440 --> 00:30:44,320
Speaker 1: And one of the things that's important here, which of

610
00:30:44,320 --> 00:30:46,280
Speaker 1: course you have, so you and I both emphasize this

611
00:30:46,320 --> 00:30:47,959
Speaker 1: is a lot in our books, is that all we

612
00:30:48,000 --> 00:30:50,640
Speaker 1: are ever seeing is our model of the world, right,

613
00:30:50,800 --> 00:30:53,960
Speaker 1: and so we don't have any direct access to what's

614
00:30:54,080 --> 00:30:54,960
Speaker 1: actually out there.

615
00:30:55,000 --> 00:30:56,160
Speaker 2: And so the fact.

616
00:30:57,480 --> 00:31:00,600
Speaker 1: You mentioned earlier the binding problem, if you mentioned it

617
00:31:00,640 --> 00:31:02,440
Speaker 1: by name. But the binding problem is this issue that

618
00:31:02,480 --> 00:31:04,480
Speaker 1: when the coffee cup is here and it's moving, how

619
00:31:04,480 --> 00:31:06,160
Speaker 1: come the color doesn't bleed off the cup?

620
00:31:06,200 --> 00:31:07,960
Speaker 2: And how come it seems like one thing and so on.

621
00:31:08,160 --> 00:31:11,200
Speaker 2: Buying problem is a poorly defined problem. Exactly. It means

622
00:31:11,240 --> 00:31:12,840
Speaker 2: a lot of different things, a lot of different people.

623
00:31:12,880 --> 00:31:15,120
Speaker 2: So you gotta be really careful a say, oh, I

624
00:31:15,360 --> 00:31:18,640
Speaker 2: let's talk about the binding problem, I might have a

625
00:31:18,640 --> 00:31:20,280
Speaker 2: different perception of what the binding problem is.

626
00:31:20,640 --> 00:31:20,880
Speaker 1: To me.

627
00:31:21,120 --> 00:31:23,200
Speaker 2: The binding problem is the one I've already discussed, which

628
00:31:23,280 --> 00:31:30,200
Speaker 2: is you have these different sensory inputs that but somehow

629
00:31:30,200 --> 00:31:33,080
Speaker 2: they lead to a single perscept and you can switch

630
00:31:33,160 --> 00:31:35,560
Speaker 2: back and forth. It's like, I don't It's like, how

631
00:31:35,600 --> 00:31:38,040
Speaker 2: do I bring these things together? How do I say

632
00:31:38,120 --> 00:31:41,080
Speaker 2: these are all the same thing? And people used to

633
00:31:41,120 --> 00:31:43,040
Speaker 2: think in the binding problems like, oh, if I have

634
00:31:43,120 --> 00:31:46,160
Speaker 2: the auditory cortex and the visual cortext and somautter centric

635
00:31:46,160 --> 00:31:50,400
Speaker 2: cortex touch, then they must all project to someplace where

636
00:31:50,480 --> 00:31:54,200
Speaker 2: they are binding together into a single model. And we

637
00:31:54,360 --> 00:31:57,440
Speaker 2: flip that on its head. They don't bind. They bind

638
00:31:57,480 --> 00:32:00,560
Speaker 2: together through just long range connections. But there's no place

639
00:32:00,600 --> 00:32:03,160
Speaker 2: I have to do that. There's no nobody's sitting on

640
00:32:03,200 --> 00:32:05,840
Speaker 2: top of it and saying, hey, what's your vote? What's

641
00:32:05,840 --> 00:32:08,200
Speaker 2: your vote? Now? It's just like, so there, we don't

642
00:32:08,240 --> 00:32:11,760
Speaker 2: need a model that incorporates all the aspects of objects.

643
00:32:11,840 --> 00:32:15,680
Speaker 2: We have independent models that we can invoke as needed,

644
00:32:15,800 --> 00:32:19,520
Speaker 2: and they all they all vote to reach common consensus.

645
00:32:19,520 --> 00:32:21,960
Speaker 2: So I have no problems navigating, you know, doing things

646
00:32:22,040 --> 00:32:23,840
Speaker 2: in the dark. I have no troubles doing things just

647
00:32:23,840 --> 00:32:25,680
Speaker 2: by vision. I have no I can do things sometimes

648
00:32:25,680 --> 00:32:28,720
Speaker 2: of audition, you know, like I know the same things

649
00:32:28,720 --> 00:32:30,800
Speaker 2: are going on. I have the same model in the world, right.

650
00:32:31,000 --> 00:32:33,120
Speaker 2: You know, if I if I'm walking at night between

651
00:32:33,120 --> 00:32:36,239
Speaker 2: my bed and the bathroom and it's pitch black, I

652
00:32:36,280 --> 00:32:38,400
Speaker 2: still have the same model in the house. I still

653
00:32:38,440 --> 00:32:40,160
Speaker 2: know where the door is going to be and everything else.

654
00:32:40,240 --> 00:32:42,560
Speaker 2: You know, if I can do with touch for the vision. Right,

655
00:32:43,200 --> 00:32:45,440
Speaker 2: So there isn't a central model that says, here's a

656
00:32:45,480 --> 00:32:47,240
Speaker 2: model of my house of touch and vision hearing. It

657
00:32:47,280 --> 00:32:49,320
Speaker 2: says all these independent models. Right.

658
00:32:49,680 --> 00:32:52,280
Speaker 1: And now the reason you called your book a thousand brains,

659
00:32:52,360 --> 00:32:53,200
Speaker 1: we call this ypods.

660
00:32:53,240 --> 00:32:56,400
Speaker 2: It's one thousand brains theary theory, right, is.

661
00:32:57,960 --> 00:33:01,120
Speaker 1: Precisely because you've got all these cortical calls and they're

662
00:33:01,280 --> 00:33:03,360
Speaker 1: each making a little model of the world, and they're

663
00:33:03,360 --> 00:33:06,000
Speaker 1: all talking to one another. Right, So you know, hi,

664
00:33:06,680 --> 00:33:08,760
Speaker 1: this is this feels like coffee cup. This looks like

665
00:33:08,800 --> 00:33:10,720
Speaker 1: coffee cup. It sounds like coffee cup when it plays down.

