WEBVTT

1
00:00:00.391 --> 00:00:04.776
There's a question that haunts them every time, which is, is my loved one still there?

2
00:00:05.456 --> 00:00:09.981
And this question is one of the hardest problems in all of neuroscience, right?

3
00:00:10.121 --> 00:00:11.142
What is consciousness?

4
00:00:11.222 --> 00:00:18.990
He used AI to reverse engineer the biological mechanism of unconsciousness in coma patients.

5
00:00:19.010 --> 00:00:21.832
AI has rediscovered known features of unconsciousness.

6
00:00:22.493 --> 00:00:26.796
And then it's also predicted two entirely new biological mechanisms.

7
00:00:27.136 --> 00:00:27.917
Hello, Internet.

8
00:00:27.937 --> 00:00:35.563
This is your captain speaking, Lester Nare, joined as always by my co-host and our resident PhD, Krishna Chowdhury.

9
00:00:36.103 --> 00:00:48.432
We are back for another deep dive episode where we are going to talk about a paper today that's trying to tackle one of the hardest questions in neuroscience about comas.

10
00:00:48.992 --> 00:00:54.558
and how AI is helping us better understand the state of deep, prolonged unconsciousness.

11
00:00:54.938 --> 00:01:00.604
This was a paper published in Nature Neuroscience by friend of the pod, Daniel Toker, et al.

12
00:01:00.824 --> 00:01:02.186
on March 26th.

13
00:01:02.386 --> 00:01:09.073
And as always, we are going to learn about the science from the ground up, because this is From First Principles.

14
00:01:24.303 --> 00:01:27.567
Every year, tens of thousands of people fall into comas.

15
00:01:28.288 --> 00:01:33.956
And it's either from traumatic brain injury or a stroke, cardiac arrest, drug overdoses.

16
00:01:34.697 --> 00:01:39.984
And whenever this happens, some patients recover and some patients don't.

17
00:01:41.132 --> 00:01:44.275
We don't really know why some recover and why some don't.

18
00:01:44.736 --> 00:01:52.745
The other thing is for families that are sitting in that intensive care unit, there's a question that haunts them every time, which is, is my loved one still there?

19
00:01:53.425 --> 00:01:57.250
And this question is one of the hardest problems in all of neuroscience.

20
00:01:58.070 --> 00:01:59.092
What is consciousness?

21
00:01:59.172 --> 00:02:08.561
It's not just philosophically hard, which that's above my pay grade and we're not going to get into, but even mechanistically and clinically, this is a very hard problem.

22
00:02:08.662 --> 00:02:11.324
What is consciousness even physically speaking?

23
00:02:11.965 --> 00:02:13.727
Obviously, it has something to do with the brain.

24
00:02:15.286 --> 00:02:18.568
The human brain has 86 billion neurons.

25
00:02:19.028 --> 00:02:22.770
It's a muscle of fat and neural tissue.

26
00:02:23.131 --> 00:02:32.656
And it's quite incredible that that, like, sack of fat and tissue is where we get this profound phenomenon of consciousness, right?

27
00:02:32.696 --> 00:02:41.201
Our ability to perceive, to feel, to think, and to have this subjective experience of ourselves and the world around us.

28
00:02:41.621 --> 00:02:42.622
And it's all just from, like...

29
00:02:43.242 --> 00:02:48.264
this sack of like blood and fat in our skull, right?

30
00:02:48.304 --> 00:02:55.286
And one can ask, so it's somewhere in the brain, but what are the specific circuits that generate consciousness?

31
00:02:55.886 --> 00:02:58.947
And if a part of that circuit breaks, how does...

32
00:03:00.153 --> 00:03:01.194
it affect consciousness?

33
00:03:01.334 --> 00:03:05.117
And then can I actually figure out how to repair it, right?

34
00:03:05.938 --> 00:03:10.342
Now, doctors can observe that a patient is conscious or unconscious, right?

35
00:03:10.362 --> 00:03:15.727
Because they're not responsive, and they can classify the depth of unconsciousness on behavioral scales.

36
00:03:16.287 --> 00:03:16.728
But when

37
00:03:17.348 --> 00:03:20.129
All we can do really is like wait for them to wake up.

38
00:03:20.549 --> 00:03:27.590
There are some new therapies right now that we're going to get into that are maybe getting at trying to wake them up.

39
00:03:28.151 --> 00:03:38.113
But at the end of the day, we still don't have a way to peer inside the machinery of awareness and start asking like, what is the thing that is broken and how do we fix it?

40
00:03:38.533 --> 00:03:38.753
Okay.

41
00:03:38.793 --> 00:03:39.733
Or how do we replace it?

42
00:03:47.835 --> 00:04:12.984
and the actual mechanism that generates that it's kind of a little bit of a yeah dark cave yeah very much a black box in that sense right and that's why i really like this new paper it's a new study that i think represents a genuine step change in that field okay so it's a team that was led by daniel toker at ucla who is as you said a friend of the pod he went to college with us at princeton undergrad and

43
00:04:13.664 --> 00:04:21.426
Then I think he went to Berkeley for a PhD and then came to UCLA and now has like a gig with UCLA and USC as a postdoc.

44
00:04:22.126 --> 00:04:26.047
He's also the brain scientist on Instagram and TikTok.

45
00:04:26.067 --> 00:04:28.568
He's got a bunch of followers, really into science communication.

46
00:04:28.888 --> 00:04:30.708
You guys should check out his page.

47
00:04:30.748 --> 00:04:31.749
It's really, really cool.

48
00:04:32.369 --> 00:04:32.609
So...

49
00:04:34.176 --> 00:04:44.927
The cool thing about this paper, the cool thing that Daniel did was he used AI to reverse engineer the biological mechanism of unconsciousness in coma patients.

50
00:04:45.388 --> 00:04:52.715
And the way that he used AI is very different from, I think, when people think about using AI to do science.

51
00:04:53.095 --> 00:04:55.117
They think like, oh, black box LLM model.

52
00:04:55.197 --> 00:04:55.497
I like...

53
00:04:56.258 --> 00:05:01.883
Train it on a bunch of data and then it's going to spit out like a classifier consciousness or not consciousness.

54
00:05:02.264 --> 00:05:03.205
There's part of that here.

55
00:05:03.525 --> 00:05:11.072
But what really I think is cool about it is there's the AI has rediscovered known features of unconsciousness.

56
00:05:11.692 --> 00:05:16.036
And then it's also predicted to entirely new biological mechanisms.

57
00:05:16.076 --> 00:05:16.217
Mm hmm.

58
00:05:17.040 --> 00:05:22.322
Because the way that he has engineered this AI is it's not a black box.

59
00:05:22.883 --> 00:05:28.825
There's actual mechanistic understanding about a model of the human brain that he's hard-coded in.

60
00:05:29.205 --> 00:05:31.186
And that's what the AI is sort of acting on.

61
00:05:31.226 --> 00:05:34.628
That's what the evolution of the AI is acting on.

62
00:05:34.688 --> 00:05:35.828
And we're going to get into that.

63
00:05:35.848 --> 00:05:37.489
But I think it's very, very cool.

64
00:05:37.509 --> 00:05:40.030
And the implications are in every direction.

65
00:05:40.590 --> 00:05:43.631
I just want to make a quick note here.

66
00:05:43.672 --> 00:05:43.872
Yeah.

67
00:05:44.212 --> 00:05:48.794
We use the term AI in a very general sense when we discuss on the pod.

68
00:05:48.874 --> 00:05:54.917
It is a very expansive area of research with a lot of different subcategories and subboxes.

69
00:05:55.437 --> 00:05:59.939
So a lot of times we will just say AI for shorthand because we have a lot to get to.

70
00:06:00.019 --> 00:06:00.759
Yeah.

71
00:06:00.880 --> 00:06:04.121
But we don't mean AI equals LLM when we do so.

72
00:06:04.221 --> 00:06:04.441
Yes.

73
00:06:04.561 --> 00:06:07.803
It is sort of talking about the entire field of research.

74
00:06:07.843 --> 00:06:08.043
Yeah.

75
00:06:08.583 --> 00:06:13.087
at a high level and we then get into the weeds of what we're talking about specifically.

76
00:06:13.868 --> 00:06:19.413
But just as a small caveat note, that's important because it is a vast, vast space that is beyond just chatbots.

77
00:06:19.674 --> 00:06:20.154
Exactly.

78
00:06:20.194 --> 00:06:22.236
And this is very much not a chatbot.

79
00:06:22.256 --> 00:06:22.436
Right.

80
00:06:22.656 --> 00:06:23.597
It's very, very cool.

81
00:06:23.677 --> 00:06:24.618
We're going to get into it, right?

82
00:06:24.658 --> 00:06:27.761
And the implications, I think, radiate in all directions, right?

83
00:06:27.841 --> 00:06:33.306
Something like 5,500 Americans currently are in a persistent vegetative state and –

84
00:06:34.267 --> 00:06:40.729
There's a larger number that's in a minimally conscious state, and this research offers a genuine scientific pathway towards treatment.

85
00:06:41.649 --> 00:06:49.432
From the philosophy of mind perspective, it provides a very mechanistic specific picture of what makes a brain conscious.

86
00:06:50.132 --> 00:06:55.254
And then for an AI perspective, it's also very cool to think about how one can use brain.

87
00:06:56.370 --> 00:07:02.094
underlying principles of AI architectures to understand something as complicated as consciousness, right?

88
00:07:02.534 --> 00:07:06.376
So it's a deeply personal question that I think this is answering.

89
00:07:06.436 --> 00:07:12.540
Like consciousness is the one thing I think that we can all be sure exists because at least we experience it.

90
00:07:12.620 --> 00:07:15.122
It's the Descartes, I think, therefore I am,

91
00:07:15.742 --> 00:07:17.242
Clearly consciousness is a thing.

92
00:07:17.282 --> 00:07:18.162
Maybe the world isn't.

93
00:07:18.483 --> 00:07:18.683
Right.

94
00:07:18.803 --> 00:07:20.223
And quantum fields aren't.

95
00:07:20.363 --> 00:07:26.184
But consciousness, like 100 percent, like I am me, right, sitting here talking in this microphone.

96
00:07:26.584 --> 00:07:29.325
And you are you and everybody agrees that like that is a thing.

97
00:07:29.805 --> 00:07:32.925
So it's a deeply personal question.

98
00:07:33.346 --> 00:07:38.987
And it's still a phenomenon that we, I think, understand the least out of everything.

99
00:07:39.047 --> 00:07:40.207
So any question.

100
00:07:41.230 --> 00:07:44.614
Light that is shed on this mystery, I think, is certainly welcome.

101
00:07:45.194 --> 00:07:45.474
All right.

102
00:07:45.535 --> 00:07:49.479
So with that in mind, let's get into some of the history behind this type of research.

103
00:07:49.499 --> 00:07:49.819
Yeah.

104
00:07:50.099 --> 00:07:52.662
And we'll start all the way with the ancient Greeks.

105
00:07:53.182 --> 00:07:53.463
Always.

106
00:07:53.743 --> 00:07:57.307
OK, because the ancient Greeks, they had some crazy ideas.

107
00:07:57.787 --> 00:08:02.011
Like Aristotle thought that the brain was just a cooling organ for the blood.

108
00:08:03.264 --> 00:08:05.246
And consciousness actually lived in the heart.

109
00:08:05.926 --> 00:08:07.448
We're going to let that slide.

110
00:08:08.008 --> 00:08:13.413
Because other people like Hippocrates, the guy from the Hippocratic Oath, right?

111
00:08:13.473 --> 00:08:20.299
He noted that patients who fell into deep, unresponsive sleep after a head trauma rarely ever recovered.

112
00:08:20.859 --> 00:08:21.120
Okay?

113
00:08:21.580 --> 00:08:24.903
And the word coma comes from deep sleep.

114
00:08:26.230 --> 00:08:26.750
in Greek.

115
00:08:27.291 --> 00:08:33.614
So they had already sort of classified normal sleep versus this pathological deep sleep coma.

116
00:08:33.674 --> 00:08:35.175
They had already kind of figured that out.

117
00:08:35.275 --> 00:08:37.656
But again, it was just sort of an observation.

118
00:08:38.177 --> 00:08:42.799
Then in the Renaissance, there came the anatomical revolution with Visalius.

119
00:08:43.119 --> 00:08:50.943
He published this massive manuscript called De Humani Corporis Fabrica in 1543.

120
00:08:51.003 --> 00:08:55.246
He was basically really into dissecting human bodies.

121
00:08:56.046 --> 00:08:59.489
and gave us incredible drawings of human anatomy.

122
00:08:59.829 --> 00:09:02.732
So from there, now we've got an understanding of brain structure, right?

123
00:09:03.032 --> 00:09:08.036
But it's kind of like the machine is getting cataloged, but we don't really know what any of the parts are doing.

124
00:09:08.216 --> 00:09:18.525
So already by the early 1600s, we both had like a philosophical framework or concepts developing about classifying what these different parts of the body are doing and how that works.

125
00:09:19.165 --> 00:09:21.448
Like how that translates behaviorally.

126
00:09:21.528 --> 00:09:21.728
Yeah.

127
00:09:21.828 --> 00:09:27.435
As well as sort of a very detailed cataloging of the physical structures.

128
00:09:27.495 --> 00:09:27.715
Yeah.

129
00:09:27.875 --> 00:09:30.578
At least as it related to the brain and other key parts of the human body.

130
00:09:31.419 --> 00:09:33.282
You know, four or five hundred years ago.

131
00:09:33.302 --> 00:09:33.522
Yes.

132
00:09:33.582 --> 00:09:34.243
Yeah, exactly.

133
00:09:34.403 --> 00:09:35.184
It was a catalog.

134
00:09:35.224 --> 00:09:35.784
We didn't know what...

135
00:09:36.365 --> 00:09:39.846
Each thing did, but there was a pretty good catalog already, right?

136
00:09:40.326 --> 00:09:44.147
And then in the 1800s, we had some pretty revolutionary ideas.

137
00:09:45.147 --> 00:09:49.768
Paul Broca in 1861, he discovered this idea of cortical localization.

138
00:09:50.128 --> 00:09:58.270
There's an area of the brain now called Broca's area, and now we know that that damage to that area causes language loss.

139
00:09:58.630 --> 00:10:02.371
We also know that electrical stimulation in the motor cortex causes muscle movements.

