1
00:00:00,000 --> 00:00:03,700
Imagine you're just sitting in a sterile, brightly

2
00:00:03,700 --> 00:00:05,900
lit laboratory. You know the kind. Oh, absolutely.

3
00:00:05,960 --> 00:00:08,759
The classic sci -fi setup. Yeah, exactly. And

4
00:00:08,759 --> 00:00:10,679
you have all these wires glued to your scalp,

5
00:00:10,779 --> 00:00:13,660
reading the electrical storms inside your brain,

6
00:00:13,859 --> 00:00:16,640
plus more wires taped to your forearm, just like

7
00:00:16,640 --> 00:00:19,399
measuring every tiny flex of your muscles. Sounds

8
00:00:19,399 --> 00:00:22,199
incredibly uncomfortable, to be honest. Right.

9
00:00:22,579 --> 00:00:25,719
But the researcher gives you one incredibly simple

10
00:00:25,719 --> 00:00:29,039
instruction. Just watch this highly precise clock

11
00:00:29,039 --> 00:00:32,320
on the wall. And whenever you feel the spontaneous,

12
00:00:32,500 --> 00:00:35,740
conscious urge to do so, flick your finger. Just

13
00:00:35,740 --> 00:00:38,200
flick it. Just flick it. That's it. And note

14
00:00:38,200 --> 00:00:41,159
the exact millisecond on the clock when you made

15
00:00:41,159 --> 00:00:44,299
that conscious choice. I mean, it sounds like

16
00:00:44,299 --> 00:00:46,619
the most basic demonstration of human free will

17
00:00:46,619 --> 00:00:49,259
possible. You're the boss. You make a decision.

18
00:00:49,579 --> 00:00:52,420
You issue a command. And your body obeys. Exactly.

19
00:00:52,939 --> 00:00:54,880
That is the assumption we literally all live

20
00:00:54,880 --> 00:00:58,079
by. But. When the neuroscientist Benjamin LaBette

21
00:00:58,079 --> 00:01:01,799
ran this exact experiment back in 1983, I think?

22
00:01:01,920 --> 00:01:04,739
Yep, 1983. Yeah, the machines told a profoundly

23
00:01:04,739 --> 00:01:07,560
disturbing story. The sensors showed that the

24
00:01:07,560 --> 00:01:09,840
brain activity preparing for the movement, like

25
00:01:09,840 --> 00:01:12,420
the engine revving up to flick the finger, started

26
00:01:12,420 --> 00:01:15,019
more than half a second before the physical movement

27
00:01:15,019 --> 00:01:18,340
actually happened. Which honestly makes sense

28
00:01:18,340 --> 00:01:21,099
mechanically. Right. But here's the part that

29
00:01:21,099 --> 00:01:22,900
totally shattered our understanding of the mind.

30
00:01:23,019 --> 00:01:26,180
Yeah, this is the crazy part. Preparatory brain

31
00:01:26,180 --> 00:01:29,840
activity started about 300 milliseconds before

32
00:01:29,840 --> 00:01:32,459
the person was even consciously aware they wanted

33
00:01:32,459 --> 00:01:36,379
to move. 300 milliseconds. So the brain was already

34
00:01:36,379 --> 00:01:38,980
initiating the action before the conscious mind

35
00:01:38,980 --> 00:01:42,959
had even decided to do it. Exactly. The conscious

36
00:01:42,959 --> 00:01:45,599
experience of willing the movement was essentially

37
00:01:45,599 --> 00:01:48,500
just a delayed echo. It was playing catch up.

38
00:01:48,719 --> 00:01:51,480
Wow. Your conscious mind was basically taking

39
00:01:51,480 --> 00:01:55,340
credit for an action that a deeper, darker, physically

40
00:01:55,340 --> 00:01:57,980
grounded system had already set in motion. Okay.

41
00:01:58,299 --> 00:02:01,560
Let's unpack this. Because that completely dismantles

42
00:02:01,560 --> 00:02:04,540
the illusion that you or I are the conscious

43
00:02:04,540 --> 00:02:07,680
CEO of our bodies, sitting in some pristine mental

44
00:02:07,680 --> 00:02:10,360
boardroom issuing orders to our limbs. It really

45
00:02:10,360 --> 00:02:13,000
does. It flips everything upside down. Right.

46
00:02:13,319 --> 00:02:15,860
And today, we are going to dismantle that illusion

47
00:02:15,860 --> 00:02:19,259
entirely. We're taking a deep dive into a massive

48
00:02:19,259 --> 00:02:22,099
paradigm shift in how we understand intelligence

49
00:02:22,099 --> 00:02:25,659
itself. We are being guided by an incredibly

50
00:02:25,659 --> 00:02:28,659
dense groundbreaking book called How the Body

51
00:02:28,659 --> 00:02:31,400
Shapes the Way We Think. A New View of Intelligence

52
00:02:31,400 --> 00:02:33,960
by Rolf Pfeiffer and Josh Bongard. That's the

53
00:02:33,960 --> 00:02:36,740
one. In our mission today is to show you why

54
00:02:36,740 --> 00:02:39,960
the brain is not a lonely supercomputer. We're

55
00:02:39,960 --> 00:02:42,479
going to explore how your physical body, its

56
00:02:42,479 --> 00:02:46,060
squishiness, its shape, its limitations, is an

57
00:02:46,060 --> 00:02:49,150
absolute non -negotiable prerequisite for you

58
00:02:49,150 --> 00:02:52,949
to be able to think at all. The stakes of this

59
00:02:52,949 --> 00:02:55,229
paradigm shift are just enormous because for

60
00:02:55,229 --> 00:02:58,710
decades, scientists, roboticists, and well, philosophers

61
00:02:58,710 --> 00:03:01,370
chased a model of intelligence that was fundamentally

62
00:03:01,370 --> 00:03:03,949
flawed. Because they ignored the body, right?

63
00:03:04,129 --> 00:03:06,969
Totally. We built early artificial intelligence

64
00:03:06,969 --> 00:03:09,270
on the assumption that the mind was the supreme

65
00:03:09,270 --> 00:03:11,789
commander and the body was just a mindless meat

66
00:03:11,789 --> 00:03:15,430
puppet. A meat puppet. I love that, but it's

67
00:03:15,430 --> 00:03:18,319
also kind of terrifying. It is. And because of

68
00:03:18,319 --> 00:03:20,639
that blind spot, we completely fail to understand

69
00:03:20,639 --> 00:03:23,860
how human beings actually move, how infants learn

70
00:03:23,860 --> 00:03:26,360
language, and how our memories actually work.

71
00:03:26,780 --> 00:03:28,479
By the end of this conversation, you are going

72
00:03:28,479 --> 00:03:31,599
to see your own mind -body connection in a completely

73
00:03:31,599 --> 00:03:34,280
alien way. Well, we have to figure out where

74
00:03:34,280 --> 00:03:36,639
this profound misunderstanding even came from

75
00:03:36,639 --> 00:03:39,800
first. Like, why do we all naturally feel like

76
00:03:39,800 --> 00:03:42,180
we're a software program trapped inside a hardware

77
00:03:42,180 --> 00:03:44,930
shell? For that, we have to lay the historical

78
00:03:44,930 --> 00:03:48,469
blame mostly at the feet of the 17th century

79
00:03:48,469 --> 00:03:51,310
French philosopher René Descartes. Descartes.

80
00:03:51,490 --> 00:03:53,569
Right. Of course. He was a brilliant thinker,

81
00:03:53,669 --> 00:03:57,030
obviously, but he gave us the phrase, I think

82
00:03:57,030 --> 00:04:00,449
therefore I am. Right. Which codified what we

83
00:04:00,449 --> 00:04:03,419
now call the Cartesian heritage. Exactly. It's

84
00:04:03,419 --> 00:04:06,060
the ultimate dualism. Descartes basically argued

85
00:04:06,060 --> 00:04:08,719
that the mind, you know, the ethereal, rational,

86
00:04:08,900 --> 00:04:11,560
thinking essence of a human, and the physical

87
00:04:11,560 --> 00:04:14,620
body were two completely separate substances.

88
00:04:14,840 --> 00:04:17,540
Two different things entirely. Yeah. To him,

89
00:04:17,620 --> 00:04:19,860
the body was just a mechanical automaton full

90
00:04:19,860 --> 00:04:22,939
of pumps and levers, and the mind was a divine,

91
00:04:23,040 --> 00:04:25,879
non -physical ghost haunting that machine. I

92
00:04:25,879 --> 00:04:28,420
mean, it is a very seductive idea, you know?

93
00:04:28,459 --> 00:04:31,860
It feeds our ego. We love feeling like our rationality

94
00:04:31,860 --> 00:04:34,720
is completely separate from the messy, sweaty,

95
00:04:35,019 --> 00:04:36,720
gravity -bound reality of the physical world.

96
00:04:37,060 --> 00:04:39,779
Oh, it was so seductive that it became the invisible

97
00:04:39,779 --> 00:04:43,160
bedrock of Western science for centuries. And

98
00:04:43,160 --> 00:04:45,379
when the computer age dawned in the mid -20th

99
00:04:45,379 --> 00:04:48,720
century, that Cartesian dualism was mapped directly

100
00:04:48,720 --> 00:04:50,860
onto the new technology. The brain became the

101
00:04:50,860 --> 00:04:52,939
software and the body became the interchangeable

102
00:04:52,939 --> 00:04:55,620
hardware. Right, which culminated in that massive

103
00:04:55,620 --> 00:04:58,439
initiative in 1956, the Dartmouth Conference.

104
00:04:58,779 --> 00:05:02,139
Yes, the conference that officially launched

105
00:05:02,139 --> 00:05:05,120
what philosopher John Hauglin later dubbed GoFi,

106
00:05:05,699 --> 00:05:08,019
good old fashioned artificial intelligence. Good

107
00:05:08,019 --> 00:05:11,019
old GoFi. And their entire approach was based

108
00:05:11,019 --> 00:05:14,420
on this top -down brain as CEO model, right?

109
00:05:14,420 --> 00:05:16,879
They believed intelligence was simply about processing

110
00:05:16,879 --> 00:05:20,620
abstract symbols. That was the core tenet. The

111
00:05:20,620 --> 00:05:23,079
founding fathers of AI looked at human intelligence

112
00:05:23,079 --> 00:05:26,279
and just assumed its absolute highest peak was

113
00:05:26,279 --> 00:05:29,540
abstract logic. Like math and chess. Exactly.

114
00:05:29,920 --> 00:05:31,560
They thought, well, if we can just build a machine

115
00:05:31,560 --> 00:05:34,620
that manipulates logical symbols, like if A equals

116
00:05:34,620 --> 00:05:38,639
B and B equals C, then A equals C, we will have

117
00:05:38,639 --> 00:05:41,500
created a synthetic mind. And to be fair to them,

118
00:05:41,500 --> 00:05:44,100
they did have some massive early victories that

119
00:05:44,100 --> 00:05:47,759
probably inflated their egos quite a bit. Oh,

120
00:05:47,759 --> 00:05:49,620
definitely. I mean, they built programs that

121
00:05:49,620 --> 00:05:52,180
could solve complex algebra and prove mathematical

122
00:05:52,180 --> 00:05:54,860
theorems. They eventually built machines that

123
00:05:54,860 --> 00:05:57,860
could play championship level chess. So from

124
00:05:57,860 --> 00:05:59,959
the outside looking in, it really looked like

125
00:05:59,959 --> 00:06:02,519
they had cracked the code of intelligence. They

126
00:06:02,519 --> 00:06:06,100
cracked a very specific type of specialized logic.

127
00:06:06,740 --> 00:06:09,980
But then they hit an unbelievable paradox. Right.

128
00:06:10,199 --> 00:06:14,000
Because the real world is messy. Exactly. The

129
00:06:14,000 --> 00:06:16,860
classical AI was phenomenally good at things

130
00:06:16,860 --> 00:06:19,800
human beings consider incredibly hard, like chess.

131
00:06:20,160 --> 00:06:22,879
But it completely failed at things a human toddler

132
00:06:22,879 --> 00:06:26,660
finds effortlessly easy. Like just walking across

133
00:06:26,660 --> 00:06:29,500
a messy living room without falling over. Or

134
00:06:29,500 --> 00:06:33,060
recognizing a mother's face in a crowd. Or just

135
00:06:33,060 --> 00:06:35,060
reaching out and picking up a fragile coffee

136
00:06:35,060 --> 00:06:38,500
cup without crushing it into dust. Why was walking

137
00:06:38,500 --> 00:06:41,360
such a nightmare for these million -dollar supercomputers?

138
00:06:41,839 --> 00:06:43,740
Because they were trying to solve the problem

139
00:06:43,740 --> 00:06:46,740
of walking using that top -down Cartesian approach.

140
00:06:47,540 --> 00:06:50,040
Oh, trying to map everything out first. Right.

141
00:06:50,120 --> 00:06:53,120
The engineers would attempt to program a perfect,

142
00:06:53,540 --> 00:06:55,779
internally updated three -dimensional map of

143
00:06:55,779 --> 00:06:57,959
the environment into the robot's central processor.

144
00:06:58,519 --> 00:07:02,379
That sounds exhausting. It is. To take a single

145
00:07:02,379 --> 00:07:05,180
step, the robot's computer had to calculate the

146
00:07:05,180 --> 00:07:08,279
exact geometric angle of its knee joint, the

147
00:07:08,279 --> 00:07:11,139
exact distance to the floor in millimeters, the

148
00:07:11,139 --> 00:07:14,079
precise torque required for the motor, and like

149
00:07:14,079 --> 00:07:17,050
the estimated friction of the surface. But the

150
00:07:17,050 --> 00:07:19,610
real world refuses to sit still for those calculations.

151
00:07:19,790 --> 00:07:21,689
I mean, the lighting changes, right? Casting

152
00:07:21,689 --> 00:07:24,329
a shadow that looks like a hole. Or the floor

153
00:07:24,329 --> 00:07:26,730
is just slightly more slippery than expected.