666
00:33:10,760 --> 00:33:13,360
Speaker 1: This is the temperature of coffee cup, and so and

667
00:33:13,440 --> 00:33:15,040
Speaker 1: so these are all talking with one another.

668
00:33:15,600 --> 00:33:18,240
Speaker 2: So so the reason they call the thousand brains. Is

669
00:33:18,280 --> 00:33:23,440
Speaker 2: that each cortical column is doing what the entire brain

670
00:33:23,520 --> 00:33:25,800
Speaker 2: is doing. Right, each quarter column is a senior model

671
00:33:25,840 --> 00:33:30,320
Speaker 2: learning system. And and when we ask where is a

672
00:33:30,360 --> 00:33:32,400
Speaker 2: model of something? We've been talking about this, where is

673
00:33:32,400 --> 00:33:35,360
Speaker 2: the model of the skull or the microphone or whatever.

674
00:33:36,160 --> 00:33:39,320
Speaker 2: So many things we know, where is that model? It's not,

675
00:33:39,600 --> 00:33:43,040
Speaker 2: it's it's in many different places. So there's a thousand

676
00:33:43,160 --> 00:33:46,760
Speaker 2: models of coffee cops, a thousand models. You don't perceive that,

677
00:33:47,680 --> 00:33:51,680
Speaker 2: but they exist. And and so it's like it's it

678
00:33:51,760 --> 00:33:54,600
Speaker 2: was really trying to capture that original idea that cortical

679
00:33:54,640 --> 00:33:57,080
Speaker 2: columns are common and that and that there's all these

680
00:33:57,120 --> 00:33:59,640
Speaker 2: different models out there that are different and they can

681
00:33:59,720 --> 00:34:02,560
Speaker 2: vote one day, they vote to reach a consensus.

682
00:34:02,680 --> 00:34:05,160
Speaker 1: Yeah, and it's certainly consistent with the idea that you know,

683
00:34:05,160 --> 00:34:07,560
Speaker 1: for example, if someone is born blind and then visual

684
00:34:07,640 --> 00:34:10,359
Speaker 1: cortex gets taken over by hearing and touches on, they

685
00:34:10,360 --> 00:34:13,239
Speaker 1: are better at hearing in touch presuming because they just

686
00:34:13,239 --> 00:34:14,800
Speaker 1: have a lot more real estates.

687
00:34:14,760 --> 00:34:17,040
Speaker 2: Voted right, right, or real estate is a lot more

688
00:34:17,080 --> 00:34:21,759
Speaker 2: practice too, right right. So it's amazing how flexible it is.

689
00:34:21,960 --> 00:34:26,399
Speaker 1: Yes, given your model of the brain. Let's talk about

690
00:34:26,440 --> 00:34:28,560
Speaker 1: AI and what you think is going on currently with

691
00:34:28,800 --> 00:34:31,120
Speaker 1: LLLMS and what that is missing.

692
00:34:31,400 --> 00:34:35,000
Speaker 2: Right, LMS flipp is interesting. Well, let me start with

693
00:34:35,040 --> 00:34:38,920
Speaker 2: the criticism AI in general. Okay, AI has always been

694
00:34:38,960 --> 00:34:42,279
Speaker 2: focused on what they call benchmarks, like how well can

695
00:34:42,320 --> 00:34:45,560
Speaker 2: you solve problems? So, how well came this system recognized images?

696
00:34:45,600 --> 00:34:47,360
Speaker 2: How well can play chess? How well can play go?

697
00:34:47,480 --> 00:34:49,839
Speaker 2: How well I can translate from one language to the other.

698
00:34:50,000 --> 00:34:52,280
Speaker 2: And you have all these benchmarks, and everyone competes against

699
00:34:52,280 --> 00:34:54,600
Speaker 2: these benchmarks that they're kind of diverse all over the place.

700
00:34:55,239 --> 00:34:58,680
Speaker 2: That's the wrong way to think about it. When we

701
00:34:59,120 --> 00:35:02,640
Speaker 2: let's use computer as an analogy. When we say something

702
00:35:02,640 --> 00:35:05,000
Speaker 2: as a computer, we don't based on what it's doing.

703
00:35:05,080 --> 00:35:07,960
Speaker 2: We based on how it works. Alan Turing and John

704
00:35:08,000 --> 00:35:10,680
Speaker 2: Fan Nordmann define what we now call a universal turning machine,

705
00:35:10,680 --> 00:35:13,200
Speaker 2: which is like, okay, if a system has memory and

706
00:35:13,640 --> 00:35:18,120
Speaker 2: a processor, and the memory has data and instructions, and

707
00:35:18,160 --> 00:35:20,200
Speaker 2: you can change the instructions and change the data, it

708
00:35:20,200 --> 00:35:23,120
Speaker 2: can do anything. And that is a computer. So I

709
00:35:23,160 --> 00:35:25,560
Speaker 2: can say my toaster is a computer, even though it's

710
00:35:25,560 --> 00:35:28,120
Speaker 2: a very limited computer, because it has one of those

711
00:35:28,120 --> 00:35:32,080
Speaker 2: things inside. Right, if it was hard coded with springs

712
00:35:32,080 --> 00:35:33,799
Speaker 2: and wires and stuff, it wouldn't be a computer. But

713
00:35:33,920 --> 00:35:37,080
Speaker 2: because it has a little microprocessor has those definitions, it's

714
00:35:37,120 --> 00:35:39,480
Speaker 2: a computer. So that's how we do it in the

715
00:35:39,480 --> 00:35:42,239
Speaker 2: computer world. We say, these are the functions that it

716
00:35:42,239 --> 00:35:44,080
Speaker 2: has to perform, and you can apply it to big problems,

717
00:35:44,080 --> 00:35:46,040
Speaker 2: little problems, different types of problems all over the place.