140
00:10:06.232 --> 00:10:06.738
Thank you.

141
00:10:07.040 --> 00:10:11.983
a machine that has specialized parts, and each of these parts is doing something, right?

142
00:10:12.784 --> 00:10:15.746
The other invention in the 1800s was electrophysiology.

143
00:10:15.786 --> 00:10:22.591
In the 1870s, Richard Caton placed electrodes on exposed skulls of rabbits and found that there was electrical activity.

144
00:10:22.911 --> 00:10:25.372
And then we see Hans Berger in 1929.

145
00:10:25.873 --> 00:10:33.278
He develops the EEG, the electroencephalogram, which is you put leads on the skull and you can actually read electrical activity.

146
00:10:33.298 --> 00:10:35.980
And he found that the brain electrical activity

147
00:10:36.720 --> 00:10:40.041
dramatically changes with the state of consciousness.

148
00:10:40.581 --> 00:10:40.841
Okay?

149
00:10:41.201 --> 00:10:45.582
So different patterns mean waking, sleep, anesthesia.

150
00:10:46.042 --> 00:10:54.123
So already now it's like, okay, the brain is doing different things during unconsciousness, so perhaps the brain is really the seat of consciousness.

151
00:10:54.323 --> 00:10:55.063
That kind of makes sense.

152
00:10:55.083 --> 00:10:55.483
Yes.

153
00:10:55.503 --> 00:10:58.464
Yeah, yeah, yeah.

154
00:10:58.504 --> 00:11:03.845
Finally, 1949, Giuseppe Morozzi and Horace Maguan, they published this landmark paper...

155
00:11:04.725 --> 00:11:08.468
And this demonstrates something called the reticular activating system.

156
00:11:09.068 --> 00:11:11.670
It's a diffuse system that runs in the brainstem.

157
00:11:12.070 --> 00:11:15.693
The brainstem is where your brain meets the rest of the body with the spinal cord.

158
00:11:15.773 --> 00:11:18.615
So it's kind of like the connecting part of the brain.

159
00:11:19.155 --> 00:11:27.481
And they found that that part of the brain, the part that connects the brain to the spinal cord, that is essential for maintaining wakefulness.

160
00:11:28.202 --> 00:11:32.045
So now we've honed in on this is the sort of general area.

161
00:11:32.928 --> 00:11:34.970
where consciousness is probably happening.

162
00:11:35.830 --> 00:11:44.617
So to make a crude analogy, we figured out the brain is the factory, and we now know that different parts of the factory have different workers.

163
00:11:45.017 --> 00:11:46.418
The engineers are downstairs.

164
00:11:47.399 --> 00:11:49.600
The sort of manufacturing line is upstairs.

165
00:11:50.021 --> 00:11:51.602
The finance department is over here.

166
00:11:53.323 --> 00:11:55.925
But we didn't really know where the boss's office was.

167
00:11:56.205 --> 00:11:56.425
Yeah.

168
00:11:56.525 --> 00:11:57.666
Where's the boss's office?

169
00:11:57.766 --> 00:11:57.947
Yeah.

170
00:11:58.087 --> 00:11:58.267
Yeah.

171
00:11:58.367 --> 00:12:00.469
It seems like that's where it's the brainstem.

172
00:12:00.589 --> 00:12:00.789
Right.

173
00:12:00.949 --> 00:12:01.590
It's that part.

174
00:12:01.650 --> 00:12:01.830
Right.

175
00:12:02.030 --> 00:12:06.814
And over the following decades, we've mapped out that brainstem in more and more detail.

176
00:12:07.234 --> 00:12:10.497
So we figured out there's something called the locus coriolis.

177
00:12:10.897 --> 00:12:12.859
There's something called a basal forebrain.

178
00:12:13.139 --> 00:12:15.481
There's something that's part of the brainstem.

179
00:12:15.881 --> 00:12:17.102
There's the hypothalamus.

180
00:12:17.182 --> 00:12:18.544
There's the raphe nuclei.

181
00:12:18.824 --> 00:12:23.548
And so all of these like sort of parts, you see where the spinal cord is going in, that brainstem.

182
00:12:24.068 --> 00:12:47.158
junction is where all of this is happening which it's it's almost like even just from a crude structural like just looking at it you would imagine if the things that are most sensitive have the most protection just locationally yes locationally that makes sense from one perspective of evolution and the other perspective of evolution i think is that like the brain sort of grew out from that part right yeah

183
00:12:47.638 --> 00:12:51.801
So it's like the oldest part of the brain is the part that connects it to the rest of the body.

184
00:12:51.821 --> 00:12:55.744
Because, I mean, let's face it, like the brain evolved to control movement.

185
00:12:56.404 --> 00:12:56.604
Right.

186
00:12:56.885 --> 00:13:01.208
And so the brain controlling the rest of the spinal cord is sort of probably how it started.

187
00:13:01.568 --> 00:13:03.949
And then like higher and higher cognition formed around that.

188
00:13:03.989 --> 00:13:04.190
Right.

189
00:13:04.430 --> 00:13:06.491
Like, oh, OK, I need a sense of place.

190
00:13:06.791 --> 00:13:09.233
I need a sense of time.

191
00:13:09.613 --> 00:13:11.395
I need a sense of problem solving.

192
00:13:11.655 --> 00:13:14.437
All of that starts forming around this like nucleus.

193
00:13:14.457 --> 00:13:14.577
Right.

194
00:13:15.097 --> 00:13:17.500
And then the prefrontal cortex being the most advanced.

195
00:13:17.520 --> 00:13:18.141
The most advanced.

196
00:13:18.161 --> 00:13:19.182
That's the one that's like...

197
00:13:19.262 --> 00:13:20.524
Right in the front.

198
00:13:20.564 --> 00:13:23.568
Which, you know, it's very cool to think about it that way.

199
00:13:23.588 --> 00:13:27.332
Just like in terms of stages of development over the millions of years of evolution.

200
00:13:27.412 --> 00:13:27.693
Okay.

201
00:13:28.013 --> 00:13:28.293
Okay.

202
00:13:28.774 --> 00:13:33.319
So now we know that it's like kind of the brainstem, but we still don't have really an idea of like...

203
00:13:34.077 --> 00:13:40.799
Okay, this thing connects to this, this thing connects to this, so this is powering that, that, that, that, and so on and so forth.

204
00:13:40.839 --> 00:13:48.182
2007, there's a pivotal moment in consciousness research because Nicholas Schiff and his colleagues published this paper in Nature.

205
00:13:48.682 --> 00:13:53.424
And it showed that continuous deep brain stimulation, this is the idea of we take an electrode...

206
00:13:53.864 --> 00:13:59.067
and we pulse it with electricity, that is going to excite the neurons near that electrode.

207
00:13:59.327 --> 00:14:01.749
It's just you're pumping electricity deep into the brain.

208
00:14:02.089 --> 00:14:03.830
That's why it's deep brain stimulation.

209
00:14:04.210 --> 00:14:09.113
And deep brain stimulation of the central thalamus helped a minimally conscious patient

210
00:14:10.113 --> 00:14:37.146
regain the ability to communicate and feed himself okay this was a very big deal because this was an ability that he had lost about six years earlier after a traumatic brain injury it's a preliminary result single patient notoriously in human consciousness studies and things like that the n is like very low single digits is you're already getting a lot because human i mean you're like experimenting on human beings right there's tons of ethical concerns yep there's not a lot of like

211
00:14:37.746 --> 00:14:39.246
You know, people who are willing to do this.

212
00:14:39.547 --> 00:14:44.148
And so even though the data is low, the fact that it worked means that there's something going on.

213
00:14:44.448 --> 00:14:46.648
So that sort of rejuvenated this research.

214
00:14:47.009 --> 00:14:53.610
And in 2010, that same author, Nicholas Schiff, he came up with something called the Mesocircuit Hypothesis.

215
00:14:54.191 --> 00:14:54.431
Okay?

216
00:14:55.451 --> 00:14:57.111
Effectively, what he's saying is the following.

217
00:14:57.131 --> 00:14:59.092
The consciousness relies on a network.

218
00:15:00.291 --> 00:15:03.736
of interactions between the cortex and the central thalamus.

219
00:15:04.036 --> 00:15:04.197
Okay.

220
00:15:04.717 --> 00:15:09.304
And that interaction is regulated by a part of the brain called the basal ganglia.

221
00:15:09.864 --> 00:15:12.268
We're going to focus on effectively three...

222
00:15:14.044 --> 00:15:15.165
regions of the brain.

223
00:15:15.405 --> 00:15:15.685
Okay.

224
00:15:16.265 --> 00:15:19.907
And I'm going to use an analogy of an electrical grid.

225
00:15:20.427 --> 00:15:20.667
Okay.

226
00:15:20.687 --> 00:15:20.887
Okay.

227
00:15:21.247 --> 00:15:22.368
A city's electrical grid.

228
00:15:22.408 --> 00:15:22.588
Yes.

229
00:15:22.768 --> 00:15:29.031
So the cerebral cortex, this is the part that's on the outside, the most sort of recently developed in humans.

230
00:15:29.571 --> 00:15:34.634
Um, this is the, when we think about the brain and all the wrinkles, that's the cerebral cortex that we're seeing on the outside.

231
00:15:35.114 --> 00:15:39.596
And that outer layer, it's responsible for higher thought, memory, sensory processing,

232
00:15:40.236 --> 00:15:50.446
And in this analogy of a city's electrical grid, this is your network of homes, podcast studios, businesses that consume the electricity, right?

233
00:15:50.686 --> 00:15:54.470
Now, the central thalamus is the deep brain relay station.

234
00:15:54.550 --> 00:15:58.033
That's the part that's sort of on the right over there in the pink.

235
00:15:58.333 --> 00:16:00.254
Just above the brainstem?

236
00:16:00.355 --> 00:16:01.435
Just above the brainstem.

237
00:16:01.475 --> 00:16:05.518
Okay, so that's getting input from the cortex, and it's also giving input to the cortex.

238
00:16:05.558 --> 00:16:06.759
Okay, so it's a two-way street.

239
00:16:07.359 --> 00:16:09.581
It's kind of like the primary power substation.

240
00:16:09.621 --> 00:16:09.821
Yes.

241
00:16:10.081 --> 00:16:15.825
It's pumping excitatory input to the cortex, but it's also getting excitatory input back from the cortex.

242
00:16:16.106 --> 00:16:17.987
So it's relaying a bunch of electricity.

243
00:16:18.327 --> 00:16:19.068
One way or the other.

244
00:16:19.228 --> 00:16:25.853
So one of the examples being if you have solar on your house, you're sending electricity back to the grid.

245
00:16:25.893 --> 00:16:27.855
It's not just you're receiving it as a practical example.

246
00:16:27.875 --> 00:16:28.395
That's very good.

247
00:16:28.415 --> 00:16:28.856
Yeah, yeah, yeah.

248
00:16:28.896 --> 00:16:29.756
No, that's exactly right.

249
00:16:30.197 --> 00:16:33.359
And this feedback loop kind of helps both run.

250
00:16:33.759 --> 00:16:33.920
Right.

251
00:16:34.080 --> 00:16:34.340
Okay?

252
00:16:34.580 --> 00:16:35.141
In some sense.

253
00:16:35.381 --> 00:16:39.884
And then finally there's the basal ganglia, and that's the striatum and the globus pallidus.

254
00:16:40.205 --> 00:16:40.365
Mm-hmm.

255
00:16:40.953 --> 00:16:49.302
That is a kind of regulatory system, and it's kind of like a voltage controller that manages the flow of power in this entire network.

256
00:16:49.662 --> 00:16:55.468
So in the summer when it gets hot, they have to decide where to send energy, where to not send energy in order to not overflow.

257
00:16:55.488 --> 00:17:01.795
Like there has to be – somebody decides, okay, we're going to shut down this neighborhood because we need to power the nuclear.

258
00:17:01.795 --> 00:17:03.216
Yeah, exactly.

259
00:17:03.236 --> 00:17:13.260
As a crude analogy, it's kind of routing signals, and it's controlling both the power stations and the homes to make sure there's balance and nothing's going out of the way.

260
00:17:13.680 --> 00:17:18.282
Okay, so that's how sort of the consciousness mechanism is working in healthy patients.

261
00:17:18.362 --> 00:17:19.583
What happens during a coma?

262
00:17:19.923 --> 00:17:22.964
Well, according to the mesocircuit hypothesis, here's what happens.

263
00:17:23.485 --> 00:17:31.048
When there's widespread brain injury, like from a traumatic brain injury or a stroke, so a bunch of neurons die because they don't get oxygen and things like that.

264
00:17:32.268 --> 00:17:34.709
your power lines are going to be down on the cortex, right?

265
00:17:34.729 --> 00:17:42.052
So you're going to get localized areas of power cuts, like we do in LA sometimes when somebody steals our copper from underground.

266
00:17:43.093 --> 00:17:49.615
But anyways, the cortex is now going to reduce its excitatory input to the thalamus.

267
00:17:49.635 --> 00:17:51.856
That connection is broken, as you can see, on the left.

268
00:17:51.896 --> 00:17:54.037
The thalamus and the cortex, that's now a dotted line.

269
00:17:54.357 --> 00:17:55.998
Instead of on the left, it's a solid line.

270
00:17:56.278 --> 00:17:57.679
Okay, the other thing is,

271
00:17:58.719 --> 00:18:04.720
There are these special neurons in the striatum called medium spiny neurons.

272
00:18:04.920 --> 00:18:09.061
And what they do, they're super prone to metabolic stress death.

273
00:18:09.201 --> 00:18:12.902
Meaning you, like, remove a little bit of oxygen and they'll just die.

274
00:18:13.222 --> 00:18:13.402
Okay?

275
00:18:13.422 --> 00:18:14.482
They're not very resilient.

276
00:18:14.862 --> 00:18:16.083
They're very finicky.

277
00:18:16.463 --> 00:18:23.844
And if those guys die, then the striatum is going to not inhibit the part that is inhibiting the thalamus.

278
00:18:24.144 --> 00:18:25.345
It's a double negative now.

279
00:18:25.485 --> 00:18:25.805
Okay.

280
00:18:25.885 --> 00:18:26.185
Okay?

281
00:18:26.485 --> 00:18:26.785
So...

