WEBVTT

1
00:00:00.223 --> 00:00:08.485
Orangutan mothers arrange play dates for their infants in order to invest in their social development.

2
00:00:08.885 --> 00:00:16.647
You're turning the patient into a collaborator for training the neural network, for training the machine learning model.

3
00:00:16.687 --> 00:00:17.447
This is so good.

4
00:00:17.827 --> 00:00:29.370
They have a robotic arm that in 50 steps in 30 minutes can set up these lenses for this optical experiment, which we've talked about previously.

5
00:00:29.890 --> 00:00:30.671
Hello, Internet.

6
00:00:30.691 --> 00:00:32.252
This is your captain speaking.

7
00:00:32.292 --> 00:00:39.398
Lester Nare joined, as always, by my co-host and our resident PhD, Krishna Chowdhury.

8
00:00:39.939 --> 00:00:45.724
We are back for a fun rundown episode this week with some special news.

9
00:00:46.364 --> 00:01:14.263
about our recent award-winning show as the recipients of a bronze signal award which we just got the news of this morning and so a big shout out to all of you in ffp nation for watching the show for giving us the inspiration a quick question for our resident phd how does it feel to be recognized for the fifth annual signal award in best podcast for science

10
00:01:15.524 --> 00:01:16.265
It feels great.

11
00:01:18.146 --> 00:01:21.089
In the sixth annual Signal Award, I would like to get gold.

12
00:01:21.169 --> 00:01:21.449
Yes.

13
00:01:21.609 --> 00:01:23.811
So that's what I'm feeling.

14
00:01:24.351 --> 00:01:31.758
Gold is on the mind and we are going to get into three science stories today for a rundown.

15
00:01:32.118 --> 00:01:34.940
For those of you who joined us during the Nobel Prizes, welcome.

16
00:01:35.281 --> 00:01:36.562
Rundowns are where we do fun.

17
00:01:36.982 --> 00:01:47.685
a little bit more of a high-level overview of some science stories that we have found interesting, starting with an animal behavior story, the parenting techniques and sacrifices of orangutan.

18
00:01:48.025 --> 00:01:55.027
Mothers followed by personalized medicine with AI using a chatbot to make epilepsy treatment more effective.

19
00:01:55.067 --> 00:01:59.048
And we're going to end with automated scientific discovery.

20
00:01:59.343 --> 00:02:05.085
The AI czars are telling us they're going to solve all human disease in 10 years.

21
00:02:05.425 --> 00:02:06.625
But what does that really mean?

22
00:02:06.705 --> 00:02:10.006
And there's an interesting story that actually came out earlier today.

23
00:02:10.026 --> 00:02:20.370
It's going to help us look at the larger ecosystem of what is going to be required to have AI really impact clinical outcomes for humanity.

24
00:02:20.570 --> 00:02:26.792
As always, we're going to talk about the science from the ground up today because this is From First Principles.

25
00:02:40.957 --> 00:02:43.044
The headline here is Irresistible.

26
00:02:43.505 --> 00:02:46.474
Orangutan mothers arrange playdates...

27
00:02:47.130 --> 00:02:51.754
for their infants in order to invest in their social development.

28
00:02:52.534 --> 00:02:55.957
I mean, as soon as I saw this, I was like, we got to cover it.

29
00:02:55.977 --> 00:02:56.697
We got to cover it.

30
00:02:56.777 --> 00:03:00.460
Because, I mean, it intrigues me in so many ways.

31
00:03:00.540 --> 00:03:15.952
As a new father, you know, I'm starting to think about the pros and cons of, like, putting my kid in daycare and, like, the social development that comes with it, but also the exposure to, like, you know, just getting sick and, like, all sorts of, like, nonsense that comes with that.

32
00:03:16.292 --> 00:03:20.793
So, you know, there's a lot hiding in that headline.

33
00:03:21.453 --> 00:03:32.335
And when I was diving into the paper, there's a lot of rich science behind the methods and things like that that I thought would be perfect for this podcast.

34
00:03:32.715 --> 00:03:37.756
There's a fact that this seems to be a deliberate decision by the orangutan mothers.

35
00:03:38.136 --> 00:03:42.457
Like, they're not just, like, casually meeting up with other mothers specifically.

36
00:03:43.003 --> 00:03:43.823
that have babies.

37
00:03:43.943 --> 00:03:45.464
This is a deliberate decision.

38
00:03:46.525 --> 00:03:50.066
There's a fact that they have to give up energy and time to do this.

39
00:03:50.526 --> 00:03:54.088
So there's a story where, you know, mothers are making sacrifices.

40
00:03:54.168 --> 00:03:57.309
It's so similar to how humans do things.

41
00:03:57.549 --> 00:03:58.130
100%.

42
00:03:59.070 --> 00:04:02.772
And the science underneath that headline is really interesting.

43
00:04:02.832 --> 00:04:10.435
Like the way they collected years of data, analyzed that data, and then they can convincingly state that, you know,

44
00:04:12.068 --> 00:04:16.492
Orangutan mothers are consciously, or not consciously, but deliberately trying to do this.

45
00:04:17.033 --> 00:04:18.414
I think it's really, really cool.

46
00:04:18.575 --> 00:04:21.337
Careful with the consciousness word with all the AI stuff going on.

47
00:04:21.357 --> 00:04:22.198
I know, right?

48
00:04:22.278 --> 00:04:23.980
It's up for debate right now.

49
00:04:24.180 --> 00:04:24.601
Exactly.

50
00:04:24.661 --> 00:04:29.866
I mean, if you ask me personally, my personal opinion is that, of course, orangutans are conscious.

51
00:04:29.946 --> 00:04:30.406
Like, just...

52
00:04:31.187 --> 00:04:34.309
Look at them and look at the way that they respond to their environment.

53
00:04:34.349 --> 00:04:35.429
They respond to each other.

54
00:04:35.810 --> 00:04:36.830
They respond to stress.

55
00:04:37.170 --> 00:04:44.754
I mean, it's obvious to me, but there's going to be people that are like, you know, saying, yeah, you don't really know what consciousness is.

56
00:04:44.854 --> 00:04:45.295
Okay.

57
00:04:45.835 --> 00:04:48.196
Anyways, let's get into some history and background.

58
00:04:48.636 --> 00:04:50.217
Orangutans, they're the great apes.

59
00:04:50.277 --> 00:04:51.258
They're part of the great apes.

60
00:04:51.298 --> 00:04:54.540
That includes gorillas, chimpanzees, bonobos, and us.

61
00:04:54.720 --> 00:04:55.000
Mm-hmm.

62
00:04:55.520 --> 00:05:08.514
The animals that are in this study are particularly the Bornean orangutans, specifically Pongo pygmaeus wormby, another taxonomical name that I'm struggling with.

63
00:05:08.914 --> 00:05:12.478
But these are the guys that live in Tuanan in central Georgia.

64
00:05:12.918 --> 00:05:32.813
kalimantan in the indonesian borneo so borneo is a giant island and there's indonesia has like a majority of that and i'll just know native uh for all of my nickelodeon fans the wild thorn berries is the first time i ever heard of the location of borneo because this family that traveled around is like wildlife videographers

65
00:05:33.514 --> 00:05:47.856
uh like that was kind of it was like i think it's literally in the theme song uh that and so it was really cool because they traveled the world but that was the first time i understood where the location was they must have been looking for orangutans in borneo because borneo is very famous for being like sort of a native

66
00:05:48.497 --> 00:05:56.424
So the family had an ape, Darwin, that could speak to Eliza because Eliza could speak to animals.

67
00:05:56.904 --> 00:05:57.405
Oh, I see.

68
00:05:57.425 --> 00:05:58.346
That's the shtick of the show.

69
00:05:58.366 --> 00:05:58.866
That's the shtick.

70
00:05:58.926 --> 00:06:01.208
And they had a, you know, Darwin was around.

71
00:06:01.788 --> 00:06:03.230
It's a good name for an ape.

72
00:06:03.890 --> 00:06:04.431
Yeah.

73
00:06:04.891 --> 00:06:06.372
Clearly was done with intent.

74
00:06:06.472 --> 00:06:07.133
Yeah, yeah, yeah.

75
00:06:07.794 --> 00:06:11.277
So orangutans are a bit different from other apes, okay?

76
00:06:11.297 --> 00:06:12.938
Because they are not that social.

77
00:06:13.839 --> 00:06:15.140
They are solitary, okay?

78
00:06:15.914 --> 00:06:22.240
In the sense that adult females, they occupy home regions that maybe overlap with neighbors.

79
00:06:22.681 --> 00:06:25.644
But in a real sense, like, they're very solitary.

80
00:06:26.064 --> 00:06:34.172
Solitary here means that the usual organization of travel and feeding, like, they don't go in groups.

81
00:06:34.492 --> 00:06:36.615
They don't assemble in packs, you know?

82
00:06:37.235 --> 00:06:39.176
So they're like Americans, very individualistic.

83
00:06:39.336 --> 00:06:40.396
Very individualistic.

84
00:06:40.576 --> 00:06:43.517
In modern Americans, very much so, right?

85
00:06:43.737 --> 00:06:45.998
Modern society, very much so, in some sense.

86
00:06:46.418 --> 00:06:51.920
And maybe that tells us something that we can gather from the study, but we'll touch on that later on, right?

87
00:06:52.720 --> 00:06:59.342
The idea is that rather than traveling every single day in a stable troop, they actually move as separate units.

88
00:06:59.783 --> 00:07:05.284
So the mothers will be separate, and if a mother has an offspring, they're going to move as separate units, okay?

89
00:07:07.285 --> 00:07:13.032
Some section of the forest that they like explore, but those are going to be mostly disjoint.

90
00:07:14.454 --> 00:07:17.558
Now, there are good ecological reasons for that.

91
00:07:17.778 --> 00:07:23.645
It actually has to do with the diet of these orangutans and like where they live.

92
00:07:25.625 --> 00:07:28.966
The diet is mostly like fruit, large fruit eaters.

93
00:07:29.306 --> 00:07:34.769
Now, fruit is something that is kind of rare in the forest, and it's sporadic.

94
00:07:34.849 --> 00:07:41.451
It's not like, you know, at a regular cadence, you can be like, that tree is definitely going to have fruit, and then this tree is definitely going to have fruit.

95
00:07:41.771 --> 00:07:43.872
So group living has its advantages, right?

96
00:07:43.892 --> 00:07:49.995
Because you can like share information, you can share protection, but the disadvantages are that

97
00:07:50.875 --> 00:07:55.897
There's a limited amount of food, and you're going to be competing for that limited amount of food, right?

98
00:07:56.817 --> 00:08:00.218
In a space where the density of food is very low.

99
00:08:00.758 --> 00:08:05.120
So that's why orangutans, I mean, from a biophysical perspective, there's a lot of...