718
00:35:46,080 --> 00:35:48,520
Speaker 2: And AI we've been focused on this idea that oh

719
00:35:48,560 --> 00:35:50,719
Speaker 2: a benchmarks, you know, and we always want to be

720
00:35:50,719 --> 00:35:53,480
Speaker 2: beat some human Well, like a dog. Almost everyone who

721
00:35:53,520 --> 00:35:56,240
Speaker 2: has the dog says it's intelligent, right, but doesn't have language,

722
00:35:56,280 --> 00:35:58,000
Speaker 2: it doesn't play standswer, it doesn't play go. But why

723
00:35:58,040 --> 00:35:59,920
Speaker 2: do we say it's intelligent because we can tell the

724
00:36:00,080 --> 00:36:01,920
Speaker 2: that dog has an eternal model of the world. It's

725
00:36:01,960 --> 00:36:03,640
Speaker 2: kind of like my internal the model. He knows where

726
00:36:03,640 --> 00:36:05,160
Speaker 2: the door is, it knows how to get go on

727
00:36:05,160 --> 00:36:09,080
Speaker 2: the walk. And so why why focus on this issue

728
00:36:09,120 --> 00:36:10,919
Speaker 2: of like, well, it's not intelligent because it doesn't play

729
00:36:10,960 --> 00:36:14,680
Speaker 2: go better than the best human player. So I think

730
00:36:14,719 --> 00:36:17,080
Speaker 2: part of the problem was that people didn't know how

731
00:36:17,120 --> 00:36:19,040
Speaker 2: brains worked, and so if you don't, what are you

732
00:36:19,080 --> 00:36:21,719
Speaker 2: going to do? Right? We don't know it. We we

733
00:36:21,840 --> 00:36:23,960
Speaker 2: know enough to build this stuff. So I think in

734
00:36:24,000 --> 00:36:25,759
Speaker 2: the future that's what's going to be. We're going to

735
00:36:25,840 --> 00:36:28,440
Speaker 2: say AI systems don't have to be like humans. They

736
00:36:28,440 --> 00:36:30,120
Speaker 2: don't even have to do the same things humans do.

737
00:36:30,360 --> 00:36:32,000
Speaker 2: Some of them are going to be very dedicated. It's

738
00:36:32,120 --> 00:36:33,920
Speaker 2: very focused tasks, and we're going to be very broad

739
00:36:33,960 --> 00:36:37,120
Speaker 2: So might be you know, engineers building space stations, all

740
00:36:37,200 --> 00:36:38,880
Speaker 2: this huge rider, but they're all going to work on

741
00:36:38,920 --> 00:36:43,160
Speaker 2: the same principles that biology has discovered. Today's AI doesn't

742
00:36:43,160 --> 00:36:45,279
Speaker 2: work on those principles, you know, most of it. If

743
00:36:45,280 --> 00:36:47,800
Speaker 2: you talk about the large language models, these are transform

744
00:36:47,800 --> 00:36:51,160
Speaker 2: well models. We feed in a string of tokens basically

745
00:36:51,360 --> 00:36:54,920
Speaker 2: words or wordlike things, and it just learns the structure

746
00:36:54,960 --> 00:36:57,680
Speaker 2: of that string, and it's very good at what it does.

747
00:36:57,960 --> 00:37:00,840
Speaker 2: But there's no inherent knowledge of the actual the world.

748
00:37:01,160 --> 00:37:03,359
Speaker 2: It doesn't have a three dimensional models of the world.

749
00:37:03,360 --> 00:37:05,759
Speaker 2: It doesn't if someone's written about it, it'll tell you

750
00:37:05,800 --> 00:37:09,160
Speaker 2: about it, but it can't experience it itself. So you

751
00:37:09,200 --> 00:37:11,239
Speaker 2: couldn't send in one of these AI systems down the

752
00:37:11,280 --> 00:37:14,440
Speaker 2: space and say, you know, go to Mars explore and

753
00:37:14,600 --> 00:37:16,279
Speaker 2: see what's out there that we can build things with

754
00:37:16,400 --> 00:37:18,239
Speaker 2: and here's some tools and start building a structure. It

755
00:37:18,280 --> 00:37:20,200
Speaker 2: just no why they're gonna do this though it's not

756
00:37:20,200 --> 00:37:23,319
Speaker 2: gonna happen, but to contact. The tools we're working on

757
00:37:23,560 --> 00:37:26,279
Speaker 2: can do that. That's what humans do, and that's the

758
00:37:26,360 --> 00:37:30,080
Speaker 2: promise of AI. It's not just you know, targeting things

759
00:37:30,120 --> 00:37:32,440
Speaker 2: that humans can do, like high level things like you know,

760
00:37:32,480 --> 00:37:35,319
Speaker 2: translating language or writing poems or things like that. It's

761
00:37:35,360 --> 00:37:37,080
Speaker 2: really how do you build the system to understands the

762
00:37:37,080 --> 00:37:39,239
Speaker 2: world and knows how to act in that world. And

763
00:37:39,600 --> 00:37:54,760
Speaker 2: that's the key.

764
00:37:56,640 --> 00:37:58,040
Speaker 1: One of the things you wrote in your book that

765
00:37:58,080 --> 00:38:01,000
Speaker 1: I thought was great was, uh, you address this issue

766
00:38:01,040 --> 00:38:04,040
Speaker 1: of the existential threat of AI that a lot of

767
00:38:04,040 --> 00:38:07,040
Speaker 1: people are banging on about and you don't think it's

768
00:38:07,040 --> 00:38:07,399
Speaker 1: a threat.