154
00:07:27,250 --> 00:07:29,689
That's exactly the problem. By the time the central

155
00:07:29,689 --> 00:07:32,050
processor finished crunching the millions of

156
00:07:32,050 --> 00:07:34,350
variables to dictate the quote -unquote perfect

157
00:07:34,350 --> 00:07:37,430
step, the physical environment had already changed.

158
00:07:37,569 --> 00:07:40,699
Wow. The robot would step down, hit an unexpected

159
00:07:40,699 --> 00:07:43,660
pebble, the central processor would get totally

160
00:07:43,660 --> 00:07:46,220
overwhelmed by the contradictory data, and the

161
00:07:46,220 --> 00:07:48,879
robot would completely freeze and tip over. The

162
00:07:48,879 --> 00:07:50,920
processing bottleneck was just insurmountable.

163
00:07:51,360 --> 00:07:53,620
Meanwhile, human beings can casually walk down

164
00:07:53,620 --> 00:07:57,339
a crowded sidewalk, drink a coffee, smoke a cigarette,

165
00:07:57,720 --> 00:08:00,939
and have a complex philosophical debate all simultaneously

166
00:08:00,939 --> 00:08:04,519
without falling over. Exactly. Or as the book

167
00:08:04,519 --> 00:08:07,439
mentions, Do the whole Ministry of Silly Walks

168
00:08:07,439 --> 00:08:10,540
routine like John Cleese. Right, right. So the

169
00:08:10,540 --> 00:08:12,959
top -down AI approach is basically like me trying

170
00:08:12,959 --> 00:08:15,360
to drive a car by staring obsessively at the

171
00:08:15,360 --> 00:08:17,800
GPS map on the dashboard. Yes. Like trying to

172
00:08:17,800 --> 00:08:19,480
perfectly calculate my coordinates on the little

173
00:08:19,480 --> 00:08:21,740
screen instead of just looking out the windshield

174
00:08:21,740 --> 00:08:24,500
at the actual road. Eventually you are going

175
00:08:24,500 --> 00:08:26,279
to crash into a parked truck because your internal

176
00:08:26,279 --> 00:08:28,639
map doesn't match the immediate reality. That

177
00:08:28,639 --> 00:08:31,579
is a phenomenal way to visualize it. The obsession

178
00:08:31,579 --> 00:08:34,620
with the internal map caused the crash. And recognizing

179
00:08:34,620 --> 00:08:37,580
this failure led to a massive rebellion in robotics

180
00:08:37,580 --> 00:08:41,220
in the 1980s. The embodied turn. The embodied

181
00:08:41,220 --> 00:08:43,899
turn. It was heavily championed by researchers

182
00:08:43,899 --> 00:08:47,360
like Rodney Brooks down at MIT. Brooks realized

183
00:08:47,360 --> 00:08:49,480
that trying to build human -level AI from the

184
00:08:49,480 --> 00:08:51,620
top down, starting with logic and chess, was

185
00:08:51,620 --> 00:08:55,159
just a fool's errand. So what did he do? He argued

186
00:08:55,159 --> 00:08:57,500
we needed to start from the bottom up. We needed

187
00:08:57,500 --> 00:09:00,379
to look at creatures that were brilliant at navigating

188
00:09:00,379 --> 00:09:02,779
the physical world, but had almost zero brain

189
00:09:02,779 --> 00:09:04,940
power. He started looking at insects, didn't

190
00:09:04,940 --> 00:09:08,340
he? Yes. Because an ant or a fly does not have

191
00:09:08,340 --> 00:09:10,899
the neural capacity to hold a 3D rendering of

192
00:09:10,899 --> 00:09:13,240
the world in its head. It literally doesn't have

193
00:09:13,240 --> 00:09:16,200
the hardware for an internal map. Right. It simply

194
00:09:16,200 --> 00:09:18,480
reacts to its immediate physical environment.

195
00:09:19,419 --> 00:09:21,419
Brooks coined a slogan that became a battle cry

196
00:09:21,419 --> 00:09:24,179
for embodied cognition, which was, the world

197
00:09:24,179 --> 00:09:27,059
is its own best model. The world is its own best

198
00:09:27,059 --> 00:09:30,360
model. Meaning, why waste precious brain energy

199
00:09:30,360 --> 00:09:32,639
trying to simulate the room in your head when

200
00:09:32,639 --> 00:09:34,399
the actual room is sitting right in front of

201
00:09:34,399 --> 00:09:36,659
you. Exactly. Just use your sensors to look at

202
00:09:36,659 --> 00:09:38,620
the room. The environment itself holds all the

203
00:09:38,620 --> 00:09:40,960
information you need. You just have to physically

204
00:09:40,960 --> 00:09:44,679
interact with it. But I'm stuck on a logistical

205
00:09:44,679 --> 00:09:48,179
problem here. OK, what's that? If we aren't heavily

206
00:09:48,179 --> 00:09:50,159
filtering and processing the world through an

207
00:09:50,159 --> 00:09:54,500
internal map, how do we avoid just short circuiting?

208
00:09:54,700 --> 00:09:57,610
Because every single second Millions of photons

209
00:09:57,610 --> 00:09:59,549
are hitting my retinas. Sound waves are vibrating

210
00:09:59,549 --> 00:10:02,769
in my ears. My skin is detecting micro fluctuations

211
00:10:02,769 --> 00:10:05,490
in temperature. It's a lot of data. Right. If

212
00:10:05,490 --> 00:10:07,730
I'm just an agent reacting to the raw physical

213
00:10:07,730 --> 00:10:10,250
world, how do I not collapse from data overload?

214
00:10:10,789 --> 00:10:13,570
What's fascinating here is that you avoid data

215
00:10:13,570 --> 00:10:16,110
overload not by building a better central processor,

216
00:10:16,690 --> 00:10:18,669
but by letting your physical body and the environment

217
00:10:18,669 --> 00:10:22,149
absorb the complexity for you. OK, wait. Absorb

218
00:10:22,149 --> 00:10:24,990
it for me? Yeah, we have to introduce a concept

219
00:10:24,990 --> 00:10:28,230
called the frame of reference problem. To understand

220
00:10:28,230 --> 00:10:30,529
how intelligence works, we must fundamentally

221
00:10:30,529 --> 00:10:32,789
distinguish between the perspective of the observer

222
00:10:32,789 --> 00:10:35,549
and the perspective of the agent actually doing

223
00:10:35,549 --> 00:10:38,509
the action. Okay, this feels like a critical

224
00:10:38,509 --> 00:10:41,470
pivot. How do we distinguish those perspectives?

225
00:10:41,909 --> 00:10:44,269
We can look at a very famous anecdote from the

226
00:10:44,269 --> 00:10:47,730
Nobel laureate Herbert Simon. He asks us to imagine

227
00:10:47,730 --> 00:10:50,490
an ant walking across a bumpy, windswept beach,

228
00:10:50,789 --> 00:10:53,600
trying to get back to its nest. Okay, I'm picturing

229
00:10:53,600 --> 00:10:56,580
an ant on a beach. If you stand over the beach

230
00:10:56,580 --> 00:10:59,159
and trace the ant's path with a pen, you will

231
00:10:59,159 --> 00:11:02,000
draw an incredibly chaotic, complex, winding,

232
00:11:02,200 --> 00:11:04,500
zigzagging line. It oops around, it doubles back,

233
00:11:04,659 --> 00:11:08,889
it scales steep inclines. Sure. If I'm the human

234
00:11:08,889 --> 00:11:11,710
looking down at that path, I would look at that

235
00:11:11,710 --> 00:11:14,549
complex geometry and assume the ant must be a

236
00:11:14,549 --> 00:11:17,149
brilliant navigator. I would assume it has a

237
00:11:17,149 --> 00:11:19,730
sophisticated cognitive engine mapping out that

238
00:11:19,730 --> 00:11:22,610
zigzag route to avoid obstacles. That is the

239
00:11:22,610 --> 00:11:25,070
trap of the observer's frame of reference. You

240
00:11:25,070 --> 00:11:27,669
project your own cognitive complexity onto the

241
00:11:27,669 --> 00:11:31,309
agent. Oh, wow. But Simon points out that from

242
00:11:31,309 --> 00:11:34,129
the ant's frame of reference, the agent's perspective,

243
00:11:34,669 --> 00:11:37,789
there is no complex map. The ant's internal rules

244
00:11:37,789 --> 00:11:42,059
are shockingly simple. Like what? Like, if obstacle

245
00:11:42,059 --> 00:11:45,179
on left, turn right. If obstacle on right, turn

246
00:11:45,179 --> 00:11:48,740
left. If smell pheromone, move forward. That's

247
00:11:48,740 --> 00:11:52,179
it? The ant is practically mindless. The complexity

248
00:11:52,179 --> 00:11:54,700
of the path does not reside in the ant's brain.

249
00:11:55,159 --> 00:11:57,759
The complexity is an illusion generated by the

250
00:11:57,759 --> 00:12:00,059
interaction between a very simple set of physical

251
00:12:00,059 --> 00:12:03,279
rules and a highly complex chaotic environment.

252
00:12:03,679 --> 00:12:05,659
The random puddles, pebbles, and twigs on the

253
00:12:05,659 --> 00:12:08,539
beach. Exactly. That is a harsh reality check.

254
00:12:08,879 --> 00:12:11,240
We project genius out of the ant, but the ant

255
00:12:11,240 --> 00:12:14,600
is basically just the DVD screensaver logo bouncing

256
00:12:14,600 --> 00:12:17,399
off the edges of the TV screen. Yes. Perfect

257
00:12:17,399 --> 00:12:19,440
analogy. It just goes until it hits something

258
00:12:19,440 --> 00:12:21,740
and then changes direction. And this isn't just

259
00:12:21,740 --> 00:12:24,279
a thought experiment. Pfeiffer and Bongard detail

260
00:12:24,279 --> 00:12:26,580
a real -world robotics experiment that proved

261
00:12:26,580 --> 00:12:29,559
this definitively, known as the Swiss robots.

262
00:12:29,919 --> 00:12:32,820
Oh right, the cleanup robots. What did these

263
00:12:32,820 --> 00:12:36,340
robots look like? They were incredibly rudimentary.

264
00:12:36,820 --> 00:12:39,460
Small wheeled cylinders, they had a motor to

265
00:12:39,460 --> 00:12:42,419
drive forward, and only two basic infrared sensors

266
00:12:42,419 --> 00:12:44,500
mounted on their front to detect if they were

267
00:12:44,500 --> 00:12:46,570
bumping into something. And that was it. That

268
00:12:46,570 --> 00:12:49,710
was their entire sensory apparatus. The researchers

269
00:12:49,710 --> 00:12:52,009
programmed them with the simplest possible reflex,

270
00:12:52,350 --> 00:12:55,149
just drive straight. If the left sensor detects

271
00:12:55,149 --> 00:12:58,049
a bump, reverse and turn right. If the right

272
00:12:58,049 --> 00:13:00,169
sensor detects a bump, reverse and turn left.

273
00:13:00,750 --> 00:13:02,870
So absolutely no cameras, no internal memory,

274
00:13:03,090 --> 00:13:05,370
no Wi -Fi connection talking to a mainframe somewhere.

275
00:13:05,710 --> 00:13:08,470
Nothing. They were functionally blind amnesiac

276
00:13:08,470 --> 00:13:12,330
snowplows. Amnesiac snowplows. Yeah. So the researchers

277
00:13:12,330 --> 00:13:14,470
placed several of these robots into an enclosed

278
00:13:14,470 --> 00:13:17,269
arena. and scattered completely randomly across

279
00:13:17,269 --> 00:13:19,470
the floor of this arena were dozens of little

280
00:13:19,470 --> 00:13:22,289
Styrofoam cubes. They turned the robots on and

281
00:13:22,289 --> 00:13:24,269
just let them blindly bump around the arena for

282
00:13:24,269 --> 00:13:26,330
a few hours. I mean, if they are just bouncing

283
00:13:26,330 --> 00:13:28,230
off things randomly, I would assume they just

284
00:13:28,230 --> 00:13:30,389
shoved the Styrofoam cubes around into an ever

285
00:13:30,389 --> 00:13:32,649
-changing chaotic mess. That is what you would

286
00:13:32,649 --> 00:13:35,289
expect, right? But when the researchers came

287
00:13:35,289 --> 00:13:38,230
back, something magical had occurred. What happened?

288
00:13:38,629 --> 00:13:42,299
The arena was totally clean. The blind robots

289
00:13:42,299 --> 00:13:44,639
had gathered all the randomly scattered Styrofoam

290
00:13:44,639 --> 00:13:47,039
cubes into neat, organized clusters in the center

291
00:13:47,039 --> 00:13:49,860
of the room. They had categorized their environment.

292
00:13:50,500 --> 00:13:52,799
How is that physically possible? You just said

293
00:13:52,799 --> 00:13:55,279
they have no memory. They don't even have a program

294
00:13:55,279 --> 00:13:56,879
that tells them what a cluster is. They don't

295
00:13:56,879 --> 00:13:59,580
even know what a cube is. They don't. The quote

296
00:13:59,580 --> 00:14:01,539
-unquote intelligence of the cleanup was not

297
00:14:01,539 --> 00:14:04,500
in the software. It was engineered entirely into

298
00:14:04,500 --> 00:14:06,940
the physical design of the robots, specifically

299
00:14:06,940 --> 00:14:09,919
the spacing of their two front sensors. The spacing.

300
00:14:10,090 --> 00:14:12,669
Yeah. The engineers intentionally placed the

301
00:14:12,669 --> 00:14:14,929
left and right sensors slightly farther apart

302
00:14:14,929 --> 00:14:17,909
than the width of a single styrofoam cube. Okay,

303
00:14:17,970 --> 00:14:20,809
wait. I think I see where this is going. So imagine

304
00:14:20,809 --> 00:14:23,389
a robot driving straight toward a lone cube.