100
00:08:07.344 --> 00:08:09.746
theories out there for why they are solitary.

101
00:08:09.806 --> 00:08:14.348
It actually has to do with the physics of the environment itself.

102
00:08:14.448 --> 00:08:21.991
There's a resource constraint that they are conscious, not consciously, but actively operating around.

103
00:08:22.371 --> 00:08:22.872
Exactly.

104
00:08:23.412 --> 00:08:25.234
This is very different from other great apes.

105
00:08:25.414 --> 00:08:29.277
Chimpanzees, very famously, they are in troops, right?

106
00:08:29.818 --> 00:08:33.881
There's a lot of resource abundance in the place that chimpanzees live, right?

107
00:08:33.901 --> 00:08:37.204
So they can actually administer in that.

108
00:08:37.684 --> 00:08:48.314
So now, in this space of solitary confinement of these orangutan individuals, the adults who are solitary in the forest, now let's consider the infant.

109
00:08:48.794 --> 00:08:49.054
Okay.

110
00:08:49.575 --> 00:08:51.856
Orangutans usually have one baby at a time.

111
00:08:52.917 --> 00:08:59.441
And usually that baby happens once every seven to eight years.

112
00:09:00.002 --> 00:09:00.362
Oh, wow.

113
00:09:00.842 --> 00:09:02.243
So they don't have siblings.

114
00:09:02.823 --> 00:09:03.144
Yeah.

115
00:09:03.344 --> 00:09:04.645
In the traditional sense, right?

116
00:09:04.685 --> 00:09:11.609
It's not like a litter of kittens or puppies where, like, they have a bunch of people their same age that they can play around with.

117
00:09:12.510 --> 00:09:13.811
These are single babies.

118
00:09:13.831 --> 00:09:14.071
Yeah.

119
00:09:15.357 --> 00:09:18.538
To a single mother, effectively, because the dad...

120
00:09:20.498 --> 00:09:21.319
Went to go get the milk.

121
00:09:21.419 --> 00:09:22.879
Yeah, yeah, yeah.

122
00:09:23.639 --> 00:09:25.600
You know, the dad is not involved.

123
00:09:25.920 --> 00:09:27.501
Let's say that.

124
00:09:27.661 --> 00:09:37.003
So this is a single mom with a single baby, no siblings, and you've got to introduce some bit of social, you know... Interaction.

125
00:09:37.063 --> 00:09:38.024
Interaction, yes.

126
00:09:38.524 --> 00:09:40.264
You have to introduce some bit of that.

127
00:09:40.344 --> 00:09:41.405
You don't have group living.

128
00:09:41.745 --> 00:09:42.765
So what do you do?

129
00:09:43.757 --> 00:09:45.017
That's the challenge, right?

130
00:09:47.998 --> 00:09:48.818
That's the challenge.

131
00:09:49.758 --> 00:09:54.419
And that creates a developmental problem inside of the ecological problem.

132
00:09:54.459 --> 00:09:55.039
That's interesting.

133
00:09:55.059 --> 00:09:56.140
Yeah, there's layers.

134
00:09:56.440 --> 00:09:56.660
Right?

135
00:09:56.960 --> 00:10:08.683
The ecological problem is, well, because of the forest that I'm living in and because of the constraints that I have depending on the fruit and all of the choices that I have for food, I need to stay independent.

136
00:10:08.723 --> 00:10:10.643
But if I have a kid...

137
00:10:11.813 --> 00:10:17.363
I need the kid to socialize in order to actually grow up and be sort of functionally accurate, right?

138
00:10:17.683 --> 00:10:21.530
Like, play is a very, very important part of upbringing.

139
00:10:22.672 --> 00:10:22.893
So...

140
00:10:24.190 --> 00:10:25.551
Play is actually very important.

141
00:10:25.571 --> 00:10:27.471
It's been studied for a very long time.

142
00:10:27.571 --> 00:10:32.893
Play is difficult science-wise because it's easy to recognize, but it's hard to quantify.

143
00:10:32.933 --> 00:10:33.554
Yeah.

144
00:10:33.594 --> 00:10:37.055
This is a photo of two bear cubs playing.

145
00:10:37.415 --> 00:10:41.337
Now here, they're probably siblings, which is why they're close to one another, right?

146
00:10:41.377 --> 00:10:42.757
And they're playing amongst one another.

147
00:10:43.097 --> 00:10:44.218
I don't think bear cubs...

148
00:10:45.038 --> 00:10:54.228
really mess with other bear cubs because mama bears are very very cautious of other bears when they are nursing and when they're bringing up

149
00:10:55.398 --> 00:10:57.238
their kids, right?

150
00:10:57.419 --> 00:10:59.499
So bears have this advantage of having siblings.

151
00:10:59.639 --> 00:11:00.599
Orangutans do not.

152
00:11:01.200 --> 00:11:09.762
Now, in orangutan research, social play means that you have non-aggressive interactions such as tagging, wrestling, chasing.

153
00:11:10.122 --> 00:11:13.143
It's the stuff that humans are pretty used to.

154
00:11:13.563 --> 00:11:16.964
And there's a long history behind why play evolved.

155
00:11:17.144 --> 00:11:20.405
I mean, one idea is that it rehearses adult skills.

156
00:11:21.105 --> 00:11:21.345
So...

157
00:11:22.385 --> 00:11:28.969
Even if the adult is not meant to be in a social setting, play can still be important.

158
00:11:29.329 --> 00:11:33.331
If the adult is meant to be in a social setting, then of course play is important, right?

159
00:11:33.371 --> 00:11:35.352
Because you're getting used to socializing.

160
00:11:35.873 --> 00:11:40.395
But you could have the argument that like, well, orangutans, they're asocial.

161
00:11:40.895 --> 00:11:41.816
Why do you need to play?

162
00:11:42.076 --> 00:11:49.000
Well, play gives you a kind of emotional regulation, and it gives you the ability to respond to something that...

163
00:11:49.330 --> 00:11:50.391
That is unexpected.

164
00:11:50.732 --> 00:11:50.932
Right.

165
00:11:50.972 --> 00:11:51.132
Right.

166
00:11:51.232 --> 00:11:54.796
If you're doing self-play, then it's kind of a self-feedback.

167
00:11:55.116 --> 00:11:55.757
You know what's coming.

168
00:11:55.877 --> 00:11:56.598
You know what's coming.

169
00:11:56.978 --> 00:12:01.463
When you're playing with someone else, you don't know what the other person is going to do.

170
00:12:01.964 --> 00:12:10.493
And that ability to respond to an environment or to a stimulus that is unpredictable, that can have a lasting effect.

171
00:12:11.033 --> 00:12:17.979
It's beneficial to your own survival, ultimately, because you want to be able to react to unknown stimuli in a way that is beneficial to your own survival.

172
00:12:17.999 --> 00:12:21.803
But that is not also an overreaction unnecessarily.

173
00:12:22.183 --> 00:12:22.643
Exactly.

174
00:12:23.004 --> 00:12:28.729
Yeah, so there is a benefit to play even if the population itself is asocial.

175
00:12:29.069 --> 00:12:32.432
That was the first thing that when I was reading the paper, I was like, that's actually...

176
00:12:32.912 --> 00:12:34.615
interesting to think about, right?

177
00:12:34.655 --> 00:12:40.283
Because it's trivial that play is important for social animals because you need to be social.

178
00:12:40.623 --> 00:12:46.872
But even for asocial animals, there is a component of play that is important for your development.

179
00:12:48.342 --> 00:12:51.045
Any case, do orangutan youngsters play?

180
00:12:51.265 --> 00:13:09.565
Well, in 2024, Julia Kunz, who is a co-author on the current study that we're going to be talking about in depth, she published this article that showed 11,000 observation hours from 99 young orangutans at two sites, one in Borneo and one in Sumatra.

181
00:13:10.186 --> 00:13:16.216
And she showed that social play was more frequent in the Sumatran site versus the Borneo site.

182
00:13:16.576 --> 00:13:21.184
But actually, if you controlled for the amount of times that...

183
00:13:22.325 --> 00:13:25.468
play was presented, it was the same.

184
00:13:25.929 --> 00:13:30.313
So if they're given the opportunity to play, youngster orangutans will play.

185
00:13:30.693 --> 00:13:36.759
So now let's try to hone in on the other part, which is how many times are they given the opportunity to play?

186
00:13:37.360 --> 00:13:38.581
That's what this paper is about.

187
00:13:38.601 --> 00:13:39.442
Because there's a difference.

188
00:13:39.982 --> 00:13:40.183
Right.

189
00:13:40.263 --> 00:13:42.945
It's like if I'm not given the opportunity, then I can't do it, right?

190
00:13:42.965 --> 00:13:43.125
Right.

191
00:13:43.906 --> 00:13:46.028
And the new paper builds on that groundwork.

192
00:13:46.708 --> 00:13:51.872
This is the paper that came out in the Proceedings of the National Academy of Sciences in October 2026.

193
00:13:52.033 --> 00:13:53.854
I see you've avoided the acronym.

194
00:13:54.414 --> 00:13:54.955
PNAS.

195
00:13:56.536 --> 00:13:57.197
That's what it is.

196
00:13:57.737 --> 00:13:58.978
That's what everyone says, okay?

197
00:13:59.178 --> 00:14:01.320
Everyone says, stop saying the acronym.

198
00:14:01.400 --> 00:14:02.381
No, no, no, no.

199
00:14:02.401 --> 00:14:03.021
That's a joke.

200
00:14:03.161 --> 00:14:03.962
Everyone says it.

201
00:14:05.203 --> 00:14:08.145
This is from Max Planck Institute of Animal Behavior.

202
00:14:08.625 --> 00:14:11.067
And there's a few authors from the University of Zurich.

203
00:14:12.088 --> 00:14:19.915
So the earlier work established that orangutan youngsters play and that the opportunities for that play varies, right?

204
00:14:20.556 --> 00:14:34.709
And here what we're saying is that the maternal association could provide for that opportunity and the moms actually make a concerted effort in providing that opportunity.

205
00:14:35.249 --> 00:14:35.489
Okay?

206
00:14:35.810 --> 00:14:37.071
That's what's so cool about this.

207
00:14:37.631 --> 00:14:37.892
Okay.

208
00:14:38.782 --> 00:14:41.504
The methods are incredible, I would say.

209
00:14:42.104 --> 00:14:54.231
So since 2003, Van Nordwyck and her team have been following mother-infant pairs at Tuanan Research Station on the island of Borneo since 2003.

210
00:14:55.532 --> 00:15:00.693
Over a decade of research, they've observed encounters between mothers that are rare.

211
00:15:01.033 --> 00:15:04.434
But when they do happen, the mothers are very cautious.

212
00:15:04.894 --> 00:15:10.155
But the kids, as soon as they see each other, they just like run at each other and start a blast.