769
00:38:07,480 --> 00:38:09,440
Speaker 2: I don't think it's a threat. I mean you have to,

770
00:38:09,719 --> 00:38:12,080
Speaker 2: you have to tease it apart because so many people

771
00:38:12,200 --> 00:38:14,960
Speaker 2: like there's different existential threats, but you know that I

772
00:38:15,000 --> 00:38:17,439
Speaker 2: one is called the alignment problem, Like all these AI

773
00:38:17,520 --> 00:38:19,680
Speaker 2: agents are gonna you know, you're gonna tell what to do,

774
00:38:19,719 --> 00:38:22,799
Speaker 2: but it won't be aligned with our values. And I'm

775
00:38:22,840 --> 00:38:26,600
Speaker 2: just saying they don't have any values and it's just

776
00:38:26,640 --> 00:38:28,960
Speaker 2: so far from reality. I just if once you understand

777
00:38:28,960 --> 00:38:30,840
Speaker 2: how brains work, you said, like I'm doing any of

778
00:38:30,880 --> 00:38:33,560
Speaker 2: that stuff. It's it's hard to it's hard for me

779
00:38:33,600 --> 00:38:36,160
Speaker 2: to give a succinct answer to this. But I don't

780
00:38:36,239 --> 00:38:39,560
Speaker 2: think that today's AI systems have any of these problems.

781
00:38:40,239 --> 00:38:42,399
Speaker 2: They're not gonna run away. They're not They're not gonna

782
00:38:42,400 --> 00:38:45,480
Speaker 2: have their own desires. They're not gonna say, hey, I'm awake,

783
00:38:45,600 --> 00:38:47,799
Speaker 2: I need to survive, you know.

784
00:38:48,880 --> 00:38:52,080
Speaker 1: Because because these current large language models are just statistical

785
00:38:52,120 --> 00:38:55,720
Speaker 1: parrots that are taking much language and spinning language back out.

786
00:38:55,880 --> 00:38:57,960
Speaker 2: Right, and you can apply they'll prime the robotics and

787
00:38:58,000 --> 00:39:02,960
Speaker 2: other things, but they're gonna be still so statistical parents exactly.

788
00:39:03,160 --> 00:39:05,719
Speaker 2: But and and by the way, they lack the human brain.

789
00:39:05,760 --> 00:39:07,640
Speaker 2: We talked earlier, the new or corts is the biggest

790
00:39:07,640 --> 00:39:08,640
Speaker 2: part of the brain, but we have a lot of

791
00:39:08,680 --> 00:39:11,920
Speaker 2: other parts of the brain, and our emotional centers and

792
00:39:12,320 --> 00:39:14,480
Speaker 2: how much of what makes us humans, our drives and

793
00:39:14,520 --> 00:39:18,279
Speaker 2: motivations are mostly not the near cortex. Right. There are

794
00:39:18,400 --> 00:39:21,560
Speaker 2: these other things. And if you provided an AI system

795
00:39:21,600 --> 00:39:24,000
Speaker 2: with those other things, I might start worrying about it.

796
00:39:24,080 --> 00:39:27,879
Speaker 2: But if you're just trying to model stuff, it's it's

797
00:39:27,920 --> 00:39:32,640
Speaker 2: no threat. It's we just assume that some AI system,

798
00:39:32,680 --> 00:39:35,399
Speaker 2: because it can spew back language, is going to think

799
00:39:35,480 --> 00:39:37,719
Speaker 2: like us and be like us and have our same motivations.

800
00:39:37,760 --> 00:39:38,640
Speaker 2: Nothing like it at all.

801
00:39:38,760 --> 00:39:41,680
Speaker 1: So tell us about the thousand brains projects and how

802
00:39:41,719 --> 00:39:42,640
Speaker 1: you're gonna make this happen.

803
00:39:42,760 --> 00:39:45,680
Speaker 2: Right, So, we kind of been working on this theory

804
00:39:45,719 --> 00:39:48,560
Speaker 2: for decades really, and maybe five or six years ago

805
00:39:48,600 --> 00:39:50,560
Speaker 2: we really had some breakthroughs and sort of all came

806
00:39:50,600 --> 00:39:53,400
Speaker 2: together and then we said, well, I always thought that

807
00:39:53,440 --> 00:39:55,280
Speaker 2: this is the way we're going to build tuly intelligent machines.

808
00:39:55,320 --> 00:39:57,040
Speaker 2: And this is at the same time as deep learning

809
00:39:57,080 --> 00:39:59,480
Speaker 2: and transformers are taken off and all this excitement about it.

810
00:40:00,440 --> 00:40:03,120
Speaker 2: But that didn't distract us. We said, Okay, let's see

811
00:40:03,120 --> 00:40:04,680
Speaker 2: if we can start building this stuff. So we for

812
00:40:04,680 --> 00:40:06,719
Speaker 2: a couple of years we had a small team that

813
00:40:06,880 --> 00:40:10,399
Speaker 2: was trying to implement the thousand brains theory, modeling quarter

814
00:40:10,440 --> 00:40:13,600
Speaker 2: u cooms, the voting, all this stuff, multisensories things, all

815
00:40:13,640 --> 00:40:17,080
Speaker 2: this stuff, were modeling it, and we decided earlier this

816
00:40:17,160 --> 00:40:19,160
Speaker 2: year that the best way to go forward was to

817
00:40:19,200 --> 00:40:22,200
Speaker 2: do this in an open source project. We haven't actually

818
00:40:22,239 --> 00:40:24,759
Speaker 2: told people we're doing this before. So we've created a

819
00:40:24,800 --> 00:40:29,160
Speaker 2: thousand Brains project. We're taking all of our our code

820
00:40:29,160 --> 00:40:31,680
Speaker 2: and putting an open source We are taking the patterns.

821
00:40:31,680 --> 00:40:32,920
Speaker 2: We have a lot of patents. We're going to make

822
00:40:33,040 --> 00:40:34,759
Speaker 2: a non a short clause. We've done that. I'm not

823
00:40:34,800 --> 00:40:37,680
Speaker 2: a short close on our patents. We have. We hire

824
00:40:37,680 --> 00:40:41,440
Speaker 2: a team of people to like open source project manager

825
00:40:41,480 --> 00:40:43,839
Speaker 2: for the outside people. We've already got quite a few

826
00:40:43,840 --> 00:40:48,239
Speaker 2: people interested. We already have received some funding from the

827
00:40:48,239 --> 00:40:50,600
Speaker 2: Gates Foundation for this significant funding and help fund the

828
00:40:50,600 --> 00:40:52,440
Speaker 2: project for a couple of years. We have there's a

829
00:40:52,480 --> 00:40:55,320
Speaker 2: guy named John hen at Connegie Mellen University who's building

830
00:40:55,320 --> 00:40:58,240
Speaker 2: silicon to implement quartical columns. So there's the people around

831
00:40:58,280 --> 00:41:00,600
Speaker 2: the world who've been excited about our work and following

832
00:41:00,640 --> 00:41:03,160
Speaker 2: it and want to join in on this day. So

833
00:41:03,160 --> 00:41:05,840
Speaker 2: we figured let's get them all together, let's build a

834
00:41:05,880 --> 00:41:10,360
Speaker 2: framework open source project. And so we built out this team.