282
00:18:28.143 --> 00:18:35.951
Because of that, so what it's doing is usually the striatum is like inhibiting the part that is putting brakes on the thalamus.

283
00:18:36.411 --> 00:18:36.652
Okay?

284
00:18:36.972 --> 00:18:41.657
But now that that brake is gone, the first brake is gone, right?

285
00:18:41.677 --> 00:18:42.117
Right.

286
00:18:42.970 --> 00:18:45.771
The second negative is going to go even more negative.

287
00:18:46.111 --> 00:18:46.371
Okay?

288
00:18:46.572 --> 00:18:48.912
And it's going to start inhibiting the thalamus even more.

289
00:18:48.953 --> 00:18:50.673
That's why that red line just got fatter.

290
00:18:50.733 --> 00:18:51.133
That makes sense.

291
00:18:51.153 --> 00:18:51.293
Right?

292
00:18:51.534 --> 00:18:55.595
So the thalamus is, one, not getting any input from the cortex because the power lines are down.

293
00:18:55.755 --> 00:18:55.935
Yes.

294
00:18:56.015 --> 00:19:02.738
And second, the part that was putting brakes on the thalamus has now gone hyperactive and putting even more brakes on the thalamus.

295
00:19:02.838 --> 00:19:04.379
So the thalamus is getting inhibited.

296
00:19:04.719 --> 00:19:09.081
It's getting shut down more and more in a coma because of these very tiny...

297
00:19:10.223 --> 00:19:12.905
Defects that have happened in exactly the wrong areas.

298
00:19:13.325 --> 00:19:13.525
Yeah.

299
00:19:14.045 --> 00:19:15.326
So this is very interesting.

300
00:19:15.346 --> 00:19:26.832
And what we're sort of saying here is the traumatic brain injuries are disrupting the normal flow between these three fundamental component parts.

301
00:19:26.852 --> 00:19:27.953
Yeah.

302
00:19:28.333 --> 00:19:37.078
The cortex, which is like the end destination, the Holmes example, the thalamus, which is sort of sending and receiving, and then the striatum and palliatum that are...

303
00:19:37.598 --> 00:19:42.201
kind of controlling the amount of what is or is not being sent and where it is or is not being sent.

304
00:19:42.221 --> 00:19:42.662
Exactly.

305
00:19:42.742 --> 00:19:49.466
And it's exactly at the right part where the thalamus, which is the relay station, is just getting signaled to shut down.

306
00:19:49.546 --> 00:19:50.267
Yeah, yeah, yeah.

307
00:19:50.287 --> 00:19:50.427
Okay?

308
00:19:50.487 --> 00:19:55.230
And when that shuts down, now this entire circuit of consciousness is getting shut down.

309
00:19:55.291 --> 00:20:00.554
It's almost like the fail-safe system that happens at a substation if there's some overload, just gets stuck in the locked position.

310
00:20:00.574 --> 00:20:00.935
Exactly.

311
00:20:00.955 --> 00:20:01.395
Exactly.

312
00:20:01.795 --> 00:20:02.236
Exactly.

313
00:20:02.296 --> 00:20:07.582
And there's some evidence that this mesocircuit hypothesis is probably correct.

314
00:20:07.942 --> 00:20:17.292
There's certain stimulant drugs, for example, amantadine, which enhances the inhibition to that break, like the part that's inhibiting the thalamus.

315
00:20:18.202 --> 00:20:20.805
amantadine inhibits that part.

316
00:20:20.946 --> 00:20:23.949
So it's no longer going to inhibit the thalamus kind of works.

317
00:20:24.029 --> 00:20:26.813
And it kind of wakes up patients from turning it to overclock.

318
00:20:26.833 --> 00:20:28.014
Yes, yes, exactly.

319
00:20:28.075 --> 00:20:31.359
So, so there, there's some mechanism that,

320
00:20:32.422 --> 00:20:35.584
where these drugs are kind of working with the mesocircuit hypothesis.

321
00:20:35.624 --> 00:20:38.966
But at the end of the day, still a hypothesis because it's not very granular, right?

322
00:20:39.006 --> 00:20:43.188
It's just like, oh, giant brain area.

323
00:20:43.308 --> 00:20:47.270
But the giant brain area has millions of neurons that are all different types.

324
00:20:47.530 --> 00:20:47.970
What type?

325
00:20:48.431 --> 00:20:49.631
What type is doing what, right?

326
00:20:50.672 --> 00:20:57.696
And also, you can't really test it directly because the bottleneck is standard animal models are really bad at mimicking neurons.

327
00:20:58.676 --> 00:21:01.458
human comas, this prolonged unconsciousness.

328
00:21:01.699 --> 00:21:08.743
And that just has to do with the fact that animal models like rodents, the brain is structurally a very different thing, right?

329
00:21:09.224 --> 00:21:10.405
Even though it's still a mammal...

330
00:21:11.405 --> 00:21:19.948
One of the things that we think happens during traumatic brain injury that causes this kind of coma is the idea of a diffuse axonal injury.

331
00:21:20.008 --> 00:21:20.548
Here's the idea.

332
00:21:20.808 --> 00:21:22.929
You've got the cerebral cortex on the outside.

333
00:21:23.669 --> 00:21:25.910
There's gray matter, which is like sort of the cell bodies.

334
00:21:26.010 --> 00:21:27.070
And then there's white matter.

335
00:21:27.090 --> 00:21:28.871
You might have heard of that gray matter and white matter.

336
00:21:29.331 --> 00:21:31.191
Gray matter is kind of like the cell bodies.

337
00:21:31.211 --> 00:21:32.652
The white matter is the axons.

338
00:21:32.952 --> 00:21:34.613
These are these like cables.

339
00:21:35.013 --> 00:21:38.674
Like, you know, the high voltage cables that come from, you know, Hoover Dam to L.A.?

340
00:21:39.374 --> 00:21:40.655
The axons are kind of like that.

341
00:21:40.675 --> 00:21:41.855
They're the high voltage cables.

342
00:21:42.215 --> 00:21:50.959
Now, if we go to a photo of the 14, what you'll see is the white matter is kind of in the middle and it's like connecting these areas of gray matter.

343
00:21:51.259 --> 00:21:59.483
But if you have a traumatic brain injury, let's say you got in a car crash or like, boom, like your brain had severe G-forces on it.

344
00:22:00.411 --> 00:22:03.734
Those axons, which are the cables, those are going to rupture first.

345
00:22:03.814 --> 00:22:04.394
Yeah.

346
00:22:04.474 --> 00:22:04.755
Right?

347
00:22:04.815 --> 00:22:05.175
Yeah.

348
00:22:05.635 --> 00:22:12.561
Now, with a small rodent, the brain is not big enough to feel that kind of G-force stress.

349
00:22:13.161 --> 00:22:13.341
Right?

350
00:22:13.462 --> 00:22:14.402
Ours is big enough.

351
00:22:14.442 --> 00:22:17.285
We're like, and we're like doing crazy things as humans.

352
00:22:17.645 --> 00:22:20.107
We're like, we're driving around at 70 miles an hour.

353
00:22:20.407 --> 00:22:23.970
We're like, we're now subject to G-forces if something goes wrong.

354
00:22:24.430 --> 00:22:24.590
Right?

355
00:22:24.671 --> 00:22:27.173
And so our brain is just not evolved.

356
00:22:28.073 --> 00:22:41.808
When we were hunter-gatherers in Africa, like evolving, the evolution didn't think, hey, I should probably make this brain, you know, withstand getting hit by another 350-pound man in an NFL game.

357
00:22:41.889 --> 00:22:42.669
Yeah, exactly.

358
00:22:42.689 --> 00:22:43.030
Exactly.

359
00:22:44.531 --> 00:23:00.708
So there's a mismatch between what the brain has designed, the environment that the brain is designed in, and then what the brain was capable of creating, which is like human beings in cars and all this other crazy stuff, skiing and then hitting a tree and things like that.

360
00:23:00.849 --> 00:23:01.009
Right.

361
00:23:01.269 --> 00:23:01.449
Right.

362
00:23:01.469 --> 00:23:04.532
So – and we simply can't test –

363
00:23:05.605 --> 00:23:06.847
In rodent models.

364
00:23:06.927 --> 00:23:07.147
Right.

365
00:23:07.328 --> 00:23:07.588
Okay.

366
00:23:07.748 --> 00:23:21.608
And this is something we talked about in a recent previous episode too about just, you know, there are a lot of opportunities to use animal models, which the purpose of which, yes, for folks who are, you know, PETA adjacent, et cetera.

367
00:23:22.649 --> 00:23:32.539
The ethical concerns about testing directly on humans are so prevalent that it's perceived as a lower ethically bad option.

368
00:23:32.579 --> 00:23:33.359
100%.

369
00:23:33.540 --> 00:23:36.543
And for an outcome that's going to save millions of lives at some point in the future.

370
00:23:36.563 --> 00:23:36.723
Yeah.

371
00:23:37.023 --> 00:23:38.104
Yeah, but here, like...

372
00:23:38.604 --> 00:23:38.964
It doesn't.

373
00:23:39.104 --> 00:23:39.464
It doesn't.

374
00:23:39.484 --> 00:23:44.327
The ethics, like, is even more problematic because, well, you're not going to get anything out of it.

375
00:23:44.407 --> 00:23:44.867
Right, right.

376
00:23:45.027 --> 00:23:45.207
Right.

377
00:23:45.267 --> 00:23:50.369
And so you're just... Yeah, you're just, like, wasting animals at that point, right?

378
00:23:50.409 --> 00:23:50.570
Right.

379
00:23:50.690 --> 00:23:53.091
And that's not something that we want to do in science research.

380
00:23:53.291 --> 00:23:53.431
Right.

381
00:23:53.451 --> 00:23:54.952
So that's one bottleneck.

382
00:23:55.112 --> 00:23:55.272
Okay.

383
00:23:55.292 --> 00:23:55.392
Okay.

384
00:23:55.772 --> 00:23:57.993
which is just animal models are not a thing.

385
00:23:58.533 --> 00:24:01.034
The second bottleneck is the diagnostic crisis.

386
00:24:01.494 --> 00:24:01.934
Okay.

387
00:24:01.994 --> 00:24:04.755
This is in general with consciousness in medicine.

388
00:24:05.115 --> 00:24:05.315
Okay.

389
00:24:05.575 --> 00:24:07.296
Consciousness has two fundamental components.

390
00:24:07.336 --> 00:24:10.057
There's wakefulness, which is like arousal, like when you're awake.

391
00:24:10.577 --> 00:24:11.798
And then there's also awareness.

392
00:24:11.838 --> 00:24:17.080
That's the qualitative experience of like me being me and seeing you and everyone else.

393
00:24:17.140 --> 00:24:17.280
Right.

394
00:24:17.300 --> 00:24:22.201
This is some qualitative subjective experience that only I have about what it's like to be me.

395
00:24:22.561 --> 00:24:22.762
Right.

396
00:24:23.182 --> 00:24:23.362
So,

397
00:24:24.442 --> 00:24:29.606
disorders of consciousness that comes from physical trauma and oxygen deprivation, there's several different types.

398
00:24:29.646 --> 00:24:33.869
So in a coma, you have neither arousal nor awareness, right?

399
00:24:33.929 --> 00:24:34.970
Because you can't be awoken.

400
00:24:35.150 --> 00:24:35.370
Yes.

401
00:24:35.630 --> 00:24:37.091
There's no sleep-wake cycle.

402
00:24:37.211 --> 00:24:37.451
Yes.

403
00:24:38.312 --> 00:24:40.133
And they fail to respond to stimuli.

404
00:24:40.313 --> 00:24:41.014
The lights are off.

405
00:24:41.154 --> 00:24:43.135
Yeah, lights are off and maybe nobody's home.

406
00:24:43.255 --> 00:24:43.436
Right.

407
00:24:43.536 --> 00:24:43.796
Okay?

408
00:24:44.356 --> 00:24:49.300
Vegetative state means it's also called unresponsive wakefulness syndrome.

409
00:24:49.640 --> 00:24:49.760
Okay.

410
00:24:49.800 --> 00:24:50.801
You have wakefulness.

411
00:24:52.516 --> 00:24:55.460
And you have a sleep-wake cycle, but perhaps you have no awareness.

412
00:24:55.821 --> 00:24:58.404
So the lights are on, but maybe nobody's home.

413
00:24:58.805 --> 00:24:59.105
Okay.

414
00:24:59.546 --> 00:25:01.068
But now it gets kind of weird.

415
00:25:01.428 --> 00:25:05.334
Like, what if they are aware, but they just can't...

416
00:25:06.917 --> 00:25:27.542
have any agency over their motor controls right so how can you tell so someone's home but they can't answer the door when you ring the doorbell exactly yeah lights are on lights are on and someone is home yeah you ring the doorbell but they can't yeah that's very good yeah exactly how do you how can you tell right this is a yeah that's subtle that's very subtle and for the

417
00:25:30.703 --> 00:25:33.086
Science paper 2006 by Adrian Owen.

418
00:25:33.386 --> 00:25:34.427
It's a landmark paper.

419
00:25:34.467 --> 00:25:35.408
It's quite incredible.

420
00:25:35.909 --> 00:25:37.731
This was a wake-up call, so to speak.

421
00:25:38.171 --> 00:25:38.451
Okay.

422
00:25:39.012 --> 00:25:48.162
He used functional MRI to ask a patient that was diagnosed in a vegetative state to imagine playing tennis and imagine walking around her home.

423
00:25:50.772 --> 00:25:56.620
And the brain activity from that fMRI study was identical to someone who was fully aware.

424
00:25:57.602 --> 00:26:00.646
So she's there in this vegetative state.

425
00:26:01.267 --> 00:26:04.731
There's no way that she can express her awareness.

426
00:26:04.851 --> 00:26:05.052
Mm-hmm.

427
00:26:05.352 --> 00:26:08.874
But if we look into the brain MRI, it's the same as someone who's aware.

428
00:26:10.335 --> 00:26:17.099
The idea being they may not be able to move their hands to touch a button or verbalize.

429
00:26:17.219 --> 00:26:18.400
Or even move their eyes.

430
00:26:18.420 --> 00:26:22.823
Or move their eyes, which is techniques that have been used before to communicate.

431
00:26:23.263 --> 00:26:26.865
Now we're just saying it's because the awareness is in the brain.

432
00:26:27.225 --> 00:26:29.647
And so you shouldn't actually need any of these secondary things.