213
00:15:10.215 --> 00:15:11.216
Yeah, they're having a blast.

214
00:15:11.536 --> 00:15:12.776
The mothers are very cautious.

215
00:15:12.816 --> 00:15:16.097
They've, you know, developed the whole adult thing of being solitary.

216
00:15:16.497 --> 00:15:19.058
The kids, they're having a great time.

217
00:15:25.759 --> 00:15:28.122
they seem to come at a cost to the mothers.

218
00:15:29.203 --> 00:15:30.224
Oh, okay.

219
00:15:30.244 --> 00:15:35.229
The cost being it's a really long and hard day for them when they have to arrange these playdates.

220
00:15:36.330 --> 00:15:38.012
Are they even arranging these playdates?

221
00:15:38.112 --> 00:15:44.138
And if they arrange these playdates, are they doing it at an expense to their own benefit, right?

222
00:15:44.158 --> 00:15:45.760
Because they could be doing other things.

223
00:15:46.200 --> 00:15:47.321
Resource gathering.

224
00:15:47.381 --> 00:15:48.481
Yeah, they could be eating.

225
00:15:48.621 --> 00:15:49.502
They could be resting.

226
00:15:49.702 --> 00:15:51.002
Instead, they're doing this, right?

227
00:15:51.303 --> 00:15:52.603
So is it a concerted effort?

228
00:15:52.983 --> 00:16:01.407
After 15 years, they've amassed 32,000 hours of observation from 31 mother-infant pairs.

229
00:16:01.768 --> 00:16:04.149
That's actually incredible, first of all, right?

230
00:16:04.429 --> 00:16:07.630
Like 15 years of work, 31,000 hours.

231
00:16:08.271 --> 00:16:10.312
That's like on the order of like a year, a year.

232
00:16:11.132 --> 00:16:14.454
of just like continuous observing orangutans.

233
00:16:14.995 --> 00:16:15.615
It's incredible.

234
00:16:15.895 --> 00:16:18.597
The basic technique is something called a focal follow.

235
00:16:19.037 --> 00:16:24.481
This is a very famous photo of Jane Goodall doing a focal follow on chimpanzees.

236
00:16:25.682 --> 00:16:36.109
The idea is that you have researchers that choose an identified individual, and then you basically follow its behavior from sundown to sunset, or from sunrise to sunset.

237
00:16:41.564 --> 00:16:42.645
where the person is.

238
00:16:42.685 --> 00:16:48.008
So you have like a GPS monitor and you're like, okay, I'm going to monitor where the orangutan is right now.

239
00:16:48.409 --> 00:16:52.231
I'm going to monitor what the animal is doing at that moment.

240
00:16:52.371 --> 00:16:53.672
So is the mother feeding?

241
00:16:54.072 --> 00:16:56.294
Is the youngster playing with whom?

242
00:16:56.814 --> 00:17:00.877
And roughly every 30 minutes you log a GPS location, right?

243
00:17:01.177 --> 00:17:07.701
So you, you get a kind of GPS of, of where this person, where, where this animal is going and you get a,

244
00:17:09.135 --> 00:17:13.417
every two minutes, some kind of log of the activity over the entire day.

245
00:17:14.058 --> 00:17:16.779
That's a rich data set.

246
00:17:16.899 --> 00:17:17.159
Yeah.

247
00:17:17.720 --> 00:17:18.480
Incredible, right?

248
00:17:18.520 --> 00:17:18.900
Yeah.

249
00:17:19.040 --> 00:17:26.224
And it requires a human being just straight up following this orangutan or this mother-infant pair for the entire day.

250
00:17:27.265 --> 00:17:27.945
Pretty incredible.

251
00:17:27.965 --> 00:17:30.066
For days and days and days.

252
00:17:30.126 --> 00:17:36.610
And it's the same individual that you have to follow for days and days because you want to have some type of longitudinal data on the same individual.

253
00:17:36.690 --> 00:17:36.890
Yeah.

254
00:17:37.792 --> 00:17:49.206
Now, the data set here includes 3,000 plus follow days, 2,000 of which were full nest to nest days, meaning like there's like mothers involved.

255
00:17:49.226 --> 00:17:49.927
Only 6.3%.

256
00:17:49.947 --> 00:17:50.448
So only 6%.

257
00:17:55.008 --> 00:17:57.551
Did they record play?

258
00:17:57.731 --> 00:17:58.251
That's incredible.

259
00:17:58.672 --> 00:18:06.540
What we're saying is over the course of all this observation, a less than 10% proportion of it has this peer play concept, has the thing we're looking for.

260
00:18:07.040 --> 00:18:11.205
And that's why you need such a large data set because it's so infrequent.

261
00:18:11.445 --> 00:18:12.166
Yeah.

262
00:18:12.186 --> 00:18:16.650
And you need a large data set because if you have infrequent events,

263
00:18:17.671 --> 00:18:21.034
In order to do good statistics, you need a large data set, right?

264
00:18:21.155 --> 00:18:23.337
Reminds me of the IceCube neutrino detector.

265
00:18:23.357 --> 00:18:23.797
100%.

266
00:18:23.857 --> 00:18:24.598
It's the same thing.

267
00:18:24.778 --> 00:18:27.060
It's the same exact paradigm, right?

268
00:18:27.240 --> 00:18:32.705
Because you need a lot of events in order to hone down on a very rare, like, per chance.

269
00:18:33.806 --> 00:18:35.448
So now let's get into the figures.

270
00:18:35.748 --> 00:18:37.029
They've collected all this data.

271
00:18:37.249 --> 00:18:38.491
What do they actually see?

272
00:18:38.911 --> 00:18:40.653
Let's start actually with figure C here.

273
00:18:40.713 --> 00:18:41.453
This is figure 1.

274
00:18:42.501 --> 00:18:49.433
In C, on the x-axis, you have the age difference between two offspring that are meeting up for a play date.

275
00:18:49.914 --> 00:18:52.037
And meeting up meaning, you know, they're playing with one another.

276
00:18:52.098 --> 00:18:55.203
This is obvious from a human that's like looking up and seeing them.

277
00:18:56.832 --> 00:19:04.922
On the y-axis is their play rate, which is some normalized sense of how often do these two individuals get together.

278
00:19:05.002 --> 00:19:07.205
We're looking at C. Okay, yes.

279
00:19:07.285 --> 00:19:07.505
Right?

280
00:19:07.686 --> 00:19:09.408
We're looking at this part right here.

281
00:19:09.428 --> 00:19:09.588
Yes.

282
00:19:09.908 --> 00:19:15.035
And there, you can see there's a negative correlation, meaning that as the age difference gets bigger...

283
00:19:15.495 --> 00:19:17.375
the play rate diminishes.

284
00:19:17.496 --> 00:19:18.056
Yeah, that makes sense.

285
00:19:18.496 --> 00:19:19.516
Kind of makes sense, right?

286
00:19:19.556 --> 00:19:21.656
My little brother and I have a 13-year age gap.

287
00:19:22.077 --> 00:19:22.297
Yeah.

288
00:19:22.317 --> 00:19:26.417
So our play rate was much lower than siblings who were much closer together.

289
00:19:26.558 --> 00:19:26.998
Exactly.

290
00:19:27.378 --> 00:19:30.118
The other cool thing is on the y-axis, it's actually exponential.

291
00:19:30.178 --> 00:19:30.998
That's a log scale.

292
00:19:31.039 --> 00:19:31.279
Okay.

293
00:19:31.299 --> 00:19:34.459
Because it goes like 10 to the minus 1, 10 to the minus 2, 10 to the minus 3.

294
00:19:34.739 --> 00:19:36.520
So this is exponential decay.

295
00:19:36.680 --> 00:19:41.521
It's a very strong effect that you're seeing, which is like, it seems that like the...

296
00:19:42.928 --> 00:19:45.931
The closer you are to age, the longer you're going to play, right?

297
00:19:46.331 --> 00:19:46.592
Okay.

298
00:19:47.332 --> 00:19:57.441
On A and B, what we're actually seeing is the effect size based on different markers that the human beings that are on the ground are actually marking.

299
00:19:57.781 --> 00:20:06.209
For example, you can track the relatedness of two individuals, and you can track the fruit availability in the...

300
00:20:07.450 --> 00:20:07.870
forest.

301
00:20:08.171 --> 00:20:10.973
Because you could imagine, right, fruit availability is a confound.

302
00:20:11.353 --> 00:20:17.198
If there's a lot of fruit, then of course the mothers are going to be foraging a lot, and so they're going to meet up a lot, right?

303
00:20:17.358 --> 00:20:20.761
You want to disentangle those and see what's the actual effect.

304
00:20:21.101 --> 00:20:24.584
Well, actually, fruit availability is not that high.

305
00:20:25.044 --> 00:20:27.146
In effect, the correlation is not that big.

306
00:20:27.526 --> 00:20:34.772
On the other hand, relatedness, like whether you're related to the other kid,

307
00:20:35.760 --> 00:20:37.181
That's high for occurrence.

308
00:20:37.801 --> 00:20:45.125
But if you control for occurrence, the amount of play that you do per occurrence is not that high.

309
00:20:46.466 --> 00:20:48.367
If you're family, I'm not about that life.

310
00:20:48.467 --> 00:20:50.228
Yeah, like we'll meet up a lot.

311
00:20:50.288 --> 00:20:51.649
Yeah, but.

312
00:20:51.709 --> 00:20:55.311
But I would play with you just as much as I would play with a friend.

313
00:20:56.412 --> 00:20:57.152
You see what I'm saying?

314
00:20:57.172 --> 00:20:57.933
Yeah, yep, yep.

315
00:20:58.093 --> 00:20:59.654
Like there's a difference here.

316
00:20:59.814 --> 00:21:03.436
Like maybe the occurrence is higher with family because like the moms know each other.

317
00:21:03.901 --> 00:21:05.782
But just because your family doesn't mean I play with you more.

318
00:21:05.922 --> 00:21:06.322
Yeah, yeah.

319
00:21:06.362 --> 00:21:10.665
I play with you the same amount that I would with like a stranger's kid, right?

320
00:21:11.645 --> 00:21:15.488
The two big anti-correlations are, again, age difference.

321
00:21:15.788 --> 00:21:16.588
That's the very top.

322
00:21:17.108 --> 00:21:22.972
It's like if the age difference is very high, then the amount of play rate is very low and the occurrence is very low.

323
00:21:23.012 --> 00:21:24.533
So it seems to be that the mothers...

324
00:21:25.355 --> 00:21:30.139
have honed in on who has the offspring that is about the same age as my kid.

325
00:21:30.179 --> 00:21:31.741
That's fascinating.

326
00:21:31.801 --> 00:21:32.501
Isn't that crazy?

327
00:21:32.521 --> 00:21:33.462
That's so good.

