Stephen Grossberg: How Does Each Brain Make A Mind? Adaptive Resonance Theory (ART) & Consciousness

1
00:00:10,120 --> 00:00:13,240
Steve, I wanna start with the
most fundamental question.

2
00:00:13,240 --> 00:00:16,560
This podcast is called Mind Body
Solution and it's obviously

3
00:00:16,560 --> 00:00:18,120
paying homage to the mind body
problem.

4
00:00:18,440 --> 00:00:21,600
But I find that definitions play
a huge role with this podcast

5
00:00:21,600 --> 00:00:24,560
and many people want to know
your definition or the guests

6
00:00:24,640 --> 00:00:27,040
definition of something.
So in this case, let's talk

7
00:00:27,040 --> 00:00:30,290
about how do you define or
differentiate between

8
00:00:30,290 --> 00:00:32,729
consciousness, the self and the
mind.

9
00:00:35,490 --> 00:00:41,450
So before I jump into such a
complex issue, I think I should

10
00:00:41,450 --> 00:00:45,330
give you a little background
about how I got started as a

11
00:00:45,330 --> 00:00:49,890
scientist who eventually thought
he had anything to say about

12
00:00:49,890 --> 00:00:57,170
consciousness, mind and self.
And I think it's a good story

13
00:00:57,170 --> 00:01:00,870
because.
The moral of the story is you're

14
00:01:00,870 --> 00:01:04,550
never too young to follow your
life's passion.

15
00:01:06,710 --> 00:01:12,590
So my life's work began in 1957,
before you were born, that I was

16
00:01:12,590 --> 00:01:18,270
17 years old and I was a college
freshman taking introductory

17
00:01:18,270 --> 00:01:24,870
psychology at Dartmouth College.
And it was in that year that I

18
00:01:24,870 --> 00:01:27,230
introduced the field of neural
networks.

19
00:01:27,940 --> 00:01:32,260
And that was because I was
inspired by basic facts about

20
00:01:32,260 --> 00:01:39,100
how humans learn.
And the psychological data led

21
00:01:39,100 --> 00:01:43,660
me to derive my first neural
models of how brains make minds.

22
00:01:43,660 --> 00:01:47,820
And you might say, well, how
does that happen?

23
00:01:48,740 --> 00:01:54,580
And I was led from mind to brain
because brain evolution needs to

24
00:01:54,580 --> 00:01:56,580
achieve behavioral success.
So.

25
00:01:57,310 --> 00:02:01,950
Without brain mechanisms, one
can't understand psychological

26
00:02:01,950 --> 00:02:06,990
functions, including learning.
But without the psychological

27
00:02:06,990 --> 00:02:11,070
functions, the brain mechanisms
have no behavioral meaning.

28
00:02:11,070 --> 00:02:15,710
So to me, I always felt the
need, since I was always

29
00:02:15,710 --> 00:02:20,550
thinking in real time about how
does this kind of learning or

30
00:02:20,550 --> 00:02:24,470
other human experience emerge
moment by moment in real time.

31
00:02:24,790 --> 00:02:28,710
I had to think about.
What's going on in there to

32
00:02:28,750 --> 00:02:32,830
enable it so?
And over the years, it's become

33
00:02:32,830 --> 00:02:40,470
clear that interactions and
within and across several brain

34
00:02:40,470 --> 00:02:44,630
regions are often needed to
generate these psychological

35
00:02:44,630 --> 00:02:47,390
behaviors.
And the behaviors are really

36
00:02:47,390 --> 00:02:51,110
emergent properties of these
interactions.

37
00:02:51,110 --> 00:02:56,010
It's not like a computer code.
But purely experimental

38
00:02:56,010 --> 00:03:01,050
approaches cannot fully
understand emergent properties.

39
00:03:01,050 --> 00:03:05,290
Even if you have 100 electrodes
in a wake behaving monkey, the

40
00:03:05,290 --> 00:03:09,490
emergent property and behavior,
There's an explanatory gap.

41
00:03:10,090 --> 00:03:16,410
And for that you need models.
So I was driven as a freshman to

42
00:03:16,410 --> 00:03:20,850
derive my first neural models.
I introduced the paradigm of

43
00:03:20,850 --> 00:03:24,170
using systems of nonlinear
differential equations.

44
00:03:24,670 --> 00:03:29,470
I introduced cell activations or
short term memory traces.

45
00:03:29,470 --> 00:03:33,950
I introduced long term memory.
Traces were adaptive weights for

46
00:03:33,950 --> 00:03:37,590
learning memory and I also
introduced what I like to call

47
00:03:37,590 --> 00:03:42,070
medium term memory, which is
activity dependent Habituation

48
00:03:42,230 --> 00:03:46,230
in a word, if you find them too
long, cells can get tired.

49
00:03:48,950 --> 00:03:52,470
But to derive the models I had
to develop a.

50
00:03:53,310 --> 00:03:58,550
Modeling method and the cycle
because you can't derive a whole

51
00:03:58,550 --> 00:04:02,550
brain all at once, and God knows
that I hardly knew anything when

52
00:04:02,550 --> 00:04:07,190
I got started.
So, because brain evolution

53
00:04:07,190 --> 00:04:11,470
needs to achieve behavioral
success, such a method would

54
00:04:11,470 --> 00:04:15,910
always start by analyzing scores
or even hundreds of

55
00:04:15,910 --> 00:04:19,950
psychological experiments, since
he's the experiments that define

56
00:04:19,950 --> 00:04:22,230
the behavior we want to
understand.

57
00:04:23,940 --> 00:04:31,180
And the method showed how to
discover design principles that

58
00:04:31,180 --> 00:04:35,460
underlie these large
psychological database.

59
00:04:35,460 --> 00:04:37,580
And you might say, well, how
does that happen?

60
00:04:38,460 --> 00:04:43,260
And here I just have to quote
the greatest of them all, Albert

61
00:04:43,260 --> 00:04:45,660
Einstein.
It was a speculative leap.

62
00:04:46,340 --> 00:04:50,540
If you live in hundreds of
experiments long enough, you can

63
00:04:50,540 --> 00:04:54,420
begin to see.
What the underlying principles

64
00:04:54,420 --> 00:05:00,980
and mechanisms are that
generated those behaviors as

65
00:05:00,980 --> 00:05:02,980
emergent properties?
It's a gift.

66
00:05:02,980 --> 00:05:06,700
It's a leap.
There is no algorithm for it.

67
00:05:06,700 --> 00:05:11,020
And the way to learn it, if
you're lucky, is to study with

68
00:05:11,020 --> 00:05:13,740
someone who does it well.
And that's what I did with all

69
00:05:13,740 --> 00:05:17,180
my students.
And then after you get the

70
00:05:17,180 --> 00:05:21,620
design principles, you've got to
derive.

71
00:05:22,340 --> 00:05:29,380
The minimal models that embody
them and then show how models

72
00:05:29,380 --> 00:05:32,780
can be used to explain lots more
psychological data than the

73
00:05:32,780 --> 00:05:35,700
hypothesis that you used to
derive them.

74
00:05:35,700 --> 00:05:37,940
Otherwise you'd just be going in
circles.

75
00:05:38,500 --> 00:05:40,460
And you might say, well, why
minimal models?

76
00:05:40,460 --> 00:05:43,780
It's the principle of parsimony.
It's Occam's razor.

77
00:05:44,580 --> 00:05:47,780
Even if I might know, oh,
there's this interneuron, and I

78
00:05:47,780 --> 00:05:51,940
haven't yet put in this or that
of the other by.

79
00:05:53,030 --> 00:05:58,910
Being honest about it and only
deriving the model that I knew

80
00:05:58,910 --> 00:06:02,710
at that time from the
postulates, that put huge

81
00:06:02,710 --> 00:06:08,510
pressure on me to derive the
next more integrative model.

82
00:06:09,030 --> 00:06:15,390
And the big surprise to me when
I was a boy of 17 is that these

83
00:06:15,390 --> 00:06:19,790
models look like, you know, they
they helped me.

84
00:06:20,540 --> 00:06:26,020
Viral data and the models look
like biological neural networks,

85
00:06:27,500 --> 00:06:30,660
and the bottom line is learn how
to think in real time.

86
00:06:30,660 --> 00:06:35,780
That's an art.
So as I indicated, no one

87
00:06:37,140 --> 00:06:41,700
derivation can derive a whole
brain, But after doing the

88
00:06:41,700 --> 00:06:44,780
derivation and explaining what I
could with that stage of

89
00:06:44,780 --> 00:06:48,600
modeling.
I could sort of begin to see the

90
00:06:48,600 --> 00:06:52,840
boundary, I like to call it,
between what I knew and what I

91
00:06:52,840 --> 00:06:55,760
didn't know.
And by understanding a little

92
00:06:55,760 --> 00:06:59,440
bit about the shape of that
boundary, I could figure out

93
00:06:59,440 --> 00:07:04,000
what new design principle or
hypothesis I left out.

94
00:07:04,560 --> 00:07:09,840
I'd put that in and then derive
the next more complex model in a

95
00:07:09,840 --> 00:07:12,680
self consistent way without
undermining.

96
00:07:13,480 --> 00:07:16,200
What I had already done, which
would have been, you know,

97
00:07:17,080 --> 00:07:20,320
frustrating.
And so the new model always

98
00:07:20,320 --> 00:07:24,520
included the old model, although
some somehow in a more

99
00:07:24,520 --> 00:07:26,840
sophisticated form.
I'll mention briefly.

100
00:07:26,840 --> 00:07:33,480
Eventually, I was led to models
that linked the lamina, or

101
00:07:34,320 --> 00:07:39,200
layered circuits of cerebral
cortex to different kinds of

102
00:07:39,200 --> 00:07:44,700
behavior, like vision.
Recognition, speech, language,

103
00:07:44,700 --> 00:07:47,340
movement.
And I never dreamed when I was a

104
00:07:47,380 --> 00:07:53,660
boy that could ever happen.
But it also was what led to

105
00:07:54,620 --> 00:07:58,940
understand how we learn, to pay
attention, to recognize and

106
00:07:58,940 --> 00:08:02,660
predict objects and events in a
changing world.

107
00:08:02,660 --> 00:08:06,060
And I say changing world because
I was always thinking about

108
00:08:06,060 --> 00:08:08,460
dynamics.
You know, we're always evolving.

109
00:08:08,460 --> 00:08:12,480
We're always growing.
And with this foundation, I was

110
00:08:12,480 --> 00:08:17,400
eventually forced into a
functional understanding of

111
00:08:17,400 --> 00:08:24,480
consciousness.
And I have to admit immediately,

112
00:08:24,480 --> 00:08:27,080
I never try to understand
consciousness.

113
00:08:27,680 --> 00:08:30,280
You don't wake up in the morning
and say, oh, hey, today I'm

114
00:08:30,280 --> 00:08:33,720
going to understand
consciousness, you know, never

115
00:08:33,720 --> 00:08:36,760
could happen.
Just as I never thought when I

116
00:08:36,760 --> 00:08:39,320
start setting psychology and
learning.

117
00:08:40,070 --> 00:08:45,070
That I would be thrown into the
norodynamics behind learning.

118
00:08:45,070 --> 00:08:49,950
You know, these are the things
that happen if you're honest and

119
00:08:50,470 --> 00:08:54,870
lucky enough to find true
principles, because they will

120
00:08:55,390 --> 00:09:02,990
never disappoint you.
I call it the gift that keeps on

121
00:09:02,990 --> 00:09:07,230
giving.
And so most of my work emerged

122
00:09:07,230 --> 00:09:11,080
from an analysis of.
Self organization, both in

123
00:09:11,080 --> 00:09:16,000
development and learning.
Notably how we can learn quickly

124
00:09:16,000 --> 00:09:19,600
without suffering from
catastrophic forgetting.

125
00:09:19,600 --> 00:09:23,400
Otherwise expressed how we can
learn quickly without forgetting

126
00:09:23,400 --> 00:09:27,240
just as quickly, which of course
was a topic very much on the

127
00:09:27,240 --> 00:09:31,600
mind of a young student.
And learning quickly without

128
00:09:31,600 --> 00:09:36,360
catastrophic forgetting, I've
come to call solving the

129
00:09:36,360 --> 00:09:40,050
stability.
Plasticity dilemma, where the

130
00:09:40,050 --> 00:09:44,690
plasticity is the learning and
the stability is the fact that

131
00:09:44,690 --> 00:09:47,930
memories are buffered so you
don't have catastrophic

132
00:09:47,930 --> 00:09:51,170
forgetting.
And that's how I was led to

133
00:09:51,170 --> 00:09:56,130
introduce adaptive resonance
theory or a RTI, just say art,

134
00:09:56,730 --> 00:10:03,690
in 1976 to solve the problem and
given that I was starting.

135
00:10:04,430 --> 00:10:08,430
In order to understand learning,
I think it's still remarkable

136
00:10:09,070 --> 00:10:15,990
that art dynamics also showed me
something of the underlying

137
00:10:15,990 --> 00:10:21,070
mechanisms of conscious states
of seeing, hearing, feeling and

138
00:10:21,070 --> 00:10:25,310
knowing.
And emphatically, art also

139
00:10:25,310 --> 00:10:29,870
explains that you're not there
to just contemplate the

140
00:10:30,230 --> 00:10:32,510
Platonically the beauty of the
world.

141
00:10:33,090 --> 00:10:37,690
You're using these conscious
states to plan and act, to

142
00:10:37,690 --> 00:10:40,290
realize valued goals.
They're essential to our

143
00:10:40,290 --> 00:10:44,410
survival.
So, and this introduction can

144
00:10:44,410 --> 00:10:47,330
lead me to try to answer your
next three questions.

145
00:10:48,050 --> 00:10:50,250
Of course, they were all
intimately linked.

146
00:10:50,970 --> 00:10:54,410
And after I do that, ask me
again and I'll comment about.

147
00:10:54,960 --> 00:10:59,240
Mind and self.
Because I'm not quite ready I

148
00:10:59,240 --> 00:11:01,600
think while we're on.
Before we go to these questions,

149
00:11:01,600 --> 00:11:05,560
I just want to say that many
scientists and philosophers that

150
00:11:05,560 --> 00:11:09,360
I've spoken to on this podcast
have spoken about you with so

151
00:11:09,360 --> 00:11:12,600
much respect.
They put your name up there with

152
00:11:12,600 --> 00:11:14,200
some of the greatest scientists
of all time.

153
00:11:14,320 --> 00:11:17,240
They call you the Newton of the
brain, the Einstein of the mind.

154
00:11:17,640 --> 00:11:19,630
And and I've noticed this is a
pattern.

155
00:11:19,630 --> 00:11:22,150
This is not something that's
just one or two people say.

156
00:11:22,150 --> 00:11:24,830
I mean, some people call you the
greatest living scientist of all

157
00:11:24,830 --> 00:11:27,030
time.
And having been reading your

158
00:11:27,030 --> 00:11:30,670
work for all very much most of
my life, I can honestly say that

159
00:11:30,670 --> 00:11:33,310
I very much agree with him with
that sentiment.

160
00:11:33,310 --> 00:11:38,270
I mean, your work is very, very
indepth intriguing, but not only

161
00:11:38,270 --> 00:11:40,750
that, you managed to bridge so
many gaps of different fields,

162
00:11:40,750 --> 00:11:45,070
this multidisciplinary approach
that I think is very vital when

163
00:11:45,070 --> 00:11:46,430
it comes to understanding
anything.

164
00:11:46,980 --> 00:11:49,180
You've managed to bridge so many
different fields together.

165
00:11:50,100 --> 00:11:52,980
Just congratulations.
Thank you.

166
00:11:53,380 --> 00:11:56,500
It's very kind to you to say
that and especially when you're

167
00:11:56,500 --> 00:12:02,100
talking to an old man.
I've been working like a dog for

168
00:12:02,100 --> 00:12:05,860
66 years.
So I like figure hearing things

169
00:12:05,860 --> 00:12:11,620
like that.
But also a key point, and I I

170
00:12:11,620 --> 00:12:14,940
can't overemphasize it, You have
to.

171
00:12:17,500 --> 00:12:23,380
Get really intimately familiar
with the large enough databases

172
00:12:24,460 --> 00:12:30,700
so that they can help you find
the simple design principles

173
00:12:31,380 --> 00:12:36,060
that can lead to in a real
insight, things you never

174
00:12:36,060 --> 00:12:40,140
dreamed to understand.
Of course, once you get that

175
00:12:40,140 --> 00:12:44,390
intuitive.
Grasping of some of the

176
00:12:44,390 --> 00:12:47,270
principles, then you have to
face the tough honesty of

177
00:12:47,270 --> 00:12:51,270
mathematics and then you have to
become technical.

178
00:12:51,270 --> 00:12:58,350
So I think my my main gift is
being able to see to the heart

179
00:12:58,430 --> 00:13:05,430
of data what what it means and
then I have enough technique to

180
00:13:05,430 --> 00:13:09,070
carry out the program and that's
one reason if not the main

181
00:13:09,070 --> 00:13:13,010
reason.
That I got a PhD in mathematics

182
00:13:13,010 --> 00:13:17,170
and one of my professorships is
professor of mathematics and

183
00:13:17,170 --> 00:13:22,170
statistics.
So I had the tools to carry out

184
00:13:22,170 --> 00:13:29,690
my intuitive ideas in a
quantitative and testable way.

185
00:13:32,050 --> 00:13:36,010
Well, Steve, on that note, let's
continue with this theme.

186
00:13:36,050 --> 00:13:38,170
Are you talking about
mathematical principles?

187
00:13:38,660 --> 00:13:40,500
Let's elaborate on the
mathematical models that

188
00:13:40,500 --> 00:13:44,260
underpin art adaptive residence
theory for everyone listening.

189
00:13:44,300 --> 00:13:47,500
And then let's also discuss a
brief overview of the theory and

190
00:13:47,500 --> 00:13:50,460
the key principles that
underline this theory, and how

191
00:13:50,460 --> 00:13:53,020
it provides this rigorous,
rigorous, sorry solution to the

192
00:13:53,020 --> 00:13:59,900
mind body problem.
OK, I'm happy to try so as I

193
00:13:59,940 --> 00:14:04,660
already noted, I was always
fascinated by.

194
00:14:05,360 --> 00:14:09,320
How humans learn so quickly
without being forced to forget

195
00:14:09,320 --> 00:14:12,560
just as quickly.
And this is the stability

196
00:14:12,560 --> 00:14:17,200
plasticity dilemma that I
introduced adaptive residency to

197
00:14:17,200 --> 00:14:20,800
solve.
But even with adaptive

198
00:14:20,800 --> 00:14:24,760
residency, you know you don't do
that waking up one day and say

199
00:14:24,840 --> 00:14:28,400
oh to under the salt, this
stability plasticity dilemma.

200
00:14:28,400 --> 00:14:32,400
Cause you don't even know that
there exists such a dilemma.

201
00:14:33,310 --> 00:14:37,870
So before I introduced art, I
introduced what many people like

202
00:14:37,870 --> 00:14:43,150
to competitive learning now, and
I did that in 1976.

203
00:14:44,710 --> 00:14:49,110
And that explains how humans in
various other mammals at very

204
00:14:49,110 --> 00:14:54,230
least learn recognition
categories, where recognition

205
00:14:54,230 --> 00:15:00,030
category is a self population
that fires selectively.

206
00:15:00,620 --> 00:15:05,380
To spatial patterns of
activation across the network of

207
00:15:05,380 --> 00:15:11,060
feature selective neurons.
So, for example, such a spatial

208
00:15:11,060 --> 00:15:15,140
pattern of feature selective
neurons might represent

209
00:15:15,420 --> 00:15:20,180
spatially organized features of
a face or of a painting.

210
00:15:20,180 --> 00:15:24,140
And if you learn mission
categories for the face and the

211
00:15:24,140 --> 00:15:29,500
painting, then you'll have cells
that can recognize your face and

212
00:15:29,500 --> 00:15:32,990
recognize.
A painting You love by Matisse.

213
00:15:34,110 --> 00:15:38,190
So a competitive learning
category can learn to

214
00:15:38,190 --> 00:15:43,190
selectively fire in response to
a particular taste or painting,

215
00:15:43,670 --> 00:15:49,110
or indeed any spatial pattern of
features that's presented to the

216
00:15:49,110 --> 00:15:53,070
network.
And So what?

217
00:15:53,070 --> 00:15:57,790
First let me explain how a
spatial pattern evacuation.

218
00:15:58,550 --> 00:16:03,350
Across feature selective cells
is computed so there isn't an

219
00:16:03,350 --> 00:16:09,030
explanatory gap here.
So individual neurons or nerve

220
00:16:09,030 --> 00:16:13,510
cells can respond selectively to
certain combinations of

221
00:16:13,510 --> 00:16:19,070
features, and when they respond,
their activity often increases.

222
00:16:19,710 --> 00:16:23,350
They're firing and such an
activity.

223
00:16:23,630 --> 00:16:26,750
I like also to call a short term
memory trace or.

224
00:16:27,310 --> 00:16:32,110
STM trace.
It's realized in this by a

225
00:16:32,110 --> 00:16:35,950
change in membrane potential.
The main reasons for short term

226
00:16:35,950 --> 00:16:39,190
memory traces.
You know the IT get I can I get

227
00:16:39,190 --> 00:16:44,230
this and that it it is.
It's hard to do in the picture

228
00:16:44,550 --> 00:16:48,030
because of the angle.
The cell gets active for a

229
00:16:48,030 --> 00:16:52,270
little while and then it loses
its activity if nothing else is

230
00:16:52,270 --> 00:16:56,740
happening.
So the pathways from these

231
00:16:56,740 --> 00:17:00,220
active features to the
categories are adaptive.