305
00:14:23,950 --> 00:14:26,190
The cube slips right between the sensors without

306
00:14:26,190 --> 00:14:29,330
triggering either of them. The robot is completely

307
00:14:29,330 --> 00:14:31,789
unaware it has hit anything, so it just keeps

308
00:14:31,789 --> 00:14:34,330
driving straight, pushing the cube along mechanically.

309
00:14:34,669 --> 00:14:37,110
Like food caught in your teeth. You don't even

310
00:14:37,110 --> 00:14:40,419
realize it's there. Precisely. But... If that

311
00:14:40,419 --> 00:14:42,639
robot happens to randomly push its cube into

312
00:14:42,639 --> 00:14:45,320
another cube lying on the floor, suddenly the

313
00:14:45,320 --> 00:14:47,620
mass of Styrofoam in front of the robot is two

314
00:14:47,620 --> 00:14:50,740
cubes wide. And two cubes are wider than the

315
00:14:50,740 --> 00:14:53,639
gap between the sensors. Instantly, the sensors

316
00:14:53,639 --> 00:14:56,399
detect the mass. The simple programming kicks

317
00:14:56,399 --> 00:14:59,840
in, bump detected, reverse and turn. The robot

318
00:14:59,840 --> 00:15:01,980
backs away, leaving the two cubes sitting there

319
00:15:01,980 --> 00:15:05,039
together and drives off in a new direction. Wow.

320
00:15:05,299 --> 00:15:08,320
Over time, simply through the statistical probability

321
00:15:08,320 --> 00:15:11,200
of these physical encounters, all the cubes get

322
00:15:11,200 --> 00:15:14,039
deposited into larger and larger piles. That

323
00:15:14,039 --> 00:15:17,919
is genuinely mind -blowing. Because from the

324
00:15:17,919 --> 00:15:20,460
perspective of a human looking down at the arena,

325
00:15:21,340 --> 00:15:23,620
those robots look like brilliant little janitors

326
00:15:23,620 --> 00:15:26,679
working together to tidy up a room. But from

327
00:15:26,679 --> 00:15:28,960
the robot's perspective, it was just driving

328
00:15:28,960 --> 00:15:31,240
straight until its face hit a wall and then backing

329
00:15:31,240 --> 00:15:34,509
up. The intelligence was literally in the empty

330
00:15:34,509 --> 00:15:37,230
physical space between the sensors. Exactly.

331
00:15:37,669 --> 00:15:40,350
And this forces us to confront a pretty humbling

332
00:15:40,350 --> 00:15:42,549
truth. If we want to understand intelligence,

333
00:15:42,710 --> 00:15:45,409
we cannot just look at the brain. We have to

334
00:15:45,409 --> 00:15:47,789
analyze the system through three distinct lenses

335
00:15:47,789 --> 00:15:50,909
or timeframes. OK, let's break those down because

336
00:15:50,909 --> 00:15:53,450
this really feels like the core framework of

337
00:15:53,450 --> 00:15:56,730
the whole philosophy. The first lens is the here

338
00:15:56,730 --> 00:15:59,299
and now. This is the immediate physical reaction

339
00:15:59,299 --> 00:16:02,200
of a body in an environment. It's the Swiss robot

340
00:16:02,200 --> 00:16:04,720
bumping into the cube, or your hand instantly

341
00:16:04,720 --> 00:16:06,799
pulling away from a hot stove before your conscious

342
00:16:06,799 --> 00:16:09,659
mind even registers the heat. Just raw physics

343
00:16:09,659 --> 00:16:12,740
and reflex. Right. The second lens is learning

344
00:16:12,740 --> 00:16:15,659
and development, which biologists call ontogenesis.

345
00:16:16,220 --> 00:16:18,460
This looks at how an agent's behavior changes

346
00:16:18,460 --> 00:16:21,000
over the course of its own lifetime. So like

347
00:16:21,000 --> 00:16:23,440
a baby flailing its arms randomly, eventually

348
00:16:23,440 --> 00:16:26,419
learning to crawl and then walk? Exactly. And

349
00:16:26,419 --> 00:16:29,960
the third lens. Evolution. Yes, evolution or

350
00:16:29,960 --> 00:16:32,899
phylogenesis. This is the grandest scale. It

351
00:16:32,899 --> 00:16:35,179
asks how the physical shape of the species, like

352
00:16:35,179 --> 00:16:37,440
the specific design of the joints, the density

353
00:16:37,440 --> 00:16:40,080
of the muscles, was shaped over millions of years

354
00:16:40,080 --> 00:16:42,759
of natural selection to make that specific type

355
00:16:42,759 --> 00:16:44,899
of learning and movement possible in the first

356
00:16:44,899 --> 00:16:47,419
place. You know, this makes me deeply suspicious

357
00:16:47,419 --> 00:16:50,039
of my own brain. I mean, if the intelligence

358
00:16:50,039 --> 00:16:53,059
of the Swiss robots is essentially a mirage,

359
00:16:53,519 --> 00:16:55,580
created by their physical shape in the environment.

360
00:16:56,279 --> 00:16:58,539
Does that mean much of human intelligence is

361
00:16:58,539 --> 00:17:01,960
an illusion too? It's a fair question. Are we

362
00:17:01,960 --> 00:17:04,140
just walking around blindly following physical

363
00:17:04,140 --> 00:17:07,900
rules while our conscious minds invent a complex

364
00:17:07,900 --> 00:17:11,079
narrative to explain it all? It absolutely suggests

365
00:17:11,079 --> 00:17:13,779
that a vast amount of what we consider complex

366
00:17:13,779 --> 00:17:16,880
human genius is actually just our bodies exploiting

367
00:17:16,880 --> 00:17:19,869
the physics of our environment for free. We are

368
00:17:19,869 --> 00:17:22,450
constantly attributing deep cognitive intent

369
00:17:22,450 --> 00:17:25,130
to things that are just following physical dynamics.

370
00:17:25,569 --> 00:17:28,289
I totally do that. Have you ever yelled at your

371
00:17:28,289 --> 00:17:30,309
laptop because you felt like it was intentionally

372
00:17:30,309 --> 00:17:33,670
hiding a file from you? All the time. Or looked

373
00:17:33,670 --> 00:17:36,410
at your dog cowering by a chewed up shoe and

374
00:17:36,410 --> 00:17:40,289
thought, he feels so guilty. We project our rich

375
00:17:40,289 --> 00:17:43,289
internal emotional world onto absolutely everything.

376
00:17:43,730 --> 00:17:46,670
We really do. The laptop doesn't hate you and

377
00:17:46,670 --> 00:17:49,549
the dog doesn't feel moral guilt. The dog just

378
00:17:49,549 --> 00:17:52,150
knows that assuming a submissive physical posture

379
00:17:52,150 --> 00:17:55,930
stops the loud human from yelling. We are chronic

380
00:17:55,930 --> 00:17:58,990
anthropomorphizers. But if the intelligence isn't

381
00:17:58,990 --> 00:18:02,079
sitting in a vault in our brains, where is the

382
00:18:02,079 --> 00:18:04,799
computation actually happening? How does the

383
00:18:04,799 --> 00:18:07,259
physical body think? Well, this brings us to

384
00:18:07,259 --> 00:18:09,819
a massive concept in embodied cognition called

385
00:18:09,819 --> 00:18:12,509
morphological computation. The basic premise

386
00:18:12,509 --> 00:18:15,390
is that you can offload incredibly complex neural

387
00:18:15,390 --> 00:18:18,049
processing directly onto the physical structures

388
00:18:18,049 --> 00:18:21,609
of your body. Wait, offload the processing to

389
00:18:21,609 --> 00:18:24,609
the body? Yeah. The shape and material of your

390
00:18:24,609 --> 00:18:26,910
body parts do the mathematical heavy lifting

391
00:18:26,910 --> 00:18:29,569
for you. I'm going to need a very grounded example

392
00:18:29,569 --> 00:18:32,829
of this. Let's look at the simple act of grasping

393
00:18:32,829 --> 00:18:35,930
a coffee mug. In classical top -down robotics,

394
00:18:36,309 --> 00:18:38,869
engineering a hand to pick up a mug is a computational

395
00:18:38,869 --> 00:18:41,470
nightmare. Because of all the angles. Right.

396
00:18:41,549 --> 00:18:43,829
If you build a robotic hand out of rigid metal,

397
00:18:44,009 --> 00:18:46,869
the central processor has to calculate a staggering

398
00:18:46,869 --> 00:18:49,430
number of variables. It must determine the exact

399
00:18:49,430 --> 00:18:52,089
geometry of the mug's curvature. It must calculate

400
00:18:52,089 --> 00:18:54,789
the precise angle of approach for every single

401
00:18:54,789 --> 00:18:57,230
mechanical joint in the fingers. It has to measure

402
00:18:57,230 --> 00:18:59,589
the friction coefficient of ceramic and then

403
00:18:59,589 --> 00:19:01,750
dictate the exact amount of pressure the motors

404
00:19:01,750 --> 00:19:04,410
must apply. So it doesn't drop the heavy mug

405
00:19:04,410 --> 00:19:06,230
but also doesn't apply so much force that it

406
00:19:06,230 --> 00:19:09,009
shatters it. Exactly. It's an astronomical math

407
00:19:09,009 --> 00:19:11,660
problem just to take a sip of coffee. But when

408
00:19:11,660 --> 00:19:14,079
I reach out and grab my mug, I'm certainly not

409
00:19:14,079 --> 00:19:16,579
doing calculus in my head. Because you are not

410
00:19:16,579 --> 00:19:19,140
using rigid metal fingers. Imagine trying to

411
00:19:19,140 --> 00:19:21,579
pick up a wet, slippery glass while wearing hard

412
00:19:21,579 --> 00:19:23,920
metal thimbles on all of your fingers. Oh, that

413
00:19:23,920 --> 00:19:26,180
would be incredibly difficult. I'd have to focus

414
00:19:26,180 --> 00:19:28,599
entirely on the exact angle of my fingertips

415
00:19:28,599 --> 00:19:32,099
so the glass wouldn't slide out. It would take

416
00:19:32,099 --> 00:19:35,240
massive mental effort. Exactly the effort the

417
00:19:35,240 --> 00:19:38,460
top -down robot uses. But in reality, you have

418
00:19:38,460 --> 00:19:41,480
soft, fleshy, deformable tissue on your fingertips.

419
00:19:42,000 --> 00:19:43,920
When you reach for the mug, you don't calculate

420
00:19:43,920 --> 00:19:47,039
its geometry. You just shove your hand against

421
00:19:47,039 --> 00:19:50,380
it? Right. The soft fat and skin of your fingertips

422
00:19:50,380 --> 00:19:53,640
physically squash and deform around the curvature

423
00:19:53,640 --> 00:19:56,500
of the ceramic. So the flesh molds to the shape

424
00:19:56,500 --> 00:19:59,339
automatically? The physical elasticity of your

425
00:19:59,339 --> 00:20:02,720
skin solves the geometric math problem instantaneously,

426
00:20:02,859 --> 00:20:06,819
for free. That is amazing. The intelligence required

427
00:20:06,819 --> 00:20:09,160
to hold the cup isn't in your brain's processing

428
00:20:09,160 --> 00:20:12,660
power. It is literally embedded in the squishiness

429
00:20:12,660 --> 00:20:15,779
of your tissue. Your skin is computing the grip.

430
00:20:16,640 --> 00:20:19,480
Here's where it gets really interesting. It's

431
00:20:19,480 --> 00:20:22,160
exactly like a mechanical door closer. What do

432
00:20:22,160 --> 00:20:24,420
you mean? Well, if you wanted to build a door

433
00:20:24,420 --> 00:20:27,140
that closes itself slowly behind you, you could

434
00:20:27,140 --> 00:20:29,680
write a thousand lines of computer code, install

435
00:20:29,680 --> 00:20:32,279
a motion sensor, put a microprocessor in the

436
00:20:32,279 --> 00:20:34,940
door frame, and connect a motorized hinge to

437
00:20:34,940 --> 00:20:37,799
calculate the swing speed. A massively complex

438
00:20:37,799 --> 00:20:40,599
top -down system. Exactly. Or you could just

439
00:20:40,599 --> 00:20:43,559
screw a coiled piece of metal, a spring, to the

440
00:20:43,559 --> 00:20:46,799
door. The spring does the exact same job, perfectly,

441
00:20:47,039 --> 00:20:49,359
every single time, without needing a battery

442
00:20:49,359 --> 00:20:52,700
or a single line of code. Morphological computation

443
00:20:52,700 --> 00:20:55,480
is nature realizing that it's cheaper to use

444
00:20:55,480 --> 00:20:58,019
a spring. That is a brilliant way to look at

445
00:20:58,019 --> 00:21:01,670
it. Nature is ruthlessly frugal. Brain tissue

446
00:21:01,670 --> 00:21:04,410
requires massive amounts of caloric energy to

447
00:21:04,410 --> 00:21:07,369
run. Evolution will not waste expensive brain

448
00:21:07,369 --> 00:21:10,029
calories calculating a problem that simple physics

449
00:21:10,029 --> 00:21:13,410
and body mechanics can solve for free. Wow. We

450
00:21:13,410 --> 00:21:15,970
see this even more dramatically in human locomotion.

451
00:21:16,390 --> 00:21:18,250
Consider what happens to your body when you are

452
00:21:18,250 --> 00:21:20,609
jogging and you suddenly hit a steep downhill

453
00:21:20,609 --> 00:21:24,309
slope. If I hit a hill, I speed up. My legs start

454
00:21:24,309 --> 00:21:26,349
turning over much faster to catch my momentum,

455
00:21:26,349 --> 00:21:29,430
so I don't face plant into the concrete. Right.