433
00:26:30.700 --> 00:26:51.550
uh meth avenues yes to communicate and the only way to make the differentiation between the uh is someone home versus someone not home reference point you have to go directly to the source yeah if all of their physical capacities yeah are are not available to them which does not necessarily mean yeah that no one is home

434
00:26:51.690 --> 00:26:52.070
Exactly.

435
00:26:52.090 --> 00:26:55.551
It's like no one's answering the door, but here it's kind of like we did an X-ray of the home.

436
00:26:55.611 --> 00:26:55.831
Yeah, yeah.

437
00:26:55.871 --> 00:26:57.011
And the person is moving around.

438
00:26:57.031 --> 00:26:57.332
Right, right.

439
00:26:57.352 --> 00:26:58.572
We have the thermal and you can see.

440
00:26:58.592 --> 00:26:59.212
Yeah, exactly.

441
00:26:59.332 --> 00:26:59.612
You know?

442
00:26:59.652 --> 00:27:00.252
Okay.

443
00:27:00.352 --> 00:27:01.593
I thought that was a very cool paper.

444
00:27:01.613 --> 00:27:02.593
That is very clever.

445
00:27:02.833 --> 00:27:03.573
It's very clever.

446
00:27:03.593 --> 00:27:07.234
And now it's a real crisis, right?

447
00:27:07.254 --> 00:27:07.374
Right.

448
00:27:07.394 --> 00:27:08.295
Because now there's a third.

449
00:27:08.315 --> 00:27:09.915
It's called minimally conscious state.

450
00:27:10.095 --> 00:27:10.335
Mm-hmm.

451
00:27:10.435 --> 00:27:10.695
Okay?

452
00:27:10.995 --> 00:27:17.177
And it turns out that 15% to 20% of patients that are classified as vegetative... Mm-hmm.

453
00:27:17.617 --> 00:27:19.778
actually might have covert conscious awareness.

454
00:27:20.499 --> 00:27:21.139
That's a lot.

455
00:27:21.199 --> 00:27:22.180
That's got to be terrifying.

456
00:27:22.200 --> 00:27:23.760
And that's got to be so terrifying.

457
00:27:24.141 --> 00:27:24.341
Right?

458
00:27:24.401 --> 00:27:28.443
Because you can hear and see everything happening around you, and you cannot engage in that environment.

459
00:27:28.463 --> 00:27:29.664
That's just unbelievable.

460
00:27:29.684 --> 00:27:32.005
And this was only 20 years ago, 2006, right?

461
00:27:32.165 --> 00:27:34.586
After this 500-year cycle we've just talked about.

462
00:27:34.606 --> 00:27:34.766
Yeah.

463
00:27:34.947 --> 00:27:42.791
So, I mean, this field is just one of those where you can really sit back and be like, wow, we really don't know much about the brain.

464
00:27:42.831 --> 00:27:42.951
Right.

465
00:27:42.971 --> 00:27:43.131
Right.

466
00:27:43.551 --> 00:27:44.431
Right, right.

467
00:27:44.451 --> 00:27:51.133
Because this is like kind of, you know, people's, this goes to the materialist, non-materialist debate.

468
00:27:51.613 --> 00:28:03.157
You know, is it that consciousness arises out of matter or is consciousness primary, which gets a little, that's, again, that was above our pay grade and not the focus of this podcast.

469
00:28:03.177 --> 00:28:03.357
Yeah.

470
00:28:03.817 --> 00:28:05.218
But it is one of those.

471
00:28:05.798 --> 00:28:07.099
It's the first question.

472
00:28:07.199 --> 00:28:07.439
Yes.

473
00:28:07.819 --> 00:28:10.100
Because once you get, you know what I mean?

474
00:28:10.320 --> 00:28:14.242
You can't, everything else downstream is really impacted by which of the two.

475
00:28:14.423 --> 00:28:15.543
Which of those two, right?

476
00:28:15.623 --> 00:28:16.584
What are we talking about?

477
00:28:16.644 --> 00:28:17.164
Right.

478
00:28:17.564 --> 00:28:18.245
Who am I?

479
00:28:18.605 --> 00:28:19.145
Who are you?

480
00:28:19.365 --> 00:28:21.926
And what the hell is going on?

481
00:28:21.946 --> 00:28:22.247
Right.

482
00:28:22.307 --> 00:28:23.667
That's how I like to say it.

483
00:28:23.827 --> 00:28:24.208
Right.

484
00:28:24.308 --> 00:28:24.828
Right.

485
00:28:25.168 --> 00:28:25.488
You know?

486
00:28:25.789 --> 00:28:33.593
And so with these kinds of studies, it's really like underscoring how much we don't know about this, right?

487
00:28:33.613 --> 00:28:33.773
Right.

488
00:28:34.033 --> 00:28:39.197
And now 15% to 20% of patients, we're saying, are classified incorrectly.

489
00:28:39.437 --> 00:28:40.577
Which is fascinating.

490
00:28:40.597 --> 00:28:45.460
Because, I mean, I'm sure after that study, a variety of... Because the fMRIs are not hugely...

491
00:28:47.073 --> 00:29:12.109
negative for a patient no so at least as a way especially for families or anybody to try um but in any event exactly so now we finally get to um daniel's paper daniel toker in nature neuroscience adversarial ai reveals mechanisms and treatments for disorders of consciousness it's a really innovative way of using machine learning and the vast amount of data that's now available

492
00:29:12.809 --> 00:29:16.111
On comatose patients, vegetative patients.

493
00:29:16.271 --> 00:29:18.992
He's made an in-silico model of coma.

494
00:29:19.392 --> 00:29:26.175
So a computer model of the brain that circumvents the lack of mouse models.

495
00:29:26.415 --> 00:29:30.136
So now I have a model in the brain that I can play with.

496
00:29:30.537 --> 00:29:34.318
A model of the brain in my computer that I'm fairly confident is quite good.

497
00:29:34.478 --> 00:29:35.419
That I can play around with.

498
00:29:35.439 --> 00:29:38.400
Because the idea is it's based off of this real-world data.

499
00:29:38.440 --> 00:29:38.940
Anonymized.

500
00:29:38.980 --> 00:29:39.180
Yeah.

501
00:29:39.260 --> 00:29:39.640
Yeah.

502
00:29:41.565 --> 00:29:41.825
data.

503
00:29:42.225 --> 00:29:42.605
Exactly.

504
00:29:42.645 --> 00:29:52.549
And then I can figure out a kind of mechanistic model about how consciousness works, what gets disrupted to make a coma, and then it becomes a kind of discovery engine.

505
00:29:52.889 --> 00:29:55.470
From that, we can mess around with it and propose new treatments.

506
00:29:55.831 --> 00:29:56.871
And he is doing all of these.

507
00:29:57.471 --> 00:29:59.072
Very well done, Daniel, by the way.

508
00:29:59.752 --> 00:30:00.113
Quite good.

509
00:30:00.973 --> 00:30:01.374
Quite good.

510
00:30:01.754 --> 00:30:03.936
Again, thebrainscientist on Instagram.

511
00:30:04.497 --> 00:30:08.461
And before we deep dive into it, let's do a little bit of housekeeping.

512
00:30:08.521 --> 00:30:08.761
Yes.

513
00:30:09.041 --> 00:30:15.748
So as always, we are so grateful for all of you to join us for these research deep dives.

514
00:30:15.948 --> 00:30:19.271
It's just an incredible, not only an incredible time to be alive, but...

515
00:30:19.951 --> 00:30:28.935
But the opportunity to really understand not only what we're learning today, but how it works and what the history of how we got here, you're not going to get it anywhere else.

516
00:30:29.495 --> 00:30:32.916
And this is why we love doing this show at From First Principles.

517
00:30:33.236 --> 00:30:39.539
And it is hugely helpful to us to continue to bring you the best and greatest breaking science research.

518
00:30:40.479 --> 00:30:56.995
As we, as I always say, fight the billionaire algorithms, a like, a share, a comment, a DM, sending it to a friend, bringing it into Journal Club is hugely, hugely helpful for us where it's, again, still just the two of us running the show to be able to give this to you multiple times every week.

519
00:30:57.435 --> 00:31:04.224
If you would like to become a patron, you can donate to the pod at ffppod.com backslash donate.

520
00:31:04.624 --> 00:31:12.434
Any and all support monetarily or non-monetarily is hugely, hugely valuable, and we really greatly appreciate it.

521
00:31:12.914 --> 00:31:16.839
I have no other show notes this week, and I'm very excited to dive back into this paper.

522
00:31:17.039 --> 00:31:17.500
Let's do it.

523
00:31:17.780 --> 00:31:18.040
Okay.

524
00:31:18.421 --> 00:31:21.304
So first thing I want to do, right?

525
00:31:21.324 --> 00:31:28.812
I want to use artificial intelligence to make a discovery engine for my brain, figure out why comas happen, and then propose treatment.

526
00:31:29.192 --> 00:31:30.974
That's the end goal of this paper, okay?

527
00:31:31.214 --> 00:31:31.354
Okay.

528
00:31:31.394 --> 00:31:38.502
First thing I want to do is train an AI detective that is capable of diagnosing consciousness from raw brainwaves.

529
00:31:38.622 --> 00:31:38.882
Okay.

530
00:31:38.902 --> 00:31:39.102
Okay?

531
00:31:39.783 --> 00:31:48.168
Just can I identify consciousness if you give me the, you know, sort of brain activity, the electrical activity of a brain?

532
00:31:48.608 --> 00:31:51.650
So he made something called a deep convolutional neural network.

533
00:31:51.690 --> 00:31:55.051
These are just our normal, you know, CNNs that we hear about.

534
00:31:55.452 --> 00:32:00.154
Giant network of artificial neurons that is trained through backpropagation.

535
00:32:00.454 --> 00:32:04.397
You know, supervised learning, which is just we've got about 680,000 10-second neurons.

536
00:32:07.138 --> 00:32:09.019
electrophysiological recordings.

537
00:32:09.179 --> 00:32:13.721
So 680,000 recordings of brains, electrical activity.

538
00:32:14.281 --> 00:32:20.704
These recordings come from humans, monkeys, rats, and bats under anesthesia, in a coma, and wakeful.

539
00:32:20.984 --> 00:32:23.965
So a swath all across mammals.

540
00:32:24.346 --> 00:32:27.327
So that means it's not going to really hone in on some human feature.

541
00:32:27.487 --> 00:32:31.789
It's going to really start thinking about what is the nature of consciousness itself.

542
00:32:32.649 --> 00:32:32.989
Right?

543
00:32:33.610 --> 00:32:35.590
And he made three discriminators.

544
00:32:37.405 --> 00:32:40.546
neural network that was going to get trained on the cortical data.

545
00:32:40.986 --> 00:32:43.607
There's a thalamic and then there's a palladial.

546
00:32:43.647 --> 00:32:49.649
That's the three sort of big brain regions that we were talking about in that mesocircuit model, right?

547
00:32:49.889 --> 00:32:55.691
The core, the cortex is your grid with all the homes and the podcast studios.

548
00:32:56.051 --> 00:33:04.833
The thalamus is the relay station and the palladial station is kind of like governing what's going on between these two.

549
00:33:05.033 --> 00:33:05.254
Right?

550
00:33:05.554 --> 00:33:05.734
Yes.

551
00:33:05.754 --> 00:33:06.334
Okay.

552
00:33:06.354 --> 00:33:06.554
Okay.

553
00:33:08.040 --> 00:33:10.121
Once he trains it, he validates it.

554
00:33:10.842 --> 00:33:11.362
It's pretty good.

555
00:33:11.802 --> 00:33:16.925
The AI score for consciousness, which is plotted on the y-axis, is...

556
00:33:17.998 --> 00:33:24.241
is correlated with the ground truth score of consciousness that it was given to each sample in the test sample.

557
00:33:24.541 --> 00:33:24.781
Okay?

558
00:33:24.981 --> 00:33:36.065
Which means that, like, the AI, this deep convolutional neural network, is given a sample that it's never seen before, and it's told to guess how conscious it is based on a number between 0 and 1.

559
00:33:36.365 --> 00:33:38.406
That number is correlated with the ground truth.

560
00:33:38.586 --> 00:33:39.667
Okay, so that means it's working.

561
00:33:39.727 --> 00:33:39.887
Yes.

562
00:33:40.787 --> 00:33:44.290
The other big one that I quite liked is from figure 1J, if you see over there.

563
00:33:44.850 --> 00:33:53.036
What that shows is that the AI was able to classify fully paralyzed ALS patients as conscious.

564
00:33:53.416 --> 00:33:57.719
Remember in ALS, we did the story where it's a motor neuron disease, right?

565
00:33:57.779 --> 00:33:59.240
All of your motor neurons go away.

566
00:33:59.480 --> 00:34:02.142
So you can still be conscious, but you won't be able to...

567
00:34:03.683 --> 00:34:04.944
fully express that.

568
00:34:05.885 --> 00:34:15.011
Here, they looked at EFIS recordings from ALS patients, and there's no significant difference between them and healthy cortical activity.

569
00:34:15.031 --> 00:34:22.897
This is actually a fascinating follow-up to our ALS story because we did a great deep dive on understanding what it actually is mechanistically.

570
00:34:22.937 --> 00:34:23.017
Yes.

571
00:34:23.057 --> 00:34:23.277
Yes.

572
00:34:24.678 --> 00:34:29.141
And it dovetails with what we said earlier, which is just because your motor functions go away.

573
00:34:29.161 --> 00:34:32.082
ALS is a perfect example of what we were just talking about.

574
00:34:32.463 --> 00:34:35.365
It doesn't necessarily mean that there's no one home.

575
00:34:35.385 --> 00:34:35.985
Yeah, yeah.

576
00:34:36.425 --> 00:34:37.366
That's very – Exactly.

577
00:34:37.406 --> 00:34:38.847
And this was able to find that, right?

578
00:34:38.887 --> 00:34:41.368
Without motor output, you can now classify.

579
00:34:41.428 --> 00:34:42.049
That's it.

580
00:34:42.069 --> 00:34:42.969
So already big win.

581
00:34:43.150 --> 00:34:43.550
Figure one.

582
00:34:43.570 --> 00:34:43.650
Yes.

583
00:34:43.870 --> 00:34:45.011
Yes, that is huge.