232
00:17:00,460 --> 00:17:06,099
They can learn the end of each
pathway has a process that's

233
00:17:06,099 --> 00:17:10,579
often called an adaptive wait or
a long term memory trace, as

234
00:17:10,579 --> 00:17:14,900
opposed to the short term memory
trace or activation.

235
00:17:15,780 --> 00:17:19,339
And the long term memory trace
learns in response to

236
00:17:19,740 --> 00:17:23,730
correlations.
Between the active feature

237
00:17:23,730 --> 00:17:28,610
selective cell that's sending
the signal down the pathway and

238
00:17:28,610 --> 00:17:33,370
the category selective cell
that's active when the proceed.

239
00:17:33,610 --> 00:17:40,130
So it's correlation learning and
the category cell learns to

240
00:17:40,130 --> 00:17:44,970
respond not just to 1 pathway,
it learns us to respond to a

241
00:17:44,970 --> 00:17:49,910
spatial pattern of activity.
Across all the feature selective

242
00:17:49,910 --> 00:17:54,270
cells that send the pathways,
we're taken together.

243
00:17:54,390 --> 00:18:00,150
All the pathways taken together
are called an adaptive filter or

244
00:18:00,150 --> 00:18:04,870
a bottom up adaptive filter
because they're enabling the

245
00:18:04,870 --> 00:18:09,110
category to respond selectively,
let's say to a phase or a

246
00:18:09,110 --> 00:18:15,590
painting OK, which has to have a
spatially distributed input, so.

247
00:18:15,950 --> 00:18:20,830
Competitive learning was good as
far as it went, and it responds

248
00:18:20,830 --> 00:18:23,710
well to sparse input
environments.

249
00:18:23,710 --> 00:18:27,830
Well, what does that mean?
In fact, even deep learning

250
00:18:27,830 --> 00:18:30,910
doesn't experience catastrophic
together in a sparse input

251
00:18:30,910 --> 00:18:35,390
environment.
What it means that the input

252
00:18:35,390 --> 00:18:39,990
patterns aren't too close
together in pattern space if you

253
00:18:39,990 --> 00:18:42,910
think of them as points in a
multidimensional space.

254
00:18:43,960 --> 00:18:48,960
And so I published a stable
sparse learning theorem about

255
00:18:48,960 --> 00:18:54,840
competitive learning in 1976,
showing if the input vector

256
00:18:54,840 --> 00:18:57,800
space is sparse.
Everything is good, you learn

257
00:18:58,000 --> 00:19:04,080
nice categories that are stable.
But I then realized that if

258
00:19:04,080 --> 00:19:08,760
there are too many input
patterns for the.

259
00:19:10,460 --> 00:19:13,540
The way you've structured the
network or the inputs that are

260
00:19:13,540 --> 00:19:16,140
presented through time are
related to each other.

261
00:19:16,660 --> 00:19:20,180
Like you might have a
statistical drift in changing

262
00:19:20,180 --> 00:19:24,140
inputs through time.
Then previously learned

263
00:19:24,140 --> 00:19:28,340
categories got washed away by
new learning, so competitive

264
00:19:28,340 --> 00:19:31,500
learning experience is
catastrophic.

265
00:19:31,500 --> 00:19:36,540
Forgetting now that isn't good,
because we know we live in a

266
00:19:36,580 --> 00:19:42,560
looming, buzzing confusion.
Where we're bombarded with

267
00:19:42,560 --> 00:19:49,520
inputs and nothing is sparse,
everything is overlapping and.

268
00:19:49,520 --> 00:19:55,320
But in 1975, as luck would have
it, I was studying reinforcement

269
00:19:55,320 --> 00:19:59,680
learning and I realized the
importance of a learned top down

270
00:19:59,680 --> 00:20:04,360
expectation.
So just like a bottom up adapted

271
00:20:04,360 --> 00:20:12,580
pathway, a top down expectation.
Has an adaptive way or a long

272
00:20:12,580 --> 00:20:16,940
term memory trace.
So you have a category and it's

273
00:20:16,940 --> 00:20:21,140
going to send the signal to a
feature selective cell up that

274
00:20:21,940 --> 00:20:26,700
and there's an adaptive way of
budding that feature selective

275
00:20:26,700 --> 00:20:29,740
cell.
And if you look at all of the

276
00:20:29,740 --> 00:20:33,940
pathways from that category to
all the feature selective cells

277
00:20:33,940 --> 00:20:38,230
and all their long term memory
traces, I say it learned the top

278
00:20:38,230 --> 00:20:43,470
down expectation because it
learned patterns of activity

279
00:20:43,470 --> 00:20:46,550
across the teacher selective
cells when the category was

280
00:20:46,550 --> 00:20:49,270
active.
And so if you read out that

281
00:20:49,270 --> 00:20:51,630
category it would expect to see
that pattern.

282
00:20:53,990 --> 00:20:57,590
So taken together, all the long
term memory traces in the top

283
00:20:57,590 --> 00:21:01,630
down pathways from a category
cell to the feature selective

284
00:21:01,630 --> 00:21:06,070
cells is therefore called a
learned expectation.

285
00:21:07,800 --> 00:21:13,040
So I then realized from my
reinforcement learning work, you

286
00:21:13,040 --> 00:21:16,200
know here we're facing
catastrophic forgetting

287
00:21:16,200 --> 00:21:20,560
competitive learning.
I then realized that adding

288
00:21:20,560 --> 00:21:24,400
these learned top down
expectations to a category

289
00:21:24,400 --> 00:21:29,080
learning network solves this
Billy plasticity dilemma i.e.

290
00:21:29,320 --> 00:21:31,360
It prevents catastrophic
forgetting.

291
00:21:32,760 --> 00:21:35,760
And so how does adaptive
resonance get into the story?

292
00:21:36,740 --> 00:21:41,140
Because when both the bottom up
adaptive filter and the top down

293
00:21:41,140 --> 00:21:45,620
expectation is simultaneously
active, you have a positive

294
00:21:45,620 --> 00:21:49,660
feedback loop and a resonance
can develop.

295
00:21:50,620 --> 00:21:55,620
So when all resonances occurs,
when they're excitatory feedback

296
00:21:55,620 --> 00:21:59,900
signals between the bottom up
and the top down pathways

297
00:22:00,340 --> 00:22:04,540
between two or more brain
regions, and when these patterns

298
00:22:05,960 --> 00:22:10,840
match their signals well enough
to cause the active cells to

299
00:22:10,840 --> 00:22:15,400
synchronize.
You know, a lot of brain rhythms

300
00:22:15,400 --> 00:22:21,400
synchronize, and because it's a
positive feedback loop they get,

301
00:22:21,440 --> 00:22:27,960
they sustain this synchronized
activity at enhanced activity

302
00:22:27,960 --> 00:22:30,320
level.
And that's enough to trigger

303
00:22:30,320 --> 00:22:35,360
learning, but in a very
selective fashion.

304
00:22:36,350 --> 00:22:40,390
And that's why I called the
theory adaptive resonance

305
00:22:40,390 --> 00:22:44,550
theory.
And you solve the catastrophic

306
00:22:44,550 --> 00:22:48,550
forgetting problem because as
I'll say a little more clearly

307
00:22:48,550 --> 00:22:53,310
in a moment, you suppress
outlier or irrelevant teachers,

308
00:22:53,750 --> 00:22:57,550
you only learn relevant stuff
and you dynamically stabilize

309
00:22:57,550 --> 00:23:01,430
that learning.
So, but let me just make a

310
00:23:01,430 --> 00:23:05,390
summary state before I move to
the next related theme.

311
00:23:05,390 --> 00:23:10,590
That art explains how multiple
processing stages in our brains

312
00:23:11,190 --> 00:23:16,510
use excitatory bottom up,
adaptive filters and top down

313
00:23:16,510 --> 00:23:22,110
learned expectations to attend.
Because the expectations help us

314
00:23:22,110 --> 00:23:25,390
to pay attention to what we
expect to see based on past

315
00:23:25,390 --> 00:23:32,110
experience, to attend,
recognize, and also predict

316
00:23:32,600 --> 00:23:36,440
objects and events in a changing
world without experiencing

317
00:23:36,440 --> 00:23:46,440
catastrophic forgetting.
So in this way, art showed that

318
00:23:46,440 --> 00:23:50,840
there's an intimate link in our
brains between processes of

319
00:23:50,840 --> 00:24:01,520
learning, expectation,
attention, resonance, and

320
00:24:01,520 --> 00:24:06,370
synchrony.
But I gradually realized that I

321
00:24:06,450 --> 00:24:11,810
was explaining a ton of data
about the parametric properties

322
00:24:12,850 --> 00:24:16,770
of real time behaviors that were
conscious behaviors.

323
00:24:17,770 --> 00:24:23,210
So through the back door, as it
were, I was led to begin to

324
00:24:23,210 --> 00:24:27,210
understand consciousness,
because these adaptive

325
00:24:27,210 --> 00:24:31,390
resonances were supporting a
particular kind of conscious

326
00:24:31,390 --> 00:24:33,550
awareness.
That I will clarify more in a

327
00:24:33,550 --> 00:24:36,230
moment.
So in other words, ought

328
00:24:36,230 --> 00:24:40,670
proposed an intimate link in our
brains between processes of

329
00:24:41,030 --> 00:24:47,870
consciousness, learning,
expectation, attention,

330
00:24:47,870 --> 00:24:53,150
resonance, and synchrony.
And those letters CLEARS spell

331
00:24:53,150 --> 00:24:57,290
out the word clears.
So I like saying that art

332
00:24:57,290 --> 00:25:01,610
mechanistically spraying links
that occurs between the clearest

333
00:25:01,610 --> 00:25:07,610
processes in our brains and it's
a lot.

334
00:25:07,610 --> 00:25:14,130
Learn top down expectations, as
I commented before, that protect

335
00:25:14,130 --> 00:25:16,970
us against catastrophic
forgetting and solve this Billy

336
00:25:16,970 --> 00:25:21,780
Plasticity dilemma.
But for that to happen, you

337
00:25:21,780 --> 00:25:25,700
can't just say, oh let's throw
in any top down expectation.

338
00:25:26,420 --> 00:25:30,460
For example, if you use the kind
of inhibitory matching that you

339
00:25:30,460 --> 00:25:34,900
have a Bayesian models, you
couldn't solve the problem and

340
00:25:34,900 --> 00:25:37,020
moreover you couldn't learn a
damn thing.

341
00:25:38,300 --> 00:25:43,740
So the kind of matching in ART
that ensures this property obeys

342
00:25:43,740 --> 00:25:47,940
what I've come to call the art
matching rule puts top down

343
00:25:47,940 --> 00:25:52,070
expectations and matching
against bottom up patterns.

344
00:25:52,870 --> 00:25:59,590
And I was led with GAIL
Carpenter to realize that it's

345
00:25:59,590 --> 00:26:07,230
realized by a top down
modulatory on center driving off

346
00:26:07,230 --> 00:26:11,310
surround network.
Well what are those words mean?

347
00:26:12,030 --> 00:26:17,070
First let me just remark that
the existence of the prediction

348
00:26:17,070 --> 00:26:23,330
at an arc matching rule that is
realized by this kind of top

349
00:26:23,330 --> 00:26:27,090
down modulatory on center offs
through our network has been

350
00:26:27,090 --> 00:26:31,770
supported by psychological,
anatomical, physiological data

351
00:26:31,770 --> 00:26:34,970
in multiple species including
bats.

352
00:26:35,730 --> 00:26:39,330
Nobuo Suga came to what
apartment once and he was very

353
00:26:39,330 --> 00:26:42,890
excited about his new results
about top down matching in bats

354
00:26:42,930 --> 00:26:46,250
and the circuit obey the art
matching rule that had been

355
00:26:46,250 --> 00:26:51,040
conserved at least from bats to
humans and monkeys.

356
00:26:52,320 --> 00:27:00,440
So matching occurs when a top
down expectation activates the

357
00:27:00,440 --> 00:27:06,720
modulatory on cineworks around
net and does so against the

358
00:27:06,960 --> 00:27:11,840
bottom up input pattern.
And the question is do they

359
00:27:11,840 --> 00:27:18,740
agree well enough for resonance
to occur or is the mismatch so

360
00:27:18,740 --> 00:27:22,020
great that they'll be
suppression and reset and we'll

361
00:27:22,020 --> 00:27:26,460
come to that in a moment, But
the modulatory on center by

362
00:27:26,460 --> 00:27:33,100
itself, let's say you just, I
want to think about, oh, will I

363
00:27:33,100 --> 00:27:37,740
see that today.
So I might want to turn on my

364
00:27:37,740 --> 00:27:44,160
tap face category and I'll read
out a prime Now if whatever read

365
00:27:44,160 --> 00:27:47,000
out of prime I sort have, I'd be
hallucinating.

366
00:27:47,720 --> 00:27:53,200
But I can have visual imagery of
that by volitionally through the

367
00:27:53,200 --> 00:27:57,000
basal ganglia helping the gain
And then you'll go from a

368
00:27:57,000 --> 00:28:00,520
subthreshold prime to a
superthreshold resonance and I

369
00:28:00,520 --> 00:28:03,680
can think about what a nice
looking young man this is.

370
00:28:03,720 --> 00:28:11,960
OK, so but so the modulatory on
center can sensitize modulator

371
00:28:12,490 --> 00:28:17,690
prime the cells to which IT
projects, but it can't fire them

372
00:28:18,290 --> 00:28:21,850
to super threshold levels
without support from a match

373
00:28:21,890 --> 00:28:26,130
that's sufficiently well matched
bottom up input pattern unless

374
00:28:26,130 --> 00:28:31,930
you use additional game control.
For example, if the expectation

375
00:28:31,930 --> 00:28:37,130
is of a friend's face and if I
then saw his face or her face,

376
00:28:37,850 --> 00:28:42,770
I'd recognize them at lower
thresholds and faster.

377
00:28:42,770 --> 00:28:46,130
So in dim lighting you might say
who is that and if it's

378
00:28:46,130 --> 00:28:50,050
unfamiliar you might not know.
But if it's a familiar person,

379
00:28:50,650 --> 00:28:55,090
you have enough matching with
the prime to bring it out of the

380
00:28:55,130 --> 00:29:00,770
background.
So the art matching rule enables

381
00:29:00,770 --> 00:29:08,380
top down expectations to select
it and and learn the critical

382
00:29:08,380 --> 00:29:12,220
feature patterns that control
successful predictions.

383
00:29:12,220 --> 00:29:18,340
Cause the modulatory on center
eventually is learning from a

384
00:29:18,340 --> 00:29:23,300
learn category to read out the
pattern of critical features

385
00:29:24,220 --> 00:29:28,780
that are predictive and it's
those features that are going to

386
00:29:28,780 --> 00:29:33,540
go into the adaptive weights in
the bottom of filters and the

387
00:29:33,540 --> 00:29:39,560
top down expectations.
So there's a link between

388
00:29:39,720 --> 00:29:44,120
attention and prediction here
through this selected learning

389
00:29:44,120 --> 00:29:50,120
process.
So to to summarize, the critical

390
00:29:50,120 --> 00:29:53,640
features are learned by the
bottom up adapter filters, the

391
00:29:53,640 --> 00:30:00,200
top down expectations that carry
the excitatory signals that

392
00:30:00,200 --> 00:30:05,130
support the resonating cells.
At this point, you know I'm

393
00:30:05,130 --> 00:30:08,410
blabbing away oh, you know, this
guy introduced art.

394
00:30:09,050 --> 00:30:11,410
You know, sure he loves it.
It's baby.

395
00:30:11,410 --> 00:30:16,250
Well, why should you believe it?
So of course there are.

396
00:30:17,050 --> 00:30:20,010
There are at least three kinds
of predictive success of which

397
00:30:20,010 --> 00:30:24,250
two of them you'd expect from
any theory.

398
00:30:24,250 --> 00:30:28,410
The first is all the
foundational hypothesis from

399
00:30:28,450 --> 00:30:32,760
which I derived art have been
supported by subsequent

400
00:30:32,760 --> 00:30:37,040
psychological and
neurobiological data, so it's

401
00:30:37,040 --> 00:30:40,520
compatible with all the data
that you need to build the

402
00:30:40,520 --> 00:30:46,160
model, and it's also explained
and simulated data from hundreds

403
00:30:46,160 --> 00:30:50,080
of other experiments, So it has
an unrivaled explanatory range.

404
00:30:50,760 --> 00:30:58,420
But there's a deeper reason why
I think art will survive in some

405
00:30:58,420 --> 00:31:00,220
full form.
Because you know, there are

406
00:31:00,220 --> 00:31:04,100
variants of art, art, one or two
or two, a or three, for the art,

407
00:31:04,100 --> 00:31:10,420
for the art map, distributed art
map, you know, getting to, you

408
00:31:10,420 --> 00:31:18,580
know, a fully embodied theory of
art, of which these are

409
00:31:19,100 --> 00:31:21,260
computationally effective
examples.

410
00:31:21,860 --> 00:31:27,660
So in art, in a famous paper in
Psychological Review that,

411
00:31:27,700 --> 00:31:34,060
published in 80I, derived art
from a thought experiment.

412
00:31:35,140 --> 00:31:39,780
And this is a thought experiment
about how any system can

413
00:31:39,780 --> 00:31:45,860
autonomously learn to correct
predictive errors in a changing

414
00:31:45,860 --> 00:31:48,380
world that's filled with
unexpected events.

415
00:31:49,140 --> 00:31:54,560
So it's not saying mind brain,
blah blah and the hypothesis of

416
00:31:54,640 --> 00:31:59,960
the thought you familiar facts
that we all know from our daily

417
00:31:59,960 --> 00:32:03,200
lives.
And the important thought

418
00:32:03,200 --> 00:32:09,280
experiment is, is that if these
facts from daily life act

419
00:32:09,280 --> 00:32:16,440
together over the millennia,
their evolutionary pressures for

420
00:32:16,560 --> 00:32:22,230
the evolution of our brains,
including how we learn to pay

421
00:32:22,230 --> 00:32:27,470
attention and recognize things.
But these familiar facts, these

422
00:32:27,470 --> 00:32:31,230
effects about the environment,
they don't say anything about

423
00:32:31,230 --> 00:32:37,550
mind or brain.
So I'm forced to conclude I see

424
00:32:37,550 --> 00:32:42,670
no way out of the thought
experiment, that artsy universal

425
00:32:42,670 --> 00:32:50,150
solution of the problem, how any
system can autonomously learn to

426
00:32:50,150 --> 00:32:52,870
recognize and predict the
changing world of children,

427
00:32:52,870 --> 00:32:59,630
unexpected events.
And from that perspective, I

428
00:32:59,630 --> 00:33:04,110
think it's all the more
remarkable that art dynamics

429
00:33:04,110 --> 00:33:09,070
also support conscious states of
seeing, hearing, feeling and

430
00:33:09,070 --> 00:33:14,710
knowing things, and explains how
we can use these conscious

431
00:33:14,710 --> 00:33:18,780
states to effectively plan and
act to realize value.

432
00:33:18,780 --> 00:33:21,940
Gold and I'll explain why that
is in a moment.

433
00:33:22,620 --> 00:33:25,340
I think you have to ask me
another question first.

434
00:33:26,740 --> 00:33:31,940
So the proposed solution of the
mind body problem, in summary,

435
00:33:32,500 --> 00:33:37,500
arises from a computational
analysis of how humans and other

436
00:33:37,500 --> 00:33:41,580
animals or autonomous who learn
in the changing world.

437
00:33:42,500 --> 00:33:46,400
And that's how consciousness cut
in the mix.

438
00:33:46,400 --> 00:33:53,200
Which makes me take for granted
that every animal or other

439
00:33:53,520 --> 00:33:58,840
Organism that can autonomously
solve the stability plasticity

440
00:33:58,840 --> 00:34:02,080
dilemma has a form of
consciousness.

441
00:34:04,040 --> 00:34:06,320
We're not that special, We're
real special.

442
00:34:06,320 --> 00:34:15,130
But and mind and self to come
back to that are possible

443
00:34:15,130 --> 00:34:18,409
because ourselves.
This build plasticity dilemma,

444
00:34:18,810 --> 00:34:22,770
because as a result of that you
can have a lifetime of learning

445
00:34:23,810 --> 00:34:31,130
that all gets embedded in
increasingly complex interacting

446
00:34:31,130 --> 00:34:37,929
awarenesses and abilities and
thoughts about how the world

447
00:34:37,929 --> 00:34:41,290
works.
And collectively, all of them

448
00:34:41,290 --> 00:34:48,960
together are our mind.
And as you go through life and

449
00:34:48,960 --> 00:34:52,840
you accumulate all of this
understanding and attach it to

450
00:34:52,840 --> 00:34:57,280
your own person, you have an
evolving self.

451
00:34:57,840 --> 00:35:02,840
So the self I have today is very
different from the self I had as

452
00:35:02,840 --> 00:35:06,960
a boy, or even when I was in my
30s or 40s.

453
00:35:07,000 --> 00:35:11,840
I'm a quite a different being in
some ways.

454
00:35:13,510 --> 00:35:15,950
Maybe I should stop there.
No, no, no.

455
00:35:15,950 --> 00:35:18,190
I love the fact that you're
going on because this podcast

456
00:35:18,190 --> 00:35:20,950
isn't about you.
So perfect.

457
00:35:20,950 --> 00:35:22,430
The longer the answers, I don't
really mind.