456
00:21:29,970 --> 00:21:31,910
In the old Cartesian model, we would explain

457
00:21:31,910 --> 00:21:35,069
that process like this. Your eyes visually detect

458
00:21:35,069 --> 00:21:37,470
the change in the slope. The optic nerve sends

459
00:21:37,470 --> 00:21:40,210
that image to the brain. The brain calculates

460
00:21:40,210 --> 00:21:43,190
the new angle of decline, realizes gravity is

461
00:21:43,190 --> 00:21:46,369
increasing, and sends a new, faster firing command

462
00:21:46,369 --> 00:21:48,930
down the spinal cord to the leg muscles. But

463
00:21:48,930 --> 00:21:51,269
that sounds way too slow. I mean, if I had to

464
00:21:51,269 --> 00:21:52,930
wait for that whole email chain to process, I'd

465
00:21:52,930 --> 00:21:55,670
already be eating asphalt. It is entirely too

466
00:21:55,670 --> 00:21:58,569
slow. What actually happens is far more elegant.

467
00:21:59,009 --> 00:22:00,869
When you hit the decline, the physical force

468
00:22:00,869 --> 00:22:03,410
of gravity pulls your torso forward. Gravity

469
00:22:03,410 --> 00:22:06,589
does the work. Yes. This mechanical force physically

470
00:22:06,589 --> 00:22:08,789
accelerates your limbs. They literally swing

471
00:22:08,789 --> 00:22:10,730
faster because gravity is pulling them. OK, so

472
00:22:10,730 --> 00:22:13,190
the body moves faster. And here is the paradigm

473
00:22:13,190 --> 00:22:15,509
-shattering part. The physical speeding up of

474
00:22:15,509 --> 00:22:17,529
your leg muscles generates mechanical tension

475
00:22:17,529 --> 00:22:20,569
that sends signals back up the spinal cord, which

476
00:22:20,569 --> 00:22:23,210
actually entrains and speeds up the neural oscillators

477
00:22:23,210 --> 00:22:25,230
in your central nervous system. Wait, wait, wait.

478
00:22:25,420 --> 00:22:28,440
The physical stretching of the leg muscles forces

479
00:22:28,440 --> 00:22:31,019
the brain's rhythm to speed up to match it. Yes.

480
00:22:31,420 --> 00:22:33,579
The body is controlling the brain. Unbelievable.

481
00:22:34,000 --> 00:22:36,200
It is not a top -down hierarchy where the brain

482
00:22:36,200 --> 00:22:38,539
sits in a throne room issuing orders. It is a

483
00:22:38,539 --> 00:22:42,200
completely flat, mutual coupling. Gravity changes

484
00:22:42,200 --> 00:22:45,779
the leg and the leg changes the brain. That completely

485
00:22:45,779 --> 00:22:49,539
reorganizes how I view myself. I am not a pilot

486
00:22:49,539 --> 00:22:52,480
flying a meat mech. The meat mech is part of

487
00:22:52,480 --> 00:22:55,269
the pilot. It really is. But how does this coupling

488
00:22:55,269 --> 00:22:57,509
actually help us make sense of the world? I mean,

489
00:22:57,809 --> 00:23:00,069
if I'm just a coupled system reacting to gravity

490
00:23:00,069 --> 00:23:02,289
and squishing my fingers, how do I actually learn

491
00:23:02,289 --> 00:23:04,890
to recognize things? You do it through a mechanism

492
00:23:04,890 --> 00:23:08,369
called sensory motor coordination. This is the

493
00:23:08,369 --> 00:23:10,509
realization that we don't just passively receive

494
00:23:10,509 --> 00:23:12,609
information from the world. We actively structure

495
00:23:12,609 --> 00:23:15,430
it through our physical movements. Let's slow

496
00:23:15,430 --> 00:23:17,930
down and unpack that, because structuring information

497
00:23:17,930 --> 00:23:21,289
sounds a bit, you know, academic. How does a

498
00:23:21,289 --> 00:23:24,039
physical movement structure data? Let's look

499
00:23:24,039 --> 00:23:26,380
at an artificial intelligence experiment conducted

500
00:23:26,380 --> 00:23:29,319
by Giorgio Mehta at MIT with a robot named Cog.

501
00:23:30,160 --> 00:23:32,619
Cog was essentially a humanoid torso with cameras

502
00:23:32,619 --> 00:23:36,279
for eyes. Okay, a torso with camera eyes. The

503
00:23:36,279 --> 00:23:38,940
goal was simple. Teach Cog to look at a messy

504
00:23:38,940 --> 00:23:41,460
cluttered desk and identify a single object like

505
00:23:41,460 --> 00:23:44,420
a toy block amidst all the junk. In classical

506
00:23:44,420 --> 00:23:46,759
AI, you just take a photograph of the desk and

507
00:23:46,759 --> 00:23:49,059
feed it into an image recognition algorithm,

508
00:23:49,119 --> 00:23:52,339
right? And classical image recognition struggles

509
00:23:52,339 --> 00:23:55,420
immensely with that. In a static 2D photograph

510
00:23:55,420 --> 00:23:57,960
of a messy desk, the pixels of a toy block blend

511
00:23:57,960 --> 00:24:00,220
seamlessly into the pixels of the notebook behind

512
00:24:00,220 --> 00:24:02,119
it, which blend into the pixels of the stapler.

513
00:24:02,720 --> 00:24:05,599
It's just a mess of pixels. Right. Without prior

514
00:24:05,599 --> 00:24:07,599
knowledge, it's almost impossible for a computer

515
00:24:07,599 --> 00:24:10,140
to know where one object ends and another begins.

516
00:24:10,700 --> 00:24:13,680
So how did Meta solve it with Kog? He had Kog

517
00:24:13,680 --> 00:24:17,160
do what human infants do. He programmed Kog to

518
00:24:17,160 --> 00:24:20,150
physically move its camera head side to side.

519
00:24:20,650 --> 00:24:23,549
How does shaking its head help? Think about the

520
00:24:23,549 --> 00:24:26,009
physics of visual perspective. When you move

521
00:24:26,009 --> 00:24:28,210
your head from side to side, objects that are

522
00:24:28,210 --> 00:24:30,349
close to you appear to move very quickly across

523
00:24:30,349 --> 00:24:32,869
your field of vision. Right. But objects that

524
00:24:32,869 --> 00:24:35,309
are far away, in the background, move very slowly.

525
00:24:35,869 --> 00:24:38,309
Oh, like looking out the window of a train. The

526
00:24:38,309 --> 00:24:40,950
trees by the tracks blur past, but the mountains

527
00:24:40,950 --> 00:24:44,470
in the distance barely move. Exactly. By generating

528
00:24:44,470 --> 00:24:47,500
physical movement, COG forces the raw, chaotic

529
00:24:47,500 --> 00:24:49,960
pixels to separate into a structured pattern.

530
00:24:50,500 --> 00:24:52,500
The pixels belonging to the block move together

531
00:24:52,500 --> 00:24:55,019
at one speed, while the background pixels move

532
00:24:55,019 --> 00:24:58,140
at a different speed. That's brilliant. The robot

533
00:24:58,140 --> 00:25:00,319
hasn't done any complex geometric processing.

534
00:25:00,839 --> 00:25:03,599
It has simply used physical motion to pre -digest

535
00:25:03,599 --> 00:25:06,440
the visual data. The movement creates the information.

536
00:25:06,920 --> 00:25:09,680
So, when I reach out and grab a cup, I'm not

537
00:25:09,680 --> 00:25:12,440
just feeling it. By sliding my physical fingers

538
00:25:12,440 --> 00:25:15,109
over the rim, while simultaneously looking at

539
00:25:15,109 --> 00:25:17,609
my hand and feeling the muscle tension in my

540
00:25:17,609 --> 00:25:20,869
arm, I am syncing up three totally different

541
00:25:20,869 --> 00:25:23,809
data streams. Yes, you are actively generating

542
00:25:23,809 --> 00:25:26,170
a correlated pattern that your brain can easily

543
00:25:26,170 --> 00:25:29,430
digest. The philosopher John Dewey actually foresaw

544
00:25:29,430 --> 00:25:32,190
this all the way back in 1896, didn't he? He

545
00:25:32,190 --> 00:25:35,069
did. He argued that the fundamental unit of behavior

546
00:25:35,069 --> 00:25:37,869
is not a passive sensory stimulus followed by

547
00:25:37,869 --> 00:25:41,269
a motor response. The fundamental unit is a continuous

548
00:25:41,269 --> 00:25:44,200
sensory motor loop. So the physical movement

549
00:25:44,200 --> 00:25:46,420
of the eye determines the quality of the sensation

550
00:25:46,420 --> 00:25:49,819
it receives. You are literally orchestrating

551
00:25:49,819 --> 00:25:51,579
your sensory reality through the movement of

552
00:25:51,579 --> 00:25:54,880
your joints. Exactly. Okay. I am fully on board

553
00:25:54,880 --> 00:25:56,519
with the idea that the body handles the physical

554
00:25:56,519 --> 00:26:00,160
here and now. My soft fingers solve the geometry

555
00:26:00,160 --> 00:26:03,299
of the cup. My legs adjust to the hill. My moving

556
00:26:03,299 --> 00:26:05,259
head separates the foreground from the background.

557
00:26:05,400 --> 00:26:08,400
Right. But here's the massive roadblock my mind

558
00:26:08,400 --> 00:26:10,970
keeps hitting. How do we get from a physical

559
00:26:10,970 --> 00:26:15,230
reflex? to a high -level abstract thought. Ah,

560
00:26:15,410 --> 00:26:18,470
now we are entering the deep waters. Right. Because

561
00:26:18,470 --> 00:26:20,730
I can understand how wiggling my arms helps me

562
00:26:20,730 --> 00:26:22,769
figure out where my body ends and the cup begins.

563
00:26:23,470 --> 00:26:26,970
But how does an infant go from wiggling its arms

564
00:26:26,970 --> 00:26:30,630
to understanding the spoken word cup? It's a

565
00:26:30,630 --> 00:26:33,210
huge leap. How do we get from sensory motor loops

566
00:26:33,210 --> 00:26:36,650
to remembering our mother's face or pondering

567
00:26:36,650 --> 00:26:39,900
the philosophical concept of infinity? How does

568
00:26:39,900 --> 00:26:43,640
meat become mind. You are describing what cognitive

569
00:26:43,640 --> 00:26:45,720
scientists call the symbol grounding problem.

570
00:26:46,240 --> 00:26:48,720
It is perhaps the most profound hurdle in artificial

571
00:26:48,720 --> 00:26:51,740
intelligence and philosophy. Let's define the

572
00:26:51,740 --> 00:26:53,619
symbol grounding problem clearly for everyone

573
00:26:53,619 --> 00:26:55,859
listening. Think about a symbol like the word

574
00:26:55,859 --> 00:26:58,960
apple or the number four. These are abstract

575
00:26:58,960 --> 00:27:01,359
representations. Inside a classical computer

576
00:27:01,359 --> 00:27:03,640
you can program a system to manipulate the word

577
00:27:03,640 --> 00:27:06,660
apple. Like it can link it to the word red or

578
00:27:06,660 --> 00:27:09,759
the word fruit. Right. It can spit out sentences

579
00:27:09,759 --> 00:27:12,579
about apples, but the computer has no idea what

580
00:27:12,579 --> 00:27:15,799
an apple actually is. The symbol is untethered.

581
00:27:16,240 --> 00:27:18,480
It floats in a void. Because the computer has

582
00:27:18,480 --> 00:27:21,099
never bitten into a crisp apple, felt the juice,

583
00:27:21,400 --> 00:27:24,059
or heard the crunch, it has no physical reference.

584
00:27:24,640 --> 00:27:27,119
Precisely. The symbol grounding problem asks,

585
00:27:27,359 --> 00:27:30,720
how do abstract symbols acquire actual felt meaning

586
00:27:30,720 --> 00:27:33,680
inside a continuous, squishy physical system

587
00:27:33,680 --> 00:27:36,480
like a human? How does a physical sensation translate

588
00:27:36,480 --> 00:27:39,339
into a mental category? The embodied cognition

589
00:27:39,339 --> 00:27:43,380
answer is shocking. It all begins with locomotion,

590
00:27:43,960 --> 00:27:46,890
movement. That sounds like an impossible leap.

591
00:27:47,650 --> 00:27:50,710
How does swimming or crawling create concepts?

592
00:27:51,250 --> 00:27:53,269
To bridge that gap, we have to look at an experiment

593
00:27:53,269 --> 00:27:55,390
that proves just how deeply the brain's baseline

594
00:27:55,390 --> 00:27:57,930
rhythm is enslaved to the physical environment.

595
00:27:58,529 --> 00:28:01,450
We need to talk about the lamprey. The lamprey.

596
00:28:01,579 --> 00:28:04,779
That is a fish, right? It is a very primitive,

597
00:28:05,039 --> 00:28:08,059
jawless, eel -like fish. It swims by undulating

598
00:28:08,059 --> 00:28:10,619
its body in an S -shape through the water. A

599
00:28:10,619 --> 00:28:13,119
neuroscientist named Sten Grillner wanted to

600
00:28:13,119 --> 00:28:15,960
study the central pattern generators in the lamprey's

601
00:28:15,960 --> 00:28:18,599
brain and spinal cord. Which are the neural circuits

602
00:28:18,599 --> 00:28:21,039
that dictate that rhythmic swimming motion. Exactly.

603
00:28:22,009 --> 00:28:24,349
So Grillner performed the ultimate Cartesian

604
00:28:24,349 --> 00:28:27,240
experiment. He carefully extracted the entire

605
00:28:27,240 --> 00:28:30,359
intact spinal cord and brain stem from the lamprey

606
00:28:30,359 --> 00:28:32,960
and placed this completely disembodied nervous

607
00:28:32,960 --> 00:28:36,500
system into a petri dish. Filled with some nutrient

608
00:28:36,500 --> 00:28:38,500
solution to keep the neurons alive, I'm guessing?