835
00:41:10,840 --> 00:41:14,239
Speaker 2: We have just been run by the women Vision Clay,

836
00:41:14,239 --> 00:41:18,360
Speaker 2: who's just brilliant, and technical side is by Dunam Neils

837
00:41:19,280 --> 00:41:22,600
Speaker 2: and and so we're just starting this, you know. So

838
00:41:22,840 --> 00:41:25,560
Speaker 2: we actually haven't officially and we've talked about it, but

839
00:41:25,560 --> 00:41:28,279
Speaker 2: we haven't officially launched yet because not everything is open

840
00:41:28,360 --> 00:41:29,840
Speaker 2: yet and we have to there's a lot of stuff

841
00:41:29,840 --> 00:41:31,160
Speaker 2: you have to do to put in to get those

842
00:41:31,200 --> 00:41:33,920
Speaker 2: whole do work. But we're going full boring this and

843
00:41:33,960 --> 00:41:38,480
Speaker 2: I think my hope is that anyone who's excited about

844
00:41:38,520 --> 00:41:40,879
Speaker 2: the work, and there's quite a few people can help

845
00:41:40,960 --> 00:41:43,600
Speaker 2: join us and work on this and propel it forward

846
00:41:43,680 --> 00:41:47,120
Speaker 2: and really created what I not only just an alternate

847
00:41:47,160 --> 00:41:49,880
Speaker 2: form of AI centory motor AI based on brain principles,

848
00:41:50,000 --> 00:41:52,120
Speaker 2: but I think what's going to be actually the ultimate,

849
00:41:52,560 --> 00:41:57,360
Speaker 2: uh primary source of AI, which is brain modeling a

850
00:41:57,400 --> 00:41:59,759
Speaker 2: thousand brains projects. This is amazing. So how do people

851
00:41:59,800 --> 00:42:00,680
Speaker 2: get involved in this?

852
00:42:01,120 --> 00:42:01,239
Speaker 1: Uh?

853
00:42:01,520 --> 00:42:04,480
Speaker 2: You can just go to our website dementa and dot

854
00:42:04,520 --> 00:42:07,040
Speaker 2: com and you know, there's a lot of information already,

855
00:42:07,239 --> 00:42:09,480
Speaker 2: a tons of information. It's like we have all the

856
00:42:09,480 --> 00:42:14,160
Speaker 2: stuff with accumulated documentation, code, videos, were all that up there,

857
00:42:14,239 --> 00:42:16,759
Speaker 2: plus tutorials and so on. So you just you can

858
00:42:16,880 --> 00:42:19,319
Speaker 2: just you can you go to our nomenta dot com.

859
00:42:19,320 --> 00:42:21,520
Speaker 2: It'll be obvious how to sign up to be informed

860
00:42:21,520 --> 00:42:23,080
Speaker 2: of what's going on or how to get involved.

861
00:42:23,080 --> 00:42:25,759
Speaker 1: Great, So some listener to this podcast is I want

862
00:42:25,760 --> 00:42:27,640
Speaker 1: to get involved and understand more about this thing. Go

863
00:42:27,640 --> 00:42:29,440
Speaker 1: to Nementa dot com and they can.

864
00:42:29,760 --> 00:42:31,920
Speaker 2: They can. First of all, they'll start, they'll sign up

865
00:42:31,960 --> 00:42:34,839
Speaker 2: to getting notified about things are happening. They can get

866
00:42:34,960 --> 00:42:38,720
Speaker 2: educated on the whole project. They can I don't think

867
00:42:38,880 --> 00:42:42,239
Speaker 2: right yet they can contribute code yet, but that will

868
00:42:42,280 --> 00:42:44,839
Speaker 2: happen within a month or so. It'll be obvious how

869
00:42:44,880 --> 00:42:47,200
Speaker 2: to get started. There's there's a lot of information to learn.

870
00:42:47,320 --> 00:42:49,440
Speaker 2: I would think if you haven't. If you haven't, you

871
00:42:49,520 --> 00:42:51,479
Speaker 2: might want to start with just by reading the book

872
00:42:51,480 --> 00:42:53,719
Speaker 2: to the Thousand Brains, because it gives you that not

873
00:42:53,800 --> 00:42:56,080
Speaker 2: only the basics of the theory, but it also gives

874
00:42:56,080 --> 00:42:57,920
Speaker 2: you the vision about how this is going to play

875
00:42:57,920 --> 00:42:58,399
Speaker 2: out over time.

876
00:42:58,600 --> 00:43:00,840
Speaker 1: And so the idea is just so I'm straight on this.

877
00:43:00,920 --> 00:43:04,280
Speaker 1: So the idea is a person can download the code

878
00:43:04,320 --> 00:43:05,760
Speaker 1: and run this model.

879
00:43:06,480 --> 00:43:10,279
Speaker 2: Right the first we're making said that you can do that.

880
00:43:10,320 --> 00:43:13,920
Speaker 2: You can run our current experiments, you can recreate them,

881
00:43:13,960 --> 00:43:16,080
Speaker 2: you can apply them different ways. Great.

882
00:43:16,560 --> 00:43:18,520
Speaker 1: So something that you and I have in common that

883
00:43:18,560 --> 00:43:21,760
Speaker 1: we are obsessed about is this idea that we're living

884
00:43:22,120 --> 00:43:25,359
Speaker 1: inside our own internal models. This is all a construction.