458
00:35:22,670 --> 00:35:24,910
I'm enjoying it.
One thing that has remained

459
00:35:24,910 --> 00:35:27,870
certain and similar throughout
your years, even though yourself

460
00:35:27,870 --> 00:35:32,830
has continuously evolved and
changed, is just your your drive

461
00:35:32,830 --> 00:35:35,950
and tenacity to continue to
publish work.

462
00:35:36,430 --> 00:35:39,310
I mean, just to put certain
things into context, you were 17

463
00:35:39,310 --> 00:35:42,350
when you had this, this famous
epiphany.

464
00:35:42,880 --> 00:35:46,560
And when this happened, you
didn't just think of something.

465
00:35:46,560 --> 00:35:49,040
You completely shifted the way
people think about the

466
00:35:49,040 --> 00:35:52,520
mathematical concepts.
You did exactly what people like

467
00:35:52,520 --> 00:35:56,440
Einstein have done.
You thought of a new way of

468
00:35:56,440 --> 00:35:59,760
thinking about concepts, so you
developed a form of mathematical

469
00:35:59,920 --> 00:36:03,520
systems too.
Well, that's why people call me

470
00:36:03,520 --> 00:36:07,280
the Newton.
Never mind, because just like

471
00:36:07,520 --> 00:36:14,580
Newton, as a boy I introduced
the basic concepts and equations

472
00:36:15,260 --> 00:36:18,420
that are used to understand the
universe of mind.

473
00:36:19,260 --> 00:36:25,100
I'm called the Einstein of the
mind because I use thought

474
00:36:25,100 --> 00:36:28,860
experiments.
Just like Einstein derived both

475
00:36:29,300 --> 00:36:32,020
special relativity and general
relativity from thought

476
00:36:32,020 --> 00:36:36,300
experiments like derived art
from the thought experiment.

477
00:36:37,080 --> 00:36:40,480
And I also derived my cognitive
motional model from a thought

478
00:36:40,480 --> 00:36:43,200
experiment and competitive
networks from thought

479
00:36:43,200 --> 00:36:45,840
experiment.
But you don't just wake up in

480
00:36:45,840 --> 00:36:49,000
the morning to say, hey, what's
my latest thought experiment?

481
00:36:49,760 --> 00:36:53,160
It's only after you already have
a deep understanding of the

482
00:36:53,160 --> 00:36:57,520
targeted class of phenomena that
you can step back and see, well,

483
00:36:57,520 --> 00:37:04,640
what are the simple ideas that
if I carry them out as a logical

484
00:37:04,640 --> 00:37:08,710
story, they're conjoint
implications.

485
00:37:09,470 --> 00:37:12,950
You get to something you never
thought you'd understand the 1st

486
00:37:12,950 --> 00:37:15,190
place.
That's the power of it.

487
00:37:15,190 --> 00:37:18,110
You know, you someone says, oh,
I don't really believe any of

488
00:37:18,110 --> 00:37:22,230
this nonsense, and then you say,
OK, read the thought experiment,

489
00:37:22,230 --> 00:37:27,430
show me where he made a mistake.
Because if you can't, then if

490
00:37:27,430 --> 00:37:30,870
you believe the simple
hypotheses, which we all know

491
00:37:31,270 --> 00:37:34,390
they're not, area die.
They're facts of life.

492
00:37:36,090 --> 00:37:39,530
Then you have to either accept
the conclusion or throw away

493
00:37:40,170 --> 00:37:42,370
your belief in the scientific
method.

494
00:37:43,090 --> 00:37:45,210
Yes.
I mean, your work's been

495
00:37:45,330 --> 00:37:46,850
transformed into different
courses.

496
00:37:46,850 --> 00:37:49,730
People are doing lots of
different variations of it.

497
00:37:49,810 --> 00:37:52,090
It has continued to evolve even
beyond you.

498
00:37:52,170 --> 00:37:54,730
It's almost as if you're I hope
so.

499
00:37:54,890 --> 00:37:56,810
No, it has.
I've just seen versions of it

500
00:37:56,810 --> 00:38:01,640
out there, so it definitely has.
Well, just with art, there are

501
00:38:01,640 --> 00:38:06,320
multiple varieties.
You know when people are

502
00:38:06,320 --> 00:38:13,440
applying art in technology, they
want the most efficient

503
00:38:13,440 --> 00:38:17,760
algorithm to solve a certain
class of problems.

504
00:38:18,760 --> 00:38:22,960
And typically, just as in our
brain, before you get to

505
00:38:22,960 --> 00:38:27,800
infratemporal cortex and
prefrontal cortex and other

506
00:38:28,370 --> 00:38:32,370
cognitive areas, there's a huge
amount of preprocessing, visual

507
00:38:32,370 --> 00:38:38,450
preprocessing, you know, retina,
lateral geniculate nucleus,

508
00:38:38,450 --> 00:38:43,570
V1V2V384.
So all visual preprocessing to

509
00:38:43,570 --> 00:38:48,330
prepare the data to be
classified and to appear basis

510
00:38:48,330 --> 00:38:51,650
for prediction.
Before that, things are just too

511
00:38:51,650 --> 00:38:59,240
noisy.
So so you know a lot of people

512
00:38:59,240 --> 00:39:03,600
are developing varieties of art
to deal with different

513
00:39:03,600 --> 00:39:06,760
engineering and AI and
technology problems. 1

514
00:39:07,400 --> 00:39:11,800
colleague, younger colleague,
but now he's a very senior guy

515
00:39:11,800 --> 00:39:15,480
in his own right.
Is Don Brunch at Missouri

516
00:39:15,640 --> 00:39:21,360
Science and Technology.
I guess it's the Missouri

517
00:39:21,360 --> 00:39:23,320
University of Science and
Technology.

518
00:39:24,080 --> 00:39:30,040
John visited us as a young man,
a very A student and he got

519
00:39:30,040 --> 00:39:35,600
fascinated with art and related
models like like mine.

520
00:39:36,800 --> 00:39:40,640
I mean I've I've been lucky to
have a number of people early on

521
00:39:41,400 --> 00:39:47,040
who fell in love with it.
Like, I don't know if you know

522
00:39:47,040 --> 00:39:55,170
who Robert Heckneilson was.
Robert used to work in a company

523
00:39:55,170 --> 00:39:59,530
in California and he would fly
to have lunch with me when I was

524
00:39:59,530 --> 00:40:02,210
still at MIT.
And we talked about neural

525
00:40:02,210 --> 00:40:06,090
networks and he founded a
company which was originally

526
00:40:06,090 --> 00:40:11,730
called, Heck Nielsen Neura
Computer, and then it became HNC

527
00:40:13,490 --> 00:40:16,250
Corporation.
Then it got eaten by a bigger

528
00:40:16,250 --> 00:40:19,970
corporation.
But you know his whole reason

529
00:40:19,970 --> 00:40:24,390
for getting in technology.
Was I inspired him when he was

530
00:40:24,390 --> 00:40:29,950
doing his PhD work?
He's been dead for a while.

531
00:40:30,030 --> 00:40:37,030
Said he died of a heart attack.
So I've been very lucky that you

532
00:40:37,030 --> 00:40:41,790
know some such pioneering souls
were inspired by another

533
00:40:41,790 --> 00:40:45,790
pioneering soul.
But there were a lot of vested

534
00:40:45,790 --> 00:40:52,090
interests who would wish I'd go
away cause either they didn't

535
00:40:52,090 --> 00:40:59,250
have the preparation to compete,
although I don't think it in

536
00:40:59,250 --> 00:41:03,170
those terms, or they were so in
love with what they were

537
00:41:03,170 --> 00:41:11,690
currently doing that they wanted
any concept that there wasn't

538
00:41:11,890 --> 00:41:16,650
this wasn't the best game in
town to go bye bye, yes so.

539
00:41:17,010 --> 00:41:19,570
When you look at this thought
experiment and you and you.

540
00:41:20,260 --> 00:41:22,300
Delve into the dynamical
systems, the mathematical

541
00:41:22,300 --> 00:41:25,500
processes you're talking about.
It's very clear, and it's easy

542
00:41:25,500 --> 00:41:27,940
to see how this sort of makes
sense when you think of

543
00:41:27,940 --> 00:41:32,420
consciousness as a verb rather
than a noun or something that's

544
00:41:32,420 --> 00:41:36,220
constantly happening, rather
than just one single thing.

545
00:41:37,060 --> 00:41:41,340
It's very intuitive, but yet it
takes such a complicated theory

546
00:41:41,340 --> 00:41:44,520
to explain the situation.
Tell me about the cognitive

547
00:41:44,520 --> 00:41:47,800
emotional motor and and how
combining it with art leads to

548
00:41:47,800 --> 00:41:50,240
fascinating results and
consciousness as well as our

549
00:41:50,240 --> 00:41:52,920
understanding of mental health
and mental illness.

550
00:41:54,640 --> 00:41:59,000
OK, that's a big question too.
But when I I really I like it.

551
00:41:59,000 --> 00:42:03,400
I hope I do justice.
So this so-called cognitive

552
00:42:03,400 --> 00:42:06,040
emotional modem model or cog and
model.

553
00:42:06,840 --> 00:42:13,440
It can also be derived from the
thought experiment, and I've

554
00:42:13,440 --> 00:42:17,520
used it to explain a lot of data
about how cognition or thinking

555
00:42:18,240 --> 00:42:25,280
and emotion or feeling interact
in order to choose actions.

556
00:42:25,280 --> 00:42:27,560
That's why it's the cognitive
emotional Motor model.

557
00:42:28,040 --> 00:42:32,960
Actions that'll realize value
holds because art by itself has

558
00:42:32,960 --> 00:42:39,120
no concept of values.
You need emotions, feelings to

559
00:42:39,160 --> 00:42:44,020
direct your thoughts, to valued
goals, worthwhile goals.

560
00:42:45,300 --> 00:42:50,100
And so it's at this point I like
distinguishing 6 different kind

561
00:42:50,100 --> 00:42:52,780
of adaptive resonances that I've
studied.

562
00:42:56,700 --> 00:42:59,700
They support different aspects
of consciousness.

563
00:42:59,700 --> 00:43:02,740
You were saying that's what's
triggering this thought with

564
00:43:02,740 --> 00:43:05,820
different functions in different
parts of our brains.

565
00:43:06,420 --> 00:43:09,580
So I'll start with cognitive
emotional resonances.

566
00:43:09,580 --> 00:43:15,270
So not surprisingly, they
support conscious feeling of

567
00:43:15,270 --> 00:43:18,430
emotions and also knowing their
source.

568
00:43:18,990 --> 00:43:24,390
Oh, I love that person.
You know, the feeling and the

569
00:43:24,510 --> 00:43:28,990
target of the feeling.
But then there are five others.

570
00:43:29,350 --> 00:43:32,030
This is not exhausted.
It's just what I've managed to

571
00:43:32,030 --> 00:43:35,790
do with colleagues.
They're what I call surface

572
00:43:35,790 --> 00:43:40,290
shroud resonances, which I'll
explain a moment support

573
00:43:40,290 --> 00:43:44,930
conscious seeing of visual
objects and scenes.

574
00:43:45,930 --> 00:43:48,570
Then there are the feature
category resonances.

575
00:43:48,570 --> 00:43:53,090
This is what the original art is
about that supports conscious

576
00:43:53,450 --> 00:43:57,770
recognition of visual objects
and scenes.

577
00:43:58,410 --> 00:44:02,170
So seeing and recognition are
different.

578
00:44:03,250 --> 00:44:08,940
So you know, seeing or hearing
or any percept is general

579
00:44:08,940 --> 00:44:12,500
purpose.
You open your eyes, you see

580
00:44:13,300 --> 00:44:19,020
someone makes a sound, you hear.
Recognition requires prior

581
00:44:19,020 --> 00:44:22,020
learning, it's not general
purpose.

582
00:44:22,820 --> 00:44:27,740
So all all that time, the
feature category resonances of

583
00:44:27,740 --> 00:44:31,860
art were about conscious
recognition and it didn't say a

584
00:44:31,900 --> 00:44:37,050
thing about seeing.
Then there are stream shroud

585
00:44:37,050 --> 00:44:41,890
resonances that support
conscious hearing of auditory

586
00:44:41,890 --> 00:44:45,890
objects and streams.
They're what I call the spectral

587
00:44:46,170 --> 00:44:51,330
pitch and tamper resonances that
support conscious recognition of

588
00:44:51,610 --> 00:44:55,130
auditory objects and streams,
including things like music.

589
00:44:56,130 --> 00:44:59,570
And they're the item list
resonances that support

590
00:44:59,970 --> 00:45:02,890
conscious recognition of speech
and language.

591
00:45:03,510 --> 00:45:06,710
And these are the foundational
kind of resonances that are

592
00:45:07,470 --> 00:45:13,390
really used big time in the
prefrontal cortex and organizing

593
00:45:14,550 --> 00:45:18,710
sequences of thoughts and
beliefs and actions.

594
00:45:22,030 --> 00:45:25,070
So I already said perception is
general purpose.

595
00:45:25,710 --> 00:45:30,430
You have it in response to both
familiar and unfamiliar events.

596
00:45:30,470 --> 00:45:35,210
But recognition is only to
familiar things you've already

597
00:45:35,210 --> 00:45:39,370
learned about, so one must never
conflate them.

598
00:45:41,210 --> 00:45:46,970
So now, without mental health, I
just have very sketchy things

599
00:45:46,970 --> 00:45:48,810
here.
Such a huge topic.

600
00:45:51,650 --> 00:45:58,510
Well, first, I think it's really
useful to have brain theories

601
00:45:58,510 --> 00:46:02,390
that explain a lot of facts
about normal or typical

602
00:46:02,390 --> 00:46:09,510
behavior, because it's only when
such processes get imbalanced or

603
00:46:09,510 --> 00:46:15,550
damaged in certain ways that the
behavior is no longer typical.

604
00:46:16,470 --> 00:46:22,990
And if it gets imbalanced enough
to lead to symptoms that people

605
00:46:22,990 --> 00:46:27,640
find not very sociable, then
we'll call it a mental disorder.

606
00:46:28,600 --> 00:46:31,760
So I was led through the back
door, just like with

607
00:46:31,760 --> 00:46:37,440
consciousness into mental
disorders by noticing, hey, if

608
00:46:37,440 --> 00:46:41,280
the following processes and
cognitive and cognitive

609
00:46:41,280 --> 00:46:45,920
emotional dynamics get
imbalanced, symptoms of

610
00:46:45,920 --> 00:46:52,040
Alzheimer's or autism or amnesia
or post traumatic stress

611
00:46:52,040 --> 00:46:57,150
disorder, etcetera, pop out.
So it was the gift that keeps on

612
00:46:57,150 --> 00:47:01,190
giving, but that gift just opens
the door to you.

613
00:47:02,110 --> 00:47:07,910
Then you have to start doing the
hard work of justifying your

614
00:47:07,910 --> 00:47:11,910
lead by explaining and
predicting really hard data,

615
00:47:12,270 --> 00:47:18,710
which I did with a number of
students in various papers and a

616
00:47:18,710 --> 00:47:25,400
theme and a lot of the work, not
the only design theme is one of

617
00:47:25,400 --> 00:47:31,800
opponent processes, which I
model using a ubiquitously

618
00:47:31,800 --> 00:47:38,880
occurring circuit that I call a
gated dipole dipole.

619
00:47:38,880 --> 00:47:45,040
Their opponent on and off cells
that are competing and the gates

620
00:47:45,040 --> 00:47:47,840
are transmitters, chemical
transmitters.

621
00:47:48,440 --> 00:47:51,680
These are like the medium term
memories I earlier mentioned

622
00:47:52,340 --> 00:47:56,740
that can habituate if you use
them, and that can cause all

623
00:47:56,740 --> 00:48:02,020
sorts of fascinating
explanation.

624
00:48:02,020 --> 00:48:11,500
So for example, a gated dipole.
Let's say I turn on a shock and

625
00:48:11,500 --> 00:48:14,340
you're a poor pigeon.
You're in a Skinner box, the

626
00:48:14,340 --> 00:48:20,310
floor is electrified, and you're
running around frantically

627
00:48:20,310 --> 00:48:23,950
trying to keep your feet off the
floor to minimize the pain,

628
00:48:23,990 --> 00:48:27,630
trying to escape.
And you accidentally bump into a

629
00:48:27,630 --> 00:48:35,670
red buzzer and the shock turns
off and you feel a wave of

630
00:48:35,670 --> 00:48:38,990
relief.
So there's an antagonistic

631
00:48:38,990 --> 00:48:44,150
rebound between pain and fear
and relief.

632
00:48:45,440 --> 00:48:50,280
And relief is a positive
motivation that you can use to

633
00:48:50,280 --> 00:48:54,280
do many positive activities,
including.

634
00:48:54,280 --> 00:48:58,480
It's one of the foundations for
the relaxation, response and

635
00:48:58,480 --> 00:49:02,440
yoga, but you can use it to
learn an escape reaction.

636
00:49:03,400 --> 00:49:07,840
So if the shock turns on again,
you can look toward the buzzer

637
00:49:07,840 --> 00:49:12,880
and run to it to shut the shock
off for escape and avoidance.

638
00:49:15,730 --> 00:49:20,090
So these opponent processes are
very important and the question

639
00:49:20,090 --> 00:49:26,770
is what is the energy source
that enables relief to be

640
00:49:26,770 --> 00:49:29,690
triggered?
Because after all, all you did

641
00:49:29,690 --> 00:49:37,250
was shut off the shock, so the
input to the fear center is off.

642
00:49:37,850 --> 00:49:40,770
Well, could just have gone quiet
and you did nothing.

643
00:49:40,850 --> 00:49:45,720
You just slump over.
But there is an arousal source

644
00:49:46,200 --> 00:49:50,480
to both the on and off channel
that tonically arouses to them.

645
00:49:51,840 --> 00:49:54,000
And when they're just being
aroused because they're

646
00:49:54,000 --> 00:49:59,240
competing, they have no output.
And if there is a fearful cue

647
00:49:59,240 --> 00:50:01,760
coming in, it upsets the
balance.

648
00:50:01,760 --> 00:50:07,760
So you get an on response and
you feel fear and pain and when

649
00:50:07,760 --> 00:50:12,360
you shut the fearful input off,
now we have arousal, but because

650
00:50:12,360 --> 00:50:17,000
of the medium term memory in the
on channel it habituated that

651
00:50:17,440 --> 00:50:21,040
transmitter got weaker.
So the net input to the on

652
00:50:21,040 --> 00:50:26,400
channel is now less than the
unhabituated input to the off

653
00:50:26,400 --> 00:50:31,040
channel, which triggers A
rebound of relief.

654
00:50:31,600 --> 00:50:36,040
And that's transient because
during that relief rebound that

655
00:50:36,040 --> 00:50:40,570
the transmit of the off channel
also habituates until they're

656
00:50:40,570 --> 00:50:44,010
equal again, and so you get a
transient rebound.

657
00:50:44,010 --> 00:50:47,770
Otherwise, if you didn't have
that situation or medium term

658
00:50:47,770 --> 00:50:53,170
memory, you'd be feeling tonic
relief forever and your life

659
00:50:53,170 --> 00:51:00,010
would be very different.
So the problem is, how do you

660
00:51:00,010 --> 00:51:05,570
choose the arousal level?
There has to be a kind of golden

661
00:51:05,570 --> 00:51:10,580
mean in the middle.
It's just enough arousal to

662
00:51:10,780 --> 00:51:16,460
activate a rebound, but not too
much arousal.

663
00:51:16,460 --> 00:51:23,220
And I I without explaining,
there's simple things I could

664
00:51:23,220 --> 00:51:25,100
tell you, but everything would
take too long.

665
00:51:25,100 --> 00:51:30,180
Then if you have too little
arousal you're you have kind of

666
00:51:30,260 --> 00:51:37,360
under aroused depression and and
then in response to a bigger

667
00:51:37,360 --> 00:51:39,880
than normal input you can be
hypersensitive.

668
00:51:39,920 --> 00:51:45,360
That's the paradox.
And you can get that in ADHD for

669
00:51:45,360 --> 00:51:51,800
example.
And if you're over aroused then

670
00:51:51,800 --> 00:51:54,840
you always have enough arousal,
but you're very insensitive,

671
00:51:54,840 --> 00:51:58,880
you're hyposensitive to inputs.
You could get that in

672
00:51:59,400 --> 00:52:03,200
schizophrenia, for example, very
insensitive.

673
00:52:04,660 --> 00:52:07,980
Anyway, so and and those
predictions, I hope I didn't say

674
00:52:07,980 --> 00:52:15,180
it normally, have been supported
by subsequent experiments

675
00:52:16,940 --> 00:52:21,660
precisely as I predicted them.
Then there are very other

676
00:52:22,060 --> 00:52:25,980
different things, like when
you're learning in art.

677
00:52:26,780 --> 00:52:29,740
I I didn't talk about
complementary computing, of the

678
00:52:29,740 --> 00:52:32,850
fact there's an intentional
system where you're learning

679
00:52:32,850 --> 00:52:34,810
categories.
And then there's an orienting

680
00:52:34,810 --> 00:52:39,970
system which is sensitive to
novelty and lets you reset your

681
00:52:39,970 --> 00:52:42,410
attentional focus if something
new happens.

682
00:52:42,410 --> 00:52:44,370
Or if you're searching for a
better answer.

683
00:52:45,250 --> 00:52:48,570
And the question is how
sensitive is your novelty

684
00:52:48,570 --> 00:52:52,090
system?
And it turns out there's a

685
00:52:52,090 --> 00:52:57,810
parameter not called vigilance.
And you know, if your vigilance

686
00:52:57,810 --> 00:53:02,790
isn't working well or breaks
down entirely, that's one of the

687
00:53:02,790 --> 00:53:05,630
things that can go wrong in
Alzheimer's and autism.