609
00:28:38,740 --> 00:28:41,660
Right. So he literally put the ghost in a dish,

610
00:28:41,920 --> 00:28:43,980
completely separated from the machine of the

611
00:28:43,980 --> 00:28:46,960
body? Yes. And he attached electrodes to the

612
00:28:46,960 --> 00:28:49,039
spinal cord to measure the neural firing rate.

613
00:28:49,380 --> 00:28:52,440
He wanted to see the pure, unadulterated thought

614
00:28:52,440 --> 00:28:55,259
of swimming. Okay, what did he find? What he

615
00:28:55,259 --> 00:28:58,119
found was incredibly strange. When isolated in

616
00:28:58,119 --> 00:29:00,440
the dish, the neural firing rate, the rhythm

617
00:29:00,440 --> 00:29:03,579
the brain uses to tell the muscles to swim, changed

618
00:29:03,579 --> 00:29:06,779
dramatically. It actually fired significantly

619
00:29:06,779 --> 00:29:09,019
faster than it ever did when it was inside the

620
00:29:09,019 --> 00:29:12,039
living fish. Why would it speed up? I mean, you'd

621
00:29:12,039 --> 00:29:15,059
assume that a brain free from the heavy dragging

622
00:29:15,059 --> 00:29:17,579
weight of a physical body would finally operate

623
00:29:17,579 --> 00:29:20,930
at its true perfect natural frequency. That is

624
00:29:20,930 --> 00:29:24,569
the Cartesian assumption. I fell for it. We all

625
00:29:24,569 --> 00:29:29,190
do. But it is entirely wrong. There is no quote

626
00:29:29,190 --> 00:29:31,349
-unquote true natural frequency for the brain

627
00:29:31,349 --> 00:29:34,730
in isolation. The brain evolved to operate against

628
00:29:34,730 --> 00:29:38,369
the physical world. Against it. In the intact

629
00:29:38,369 --> 00:29:40,970
animal, as the lamprey flexes its muscles to

630
00:29:40,970 --> 00:29:43,309
swim, those muscles have to push against the

631
00:29:43,309 --> 00:29:45,329
physical density and resistance of the water.

632
00:29:45,589 --> 00:29:48,430
Because water is heavy. It pushes back. That

633
00:29:48,430 --> 00:29:50,430
physical resistance of the water pushing back

634
00:29:50,430 --> 00:29:52,509
against the body generates sensory feedback.

635
00:29:52,779 --> 00:29:55,400
That feedback travels from the skin and muscles

636
00:29:55,400 --> 00:29:57,559
back into the spinal cord. And what does it do

637
00:29:57,559 --> 00:30:00,519
there? That physical feedback physically slows

638
00:30:00,519 --> 00:30:03,740
down and stabilizes the neural rhythm. The brain's

639
00:30:03,740 --> 00:30:06,519
functioning relies entirely on the physical feedback

640
00:30:06,519 --> 00:30:08,759
of the body wrestling with the physics of water.

641
00:30:09,099 --> 00:30:11,480
Wow. Without the water, the brain basically spins

642
00:30:11,480 --> 00:30:13,640
out of control like an engine with a snapped

643
00:30:13,640 --> 00:30:16,359
transmission belt. You literally cannot understand

644
00:30:16,359 --> 00:30:18,640
the brain without understanding the water it

645
00:30:18,640 --> 00:30:22,980
swims in. That is deeply unsettling. My brain

646
00:30:22,980 --> 00:30:25,180
only functions correctly because gravity is currently

647
00:30:25,180 --> 00:30:27,779
pushing my posture down into this chair. It's

648
00:30:27,779 --> 00:30:31,769
true. But let's scale this up. We have a brain,

649
00:30:31,890 --> 00:30:34,930
stabilized by physics. How does a human baby

650
00:30:34,930 --> 00:30:37,490
use this to learn concepts? The book explores

651
00:30:37,490 --> 00:30:40,150
this through the lens of early robotics, specifically

652
00:30:40,150 --> 00:30:43,289
a four -legged robot aptly named Puppy. Puppy?

653
00:30:43,569 --> 00:30:46,349
Yeah. Puppy was not a highly intelligent robot.

654
00:30:46,470 --> 00:30:48,789
It was built primarily with springs and simple

655
00:30:48,789 --> 00:30:51,660
motors to study how physical bodies move. But

656
00:30:51,660 --> 00:30:53,799
researchers use the mechanics of how puppy learns

657
00:30:53,799 --> 00:30:56,640
to walk to explain how human infants develop

658
00:30:56,640 --> 00:30:59,559
what we call a body image. A physical map of

659
00:30:59,559 --> 00:31:02,299
themselves? Yes. How does that happen? It happens

660
00:31:02,299 --> 00:31:04,920
through a mechanism called an efference copy.

661
00:31:05,640 --> 00:31:08,980
An efference copy? Imagine a human infant lying

662
00:31:08,980 --> 00:31:11,339
on its back, flailing its arms and legs randomly.

663
00:31:11,640 --> 00:31:14,519
It seems chaotic. But every time the infant's

664
00:31:14,519 --> 00:31:16,859
brain sends a motor command down the spinal cord,

665
00:31:17,140 --> 00:31:20,660
say, swing right arm violently, the brain simultaneously

666
00:31:20,660 --> 00:31:23,180
generates a carbon copy of that command. That's

667
00:31:23,180 --> 00:31:25,480
the efference copy. Right. And it routes it back

668
00:31:25,480 --> 00:31:28,259
into its own sensory processing centers. So it's

669
00:31:28,259 --> 00:31:30,319
like the brain is leaving a paper trail for itself.

670
00:31:30,819 --> 00:31:32,920
It issues a command to the arm, but also sends

671
00:31:32,920 --> 00:31:35,420
a memo to the sensory department saying, hey,

672
00:31:35,619 --> 00:31:37,880
heads up, I just told the arm to swing. Be on

673
00:31:37,880 --> 00:31:40,539
the lookout for what happens next. That is a

674
00:31:40,539 --> 00:31:43,900
brilliant analogy. The memo is filed. Then the

675
00:31:43,900 --> 00:31:46,730
physical arm swings. The baby's eyes see the

676
00:31:46,730 --> 00:31:49,549
blurry hand across its field of vision. The skin

677
00:31:49,549 --> 00:31:52,130
on the arm feels the rush of air. The muscles

678
00:31:52,130 --> 00:31:54,549
and tendons feel the physical stretch. All of

679
00:31:54,549 --> 00:31:57,650
this rich, multimodal sensory data rushes back

680
00:31:57,650 --> 00:32:00,210
up into the brain. And the brain compares the

681
00:32:00,210 --> 00:32:03,369
sensory data to the memo. Yes. It matches the

682
00:32:03,369 --> 00:32:05,269
incoming sensory data with the efference copy.

683
00:32:06,069 --> 00:32:09,329
It forms a hardwired neurological link. It learns,

684
00:32:09,769 --> 00:32:12,630
ah, when I issue this specific motor command,

685
00:32:13,019 --> 00:32:16,380
that specific sensory feeling happens. Through

686
00:32:16,380 --> 00:32:19,079
millions of these loops, the flailing baby is

687
00:32:19,079 --> 00:32:21,000
physically mapping the boundaries of its own

688
00:32:21,000 --> 00:32:24,200
body. It is discovering cause and effect through

689
00:32:24,200 --> 00:32:27,400
sheer physical collision. And because the baby's

690
00:32:27,400 --> 00:32:29,940
body is constrained by physical laws, you know,

691
00:32:30,059 --> 00:32:31,819
joints only bend certain ways, gravity always

692
00:32:31,819 --> 00:32:34,359
pulls down, these movements aren't truly random.

693
00:32:34,480 --> 00:32:37,029
Right, there are limits. The physical constraints

694
00:32:37,029 --> 00:32:39,430
cause the movements to naturally fall into repeated

695
00:32:39,430 --> 00:32:42,490
stable patterns. In dynamic systems theory, these

696
00:32:42,490 --> 00:32:44,390
stable patterns are called attractor states.

697
00:32:45,130 --> 00:32:47,569
Let me see if I can translate dynamic systems

698
00:32:47,569 --> 00:32:49,450
theory into something I can picture. Give it

699
00:32:49,450 --> 00:32:52,089
a shot. An attractor state is like if you take

700
00:32:52,089 --> 00:32:55,250
a large, smooth bowl and drop a marble into it.

701
00:32:55,670 --> 00:32:57,210
It doesn't matter where you drop the marble on

702
00:32:57,210 --> 00:32:59,029
the edge. It's going to roll around chaotically

703
00:32:59,029 --> 00:33:01,910
for a bit, but the physical slope of the bowl

704
00:33:01,910 --> 00:33:04,349
guarantees it will eventually settle into the

705
00:33:04,349 --> 00:33:06,670
exact same spot. at the very bottom. The bottom

706
00:33:06,670 --> 00:33:09,269
of the bowl is the attractor state. Yes. That

707
00:33:09,269 --> 00:33:11,869
is exactly what an attractor state is. The anatomy

708
00:33:11,869 --> 00:33:14,289
of the baby's body acts like the slopes of the

709
00:33:14,289 --> 00:33:17,109
bowl. No matter how the baby flails, the physical

710
00:33:17,109 --> 00:33:19,410
constraints eventually guide the neural firing

711
00:33:19,410 --> 00:33:22,670
into stable, repeatable patterns. And once the

712
00:33:22,670 --> 00:33:25,430
brain maps these attractor states, it has formed

713
00:33:25,430 --> 00:33:28,849
physical categories. Yes. It categorizes me versus

714
00:33:28,849 --> 00:33:32,109
not me. It categorizes hard floor versus soft

715
00:33:32,109 --> 00:33:35,460
blanket. Okay, I see how we get physical categories,

716
00:33:36,059 --> 00:33:38,880
but we are still a long way from poetry in calculus.

717
00:33:39,799 --> 00:33:43,480
How do we jump from hard floor to complex abstract

718
00:33:43,480 --> 00:33:46,000
thought? This is where we encounter the Lakoff

719
00:33:46,000 --> 00:33:48,680
Nunez hypothesis, which is arguably the most

720
00:33:48,680 --> 00:33:51,660
mind -bending concept in the entire book. Okay,

721
00:33:51,980 --> 00:33:54,730
lay it on me. George Lakoff and Rafael Núñez

722
00:33:54,730 --> 00:33:57,670
argue that there is no magical jump. They propose

723
00:33:57,670 --> 00:34:00,029
that even the most abstract, high -level human

724
00:34:00,029 --> 00:34:02,950
concepts, including complex mathematics, are

725
00:34:02,950 --> 00:34:05,869
actually just elaborate metaphors built directly

726
00:34:05,869 --> 00:34:08,949
on top of our physical, bodily experiences. Wait,

727
00:34:09,110 --> 00:34:12,150
math is a physical metaphor. How? Let's start

728
00:34:12,150 --> 00:34:14,730
with language and emotion. Think about how we

729
00:34:14,730 --> 00:34:17,210
describe human affection. We say, give them my

730
00:34:17,210 --> 00:34:19,809
warm regards, or she has a warm personality.

731
00:34:20,030 --> 00:34:22,780
Right. If someone is cruel, we say, he gave me

732
00:34:22,780 --> 00:34:26,280
the cold shoulder or she is icy. Why do we use

733
00:34:26,280 --> 00:34:28,699
temperature to describe a purely emotional state?

734
00:34:29,400 --> 00:34:31,300
Because, I mean, it's just a figure of speech.

735
00:34:32,170 --> 00:34:35,969
It is not just a figure of speech. It is an embodied

736
00:34:35,969 --> 00:34:39,610
metaphor. As vulnerable human infants, our very

737
00:34:39,610 --> 00:34:43,210
first primal experience of safety, survival and

738
00:34:43,210 --> 00:34:46,130
affection was the literal physical body heat

739
00:34:46,130 --> 00:34:48,349
of our caregivers holding us against their skin.

740
00:34:48,550 --> 00:34:51,170
Wow. The physical sensation of temperature became

741
00:34:51,170 --> 00:34:54,030
neurologically fused with the feeling of security.

742
00:34:54,789 --> 00:34:56,969
Years later, when we want to express the abstract

743
00:34:56,969 --> 00:35:00,070
concept of affection, we reach for the foundational

744
00:35:00,070 --> 00:35:02,269
physical attractor state that built it, which

745
00:35:02,269 --> 00:35:05,309
is warmth. The abstract thought is anchored in

746
00:35:05,309 --> 00:35:09,360
the flesh. Oh my god. That makes so much sense.

747
00:35:10,309 --> 00:35:12,969
But how does that apply to math? Math is supposed

748
00:35:12,969 --> 00:35:15,289
to be the universal, pure logic of the cosmos.

749
00:35:16,179 --> 00:35:19,000
But our understanding of it is strictly human.

750
00:35:19,340 --> 00:35:22,099
Think about addition. How do we first understand

751
00:35:22,099 --> 00:35:25,500
what adding means? We physically manipulate the

752
00:35:25,500 --> 00:35:28,539
world. We use our hands. A child sits on the

753
00:35:28,539 --> 00:35:30,860
floor and physically takes a wooden block, moves

754
00:35:30,860 --> 00:35:33,320
their arm, and places it next to another block.

755
00:35:33,780 --> 00:35:36,760
They see the pile grow physically larger. The

756
00:35:36,760 --> 00:35:39,539
abstract concept of addition is a metaphor derived

757
00:35:39,539 --> 00:35:42,519
from the physical action of object collection.

758
00:35:42,960 --> 00:35:46,280
What about something ungraspable like the concept

759
00:35:46,280 --> 00:35:48,980
of infinity? I can't put infinity in a pile.

760
00:35:49,329 --> 00:35:53,440
But you can walk. Walk? The concept of infinity

761
00:35:53,440 --> 00:35:56,239
is grounded in the physical experience of continuous

762
00:35:56,239 --> 00:35:58,820
locomotion. You have the bodily experience of

763
00:35:58,820 --> 00:36:01,159
taking a step forward on a flat plane. You know

764
00:36:01,159 --> 00:36:04,119
what it feels like to physically move. Okay.