328
00:21:33.803 --> 00:21:42.230
Like, I mean, they're keeping track of who has the right age group of kid.

329
00:21:43.191 --> 00:21:47.615
Now, the other cool thing, the ranges overlap for moms.

330
00:21:47.855 --> 00:21:50.858
Because I told you, right, the moms have a...

331
00:21:52.136 --> 00:21:58.422
They're solitary, which means that the range of forests that they forage might not overlap all that much.

332
00:21:58.883 --> 00:22:03.928
But overlap is higher for moms that have similar age group kids.

333
00:22:04.608 --> 00:22:05.369
Oh, that's really good.

334
00:22:05.529 --> 00:22:05.769
Right?

335
00:22:05.809 --> 00:22:06.570
That's really good.

336
00:22:06.690 --> 00:22:08.072
Because you have this data.

337
00:22:08.112 --> 00:22:08.352
Yeah.

338
00:22:08.392 --> 00:22:10.794
Over 15 years, you've been collecting GPS data.

339
00:22:10.914 --> 00:22:11.235
Yeah.

340
00:22:11.275 --> 00:22:16.700
You can actually tell where the moms forage, and the overlap is higher, it's stronger, right?

341
00:22:17.142 --> 00:22:19.483
for two moms that have similar aged kids.

342
00:22:20.143 --> 00:22:24.005
Which is pointing to this idea of some intentionality of going to that place of overlap.

343
00:22:24.045 --> 00:22:24.505
Yes.

344
00:22:24.585 --> 00:22:31.768
Because I know the other mom that comes around here, their little Bobby and my little Timmy are about the same age.

345
00:22:31.968 --> 00:22:32.969
About the same age, right?

346
00:22:32.989 --> 00:22:35.050
So they can play around.

347
00:22:35.910 --> 00:22:39.412
This entire paper is going into that intentionality.

348
00:22:39.472 --> 00:22:40.392
Yeah, that's really good.

349
00:22:40.412 --> 00:22:42.573
Can we prove intentionality?

350
00:22:43.175 --> 00:22:44.377
In orangutan mothers.

351
00:22:44.837 --> 00:22:45.478
That's the point.

352
00:22:46.079 --> 00:23:01.557
And the way they've structured these questions and the data points they're tracking in order to reach the conclusions about intentionality are very clever in the areas that they're trying to cover to remove other confounding variables.

353
00:23:01.597 --> 00:23:01.998
Exactly.

354
00:23:02.018 --> 00:23:02.178
Exactly.

355
00:23:02.370 --> 00:23:12.419
Yeah, and that's, I've always find this fascinating because like, especially with animal behavior, it's very hard to think of all the confounds, but these guys are doing a pretty good job.

356
00:23:12.459 --> 00:23:13.640
It's quite nice.

357
00:23:14.140 --> 00:23:15.501
The next figure is figure two.

358
00:23:15.901 --> 00:23:19.525
Here we talk about what it costs for the mothers.

359
00:23:20.165 --> 00:23:20.365
Right.

360
00:23:21.667 --> 00:23:24.430
Here, the researchers are comparing three different categories.

361
00:23:24.810 --> 00:23:27.974
There's days when the mother was alone with her offspring.

362
00:23:28.654 --> 00:23:32.959
There's days when another orangutan is nearby, but there's no play.

363
00:23:33.720 --> 00:23:36.303
And then there's days when there is infant play.

364
00:23:36.623 --> 00:23:39.024
Okay, those are the three sort of categories.

365
00:23:39.865 --> 00:23:53.992
The middle category is really important because here we're simply disentangling between, like, okay, if another orangutan is nearby, does that have an effect versus another orangutan is nearby and my infant is playing, right?

366
00:23:54.613 --> 00:23:56.453
There's two different things that are happening.

367
00:23:57.034 --> 00:23:59.595
And here we're seeing an actual difference, right?

368
00:23:59.695 --> 00:24:04.758
On the left-hand side, this is the daily amount of path that the mother takes.

369
00:24:05.742 --> 00:24:11.283
When the kid is playing, the path length that the mother traversed in the forest is a lot longer.

370
00:24:11.844 --> 00:24:13.144
Almost 20% longer.

371
00:24:14.304 --> 00:24:18.965
Which means that she's exerting more effort to actually go from one place to another.

372
00:24:19.786 --> 00:24:24.947
On the right-hand side, figure 2c, that is the proportion of time feeding.

373
00:24:26.567 --> 00:24:29.248
This is what's most important to me and most salient to me.

374
00:24:30.066 --> 00:24:40.530
The amount of time that the mother spent feeding herself was lower on days when her infant played with another friend.

375
00:24:40.690 --> 00:24:51.414
There's more caloric deficit from the amount of movement while having a lower amount of caloric intake at that stage where you're having this higher caloric deficit.

376
00:24:51.534 --> 00:24:52.294
Yeah, yeah, yeah.

377
00:24:52.414 --> 00:24:53.695
It's a twofold... Yeah.

378
00:24:55.295 --> 00:24:55.575
Yeah.

379
00:24:55.635 --> 00:24:56.016
Effect.

380
00:24:56.136 --> 00:24:56.416
Yeah.

381
00:24:57.536 --> 00:25:03.100
And, you know, as human beings, part of this is going to feel like, oh, obviously.

382
00:25:03.140 --> 00:25:03.260
Yeah.

383
00:25:03.280 --> 00:25:06.522
We talk about childcare and all these things.

384
00:25:06.562 --> 00:25:06.822
Yeah.

385
00:25:06.942 --> 00:25:09.944
And you are experiencing this personally right now and going through that.

386
00:25:10.384 --> 00:25:10.604
Yes.

387
00:25:10.664 --> 00:25:10.944
Okay.

388
00:25:10.964 --> 00:25:11.985
With human beings, for sure.

389
00:25:12.005 --> 00:25:12.505
Yeah.

390
00:25:12.525 --> 00:25:12.665
Yeah.

391
00:25:12.685 --> 00:25:12.825
Yeah.

392
00:25:12.845 --> 00:25:13.085
Yeah.

393
00:25:13.166 --> 00:25:15.207
But we're trying to empirically show this.

394
00:25:16.770 --> 00:25:20.431
in a different species that does not have all of the things that we have.

395
00:25:20.471 --> 00:25:24.712
And so we have to ask what may seem like basic questions.

396
00:25:24.852 --> 00:25:25.132
Yeah.

397
00:25:25.252 --> 00:25:27.792
Because we don't know unless we have the data.

398
00:25:28.032 --> 00:25:28.332
Right.

399
00:25:28.532 --> 00:25:31.813
And that, I think, is what's key about this paper, right?

400
00:25:31.873 --> 00:25:37.254
It's like with human beings, I can just ask mothers, and they'll tell me, yeah, it's hard, right?

401
00:25:37.374 --> 00:25:42.975
But here, I think it's really cool that using data from behavior over 15 years is...

402
00:25:43.897 --> 00:25:52.259
we can empirically show that orangutan mothers are kind of feeling the same stresses that human mothers feel, right?

403
00:25:52.459 --> 00:25:55.940
The last figures that I'm going to talk about I think are really, really cool.

404
00:25:56.360 --> 00:26:02.202
The next step is really asking whether mothers' movements change before and after play, okay?

405
00:26:02.502 --> 00:26:03.762
Like, is it intentional?

406
00:26:03.782 --> 00:26:05.703
You know when we talk about, like, intentionality?

407
00:26:05.743 --> 00:26:05.963
Yeah.

408
00:26:06.123 --> 00:26:06.583
Well, here...

409
00:26:07.556 --> 00:26:16.901
What we do is first we define each mother's reference core using the average of her GPS location from the preceding year.

410
00:26:16.921 --> 00:26:17.922
365 days.

411
00:26:18.262 --> 00:26:19.602
So we've got a range, right?

412
00:26:19.622 --> 00:26:24.105
Because we need to be like, where was this mother before the foraging?

413
00:26:24.145 --> 00:26:25.726
Like, what is her control?

414
00:26:26.726 --> 00:26:29.067
And we're like, okay, this particular mother...

415
00:26:31.453 --> 00:26:34.878
Mindy goes in the green region of the forest.

416
00:26:35.419 --> 00:26:40.246
And another mother, Janak, goes in another region of the forest.

417
00:26:40.306 --> 00:26:44.151
And there's a little bit of overlap because they have similar aged infants.

418
00:26:44.331 --> 00:26:44.552
Okay.

419
00:26:44.572 --> 00:26:45.874
We've already covered that part, right?

420
00:26:46.454 --> 00:26:46.675
Now,

421
00:26:48.107 --> 00:26:52.433
Let's talk about the day when the infant play happened.

422
00:26:52.953 --> 00:26:53.194
Okay?

423
00:26:53.514 --> 00:26:55.937
Let's look at the day before and the day after.

424
00:26:56.558 --> 00:27:04.328
The day before, you can see both mothers actually venturing into each other's zones, which means there's planning involved.

425
00:27:04.448 --> 00:27:04.729
Mm-hmm.

426
00:27:05.149 --> 00:27:07.130
They're scoping the scene out.

427
00:27:07.150 --> 00:27:09.412
Yeah, they're like, let's try.

428
00:27:09.892 --> 00:27:15.456
Let's try to find the other woman who has the kid that's the similar age as mine.

429
00:27:15.856 --> 00:27:17.357
And then let's try to meet up.

430
00:27:17.878 --> 00:27:19.779
That's the yellow.

431
00:27:19.919 --> 00:27:23.121
You see Mindy going into Jannik's zone.

432
00:27:23.422 --> 00:27:26.644
And then the day of the play, they actually play around.

433
00:27:27.084 --> 00:27:29.826
And you see where the play starts and where the play ends.

434
00:27:30.406 --> 00:27:32.868
And they're moving around in their territory.

435
00:27:33.551 --> 00:27:34.412
And the day after.

436
00:27:34.552 --> 00:27:35.193
Incredible.

437
00:27:35.593 --> 00:27:37.154
The day after Mindy's like, you know what?

438
00:27:37.435 --> 00:27:37.795
We're good.

439
00:27:37.815 --> 00:27:38.215
That was enough.

440
00:27:39.056 --> 00:27:39.797
We're good.

441
00:27:39.857 --> 00:27:40.257
We're good.

442
00:27:40.277 --> 00:27:43.100
We're going to hang out entirely in my own zone.

443
00:27:43.160 --> 00:27:47.765
This is incredibly, like, it's so visceral and visual, because you can see it visually.

444
00:27:47.865 --> 00:27:48.085
Yeah.

445
00:27:48.425 --> 00:27:51.188
You know, and again, the pre, during, and post.

446
00:27:51.628 --> 00:27:55.032
This is why anytime you do data capture, like, you want the pre and post.

447
00:27:55.052 --> 00:27:55.852
You want controls.