688
00:53:06,950 --> 00:53:11,070
And you know, after a number of
years of doing vigilance control

689
00:53:12,190 --> 00:53:16,550
simulations and explanations
with GAIL Carpenter and John

690
00:53:16,550 --> 00:53:21,750
Merrill and some other people,
Max Resachi and I in 2008 were

691
00:53:21,750 --> 00:53:28,350
able to show how the nucleus
base Alice of minors responds to

692
00:53:28,910 --> 00:53:36,730
mismatches, giving you an
arousal burst into layer five

693
00:53:36,730 --> 00:53:42,490
across major parts of cerebral
cortex and releases

694
00:53:42,490 --> 00:53:45,770
acetylcholine.
And that can cause a reset.

695
00:53:46,490 --> 00:53:52,810
And if that if the tonic and
fasic vigilance controls is not

696
00:53:53,650 --> 00:53:58,370
tuned right, you can have big
problems with mental disorders.

697
00:54:00,200 --> 00:54:03,320
But you know, I I make that
statement but to explain it you

698
00:54:03,320 --> 00:54:05,280
know we'd be here till the cows
come home.

699
00:54:06,680 --> 00:54:13,000
But, but I do, having mentioned
surface route resonances, I

700
00:54:13,000 --> 00:54:21,120
think it's an opportunity to
comment about why there's a

701
00:54:21,120 --> 00:54:29,900
link, for example, between
conscious seeing and looking and

702
00:54:29,900 --> 00:54:37,340
reaching or between conscious
hearing and communication and

703
00:54:37,340 --> 00:54:45,380
speaking and so on.
So let me go right into the

704
00:54:45,380 --> 00:54:51,300
surface shroud resonances
whereby we consciously see

705
00:54:51,900 --> 00:54:56,660
visual objects and scenes.
Explain why I think that's true

706
00:54:57,920 --> 00:55:03,920
and you know some of the facts
behind it and why the conscious

707
00:55:03,920 --> 00:55:07,160
seeing is used to control
action.

708
00:55:09,120 --> 00:55:12,080
And we'll start at the retina,
you know, the photosensitive

709
00:55:12,080 --> 00:55:16,160
retina, That's where light is
registered in an eye.

710
00:55:16,960 --> 00:55:21,360
And then the light activated
signals from the retinal cells

711
00:55:21,360 --> 00:55:23,080
that are all over the back of
the eye.

712
00:55:23,800 --> 00:55:28,370
They come together in axons or
pathways, and they're bundled

713
00:55:28,370 --> 00:55:33,450
together into the optic nerve so
they can go up to the brain and

714
00:55:33,450 --> 00:55:37,530
give you spatially patterned
signals like it's a face.

715
00:55:37,530 --> 00:55:42,330
It's a pain thing, so it goes up
to the optic nerve, to the

716
00:55:42,330 --> 00:55:53,210
brain.
But that's why the retina has

717
00:55:53,210 --> 00:55:59,930
the big blind spot, the spot
where visual signals aren't

718
00:55:59,930 --> 00:56:04,050
registered, because that's the
place on the retina where you're

719
00:56:04,050 --> 00:56:07,570
forming the optic nerve.
And you might say big deal, who

720
00:56:07,570 --> 00:56:11,050
cares?
Well, no lights registered

721
00:56:11,050 --> 00:56:13,890
positions of the blind spot.
But the blind spot, if you look

722
00:56:13,890 --> 00:56:22,410
down on a retinal image, the
biggest, the fovea and the fovea

723
00:56:22,410 --> 00:56:25,170
is the part of the retina which
we have.

724
00:56:25,170 --> 00:56:30,290
Our high visual acuity in the we
do a few psychotic eye movements

725
00:56:30,290 --> 00:56:34,690
every second is to point off
ovia to the objects that

726
00:56:34,690 --> 00:56:36,850
interest us so that we can
really see them.

727
00:56:38,850 --> 00:56:42,970
So the bottom line is, we could
not look at or reach for an

728
00:56:42,970 --> 00:56:47,490
object that happened to be
registered on our retinas at the

729
00:56:47,490 --> 00:56:51,250
position of the blind spot
without further processing.

730
00:56:52,730 --> 00:56:56,740
But that could lead to disaster
because it's as big as your

731
00:56:56,740 --> 00:57:00,980
fovea where you do all your
important visual processing.

732
00:57:02,060 --> 00:57:06,420
So you have to ask, why aren't
we aware of the missing visual

733
00:57:06,420 --> 00:57:11,220
signals through the blind spot?
And how do we complete the

734
00:57:11,220 --> 00:57:14,700
retinal representation so that
you can look and reach for

735
00:57:14,700 --> 00:57:19,780
objects in these positions.
And here you know you can only

736
00:57:20,060 --> 00:57:25,900
marvel at the beauty of or
design.

737
00:57:25,900 --> 00:57:30,580
Cause one fact that people may
not know is that even though we

738
00:57:30,580 --> 00:57:34,420
think if I'm looking at you as a
stationary image, I think

739
00:57:34,940 --> 00:57:38,620
everything's stationary but no.
Why are you jiggling very

740
00:57:38,620 --> 00:57:45,260
rapidly in the orbit?
And these jiggling movements are

741
00:57:45,260 --> 00:57:48,940
refreshing signals that I'm
getting from stationary objects

742
00:57:49,820 --> 00:57:54,260
so that I can keep seeing you.
Otherwise you would fade, you

743
00:57:54,260 --> 00:58:00,300
would have gone spilt.
And this transient signals that

744
00:58:00,300 --> 00:58:04,940
are going into response to this
jiggle are refreshing the visual

745
00:58:04,940 --> 00:58:10,820
signals so you could see.
And did you see Jurassic Park?

746
00:58:12,340 --> 00:58:16,260
Well that is a very funny, funny
example of this.

747
00:58:16,260 --> 00:58:21,900
Cause at a certain point there's
the ravenous T Rex blooming over

748
00:58:22,540 --> 00:58:29,530
the guy and they hollered him.
Stand still, can't see unless

749
00:58:29,530 --> 00:58:33,970
there's relative motion.
But of course the T Rex just

750
00:58:33,970 --> 00:58:38,730
moved its head, saw him and
devoured him in one big bite.

751
00:58:39,610 --> 00:58:44,570
So.
So even with the refresh you

752
00:58:44,570 --> 00:58:49,610
still have an incomplete retinal
image with a big hole in it.

753
00:58:50,570 --> 00:58:55,510
And how is that image completed
and just here I could?

754
00:58:55,510 --> 00:58:57,150
Both eyes as well, Yeah.
Yeah.

755
00:58:58,030 --> 00:59:01,390
Oh, in both eyes, yeah.
I just want, you know, I want to

756
00:59:01,390 --> 00:59:04,430
talk about one thing at a time.
But it's happening in both eyes.

757
00:59:04,750 --> 00:59:13,670
We have two big holes and you
complete the retinal image of

758
00:59:13,670 --> 00:59:20,870
the blind spot higher up in our
dual cortex by multiple stages

759
00:59:21,660 --> 00:59:25,020
of what I call boundary
completion and surface filling.

760
00:59:25,020 --> 00:59:28,860
In.
Cause I've shown that 3D

761
00:59:28,860 --> 00:59:34,460
boundaries and surfaces properly
understood are the functional

762
00:59:34,460 --> 00:59:41,620
units of conscious vision.
And I can't give you a now a

763
00:59:41,620 --> 00:59:46,380
whole theory of how boundaries
and surfaces are formed.

764
00:59:46,380 --> 00:59:52,010
But I can say that the process
whereby incomplete boundary and

765
00:59:52,010 --> 00:59:57,130
surface representations are
completed takes multiple

766
00:59:57,130 --> 01:00:02,330
processing stages, which I like
to call a hierarchical

767
01:00:02,770 --> 01:00:07,570
resolution of uncertainty.
You know their principles of

768
01:00:07,570 --> 01:00:11,930
uncertainty, complementarity and
resonance in our brains.

769
01:00:11,930 --> 01:00:15,690
This is one of the uncertainty
principles and if you think

770
01:00:15,690 --> 01:00:18,710
about physics.
Lots of physical theories,

771
01:00:18,710 --> 01:00:23,870
notably quantum theory, have
principles of uncertainty,

772
01:00:23,870 --> 01:00:28,910
complementarity, and resonance.
And so you are fascinated by the

773
01:00:28,910 --> 01:00:33,150
question of, well, how are our
principles related?

774
01:00:33,150 --> 01:00:37,870
If it all to the principles if
the physical world with which

775
01:00:37,870 --> 01:00:42,710
our our ancestors have
ceaselessly interacted while our

776
01:00:42,710 --> 01:00:47,190
brains were evolving?
And that is a question for the

777
01:00:47,670 --> 01:00:53,350
next generation okay.
But how does the brain know at

778
01:00:53,350 --> 01:00:59,910
what stage of this hierarchical
resolution of uncertainty the

779
01:00:59,910 --> 01:01:05,070
cortical surface representation
is complete enough to control

780
01:01:05,070 --> 01:01:12,150
looking and reaching?
Turns out I believe that happens

781
01:01:12,150 --> 01:01:17,980
in prestride cortical area V4.
Remember, there's V1V2V3 a V4,

782
01:01:19,140 --> 01:01:24,220
and I claim that a resonance
with a completed surface

783
01:01:24,220 --> 01:01:31,340
representation in V4 and the
next processing stage in

784
01:01:31,340 --> 01:01:36,380
posterior parietal cortex lights
up The surface representation,

785
01:01:36,380 --> 01:01:40,980
chooses it and renders it
consciously visible.

786
01:01:41,820 --> 01:01:45,570
And you're going to use the
consciously seen surface to

787
01:01:45,570 --> 01:01:49,770
control looking in region, not
the partial crap in all the

788
01:01:49,770 --> 01:01:55,690
previous stages.
And moreover, that the spatial

789
01:01:55,690 --> 01:02:01,890
attention from posterior potal
cortex 34 fits itself to the

790
01:02:01,890 --> 01:02:05,130
shape of the surface.
It forms a kind of shroud around

791
01:02:05,130 --> 01:02:12,890
it, an intentional shroud with
Chris Christopher Tyler named a

792
01:02:12,890 --> 01:02:15,820
shroud, but he was thinking of a
different thing.

793
01:02:17,980 --> 01:02:23,460
So I call this a surface shroud.
Resonance lights up the surface,

794
01:02:23,460 --> 01:02:28,140
makes it visually conscious, and
that's used for looking and

795
01:02:28,140 --> 01:02:32,020
reaching via the posterior
cortex.

796
01:02:32,020 --> 01:02:41,940
So top down from PPC to V4, you
have spatial attention that

797
01:02:41,940 --> 01:02:46,040
forms the shroud that enabled
the surface route resonance to

798
01:02:46,040 --> 01:02:49,680
occur.
But when it's resonating bottom

799
01:02:49,680 --> 01:02:54,400
up from PPC to the rest of the
brain, it's known it helps

800
01:02:54,400 --> 01:02:59,040
regulate looking and reaching.
So PPC is an interface between

801
01:02:59,040 --> 01:03:05,280
conscious seeing and action.
So I call it This resonance is

802
01:03:05,280 --> 01:03:10,100
surface route resonance because
the spatial attention that

803
01:03:10,100 --> 01:03:13,700
lights up the surface fits it to
the shape of the surface like a

804
01:03:13,700 --> 01:03:17,100
shroud.
And the surface shroud resonance

805
01:03:17,100 --> 01:03:23,180
supports conscious seeing.
And the shroud in PPC whose

806
01:03:23,180 --> 01:03:27,940
shape is now mirroring the
surface, knows where to direct

807
01:03:27,940 --> 01:03:32,780
signals to look and reach.
But you have to solve the number

808
01:03:32,780 --> 01:03:37,100
of coordinate change problems to
do it, because you're looking

809
01:03:37,100 --> 01:03:39,940
and head center coordinates,
You're reaching and body center

810
01:03:39,940 --> 01:03:42,860
coordinates.
And with colleagues like Dan

811
01:03:42,860 --> 01:03:47,660
Bullock and the Grieve and Frank
Gunther, we spend a lot of work

812
01:03:47,660 --> 01:03:52,300
showing how we self organize
these coordinate representations

813
01:03:52,820 --> 01:04:00,500
in bodies whose relationships
between eyes and neck and

814
01:04:01,180 --> 01:04:07,060
shoulder are continually
changing during our lives.

815
01:04:07,980 --> 01:04:11,180
Steven, with regards to that,
sorry to interrupt.

816
01:04:11,420 --> 01:04:14,980
Have you guys thought about how
unexpected visual cues might

817
01:04:14,980 --> 01:04:17,700
interact or disrupt this
process?

818
01:04:18,380 --> 01:04:19,980
Well, well, yeah.
I didn't go.

819
01:04:20,340 --> 01:04:24,980
I focused so much on the art
attentional system for learning,

820
01:04:25,540 --> 01:04:31,500
but an unexpected cue will
mismatch ongoing processing.

821
01:04:32,000 --> 01:04:34,800
If there's a biggest enough
mismatch, it'll activate the

822
01:04:34,840 --> 01:04:39,920
orienting system, which is
either depending on the part of

823
01:04:39,920 --> 01:04:46,040
the brain and or hippocampal
region or nonspecific calamic

824
01:04:46,040 --> 01:04:49,560
region.
And then you'll get a burst of

825
01:04:49,560 --> 01:04:55,360
nonspecific arousal and novelty
reaction, you know, and that

826
01:04:55,360 --> 01:04:59,280
will drive a search for a better
matching category.

827
01:04:59,280 --> 01:05:03,510
So if, for example, an
unexpected but familiar category

828
01:05:03,510 --> 01:05:07,390
comes in, there would be a quick
reset and a shift to the

829
01:05:07,390 --> 01:05:10,910
attention would occur to the
right category in one step.

830
01:05:10,910 --> 01:05:19,310
There's no generalized search.
But if there were, no if there

831
01:05:19,310 --> 01:05:23,710
were no familiar category.
If it was really novel, there'd

832
01:05:23,710 --> 01:05:28,390
be a brief search.
But when there was nothing

833
01:05:28,390 --> 01:05:33,400
nearby, the system is designed
to choose an uncommitted

834
01:05:33,480 --> 01:05:38,400
population and begin learning a
new category.

835
01:05:39,400 --> 01:05:47,720
And so, yes, there's the
processes of attention, the

836
01:05:47,720 --> 01:05:50,800
attentional and orienting
systems in art or

837
01:05:52,160 --> 01:05:59,930
computationally complementary.
And let me just leave it at

838
01:05:59,930 --> 01:06:03,530
that.
And you can, it's not such a

839
01:06:03,530 --> 01:06:09,410
deep thing, but you can go
through a few properties of each

840
01:06:09,410 --> 01:06:11,810
and you could see they're
manifestly complementary.

841
01:06:11,810 --> 01:06:17,050
They fit together like yin fits
together with gang or puzzle

842
01:06:17,050 --> 01:06:19,850
pieces fit together.
And that's one of the beautiful

843
01:06:19,850 --> 01:06:23,570
things.
So many of the processes I've

844
01:06:24,450 --> 01:06:30,090
understood computationally
complementary, like yin, Yang,

845
01:06:30,690 --> 01:06:38,010
and I think of them as emerging
in early development as a kind

846
01:06:38,010 --> 01:06:43,850
of global symmetry breaking in
morphogenesis, which I don't

847
01:06:44,690 --> 01:06:47,730
fully understand.
I've written down some of the

848
01:06:47,730 --> 01:06:52,730
targets of morphogenesis.
I've also written some, you

849
01:06:52,730 --> 01:06:59,980
know, basic but and complete
papers about early development

850
01:06:59,980 --> 01:07:05,140
with various students.
But to fully explain the

851
01:07:05,180 --> 01:07:11,620
unfolding of these complementary
systems will need a lot of very

852
01:07:11,620 --> 01:07:14,620
productive work.
So it's another big unsolved

853
01:07:14,620 --> 01:07:18,940
problem for the new generation.
Steve, tell me how do these

854
01:07:18,940 --> 01:07:22,620
neural models combine?
To explain how we could design

855
01:07:22,620 --> 01:07:26,300
autonomous adaptive intelligence
intelligence into algorithms and

856
01:07:26,300 --> 01:07:29,620
mobile robots for engineering
technology and just artificial

857
01:07:29,620 --> 01:07:35,540
intelligence in general, okay so
and this will I think give you a

858
01:07:35,740 --> 01:07:38,140
little of the greater detail you
wanted.

859
01:07:39,420 --> 01:07:45,980
So all the neural models that
I've discovered and developed

860
01:07:46,540 --> 01:07:53,310
and illustrated operate
autonomously, and the ones that

861
01:07:53,310 --> 01:07:58,710
learn can learn unsupervised,
just in response to inputs from

862
01:07:58,710 --> 01:08:03,630
the world, or supervised where
you can.

863
01:08:04,910 --> 01:08:09,110
The supervision is provided by
feedback about predictive

864
01:08:09,110 --> 01:08:13,830
success in the world.
Like, you know, I look at a

865
01:08:13,830 --> 01:08:21,970
letter that I think is a letter
me because it has, you know, 2

866
01:08:22,410 --> 01:08:24,850
two things and a little thing at
the bottom.

867
01:08:25,490 --> 01:08:27,290
And my teacher says no, it's an
app.

868
01:08:28,210 --> 01:08:34,609
And so you know that she doesn't
have to go further.

869
01:08:34,649 --> 01:08:38,890
Just through examples like that
I can realize that, you know,

870
01:08:38,890 --> 01:08:40,850
ignore that little thing on the
bottom.

871
01:08:43,330 --> 01:08:46,170
So all the perceptual and
cognitive processes.

872
01:08:46,170 --> 01:08:49,720
And this is the point of
scientific general purpose,

873
01:08:50,760 --> 01:08:55,200
meaning they can respond to any
input pattern in their

874
01:08:55,520 --> 01:09:00,160
particular domain and respond
adaptively to these general

875
01:09:00,160 --> 01:09:03,560
purpose inputs and changing
environments.

876
01:09:05,319 --> 01:09:08,600
And that is, you know, I want to
emphasize a work in progress, of

877
01:09:08,640 --> 01:09:14,080
course.
But let me illustrate the power

878
01:09:14,080 --> 01:09:21,200
of what we do know by noting
that what I call the paradigm of

879
01:09:21,200 --> 01:09:26,160
laminar computing, which
clarifies how multiple parts of

880
01:09:26,160 --> 01:09:31,479
the brain can interact
seamlessly together because

881
01:09:31,479 --> 01:09:36,640
they're designed as variations
of this single canonical

882
01:09:36,640 --> 01:09:41,960
cortical circuit, even though
they may be doing vision

883
01:09:42,560 --> 01:09:49,380
recognition, speech language
just be by, you know, fine

884
01:09:49,380 --> 01:10:01,380
tuning A canonical circuit.
And so I'll jump ahead and say

885
01:10:01,820 --> 01:10:06,380
think about what that means for
the future of technology.

886
01:10:07,460 --> 01:10:12,540
Of course you could have one
canonical VLSI chip, and if you

887
01:10:12,540 --> 01:10:17,230
do, the variations of that chip,
detective, vision, speech,

888
01:10:17,230 --> 01:10:26,990
language, whatsoever, cognition.
Because it's the same chip, you

889
01:10:26,990 --> 01:10:31,390
can connect it all in a self
consistent, increasingly

890
01:10:31,390 --> 01:10:34,190
autonomous system.
It all fits together.

891
01:10:34,190 --> 01:10:38,390
It's the same architecture
repeated over and over.

892
01:10:39,110 --> 01:10:44,510
And to to put in perspective,
the cerebral cortex.

893
01:10:44,510 --> 01:10:49,070
You know, the little Gray cells
of Ichio Poirot is the seed of

894
01:10:49,070 --> 01:10:53,390
an all higher intelligence in
all modalities.

895
01:10:54,310 --> 01:10:58,630
And remarkably, let me
emphasize, variations of a

896
01:10:58,630 --> 01:11:03,870
single canonical cortical
circuit gives rise to all higher

897
01:11:03,870 --> 01:11:08,590
autobiological intelligence as
emergent or interactive

898
01:11:08,590 --> 01:11:12,680
properties of how several
cortical areas interact

899
01:11:12,680 --> 01:11:18,440
together.
So the cell dense cerebral

900
01:11:18,440 --> 01:11:23,040
cortex organized in layers
usually in perceptual of

901
01:11:23,040 --> 01:11:28,480
cognitive cortex 6 main layers
with characteristics of lamina

902
01:11:29,800 --> 01:11:36,360
and the laminate computing model
explains how this laminate

903
01:11:36,360 --> 01:11:41,320
design unifies.
And here's where the it's very

904
01:11:41,320 --> 01:11:45,800
important for the future of
natural intelligence computing.

905
01:11:46,200 --> 01:11:49,520
Unifies the best properties of
feed forward and feedback

906
01:11:49,520 --> 01:11:53,440
processing, digital and analog
processing.

907
01:11:54,200 --> 01:11:59,760
Bottom up preattentive
data-driven processing like our

908
01:11:59,760 --> 01:12:04,800
adaptive filters and top down
attentive hypothesis driven

909
01:12:05,560 --> 01:12:11,360
processing like I learned
expectations and if it could be

910
01:12:11,360 --> 01:12:18,040
embodied in a the LSI ship
suitably specialized to embody

911
01:12:18,040 --> 01:12:25,640
these different intelligent
capabilities, we can move toward

912
01:12:26,080 --> 01:12:31,560
general purpose autonomous
adapted intelligence.