765
00:36:04,480 --> 00:36:07,039
You can then project that physical feeling conceptually

766
00:36:07,039 --> 00:36:09,519
and imagine taking another step and another step

767
00:36:09,519 --> 00:36:12,239
without ever stopping. The math is just a highly

768
00:36:12,239 --> 00:36:14,639
refined metaphor of the body's movement through

769
00:36:14,639 --> 00:36:17,420
space. We can only conceptualize what our bodies

770
00:36:17,420 --> 00:36:19,929
give us the tools to understand. I have to push

771
00:36:19,929 --> 00:36:22,650
back on this, because the implications are, frankly,

772
00:36:22,969 --> 00:36:26,210
terrifying for modern technology. How so? If

773
00:36:26,210 --> 00:36:28,670
every single abstract thought we have, from the

774
00:36:28,670 --> 00:36:30,530
feeling of heartbreak to the logic of addition,

775
00:36:30,969 --> 00:36:34,289
is inextricably grounded in the specific, squishy,

776
00:36:34,570 --> 00:36:37,110
gravity -bound experience of a human body, what

777
00:36:37,110 --> 00:36:39,510
does that mean for artificial intelligence? Uh,

778
00:36:39,510 --> 00:36:42,309
yes. We have these massive, large language models

779
00:36:42,309 --> 00:36:45,409
sitting in server farms right now, writing poetry

780
00:36:45,409 --> 00:36:47,969
and passing the bar exam. But if they don't have

781
00:36:47,969 --> 00:36:50,289
a body, Do they actually understand anything

782
00:36:50,289 --> 00:36:54,070
they are saying? Can a disembodied AI ever truly

783
00:36:54,070 --> 00:36:57,329
understand human language? According to the framework

784
00:36:57,329 --> 00:36:59,650
of embodied cognition, the answer is a definitive

785
00:36:59,650 --> 00:37:03,789
no. Really? A disembodied AI, no matter how vast

786
00:37:03,789 --> 00:37:06,429
its neural network, is fundamentally just a master

787
00:37:06,429 --> 00:37:09,030
of statistical correlation. It can analyze a

788
00:37:09,030 --> 00:37:11,449
trillion documents and learn that the word heartbreak

789
00:37:11,449 --> 00:37:14,050
often appears near the words tears, chest, and

790
00:37:14,050 --> 00:37:16,570
pain. They can string those words together beautifully.

791
00:37:16,829 --> 00:37:19,329
But it doesn't feel the physical tightening of

792
00:37:19,329 --> 00:37:21,329
the chest. It doesn't know the physical weight

793
00:37:21,329 --> 00:37:24,010
of gravity when you are depressed. It's like

794
00:37:24,010 --> 00:37:26,429
a person who has been blind since birth, perfectly

795
00:37:26,429 --> 00:37:28,409
memorizing the dictionary definition of the color

796
00:37:28,409 --> 00:37:32,190
red, but having absolutely no internal experience

797
00:37:32,190 --> 00:37:36,320
of redness. Precisely. Without a physical body

798
00:37:36,320 --> 00:37:38,199
moving through a physical world to ground those

799
00:37:38,199 --> 00:37:40,840
symbols and actual consequences and sensations,

800
00:37:41,400 --> 00:37:44,880
it is just a highly complex parlor trick. The

801
00:37:44,880 --> 00:37:47,159
AI lacks the fundamental frame of reference.

802
00:37:47,639 --> 00:37:50,739
The symbols are ungrounded. That completely changes

803
00:37:50,739 --> 00:37:52,820
how we should view our interactions with chatbots.

804
00:37:53,500 --> 00:37:56,039
They aren't minds, they're just incredibly convincing

805
00:37:56,039 --> 00:37:58,619
mirrors reflecting our own embodied language

806
00:37:58,619 --> 00:38:02,119
back at us. Exactly right. Okay, let's zoom out

807
00:38:02,119 --> 00:38:04,840
for a second. We've seen how development happens

808
00:38:04,840 --> 00:38:07,320
over a single lifetime, from a baby flailing

809
00:38:07,320 --> 00:38:10,280
its arms, to building attractor states, to inventing

810
00:38:10,280 --> 00:38:13,000
calculus. But this intricate coupling between

811
00:38:13,000 --> 00:38:15,280
the shape of the body and the wiring of the brain

812
00:38:15,280 --> 00:38:18,900
had to be designed somehow. Right. How does nature

813
00:38:18,900 --> 00:38:21,639
forge these systems over millions of years? To

814
00:38:21,639 --> 00:38:23,860
answer that, we have to look at the third time

815
00:38:23,860 --> 00:38:26,960
frame, which is evolution. And what's fascinating

816
00:38:26,960 --> 00:38:29,619
is that computer scientists are no longer just

817
00:38:29,619 --> 00:38:32,780
studying biological evolution. They are actively

818
00:38:32,780 --> 00:38:35,360
using it as an engineering tool. They use something

819
00:38:35,360 --> 00:38:38,380
called artificial evolution to design robots.

820
00:38:38,719 --> 00:38:40,840
Instead of a human engineer sitting in a CAD

821
00:38:40,840 --> 00:38:43,880
program designing a robot joint by joint, they

822
00:38:43,880 --> 00:38:46,920
let a Darwinian algorithm do the work? Exactly.

823
00:38:47,760 --> 00:38:49,860
Because human engineers are biased. We always

824
00:38:49,860 --> 00:38:52,800
try to build top -down Cartesian machines. We

825
00:38:52,800 --> 00:38:55,639
force robots to look like us. But evolution has

826
00:38:55,639 --> 00:38:58,179
no ego. How does that actually work in practice?

827
00:38:58,719 --> 00:39:01,320
Well, the book highlights an incredible experiment

828
00:39:01,320 --> 00:39:04,170
by Josh Bongard, one of the authors. Bongard

829
00:39:04,170 --> 00:39:06,750
created a physics simulator in a computer, a

830
00:39:06,750 --> 00:39:09,510
virtual world with gravity and friction. What

831
00:39:09,510 --> 00:39:12,030
did he put in this virtual world? He generated

832
00:39:12,030 --> 00:39:14,769
simple virtual creatures. They looked like random

833
00:39:14,769 --> 00:39:17,010
collections of geometric blocks connected by

834
00:39:17,010 --> 00:39:20,329
basic hinge joints. He gave each creature a tiny

835
00:39:20,329 --> 00:39:22,750
randomized artificial neural network to control

836
00:39:22,750 --> 00:39:26,050
its hinges. OK. And he gave the algorithm one

837
00:39:26,050 --> 00:39:29,050
single simple metric for success. Push a large

838
00:39:29,050 --> 00:39:31,769
target block as far forward as possible in 10

839
00:39:31,769 --> 00:39:34,320
seconds. If they were totally random, Generation

840
00:39:34,320 --> 00:39:37,300
1 must have been a spectacular failure. Oh, it

841
00:39:37,300 --> 00:39:40,500
was a disaster. Thousands of virtual creatures

842
00:39:40,500 --> 00:39:43,079
just flailed wildly. Most of them immediately

843
00:39:43,079 --> 00:39:46,079
fell over and twitched on the floor. Survival

844
00:39:46,079 --> 00:39:49,320
of the fittest twitchers. But purely by statistical

845
00:39:49,320 --> 00:39:52,460
chance, a few of those flailing creatures accidentally

846
00:39:52,460 --> 00:39:54,659
twitched in a way that bumped the target block

847
00:39:54,659 --> 00:39:57,639
a fraction of a millimeter. And the evolutionary

848
00:39:57,639 --> 00:40:00,460
algorithm observed this. Right. It selected those

849
00:40:00,460 --> 00:40:02,960
mildly successful creatures, essentially killed

850
00:40:02,960 --> 00:40:05,440
off the rest, and bred the winners to create

851
00:40:05,440 --> 00:40:08,340
the next generation, injecting slight random

852
00:40:08,340 --> 00:40:11,260
mutations into their digital DNA. And it ran

853
00:40:11,260 --> 00:40:14,860
this cycle over and over. Hundreds and hundreds

854
00:40:14,860 --> 00:40:18,159
of generations. But here is the critical breakthrough

855
00:40:18,159 --> 00:40:21,019
Bongard achieved. The algorithm wasn't just allowed

856
00:40:21,019 --> 00:40:23,699
to mutate the neural network, the brains of the

857
00:40:23,699 --> 00:40:26,619
creatures. It was simultaneously allowed to mutate

858
00:40:26,619 --> 00:40:29,090
their morphology. It could change their physical

859
00:40:29,090 --> 00:40:32,130
bodies. Yes. It could change the physical length

860
00:40:32,130 --> 00:40:34,610
of their limbs, the size of their bodies, the

861
00:40:34,610 --> 00:40:37,469
angle of their joints. The brain and the body

862
00:40:37,469 --> 00:40:40,460
co -evolved. What did the final creatures look

863
00:40:40,460 --> 00:40:43,480
like? The results were stunning. Over generations,

864
00:40:43,739 --> 00:40:46,440
the evolutionary algorithm discovered that to

865
00:40:46,440 --> 00:40:49,099
push a heavy block, you need physical mass. So

866
00:40:49,099 --> 00:40:51,480
it evolved creatures with much larger, heavier

867
00:40:51,480 --> 00:40:54,539
bodies. It evolved specific limb structures that

868
00:40:54,539 --> 00:40:56,679
dragged along the ground to maximize friction

869
00:40:56,679 --> 00:40:59,780
and leverage. It designed an entire morphological

870
00:40:59,780 --> 00:41:01,980
system that a human engineer never would have

871
00:41:01,980 --> 00:41:04,099
dreamed up. It proved mathematically that changing

872
00:41:04,099 --> 00:41:06,239
the physical shape of the body is often a more

873
00:41:06,239 --> 00:41:07,940
efficient solution than adding more computing

874
00:41:07,940 --> 00:41:10,300
power to the brain. The body solves the problem

875
00:41:10,300 --> 00:41:12,260
before the brain even has to think about it.

876
00:41:12,639 --> 00:41:14,940
And evolution doesn't just work on isolated individuals.

877
00:41:15,480 --> 00:41:18,440
It works on massive populations. which leads

878
00:41:18,440 --> 00:41:20,320
to some of the most bizarre and beautiful examples

879
00:41:20,320 --> 00:41:24,239
of embodied intelligence in nature, swarm intelligence.

880
00:41:24,739 --> 00:41:28,440
Yes, swarm intelligence. When we look at an ant

881
00:41:28,440 --> 00:41:31,059
colony, we are forced to completely abandon the

882
00:41:31,059 --> 00:41:33,179
idea that intelligence is contained within a

883
00:41:33,179 --> 00:41:35,960
single brain. Because an individual ant is incredibly

884
00:41:35,960 --> 00:41:38,039
simple, as we discussed with the ant on the beach.

885
00:41:38,360 --> 00:41:42,059
But a colony of a million ants... behaves like

886
00:41:42,059 --> 00:41:45,280
a highly intelligent superorganism. An ant colony

887
00:41:45,280 --> 00:41:48,000
can systematically explore a forest, find the

888
00:41:48,000 --> 00:41:50,800
absolute shortest most mathematically efficient

889
00:41:50,800 --> 00:41:54,159
path to a food source, dynamically reroute around

890
00:41:54,159 --> 00:41:56,780
obstacles, and build architecturally complex

891
00:41:56,780 --> 00:41:59,960
nests with precise temperature control. How is

892
00:41:59,960 --> 00:42:01,940
that possible if the individual ant is basically

893
00:42:01,940 --> 00:42:04,719
a Roomba bumping into walls? They achieve this

894
00:42:04,719 --> 00:42:07,769
through a process called stigmurgy. Stigmurgy

895
00:42:07,769 --> 00:42:09,809
is the concept of communicating and coordinating

896
00:42:09,809 --> 00:42:12,210
entirely through the modification of the physical

897
00:42:12,210 --> 00:42:15,489
environment. So the ants aren't having tiny committee

898
00:42:15,489 --> 00:42:17,530
meetings or talking to each other? They aren't

899
00:42:17,530 --> 00:42:19,989
directly communicating at all. It works like

900
00:42:19,989 --> 00:42:22,630
this. An ant wanders randomly until it finds

901
00:42:22,630 --> 00:42:25,610
a dead caterpillar. It grabs a piece and walks

902
00:42:25,610 --> 00:42:28,550
back to the nest. As it walks, it secretes a

903
00:42:28,550 --> 00:42:30,730
chemical pheromone from its abdomen, leaving

904
00:42:30,730 --> 00:42:33,289
a physical trail on the dirt. Like Hansel and

905
00:42:33,289 --> 00:42:36,489
Gretel dropping breadcrumbs. Exactly. Now other

906
00:42:36,489 --> 00:42:39,050
ants are wandering blindly. If they happen to

907
00:42:39,050 --> 00:42:41,469
cross this chemical trail, their simple physical

908
00:42:41,469 --> 00:42:44,949
programming kicks in. Follow the smell. As they

909
00:42:44,949 --> 00:42:47,489
follow the trail to the food and back, they also

910
00:42:47,489 --> 00:42:50,309
secrete pheromones. So every ant that uses the

911
00:42:50,309 --> 00:42:53,090
path makes the path stronger and more attractive

912
00:42:53,090 --> 00:42:55,369
to the next ant? It's a positive feedback loop.