885
00:43:26,000 --> 00:43:27,880
Speaker 1: And you had a line in the book that I loved,

886
00:43:27,920 --> 00:43:32,680
Speaker 1: which is that if you had different sensors for picking

887
00:43:32,760 --> 00:43:35,600
Speaker 1: up different information in the world, we would have a

888
00:43:35,680 --> 00:43:39,880
Speaker 1: different perceptual experience, a completely different experience of the universe.

889
00:43:40,080 --> 00:43:45,880
Speaker 2: Well maybe completely, not completely, Like a blind person is

890
00:43:46,000 --> 00:43:48,399
Speaker 2: we're learning the worth of touch, and a person who

891
00:43:49,480 --> 00:43:53,880
Speaker 2: is deaferent, a person maybe who has sensory problems on

892
00:43:54,000 --> 00:43:56,000
Speaker 2: his hand, they will end up with a similar structure.

893
00:43:56,239 --> 00:43:58,440
Speaker 1: Sorry, but what I mean is not in terms of

894
00:43:58,520 --> 00:44:00,400
Speaker 1: how we pick up on the visible light ratee. But

895
00:44:00,719 --> 00:44:02,680
Speaker 1: I pick up on infrared and you pick up on

896
00:44:02,840 --> 00:44:03,680
Speaker 1: radio waves.

897
00:44:03,680 --> 00:44:06,040
Speaker 2: Okay, right, you might if you if you really did that,

898
00:44:06,280 --> 00:44:07,919
Speaker 2: then you would have a different view of the world,

899
00:44:07,960 --> 00:44:10,840
Speaker 2: like it's often you know, like it take the issue

900
00:44:10,840 --> 00:44:13,399
Speaker 2: of color. It's often said that bees, you know, seeing

901
00:44:13,400 --> 00:44:16,160
Speaker 2: the ultra violet and we don't. So what looks like

902
00:44:16,280 --> 00:44:18,400
Speaker 2: toss is a white flower to them? Is this beautifully

903
00:44:18,400 --> 00:44:20,120
Speaker 2: colorful variegated flower.

904
00:44:20,920 --> 00:44:22,799
Speaker 1: But let's say you saw a totally different part of

905
00:44:22,800 --> 00:44:26,560
Speaker 1: the electromagic spectrum and so you see in the microwave range.

906
00:44:27,000 --> 00:44:29,319
Speaker 1: Question is would would we have color at all?

907
00:44:29,480 --> 00:44:32,360
Speaker 2: I don't know, It's hard to say, right, there's a

908
00:44:32,480 --> 00:44:37,399
Speaker 2: there's an underlying really interesting philosophical problem called qualitia, which

909
00:44:37,440 --> 00:44:40,840
Speaker 2: is like, why does color feel like color? Right? And

910
00:44:40,880 --> 00:44:45,279
Speaker 2: it doesn't feel like sounds or tactile sensations. And it's

911
00:44:45,320 --> 00:44:48,279
Speaker 2: an interesting challenge to understand that. I've written about it

912
00:44:48,320 --> 00:44:48,560
Speaker 2: a bit.

913
00:44:49,440 --> 00:44:51,680
Speaker 1: Yeah, do you have a hypothesis about this, I'll tell

914
00:44:51,680 --> 00:44:54,279
Speaker 1: you what mine is, But it is it's always sort

915
00:44:54,320 --> 00:44:56,080
Speaker 1: of half one, which is I think it's about the

916
00:44:56,320 --> 00:44:59,840
Speaker 1: structure of the data coming in defines the quality.

917
00:45:00,080 --> 00:45:02,239
Speaker 2: I don't know why or how that's true.

918
00:45:01,960 --> 00:45:04,359
Speaker 1: But you know, with the eyes, you've got two two

919
00:45:04,440 --> 00:45:08,440
Speaker 1: dimensional sheets of data coming in, and so vision feels

920
00:45:08,680 --> 00:45:12,120
Speaker 1: like something with hearing, it's a one dimensional signals just

921
00:45:12,120 --> 00:45:14,160
Speaker 1: going up and down and vibringing your ear drum. That

922
00:45:14,360 --> 00:45:17,080
Speaker 1: feels like something. You don't confuse vision with hearing. That's

923
00:45:17,280 --> 00:45:20,399
Speaker 1: like completely different worlds to you. My interest has been

924
00:45:20,440 --> 00:45:23,960
Speaker 1: in what happens when we feed news structures. We've done

925
00:45:23,960 --> 00:45:25,160
Speaker 1: a lot of interesting stuff in this area.

926
00:45:25,280 --> 00:45:28,640
Speaker 2: Exactly would you have a completely new quality? Is it possible? So?

927
00:45:28,840 --> 00:45:32,080
Speaker 2: I mean, certainly you can imagine. First of all, I

928
00:45:32,120 --> 00:45:36,719
Speaker 2: agree with you again, it's all spikes, right, So there's

929
00:45:36,719 --> 00:45:41,319
Speaker 2: nothing there's no color spikes, there's no heat, spikes. It's

930
00:45:41,360 --> 00:45:45,719
Speaker 2: just spikes. And so obviously the different quality it has

931
00:45:45,719 --> 00:45:48,279
Speaker 2: to come about somehow from the structure of the data

932
00:45:48,440 --> 00:45:52,359
Speaker 2: spatially and temporarily, and also sensory motory, you know, it's

933
00:45:52,360 --> 00:45:54,640
Speaker 2: like how things change as you move, and I think

934
00:45:54,640 --> 00:45:56,200
Speaker 2: that's a big part of it. So I agree with

935
00:45:56,280 --> 00:45:58,600
Speaker 2: on a fundamental level that it has to be some

936
00:45:58,840 --> 00:46:04,120
Speaker 2: in the data and it's nothing else. And we can

937
00:46:04,120 --> 00:46:06,560
Speaker 2: then ask ourselves something like, well, imagine you've been blind

938
00:46:06,680 --> 00:46:08,880
Speaker 2: your whole life. You don't have a sense of color.