913
01:12:32,040 --> 01:12:34,760
And because they're chips, they
can go in everything from

914
01:12:34,760 --> 01:12:38,220
algorithms and machines.
The mobile robots cause their

915
01:12:38,220 --> 01:12:41,740
light.
It'll take a big, big effort,

916
01:12:41,740 --> 01:12:46,260
and Google seems to be
fascinated by deep learning.

917
01:12:46,260 --> 01:12:50,460
But if Google wants to make
trillions of dollars, if they do

918
01:12:50,460 --> 01:12:53,380
it right, this is where the
revolution is.

919
01:12:53,380 --> 01:12:59,860
The revolution is in autonomy,
not in semi classical steepest

920
01:12:59,860 --> 01:13:03,060
descent.
That goes back to Friedrich

921
01:13:03,100 --> 01:13:07,760
Gaus.
So as I said, there are 6 main

922
01:13:07,760 --> 01:13:12,800
layers in all cortical areas
that support vision, object

923
01:13:12,800 --> 01:13:16,280
recognition, speech, language
and cognition.

924
01:13:17,480 --> 01:13:23,280
The layers are a little
different in action areas and

925
01:13:23,640 --> 01:13:27,360
they interact.
The layers interact in a single

926
01:13:27,360 --> 01:13:31,160
design for bottom up,
horizontal, and top down

927
01:13:31,160 --> 01:13:33,590
pathways.
I've been talking mostly about

928
01:13:33,590 --> 01:13:36,630
bottom up and top down, but we
need horizontal too.

929
01:13:37,710 --> 01:13:41,070
So as our earliest stage, the
bottom up pathways are adaptive

930
01:13:41,070 --> 01:13:45,270
filters that carry out the
learning of recognition

931
01:13:45,270 --> 01:13:50,470
categories and the top down are
expectations that going to focus

932
01:13:50,470 --> 01:13:56,630
attention on the critical
features, the ones that you're

933
01:13:56,630 --> 01:13:59,510
paying attention to that control
predictive success.

934
01:14:00,250 --> 01:14:05,330
But the horizontal interactions
between multiple cortical

935
01:14:05,530 --> 01:14:11,290
columns link cells into
groupings that carry out

936
01:14:11,290 --> 01:14:15,370
different functions across the
brain and remember like mention

937
01:14:15,370 --> 01:14:22,450
boundaries and surfaces earlier.
Well, such horizontal

938
01:14:22,450 --> 01:14:28,740
interactions are what enable us
to form receptual boundaries

939
01:14:28,740 --> 01:14:33,500
around objects with a kind of
cell that I introduced called a

940
01:14:34,020 --> 01:14:41,420
bipole cell.
And anyway, so how the

941
01:14:41,420 --> 01:14:48,580
neocortical cells interact
within and across layers is what

942
01:14:48,580 --> 01:14:53,420
the autonomous adaptive
intelligence problem reduces to

943
01:14:54,060 --> 01:14:56,780
in this canonical cortical
circuit.

944
01:14:57,340 --> 01:15:02,020
So it's a clearly defined
problem which is immense amount

945
01:15:02,020 --> 01:15:06,820
of compatible data.
In particular I love to say that

946
01:15:06,900 --> 01:15:12,060
a pre attentive grouping like a
boundary which can form

947
01:15:12,060 --> 01:15:14,540
automatically even before you
pay attention to it.

948
01:15:14,900 --> 01:15:17,740
It's the difference between
perception and recognition.

949
01:15:18,580 --> 01:15:24,300
A pre attentive grouping is its
own attentional prime due to the

950
01:15:24,300 --> 01:15:31,010
way in which horizontal and
feedback interact within the

951
01:15:31,010 --> 01:15:39,010
laminar circuits.
Anyway, I think that's all I

952
01:15:39,010 --> 01:15:43,650
want to say about that except
chapter 10 of my magnum opus

953
01:15:43,650 --> 01:15:47,170
which I haven't mentioned.
I should mention it because it

954
01:15:47,810 --> 01:15:52,410
it's called Conscious Mind
Resonant Brain, How each brain

955
01:15:52,410 --> 01:15:56,190
makes a mind.
It was published in 2021 by

956
01:15:56,190 --> 01:16:01,550
Oxford University Press and it
tries to give a self-contained

957
01:16:01,550 --> 01:16:07,870
and nontechnical overview and
synthesis of not only my work

958
01:16:07,870 --> 01:16:11,950
and hundreds of collaborators
over the past 50 some odd years,

959
01:16:11,950 --> 01:16:17,270
but also explains and unifies
the understanding of the work of

960
01:16:17,270 --> 01:16:22,760
hundreds of other scientists.
It's not just about me.

961
01:16:22,800 --> 01:16:28,320
And I'm happy to say that the
book won the 2022 Pros Book

962
01:16:28,320 --> 01:16:32,160
Award in Neuroscience of the
Association of American

963
01:16:32,160 --> 01:16:35,280
Publishers.
And if you look it up on Amazon,

964
01:16:35,280 --> 01:16:39,560
you'll see it's dirt cheap.
You can even it's almost 800

965
01:16:39,560 --> 01:16:43,080
pages with over 600 color
figures.

966
01:16:43,080 --> 01:16:49,240
That kindle is just 17 bucks and
the hard copy is just 33 bucks

967
01:16:49,950 --> 01:16:54,390
because I spent thousands of
dollars my own money to make it

968
01:16:54,390 --> 01:16:58,150
affordable, especially the young
people like students.

969
01:16:59,230 --> 01:17:04,350
And as to being self-contained
and nontechnical, it's not an

970
01:17:04,430 --> 01:17:09,230
easy read because this is the
most complex system we can

971
01:17:09,230 --> 01:17:12,950
study.
But I have friends who are a

972
01:17:12,950 --> 01:17:21,210
rabbi, a pastor, a visual
artist, the gallery owner, a

973
01:17:21,210 --> 01:17:26,410
social worker, a lawyer who know
no science, who've enjoyed

974
01:17:26,410 --> 01:17:29,890
reading parts of it.
And I say parts of it because I

975
01:17:29,890 --> 01:17:32,170
know it's a long book and
everyone's busy.

976
01:17:32,170 --> 01:17:40,170
And if you read the introduction
and preface you can then jump to

977
01:17:40,170 --> 01:17:44,410
any chapter you like.
Cause I wrote the chapters to be

978
01:17:45,110 --> 01:17:47,430
readable independently of each
other.

979
01:17:47,430 --> 01:17:51,710
And if you want to read more
about or you'd read chapter 5.

980
01:17:52,270 --> 01:17:57,790
If you want to read more about
how we consciously see, you'd

981
01:17:57,790 --> 01:18:03,830
read chapters one through 3.
I think that was a very good

982
01:18:04,270 --> 01:18:06,350
approach to the book as well,
Steve, because I remember when I

983
01:18:06,350 --> 01:18:09,350
was doing my my dissertation
whenever I.

984
01:18:09,910 --> 01:18:12,510
Cuz I read your book once before
and then I remember while

985
01:18:12,510 --> 01:18:15,390
working later on, any time I
wanted to figure out something

986
01:18:15,390 --> 01:18:17,590
about a specific topic, I was
able to jump to that specific

987
01:18:17,590 --> 01:18:20,190
chapter.
So that approach is not just me.

988
01:18:20,190 --> 01:18:22,790
Yeah, it's very effective.
Almost like having a little

989
01:18:23,070 --> 01:18:29,670
dictionary for the mind.
Well, you know, in the hype it's

990
01:18:29,670 --> 01:18:34,390
called Principia of mind.
Just like Newton had his own

991
01:18:34,430 --> 01:18:40,540
Principia.
But, and it's a anyway, I worked

992
01:18:40,540 --> 01:18:47,300
very hard on it and I think it
came out pretty well given the

993
01:18:47,300 --> 01:18:54,020
challenge of writing it.
So you want me to go through the

994
01:18:54,020 --> 01:18:56,620
thought experiment you want?
Yeah, I think let's do that.

995
01:18:56,620 --> 01:18:59,340
Let's go through that.
I mean, this thought experiment

996
01:18:59,420 --> 01:19:02,020
is, it's amazing.
So run me through the thought

997
01:19:02,020 --> 01:19:04,940
experiment.
How any system can autonomously

998
01:19:04,940 --> 01:19:07,780
learn to correct predictive
errors in a changing world that

999
01:19:07,780 --> 01:19:09,420
is filled with unexpected
events?

1000
01:19:11,100 --> 01:19:13,740
And let me thank you for reading
my work so carefully.

1001
01:19:13,740 --> 01:19:17,500
You use my very words.
Well, this is from our e-mail,

1002
01:19:17,500 --> 01:19:24,140
so I just found it a lot easier
to do there Is there, there is

1003
01:19:25,500 --> 01:19:28,020
there.
So there are multiple steps in

1004
01:19:28,020 --> 01:19:30,500
the thought experiment, and I
can't do them all here.

1005
01:19:32,240 --> 01:19:38,080
And I can now ever mention that
the main ideas are very, very

1006
01:19:38,080 --> 01:19:43,920
simple and there are two
questions that drive the thought

1007
01:19:43,920 --> 01:19:48,960
experiment, and they all have to
do with Principle of Sufficient

1008
01:19:48,960 --> 01:19:51,640
reason.
It's all an exercise in pure

1009
01:19:51,640 --> 01:19:58,880
logic, and it's a celebration of
role of reason in scientific

1010
01:19:58,880 --> 01:20:01,670
theorizing.
The first question is how can

1011
01:20:01,670 --> 01:20:08,070
the coding error be corrected if
no individual cell knows that

1012
01:20:08,070 --> 01:20:14,470
one has occurred?
You know cells are idiots, and

1013
01:20:14,470 --> 01:20:18,350
von Neumann realized that.
Von Neumann commented on the

1014
01:20:18,350 --> 01:20:24,590
fact that brain cells
individually are pretty stupid.

1015
01:20:25,470 --> 01:20:27,790
How do you get this collective
intelligence?

1016
01:20:27,870 --> 01:20:32,280
And the second point, the
symbols will compress

1017
01:20:32,280 --> 01:20:36,560
categories, you know that
response selectively to stuff by

1018
01:20:36,560 --> 01:20:40,520
themselves they can't detect
errors, they don't know what

1019
01:20:40,520 --> 01:20:42,440
activated them was right or
wrong.

1020
01:20:43,840 --> 01:20:48,440
So then how do you correct
errors and that drives you to

1021
01:20:48,440 --> 01:20:53,560
top down expectations because
you have to see is what's going

1022
01:20:53,560 --> 01:20:56,400
on down there, what was going on
down there when you learn the

1023
01:20:56,400 --> 01:20:59,040
category.
So it's very simple minded.

1024
01:21:00,220 --> 01:21:04,220
And so these trivial statements
gain their power from the

1025
01:21:04,220 --> 01:21:06,820
logical story that is a thought
experiment.

1026
01:21:06,940 --> 01:21:10,420
It's a celebration of logic and
science.

1027
01:21:11,940 --> 01:21:14,100
That's all I really want to say
about that.

1028
01:21:14,820 --> 01:21:19,420
I think people, yeah, if people
don't want to buy the book.

1029
01:21:19,420 --> 01:21:26,660
If you look at my 1980 paper in
Psychological Review, I have

1030
01:21:26,660 --> 01:21:34,240
that and over 500 papers on my
web page, which is sites.

1031
01:21:34,240 --> 01:21:45,960
That's SITES dot BU for Boston,
university dot Edu for education

1032
01:21:46,200 --> 01:21:53,640
slash Steve GSTED e.g.
If you go to sites dot EU dot

1033
01:21:53,640 --> 01:21:58,660
EDU/CG, that's my web page.
There are hundreds of papers,

1034
01:21:58,980 --> 01:22:03,220
1980 You can download my
Psychological Review paper for

1035
01:22:03,220 --> 01:22:06,220
free.
You don't have to pay 17 bucks

1036
01:22:06,220 --> 01:22:10,020
for my book.
And there are.

1037
01:22:11,300 --> 01:22:15,500
There are all all of those
papers there, as well as many

1038
01:22:17,340 --> 01:22:20,060
videos of keynote lectures and
interviews.

1039
01:22:20,060 --> 01:22:23,780
And I hope I can put this on my
web page eventually.

1040
01:22:24,210 --> 01:22:25,850
I'd appreciate that.
And I'll definitely put a link

1041
01:22:25,850 --> 01:22:28,650
to those to that, to that
website in the description as

1042
01:22:28,650 --> 01:22:30,210
well.
Steve, Steve, I've gone thank

1043
01:22:30,210 --> 01:22:31,370
you.
Something I've been thinking

1044
01:22:31,370 --> 01:22:35,290
about when as a kid, when you
think about people like Bayes

1045
01:22:35,290 --> 01:22:38,250
and Bayesian predictive brains,
this is not in the schedule, but

1046
01:22:38,250 --> 01:22:39,650
it's something I'm curious
about.

1047
01:22:39,810 --> 01:22:41,970
At what point were you thinking
like this?

1048
01:22:41,970 --> 01:22:44,450
Doesn't this predictive
approach, this hierarchical

1049
01:22:44,450 --> 01:22:47,610
Bayesian brain approach just
does not suffice for me?

1050
01:22:49,730 --> 01:22:54,370
Well remember I did Art which is
a predictive theory starting in

1051
01:22:54,370 --> 01:23:02,730
76 and all of the Fed like
interest in the Bayes rule was

1052
01:23:02,730 --> 01:23:09,690
much later.
And you know the Bayes rule as

1053
01:23:09,690 --> 01:23:17,370
used in a statistical estimation
is a valid approach.

1054
01:23:17,370 --> 01:23:21,890
But what is the Bayes rule?
It's the probability of two

1055
01:23:21,890 --> 01:23:25,860
events A and B written in two
ways.

1056
01:23:26,500 --> 01:23:32,500
The probability of B given A
times the probability of A, or

1057
01:23:32,500 --> 01:23:36,900
the probability of A given B
times the probability of B.

1058
01:23:37,180 --> 01:23:42,540
It's just a tautology.
Take that equality between those

1059
01:23:42,540 --> 01:23:46,860
two ways of probability A given
B times probability equal

1060
01:23:46,860 --> 01:23:51,110
probability of B given A.
It's probably day, and divide,

1061
01:23:51,110 --> 01:23:54,230
let's say both sides by
probability a day and maximize.

1062
01:23:54,630 --> 01:23:59,070
That's the Bayes rule.
It's a tautology, but it's used

1063
01:23:59,070 --> 01:24:02,470
in statistical estimation.
Some of my good friends are good

1064
01:24:02,470 --> 01:24:06,590
statisticians who've used it,
but it says nothing about the

1065
01:24:06,590 --> 01:24:11,750
world.
It's just the tautology.

1066
01:24:12,550 --> 01:24:17,530
Anything that you want to say
about in physics or chemistry or

1067
01:24:17,570 --> 01:24:21,210
biology or mind and brain,
you've got to add.

1068
01:24:22,250 --> 01:24:27,050
You'll notice I don't think it's
used in physics or chemistry or

1069
01:24:27,050 --> 01:24:33,810
biology, and there's nothing in
the heuristics of the Bayes rule

1070
01:24:34,690 --> 01:24:38,330
that suggests what to add.
And that's when people just

1071
01:24:38,330 --> 01:24:43,190
start throwing in the kitchen
and sink and trying to fit it to

1072
01:24:43,190 --> 01:24:45,630
what is basically just the
topology.

1073
01:24:46,190 --> 01:24:50,590
I just think that's sad, given
that people like me and hundreds

1074
01:24:50,590 --> 01:24:54,950
of other people emulating,
developing, varying, improving,

1075
01:24:55,510 --> 01:24:58,390
it's all over the literature.
Why are you doing this?

1076
01:24:59,110 --> 01:25:02,350
You know it's trying to get
something for nothing.

1077
01:25:02,350 --> 01:25:06,270
You want to turn the magical
level, and you'll get out more

1078
01:25:06,270 --> 01:25:08,980
than you put in.
There's a difference between the

1079
01:25:08,980 --> 01:25:11,860
gift that keeps on giving when
you have foundations that are

1080
01:25:11,860 --> 01:25:15,540
already explanatory and
predictive, and getting

1081
01:25:15,540 --> 01:25:17,700
something for nothing when you
don't know what the hell you're

1082
01:25:17,700 --> 01:25:22,580
talking about.
In fact, I'll give you an old

1083
01:25:22,580 --> 01:25:25,700
example of that.
I do.

1084
01:25:25,700 --> 01:25:29,660
You know who Renee Tom was.
Catastrophe Theory.

1085
01:25:30,340 --> 01:25:37,170
Well, I was invited to a big
conference about catastrophe

1086
01:25:37,210 --> 01:25:40,290
theory.
And they invited me because, you

1087
01:25:40,290 --> 01:25:44,530
know, I was already the
preeminent model of mind and

1088
01:25:44,530 --> 01:25:51,130
brain.
And I I was sharing an office

1089
01:25:51,130 --> 01:25:58,090
with a young mathematician and
he literally said to me, isn't

1090
01:25:58,090 --> 01:26:03,460
it wonderful that by knowing the
elementary catastrophes I can

1091
01:26:03,460 --> 01:26:06,900
write about biology, but I don't
have to learn any biology.

1092
01:26:06,900 --> 01:26:11,260
It was really a case.
He said it he's a very

1093
01:26:11,260 --> 01:26:14,260
intelligent young man, case of
trying to get something for

1094
01:26:14,260 --> 01:26:17,700
nothing.
And then you know, Tom would

1095
01:26:18,340 --> 01:26:21,780
make screeping statements about
applying.

1096
01:26:22,500 --> 01:26:27,180
You know, we took this one out
of the other thing and and I had

1097
01:26:27,180 --> 01:26:31,930
a chat with him.
So I said, you know, he's

1098
01:26:31,930 --> 01:26:34,610
interested.
I think maybe with language was

1099
01:26:34,930 --> 01:26:38,090
something.
So I, you know, one of my early

1100
01:26:38,090 --> 01:26:41,690
loves was understanding verbal
learning.

1101
01:26:43,730 --> 01:26:49,530
And so I I asked him, well, how
would you, for example, explain

1102
01:26:49,530 --> 01:26:53,130
the boat serial position curve
in serial verbal learning?

1103
01:26:53,770 --> 01:26:57,170
First he never heard of it, but
he said I don't care about data.

1104
01:26:57,850 --> 01:27:00,920
He said that to me.
I don't care about data.

1105
01:27:01,600 --> 01:27:06,880
You'd never know it from hearing
him talk anyway.

1106
01:27:06,880 --> 01:27:09,560
So that was very, very sad, I
thought.

1107
01:27:12,120 --> 01:27:16,040
But he was a great differential
geometer which just he reached a

1108
01:27:16,040 --> 01:27:19,880
certain age and he wanted to
become a philosophy king.

1109
01:27:20,680 --> 01:27:26,400
He I think he would have would
have been better if he tempered

1110
01:27:26,960 --> 01:27:32,510
his work to he had postulates
that were really good for things

1111
01:27:32,510 --> 01:27:36,150
like turbulence.
He should have just left well

1112
01:27:36,150 --> 01:27:39,830
enough alone, because mind brain
has nothing to do with his

1113
01:27:39,830 --> 01:27:43,110
postulates, the elementary
catastrophes.

1114
01:27:44,550 --> 01:27:46,910
I think I mean you you clearly
are different in that way.

1115
01:27:46,910 --> 01:27:49,350
I mean you're you're actually
very much a polymath in that

1116
01:27:49,350 --> 01:27:51,350
regard.
When you think about you being

1117
01:27:51,350 --> 01:27:53,750
an emiratist professor, it's in
so many different fields.

1118
01:27:53,750 --> 01:27:56,470
When you see your, your buy is
going to be quite when.

1119
01:27:56,630 --> 01:28:00,860
When I put it down, it's because
of the universality of the

1120
01:28:00,860 --> 01:28:03,220
subject matter that I've
illustrated.

1121
01:28:03,980 --> 01:28:09,620
I've been led sometimes kicking
and screaming into areas I never

1122
01:28:09,660 --> 01:28:12,980
thought I'd know anything about,
including consciousness.

1123
01:28:13,980 --> 01:28:17,500
In a way.
I've been forced into these

1124
01:28:17,540 --> 01:28:23,580
fields by the power of the
theory and and I I feel very

1125
01:28:23,580 --> 01:28:27,140
lucky to have been put in that
position.

1126
01:28:27,140 --> 01:28:33,020
But, you know, I do think that
I've been true to the data.

1127
01:28:33,980 --> 01:28:41,180
I I'm not a slavish reductionist
by any means, but I believe that

1128
01:28:41,260 --> 01:28:49,460
the majesty of evolution,
including of our minds, provides

1129
01:28:49,460 --> 01:28:54,580
us with a richness that we just
left to do the best we can.

1130
01:28:54,580 --> 01:28:57,860
You know it.
Is it is it rewarding to know?

1131
01:28:57,860 --> 01:29:00,660
I mean you've you've had it out
for such a long time and you can

1132
01:29:00,660 --> 01:29:03,620
see how it's been applied and
and practically used at this

1133
01:29:03,620 --> 01:29:05,540
point.
How does it feel to see it

1134
01:29:05,540 --> 01:29:11,900
continuously grow and and and
you know, I I love it when you

1135
01:29:11,900 --> 01:29:15,540
tell me things like that and I
occasionally notice things like

1136
01:29:15,540 --> 01:29:18,780
that.
But I noticed just as often

1137
01:29:19,740 --> 01:29:24,460
young investigators who know
nothing about the work working

1138
01:29:24,460 --> 01:29:28,140
on similar problems.
And I think the problem is that

1139
01:29:28,940 --> 01:29:35,100
all too often today
investigators might think they

1140
01:29:35,100 --> 01:29:38,420
only need to read the last five
years of research.