913
00:42:55,670 --> 00:42:57,610
But here is where the physical environment does

914
00:42:57,610 --> 00:43:01,400
the actual computation. Okay, how? Let's say

915
00:43:01,400 --> 00:43:04,579
there are two paths to the food, one long winding

916
00:43:04,579 --> 00:43:07,800
path and one short straight path. The pheromone

917
00:43:07,800 --> 00:43:11,079
chemicals naturally evaporate over time. Ants

918
00:43:11,079 --> 00:43:13,420
taking the shorter path make the round trip much

919
00:43:13,420 --> 00:43:16,519
faster. Because it's shorter. Because they are

920
00:43:16,519 --> 00:43:18,679
traveling faster, they lay down pheromones at

921
00:43:18,679 --> 00:43:21,000
a higher rate than the evaporation can erase

922
00:43:21,000 --> 00:43:24,300
them. The shorter path quickly becomes a blazing

923
00:43:24,300 --> 00:43:27,099
chemical highway, while the longer path slowly

924
00:43:27,099 --> 00:43:30,219
evaporates into nothing. The ants didn't calculate

925
00:43:30,219 --> 00:43:33,099
geometry. The physical evaporation rate of the

926
00:43:33,099 --> 00:43:35,860
chemical in the dirt did the math for them. Yes.

927
00:43:36,659 --> 00:43:38,820
The intelligence of the colony is not in the

928
00:43:38,820 --> 00:43:41,019
ants. The intelligence is literally stored in

929
00:43:41,019 --> 00:43:44,059
the dirt. The physical environment is the hard

930
00:43:44,059 --> 00:43:46,420
drive, the processor, and the communication network

931
00:43:46,420 --> 00:43:49,500
all at once. That makes me look at technology

932
00:43:49,500 --> 00:43:52,610
in a completely different way. And engineers

933
00:43:52,610 --> 00:43:54,309
are actually trying to build robots that work

934
00:43:54,309 --> 00:43:57,469
like this, right? The modular robots? Yes, systems

935
00:43:57,469 --> 00:44:00,469
like the M -Tran robots developed in Japan. Instead

936
00:44:00,469 --> 00:44:03,050
of building one massive, complex humanoid robot

937
00:44:03,050 --> 00:44:05,730
with a centralized brain, they build dozens of

938
00:44:05,730 --> 00:44:09,369
identical, relatively simple robotic cubes. So

939
00:44:09,369 --> 00:44:12,590
a single block can't really do much. But they

940
00:44:12,590 --> 00:44:15,550
are programmed to interact. If a swarm of these

941
00:44:15,550 --> 00:44:17,789
blocks needs to move through a narrow pipe, they

942
00:44:17,789 --> 00:44:20,030
latch onto each other end to end to form a snake,

943
00:44:20,429 --> 00:44:22,389
and their movements ripple down the chain. That's

944
00:44:22,389 --> 00:44:24,639
creepy but cool. If that snake drops out of the

945
00:44:24,639 --> 00:44:26,679
pipe onto a flat floor and needs to move quickly,

946
00:44:26,820 --> 00:44:28,820
the blocks detach, reconfigure their connections,

947
00:44:29,260 --> 00:44:31,639
and physically transform into a four -legged

948
00:44:31,639 --> 00:44:33,739
walking machine. So what does this all mean for

949
00:44:33,739 --> 00:44:36,400
how we build the future? It feels like we have

950
00:44:36,400 --> 00:44:38,500
to completely let go of our desire for control.

951
00:44:38,780 --> 00:44:41,199
We absolutely do. It's like a master sculptor

952
00:44:41,199 --> 00:44:43,760
who decides to throw away his chisel. Instead

953
00:44:43,760 --> 00:44:46,380
of carving the stone by force, he just sets the

954
00:44:46,380 --> 00:44:48,940
stone outside. studies the fluid dynamics of

955
00:44:48,940 --> 00:44:51,260
the wind and the rain and lets the physical elements

956
00:44:51,260 --> 00:44:53,559
carve the perfect aerodynamic shape over decades.

957
00:44:54,159 --> 00:44:56,679
You set the physical parameters and you let the

958
00:44:56,679 --> 00:45:00,000
intelligence naturally emerge. It is a profoundly

959
00:45:00,000 --> 00:45:02,780
humbling way to design. You stop playing a top

960
00:45:02,780 --> 00:45:05,400
-down god and you start partnering with physics.

961
00:45:06,159 --> 00:45:08,099
Okay, we are coming into the final stretch of

962
00:45:08,099 --> 00:45:11,320
our deep dive here. We have shattered the illusion

963
00:45:11,320 --> 00:45:14,159
of the brain as a CEO. We've looked at robots

964
00:45:14,159 --> 00:45:16,539
and ants. Now I want to bring this all the way

965
00:45:16,539 --> 00:45:19,079
home to the most intimate, deeply personal part

966
00:45:19,079 --> 00:45:21,789
of the human experience. I want to talk about

967
00:45:21,789 --> 00:45:24,809
human memory. If there is one concept that the

968
00:45:24,809 --> 00:45:26,949
Cartesian view of the brain has completely dominated,

969
00:45:27,489 --> 00:45:30,309
it is our understanding of memory. We universally

970
00:45:30,309 --> 00:45:33,269
rely on what psychologists call the storehouse

971
00:45:33,269 --> 00:45:36,130
metaphor. Which is the idea that my brain is

972
00:45:36,130 --> 00:45:38,530
basically a giant filing cabinet or a computer

973
00:45:38,530 --> 00:45:40,980
hard drive. Right. Like, I experience my fifth

974
00:45:40,980 --> 00:45:43,500
birthday party, my brain encodes that experience

975
00:45:43,500 --> 00:45:46,219
into a static file, puts it in a manila folder,

976
00:45:46,679 --> 00:45:49,260
and stores it in the cabinet. Decades later,

977
00:45:49,579 --> 00:45:51,239
when I want to remember it, I just go to the

978
00:45:51,239 --> 00:45:53,039
cabinet and pull the exact same file out to look

979
00:45:53,039 --> 00:45:56,280
at it. We treat memories as static objects, things

980
00:45:56,280 --> 00:45:59,159
we possess. But the theory of embodied cognition,

981
00:45:59,579 --> 00:46:01,460
particularly through the work of psychologist

982
00:46:01,460 --> 00:46:04,619
Arthur Glennberg, completely destroys the storehouse

983
00:46:04,619 --> 00:46:08,690
metaphor. How so? Glennberg proposes the theory

984
00:46:08,690 --> 00:46:11,449
of ecological memory. He argues that we have

985
00:46:11,449 --> 00:46:14,030
to remember why memory evolved in the first place.

986
00:46:14,690 --> 00:46:16,710
Memory did not evolve so human beings could sit

987
00:46:16,710 --> 00:46:19,230
at a desk and memorize a list of vocabulary words

988
00:46:19,230 --> 00:46:21,969
for a standardized test. Memory evolved so we

989
00:46:21,969 --> 00:46:24,230
wouldn't get eaten by lions, right? It evolved

990
00:46:24,230 --> 00:46:27,650
to help us survive. Exactly. Memory evolved strictly

991
00:46:27,650 --> 00:46:30,670
in the service of physical action. Memory is

992
00:46:30,670 --> 00:46:32,710
for doing things in the physical environment.

993
00:46:33,190 --> 00:46:35,469
Therefore, memory is inextricably linked to your

994
00:46:35,469 --> 00:46:37,889
physical posture, your environment, and your

995
00:46:37,889 --> 00:46:40,389
immediate sensory state. Because memory is not

996
00:46:40,389 --> 00:46:43,889
a file you retrieve, it is a state you physically

997
00:46:43,889 --> 00:46:47,829
reconstruct. Yes. Wait, I reconstructed? I don't

998
00:46:47,829 --> 00:46:50,250
just watch a playback of the past? No. Because

999
00:46:50,250 --> 00:46:52,789
your brain is a highly plastic dynamic system,

1000
00:46:53,230 --> 00:46:55,250
every single time you recall a memory, you are

1001
00:46:55,250 --> 00:46:58,150
not pulling a dusty file from a drawer. You are

1002
00:46:58,150 --> 00:47:00,309
actively rebuilding that neural pattern from

1003
00:47:00,309 --> 00:47:02,889
scratch right in the present moment using your

1004
00:47:02,889 --> 00:47:05,349
current neural pathways. Wow. And because you

1005
00:47:05,349 --> 00:47:07,690
are rebuilding it in the present, your current

1006
00:47:07,690 --> 00:47:10,090
physical environment bleeds into the reconstruction.

1007
00:47:10,829 --> 00:47:13,429
So the very act of remembering something physically

1008
00:47:13,429 --> 00:47:15,989
alters the memory network itself. This is called

1009
00:47:15,989 --> 00:47:19,429
recategorization. You never actually remember

1010
00:47:19,429 --> 00:47:21,889
the exact same event the exact same way twice.

1011
00:47:22,210 --> 00:47:24,329
The first time you remembered it, the act of

1012
00:47:24,329 --> 00:47:28,190
remembering change the wiring. That's wild. Your

1013
00:47:28,190 --> 00:47:30,369
memories are dynamic, living things, and they

1014
00:47:30,369 --> 00:47:32,710
are heavily reliant on the external physical

1015
00:47:32,710 --> 00:47:34,690
environment to hold them together. Oh my gosh,

1016
00:47:34,750 --> 00:47:37,030
wait. Have you ever been sitting in your living

1017
00:47:37,030 --> 00:47:39,929
room, realized you wanted a glass of water, stood

1018
00:47:39,929 --> 00:47:42,789
up, walked into the kitchen, and the absolute

1019
00:47:42,789 --> 00:47:44,989
second you crossed the threshold into the kitchen,

1020
00:47:44,989 --> 00:47:47,789
your mind went completely blank. You stand there

1021
00:47:47,789 --> 00:47:49,989
staring at the fridge, having no idea why you

1022
00:47:49,989 --> 00:47:52,469
walked in there. A universal human experience.

1023
00:47:52,969 --> 00:47:55,349
We've all been there. But then you give up. You

1024
00:47:55,349 --> 00:47:57,369
turn around, walk back into the living room,

1025
00:47:57,530 --> 00:47:59,449
sit down in the exact same chair, assume the

1026
00:47:59,449 --> 00:48:01,989
exact same posture, and instantly you remember

1027
00:48:01,989 --> 00:48:04,969
water. Instantly. I always thought I was just

1028
00:48:04,969 --> 00:48:07,690
losing my mind. But under the theory of ecological

1029
00:48:07,690 --> 00:48:10,670
memory, my living room is actually acting as

1030
00:48:10,670 --> 00:48:13,110
my hard drive. That is exactly what is happening.

1031
00:48:13,590 --> 00:48:15,730
The intention to get water was not just stored

1032
00:48:15,730 --> 00:48:18,630
in the isolated vault of your skull. The memory

1033
00:48:18,630 --> 00:48:21,590
was distributed across the entire sensory motor

1034
00:48:21,590 --> 00:48:23,769
state of that moment. It was tied to the chair.

1035
00:48:23,900 --> 00:48:26,159
It was tied to the physical angle of your hips

1036
00:48:26,159 --> 00:48:28,820
in the chair. It was tied to the visual cue of

1037
00:48:28,820 --> 00:48:31,320
the television screen in front of you. The external

1038
00:48:31,320 --> 00:48:33,719
environment provided the physical scaffolding

1039
00:48:33,719 --> 00:48:36,119
that held the memory pattern stable. And when

1040
00:48:36,119 --> 00:48:38,500
I walked into the kitchen, I completely changed

1041
00:48:38,500 --> 00:48:41,039
the physical environment. The visual cues vanished.

1042
00:48:41,360 --> 00:48:43,980
The physical posture changed. I knocked the scaffolding

1043
00:48:43,980 --> 00:48:45,840
out from under the memory, and it collapsed.

1044
00:48:46,539 --> 00:48:48,519
But when you returned to the chair, the physical

1045
00:48:48,519 --> 00:48:50,800
environment provided the structure again, and

1046
00:48:50,800 --> 00:48:53,079
the brain effortlessly reconstructed the intention.

1047
00:48:53,460 --> 00:48:56,239
Your mind is literally leaking out into the room

1048
00:48:56,239 --> 00:49:00,019
around you. That is profoundly beautiful. But

1049
00:49:00,019 --> 00:49:03,539
it also blurs the line between where I end and

1050
00:49:03,539 --> 00:49:06,099
the world begins. And scientists are taking that

1051
00:49:06,099 --> 00:49:08,199
blurred line to its absolute extreme in medicine

1052
00:49:08,199 --> 00:49:11,239
right now, aren't they? Creating literal cyborgs.

1053
00:49:11,619 --> 00:49:14,340
Yes, they are. If we accept that a biological

1054
00:49:14,340 --> 00:49:16,820
brain is desperately seeking a physical body

1055
00:49:16,820 --> 00:49:18,920
to interact with the world, what happens if we

1056
00:49:18,920 --> 00:49:22,170
swap the bodies? Okay. Tell me about this. Researchers

1057
00:49:22,170 --> 00:49:25,010
like Ferdinando Musa -Ivaldi and Steve Potter

1058
00:49:25,010 --> 00:49:27,349
are doing groundbreaking work at the intersection

1059
00:49:27,349 --> 00:49:30,170
of living biology and machines. They have conducted

1060
00:49:30,170 --> 00:49:32,309
experiments where they take living neural tissue,

1061
00:49:32,750 --> 00:49:35,070
actual biological neurons, and culture them in

1062
00:49:35,070 --> 00:49:37,610
a dish embedded with a grid of microelectrodes.