448
00:27:56.053 --> 00:27:58.175
Of the event you're tracking for.

449
00:27:58.835 --> 00:28:04.941
And it's the distinction between the day before and the day after is so in your face.

450
00:28:04.961 --> 00:28:06.342
Yeah.

451
00:28:07.583 --> 00:28:11.987
And they have statistics to prove this because this is obviously an anecdotal thing, right?

452
00:28:12.007 --> 00:28:12.188
Sure.

453
00:28:13.008 --> 00:28:15.411
One of the statistics is this directional bias.

454
00:28:15.491 --> 00:28:18.754
I mean, these are the statistics on figure 4e.

455
00:28:18.794 --> 00:28:23.198
They're actually showing the difference between like the correlation day before, day after –

456
00:28:23.656 --> 00:28:25.700
There is a huge difference in the behavior.

457
00:28:25.740 --> 00:28:26.000
Yes.

458
00:28:26.721 --> 00:28:29.647
The directional difference is what's really cool.

459
00:28:29.807 --> 00:28:35.437
In the day before, they're moving towards the other...

460
00:28:35.907 --> 00:28:58.573
territory that's the yellow compared to the control on the left yes and the day after they're moving away yeah yeah and it's you know i i one of the things that as we've gone through so many episodes of our last year almost and a half now is really getting a strong understanding of the

461
00:29:01.277 --> 00:29:13.286
the types of data that are gathered and the type of analysis that's done on that data and understanding how those three component parts connect to conclusions you want to draw.

462
00:29:13.326 --> 00:29:15.728
And with animal behavior, especially we've done a lot of these stories.

463
00:29:16.329 --> 00:29:19.231
It's a very complex, all biology is complicated.

464
00:29:20.492 --> 00:29:24.695
Now we're adding behavior on top of just the biological substrate.

465
00:29:25.556 --> 00:29:28.598
And when you see something like that, it's, it's very, it's very,

466
00:29:32.233 --> 00:29:36.457
I don't want to say the word convincing, but it is present.

467
00:29:36.797 --> 00:29:37.037
Yeah.

468
00:29:37.438 --> 00:29:49.328
I mean, to me, it's like it's confirming a notion that I already had, which is that like, of course, these animals are like doing all these things.

469
00:29:49.448 --> 00:29:49.689
Right.

470
00:29:49.749 --> 00:29:57.175
But it's just it's pretty awesome that like with 15 years of data and all of this effort, we can like quantify these things.

471
00:29:57.275 --> 00:29:59.177
I think it's I think it's absolutely amazing.

472
00:29:59.337 --> 00:30:02.798
This makes me think of the whales have culture story we did.

473
00:30:03.498 --> 00:30:10.741
And we were having this very emotive reaction about it because we were showing something that might feel obvious.

474
00:30:10.981 --> 00:30:11.201
Yeah.

475
00:30:11.441 --> 00:30:14.442
If you're just like, of course, whales have culture and things like that.

476
00:30:14.462 --> 00:30:15.522
But like, how do we know?

477
00:30:15.542 --> 00:30:16.023
Yeah.

478
00:30:16.603 --> 00:30:17.583
I think that's kind of cool.

479
00:30:17.683 --> 00:30:21.084
And I think it's cool that we can like say this is how we know.

480
00:30:21.184 --> 00:30:22.085
Yeah.

481
00:30:22.605 --> 00:30:25.486
Like if you were to say, oh, the orangutan mothers are solitary, but they...

482
00:30:26.507 --> 00:30:32.331
The headline, like, just looked at, oh, how do you know that orangutan mothers are setting up playdates for their kids?

483
00:30:32.351 --> 00:30:35.834
Well, actually, here's the one, two, three, four.

484
00:30:36.635 --> 00:30:39.277
Here's how we removed other elements from it.

485
00:30:39.918 --> 00:30:42.980
And it's well-constructed.

486
00:30:43.000 --> 00:30:44.362
The methods are well-constructed.

487
00:30:44.602 --> 00:30:45.062
Yeah, yeah.

488
00:30:45.242 --> 00:30:46.784
I really quite enjoyed this, yeah.

489
00:30:47.284 --> 00:30:54.170
The evidence here supports active maternal facilitation and is consistent with planning.

490
00:30:54.431 --> 00:30:56.138
as we were talking about just now.

491
00:30:56.519 --> 00:30:58.789
Now, the broader scientific contribution...

492
00:30:59.622 --> 00:31:11.292
concerns paternal investment because in evolutionary biology, we think about like investment, meaning effort or resources devoted to an offspring that have a cost or trade-off for the parent, right?

493
00:31:11.352 --> 00:31:14.734
It doesn't require some kind of conscious calculation.

494
00:31:15.835 --> 00:31:19.678
We usually think about it in terms of nursing, caring, and protection.

495
00:31:19.939 --> 00:31:21.520
Those are the obvious examples.

496
00:31:22.200 --> 00:31:22.721
In this case,

497
00:31:29.246 --> 00:31:36.728
as something that we need to talk about when we talk about paternal, parental investment, right?

498
00:31:39.429 --> 00:31:46.771
The fact is that like a mother can shape development by changing which other animals her youngster has a chance to meet.

499
00:31:48.451 --> 00:31:49.751
It's part of a growing movement.

500
00:31:50.411 --> 00:31:56.273
There's been a growing movement in the science of behavioral analysis these days where, um,

501
00:31:58.258 --> 00:32:11.245
The biologists are now starting to study solitary mammals, and they're saying actually play and socialization is important even if these solitary mammals are kind of doing their own thing.

502
00:32:11.546 --> 00:32:13.547
Like you can't count out social...

503
00:32:14.872 --> 00:32:21.157
and social interactions just because, you know, by tradition, these animals are solitary.

504
00:32:21.818 --> 00:32:27.122
Now, there is a second implication, which I found quite intriguing.

505
00:32:27.583 --> 00:32:30.025
It was in the discussion in their paper.

506
00:32:31.746 --> 00:32:34.108
As an Indian, I found it quite intriguing.

507
00:32:34.588 --> 00:32:35.109
Here's why.

508
00:32:36.630 --> 00:32:42.355
So group living means that, you know, young animals...

509
00:32:43.770 --> 00:32:47.532
they meet potential partners almost automatically.

510
00:32:47.893 --> 00:32:55.477
If you live in a group, you get introduced to potential partners pretty regularly.

511
00:32:56.258 --> 00:33:06.284
If you're like an orangutan and you're solitary and your mom's solitary, which means that you are solitary because you're tied to your mom, you're not going to meet a lot of the opposite sex

512
00:33:07.553 --> 00:33:08.974
on an organic basis, right?

513
00:33:08.994 --> 00:33:11.916
Just from Brownian motion of like the orangutans.

514
00:33:12.336 --> 00:33:22.763
So perhaps in a dispersed society, someone has to bear the cost of making these encounters possible.

515
00:33:23.203 --> 00:33:23.463
Right.

516
00:33:23.603 --> 00:33:26.145
Kind of like an arranged marriage system.

517
00:33:26.265 --> 00:33:26.625
Yeah.

518
00:33:26.705 --> 00:33:27.746
Right?

519
00:33:27.966 --> 00:33:30.267
And this is in the discussion of their paper.

520
00:33:30.327 --> 00:33:34.430
They're saying, look, like, you know, in a solitary society,

521
00:33:36.549 --> 00:33:38.410
This is an evolutionary advantage.

522
00:33:38.450 --> 00:33:39.051
Yeah.

523
00:33:39.091 --> 00:33:39.931
Right.

524
00:33:40.031 --> 00:33:44.194
Because the moms like, okay, if we get real nitty gritty down to it, right?

525
00:33:44.234 --> 00:33:46.936
The moms care about their genes being passed on.

526
00:33:46.976 --> 00:33:50.758
This is about evolutionary fitness, Darwinian fitness, so on and so forth.

527
00:33:50.798 --> 00:33:50.979
Right.

528
00:33:51.659 --> 00:33:57.543
That's going to require their kid in a solitary society to like kind of ingratiate themselves with.

529
00:33:58.083 --> 00:34:16.513
individuals of the similar age right what if what if that's the motive yeah yeah yeah right right look look that's the question now we can ask that question now and and the more data we have we can ask that question and that's actually what's next right is whether the investment actually ultimately pays off

530
00:34:17.615 --> 00:34:21.501
In an increased fertility of the offspring.

531
00:34:22.282 --> 00:34:23.985
We don't have the answer to that question yet.

532
00:34:24.505 --> 00:34:32.457
But as an Indian son, that kind of reminded me a little bit about some of the struggles that I've had.

533
00:34:33.498 --> 00:34:35.680
This is so, I love the connection.

534
00:34:35.700 --> 00:34:36.720
You know, it's funny.

535
00:34:36.860 --> 00:34:37.541
Yes, yes.

536
00:34:38.061 --> 00:34:45.865
I will make, and this is a big spoiler alert for anyone who's not watched the new Netflix TV show below.

537
00:34:46.385 --> 00:34:47.966
So, spoiler alert.

538
00:34:48.807 --> 00:34:50.888
Fast forward like a minute or two.

539
00:34:52.229 --> 00:34:54.030
What's funny about what you just said is...

540
00:34:54.770 --> 00:35:08.615
In this show below, which is about this sort of local town in Newfoundland that gets moved around because some rich entities want access to resources where they live.

541
00:35:09.895 --> 00:35:15.997
The main character got moved when he was a kid and now he's an adult in an island next to the island he grew up on.

542
00:35:17.178 --> 00:35:18.760
in this Bay Area.

543
00:35:18.780 --> 00:35:30.833
And one of the plot lines in the show is that they thought where they were moved from was abandoned, but there was actually a family, a family and a few other individuals that had stayed there that were solitary.

544
00:35:31.314 --> 00:35:35.939
And there was a daughter that the dad was struggling to find

545
00:35:36.897 --> 00:36:03.512
a partner for and there was this sort of scheme where the main character's son got whisked away to the solitary island to be the potential partner and this idea of the the parent having this intent around this evolutionary event like one thing there it was it was and it was a core tension point in the show that had anyway but like it

546
00:36:04.252 --> 00:36:11.137
I'm just bringing a popular culture reference to something that you're talking about here because it's literally narrative storylines.

547
00:36:11.357 --> 00:36:11.477
Yeah.

548
00:36:11.497 --> 00:36:13.358
And like the number two show on Netflix right now.

549
00:36:13.418 --> 00:36:13.858
Yeah, yeah.

550
00:36:14.179 --> 00:36:16.320
And that's what the authors are proposing here, right?

551
00:36:16.440 --> 00:36:20.603
In their discussion, they're saying like, look, like as a solitary society –

552
00:36:21.468 --> 00:36:24.549
The moms are also thinking, look, I kind of want grandkids.

553
00:36:24.569 --> 00:36:25.569
I would love that.