584
00:34:45.091 --> 00:34:45.291
Right?

585
00:34:45.651 --> 00:34:45.932
That's good.

586
00:34:45.952 --> 00:34:46.332
That is huge.

587
00:34:46.532 --> 00:34:46.812
Okay.

588
00:34:47.052 --> 00:34:49.294
Now next, we've made a detector.

589
00:34:49.694 --> 00:34:51.295
We've made a detector of consciousness.

590
00:34:51.435 --> 00:34:51.636
Yes.

591
00:34:51.716 --> 00:34:58.721
Now that we have a consciousness detector, let's make a brain simulator that can make my own brain signals.

592
00:34:58.981 --> 00:34:59.261
Okay.

593
00:34:59.361 --> 00:34:59.601
Okay?

594
00:34:59.902 --> 00:35:04.105
And this is where we get into the real why there's AI in the title.

595
00:35:04.365 --> 00:35:04.625
Okay?

596
00:35:04.805 --> 00:35:05.086
Okay?

597
00:35:05.366 --> 00:35:11.531
Because just making a deep convolutional neural network that will classify, that's just an autonomous classifier.

598
00:35:11.551 --> 00:35:12.011
Right, right.

599
00:35:12.031 --> 00:35:13.072
That's like 10 years old.

600
00:35:13.092 --> 00:35:13.753
What are you doing?

601
00:35:13.793 --> 00:35:14.093
Right?

602
00:35:14.233 --> 00:35:16.595
You're not going to get into nature neuroscience like that.

603
00:35:16.755 --> 00:35:17.836
And Daniel knows that.

604
00:35:18.076 --> 00:35:19.597
And this is for the social clips.

605
00:35:19.638 --> 00:35:22.400
If we end up clipping the first half without the second half...

606
00:35:23.371 --> 00:35:25.573
Please watch the pod so you get the full story.

607
00:35:25.613 --> 00:35:27.314
Yeah, so they're not like, that's not what I mean.

608
00:35:27.334 --> 00:35:28.915
I can't believe that's a nation you're not saying.

609
00:35:29.436 --> 00:35:30.457
Yes, we know.

610
00:35:30.557 --> 00:35:33.359
We can't fit an hour-long thing onto Instagram.

611
00:35:33.519 --> 00:35:33.739
Yeah.

612
00:35:34.100 --> 00:35:37.402
Okay, so now what he's going to do is make a brain simulator.

613
00:35:37.422 --> 00:35:37.803
Mm-hmm.

614
00:35:38.483 --> 00:35:46.569
He's going to combine that discriminator that he got, the AI, the consciousness detector, with something called a generative adversarial network.

615
00:35:46.629 --> 00:35:47.470
These are GANs.

616
00:35:47.870 --> 00:35:53.294
Traditionally with GANs, I mean, they're kind of used to make, like, generative AI, like, photographs and things like that.

617
00:35:53.594 --> 00:35:54.235
Here's how it works.

618
00:35:54.255 --> 00:35:58.678
So this is very different from, like, a diffusion model, which is usually kind of used in the zeitgeist.

619
00:35:58.718 --> 00:36:00.800
GANs kind of didn't have popularity.

620
00:36:00.840 --> 00:36:02.141
They had popularity for a while.

621
00:36:02.481 --> 00:36:06.764
And then diffusion kind of went up and became the generative network.

622
00:36:06.824 --> 00:36:07.665
sort of paradigm.

623
00:36:07.785 --> 00:36:09.928
But then now GANs are kind of back up.

624
00:36:10.249 --> 00:36:12.371
We don't really know which architecture is going to win in the end.

625
00:36:12.612 --> 00:36:13.453
But here's the idea.

626
00:36:13.713 --> 00:36:13.954
Okay?

627
00:36:14.374 --> 00:36:20.102
So you've got something called a forger, which is trying to forge...

628
00:36:21.282 --> 00:36:22.843
real life data.

629
00:36:22.963 --> 00:36:25.944
For example, in this case, what we're trying to do is create real faces.

630
00:36:26.304 --> 00:36:33.267
So we're going to have a data set of real faces, and then we're going to have a forger, that's the generator over there, and that's going to generate faces.

631
00:36:34.567 --> 00:36:36.768
And then we're going to have a discriminator,

632
00:36:37.428 --> 00:36:44.495
On the other side, that is going to take the real data and going to take the forged data and try to figure out which one is forged and which one is real.

633
00:36:44.595 --> 00:36:46.297
Now, in the beginning, it's going to be obvious.

634
00:36:46.377 --> 00:36:47.158
In the Beninging.

635
00:36:47.198 --> 00:36:47.398
Yeah.

636
00:36:47.638 --> 00:36:50.081
In the beginning, because the forger doesn't have any training.

637
00:36:50.101 --> 00:36:52.043
He's just going to be making up random nonsense.

638
00:36:52.083 --> 00:36:52.283
Right.

639
00:36:52.323 --> 00:36:55.806
So the discriminator is going to be able to look and be like, oh, yeah, this is real.

640
00:36:55.827 --> 00:36:56.387
This is fake.

641
00:36:56.687 --> 00:36:59.050
But as the forger gets better and better, because...

642
00:36:59.450 --> 00:37:07.653
That output of, hey, this is fake, you keep making fake stuff, I want the real stuff, that output is getting back propagated through to the forger.

643
00:37:07.673 --> 00:37:09.714
And the forger is thinking, okay, how do I get better?

644
00:37:09.794 --> 00:37:10.514
How do I get better?

645
00:37:10.914 --> 00:37:19.237
Pretty soon after multiple rounds of training, the forger is going to get good at making the real data such that it fools...

646
00:37:19.677 --> 00:37:20.477
You're a discriminator.

647
00:37:20.537 --> 00:37:20.738
Yes.

648
00:37:20.938 --> 00:37:21.198
Okay?

649
00:37:21.438 --> 00:37:26.540
That's the generative part, and the adversarial part is you've got two networks that are adversaries.

650
00:37:26.580 --> 00:37:31.762
There's a detective that's trying to detect the fakes, and then there's a forger that's trying to create even better fakes.

651
00:37:32.123 --> 00:37:33.583
That's the adversarial part, right?

652
00:37:34.023 --> 00:37:44.268
For anyone who's ever used image generation, particularly mid-journey, this structure is exactly the reason why you'll see a blurry, weird thing happening.

653
00:37:44.328 --> 00:37:56.239
when you first initiate the image generation, and then it's going through the cycle, you just talked about it, the fidelity gets closer and closer and closer over until it gives you the final, like produced like image.

654
00:37:56.880 --> 00:37:59.422
This is like the whole stable diffusion kind of came

655
00:37:59.782 --> 00:38:02.044
Yeah, and that one's slightly different.

656
00:38:02.064 --> 00:38:13.572
Slightly different because – so what you're talking about is very, very close to generative adversarial networks because at the end of the day, how the forger creates the fake stuff is it starts with some kind of random noise, which is what you see there.

657
00:38:13.932 --> 00:38:18.115
And then it starts manipulating that noise to create the real thing.

658
00:38:18.835 --> 00:38:25.297
So what you're seeing there is the act of the forger, in some sense, going from noise to the real thing.

659
00:38:25.577 --> 00:38:29.538
But the training part is something that's already been done by mid-journey on the backend.

660
00:38:29.598 --> 00:38:29.778
100%.

661
00:38:29.858 --> 00:38:33.799
And so that's the generative part.

662
00:38:34.119 --> 00:38:34.299
Yes.

663
00:38:34.319 --> 00:38:34.980
Okay, great.

664
00:38:35.080 --> 00:38:35.560
So now...

665
00:38:37.842 --> 00:38:38.062
Great.

666
00:38:38.082 --> 00:38:41.325
That's how you make a fake Picasso and a real Picasso and things like that.

667
00:38:41.586 --> 00:38:43.447
How are we going to use that to figure out coma?

668
00:38:43.908 --> 00:38:44.168
Yes.

669
00:38:44.188 --> 00:38:45.269
All right.

670
00:38:45.309 --> 00:38:53.817
This is where Dan comes in hot with an interpretable biophysically grounded mean field model.

671
00:38:54.057 --> 00:38:54.357
Okay.

672
00:38:54.457 --> 00:38:55.358
And that is what is cool.

673
00:38:55.398 --> 00:38:57.821
So on the upper left, that's A, figure A.

674
00:38:58.041 --> 00:38:58.281
Yes.

675
00:38:58.301 --> 00:38:58.481
Okay.

676
00:38:58.881 --> 00:39:02.925
On the left-hand side, we've got our DCNN, the convolutional neural network.

677
00:39:02.945 --> 00:39:03.746
That's our detector.

678
00:39:04.066 --> 00:39:06.548
that we had trained previously in figure one, right?

679
00:39:06.788 --> 00:39:12.192
There's other detectors in here that'll actually figure out, is this real or fake data?

680
00:39:12.452 --> 00:39:15.414
That one is saying, is this unconscious or conscious data?

681
00:39:15.534 --> 00:39:17.456
There's other ones that are not in this backend.

682
00:39:17.476 --> 00:39:21.579
You have to go to the supplement to see it, where it's like, is this real data or fake data?

683
00:39:21.599 --> 00:39:22.880
Because first we want to just create...

684
00:39:23.360 --> 00:39:24.161
Real looking data.

685
00:39:24.221 --> 00:39:28.224
Then we'll worry about, okay, am I creating conscious data or unconscious data, right?

686
00:39:28.464 --> 00:39:30.266
So there's multiple steps to this process.

687
00:39:30.326 --> 00:39:34.929
But the key is that the forger is not a black box, okay?

688
00:39:35.230 --> 00:39:41.775
The forger is actually a three component within which there's multiple components.

689
00:39:41.815 --> 00:39:42.015
Yes.

690
00:39:42.055 --> 00:39:49.122
But you know that mesocircuit hypothesis of like there's a cortex, there's the basal ganglia, there's the thalamus, and these guys are talking to one another?

691
00:39:49.583 --> 00:39:53.027
Each of those boxes is its own little neural network, right?

692
00:39:53.547 --> 00:40:01.095
And it's a biophysically realistic neural network in the sense that it's not just a bunch of artificial neurons that start at noise.

693
00:40:01.576 --> 00:40:02.496
What it's doing is...

694
00:40:03.461 --> 00:40:06.223
It's saying, okay, how does a biophysical neural network work?

695
00:40:06.523 --> 00:40:15.210
Well, some input comes in, then there's going to be like channels, like, you know, some neurotransmitter is going to have some characteristic timescale.

696
00:40:15.510 --> 00:40:18.392
Other neurotransmitters are going to have other characteristic timescales.

697
00:40:18.652 --> 00:40:22.015
There's going to be excitatory neurons, which do positive feedback.

698
00:40:22.295 --> 00:40:26.038
There's going to be negative inhibitory neurons that do negative feedback.

699
00:40:26.638 --> 00:40:34.187
These I can hook up mathematically using differential equations, and I can create a biophysically realistic model.

700
00:40:34.507 --> 00:40:36.309
It is no longer a black box.

701
00:40:36.870 --> 00:40:46.401
And instead of using backpropagation to change the weights between each neuron, what I'm going to do is use something called a genetic algorithm to...

702
00:40:46.601 --> 00:40:53.569
to change the parameters of my biophysically realistic model, meaning how many excitatory neurons are there?

703
00:40:53.809 --> 00:40:55.291
How many inhibitory neurons are there?

704
00:40:55.351 --> 00:40:57.073
How is the connectivity, right?

705
00:40:57.333 --> 00:40:58.515
What is the timescale?

706
00:40:59.938 --> 00:41:25.870
of the interaction things like that things that are actually relevant when we think about what is a brain doing and is in some sense the reason why it made it's like the that algorithm is necessarily bounded by what the realistic value ranges are exactly and that's why it's not a black box because you know that oh it's either going to be excitatory neurons are either going to go between x and y yeah and that's a realistic normalized range

707
00:41:25.930 --> 00:41:26.370
Exactly.

708
00:41:26.410 --> 00:41:27.050
That's one of them.

709
00:41:27.110 --> 00:41:32.393
And the other one is we can literally point to components of the model and be like, oh, it's the excitation and the thalamus.

710
00:41:32.853 --> 00:41:33.313
Ah, yeah, yeah.

711
00:41:33.333 --> 00:41:35.134
That's right.

712
00:41:35.194 --> 00:41:37.014
I can literally be like, oh, it's this part.

713
00:41:37.235 --> 00:41:37.395
Yeah.

714
00:41:37.615 --> 00:41:41.876
With again, it's like with normal neural networks, I don't have no idea.

715
00:41:41.896 --> 00:41:42.357
Yeah.

716
00:41:42.877 --> 00:41:46.138
Like, you know, I type in make me a cat on a horse.

717
00:41:46.578 --> 00:41:50.480
I have no idea in the trillions of parameters where it decided.

718
00:41:51.462 --> 00:41:55.788
cat, where it decided horse, how it figured out to put the cat on top of the horse.

719
00:41:56.109 --> 00:41:59.033
Here, literally, I could look into the parameters.

720
00:41:59.293 --> 00:42:02.978
I can see how they evolve with training and figure out what is actually going on.

721
00:42:03.299 --> 00:42:06.083
And then make that extrapolation from there to the brain.

722
00:42:07.403 --> 00:42:08.324
That's very good, I think.

723
00:42:08.364 --> 00:42:09.044
That's very good.

724
00:42:09.104 --> 00:42:09.304
Right?

725
00:42:09.404 --> 00:42:09.965
Yeah.

726
00:42:10.125 --> 00:42:12.186
And that's one of the things I like a lot.

727
00:42:12.246 --> 00:42:12.626
Yeah.

728
00:42:12.666 --> 00:42:12.867
Right?

729
00:42:13.147 --> 00:42:18.750
This loss function, he had a loss function that sort of, like, figures out, right, how to change the parameters.

730
00:42:19.511 --> 00:42:25.855
And this loss function incorporates the outputs from the networks that's trained to classify real versus synthetic data.

731
00:42:26.195 --> 00:42:29.838
It also has outputs from the consciousness detector that we had done earlier.

732
00:42:30.198 --> 00:42:32.319
It's also got a way to identify seizures.

733
00:42:32.339 --> 00:42:33.080
Yeah.

734
00:42:33.460 --> 00:42:38.805
And as you said, it's got these empirical constraints on the firing rates of these neurons, as you said, like, right?