1063
00:49:37,769 --> 00:49:39,869
Kind of like the lamprey experiment, but with

1064
00:49:39,869 --> 00:49:42,570
raw neurons. Yes. But then they connect those

1065
00:49:42,570 --> 00:49:44,769
electrodes to a physical robot with wheels and

1066
00:49:44,769 --> 00:49:47,460
sensors in another room. What? The spontaneous

1067
00:49:47,460 --> 00:49:49,940
electrical firing of the biological neurons is

1068
00:49:49,940 --> 00:49:52,199
translated into motor commands that make the

1069
00:49:52,199 --> 00:49:55,079
robot drive around. They gave a cluster of brain

1070
00:49:55,079 --> 00:49:58,639
cells a robotic meat suit. They did. But the

1071
00:49:58,639 --> 00:50:01,400
incredible part is the sensory motor loop. When

1072
00:50:01,400 --> 00:50:04,480
the robot sensors detect an obstacle, that data

1073
00:50:04,480 --> 00:50:07,179
is translated into electrical pulses and zapped

1074
00:50:07,179 --> 00:50:09,840
back into the living neurons in the dish. So

1075
00:50:09,840 --> 00:50:13,949
the neurons feel what the robot feels. What do

1076
00:50:13,949 --> 00:50:16,909
the cells do? The living neural network physically

1077
00:50:16,909 --> 00:50:20,269
restructures itself. The synapses grow and change

1078
00:50:20,269 --> 00:50:22,570
connections to adapt to the specific physical

1079
00:50:22,570 --> 00:50:24,809
shape and sensors of their new robotic body.

1080
00:50:25,170 --> 00:50:27,630
The biology eagerly molds itself to the mechanics.

1081
00:50:28,170 --> 00:50:31,090
This proves that the brain is profoundly plastic.

1082
00:50:31,530 --> 00:50:34,429
It is an engine waiting for a morphology to drive.

1083
00:50:34,570 --> 00:50:36,590
The implications for human prosthetics are staggering.

1084
00:50:37,630 --> 00:50:40,010
When we attach a robotic arm to an amputee, we

1085
00:50:40,010 --> 00:50:42,230
aren't just strapping a tool to their body. No,

1086
00:50:42,230 --> 00:50:44,610
we are doing much more. If they can receive sensory

1087
00:50:44,610 --> 00:50:47,250
feedback from the artificial fingertips, their

1088
00:50:47,250 --> 00:50:49,949
biological brain will physically rewire itself

1089
00:50:49,949 --> 00:50:53,079
to encompass the metal and plastic. The machine

1090
00:50:53,079 --> 00:50:55,239
literally becomes part of their cognitive self.

1091
00:50:55,679 --> 00:50:58,380
We are extending the embodied mind. But this

1092
00:50:58,380 --> 00:51:00,260
incredibly tight link between our physical bodies

1093
00:51:00,260 --> 00:51:02,639
and our minds also makes us highly vulnerable

1094
00:51:02,639 --> 00:51:06,480
to manipulation. Manipulation? How so? Especially

1095
00:51:06,480 --> 00:51:09,340
as we move into an era of social robotics. Ah,

1096
00:51:09,380 --> 00:51:11,860
right. We are going to be sharing our homes with

1097
00:51:11,860 --> 00:51:15,159
robots very soon. And the book warns about something

1098
00:51:15,159 --> 00:51:17,980
called the unbearable automaticity of being.

1099
00:51:18,820 --> 00:51:20,900
Yes. That phrase comes from the psychologist

1100
00:51:20,900 --> 00:51:24,300
John Barr. It highlights a humbling fact. A massive

1101
00:51:24,300 --> 00:51:27,360
portion of our complex social behaviors, empathy,

1102
00:51:27,699 --> 00:51:30,420
bonding, trust, are not driven by high -level

1103
00:51:30,420 --> 00:51:33,260
rational thought. They are driven by deeply embedded

1104
00:51:33,260 --> 00:51:36,539
automatic physical reflexes. Like contagious

1105
00:51:36,539 --> 00:51:39,639
yawning. You see someone yawn and your jaw just

1106
00:51:39,639 --> 00:51:42,360
drops open before you even consciously register

1107
00:51:42,360 --> 00:51:45,179
it. Or, if someone makes eye contact and smiles

1108
00:51:45,179 --> 00:51:47,400
warmly at you, the muscles in your face will

1109
00:51:47,400 --> 00:51:50,239
instinctively begin to micro -smile in return.

1110
00:51:50,659 --> 00:51:53,780
It is an automatic, embodied reflex designed

1111
00:51:53,780 --> 00:51:57,380
to build social cohesion. And roboticists are

1112
00:51:57,380 --> 00:52:00,639
keenly aware of this? They are. They are building

1113
00:52:00,639 --> 00:52:03,579
humanoid companion robots, like Cynthia Brazil's

1114
00:52:03,579 --> 00:52:06,900
Kismet at MIT, or the parotherapeutic seal used

1115
00:52:06,900 --> 00:52:09,780
in nursing homes, that actively exploit these

1116
00:52:09,780 --> 00:52:12,159
physical reflexes. How do they exploit them?

1117
00:52:12,300 --> 00:52:14,980
A robot doesn't need artificial general intelligence

1118
00:52:14,980 --> 00:52:17,360
to make a human being fall in love with it. It

1119
00:52:17,360 --> 00:52:19,719
just needs to trigger your physical hardware.

1120
00:52:21,000 --> 00:52:23,900
Wow. Kismet was essentially a robotic head with

1121
00:52:23,900 --> 00:52:27,420
big expressive eyes and ears. It had a very simple

1122
00:52:27,420 --> 00:52:30,579
program. If a human face looks at me, tilt my

1123
00:52:30,579 --> 00:52:32,960
head slightly to the side, widen my mechanical

1124
00:52:32,960 --> 00:52:36,000
eyes, and track their movement. And we fall for

1125
00:52:36,000 --> 00:52:39,079
it instantly. We fall for it completely. The

1126
00:52:39,079 --> 00:52:41,719
human brain sees those physical cues, the widened

1127
00:52:41,719 --> 00:52:44,340
eyes, the attentive posture, and our automatic

1128
00:52:44,340 --> 00:52:47,940
empathy reflexes fire wildly. We feel an intense,

1129
00:52:48,340 --> 00:52:51,119
undeniable emotional bond with a machine that

1130
00:52:51,119 --> 00:52:53,880
has the cognitive depth of a toaster. Because

1131
00:52:53,880 --> 00:52:56,039
our bodies are doing the thinking, and our bodies

1132
00:52:56,039 --> 00:52:58,940
say, big eyes looking at me means a soulless

1133
00:52:58,940 --> 00:53:02,610
present. But why is that a danger? I mean, if

1134
00:53:02,610 --> 00:53:05,510
a robot's seal brings comfort to an elderly patient,

1135
00:53:05,630 --> 00:53:08,150
isn't that a good thing? It is beneficial in

1136
00:53:08,150 --> 00:53:10,989
specific contexts, but it introduces a massive

1137
00:53:10,989 --> 00:53:13,510
risk. The frame of reference problem strikes

1138
00:53:13,510 --> 00:53:17,349
again. Okay, explain that. Because the robot

1139
00:53:17,349 --> 00:53:20,349
moves like us and triggers our social reflexes,

1140
00:53:20,710 --> 00:53:23,190
we project human -like common sense onto it.

1141
00:53:23,530 --> 00:53:25,469
We assume that because the robot understands

1142
00:53:25,469 --> 00:53:28,230
eye contact, it must understand the world the

1143
00:53:28,230 --> 00:53:32,070
way we do. Right. If my robotic housekeeper smiles

1144
00:53:32,070 --> 00:53:34,650
at me and gently tilt its head when I speak,

1145
00:53:35,110 --> 00:53:37,230
I automatically assume it also possesses the

1146
00:53:37,230 --> 00:53:39,550
common sense to know that you shouldn't put the

1147
00:53:39,550 --> 00:53:42,719
family cat in the dishwasher to clean it. Exactly.

1148
00:53:43,139 --> 00:53:45,539
But the robot does not have common sense. Why?

1149
00:53:46,559 --> 00:53:48,539
Because common sense is not a list of rules.

1150
00:53:48,980 --> 00:53:52,079
It is an intuitive understanding of the physical

1151
00:53:52,079 --> 00:53:55,199
world derived from having a squishy, vulnerable

1152
00:53:55,199 --> 00:53:59,079
human body. The robot has metal joints. It does

1153
00:53:59,079 --> 00:54:01,519
not feel physical pain. It does not feel temperature.

1154
00:54:02,039 --> 00:54:05,019
It does not fear death. Because its physical

1155
00:54:05,019 --> 00:54:08,000
embodiment is fundamentally alien to ours, its

1156
00:54:08,000 --> 00:54:09,820
understanding of the world will be fundamentally

1157
00:54:09,820 --> 00:54:12,699
alien to ours. So if we forget the lessons of

1158
00:54:12,699 --> 00:54:15,679
embodied cognition, we will build machines that

1159
00:54:15,679 --> 00:54:18,820
can perfectly mimic human empathy, but fail us

1160
00:54:18,820 --> 00:54:21,940
in catastrophic practical ways because they lack

1161
00:54:21,940 --> 00:54:24,699
the physical grounding to understand human fragility.

1162
00:54:25,480 --> 00:54:30,119
That is the danger, exactly. This has been a

1163
00:54:30,119 --> 00:54:32,440
truly mind -bending journey. I mean, we started

1164
00:54:32,440 --> 00:54:34,380
with Rene Descartes sitting by his fireplace,

1165
00:54:35,219 --> 00:54:37,380
completely severing the human mind from the messy

1166
00:54:37,380 --> 00:54:40,079
reality of the physical world. A big mistake.

1167
00:54:40,969 --> 00:54:43,849
We saw how that Cartesian error led early AI

1168
00:54:43,849 --> 00:54:46,469
researchers to build incredibly smart chess computers

1169
00:54:46,469 --> 00:54:49,329
that couldn't walk across a room. We watched

1170
00:54:49,329 --> 00:54:52,150
blind Swiss robots act like brilliant janitors

1171
00:54:52,150 --> 00:54:55,610
just by bumping into things. The amnesiac snowplows.

1172
00:54:55,929 --> 00:54:58,539
Right. We realized our squishy fingertips are

1173
00:54:58,539 --> 00:55:01,980
solving complex geometry for free. We saw a lamprey's

1174
00:55:01,980 --> 00:55:04,119
disembodied brain spin out of control without

1175
00:55:04,119 --> 00:55:06,840
the heavy resistance of water. And we watched

1176
00:55:06,840 --> 00:55:09,480
evolutionary algorithms mutate physical bodies

1177
00:55:09,480 --> 00:55:12,199
to solve puzzles. We discovered that math is

1178
00:55:12,199 --> 00:55:14,920
just a metaphor for walking. And we realized

1179
00:55:14,920 --> 00:55:17,099
that our very memories are physically propped

1180
00:55:17,099 --> 00:55:20,539
up by the living rooms we sit in. It really is

1181
00:55:20,539 --> 00:55:23,500
a complete paradigm shift. Intelligence is not

1182
00:55:23,500 --> 00:55:26,349
just a brain floating in a vat. It is not an

1183
00:55:26,349 --> 00:55:29,230
ethereal software program running on interchangeable

1184
00:55:29,230 --> 00:55:33,190
meat hardware. Intelligence is a delicate, beautiful,

1185
00:55:33,710 --> 00:55:36,150
and continuous physical dance between a central

1186
00:55:36,150 --> 00:55:39,010
nervous system, a squishy body constrained by

1187
00:55:39,010 --> 00:55:42,070
gravity, and an unpredictable, messy environment.

1188
00:55:42,670 --> 00:55:46,210
Perfectly said. So, here is my challenge to you

1189
00:55:46,210 --> 00:55:49,929
today. Sometime in the next few hours, just stop

1190
00:55:49,929 --> 00:55:52,309
and pay close attention to your own physical

1191
00:55:52,309 --> 00:55:55,269
movements. Notice how your fingers instinctively

1192
00:55:55,269 --> 00:55:57,210
deform and wrap perfectly around the handle of

1193
00:55:57,210 --> 00:55:59,969
your coffee mug without you doing a single ounce

1194
00:55:59,969 --> 00:56:03,309
of conscious math. Notice how your stride effortlessly

1195
00:56:03,309 --> 00:56:05,530
changes when you walk down a flight of stairs.

1196
00:56:06,070 --> 00:56:08,369
Realize that your body is not just a dumb vehicle

1197
00:56:08,369 --> 00:56:11,150
carrying your brain around. Your body is actively

1198
00:56:11,150 --> 00:56:14,170
doing the thinking for you. And if you'll indulge

1199
00:56:14,170 --> 00:56:16,889
me, I would like to leave you with one final

1200
00:56:16,889 --> 00:56:19,269
lingering question to turn over in your mind.

1201
00:56:19,730 --> 00:56:21,610
Please do. We have established that the very

1202
00:56:21,610 --> 00:56:24,150
foundation of human intelligence, of our deep

1203
00:56:24,150 --> 00:56:26,489
emotions, of our language, and even of our highest

1204
00:56:26,489 --> 00:56:29,469
mathematics, is built entirely upon the specific

1205
00:56:29,469 --> 00:56:32,070
shape, squishiness, and physical vulnerabilities

1206
00:56:32,070 --> 00:56:35,789
of the human form. Yes. If that is true, what

1207
00:56:35,789 --> 00:56:38,730
does that mean for our current frantic pursuit

1208
00:56:38,730 --> 00:56:41,889
of artificial super -intelligence? Oh man. If

1209
00:56:41,889 --> 00:56:44,809
we ever do succeed in creating a god -like AI,

1210
00:56:45,070 --> 00:56:47,550
floating purely in a digital cloud, completely

1211
00:56:47,550 --> 00:56:50,349
devoid of a physical body, Will it even be capable

1212
00:56:50,349 --> 00:56:53,010
of understanding what it means to be human? Or

1213
00:56:53,010 --> 00:56:56,110
will we be completely, utterly alien to each

1214
00:56:56,110 --> 00:56:58,349
other? You know, when you buy a new computer,

1215
00:56:58,809 --> 00:57:01,969
you expect a comforting, clear division. The

1216
00:57:01,969 --> 00:57:04,429
hardware just sits there, and the invisible software

1217
00:57:04,429 --> 00:57:08,650
runs the show. But maybe, well, maybe the software

1218
00:57:08,650 --> 00:57:11,769
is just a ghost, and the hardware was the intelligence

1219
00:57:11,769 --> 00:57:14,010
all along. Until next time.