554
00:36:25.649 --> 00:36:28.430
And even these orangutan moms are thinking the same thing.

555
00:36:28.470 --> 00:36:35.572
They're like, look, I want orangutan grandkids, which means that you're going to need to meet people of similar age.

556
00:36:36.452 --> 00:36:41.073
It kind of explains the negative correlation between age difference and meet up.

557
00:36:41.433 --> 00:36:42.234
Very good point.

558
00:36:42.274 --> 00:36:46.675
It kind of explains the intentionality behind all of this.

559
00:36:47.636 --> 00:36:48.996
I think – It's quite nice.

560
00:36:49.056 --> 00:36:51.477
So if – I like the theory, you know?

561
00:36:51.817 --> 00:37:00.079
And of course, like there's going to be a lot more data that's going to be involved because what you're going to have to do as scientists, you're going to have to now follow the offspring.

562
00:37:00.199 --> 00:37:00.879
Yeah, right.

563
00:37:01.239 --> 00:37:05.100
And see what their reproductive success is over several decades, right?

564
00:37:05.200 --> 00:37:07.161
This is a long, longitudinal study.

565
00:37:07.241 --> 00:37:07.821
Yes.

566
00:37:07.841 --> 00:37:09.361
A long, long, longitudinal study.

567
00:37:09.421 --> 00:37:10.881
Exactly.

568
00:37:10.961 --> 00:37:13.882
But I'm willing to bet that I think that's what's happening.

569
00:37:17.763 --> 00:37:40.271
very good very good i just want to know because usually in the context of below the mother was dead which is why the dad was doing the job and so that's part of the gender difference there but like obviously here it's like all moms it's the moms yeah it's here it's generally speaking generally speaking general speaking look like we we know how it is okay even in humans it's it's usually the mom that's how it goes yeah

570
00:37:41.032 --> 00:37:42.393
Fantastic first story.

571
00:37:43.174 --> 00:37:51.682
The animal behavior stories are always so interesting because it's kind of not obvious for a layman to understand, like, how would you even go about asking the question?

572
00:37:51.742 --> 00:37:51.922
Yeah.

573
00:37:52.022 --> 00:37:53.123
And they're so clever at it.

574
00:37:53.704 --> 00:38:01.231
We are going to transition into our next story, which is going to be about AI chatbots and epilepsy.

575
00:38:01.251 --> 00:38:03.033
But before we do so, I'm going to do a quick...

576
00:38:03.113 --> 00:38:06.794
Quick commercial break here for some housekeeping.

577
00:38:07.114 --> 00:38:08.855
Welcome for those of you who are joining us.

578
00:38:09.175 --> 00:38:13.596
This is From First Principles, the best science show on the planet.

579
00:38:14.196 --> 00:38:18.518
I have one very amazing announcement about how you can watch the show.

580
00:38:18.918 --> 00:38:26.840
If you're an audio listener or have been an audio listener, we do a lot of overlays and visuals in our show to kind of talk through the fundamental science, the methods.

581
00:38:27.161 --> 00:38:31.662
It really helps you get an understanding of why these stories are so important.

582
00:38:32.142 --> 00:38:42.430
And I'm very proud to announce that finally, after weeks of promo, that the podcast is now available in video on Apple Podcasts.

583
00:38:42.810 --> 00:38:49.775
So for those of you who have iPhones or listen previously on Apple, you can now watch the show.

584
00:38:50.196 --> 00:38:53.198
It's also available in video on Spotify and YouTube.

585
00:38:53.758 --> 00:38:57.661
On social, we have so many clips that are a way that so many of you listen to us.

586
00:38:57.721 --> 00:38:59.463
If you've come from an Instagram clip...

587
00:38:59.883 --> 00:39:03.304
and are now watching the full podcast, leave a comment below letting us know.

588
00:39:03.624 --> 00:39:05.964
You've come from TikTok, Facebook, etc.

589
00:39:06.004 --> 00:39:11.326
We'd love to know that you've made the transition to the full show.

590
00:39:11.606 --> 00:39:12.266
It's super helpful.

591
00:39:12.286 --> 00:39:19.348
You can always get all of the research papers that we talk about at our website, ffppod.com.

592
00:39:19.668 --> 00:39:27.370
And we are now going to move into our second story of the rundown, AI, chatbots, epilepsy.

593
00:39:27.490 --> 00:39:28.830
How do they all connect?

594
00:39:30.887 --> 00:39:33.809
Well, we've covered a lot of AI doom and gloom.

595
00:39:33.929 --> 00:39:34.169
Yes.

596
00:39:35.830 --> 00:39:38.091
Because, well, there's a lot of AI doom and gloom.

597
00:39:38.191 --> 00:39:38.512
Yes.

598
00:39:39.252 --> 00:39:49.418
But this one, I think, is genuinely uplifting because it's the promise of AI that got me excited about it in the first place all those years ago when we got the first chat GPT moment.

599
00:39:49.959 --> 00:39:52.320
Everyone was like, oh, this is going to change the world.

600
00:39:52.620 --> 00:39:53.161
The future.

601
00:39:53.321 --> 00:39:55.062
You know, it's going to be the future.

602
00:39:55.122 --> 00:39:56.503
This isn't the apocalypse.

603
00:39:56.583 --> 00:39:57.843
This is actually great.

604
00:39:58.404 --> 00:39:59.965
Now it seems like, oh, I don't know.

605
00:39:59.985 --> 00:40:00.085
Yeah.

606
00:40:01.186 --> 00:40:04.750
In any case, this particular story, it's about personalized medicine.

607
00:40:04.791 --> 00:40:06.733
This time it's for epilepsy patients.

608
00:40:07.514 --> 00:40:09.817
The story comes out of the University of Pennsylvania.

609
00:40:10.438 --> 00:40:17.887
The headline reads, AI platform allows real-time conversation between patients and their medical devices.

610
00:40:19.133 --> 00:40:20.234
I think it's really, really cool.

611
00:40:20.254 --> 00:40:20.894
This is interesting.

612
00:40:20.934 --> 00:40:27.938
Now, in this case, when I say personalized medicine, I'm talking about really a personalized machine for medicine.

613
00:40:28.298 --> 00:40:28.559
Okay?

614
00:40:28.879 --> 00:40:30.720
We've covered things like this in the past.

615
00:40:30.760 --> 00:40:32.961
We have pacemakers that regulate your heart.

616
00:40:33.442 --> 00:40:36.103
We have cochlear implants that digitize sound.

617
00:40:36.163 --> 00:40:42.347
And we've got deep brain stimulators that suppress the tremors of Parkinson's disease and epilepsy.

618
00:40:42.367 --> 00:40:42.487
Okay.

619
00:40:42.907 --> 00:40:53.850
Now, these are marvels of modern engineering, but they're all entirely silent to the patient that is housing these machines, right?

620
00:40:54.350 --> 00:41:03.253
Like, if you get a medical implant today, the device records your physiological data, and then it stores it in some black box, effectively, right?

621
00:41:03.313 --> 00:41:05.413
Some kind of USB drive or something.

622
00:41:05.794 --> 00:41:09.615
And then months go by, weeks go by, depending on what the treatment is.

623
00:41:09.915 --> 00:41:10.815
You go to the clinic.

624
00:41:11.135 --> 00:41:30.893
the physician downloads whatever the data is and then only the physician really understands what's going on and then he or she or they are going to be going through and understanding and you're essentially locked out of that loop that's not actually really good i never thought about that yeah like the patient who is creating the medical data

625
00:41:31.794 --> 00:41:33.094
is out of the loop, right?

626
00:41:33.114 --> 00:41:39.656
The physician kind of just like uses the patient for the data and then tries to understand the patient from the data.

627
00:41:39.976 --> 00:41:43.297
But there's no real feedback that's happening, right?

628
00:41:44.017 --> 00:41:48.398
Nowhere is this disconnect even more critical than with the treatment of epilepsy.

629
00:41:48.458 --> 00:41:51.878
Globally, around 70 million people live with epilepsy.

630
00:41:52.159 --> 00:41:58.300
Epilepsy, a third of them get seizures that simply do not respond to medication.

631
00:41:58.320 --> 00:41:58.680
Right.

632
00:41:59.080 --> 00:42:05.063
A seizure is basically a temporary episode of abnormal, excessive activity in the brain.

633
00:42:05.143 --> 00:42:06.944
It's like synchronized neural networks.

634
00:42:07.544 --> 00:42:16.889
It can sometimes make a person experience convolutions, brief loss of awareness, unusual sensations, disturbance in memory.

635
00:42:17.329 --> 00:42:20.330
There's a myriad of different things that a seizure can do.

636
00:42:20.830 --> 00:42:27.554
And a single seizure does not automatically mean epilepsy, but if you're prone to this recurrence,

637
00:42:28.475 --> 00:42:33.537
over time, then it's reason to diagnose with epilepsy.

638
00:42:34.077 --> 00:42:39.119
Now, sometimes drugs work, but then there are drug-resistant patients.

639
00:42:39.580 --> 00:42:43.741
As with almost every brain disease, there are drug-resistant patients.

640
00:42:44.602 --> 00:42:50.724
In this case, one of the ways that we try to combat it is with implantable neuromodulation devices.

641
00:42:51.204 --> 00:42:51.925
This is where...

642
00:42:53.239 --> 00:43:02.545
We stick an electrode into the brain that has the ability to read the electrical signals and then pump electricity in.

643
00:43:02.985 --> 00:43:16.374
So what this thing is going to do is read brainwaves, and in principle it can detect seizures using the brainwave profile, and then it can also provide current into the brain to mess with the neural circuits and then perhaps stop the seizure.

644
00:43:18.130 --> 00:43:27.656
And the history of these types of devices has a rich, rich history.

645
00:43:28.777 --> 00:43:35.501
You've got the earliest FDA-approved devices in 1997, something like the vagus nerve stimulation.

646
00:43:36.202 --> 00:43:37.663
This is purely open loop.

647
00:43:37.743 --> 00:43:42.246
This is something where you actually stimulate the vagus nerve that's down near your neck.

648
00:43:42.586 --> 00:43:42.726
Mm-hmm.

649
00:43:43.126 --> 00:43:44.147
Not actually in the brain.

650
00:43:44.947 --> 00:43:47.988
And what you're actually doing, you can think of it as like a lawnmower.

651
00:43:48.489 --> 00:43:49.169
Not a lawnmower.

652
00:43:49.269 --> 00:43:50.370
A lawn sprinkler.

653
00:43:50.750 --> 00:43:50.990
Okay.

654
00:43:51.070 --> 00:43:51.310
Okay?

655
00:43:51.570 --> 00:43:54.652
That just like turns on and then it like sprinkles.

656
00:43:55.872 --> 00:43:57.553
It's doing that but with electricity.

657
00:43:58.694 --> 00:44:00.495
This is what we would call open loop.