735
00:42:39.666 --> 00:42:41.147
Excitation can't be that high.

736
00:42:41.287 --> 00:42:44.670
Inhibitory neurons are usually higher firing rate than excitation.

737
00:42:44.930 --> 00:42:45.851
Things like that, right?

738
00:42:45.931 --> 00:42:47.773
And the interregional communication patterns.

739
00:42:47.873 --> 00:42:48.133
Right.

740
00:42:48.313 --> 00:42:50.035
Like excitation is long range.

741
00:42:50.335 --> 00:42:51.976
Inhibition is usually short range.

742
00:42:52.196 --> 00:42:53.998
This is stuff you can bake into the model.

743
00:42:54.298 --> 00:42:54.979
This is quite nice.

744
00:42:55.359 --> 00:43:22.799
This is quite nice I love the fact that it's not black box Because that's one of the things that I hate about Like just large neural networks I have no idea where anything is happening And it's also so different from the culture of software Since its beginning Which has been that everything is explicit And you can literally point to Where exactly, like the trace callback Of where something is coming from This is the going back to that type Which is great This is like, yeah, fantastic Fantastic

745
00:43:23.919 --> 00:43:34.865
And so once this genetic algorithm has trained my brain maker, it's an artificial brain, now we can ask, okay, how about we reprogram the objective now?

746
00:43:35.165 --> 00:43:35.445
Okay.

747
00:43:35.485 --> 00:43:37.306
It was just making real world data.

748
00:43:37.806 --> 00:43:41.408
Now I reprogram the objective to make comatose data.

749
00:43:41.668 --> 00:43:45.590
So I can use that first neural network to be like, give me coma.

750
00:43:45.610 --> 00:43:45.750
Right.

751
00:43:46.050 --> 00:43:46.230
Right.

752
00:43:46.970 --> 00:43:47.191
Yes.

753
00:43:47.291 --> 00:43:47.471
Right.

754
00:43:47.571 --> 00:43:52.835
And then now that genetic algorithm is going to tweak the parameters such that I get coma type data.

755
00:43:52.895 --> 00:43:53.155
Yes.

756
00:43:53.615 --> 00:44:03.002
And just I just want to bring this back in because we have done the training and done the validation based off of a ground truth, real source of actual information.

757
00:44:03.022 --> 00:44:03.162
Yeah.

758
00:44:03.202 --> 00:44:05.224
We have all of those 10 second long clips.

759
00:44:05.384 --> 00:44:06.204
Right.

760
00:44:06.344 --> 00:44:11.168
That's like I just it's not making this from whole cloth.

761
00:44:11.268 --> 00:44:11.628
No.

762
00:44:12.128 --> 00:44:13.789
That's a really important concept.

763
00:44:13.829 --> 00:44:15.930
Yeah, and there's multiple stages of training.

764
00:44:16.130 --> 00:44:16.890
Right.

765
00:44:16.970 --> 00:44:22.313
He's figured out a way to, you know, you can't just tell a biophysical model, make me coma.

766
00:44:22.473 --> 00:44:22.693
Right.

767
00:44:22.873 --> 00:44:23.113
Right.

768
00:44:23.293 --> 00:44:27.055
You got to first be like, no, no, no, let's make just normal data.

769
00:44:27.195 --> 00:44:33.918
Give me data that can fool even me to thinking, I don't know if that's from a real patient or from my brain simulation.

770
00:44:34.198 --> 00:44:34.358
Right?

771
00:44:34.378 --> 00:44:34.538
Right.

772
00:44:35.098 --> 00:44:36.878
I'm sure he did like checks just on his own.

773
00:44:37.018 --> 00:44:37.198
Yes.

774
00:44:37.278 --> 00:44:40.199
Where like after the whole thing, I mean, you look at it, it's like, okay, that looks pretty good.

775
00:44:40.219 --> 00:44:40.359
Yes.

776
00:44:40.499 --> 00:44:40.679
Right.

777
00:44:40.719 --> 00:44:43.079
And then you go, okay, now give me a coma.

778
00:44:43.459 --> 00:44:46.060
And he was successfully able to recreate known phenotypes.

779
00:44:46.080 --> 00:44:49.100
Like on the right-hand side, we've got high voltage delta oscillations.

780
00:44:49.140 --> 00:44:50.561
This is something that happens in coma.

781
00:44:50.841 --> 00:44:53.721
Like the slow sort of oscillations that are super high voltage.

782
00:44:54.001 --> 00:44:56.802
We've got burst suppression that's on the bottom side.

783
00:44:56.942 --> 00:44:57.402
Yeah.

784
00:44:57.682 --> 00:45:00.442
Big bursts and then just seconds of inactivity.

785
00:45:00.762 --> 00:45:03.303
Again, something that's there in coma patients.

786
00:45:03.643 --> 00:45:04.243
What's cool is,

787
00:45:04.943 --> 00:45:15.654
There's no explicit programming of prior neuroanatomical knowledge on what is the difference between a coma versus a normal brain.

788
00:45:15.794 --> 00:45:16.635
Okay?

789
00:45:16.775 --> 00:45:22.381
Just the genetic algorithm and that detector that I had of coma versus conscious.

790
00:45:22.441 --> 00:45:22.721
Yes.

791
00:45:23.762 --> 00:45:32.185
Just that and the genetic algorithm has now independently discovered some of the core tenets of that mesocircuit hypothesis, right?

792
00:45:32.345 --> 00:45:33.985
What did the mesocircuit hypothesis said?

793
00:45:34.165 --> 00:45:40.767
It said that if I weaken cortical drive coming in from the cortex, that's going to cause a bunch of random crap.

794
00:45:41.067 --> 00:45:44.768
That's what my model is doing.

795
00:45:45.068 --> 00:45:48.670
So as a first pass, it's already kind of on the right track.

796
00:45:48.690 --> 00:45:49.510
You can smell...

797
00:45:50.270 --> 00:46:16.137
it in the water like there's i'm on the right track because right the normal stuff that i kind of know about the mesocircuit that i hadn't baked in explicitly it's already figuring that out it is in it it is being it is able to intuit uh what we've already done over this last five six hundred years that we talked about um out of the box without this is sort of what people say about benchmarks and ai models it's like

798
00:46:16.177 --> 00:46:19.038
well, if you train it on the benchmarks, then it's going to do good on the benchmarks.

799
00:46:19.398 --> 00:46:26.141
And so the point here, it's not, the answers to the test were not provided as a part of training.

800
00:46:26.481 --> 00:46:33.424
And so it's getting answers that are not tainted by sort of giving them the cheat sheet beforehand.

801
00:46:33.444 --> 00:46:33.704
Yeah, yeah.

802
00:46:33.704 --> 00:46:41.588
Yeah, it's just looking at how do I make a coma and already it's recreating parts of the mesocircuit hypothesis that we know are probably true.

803
00:46:41.668 --> 00:46:41.928
Right.

804
00:46:42.088 --> 00:46:42.329
Okay.

805
00:46:42.629 --> 00:46:42.829
Right.

806
00:46:42.869 --> 00:46:44.490
So that's already impressive.

807
00:46:44.570 --> 00:46:47.791
The next part is the impressive part, which is what is the new stuff?

808
00:46:48.172 --> 00:46:50.693
I just want to say this is all already very impressive.

809
00:46:50.893 --> 00:46:53.695
And just as another caveat, we're talking about Daniel specifically.

810
00:46:53.715 --> 00:46:56.476
This was obviously a team and this is incredible work by everybody on the team.

811
00:46:56.616 --> 00:47:20.962
yeah yeah yeah everybody is our friend so that's why we're talking about daniel yeah all right all right so but clearly yeah this is this is very i mean there takes a lot of data it takes a lot of like it takes a lot of brain it's always a team effort guys every time we're talking that's always a team yeah yeah so let's get back to daniel yeah okay um here's here's what he's gonna do next okay okay so

812
00:47:22.467 --> 00:47:37.980
The one part of the mesocircuit hypothesis, which is like cortical drive goes down, so the normal power grid is down, or there's elevated firing in the palladial population, so that's going to inhibit the thalamus even more and things like that.

813
00:47:38.260 --> 00:47:39.781
That stuff is already corroborated.

814
00:47:40.022 --> 00:47:42.404
Stuff we already knew, this thing is kind of confirming.

815
00:47:42.484 --> 00:47:42.964
Okay.

816
00:47:43.424 --> 00:47:43.604
Now—

817
00:47:44.305 --> 00:47:45.226
Does it do anything new?

818
00:47:45.426 --> 00:47:46.247
Well, yes.

819
00:47:46.768 --> 00:47:50.732
There's something called the selective disruption of the indirect pathway.

820
00:47:50.772 --> 00:47:54.536
This is a new prediction from the model.

821
00:47:54.776 --> 00:47:55.437
Here's what's happening.

822
00:47:55.817 --> 00:48:03.225
So it turns out the prediction is that the coma is actually driven by the degradation of the...

823
00:48:04.378 --> 00:48:05.979
Certain medium spiny neurons.

824
00:48:06.039 --> 00:48:08.180
Remember those neurons that I was telling you about?

825
00:48:08.300 --> 00:48:10.581
They just die if there's any stress whatsoever.

826
00:48:10.641 --> 00:48:14.503
Or a little bit of loxia deprivation immediately.

827
00:48:14.523 --> 00:48:16.504
They're just like, okay, I'm done.

828
00:48:17.645 --> 00:48:28.170
There's selective degradation of a certain type of medium spiny neuron in the striatum that is projecting to the part that's putting on brakes to the thalamus.

829
00:48:28.770 --> 00:48:32.212
And this very specific pathway is

830
00:48:33.350 --> 00:48:33.730
is new.

831
00:48:34.151 --> 00:48:34.411
Okay?

832
00:48:34.431 --> 00:48:39.674
The classical mesocircuit hypothesis is just like, oh, there's like striatal dysfunction.

833
00:48:40.075 --> 00:48:40.255
Right?

834
00:48:40.355 --> 00:48:42.196
It's like the whole brain region.

835
00:48:42.256 --> 00:48:42.556
Yeah, yeah.

836
00:48:42.616 --> 00:48:42.796
Right?

837
00:48:43.757 --> 00:48:55.505
This thing is saying that the indirect pathway, which is specifically this D2 receptor expressing medium spiny neuron, this very specific subpopulation in the striatum, that's the part

838
00:48:56.125 --> 00:48:56.946
That is getting weakened.

839
00:48:57.867 --> 00:49:02.531
And when that gets weakened, that's going to ultimately suppress the thalamus at the end of the day.

840
00:49:02.951 --> 00:49:04.132
Like that thing that I was telling you about.

841
00:49:04.152 --> 00:49:06.434
But now it's identified a little part.

842
00:49:06.494 --> 00:49:08.355
It's not just all of L.A. County.

843
00:49:08.395 --> 00:49:09.757
We've got it down to a street block.

844
00:49:10.117 --> 00:49:12.419
Yeah, and a specific type of house type thing.

845
00:49:12.479 --> 00:49:12.659
Right.

846
00:49:12.739 --> 00:49:12.939
Right?

847
00:49:13.380 --> 00:49:15.942
Okay, so how do you... Fine, that's your prediction.

848
00:49:16.683 --> 00:49:19.125
Predictions are only as good as the data that...

849
00:49:19.685 --> 00:49:20.186
Corroborate.

850
00:49:20.426 --> 00:49:20.626
Right.

851
00:49:20.706 --> 00:49:21.727
So how do we corroborate it?

852
00:49:21.767 --> 00:49:23.829
We use something called diffusion tensor imaging.

853
00:49:24.189 --> 00:49:26.651
So this is an advanced form of, like, MRI.

854
00:49:26.751 --> 00:49:27.732
Oh, this is beautiful.

855
00:49:28.252 --> 00:49:29.914
What we can look at is trace.

856
00:49:30.094 --> 00:49:30.775
This is very cool.

857
00:49:30.795 --> 00:49:42.064
What you can do with MRI is magnetic resonance imaging, right, where you look at, like, hydrogen atoms in whatever biological tissue, and you can actually...

858
00:49:43.045 --> 00:49:45.646
See that using this magnetic resonance imaging technique.

859
00:49:45.666 --> 00:49:50.108
I'm not going to get into how MRI works, but effectively we can trace hydrogen atoms.

860
00:49:50.188 --> 00:49:51.469
Okay.

861
00:49:51.509 --> 00:49:52.889
Water has a bunch of hydrogen atoms.

862
00:49:53.249 --> 00:49:53.489
Okay.

863
00:49:53.730 --> 00:50:01.473
So we can measure the directional diffusion rate of water molecules along the axons, along those fibers.

864
00:50:01.673 --> 00:50:01.833
Mm-hmm.

865
00:50:02.513 --> 00:50:19.620
Those neurons right and we can measure in white matter the water is going to move faster along the length of the axon Then across and so we have a directional idea of like how Diffusion of water is going and diffusion of water is a proxy for kind of how neurons talk to one another Where is it?

866
00:50:19.660 --> 00:50:20.940
How's the traffic and flowing right?

867
00:50:20.980 --> 00:50:21.720
Yeah, exactly.

868
00:50:22.041 --> 00:50:22.701
So what they did was

869
00:50:23.321 --> 00:50:30.224
was use this DTI, diffusion tensor imaging, of 51 patients with disorders of consciousness.

870
00:50:30.824 --> 00:50:37.027
And you can show that there's significantly lower striatum to GPE streamlines.

871
00:50:37.127 --> 00:50:39.048
That's the break on the thalamus.

872
00:50:39.188 --> 00:50:45.810
That prediction is lower in vegetative state versus minimal consciousness patients.

873
00:50:45.850 --> 00:50:47.011
No way!

874
00:50:47.051 --> 00:50:47.291
Yeah.

875
00:50:47.751 --> 00:50:49.232
So just to take a step back, right?

876
00:50:50.292 --> 00:50:51.333
From this model...

877
00:50:52.153 --> 00:51:07.678
it made a prediction about where specifically in this region of the brain that is controlling the flow between the thalamus and the cortex is the point where the degradation of this specific area is what's actually driving the problem.

878
00:51:07.718 --> 00:51:11.199
And this connection is where we would find it.

879
00:51:11.299 --> 00:51:15.440
And so it said this is the road and the house where the problems are.