1141
01:29:38,420 --> 01:29:41,660
Or maybe maybe if they real
scholars 10 years.

1142
01:29:42,540 --> 01:29:45,540
But I made some of these
discoveries 3040, fifty years

1143
01:29:45,540 --> 01:29:52,400
ago and they're still holding.
And then I think that's okay, No

1144
01:29:52,400 --> 01:29:54,600
one can know everything.
But you know, if I see someone's

1145
01:29:54,600 --> 01:30:00,200
making a claim that is
particularly irksome because I

1146
01:30:00,200 --> 01:30:04,000
worked so hard to explain it
much better than they were even

1147
01:30:04,000 --> 01:30:07,800
doing, now I'll write them a
plight and friendly letter

1148
01:30:07,800 --> 01:30:11,000
saying, hey, you might be
interested in knowing this

1149
01:30:11,000 --> 01:30:15,920
result.
And then that's where you can

1150
01:30:15,920 --> 01:30:20,940
tell a person's medal.
You know, if they write back and

1151
01:30:20,980 --> 01:30:24,380
even say, well, thank you very
much, I'll read it.

1152
01:30:24,820 --> 01:30:26,260
Thanks for calling my attention
to it.

1153
01:30:26,660 --> 01:30:28,100
That's the most you can hope
for.

1154
01:30:29,060 --> 01:30:34,820
But if they just choose to
ignore it and you know, you're

1155
01:30:34,820 --> 01:30:40,060
dealing with a plagiarist.
So a lot of doing science is

1156
01:30:41,020 --> 01:30:47,940
having good values.
I I felt that I have to keep my

1157
01:30:48,840 --> 01:30:55,120
emotions pure so I don't do
things that are not

1158
01:30:56,200 --> 01:31:02,840
intellectually valid because of,
you know, petty emotions.

1159
01:31:03,360 --> 01:31:05,840
And I think you're in a
difficult position because when

1160
01:31:05,840 --> 01:31:08,200
you're so focused on the work, I
mean you've said this numerous

1161
01:31:08,200 --> 01:31:09,880
times as well.
But when you're so focused on

1162
01:31:09,880 --> 01:31:12,320
doing the work moving on to the
next step and trying to get on

1163
01:31:12,320 --> 01:31:15,920
to the next field, you're not
really self promoting as much as

1164
01:31:16,000 --> 01:31:18,160
other scientists would.
Well that's right.

1165
01:31:18,160 --> 01:31:24,680
Like like Tayevo Cahoonen, who
you know people credit with self

1166
01:31:24,680 --> 01:31:28,800
organizing map will know.
I did it before him and so did

1167
01:31:29,080 --> 01:31:33,240
Christopher on the Mossberg and
I worked on in the mid 70s and

1168
01:31:33,240 --> 01:31:38,400
Tayevo started writing about an
80 to an 84 and I was.

1169
01:31:39,400 --> 01:31:45,470
I was the session chair where he
presented his work and in the

1170
01:31:45,470 --> 01:31:50,030
discussion period I wrote down
his claims against things I've

1171
01:31:50,030 --> 01:31:57,110
already published point by point
and he ignored it.

1172
01:32:01,870 --> 01:32:05,950
I even helped him.
I helped him to get up a big

1173
01:32:07,990 --> 01:32:12,180
promotion in by the government
of Finland even after he did

1174
01:32:12,180 --> 01:32:16,780
that, and I think what an
asshole I am, you know he wasn't

1175
01:32:16,780 --> 01:32:20,220
worthy of it because of his
ethics.

1176
01:32:21,140 --> 01:32:23,740
Are there any?
It was all it was, all he ever

1177
01:32:23,740 --> 01:32:25,900
did.
And I had moved on.

1178
01:32:28,220 --> 01:32:31,260
I can't.
I can't defend all my children.

1179
01:32:31,740 --> 01:32:34,540
They're out in the world.
While you're here, are there any

1180
01:32:34,580 --> 01:32:37,210
other?
Pieces of work or bodies of work

1181
01:32:37,210 --> 01:32:40,490
that you have, that work that
you feel should be brought up,

1182
01:32:40,490 --> 01:32:43,770
that you have taught the world
before someone else.

1183
01:32:43,770 --> 01:32:45,850
Or do you want to just go on to
the next question?

1184
01:32:45,850 --> 01:32:47,050
I know we're going off track a
little.

1185
01:32:47,490 --> 01:32:52,650
Well I can.
I can cuz I'm so why don't we?

1186
01:32:52,970 --> 01:32:56,690
For someone who's why don't we
quickly let's try to speed up.

1187
01:32:56,930 --> 01:32:58,730
Okay, cool.
Let's go to the next question.

1188
01:32:58,730 --> 01:33:01,770
Tell me about, I don't know
brands.

1189
01:33:02,920 --> 01:33:04,400
Yes.
How do our brains support such

1190
01:33:04,400 --> 01:33:08,120
vital human qualities such as
creativity, morality and

1191
01:33:08,120 --> 01:33:10,160
religion?
And why do so many people

1192
01:33:10,440 --> 01:33:13,320
persist in superstitious,
irrational, and self defeating

1193
01:33:13,320 --> 01:33:15,840
behaviors in certain social
environments?

1194
01:33:17,360 --> 01:33:24,080
OK, so here I'm going to quote
my magnum opus page 14, so I

1195
01:33:24,080 --> 01:33:26,920
have to read it.
Examples of cognitive emotional

1196
01:33:26,920 --> 01:33:30,710
interactions ubiquitous.
For example, few of us would say

1197
01:33:30,710 --> 01:33:35,630
that the wallpaper on a dining
room wall causes food to appear

1198
01:33:36,070 --> 01:33:39,950
on the dining room table, even
though it's present whenever

1199
01:33:39,950 --> 01:33:43,750
food is served.
On the other hand, if you went

1200
01:33:43,750 --> 01:33:47,670
into the dining room only when
food was served, you might very

1201
01:33:47,670 --> 01:33:51,150
well associate the distinctive
wallpaper in the dining room

1202
01:33:51,150 --> 01:33:53,550
with eating.
You know, if you keep if you

1203
01:33:53,550 --> 01:33:56,110
walk through the room.
But between meals, you won't do

1204
01:33:56,110 --> 01:33:59,360
that.
In fact, many superstitious

1205
01:33:59,360 --> 01:34:03,880
societal rituals, such as rain
dances or primitive medical

1206
01:34:03,880 --> 01:34:07,840
practices that may do more harm
than good, may have arisen

1207
01:34:07,840 --> 01:34:12,080
throughout history due to
accidental correlations between

1208
01:34:12,080 --> 01:34:15,400
events.
Even a low probability of reward

1209
01:34:15,400 --> 01:34:18,920
can, under the proper
circumstance that leads some

1210
01:34:18,920 --> 01:34:22,400
people to indulge in
superstitious behaviors,

1211
01:34:22,400 --> 01:34:26,040
maintain unhealthy
relationships, become chronic

1212
01:34:26,040 --> 01:34:29,640
gamblers, or even pursue
fetishistic and stuff to behave

1213
01:34:29,640 --> 01:34:33,920
it.
And my book says a bit, wow, but

1214
01:34:33,920 --> 01:34:38,520
so the main issue is how do we
learn which events are causal

1215
01:34:38,520 --> 01:34:43,000
and which are accidental?
And like the cognitive,

1216
01:34:43,000 --> 01:34:47,440
emotional work on reinforcement,
learning helps us to focus

1217
01:34:47,440 --> 01:34:52,720
attention on those combinations
of events that predict value,

1218
01:34:52,720 --> 01:34:57,120
goals, and the other events that
are outliers and are predictably

1219
01:34:57,120 --> 01:35:00,120
irrelevant.
And that's really all about how

1220
01:35:00,120 --> 01:35:04,760
you explain potential blocking,
which goes back to Ivan Pavlov.

1221
01:35:05,320 --> 01:35:09,880
So I explained how that works.
So we can go on to painters.

1222
01:35:09,880 --> 01:35:12,760
Want to go on to painters when
you want to ask me a question?

1223
01:35:13,320 --> 01:35:18,910
This question was from OGI.
Yeah, Ogi ogas.

1224
01:35:18,910 --> 01:35:20,950
Oh, so you know, okie oh.
Yeah.

1225
01:35:20,950 --> 01:35:23,350
So we spoke about when we spoke.
He's the one.

1226
01:35:23,750 --> 01:35:26,710
He's one of the people who says
I'm the smartest guy ever.

1227
01:35:26,710 --> 01:35:29,750
He says you are the greatest
scientist in history, he goes.

1228
01:35:29,750 --> 01:35:33,990
He goes on a limb of all time,
not just moving well, that well,

1229
01:35:33,990 --> 01:35:38,270
you know, such totality
statements were always risky.

1230
01:35:38,270 --> 01:35:41,350
Always, no matter what the
subject anyway.

1231
01:35:41,630 --> 01:35:44,350
But I I love it.
I appreciate it.

1232
01:35:44,710 --> 01:35:47,590
I hope it's.
I hope it's partly true.

1233
01:35:47,590 --> 01:35:50,390
OK, let's put it that way.
OK, so.

1234
01:35:50,590 --> 01:35:52,470
He said he wants today.
He says he knows you love

1235
01:35:52,470 --> 01:35:55,030
talking about art, and he knows
you like talking about art and

1236
01:35:55,030 --> 01:35:56,550
painters when you talk about the
mind.

1237
01:35:57,550 --> 01:36:00,110
Maybe you could take your
favorite examples of how your

1238
01:36:00,110 --> 01:36:02,190
research illuminates a
particular painting.

1239
01:36:04,070 --> 01:36:10,110
OK, so I'll give you very short
and I like to say that Henri

1240
01:36:10,110 --> 01:36:15,690
Matisse, whose work I love, knew
that all boundaries are

1241
01:36:15,690 --> 01:36:19,570
invisible.
Indeed, there are many percepts

1242
01:36:19,570 --> 01:36:22,850
that we recognize by completing
invisible boundaries.

1243
01:36:23,290 --> 01:36:27,170
I haven't explained why
boundaries are invisible in the

1244
01:36:27,170 --> 01:36:30,210
boundary stream.
Visibility, conscious

1245
01:36:30,210 --> 01:36:33,410
visibility, the property of
surfaces which we have discussed

1246
01:36:33,450 --> 01:36:38,130
a little and you know an example
of completion of invisible

1247
01:36:38,130 --> 01:36:40,690
boundaries in order to
understand an object.

1248
01:36:41,370 --> 01:36:46,130
You know the famous examples of
the Dalmatian and snow where all

1249
01:36:46,130 --> 01:36:48,970
you have.
You know the Dalmatian has big

1250
01:36:48,970 --> 01:36:51,330
black spots and the rest of it
coats white.

1251
01:36:51,850 --> 01:36:55,490
And if you see it in the
background of white snow, when

1252
01:36:55,490 --> 01:36:58,610
you first look at it just looks
like black blotches.

1253
01:36:59,210 --> 01:37:02,650
But after a little while,
invisible boundaries form

1254
01:37:02,650 --> 01:37:07,130
between the black blotches that
give you enough of a cursive to

1255
01:37:07,130 --> 01:37:11,320
the shape of the Dalmatian.
So you recognize the Dalmatian

1256
01:37:11,320 --> 01:37:13,880
and snow.
So that's the sort of thing.

1257
01:37:15,280 --> 01:37:19,560
And that's a serious ability in
order to deal with invisible

1258
01:37:19,560 --> 01:37:24,000
boundaries, like it helps to
escape predators.

1259
01:37:24,600 --> 01:37:30,600
If you have you know a predator
running in dapple light, dappled

1260
01:37:31,520 --> 01:37:34,240
of partially camouflaged
environments, boundary

1261
01:37:34,240 --> 01:37:41,080
completion can let you see the
recognize and see the project

1262
01:37:41,080 --> 01:37:45,680
before it's too late.
And so Matisse had many

1263
01:37:45,680 --> 01:37:50,840
paintings based on the
observer's brain completing

1264
01:37:50,840 --> 01:37:53,600
invisible boundaries to
recognize what's going on.

1265
01:37:54,160 --> 01:37:58,600
Some of my favorites he
published in 1905 when he was a

1266
01:37:58,600 --> 01:38:02,440
young man.
So that that's what I want to

1267
01:38:02,440 --> 01:38:05,800
say about that.
I could go on and on my book in

1268
01:38:05,800 --> 01:38:12,950
the in the part of my book where
I discussed visually how we see,

1269
01:38:12,950 --> 01:38:19,470
I'm really fascinated by how
artists make paintings and how

1270
01:38:19,470 --> 01:38:22,390
humans consciously see
paintings.

1271
01:38:22,430 --> 01:38:26,750
And I discussed the work of, I
think it was 14 different

1272
01:38:26,750 --> 01:38:32,230
artists, including Mattis and
Monet and Rembrandt.

1273
01:38:32,230 --> 01:38:37,700
And I I'm, I'm my mind is not
focusing.

1274
01:38:38,380 --> 01:38:44,340
And what it what the instinctive
discoveries they made about how

1275
01:38:45,100 --> 01:38:50,180
our minds consciously see that
enabled them to develop a

1276
01:38:50,180 --> 01:38:54,780
personal style that emphasize
some processes rather than

1277
01:38:54,780 --> 01:38:57,860
others.
And I like talking about style

1278
01:38:57,860 --> 01:39:01,460
from that perspective going on
into the mechanism.

1279
01:39:04,800 --> 01:39:05,960
Sorry, Steve, You're gonna
continue?

1280
01:39:07,000 --> 01:39:08,080
No, I'm done.
I'm.

1281
01:39:09,480 --> 01:39:11,640
Done.
What do you think about Ogi and

1282
01:39:11,640 --> 01:39:14,000
sighs work?
I mean, continuing your your

1283
01:39:14,000 --> 01:39:15,360
work.
I mean, they really, really.

1284
01:39:15,440 --> 01:39:17,680
Oh, I don't.
I don't do that.

1285
01:39:17,880 --> 01:39:22,440
I don't talk about other people.
OK, No, it's just it's not fair.

1286
01:39:22,440 --> 01:39:25,520
If I said something wonderful
and you read and don't like it,

1287
01:39:26,080 --> 01:39:29,360
I see you feel annoyed.
And if I say it's lousy and you

1288
01:39:29,880 --> 01:39:34,010
don't read it and you would have
liked it, it's a no win

1289
01:39:34,010 --> 01:39:37,650
situation.
So I think people who are

1290
01:39:37,650 --> 01:39:40,810
interested in the topic they're
writing on should read the book,

1291
01:39:42,050 --> 01:39:45,330
should read it.
But you know, I I I don't

1292
01:39:45,330 --> 01:39:47,570
evaluate other people's work in
public.

1293
01:39:48,250 --> 01:39:53,850
And in fact, I was an editor in
chief for many years, and my job

1294
01:39:54,330 --> 01:39:59,090
was to choose reviewers who
shouldn't be prejudiced for or

1295
01:39:59,090 --> 01:40:02,210
against work, but who should be
informed about the background.

1296
01:40:03,650 --> 01:40:08,930
And if 3 reviewers had a similar
opinion I supported, their

1297
01:40:08,930 --> 01:40:13,090
decision wasn't for me to say
So.

1298
01:40:13,810 --> 01:40:18,930
You know, if the review is of
OGI and size book like the book,

1299
01:40:19,650 --> 01:40:22,730
that's the review.
No response.

1300
01:40:22,970 --> 01:40:25,170
Sorry, Steve, I was actually
just asking in general their

1301
01:40:25,170 --> 01:40:28,170
approach of taking your work
even further and trying to.

1302
01:40:29,160 --> 01:40:31,560
I love it when people do that.
I do.

1303
01:40:31,920 --> 01:40:33,520
I wasn't asking for a review of
the book.

1304
01:40:33,520 --> 01:40:41,960
Actually, no, I'm very I'm.
It's nothing more fulfilling to

1305
01:40:42,040 --> 01:40:45,880
an old man than to have people
think your work is useful enough

1306
01:40:45,880 --> 01:40:51,200
to spend their time trying to do
more or improve, develop,

1307
01:40:51,200 --> 01:40:53,720
whatever.
Let's see.

1308
01:40:53,840 --> 01:40:56,800
So you want to ask about
challenges or criticisms?

1309
01:40:57,550 --> 01:41:01,350
So what challenges or criticisms
has the theory faced since its

1310
01:41:01,350 --> 01:41:03,150
inception, and how have you
addressed them?

1311
01:41:04,230 --> 01:41:09,870
You can go through the major
ones that come to mind, so I'll

1312
01:41:09,870 --> 01:41:20,150
give you 3 part answer.
So I've usually been far ahead

1313
01:41:20,750 --> 01:41:25,750
of my audience in anticipating
problems that I have not yet

1314
01:41:25,750 --> 01:41:30,370
solved and then solving them
before other people realized

1315
01:41:30,370 --> 01:41:34,450
they were even problems like
this stability plasticity

1316
01:41:34,450 --> 01:41:37,450
dilemma, which still has not
been solved by deep learning or

1317
01:41:37,490 --> 01:41:42,170
back propagation.
So that avoided criticism

1318
01:41:42,170 --> 01:41:47,010
because nobody even knew what
the hell I was doing, You know,

1319
01:41:47,010 --> 01:41:52,570
that was in the early years.
I got challenges from people who

1320
01:41:52,570 --> 01:41:56,180
didn't understand my work and
they wanted to push their

1321
01:41:56,180 --> 01:42:00,340
product as long as they can to
criticize it.

1322
01:42:00,340 --> 01:42:03,460
So people would read their work,
which was usually weaker.

1323
01:42:07,340 --> 01:42:13,460
Yeah, so, but you know, then
their work would fade away and

1324
01:42:13,460 --> 01:42:17,660
mine wouldn't with any.
Aspects of it that anybody who

1325
01:42:17,660 --> 01:42:20,580
really stumped you or really got
you to think about it

1326
01:42:21,100 --> 01:42:26,280
differently.
I live in the data.

1327
01:42:28,760 --> 01:42:35,120
If someone would show me first,
I often would.

1328
01:42:35,560 --> 01:42:40,760
I'm always reading data at this
stage of my life.

1329
01:42:40,760 --> 01:42:44,600
I'll often read data and I'll
say, oh, that's just this,

1330
01:42:44,720 --> 01:42:47,840
that's just that.
And I could see, you know, well

1331
01:42:47,840 --> 01:42:53,460
to to write a paper on it.
It's a lot of effort, but I know

1332
01:42:53,460 --> 01:42:57,540
what the data means.
If I see a fact which is very

1333
01:42:57,540 --> 01:43:03,420
unusual now that I don't know
what the hell is going on, I am

1334
01:43:03,420 --> 01:43:09,140
consumed.
It will own me until I can put

1335
01:43:09,140 --> 01:43:11,620
it in a context where I can
explain it.

1336
01:43:12,780 --> 01:43:14,980
So that's just the way I
operate.

1337
01:43:14,980 --> 01:43:17,740
It's not what people say about
it.

1338
01:43:18,380 --> 01:43:24,160
I know the the evidence for my
models, and I know they're

1339
01:43:24,160 --> 01:43:30,360
incomplete.
But if I see something that's

1340
01:43:30,360 --> 01:43:34,360
qualitatively outside the range,
and it's something that

1341
01:43:34,360 --> 01:43:41,720
interests me, I will.
I will passionately commit

1342
01:43:41,720 --> 01:43:45,800
myself to understanding more.
Sometimes the database is just

1343
01:43:45,800 --> 01:43:49,560
too small.
It could and it might not even

1344
01:43:49,560 --> 01:43:54,810
be a fact.
You need really to study whole

1345
01:43:54,810 --> 01:43:58,490
databases.
You need enough evidence, cause

1346
01:43:58,490 --> 01:44:04,890
anybody can try to explain one
fact in 20 ways, try to explain

1347
01:44:05,450 --> 01:44:10,450
100 facts in one way, you know.
That's where the challenge is.

1348
01:44:10,450 --> 01:44:15,610
It's trying to see how well the
constraints can be harmoniously

1349
01:44:16,170 --> 01:44:17,790
resolved.
Yeah.

1350
01:44:17,950 --> 01:44:21,230
Are there any recent
developments or extensions of

1351
01:44:21,230 --> 01:44:23,150
the theory that you find
particularly exciting?

1352
01:44:23,470 --> 01:44:28,310
And how do you envision?
Yeah, I don't know if what you

1353
01:44:28,310 --> 01:44:31,270
wouldn't?
Well, I posted it on

1354
01:44:31,270 --> 01:44:35,350
connectionist CD net and
LinkedIn, but just a few days

1355
01:44:35,350 --> 01:44:40,150
ago did an article that I really
liked a lot and it represents

1356
01:44:40,150 --> 01:44:44,030
the synthesis of many of my
earlier discoveries.

1357
01:44:44,030 --> 01:44:50,840
You know, everything is
integrative and the title is how

1358
01:44:50,840 --> 01:44:57,560
children learn to understand
language meanings called in a

1359
01:44:57,560 --> 01:45:02,560
neural model of adult child
multimodal interactions in real

1360
01:45:02,560 --> 01:45:04,760
time.
And I'll say a little more about

1361
01:45:04,760 --> 01:45:06,920
that.
But you know what's very much in

1362
01:45:06,920 --> 01:45:12,600
the news now is ChatGPT, which
is just a big look up table

1363
01:45:14,360 --> 01:45:20,120
where people have thrown in
millions of tidbits and facts

1364
01:45:20,120 --> 01:45:24,680
and and you add on a
probabilistic algorithm to try

1365
01:45:24,680 --> 01:45:28,240
to predict the next 5 words
after the last five words, has

1366
01:45:28,240 --> 01:45:34,920
no concept of meaning.
But I'm discussing meaning and

1367
01:45:34,920 --> 01:45:37,520
I'll I'll read the abstract to
you in a minute.