658
00:44:00.635 --> 00:44:00.895
Okay.

659
00:44:01.155 --> 00:44:04.316
Because there's no feedback coming from the brain.

660
00:44:04.777 --> 00:44:07.058
And there's no, it's just, it's on a timer.

661
00:44:07.078 --> 00:44:07.298
Yeah.

662
00:44:08.463 --> 00:44:09.464
And it's just like going.

663
00:44:09.905 --> 00:44:14.830
It's just stimulating the vagus nerve at regular intervals.

664
00:44:15.451 --> 00:44:16.752
It doesn't know if the grass is growing.

665
00:44:16.792 --> 00:44:19.275
It doesn't know what patches of the grass are growing more than others.

666
00:44:19.395 --> 00:44:20.236
Any of that kind of stuff.

667
00:44:20.316 --> 00:44:23.540
Any of that kind of stuff is just blindly implementing.

668
00:44:23.680 --> 00:44:24.461
I know it needs water.

669
00:44:24.641 --> 00:44:24.841
Yeah.

670
00:44:25.682 --> 00:44:26.563
And, and, and.

671
00:44:28.497 --> 00:44:29.378
It had its effects.

672
00:44:29.458 --> 00:44:29.698
Sure.

673
00:44:29.939 --> 00:44:30.539
And it was great.

674
00:44:31.260 --> 00:44:36.044
2013, there was a massive leap into closed-loop systems.

675
00:44:36.424 --> 00:44:52.818
So this is the RNS system, which is the responsive neurostimulation, meaning I've got some kind of read mechanism that is going to maybe identify that a seizure is happening and then only stimulate when the seizure is happening, right?

676
00:44:53.099 --> 00:44:55.501
So this is where the responsive thing comes in.

677
00:44:55.521 --> 00:44:55.661
Mm-hmm.

678
00:44:57.465 --> 00:45:01.708
Now, instead of open loop, we've got a closed loop kind of thing, right?

679
00:45:02.168 --> 00:45:04.450
The device knows when the seizure is coming.

680
00:45:05.191 --> 00:45:10.554
But at the same time, it's not closed in the sense that the patient is aware of what is going on, right?

681
00:45:10.935 --> 00:45:15.018
You can't tell the patient, maybe you need to sit down.

682
00:45:15.658 --> 00:45:16.899
Maybe you need to pull the car over.

683
00:45:16.959 --> 00:45:18.000
Alert, alert.

684
00:45:18.060 --> 00:45:19.181
Like, yeah, alert alert.

685
00:45:19.921 --> 00:45:21.183
You're about to have a seizure.

686
00:45:21.343 --> 00:45:21.543
Right.

687
00:45:21.823 --> 00:45:24.006
Which would be obviously great.

688
00:45:24.306 --> 00:45:24.906
That would be great.

689
00:45:25.347 --> 00:45:28.470
And that's where the breakthrough currently comes in hand.

690
00:45:28.931 --> 00:45:32.795
The patient was never involved in any of the communication cycle of this.

691
00:45:32.895 --> 00:45:34.236
Obviously, we weren't there yet.

692
00:45:34.437 --> 00:45:34.657
Yeah.

693
00:45:35.077 --> 00:45:35.318
But...

694
00:45:35.858 --> 00:45:41.545
Because it required like a human being with a trained experience to understand all of these things.

695
00:45:42.066 --> 00:45:50.236
Now, with the advent of machine learning, you can put the patient in the loop of his or her own medical degree.

696
00:45:50.980 --> 00:45:51.500
Destiny.

697
00:45:51.560 --> 00:45:51.960
Yeah.

698
00:45:52.361 --> 00:45:53.381
It's pretty incredible.

699
00:45:53.401 --> 00:46:03.726
This is a really big deal because I think it also – there's this whole era of like quantified self and all this medical information from all these devices we have around us.

700
00:46:03.986 --> 00:46:06.267
It's going to help you improve your well-being.

701
00:46:06.967 --> 00:46:12.470
But part of – and that's a different thing than medical and clinical treatments and solutions.

702
00:46:12.490 --> 00:46:13.711
I'm not saying they're the same thing.

703
00:46:13.991 --> 00:46:21.355
But there's been this big promise with the rise of technology as bringing us closer to understanding our own bodies in real time.

704
00:46:22.516 --> 00:46:32.462
And this is a perfect example of how that can have a hugely important impact for epilepsy.

705
00:46:32.482 --> 00:46:32.582
Yeah.

706
00:46:32.842 --> 00:46:33.523
Yeah.

707
00:46:34.043 --> 00:46:36.765
And I think it's not only limited to epilepsy.

708
00:46:37.066 --> 00:46:48.776
This particular case is taking epilepsy as kind of like this microcosm because it's such a rich data set in terms of like, you know, you get intracranial stimulation.

709
00:46:49.316 --> 00:46:50.958
There's a lot of data that's going on.

710
00:46:51.118 --> 00:46:56.923
It requires a lot of expertise that perhaps a machine learning model can elucidate.

711
00:46:57.315 --> 00:47:00.116
But it's absolutely an incredible paper.

712
00:47:00.156 --> 00:47:02.217
It's a team out of the University of Pennsylvania.

713
00:47:02.537 --> 00:47:09.500
And what they've done is fundamentally re-imagine the patient-device relationship by bridging

714
00:47:10.565 --> 00:47:18.948
this EEG that comes with the patients of epilepsy with agentic large language models, right?

715
00:47:18.988 --> 00:47:19.869
That's the key.

716
00:47:19.889 --> 00:47:27.552
You're tying in the brainwaves from inside the brain of these epileptic patients with now agentic models

717
00:47:28.472 --> 00:47:31.093
that use large language models.

718
00:47:31.173 --> 00:47:37.896
This is the promise that has been, again, large language models are not just a tool in the context of a chat bot.

719
00:47:38.076 --> 00:47:38.336
Yes.

720
00:47:38.916 --> 00:47:40.037
Is kind of the point here.

721
00:47:40.237 --> 00:47:43.098
Yeah, they've built a kind of interface that allows...

722
00:47:43.488 --> 00:47:52.797
the patient to maybe interface with the chatbot, but the chatbot now has access to all of the data from within the patient's brain itself.

723
00:47:52.957 --> 00:47:53.298
Which is...

724
00:47:53.318 --> 00:47:54.739
It's incredible, dude.

725
00:47:54.899 --> 00:47:55.340
Yeah.

726
00:47:55.800 --> 00:48:03.168
Now, I actually, like, you know, from a personal standpoint, my PhD had to do with...

727
00:48:03.930 --> 00:48:05.151
electrophysiology, right?

728
00:48:05.211 --> 00:48:17.041
Where, you know, you're taking electrodes, you're sticking it inside the rodent brain while the rodent is freely behaving inside mazes, inside, you know, behavioral tasks.

729
00:48:18.443 --> 00:48:18.743
One of the...

730
00:48:21.389 --> 00:48:30.858
The bane of my existence during my PhD was deciphering between real neural signal and all of the artifacts that come with behavior.

731
00:48:31.599 --> 00:48:36.423
Because if the rodent is chewing, that's going to cause electrical signal, right?

732
00:48:36.964 --> 00:48:41.929
If the rodent is going into like a micro nap, that's going to cause electrical signal.

733
00:48:43.563 --> 00:48:46.790
This problem is real even for epilepsy patients, right?

734
00:48:46.830 --> 00:48:52.884
Because if you've got electrodes that are going into the brain, there's a giant signal-to-noise problem.

735
00:48:54.232 --> 00:48:58.635
Because brainwaves are incredibly chaotic and highly variable.

736
00:48:58.995 --> 00:49:05.338
Here we're seeing brainwaves where chewing artifacts come in.

737
00:49:05.478 --> 00:49:08.580
All of those jagged things, that's when the person is chewing.

738
00:49:08.760 --> 00:49:09.661
That's crazy.

739
00:49:09.901 --> 00:49:22.028
If you have a really terrible machine learning classifier that just thinks you're having a seizure every time you chew, that's not going to be great for two reasons.

740
00:49:22.815 --> 00:49:28.976
There's this thing called alarm fatigue, which is basically the boy who cried wolf, right?

741
00:49:30.637 --> 00:49:36.018
If every time the thing just says, hey, you're going to have a seizure and you're not having a seizure, there's so many false positives.

742
00:49:36.438 --> 00:49:38.819
Well, then you're going to just disregard the thing, right?

743
00:49:39.939 --> 00:49:43.680
And two, if you're kind of someone who has seizures and has anxiety...

744
00:49:44.580 --> 00:49:50.527
That's going to add to your anxiety and it's going to like be detrimental to your wellbeing at the end of the day.

745
00:49:50.567 --> 00:49:55.813
Like the whole point is to get better and like having these alerts is making it worse.

746
00:49:56.313 --> 00:49:56.514
Right.

747
00:49:56.554 --> 00:50:00.018
You're constantly put in a state of a heightened, exactly.

748
00:50:00.038 --> 00:50:00.979
You know, um,

749
00:50:01.319 --> 00:50:02.280
Defense mode, anyway.

750
00:50:02.300 --> 00:50:02.460
Right.

751
00:50:02.740 --> 00:50:19.631
And the reason why this is so hard is because historically, tuning these algorithms requires an expert neurologist to manually go through all of the data from your brain and be like, yep, that was you chewing, that was you not chewing, that was an actual seizure right there.

752
00:50:20.252 --> 00:50:20.432
Right?

753
00:50:20.632 --> 00:50:23.734
You got to have some MD be like, that was an actual seizure.

754
00:50:24.435 --> 00:50:26.776
We got to tag that, not tag the other thing.

755
00:50:27.196 --> 00:50:27.557
Now...

756
00:50:28.923 --> 00:50:31.004
You flip the script entirely, right?

757
00:50:31.264 --> 00:50:38.807
You're turning the patient into a collaborator for training the neural network, for training the machine learning model.

758
00:50:38.867 --> 00:50:39.648
This is so good.

759
00:50:40.728 --> 00:50:42.069
Dude, I think it's so cool.

760
00:50:42.329 --> 00:50:45.890
I mean, there are things that are involved here that they were very careful about.

761
00:50:45.910 --> 00:50:47.431
You've got to be HIPAA compliant.

762
00:50:47.971 --> 00:50:56.115
And these guys at UPenn, they made a system that features a HIPAA-compliant instance of the GPT-40 mini.

763
00:50:56.755 --> 00:51:00.237
So this is one of those mini models that OpenAI has, right?

764
00:51:00.457 --> 00:51:04.099
Which means that you can, like, put it on a local thing.

765
00:51:04.139 --> 00:51:04.299
Right.

766
00:51:04.319 --> 00:51:05.259
It doesn't need to go to the cloud.

767
00:51:05.339 --> 00:51:05.699
It doesn't.