880
00:51:16.360 --> 00:51:17.101
Arise out of.

881
00:51:17.642 --> 00:51:23.008
And then we said, okay, well, we have all this actual MRI data from real patients.

882
00:51:23.769 --> 00:51:34.782
So can we look at this road, this house and real patients that the model predicted you would see degradation in an outsized way as compared to surrounding area?

883
00:51:35.403 --> 00:51:36.024
Yeah.

884
00:51:36.064 --> 00:51:39.366
I mean, no, they compared vegetative versus minimally cognitive.

885
00:51:39.606 --> 00:51:39.986
Right, right.

886
00:51:40.046 --> 00:51:41.247
Because we've got those two sets.

887
00:51:41.487 --> 00:51:41.747
Yes.

888
00:51:41.967 --> 00:51:44.529
And if there's a difference, there should be some statistically...

889
00:51:44.729 --> 00:51:46.110
Some delta between those two.

890
00:51:46.430 --> 00:51:51.914
The vegetative will have more of the degradation than the minimally... Oh, my God.

891
00:51:51.954 --> 00:51:52.894
Yeah, yeah.

892
00:51:52.914 --> 00:51:56.977
And so on the left-hand side, we see that significance.

893
00:51:57.097 --> 00:51:58.238
There is a significant difference.

894
00:51:58.598 --> 00:52:00.319
On the right-hand side, he did sneak the sin.

895
00:52:00.599 --> 00:52:01.880
The p-value is only 0.07.

896
00:52:01.900 --> 00:52:04.062
So it's not significant.

897
00:52:04.082 --> 00:52:04.222
Yeah.

898
00:52:05.363 --> 00:52:05.823
But it's there.

899
00:52:06.124 --> 00:52:09.247
And perhaps if you were to pull the two data, it would become even more significant.

900
00:52:09.527 --> 00:52:13.311
But there's at least something.

901
00:52:13.371 --> 00:52:14.272
And the N is low.

902
00:52:15.033 --> 00:52:21.840
So obviously significance is not... We're not going to get Higgs boson 10 to the minus 5 significance here.

903
00:52:21.900 --> 00:52:25.744
Where literally to do that, they had billions of...

904
00:52:26.525 --> 00:52:28.445
particles colliding with each other, right?

905
00:52:28.465 --> 00:52:36.407
So you'll never get the significance level of traditional physics, but the effect, it seems, is there, right?

906
00:52:36.647 --> 00:52:39.188
And with more and more data, perhaps it's going to be an even bigger effect.

907
00:52:39.288 --> 00:52:40.408
Makes total sense.

908
00:52:40.488 --> 00:52:48.030
It was really kind of a proof of, like, minimally viable proof that this hypothesis of this little...

909
00:52:49.170 --> 00:52:51.631
Prediction that I have, it's probably true.

910
00:52:51.711 --> 00:52:53.051
It was good enough for nature neuroscience.

911
00:52:53.091 --> 00:52:53.331
Yes.

912
00:52:53.511 --> 00:52:53.891
Not bad.

913
00:52:54.091 --> 00:52:55.072
Exactly.

914
00:52:55.112 --> 00:53:12.957
The second thing that they did was predict, this whole AI architecture predicted, that there's increased synaptic coupling between inhibitory interneurons, meaning those negative feedback neurons that I was telling you about, these are sometimes called fast spiking PV plus neurons in the cerebral cortex.

915
00:53:13.297 --> 00:53:17.238
Those negative feedback neurons in our cerebral cortex have a lot of...

916
00:53:18.727 --> 00:53:19.969
Coupling in between them.

917
00:53:20.109 --> 00:53:24.254
So it's like the negative feedback is coupled to the negative feedback, which is coupled to the negative feedback.

918
00:53:24.575 --> 00:53:24.735
Right?

919
00:53:24.775 --> 00:53:27.379
That is what the AI model is predicting.

920
00:53:28.840 --> 00:53:31.404
So to test this, you go to transcriptomics.

921
00:53:32.005 --> 00:53:34.788
Transcriptomics means I'm going to now read...

922
00:53:35.429 --> 00:53:38.992
the mRNA that is in my cerebral cortex, okay?

923
00:53:39.172 --> 00:53:47.858
And I'm going to see what genes are being expressed in patients that have vegetative state and patients that don't.

924
00:53:47.938 --> 00:53:49.619
Okay, healthy versus those in coma.

925
00:53:49.819 --> 00:53:50.039
Yes.

926
00:53:50.800 --> 00:53:56.904
What they find is there's an upregulation of two genes, the VGF and the SCG2.

927
00:53:57.504 --> 00:54:01.047
These are genes that, when expressed in these interneurons,

928
00:54:02.147 --> 00:54:08.088
They drive synaptogenesis, meaning they drive this feedback mechanism, this coupling.

929
00:54:08.108 --> 00:54:09.049
Yeah, yeah.

930
00:54:09.189 --> 00:54:09.449
That.

931
00:54:09.809 --> 00:54:10.009
Right?

932
00:54:10.229 --> 00:54:11.329
This is really...

933
00:54:11.349 --> 00:54:12.309
This one's pretty significant.

934
00:54:12.329 --> 00:54:13.690
Yeah.

935
00:54:13.730 --> 00:54:14.630
This one's way more than...

936
00:54:15.410 --> 00:54:16.230
In both those genes.

937
00:54:16.270 --> 00:54:16.950
Yeah, yeah, yeah.

938
00:54:16.970 --> 00:54:17.130
Right?

939
00:54:17.751 --> 00:54:19.931
And this is, again, data that's out there.

940
00:54:20.171 --> 00:54:21.551
Right, right, right.

941
00:54:21.671 --> 00:54:22.332
Right?

942
00:54:22.392 --> 00:54:22.812
Exactly.

943
00:54:22.832 --> 00:54:23.832
And I think the...

944
00:54:24.312 --> 00:54:27.733
This is kind of, again, in terms of like a frame of reference for...

945
00:54:28.893 --> 00:54:43.987
I've seen the conversation around quote-unquote AI be very different within some corners of the science community versus the general public because a lot of researchers view it in this way where it's like, okay, this can be a sort of intermediary layer.

946
00:54:44.827 --> 00:54:56.417
where I can rapidly prototype and generate predictions and rapidly be able to test that against real data in a way where I don't have to abuse my postdocs.

947
00:54:57.178 --> 00:54:58.819
Yeah, yeah, yeah.

948
00:54:58.960 --> 00:55:00.081
Just abuse the undergrads.

949
00:55:03.824 --> 00:55:23.777
here's a thousand images labeled right no right sorry undergrads are always gonna that's that's tough yeah it's a tough life we've all been there we all want those letters of recommendation you know you gotta you gotta work the sweat this is really interesting though because now there's the predictions are also on two very different planes

950
00:55:25.319 --> 00:55:26.880
Yes, that's a very good point.

951
00:55:26.900 --> 00:55:27.500
You know what I mean?

952
00:55:27.540 --> 00:55:33.005
Like in terms of what's... Yeah, because when we think about those three components of our city, right?

953
00:55:33.045 --> 00:55:38.009
The cerebral cortex is the homes and the businesses.

954
00:55:38.690 --> 00:55:41.692
That PV plus interneuron thing has to do with that part.

955
00:55:41.772 --> 00:55:41.972
Right.

956
00:55:42.353 --> 00:55:42.933
And then the...

957
00:55:45.315 --> 00:55:50.278
The brain imaging part, that one had to do with the thalamus and the relay stations.

958
00:55:50.298 --> 00:55:50.618
Correct.

959
00:55:50.638 --> 00:55:51.178
Right, right, right.

960
00:55:51.198 --> 00:55:54.380
So all three parts of the mesocircuit hypothesis are working together.

961
00:55:54.640 --> 00:55:57.581
And this thing is giving predictions, as you said, on all of them.

962
00:55:57.681 --> 00:55:57.962
Right.

963
00:55:58.022 --> 00:55:58.262
Right.

964
00:55:58.362 --> 00:55:58.622
Right.

965
00:55:59.242 --> 00:56:01.864
In very interesting ways.

966
00:56:01.884 --> 00:56:06.626
And again, this continues to go back to the fact that we have all these existing tools like transcriptomics.

967
00:56:06.986 --> 00:56:07.246
Mm-hmm.

968
00:56:10.648 --> 00:56:32.404
enables the ability to know what genes are being expressed yeah and one thing i just want to say is like to the team that that made this paper happen right i admire the resourcefulness right because it's one thing to have an ai model and make predictions right and it's another thing to think okay here are the predictions how can i

969
00:56:33.264 --> 00:56:34.805
make the argument that this is real.

970
00:56:34.965 --> 00:56:35.265
Right.

971
00:56:35.305 --> 00:56:36.005
How can I check?

972
00:56:36.065 --> 00:56:37.245
It's always about checking.

973
00:56:37.265 --> 00:56:37.425
Yes.

974
00:56:37.766 --> 00:56:40.927
And they were so resourceful that they found these kinds of data sets.

975
00:56:41.087 --> 00:56:43.968
You know, maybe they talked, I don't know exactly the details.

976
00:56:44.168 --> 00:56:47.449
I don't know if these are open or not, but you, you go talk to somebody who has that data set.

977
00:56:47.469 --> 00:56:47.649
Right.

978
00:56:47.689 --> 00:56:50.990
You meet someone at a conference and they're like, Hey, actually you can look into mine.

979
00:56:51.910 --> 00:57:21.032
Put me as an author You know so It's very clever I think that part is nice The whole from the ideation Of like let's give it a try To then how are we going to be able to Prove to reviewer two That this is not all nonsense Yeah yeah exactly I got lunch with Daniel about three weeks ago And he was telling me about this paper And he told me like The thing predicted this And the first thing I asked well yeah but like

980
00:57:21.293 --> 00:57:21.580
you know,

981
00:57:22.250 --> 00:57:25.592
How do you know the model isn't just bullshitting?

982
00:57:25.812 --> 00:57:27.394
And then he told me about these two techniques.

983
00:57:27.454 --> 00:57:29.255
I was like, okay, that's actually pretty dope.

984
00:57:29.415 --> 00:57:29.875
Yeah.

985
00:57:29.915 --> 00:57:32.177
Well, as soon as it's out, we'll cover it.

986
00:57:32.577 --> 00:57:32.837
You know?

987
00:57:34.418 --> 00:57:36.560
So, yeah, this was pretty cool.

988
00:57:36.600 --> 00:57:43.565
Okay, the final thing we're going to talk about is a proposed strategy for awakening patients with coma, right?

989
00:57:43.585 --> 00:57:43.745
Right.

990
00:57:43.845 --> 00:57:46.106
Because at the end of the day, that's what really matters.

991
00:57:46.186 --> 00:57:49.909
Clinically, that's how it started, and that's how we're going to end it.

992
00:57:50.289 --> 00:57:50.529
So...

993
00:57:51.650 --> 00:57:56.256
The intervention that they looked for and how to test was deep brain stimulation.

994
00:57:56.276 --> 00:58:03.866
This is, again, the idea of you test, you put in an electrode deep into the brain, and then you give it electrical activity.

995
00:58:04.818 --> 00:58:06.839
And you try to wake up the neurons, okay?

996
00:58:07.680 --> 00:58:15.744
Now, the researchers here, what they did was test a bunch of targets in their biophysical model and say, what if I provide stimulation here?

997
00:58:16.005 --> 00:58:17.425
What if I provide stimulation here?

998
00:58:17.726 --> 00:58:22.368
What's the best target in my three-component model, right?

999
00:58:22.388 --> 00:58:23.369
Because there's a bunch.

1000
00:58:23.389 --> 00:58:24.009
There's cortex.

1001
00:58:24.029 --> 00:58:24.630
There's thalamus.

1002
00:58:24.650 --> 00:58:26.351
There's subthalamic nuclei.

1003
00:58:26.391 --> 00:58:27.892
There's the palladium.

1004
00:58:28.672 --> 00:58:31.873
And each of those has their own little subpopulations.

1005
00:58:32.213 --> 00:58:35.595
So we can get really granular now, right?

1006
00:58:35.935 --> 00:58:39.436
We can get really granular and try to think what is the best target.

1007
00:58:39.756 --> 00:58:40.536
This is so good.

1008
00:58:40.956 --> 00:58:42.937
They settled on the subthalamic nucleus.

1009
00:58:43.437 --> 00:58:49.019
It had an overwhelmingly standout result for consciousness recovery in their model.

1010
00:58:49.239 --> 00:58:49.500
Okay.

1011
00:58:49.560 --> 00:58:49.800
Okay?

1012
00:58:50.200 --> 00:58:52.421
Then again, they tested this with humans.

1013
00:58:53.641 --> 00:58:59.326
They tested 130 hertz stimulation in the subthalamic nucleus in six awake human patients.

1014
00:59:00.007 --> 00:59:13.399
And the cortical CNN, the cortical consciousness detector that they had previously trained, detected a significant shift towards optimal consciousness.

1015
00:59:14.601 --> 00:59:15.282
This is crazy.

1016
00:59:15.302 --> 00:59:17.803
With only stimulation in the STN.

1017
00:59:18.104 --> 00:59:21.526
If they went anywhere else, only the sub-dilemic nucleus showed this effect.

1018
00:59:21.926 --> 00:59:27.811
If they did it in the cortex, if they did it in the palladium, didn't show any significant increase in consciousness.

1019
00:59:27.971 --> 00:59:28.231
Yes.

1020
00:59:28.391 --> 00:59:29.612
And that's what the model predicted.

1021
00:59:29.892 --> 00:59:32.453
So are you saying we found the boss's office?

1022
00:59:32.613 --> 00:59:35.575
Yeah, it seems we found... We may have found the boss's office.

1023
00:59:35.595 --> 00:59:38.116
Where if you turn the lights on and off, the boss will wake up.

1024
00:59:38.136 --> 00:59:41.298
Yeah, right, right, right.

1025
00:59:41.558 --> 00:59:41.978
This is...

1026
00:59:42.178 --> 00:59:45.400
I'm so mad because it's so clever.

1027
00:59:45.660 --> 00:59:54.645
Each of the three layers, also in how we went through it, of like detection, characterization, and evaluation almost is like, is so good.

1028
00:59:55.325 --> 00:59:56.506
Because I think...