1368
01:45:38,280 --> 01:45:43,340
And last year wrote an article I
like very much for understanding

1369
01:45:43,340 --> 01:45:47,300
the brain dynamics of music
learning and conscious

1370
01:45:47,300 --> 01:45:51,900
performance of lyrics and
melodies with variable rhythms

1371
01:45:51,900 --> 01:45:57,900
and beats.
And that is on my web page sites

1372
01:45:57,900 --> 01:46:03,780
WW EDU/DG for anyone who wants
to look at I Do everything Open

1373
01:46:03,780 --> 01:46:09,900
Access these days.
And it shows, you know, this

1374
01:46:09,900 --> 01:46:16,960
whole In addition to explaining
data about musical lyrics and

1375
01:46:19,840 --> 01:46:24,040
melodies and rhythms and beats,
it also shows how this

1376
01:46:24,040 --> 01:46:30,360
competence really coops
previously existing foundational

1377
01:46:30,360 --> 01:46:37,800
capabilities.
You know, it's like, why do we

1378
01:46:37,800 --> 01:46:41,500
move to the beat?
You know, things like that that

1379
01:46:41,500 --> 01:46:44,900
get explained.
And before that, as I've

1380
01:46:44,900 --> 01:46:49,140
indicated with the discussion of
Matisse and other artists, I did

1381
01:46:49,660 --> 01:46:53,660
stuff on how visual artists make
paintings and how humans

1382
01:46:53,660 --> 01:46:59,140
consciously see them.
So you know, I'm very into

1383
01:47:02,340 --> 01:47:07,700
these, you know, triumphs of our
civilization of the human

1384
01:47:07,700 --> 01:47:10,690
spirit.
Let me read you the abstract

1385
01:47:10,690 --> 01:47:13,090
about meanings if you indulge
me.

1386
01:47:18,330 --> 01:47:22,210
This article describes the
biological neural network model

1387
01:47:22,210 --> 01:47:26,210
that can be used to explain how
children learn to understand

1388
01:47:26,210 --> 01:47:31,650
language meetings about the
perceptual and affective events

1389
01:47:32,250 --> 01:47:33,970
that they consciously
experience.

1390
01:47:33,970 --> 01:47:37,890
Like talked about perception.
I've talked about cognitive

1391
01:47:37,890 --> 01:47:41,490
emotional dynamics.
I need that of it.

1392
01:47:41,650 --> 01:47:44,450
This kind of learning often
occurs when a child interacts

1393
01:47:44,450 --> 01:47:49,610
with an adult teacher to learn
language meanings about events

1394
01:47:49,610 --> 01:47:53,290
that they experienced together.
And here's really the bottom

1395
01:47:53,290 --> 01:47:55,970
line.
Multiple types of self

1396
01:47:55,970 --> 01:47:59,690
organizing brain processes are
involved in learning language

1397
01:47:59,690 --> 01:48:04,130
meanings, including processes
that control conscious visual

1398
01:48:04,130 --> 01:48:08,900
perception, joint attention,
object learning, conscious

1399
01:48:08,900 --> 01:48:14,500
recognition, cognitive working
memory, cognitive planning,

1400
01:48:15,940 --> 01:48:19,860
emotion, cognitive emotional
interactions, volition, and goal

1401
01:48:19,860 --> 01:48:23,340
oriented actions.
These are all things that were

1402
01:48:23,340 --> 01:48:28,220
there but I could synthesize.
That's how an old man stays

1403
01:48:28,220 --> 01:48:32,340
productive.
The article shows that all these

1404
01:48:32,340 --> 01:48:36,390
brain processes interact to
enable learning language

1405
01:48:36,390 --> 01:48:39,710
meanings to occur and it
contrasts these human

1406
01:48:39,710 --> 01:48:44,670
capabilities with AI models like
chat, GTD.

1407
01:48:45,910 --> 01:48:51,830
And to emphasize that contrast,
they call the my model chat some

1408
01:48:52,230 --> 01:48:59,750
chatsome where some abbreviates
self organized meaning so.

1409
01:49:01,110 --> 01:49:05,000
I actually saw on LinkedIn.
And I was meant to to watch.

1410
01:49:05,000 --> 01:49:06,760
But I was at work.
I was meant to actually read

1411
01:49:06,760 --> 01:49:09,200
that paper.
Well, thank you.

1412
01:49:09,280 --> 01:49:12,920
I hope some people do.
Well, you know, I think I put

1413
01:49:12,920 --> 01:49:15,040
it.
I think I posted it on.

1414
01:49:15,720 --> 01:49:18,960
I think, I think it was
yesterday or the day before.

1415
01:49:19,360 --> 01:49:21,760
I might have posted it a day or
so ago.

1416
01:49:22,280 --> 01:49:26,880
Over 2000 people have seen it,
yeah, which is nice.

1417
01:49:27,160 --> 01:49:29,440
Yeah, I'm definitely looking
forward to that as well and also

1418
01:49:29,440 --> 01:49:32,790
share it as well.
Tell me, Steve, before, I mean

1419
01:49:32,870 --> 01:49:34,990
after we concluded this final
question, I mean, I have a few

1420
01:49:34,990 --> 01:49:38,350
more, but your overall vision,
that's OK And hope for the study

1421
01:49:38,350 --> 01:49:40,990
of mine in the future and your
message to future philosophers

1422
01:49:40,990 --> 01:49:42,270
and scientists joining the
field.

1423
01:49:44,110 --> 01:49:50,110
So I have two initial thoughts.
First, read the published

1424
01:49:50,150 --> 01:49:56,070
literature in the age of Google.
There's no excuse for trying to

1425
01:49:56,670 --> 01:50:01,200
rediscover the wheel or not to
put to find a point on it.

1426
01:50:02,000 --> 01:50:05,400
You don't want to waste a lot of
your life just plagiarizing

1427
01:50:06,040 --> 01:50:09,520
because it won't really help you
in the long run.

1428
01:50:10,880 --> 01:50:14,040
And if you want to understand
the latest theoretical concepts

1429
01:50:14,040 --> 01:50:18,200
and models about how brain makes
minds and certainly include in

1430
01:50:18,200 --> 01:50:24,740
your reading work that my
colleagues have done and I wrote

1431
01:50:24,740 --> 01:50:28,060
the magnum opus Conscious Mind
Resonant Brain how each brain

1432
01:50:28,060 --> 01:50:31,500
makes mind cause a lot of people
say you know I love your work

1433
01:50:31,500 --> 01:50:33,300
but it's scattered all over the
place.

1434
01:50:33,740 --> 01:50:37,820
Now you have at least that as a
synthesizing.

1435
01:50:38,180 --> 01:50:42,300
And then if you want details,
archival articles.

1436
01:50:42,300 --> 01:50:50,760
I have over 560 archival papers
on my web page and I have no

1437
01:50:51,000 --> 01:50:54,560
videos of keynote lectures I've
given about this and that, all

1438
01:50:55,120 --> 01:50:59,360
well which I try to make is
self-contained as possible.

1439
01:51:00,320 --> 01:51:05,440
And then the second point, which
I think for young people is very

1440
01:51:05,440 --> 01:51:09,360
important, is find a problem
you're passionate about.

1441
01:51:11,240 --> 01:51:14,160
One that you're willing to spend
the hundreds of hours that

1442
01:51:14,160 --> 01:51:17,280
you're going to need to learn
and think about it.

1443
01:51:17,980 --> 01:51:19,260
Anything.
If you're going to make any

1444
01:51:19,260 --> 01:51:24,300
progress at all and not everyone
is going to be passionate about

1445
01:51:24,300 --> 01:51:29,020
a problem on mind and brain
oriented science, well then

1446
01:51:29,020 --> 01:51:32,220
don't become a theorist, don't
become an experimentalist.

1447
01:51:33,140 --> 01:51:38,140
You know, for example if you
still like reading stuff in

1448
01:51:38,140 --> 01:51:44,550
science or other areas, 1 noble
profession is teaching, which is

1449
01:51:44,550 --> 01:51:51,030
very satisfying profession.
I always love to teach and there

1450
01:51:51,030 --> 01:51:56,710
are already all innumerable
productive careers you can carry

1451
01:51:56,710 --> 01:52:00,750
out that aren't in academia.
Whoever you are or troubled

1452
01:52:00,750 --> 01:52:07,190
world needs your talents.
So just try to do some good in

1453
01:52:07,190 --> 01:52:09,150
the world.
You know, we're only here for a

1454
01:52:09,150 --> 01:52:16,070
short time.
Try to make it so you know one

1455
01:52:16,070 --> 01:52:20,390
concept of immortality is people
will have positive thoughts

1456
01:52:20,390 --> 01:52:26,270
about you being here for the
brief time you were so good

1457
01:52:26,270 --> 01:52:28,910
works.
You know, that's wonderful

1458
01:52:28,910 --> 01:52:31,470
advice, Steve.
I mean, you've got an h-index up

1459
01:52:31,470 --> 01:52:33,870
around 130.
You've been cited thousands of

1460
01:52:33,870 --> 01:52:36,680
times.
You, you continue to pump out

1461
01:52:36,680 --> 01:52:38,000
work.
I mean, what what's the next

1462
01:52:38,000 --> 01:52:40,520
step for you?
What are you or how old are you

1463
01:52:40,520 --> 01:52:41,960
at the moment and where are you
headed?

1464
01:52:41,960 --> 01:52:44,320
Like what is the plan here?
Because you just keep going.

1465
01:52:44,680 --> 01:52:53,520
I I am 83, my mom lived to 101,
but I have I have serious

1466
01:52:53,560 --> 01:52:56,920
asthma.
So if my asthma doesn't kill me,

1467
01:52:56,920 --> 01:53:00,640
there's no history of dementia
in the family.

1468
01:53:00,640 --> 01:53:06,180
If if I don't get sick, I I'll
probably be productive for maybe

1469
01:53:06,180 --> 01:53:11,060
10 more years when I was.
What I'm doing now is writing

1470
01:53:11,060 --> 01:53:13,620
another book for a general
audience.

1471
01:53:14,300 --> 01:53:16,340
And is there anything you can
say about that book or is that

1472
01:53:16,340 --> 01:53:21,420
still just spoiler free?
No, I I I don't like talking

1473
01:53:21,420 --> 01:53:27,340
about things I haven't yet done
because I might not do them, you

1474
01:53:27,340 --> 01:53:30,280
know?
Yeah, I might change my mind.

1475
01:53:31,120 --> 01:53:34,280
I don't know.
You know, I'm fascinated writing

1476
01:53:34,280 --> 01:53:37,800
the book.
I'm enjoying it.

1477
01:53:38,600 --> 01:53:45,440
But you know, if tomorrow for
some reason something goes pop

1478
01:53:45,440 --> 01:53:49,880
in my mind and I have a another
big idea and I say, God, I

1479
01:53:49,880 --> 01:53:54,080
really got to follow this out.
I'll drop the book and and do

1480
01:53:54,080 --> 01:53:58,310
that project if I can.
Because I'm retired, I've been

1481
01:53:58,310 --> 01:54:06,430
emeritus for since 2019.
I no longer have big research

1482
01:54:06,430 --> 01:54:09,470
grants, and along with that, I
don't have to think of them and

1483
01:54:09,470 --> 01:54:15,270
write them and and get PhD
students and post doc and train

1484
01:54:15,270 --> 01:54:18,990
them to work together.
And, you know, so I can work

1485
01:54:18,990 --> 01:54:22,950
solo, but that means everything
that I'm doing, I have to do

1486
01:54:23,780 --> 01:54:28,420
with the foundation in place
that's been built over 66 years.

1487
01:54:29,020 --> 01:54:32,340
So far the foundation's been
very useful, like, for the music

1488
01:54:32,340 --> 01:54:38,820
work, the artwork, the meaning
work.

1489
01:54:41,820 --> 01:54:44,100
But I don't know what's going to
happen next.

1490
01:54:44,100 --> 01:54:47,020
I really don't.
I I love writing this book.

1491
01:54:47,740 --> 01:54:51,460
It'll keep me busy if I'm brain
dead for a new, new, new, new

1492
01:54:51,460 --> 01:54:55,220
idea.
But but if I have an idea, I'll

1493
01:54:55,220 --> 01:54:57,620
go for it.
But I don't.

1494
01:54:59,300 --> 01:55:04,460
Yeah, What I may do have to
encumber myself.

1495
01:55:05,860 --> 01:55:11,220
For example, I'm 83, so it's
unfair to even think about

1496
01:55:11,540 --> 01:55:13,980
taking on a graduate student or
post doc.

1497
01:55:14,700 --> 01:55:16,820
That's at least the three-year
commitment.

1498
01:55:17,620 --> 01:55:19,860
I could be dead in three years.
It's not right.

1499
01:55:19,860 --> 01:55:27,550
And I I certainly I'm not
interested in writing grants.

1500
01:55:30,750 --> 01:55:31,950
Tell me, Stevie, is this?
Was it taught?

1501
01:55:31,950 --> 01:55:35,030
What?
Is this book specifically a way

1502
01:55:35,030 --> 01:55:37,510
to explore new ideas, or is it
something that you want to

1503
01:55:37,510 --> 01:55:39,830
clarify and just add on old
ones?

1504
01:55:40,710 --> 01:55:44,270
Well, it's a synthesis of things
that I already know.

1505
01:55:45,470 --> 01:55:49,990
But you know, the synthesis
isn't just what you already

1506
01:55:49,990 --> 01:55:52,130
know, especially in your where
there can be.

1507
01:55:52,210 --> 01:55:55,610
So much information installed,
yeah.

1508
01:55:55,610 --> 01:55:58,770
So for me it's like light
reading.

1509
01:55:58,970 --> 01:56:05,010
I'm enjoying it, but I try to be
very careful with style and

1510
01:56:05,610 --> 01:56:08,330
clarity and logical development
of ideas.

1511
01:56:08,330 --> 01:56:11,890
So we'll see how it goes.
I don't know.

1512
01:56:12,970 --> 01:56:15,690
And one of the reasons when I
started this podcast, Steve, was

1513
01:56:15,690 --> 01:56:19,870
because when I went.
To school, university, etc.

1514
01:56:19,870 --> 01:56:22,630
And I spoke about my favorite
scientists and all these people

1515
01:56:22,630 --> 01:56:25,550
who write amazing books and
publish brilliant journals.

1516
01:56:25,550 --> 01:56:27,550
Like yourself.
A lot of people don't really

1517
01:56:27,550 --> 01:56:29,870
know about these people.
A lot of the times it's these

1518
01:56:29,870 --> 01:56:31,510
are the people who they're not
really talking about.

1519
01:56:31,510 --> 01:56:34,630
And Someone Like You, in your
case, even though you've been

1520
01:56:34,750 --> 01:56:37,630
cited thousands of times, you're
an emeritus professor, you've

1521
01:56:37,630 --> 01:56:40,510
been given several awards.
There's still so many things

1522
01:56:40,510 --> 01:56:43,070
you're not recognized for, and
this was a way to.

1523
01:56:43,470 --> 01:56:45,430
Provide people with that
platform to talk about those.

1524
01:56:45,430 --> 01:56:47,430
Well, that's true.
You know, in fact, you were

1525
01:56:47,430 --> 01:56:52,830
saying, I think, what is it,
around 83,000 citations.

1526
01:56:53,630 --> 01:56:59,430
But if you look at the people
who have been credited with a

1527
01:56:59,430 --> 01:57:05,630
lot of the ideas I had first and
developed, and you estimate how

1528
01:57:05,630 --> 01:57:11,870
many citations they got from
using my work, I would have

1529
01:57:11,870 --> 01:57:14,230
three or four times that many
citations.

1530
01:57:14,830 --> 01:57:20,070
But citations aren't the reason
to work.

1531
01:57:21,070 --> 01:57:23,310
You know, I'm in it for the long
haul.

1532
01:57:23,310 --> 01:57:28,910
I have in theater brick wall.
And to avoid hitting a brick

1533
01:57:28,910 --> 01:57:39,210
wall, you you have to study
problems rich enough and give

1534
01:57:39,210 --> 01:57:43,530
enough evidence for dealing with
them that they'll carry you to

1535
01:57:43,530 --> 01:57:48,690
New and exciting ideas.
Do.

1536
01:57:49,850 --> 01:57:52,490
You feel there are any specific
ideas that you felt were

1537
01:57:52,810 --> 01:57:55,650
rightfully yours and you brought
those models out first, or that

1538
01:57:55,650 --> 01:58:06,250
you want to mention at all?
Well, I and Maltsberg, although

1539
01:58:06,250 --> 01:58:10,090
I did the correct version, made
competitive learning.

1540
01:58:10,890 --> 01:58:14,050
I introduced adaptive resonance
theory.

1541
01:58:15,090 --> 01:58:18,170
I've had many wonderful
collaborators developing and

1542
01:58:18,170 --> 01:58:23,130
still developing.
I introduced the basic equations

1543
01:58:23,130 --> 01:58:29,410
for on center or surround
networks, which solved another

1544
01:58:29,410 --> 01:58:31,690
dilemma, the noise saturation
dilemma.

1545
01:58:31,690 --> 01:58:33,490
And I had a thought experiment
to that.

1546
01:58:34,290 --> 01:58:39,850
Basically, if you look at my
work, there are a few equations,

1547
01:58:40,290 --> 01:58:42,770
fundamental equations.
They're mine.

1548
01:58:43,610 --> 01:58:50,930
I introduced them in 1957.
I introduced the paradigm of

1549
01:58:50,930 --> 01:58:54,730
nonlinear systems with
differential equations to link

1550
01:58:55,450 --> 01:58:58,010
brain to mind through neural
networks.

1551
01:58:59,290 --> 01:59:04,810
In 1957, I introduced the
somewhat larger number of

1552
01:59:04,810 --> 01:59:10,150
modules on micro circuits.
This is all on my Wikipedia

1553
01:59:10,150 --> 01:59:12,510
page.
If you ever read Steven

1554
01:59:12,510 --> 01:59:18,510
Grossberg, Wikipedia and, for
example, Shunting on center

1555
01:59:18,510 --> 01:59:20,950
restaurant networks, recurring
and nonrecurrent.

1556
01:59:21,910 --> 01:59:28,470
They are used in multiple parts
of the brain in including in

1557
01:59:28,470 --> 01:59:34,550
walking, running, cognition,
perception, emotion.

1558
01:59:35,270 --> 01:59:40,720
It's a very foundational kind of
circuit.

1559
01:59:40,720 --> 01:59:47,000
I've introduced all the basic
modules very early on and then

1560
01:59:47,440 --> 01:59:52,640
what I do is I combine
specialized versions of the

1561
01:59:52,640 --> 01:59:58,400
microcircuits and equations and
what I call modal architectures,

1562
01:59:58,400 --> 02:00:04,280
modal for different modalities,
vision, audition, etc etc.

1563
02:00:05,550 --> 02:00:08,750
And that's my legacy, the
equations, the modules and the

1564
02:00:08,910 --> 02:00:14,870
architectures, in a word, that
is the theory, that's how it's

1565
02:00:14,870 --> 02:00:19,070
organized.
And and and yeah, I've gotten

1566
02:00:19,390 --> 02:00:23,110
partial credit for it and I am
grateful for it.

1567
02:00:23,110 --> 02:00:28,550
And if I if you work for that,
you'll never never keep going

1568
02:00:29,190 --> 02:00:32,630
cause the whole point is to be
ahead of the game.

1569
02:00:35,490 --> 02:00:39,450
I'll do my best to at least
provide some people with certain

1570
02:00:39,450 --> 02:00:41,650
things just to at least help you
grow this knowledge, because I

1571
02:00:41,650 --> 02:00:46,610
think people need to know.
I think they'd enjoy knowing,

1572
02:00:46,610 --> 02:00:50,570
yeah.
I mean, a lot of people who were

1573
02:00:50,570 --> 02:00:53,410
drawn to my book have enjoyed
reading parts of it.

1574
02:00:54,330 --> 02:00:59,330
And and you know a lot of them
who want to know about things

1575
02:00:59,330 --> 02:01:03,440
they didn't think anyone knew
about, they could just at least

1576
02:01:03,440 --> 02:01:07,960
read titles and abstracts on my
web page.

1577
02:01:08,000 --> 02:01:11,440
And then if they want to get
into it, I always start a paper

1578
02:01:11,440 --> 02:01:15,960
by talking about what is the big
problem, what are the main

1579
02:01:15,960 --> 02:01:21,760
facts, Very non-technical.
I always talk to my students

1580
02:01:21,760 --> 02:01:26,040
about writing a paper that
there's an exponential loss of

1581
02:01:26,080 --> 02:01:30,480
readers with every sentence.
So if you can't get all the main

1582
02:01:30,480 --> 02:01:36,350
ideas of your paper into the
first paragraph, you've lost it.

1583
02:01:36,950 --> 02:01:41,590
But then, of course, as I've
indicated, after the simple

1584
02:01:41,590 --> 02:01:47,310
heuristics, there's the tough
demands of mathematical rigor.

1585
02:01:48,510 --> 02:01:53,630
Equations don't lie.
I've personally seen your change

1586
02:01:53,630 --> 02:01:56,350
over the years.
Being from, I've read a lot of

1587
02:01:56,350 --> 02:01:58,870
your work, watched you for a
long time as well, but you can

1588
02:01:58,870 --> 02:02:00,950
see that as the years have gone
by, you've become.