939
00:46:09,400 --> 00:46:11,680
Speaker 2: You've never experienced color, and so to you would be

940
00:46:11,800 --> 00:46:14,120
Speaker 2: kind of mysterious things. Someone can say, well, can't you

941
00:46:14,160 --> 00:46:16,920
Speaker 2: tell that's that's you know, that's this type of orange

942
00:46:16,920 --> 00:46:19,600
Speaker 2: and that time? What are you talking about? Right? They'd

943
00:46:19,600 --> 00:46:21,480
Speaker 2: have to accept that you have some super sense and

944
00:46:21,520 --> 00:46:24,120
Speaker 2: the world looks different to you because you have vision

945
00:46:24,200 --> 00:46:26,479
Speaker 2: and I don't. And they may be able to touch

946
00:46:26,560 --> 00:46:28,640
Speaker 2: things that you know, sense things that I don't. Sign,

947
00:46:28,680 --> 00:46:30,000
Speaker 2: So I could be able to try to read braille

948
00:46:30,160 --> 00:46:32,160
Speaker 2: if you're not a braille reader, that feels like what

949
00:46:32,440 --> 00:46:35,759
Speaker 2: the stuff? It's a blur? Right? So they oh, no,

950
00:46:35,840 --> 00:46:38,359
Speaker 2: I feel everything there, right, So we can we can

951
00:46:38,400 --> 00:46:40,400
Speaker 2: just ask ourselves a questions like what's the world like

952
00:46:40,440 --> 00:46:43,360
Speaker 2: to different people, and sometimes we'll end up with a

953
00:46:43,400 --> 00:46:45,719
Speaker 2: similar model, like yeah, well you and I would have

954
00:46:45,760 --> 00:46:48,200
Speaker 2: no matter what censors you have, we'd have the model

955
00:46:48,200 --> 00:46:51,879
Speaker 2: of physical structure of a coffee cup. But other times

956
00:46:51,920 --> 00:46:54,080
Speaker 2: it could be quite different, you know, and certain if

957
00:46:54,080 --> 00:46:58,480
Speaker 2: you start like sensing parts of the radio spectrum or

958
00:46:59,400 --> 00:47:01,480
Speaker 2: other things, just be you know. One of the things

959
00:47:01,480 --> 00:47:03,200
Speaker 2: I always wondered, like, what would be like if you

960
00:47:03,200 --> 00:47:06,680
Speaker 2: had if you had smell sensors stand under your fingers, right,

961
00:47:07,160 --> 00:47:10,680
Speaker 2: and then everything you touch. Well, we kind of we have.

962
00:47:10,800 --> 00:47:13,560
Speaker 2: We have temperature sensors, and we have tapping with all

963
00:47:13,640 --> 00:47:15,480
Speaker 2: kinds of But what I could smell like you could

964
00:47:15,480 --> 00:47:18,640
Speaker 2: tell chemicals that were on the surface of objects. This

965
00:47:18,719 --> 00:47:21,120
Speaker 2: is what dogs do. You know. Dogs they don't just smell.

966
00:47:21,360 --> 00:47:23,319
Speaker 2: They stick their nose right on the thing and they smell.

967
00:47:23,400 --> 00:47:25,720
Speaker 2: They moved to the next bot it smell. Dogs build

968
00:47:25,760 --> 00:47:28,920
Speaker 2: this freedom, actual structure of smells. We don't have that

969
00:47:28,960 --> 00:47:30,919
Speaker 2: smell for us. It's kind of like wasting in from

970
00:47:30,920 --> 00:47:33,920
Speaker 2: some direction, right. Dogs have this incredible model of the

971
00:47:33,920 --> 00:47:36,799
Speaker 2: world smell mod and it's hard to imagine what it is.

972
00:47:36,840 --> 00:47:39,160
Speaker 2: But I'm sure they have it, so I think it's

973
00:47:39,239 --> 00:47:42,600
Speaker 2: fun to think about these things. I don't you know,

974
00:47:42,640 --> 00:47:44,719
Speaker 2: in the future will build machines that perceive the world

975
00:47:44,719 --> 00:47:46,120
Speaker 2: different than we do. But that'll be great.

976
00:47:46,640 --> 00:47:50,920
Speaker 1: Yeah, Okay, Jeff, this has been wonderful.

977
00:47:50,960 --> 00:47:53,560
Speaker 2: Thank you for being here too. Thanks David. It's always

978
00:47:53,560 --> 00:47:55,120
Speaker 2: great talking to you and I enjoy it. It's a

979
00:47:55,160 --> 00:47:58,920
Speaker 2: lot of fun we were and I love your podcast.

980
00:47:58,960 --> 00:48:06,920
Speaker 1: So that was Jeff Hawkins, theoretician and author of A

981
00:48:07,120 --> 00:48:08,200
Speaker 1: Thousand Brains.

982
00:48:08,680 --> 00:48:08,879
Speaker 2: Now.

983
00:48:08,920 --> 00:48:12,000
Speaker 1: I love his model because it builds on previous research

984
00:48:12,080 --> 00:48:15,239
Speaker 1: and gives us a possible starting point for how this

985
00:48:15,320 --> 00:48:18,759
Speaker 1: whole system might be working. This is a view of

986
00:48:18,800 --> 00:48:21,400
Speaker 1: the brain in which you don't have just a single

987
00:48:21,560 --> 00:48:25,200
Speaker 1: model of the world being constructed, but hundreds of thousands

988
00:48:25,200 --> 00:48:29,680
Speaker 1: of little models, each viewing the world through their little straw.

989
00:48:29,840 --> 00:48:33,600
Speaker 1: And these models are independent, but they're not completely independent,

990
00:48:33,680 --> 00:48:36,840
Speaker 1: so they communicate with each other and they vote, and

991
00:48:36,880 --> 00:48:40,720
Speaker 1: in this way, the whole system converges on its best

992
00:48:40,960 --> 00:48:44,120
Speaker 1: guess of what's going on out there in the world.