1029
00:59:57.647 --> 01:00:18.248
The part that's getting me about this piece is because you've set up like the genomic algorithm and the model, the brain simulator, the way they thought to set it up that way means that you almost have this like alpha fold adjacent kind of thing where you can in silica like test stuff.

1030
01:00:18.348 --> 01:00:18.588
Yeah.

1031
01:00:20.230 --> 01:00:24.112
To try to narrow the sandbox of where to look.

1032
01:00:24.272 --> 01:00:24.752
Exactly.

1033
01:00:24.812 --> 01:00:25.012
Right?

1034
01:00:25.052 --> 01:00:27.293
And, like, that's so valuable.

1035
01:00:27.333 --> 01:00:27.513
Yeah.

1036
01:00:27.773 --> 01:00:28.473
Yeah, exactly.

1037
01:00:28.493 --> 01:00:30.314
It doesn't necessarily have to be the exact answer.

1038
01:00:30.534 --> 01:00:30.714
No.

1039
01:00:30.894 --> 01:00:35.696
But if it's even somewhat directionally correct, that's unbelievably valuable.

1040
01:00:35.756 --> 01:00:36.657
Yeah, dude, it's great.

1041
01:00:36.757 --> 01:00:43.326
And one thing that one thing that's kind of funny is so Daniel, he is himself a science communicator.

1042
01:00:43.406 --> 01:00:43.586
Right.

1043
01:00:43.606 --> 01:00:45.288
So he's got an Instagram and a TikTok.

1044
01:00:45.568 --> 01:00:51.976
And he did like a short three minute video being like, hey, my paper came out and he kind of explained it at a very high level.

1045
01:00:51.996 --> 01:00:53.859
I mean, here we've gone really in depth.

1046
01:00:54.219 --> 01:00:54.459
So.

1047
01:00:55.080 --> 01:01:01.742
At that high level, he was describing his generative AI framework, and there was someone in the comment who was like, that's not really AI, right?

1048
01:01:01.782 --> 01:01:03.542
Because you're using a biophysical model.

1049
01:01:03.802 --> 01:01:06.123
And it's like, dude, that's the point.

1050
01:01:06.163 --> 01:01:06.743
That's the point.

1051
01:01:06.803 --> 01:01:13.445
It's like without a biophysical model, I wouldn't be able to make these predictions of, oh, it's the PV plus neurons in the cortex.

1052
01:01:13.745 --> 01:01:17.026
It's this highway in between the striatum and the palladium.

1053
01:01:17.406 --> 01:01:21.327
It's the subthalamic nucleus that we need to probe, right?

1054
01:01:21.387 --> 01:01:22.207
It's just going to be subthalamic.

1055
01:01:22.387 --> 01:01:40.681
random neuron in some random layer of a trillion parameter model what good is that the the a hundred percent i still come back to one of the there's so many interesting insights in this entire papers architecture like the experimental design architecture but i think a key piece was uh

1056
01:01:42.222 --> 01:01:50.371
You can get traceability and then subsequently reproducibility because of that brain simulator.

1057
01:01:50.431 --> 01:01:52.453
Yeah.

1058
01:01:52.934 --> 01:01:57.679
And the way in which it is explicit, it is...

1059
01:01:59.040 --> 01:02:01.923
What's the, there's the terminology, and this may not be correct.

1060
01:02:02.163 --> 01:02:12.852
There's this terminology where probabilistic versus deterministic, and it's more on the deterministic side and less on the probabilistic side, which in science, it's a good thing.

1061
01:02:12.992 --> 01:02:15.054
Yeah, it's very much a good thing.

1062
01:02:15.414 --> 01:02:23.682
And then finally we now have an idea that it should be the sub-dalamic nucleus that we target, which is this tiny, like, lentil-sized structure

1063
01:02:24.322 --> 01:02:24.542
Right?

1064
01:02:25.142 --> 01:02:27.143
It's actually also quite incredible if you think about it.

1065
01:02:27.183 --> 01:02:28.803
Like, that is the boss's office.

1066
01:02:29.743 --> 01:02:29.983
Right.

1067
01:02:30.143 --> 01:02:30.363
Right?

1068
01:02:30.543 --> 01:02:32.504
For at least stimulation of the brain.

1069
01:02:32.544 --> 01:02:34.404
Consciousness could be a distributive thing.

1070
01:02:34.784 --> 01:02:46.547
But just stimulating that tiny structure of maybe, let's say, hundreds of thousands of neurons at the max is going to wake up a brain that is 86 billion neurons.

1071
01:02:47.627 --> 01:02:48.627
Right?

1072
01:02:48.647 --> 01:02:50.068
Consciousness is a crazy thing.

1073
01:02:50.248 --> 01:02:51.588
And we're getting closer to...

1074
01:02:52.668 --> 01:02:54.410
So would it actually a correct analysis?

1075
01:02:54.430 --> 01:03:03.038
Because there's going to be probably some thoughts about, oh, like consciousness is, like you discussed at the beginning, a very expansive thing.

1076
01:03:03.598 --> 01:03:10.445
And really all we're identifying in this last piece is, I keep using the grill lighter example.

1077
01:03:10.465 --> 01:03:16.030
This might just be the igniter, the little button you press on the grill to turn consciousness on.

1078
01:03:16.190 --> 01:03:16.410
Yeah.

1079
01:03:16.831 --> 01:03:19.153
But that's all we're saying right now.

1080
01:03:19.173 --> 01:03:21.535
I don't know what the grill being on means.

1081
01:03:21.875 --> 01:03:22.035
Right.

1082
01:03:22.055 --> 01:03:23.156
We don't know how big the grill is.

1083
01:03:23.357 --> 01:03:24.778
Is it a black stone?

1084
01:03:24.818 --> 01:03:25.078
Yeah.

1085
01:03:25.198 --> 01:03:27.040
Is it whatever?

1086
01:03:27.460 --> 01:03:27.861
Exactly.

1087
01:03:27.881 --> 01:03:35.828
We're just saying we may have found where we can wake an unconscious from this vegetative or unconscious state.

1088
01:03:35.868 --> 01:03:36.669
Yes.

1089
01:03:36.849 --> 01:03:40.092
And the interesting dovetail of how.

1090
01:03:42.192 --> 01:03:59.649
The 2007 paper we talked about that identified this idea of minimally waking consciousness was actually a key step in order for this to actually also work because that was where we started to zoom in and be able to validate the mesocentric model.

1091
01:03:59.709 --> 01:04:02.392
It's just like everything has to work.

1092
01:04:03.183 --> 01:04:04.243
Yeah, one after the other.

1093
01:04:04.263 --> 01:04:05.203
It has to stack.

1094
01:04:05.403 --> 01:04:05.904
Right, yeah.

1095
01:04:06.144 --> 01:04:07.984
In order for this to even be possible.

1096
01:04:08.104 --> 01:04:08.324
Yeah.

1097
01:04:08.384 --> 01:04:09.464
This was really quite nice.

1098
01:04:09.824 --> 01:04:10.565
It was a good paper.

1099
01:04:11.685 --> 01:04:15.245
Daniel does a lot of really cool work with Koma.

1100
01:04:15.305 --> 01:04:17.866
He's actually also in an organoid lab at UCLA.

1101
01:04:18.146 --> 01:04:22.347
So one of these days, we'll come by UCLA and see your lab, Daniel.

1102
01:04:22.467 --> 01:04:24.447
But until then, nicely done.

1103
01:04:24.807 --> 01:04:26.108
Very, very nicely done.

1104
01:04:26.408 --> 01:04:27.168
Obviously, the...

1105
01:04:28.620 --> 01:04:32.282
This is early, but there's a huge...

1106
01:04:32.602 --> 01:04:46.708
The unlocks that are there, you know, you could talk about it forever, just in terms of not only the clinical or medical context, but even for the philosophy of mind kind of stuff, having some mechanistic details.

1107
01:04:47.008 --> 01:04:48.169
Yes, that's big.

1108
01:04:48.469 --> 01:04:49.149
And I think...

1109
01:04:51.150 --> 01:05:02.297
The part that is going to be most influential for this paper is actually the way that he went about using this biophysical model.

1110
01:05:02.377 --> 01:05:03.318
Yes, I totally agree.

1111
01:05:03.338 --> 01:05:03.878
And tweaking.

1112
01:05:04.078 --> 01:05:04.779
I totally agree.

1113
01:05:04.819 --> 01:05:06.980
That's a very interesting way to do things.

1114
01:05:07.040 --> 01:05:07.721
I totally agree.

1115
01:05:08.061 --> 01:05:09.702
And I can already imagine...

1116
01:05:11.047 --> 01:05:21.349
it being used for all sorts of stuff, material science, like physical things, you know, like we've got really good models about how atoms work with each other.

1117
01:05:21.389 --> 01:05:21.829
Exactly.

1118
01:05:22.269 --> 01:05:30.090
Make a GAN genetic network type thing, you know, like that is the part that is really like, I think mechanistically very interesting to me.

1119
01:05:30.130 --> 01:05:31.230
The methods is very nice.

1120
01:05:31.270 --> 01:05:39.652
I mean, obviously the results are incredible, but to me, the, the one I'm most impressed about is the way that he used this biophysical model.

1121
01:05:39.672 --> 01:05:39.732
Um,

1122
01:05:40.942 --> 01:05:51.387
As someone who's coming more from the technology and software world, it is naturally what I'm able to better engage in at a deep level.

1123
01:05:52.348 --> 01:05:53.468
But I agree with you.

1124
01:05:53.508 --> 01:06:02.612
I think they've basically created a conceptual framework around how to design things.

1125
01:06:04.133 --> 01:06:14.983
ML and AI architecture as a, as an end to end system and process that is subject matter agnostic.

1126
01:06:15.364 --> 01:06:15.504
Mm hmm.

1127
01:06:16.175 --> 01:06:16.675
Yeah, yeah.

1128
01:06:17.036 --> 01:06:17.937
Like it could work.

1129
01:06:17.957 --> 01:06:18.417
Yeah, yeah, yeah.

1130
01:06:18.457 --> 01:06:19.238
That thing can work.

1131
01:06:19.578 --> 01:06:23.321
And a whole variety of – the brain was just the first target.

1132
01:06:23.681 --> 01:06:27.404
But the exact same conceptual framework could work in a variety of different areas.

1133
01:06:27.484 --> 01:06:27.705
Yeah.

1134
01:06:27.725 --> 01:06:30.747
Particularly the data feedback loop and validation process, et cetera.

1135
01:06:31.248 --> 01:06:33.209
Really, really – I just – this was great.

1136
01:06:33.369 --> 01:06:33.830
Yeah, good stuff.

1137
01:06:33.850 --> 01:06:33.970
Yeah.

1138
01:06:34.130 --> 01:06:36.272
Well done, Daniel Toker et al.

1139
01:06:36.332 --> 01:06:40.335
Again, this was in Nature Neuroscience on March 24th.

1140
01:06:40.735 --> 01:06:41.896
Fresh hot off the presses.

1141
01:06:42.036 --> 01:06:53.706
Again, there is nowhere else on the internet, not even on the direct pages of the authors themselves, where you're going to get this level of excitement, passion, and deep dive.

1142
01:06:54.246 --> 01:06:54.867
They need a break.

1143
01:06:54.907 --> 01:06:55.868
Daniel needs a break anyway.

1144
01:06:55.888 --> 01:06:58.870
So it's our job to do this and step in.

1145
01:06:58.970 --> 01:07:00.992
Yeah, we'll see what he thinks about it.

1146
01:07:01.392 --> 01:07:02.653
You actually guys did pretty well.

1147
01:07:02.673 --> 01:07:03.134
You did pretty well.

1148
01:07:03.474 --> 01:07:05.656
No, just another fantastic paper.

1149
01:07:05.896 --> 01:07:18.247
And we just really want to thank those of you who have stayed this long and listened to the episode for staying with us and enjoying this journey of curiosity and discovery.

1150
01:07:18.367 --> 01:07:21.670
I will do a brief pause to ask if we would like to do a comment.

1151
01:07:23.597 --> 01:07:24.878
for the audience.

1152
01:07:26.459 --> 01:07:26.799
Gan.

1153
01:07:27.179 --> 01:07:28.400
Gan.

1154
01:07:28.440 --> 01:07:29.580
This has become our shtick.

1155
01:07:29.660 --> 01:07:32.501
We can't think of anything.

1156
01:07:32.601 --> 01:07:33.662
What's another alternate...

1157
01:07:36.019 --> 01:07:37.139
full form of GAN.

1158
01:07:37.179 --> 01:07:39.140
Yeah, acronym definition for what?

1159
01:07:39.201 --> 01:07:40.821
Generative Adversarial Networks.

1160
01:07:40.941 --> 01:07:42.422
Come on, you guys can think of something better.

1161
01:07:42.442 --> 01:07:43.403
Something great.

1162
01:07:43.463 --> 01:07:44.583
We really appreciate you all.

1163
01:07:44.643 --> 01:07:51.267
My name is Lester Nare, joined as always by my co-host and our resident PhD, Krishna Chowdhury.

1164
01:07:51.587 --> 01:07:53.448
This is the last reminder I'll make on this.

1165
01:07:53.528 --> 01:07:57.830
We are moving to single-story episodes multiple times a week.

1166
01:07:58.231 --> 01:08:00.732
We have gotten confirmation from our deep listeners that...

1167
01:08:01.072 --> 01:08:02.733
You do appreciate it and you like it.

1168
01:08:02.873 --> 01:08:06.094
And there was a good idea that we will think through.

1169
01:08:06.114 --> 01:08:14.356
I don't know if actually, I don't even know if you saw this, but the rundown, when we do the rundown at the end of the week, we don't really go in depth.

1170
01:08:14.716 --> 01:08:19.178
And there's a lot of folks who would like us to go in depth or any number of those particular stories.

1171
01:08:19.238 --> 01:08:19.418
So,

1172
01:08:20.098 --> 01:08:26.820
We will look into having basically a patron system to vote on what rundown story should we follow up on for a deep dive.

1173
01:08:27.121 --> 01:08:27.881
It's a great idea.

1174
01:08:28.801 --> 01:08:32.022
Excuse me for not remembering the commenter who said it, but you can comment again.

1175
01:08:32.082 --> 01:08:32.743
We do see it.

1176
01:08:32.783 --> 01:08:33.483
Thank you so much.

1177
01:08:34.603 --> 01:08:36.644
We will see you all later this week.