1589
02:02:01,580 --> 02:02:04,100
Much better at being more
self-contained and actually

1590
02:02:04,100 --> 02:02:06,500
doing this and it.
It's testament to how much you

1591
02:02:06,500 --> 02:02:08,580
actually know about the topic
over these years and how much

1592
02:02:08,580 --> 02:02:13,980
you've learned even more Well,
on the notion of self-contained

1593
02:02:13,980 --> 02:02:18,460
also depends on what you know.
That's true, because I might

1594
02:02:18,460 --> 02:02:21,100
think it's perfectly
self-contained.

1595
02:02:21,500 --> 02:02:24,500
I find that.
But if you know nothing about

1596
02:02:24,500 --> 02:02:28,060
the field, you might say, what
planet is this person coming

1597
02:02:28,060 --> 02:02:29,980
from?
I think that's but how long?

1598
02:02:30,510 --> 02:02:32,910
How long have you been
interested in the mind?

1599
02:02:32,910 --> 02:02:35,830
I have no idea.
So for me, it's been for as long

1600
02:02:35,830 --> 02:02:38,710
as I can remember.
I was first inspired to read

1601
02:02:38,710 --> 02:02:42,390
about this when I read Oliver
Sacks book The Man Who Mistook

1602
02:02:42,390 --> 02:02:46,390
His Wife for a Hat and I read,
but he's a wonderful expirator.

1603
02:02:46,710 --> 02:02:49,750
Yeah, so I remember in that
moment just we're having this

1604
02:02:49,790 --> 02:02:53,550
urge to learn about the mind.
And it hasn't stopped since.

1605
02:02:53,830 --> 02:02:57,190
But I still remember back in
those days when I read some of

1606
02:02:57,190 --> 02:02:59,210
your work.
It was a period where I'd

1607
02:02:59,210 --> 02:03:03,450
forgotten about it and then
slowly reintroduced myself to it

1608
02:03:04,330 --> 02:03:08,490
and realized how much better it
was as an older man of it.

1609
02:03:09,450 --> 02:03:11,690
Well, well.
My understanding of all of the

1610
02:03:11,690 --> 02:03:18,610
Sex Real Page turn is about
really fascinating topics, but

1611
02:03:18,610 --> 02:03:21,930
doesn't really explain how
anything works.

1612
02:03:22,810 --> 02:03:26,420
So we might have a book about
hallucinations, yes.

1613
02:03:32,780 --> 02:03:36,820
And the same is true for any
mental disorders that he might

1614
02:03:36,820 --> 02:03:27,700
write about.
But you have to come to me to

1615
02:03:27,700 --> 02:03:39,620
understand how that happens.
You know, and as I said, I

1616
02:03:39,620 --> 02:03:44,620
didn't just say Grand can
understand autism or whatever.

1617
02:03:45,180 --> 02:03:48,580
It sort of has to come out of
the wash through the back door,

1618
02:03:49,140 --> 02:03:54,180
has to be implied by what you've
already done, and it opens the

1619
02:03:54,180 --> 02:04:00,620
gateway to a new area.
But I've been doing that for so

1620
02:04:00,620 --> 02:04:04,620
long now, you know, I've gone
through maybe all the doors I'm

1621
02:04:04,620 --> 02:04:07,860
never gonna go through now.
Pretty big house.

1622
02:04:08,260 --> 02:04:13,980
Yeah, tell me, Steve, I mean,
obviously closing off, but who

1623
02:04:14,220 --> 02:04:17,260
growing up and doing this?
I mean, who inspired you?

1624
02:04:17,260 --> 02:04:18,380
Who?
Who were the people?

1625
02:04:18,380 --> 02:04:21,180
Who, the scientists and
philosophers in particular, that

1626
02:04:21,180 --> 02:04:26,960
really got you going?
Well, you know, I am a pioneer

1627
02:04:27,760 --> 02:04:32,920
and pioneers sort of come out of
nowhere, but I was lucky.

1628
02:04:34,280 --> 02:04:43,520
Well, first I grew up in New
York at a time when it was very

1629
02:04:43,520 --> 02:04:46,160
hard for Jewish kids to get in
college.

1630
02:04:46,680 --> 02:04:54,950
It was a Jewish quota, so we
worked super hard to get good

1631
02:04:54,950 --> 02:05:01,230
enough scores on general tests
and grades that we might get a

1632
02:05:01,230 --> 02:05:03,630
scholarship to go to a good
school.

1633
02:05:04,710 --> 02:05:09,070
Of course there was CCNY in New
York was a wonderful school that

1634
02:05:09,070 --> 02:05:13,150
a lot of poor kids whose parents
were immigrants or they were

1635
02:05:13,150 --> 02:05:16,590
immigrants themselves, went to,
and they became very often

1636
02:05:16,590 --> 02:05:20,910
famous scholars and scientists.
But I really wanted to get the

1637
02:05:20,910 --> 02:05:27,310
hell out in New York, and so I
went to Dartmouth because first,

1638
02:05:27,310 --> 02:05:30,750
they gave me the biggest
scholarship, but second, they

1639
02:05:30,750 --> 02:05:35,230
had something they called the
senior fellowship, which meant

1640
02:05:35,230 --> 02:05:41,230
in your senior year you could
not take courses, you could just

1641
02:05:41,230 --> 02:05:44,430
do research.
No, I didn't know what research

1642
02:05:44,430 --> 02:05:47,590
was.
But I did know that I really,

1643
02:05:47,590 --> 02:05:51,950
really needed the year before I
got out of college to figure out

1644
02:05:51,950 --> 02:05:53,550
what to do with the rest of my
life.

1645
02:05:54,830 --> 02:06:00,230
So when I went to Dartmouth, as
I indicated at the beginning, I

1646
02:06:00,230 --> 02:06:04,950
took introductory psychology and
my head exploded and I started

1647
02:06:04,950 --> 02:06:08,190
making discoveries.
But that could have led in a

1648
02:06:08,190 --> 02:06:10,710
very different direction than it
did.

1649
02:06:11,390 --> 02:06:16,150
But luckily I had two wonderful
mentors.

1650
02:06:16,940 --> 02:06:20,140
One is John Kemeny.
Have you ever heard of John

1651
02:06:20,140 --> 02:06:23,980
Kemeny?
Well he was Albert Einstein's

1652
02:06:23,980 --> 02:06:30,020
last assistant.
He Co he Co invented the basic

1653
02:06:30,740 --> 02:06:34,900
computer language.
He Co invented computer time

1654
02:06:34,900 --> 02:06:39,820
sharing systems.
So he was very into computing

1655
02:06:41,580 --> 02:06:46,310
and having been so close to
Einstein, he knew what it meant

1656
02:06:46,310 --> 02:06:51,830
to be a loner.
And he also wrote a wonderful

1657
02:06:51,990 --> 02:06:55,150
book called The Philosopher
Looks at Science.

1658
02:06:56,310 --> 02:07:02,270
And I was lucky to take a course
that read out of that book when

1659
02:07:02,270 --> 02:07:07,350
I was a sophomore.
And until that point coming out

1660
02:07:07,350 --> 02:07:14,370
of New York where your ranking
like it's Stuyvesant where I was

1661
02:07:14,370 --> 02:07:17,690
valedictorian.
I don't know if you've heard of

1662
02:07:17,690 --> 02:07:21,410
Stuyvesant but being some people
think being valedictorian

1663
02:07:21,410 --> 02:07:23,610
Stuyvesant is the hardest thing
you'll ever do.

1664
02:07:24,650 --> 02:07:29,330
But I was at a party after
graduation and the guy who was a

1665
02:07:29,330 --> 02:07:32,570
little drunk came up to me.
It was true, you know,

1666
02:07:32,610 --> 02:07:35,850
Stuyvesant, Bronx, Einstein said
you're Grossburg.

1667
02:07:36,650 --> 02:07:40,210
And I said, yeah, I've been mad.
And he said, yeah, but I've

1668
02:07:40,210 --> 02:07:44,010
hated you my whole year time at
Stardisson.

1669
02:07:44,850 --> 02:07:48,690
I said, why'd you hate me so
Well, I wanted you to die

1670
02:07:49,170 --> 02:07:53,210
because my grade, my ranking
would go up.

1671
02:07:53,210 --> 02:08:00,690
I want it was so competitive
around 92 point, up to the 100th

1672
02:08:00,690 --> 02:08:06,650
decimal point.
There was 700 students out of a

1673
02:08:06,650 --> 02:08:11,700
class of 1500.
So 100th of a decimal point of

1674
02:08:11,700 --> 02:08:15,980
your cumulatives grade could
mean you wouldn't get in any

1675
02:08:15,980 --> 02:08:17,820
college.
Oh wow.

1676
02:08:18,060 --> 02:08:21,500
Yeah, if you were a Jew and most
of us were Jews.

1677
02:08:21,820 --> 02:08:23,700
And that's most of us.
And you were always first.

1678
02:08:24,180 --> 02:08:27,060
In your class, I said.
On top of that, you were always

1679
02:08:27,060 --> 02:08:29,580
first in your class.
I mean, this is basically

1680
02:08:30,140 --> 02:08:33,100
basically, yeah, a pet.
So, so anyway.

1681
02:08:33,660 --> 02:08:41,480
So I knew I had to stop learning
to compete, which was a survival

1682
02:08:41,480 --> 02:08:49,080
skill, and Start learning out of
love and my seat in my freshman

1683
02:08:49,080 --> 02:08:53,080
year.
I made my discoveries, and I

1684
02:08:53,080 --> 02:08:58,480
took Kennedy's classes
sophomore, and he wrote that I

1685
02:08:59,640 --> 02:09:03,480
wrote the best essays he had
ever written he had ever read

1686
02:09:03,840 --> 02:09:10,730
without qualification, and in my
introductory economics course.

1687
02:09:10,730 --> 02:09:15,810
My final was so sensational,
apparently, that my teacher said

1688
02:09:15,810 --> 02:09:20,530
it was at the level of a PhD
thesis and my brother, who was a

1689
02:09:20,530 --> 02:09:24,930
different college, heard about
it, which didn't make it easy

1690
02:09:24,930 --> 02:09:32,330
for him to have me as a brother.
So I I worked harder than you

1691
02:09:32,330 --> 02:09:39,350
should ever work for knowledge.
But and then my other mentor was

1692
02:09:39,350 --> 02:09:41,790
Al Hast of, who was chairman of
psychology.

1693
02:09:41,790 --> 02:09:47,150
So our chairman of math gem of
psychology and Al went on to

1694
02:09:48,710 --> 02:09:51,350
become a Dean and Provost at
Stanford.

1695
02:09:51,350 --> 02:09:56,070
He was a wonderful man.
So those are the two pillars of

1696
02:09:56,070 --> 02:09:58,310
my work, mathematics and
psychology.

1697
02:09:58,310 --> 02:10:02,230
And I became the first joint
major in mathematics and

1698
02:10:02,230 --> 02:10:06,190
psychology at Darkness.
And in my senior fellowship

1699
02:10:06,190 --> 02:10:12,310
year, I worked like a dog to
further develop the discoveries

1700
02:10:12,310 --> 02:10:19,590
I started making as a freshman.
And then I try to go to a place

1701
02:10:19,590 --> 02:10:22,230
where they would let me keep
working.

1702
02:10:23,670 --> 02:10:29,830
And they were only two places in
the world at the time that had

1703
02:10:29,830 --> 02:10:34,110
programs at what was called
mathematical psychology.

1704
02:10:35,880 --> 02:10:41,160
But for example, the one closest
to psychology was at Stanford.

1705
02:10:42,680 --> 02:10:46,360
It was run by William Estes, who
was a very great

1706
02:10:46,360 --> 02:10:49,920
experimentalist.
He did wonderful work on

1707
02:10:49,920 --> 02:10:54,720
learning with the BF Skinner in
1941.

1708
02:10:54,720 --> 02:10:59,240
They published Anyway thought I
can go there, they'll get me.

1709
02:10:59,240 --> 02:11:05,800
But they didn't, because Bill
was an experimentalist, ever had

1710
02:11:05,800 --> 02:11:15,920
made so he could do was to
develop an iron model so urns

1711
02:11:17,960 --> 02:11:22,560
can can you hear?
It's getting slightly, it's

1712
02:11:22,560 --> 02:11:26,680
slightly blurry.
OK, I think one SEC.

1713
02:11:31,900 --> 02:11:37,620
Can you hear me now?
Yeah, my Earpods died.

1714
02:11:41,700 --> 02:11:48,380
So an urn model.
You have an urn with elements

1715
02:11:48,380 --> 02:11:52,220
that are learned in urn with
elements that are forgotten.

1716
02:11:52,900 --> 02:11:56,740
You talk about the probability
of learning.

1717
02:11:56,740 --> 02:11:59,980
You go from unlearned to
learned, the probability of

1718
02:11:59,980 --> 02:12:02,060
forgetting, going from learning
to unlearn.

1719
02:12:02,700 --> 02:12:06,500
So he expresses in terms of
simple matrices.

1720
02:12:07,380 --> 02:12:09,500
And that was as far as Bill
could go.

1721
02:12:09,500 --> 02:12:18,300
It was a group data, stationary
probabilities, and I walked in

1722
02:12:18,300 --> 02:12:22,260
talking about real time
autonomous adaptation of

1723
02:12:22,260 --> 02:12:27,180
individuals in a changing world.
They didn't know what the hell

1724
02:12:27,180 --> 02:12:30,640
to do with me.
And anyway, there were a whole

1725
02:12:30,640 --> 02:12:36,000
series of mishaps.
So finally there's another cute

1726
02:12:36,000 --> 02:12:38,120
story.
I ended up at the Rockefeller

1727
02:12:38,160 --> 02:12:43,600
Institute for Medical Research
in New York, where I got my PhD

1728
02:12:43,600 --> 02:12:47,640
for proving global limit
theorems about learning and

1729
02:12:47,640 --> 02:12:50,960
memory.
And that was an accident.

1730
02:12:50,960 --> 02:12:52,840
You know, I was very unhappy at
Stanford.

1731
02:12:52,840 --> 02:12:56,560
So I wrote a letter to the
Rockefeller Institute because I

1732
02:12:56,560 --> 02:12:58,680
heard they were taking on some
students.

1733
02:12:59,650 --> 02:13:04,410
And the next thing I knew, they
invited me to fly back to New

1734
02:13:04,410 --> 02:13:08,170
York for an interview because
they checked me out and I would

1735
02:13:08,170 --> 02:13:10,490
have gone anyway.
I wanted to see my family.

1736
02:13:10,490 --> 02:13:16,930
My parents live in Queens, and
Rockefeller had a ton of money,

1737
02:13:17,930 --> 02:13:20,690
and their trustees recommended
you got it.

1738
02:13:21,410 --> 02:13:23,130
You've had a lot of Nobel
laureates.

1739
02:13:23,130 --> 02:13:25,610
Now it's time to teach a new
generation.

1740
02:13:26,720 --> 02:13:32,720
And so my interview was I had
lunch with Mark Kotz, KACA very

1741
02:13:32,720 --> 02:13:37,760
famous probability theorist and
functional analyst, a fellow

1742
02:13:37,760 --> 02:13:43,760
Hungarian.
And we had it in this amazing

1743
02:13:44,640 --> 02:13:53,240
paneled room overlooking gardens
with art and paintings on the

1744
02:13:53,240 --> 02:13:59,640
walls, linen, silverware.
Anyway, the big big Mark told me

1745
02:14:00,400 --> 02:14:05,800
was Steve never grow old.
That was our interview.

1746
02:14:06,800 --> 02:14:10,520
And then I was ushered in to the
president's office.

1747
02:14:10,520 --> 02:14:20,720
The president with deadly Bronc
PRONK, who was family, founded

1748
02:14:20,720 --> 02:14:25,560
the Bronx, the borough of the
Bronx in New York and Colonial

1749
02:14:25,560 --> 02:14:27,800
times.
He had previously been

1750
02:14:27,800 --> 02:14:31,280
president.
Johns Hopkins and I go into his

1751
02:14:31,280 --> 02:14:36,800
fabulously, elegantly furnished
office just as the sun is

1752
02:14:36,800 --> 02:14:41,360
setting, casting dramatic
shadows over the room.

1753
02:14:42,120 --> 02:14:45,880
And he sits me on this lovely
sofa and he looks me in the eye.

1754
02:14:45,880 --> 02:14:51,330
And he said, Steve, if we fun
that you would you want to come

1755
02:14:51,330 --> 02:14:53,610
to Rockefeller?
That was my interview.

1756
02:14:54,050 --> 02:14:58,090
Oh geez.
And I said yes, it's so.

1757
02:14:58,570 --> 02:15:02,250
At what point did you realize OK
you are you are definitely above

1758
02:15:02,250 --> 02:15:04,370
the wrist and kind of onto
something seriously big.

1759
02:15:08,050 --> 02:15:12,810
I always knew I had a gift.
In fact, when I was in public

1760
02:15:12,810 --> 02:15:19,200
school, my friend said I was
born with a golden spoon in my

1761
02:15:19,200 --> 02:15:22,440
mouth because I was good at
everything.

1762
02:15:23,880 --> 02:15:28,280
It wasn't to be showy, in fact.
And the second part of the

1763
02:15:28,280 --> 02:15:34,280
sentence is, and you're a nice
guy, so it wasn't something that

1764
02:15:34,280 --> 02:15:38,720
I I felt was a reason to show
off.

1765
02:15:38,720 --> 02:15:46,400
In fact, I was just was good at
everything I did and including

1766
02:15:46,400 --> 02:15:52,760
our like I want a competitive
award to study at the Museum of

1767
02:15:52,760 --> 02:15:57,560
Modern Art when I was in high
school etcetera.

1768
02:15:57,560 --> 02:16:03,120
So I was good at these things.
I was playing within the first

1769
02:16:03,120 --> 02:16:07,520
few months of taking piano
Rhapsody and who and Debussy and

1770
02:16:07,520 --> 02:16:13,680
Bach.
So I'm gifted, but I realized

1771
02:16:13,680 --> 02:16:19,700
that I really can't do world
class work in those things.

1772
02:16:19,820 --> 02:16:24,940
For starters, I don't have the
hands of a great classical

1773
02:16:24,940 --> 02:16:27,900
pianist.
Then I started too late.

1774
02:16:27,900 --> 02:16:31,220
I started piano, you know, as
much older.

1775
02:16:31,220 --> 02:16:34,820
Most people start when they're
five or six if they become

1776
02:16:34,820 --> 02:16:41,740
constant or it's not everyone.
But I realized what I can do is

1777
02:16:41,740 --> 02:16:46,190
I have a very good mind for
imagining other worlds.

1778
02:16:47,270 --> 02:16:50,790
And that's what I did.
I found the world that was so

1779
02:16:50,790 --> 02:16:54,990
rich I could imagine it for the
rest of my life.

1780
02:16:55,910 --> 02:16:57,990
And I didn't.
I was lucky that it happened

1781
02:16:57,990 --> 02:17:01,190
when I was a freshman.
I think I might have been my

1782
02:17:01,190 --> 02:17:05,230
first term.
A lot of people have existential

1783
02:17:05,230 --> 02:17:08,190
struggles.
My struggle wasn't what I wanted

1784
02:17:08,190 --> 02:17:12,510
to do, which I was passionately
committed to, is how to survive

1785
02:17:12,510 --> 02:17:16,180
it.
Yeah, it's just like if someone

1786
02:17:16,180 --> 02:17:21,620
says that's green, but I know
it's red, I have a choice.

1787
02:17:22,299 --> 02:17:25,660
Either I accept that it's red or
I go crazy.

1788
02:17:26,780 --> 02:17:30,860
I knew what I saw.
I saw my discoveries before I

1789
02:17:30,860 --> 02:17:36,660
had any equation for them, and I
followed my heart and my my

1790
02:17:36,660 --> 02:17:38,980
imagination.
I don't know.

1791
02:17:39,420 --> 02:17:44,049
Not sure if this is very
informative, but so I had some

1792
02:17:44,049 --> 02:17:48,129
real high points and some real
difficulties by people wanted to

1793
02:17:48,129 --> 02:17:56,370
throw me out of Rockefeller for
nefarious for nefarious reasons.

1794
02:17:56,370 --> 02:18:00,809
They failed mostly because they
didn't understand what I was

1795
02:18:00,809 --> 02:18:06,570
doing and it made them feel, you
know, unqualified.

1796
02:18:07,850 --> 02:18:11,480
But I got my PhD from
Rockefeller, and then Marcottes

1797
02:18:11,480 --> 02:18:15,040
and John Carova, who are both
professors there, wrote me

1798
02:18:15,040 --> 02:18:20,760
glowing reviews.
And I I've always been lucky

1799
02:18:20,760 --> 02:18:23,799
that way.
Like when I got to Stanford,

1800
02:18:26,680 --> 02:18:30,080
Richard Atkinson, who then went
on to become president of the

1801
02:18:30,080 --> 02:18:33,280
National Academy of Sciences,
who was a professor of

1802
02:18:33,280 --> 02:18:37,000
psychology, said, oh, the other
one, Kennedy said, is going to

1803
02:18:37,000 --> 02:18:41,180
save method of psychology.
So I always had someone who was,

1804
02:18:42,100 --> 02:18:45,700
and cots wrote to MIT that I was
the most creative person he'd

1805
02:18:45,700 --> 02:18:49,299
ever met.
So these people, you know, came

1806
02:18:49,299 --> 02:18:52,139
through when it really counted.