993
00:48:44,600 --> 00:48:48,520
Speaker 1: And by this mechanism we construct a full three dimensional

994
00:48:48,560 --> 00:48:52,000
Speaker 1: representation of the environment around us, with its sites and

995
00:48:52,120 --> 00:48:55,600
Speaker 1: sounds and three dimensional structure. So this gives us a

996
00:48:55,920 --> 00:48:59,719
Speaker 1: clear framework for thinking about the neocortex. Now, we might

997
00:48:59,719 --> 00:49:02,160
Speaker 1: not know, oh for a while, if this answers everything,

998
00:49:02,600 --> 00:49:06,000
Speaker 1: or it needs some tweaking, or if there are far

999
00:49:06,160 --> 00:49:09,719
Speaker 1: better models coming down the pike. But what I absolutely

1000
00:49:09,800 --> 00:49:13,520
Speaker 1: love about this is that this is where the endeavor

1001
00:49:13,560 --> 00:49:19,600
Speaker 1: of science shines. Taking something that seems insanely complex, eighty

1002
00:49:19,600 --> 00:49:23,680
Speaker 1: six billion neurons with two hundred trillion connections, something of

1003
00:49:24,080 --> 00:49:29,239
Speaker 1: such vast complexity that it bankrupts our language, and saying, wait,

1004
00:49:29,600 --> 00:49:32,799
Speaker 1: what if there's a really simple principle at work here?

1005
00:49:32,840 --> 00:49:35,839
Speaker 1: What if there's a way that we could reduce all

1006
00:49:35,920 --> 00:49:39,080
Speaker 1: that complexity by just looking at this from a new angle.

1007
00:49:39,600 --> 00:49:42,040
Speaker 1: So let me give an analogy here. Just think about

1008
00:49:42,040 --> 00:49:44,759
Speaker 1: what it would be like if you had a magical

1009
00:49:44,960 --> 00:49:48,319
Speaker 1: microscope with which you could look into a cell and

1010
00:49:48,400 --> 00:49:51,520
Speaker 1: into the nucleus in the middle. What you would see

1011
00:49:51,600 --> 00:49:56,440
Speaker 1: is mind boggling complexity. There. You'd see millions or billions

1012
00:49:56,440 --> 00:50:01,279
Speaker 1: of molecules racing around and interacting and doing god knows what,

1013
00:50:01,760 --> 00:50:04,520
Speaker 1: and you'd say, wow, there's no way.

1014
00:50:04,360 --> 00:50:05,719
Speaker 2: We're ever going to understand this.

1015
00:50:06,560 --> 00:50:10,520
Speaker 1: But then Krick and Watson come along and say, actually,

1016
00:50:10,560 --> 00:50:15,600
Speaker 1: the important thing is this DNA molecule and keeping the

1017
00:50:15,840 --> 00:50:17,680
Speaker 1: order of these base.

1018
00:50:17,480 --> 00:50:20,640
Speaker 2: Pairs, and all the rest is housekeeping.

1019
00:50:21,520 --> 00:50:26,239
Speaker 1: And suddenly the fog of confusion lifts. Now something that

1020
00:50:26,360 --> 00:50:29,480
Speaker 1: seemed well beyond us can be described in a sentence

1021
00:50:29,560 --> 00:50:34,440
Speaker 1: or two, and science leaps forward and things move fast

1022
00:50:34,480 --> 00:50:36,719
Speaker 1: from there. I worked with Francis Crik when I was

1023
00:50:36,719 --> 00:50:40,440
Speaker 1: in my postdoctoral years, and now I look around me

1024
00:50:40,560 --> 00:50:44,000
Speaker 1: at Stanford and Silicon Valley, and there are thousands of

1025
00:50:44,200 --> 00:50:49,040
Speaker 1: laboratories and companies doing amazing work with genomes, and their

1026
00:50:49,160 --> 00:50:54,480
Speaker 1: existence results entirely from this one simplifying insight about DNA

1027
00:50:54,840 --> 00:50:59,280
Speaker 1: in nineteen fifty three, that new model that suddenly clarified

1028
00:50:59,760 --> 00:51:04,040
Speaker 1: what what is happening inside the nucleus. By the same token,

1029
00:51:04,160 --> 00:51:07,080
Speaker 1: this is what we're trying to do with the brain.

1030
00:51:07,320 --> 00:51:13,520
Speaker 1: Brains appear to be ferociously complex, and yet we have

1031
00:51:13,680 --> 00:51:15,480
Speaker 1: lots of brains running around the planet.

1032
00:51:15,480 --> 00:51:17,480
Speaker 2: We've got eight point two billion of them.

1033
00:51:18,000 --> 00:51:22,719
Speaker 1: So something must be straightforward about their architecture, or else

1034
00:51:22,800 --> 00:51:25,680
Speaker 1: Mother Nature wouldn't be able to build these over and

1035
00:51:25,719 --> 00:51:30,480
Speaker 1: over with such reliability. You couldn't drop this massive quantity

1036
00:51:30,560 --> 00:51:33,720
Speaker 1: into the world and have them all functioning well unless

1037
00:51:33,719 --> 00:51:38,600
Speaker 1: there was something pretty uncomplicated about building and running a brain.

1038
00:51:39,560 --> 00:51:44,000
Speaker 1: So that is the overarching game of science to take

1039
00:51:44,040 --> 00:51:50,080
Speaker 1: the overwhelming complexity around us and to find new angles

1040
00:51:50,160 --> 00:51:58,120
Speaker 1: to look at things to reveal simplicity. Go to eagleman

1041
00:51:58,200 --> 00:52:02,000
Speaker 1: dot com slash podcast for more information and find further reading.

1042
00:52:02,520 --> 00:52:05,360
Speaker 1: Send me an email at podcasts at eagleman dot com

1043
00:52:05,360 --> 00:52:08,880
Speaker 1: with questions or discussion, and check out and subscribe to

1044
00:52:09,080 --> 00:52:12,480
Speaker 1: Inner Cosmos on YouTube for videos of each episode and

1045
00:52:12,520 --> 00:52:16,520
Speaker 1: to leave comments until next time. I'm David Eagleman, and

1046
00:52:16,560 --> 00:52:18,319
Speaker 1: this is Inner Cosmos.