768
00:51:05.739 --> 00:51:05.940
Yeah.

769
00:51:06.000 --> 00:51:06.200
Yeah.

770
00:51:06.220 --> 00:51:08.241
So now patient data is not going to the cloud.

771
00:51:08.561 --> 00:51:08.921
And, like...

772
00:51:09.281 --> 00:51:33.343
you know the the the levant al poge and buck master uh a problem nightmare where right where it's like they if you guys don't know the navier stokes episode right buck master was talking to open ai codex about his solution for navier stokes and now lo and behold lo and behold a week later yeah open ai came out with navier stokes right like i

773
00:51:33.623 --> 00:51:34.004
I don't know.

774
00:51:34.024 --> 00:51:35.268
I don't know how that happened.

775
00:51:35.528 --> 00:51:39.961
Well, in this case, they've got a GPT-40 mini that is local, right?

776
00:51:40.021 --> 00:51:42.187
So there's no way that HIPAA...

777
00:51:43.305 --> 00:51:45.706
That the data from the patient is going into the cloud.

778
00:51:46.406 --> 00:51:53.368
And it's not trivial to be able to execute something like this on one of these smaller mini models.

779
00:51:53.408 --> 00:51:55.028
Like that's been part of the challenge.

780
00:51:55.068 --> 00:51:55.508
Exactly.

781
00:51:55.528 --> 00:51:57.949
Generally is you want the capability.

782
00:51:58.029 --> 00:51:58.189
Yep.

783
00:51:58.309 --> 00:52:03.690
As you go to smaller parameter models, you lose some of the capability you're looking for in an LLM.

784
00:52:03.950 --> 00:52:09.932
So to be able to work within that constraint and still get the solution is, I think, an interesting part of the solution.

785
00:52:11.095 --> 00:52:12.919
the success here.

786
00:52:13.119 --> 00:52:13.560
Exactly.

787
00:52:13.901 --> 00:52:19.152
And one thing that I want to note here is the LLM is not actually doing the medical diagnosing.

788
00:52:19.921 --> 00:52:20.201
Okay?

789
00:52:20.641 --> 00:52:27.863
They've got a specific model, a neural network that they've trained for earlier purposes that is doing the medical diagnosis.

790
00:52:27.883 --> 00:52:34.224
Like, there's a separate model that is, like, taking in the brainwaves and being like, hey, is this epilepsy?

791
00:52:34.244 --> 00:52:34.784
Is this not?

792
00:52:34.804 --> 00:52:35.424
Blah, blah, blah.

793
00:52:36.184 --> 00:52:41.865
The LLM is being used to interface with the patient because that's the whole point of the language, right?

794
00:52:41.905 --> 00:52:44.906
It's like, as a patient, I want to use natural language.

795
00:52:46.023 --> 00:52:48.144
To interact with whatever data modality.

796
00:52:48.164 --> 00:52:49.004
This is so crazy.

797
00:52:49.444 --> 00:52:49.684
Right?

798
00:52:50.124 --> 00:52:51.965
So it's really agent-based.

799
00:52:52.285 --> 00:52:54.245
Like, there's the agent that's the LLM.

800
00:52:54.806 --> 00:52:58.147
Then there's the agent that's, like, has nothing to do with English.

801
00:52:58.367 --> 00:52:58.507
Right.

802
00:52:58.527 --> 00:53:00.907
But just trying to classify brainwaves.

803
00:53:02.988 --> 00:53:03.268
Mate.

804
00:53:03.695 --> 00:53:06.678
Yeah, I think it's really cool.

805
00:53:06.718 --> 00:53:07.678
So you can imagine, right?

806
00:53:07.738 --> 00:53:12.062
Like the deep learning model detects anomalous brain synchrony.

807
00:53:12.823 --> 00:53:17.226
The AI then texts the patient, hey, did you just experience a seizure?

808
00:53:18.267 --> 00:53:18.727
This is so crazy.

809
00:53:18.828 --> 00:53:23.131
And then the patient is like, no, that was just me like jumping up and down.

810
00:53:23.411 --> 00:53:25.173
Or actually that felt weird.

811
00:53:25.193 --> 00:53:26.154
That might've been a seizure.

812
00:53:26.314 --> 00:53:31.378
And now it's, again, you're taking the cycle of training goes from,

813
00:53:31.778 --> 00:53:33.780
Patient collects data over X period of time.

814
00:53:34.080 --> 00:53:35.221
It goes to the physician.

815
00:53:35.461 --> 00:53:42.426
The physician spends X period of time with some assistant or someone else that's tagging manually.

816
00:53:43.106 --> 00:53:48.530
The physician doesn't even know the personal experience of the patient who's experiencing the seizure.

817
00:53:48.570 --> 00:53:49.531
So they're just looking at it.

818
00:53:49.551 --> 00:53:57.196
And based on our past things, this is what normally – It actually – it has both the combination of bringing the patient into their own –

819
00:53:59.017 --> 00:54:21.378
giving them a sense of control in their own experience but it also has a very functional benefit to improving the detection because the part of the challenge is always patient translating their experience to the physician and then like that back and forth which there's it's a very noisy signal the noise there is also not very right right right right

820
00:54:22.058 --> 00:54:23.959
I mean, it's absolutely incredible.

821
00:54:24.059 --> 00:54:28.301
I think we're standing at a new dawn of medical technology.

822
00:54:28.321 --> 00:54:40.086
We're moving away from passive hardware that treats patients as biological substrates and towards semi-autonomous conversational partners that help us navigate our own biology.

823
00:54:40.486 --> 00:54:42.807
Like, imagine you have agency over...

824
00:54:44.853 --> 00:54:47.858
your own kind of outcome in some sense.

825
00:54:48.098 --> 00:54:49.720
And I want to make a small note here.

826
00:54:50.241 --> 00:54:58.293
There's, I know there's a lot of people have a lot of feelings about specific providers of foundational models.

827
00:54:58.373 --> 00:54:59.414
Yeah.

828
00:54:59.575 --> 00:55:00.776
I think part of the point here is, is,

829
00:55:01.297 --> 00:55:04.259
This could be done on an open source mini model.

830
00:55:04.339 --> 00:55:07.100
It's not open AI mini specific.

831
00:55:07.300 --> 00:55:09.482
It happens to be what they used in this case.

832
00:55:09.782 --> 00:55:12.363
The large language model is doing a very small amount.

833
00:55:12.684 --> 00:55:12.904
Right.

834
00:55:12.944 --> 00:55:13.444
Honestly.

835
00:55:13.484 --> 00:55:13.644
Right.

836
00:55:13.684 --> 00:55:13.864
Right.

837
00:55:14.485 --> 00:55:24.070
But I just wanted to make that distinction that now that we understand it can be done, this can be extrapolated to whatever provider you prefer.

838
00:55:24.090 --> 00:55:25.391
Yes, exactly.

839
00:55:25.651 --> 00:55:25.871
Yeah.

840
00:55:26.679 --> 00:55:27.280
It's incredible.

841
00:55:27.360 --> 00:55:35.288
I want to take a brief note about how they actually did it because even though this is a rundown episode, I want to get into some of the— We got to hit two hours.

842
00:55:35.328 --> 00:55:35.988
You know how it is.

843
00:55:36.469 --> 00:55:36.789
Yeah.

844
00:55:36.809 --> 00:55:37.089
Yeah.

845
00:55:37.250 --> 00:55:37.910
I'm sorry, guys.

846
00:55:38.050 --> 00:55:43.536
But these Penn researchers, they recruited 13 adults in the epilepsy monitoring unit.

847
00:55:43.556 --> 00:55:49.862
The setting here matters because it's an epilepsy monitoring unit, which is one of these things where you go in—

848
00:55:50.242 --> 00:55:51.343
High quality recording.

849
00:55:51.403 --> 00:55:52.703
There's clinical supervision.

850
00:55:53.123 --> 00:55:54.764
There's detailed event review.

851
00:55:55.264 --> 00:56:04.808
And also sometimes they even take reduced medication to induce more and more epilepsy and more and more seizures to actually get more and more data.

852
00:56:05.108 --> 00:56:06.929
So that matters a little bit.

853
00:56:07.109 --> 00:56:07.369
Right?

854
00:56:07.389 --> 00:56:08.030
Sure.

855
00:56:08.250 --> 00:56:10.172
The platform, actually, there's three parts.

856
00:56:10.552 --> 00:56:13.234
There's a stream of EEG and clinical information.

857
00:56:13.795 --> 00:56:17.838
And then there's software that extracts whatever the useful measurements are.

858
00:56:18.199 --> 00:56:22.162
And then that goes into a phone interface that then connects to the patients.

859
00:56:22.442 --> 00:56:29.048
And the patients can be just in the epilepsy monitoring unit with their phone talking to this agent.

860
00:56:29.381 --> 00:56:33.422
That is also talking to like the models that are interpreting data from his brain.

861
00:56:33.743 --> 00:56:34.763
This is so good.

862
00:56:35.283 --> 00:56:37.304
Dude, it's absolutely incredible.

863
00:56:37.324 --> 00:56:38.344
It's clever too.

864
00:56:38.424 --> 00:56:38.624
Yeah.

865
00:56:38.684 --> 00:56:39.445
It's so obvious.

866
00:56:39.465 --> 00:56:41.045
Like it's so obvious in retrospect.

867
00:56:41.065 --> 00:56:41.445
Yeah.

868
00:56:41.565 --> 00:56:43.966
And there's so many multiple modalities here.

869
00:56:44.046 --> 00:56:50.209
Like there's a system that estimates sleep stages so that that is not confounding the algorithm.

870
00:56:50.509 --> 00:56:56.411
There's a system that estimates medication because there's medication information that

871
00:56:56.771 --> 00:56:58.173
in the patient's notes, right?

872
00:56:58.533 --> 00:57:12.832
That'll tell you, okay, the patient took the medication at this time, and then there's a model that tells you how the medication is deteriorating with time, and that's going to inform this ensemble of machine learning

873
00:57:14.190 --> 00:57:14.570
Models.

874
00:57:15.091 --> 00:57:19.634
We have all these inputs into the model that gets to decide, do I set the alarm on or off?

875
00:57:20.394 --> 00:57:28.139
And you also now have the feedback of the patient, which is arguably one of the most important inputs into that, let's call it detection system.

876
00:57:29.660 --> 00:57:33.303
And to be able to have them all connected and streaming real time.

877
00:57:33.623 --> 00:57:35.826
Yeah, it's absolutely very, very cool.

878
00:57:35.866 --> 00:57:39.371
And the result is a structured collection of measurements, right?

879
00:57:39.652 --> 00:57:43.678
You can have these measurements be queried, plotted, and discussed.

880
00:57:44.038 --> 00:57:50.267
And the chatbot and the seizure detector are different models that are doing different jobs, but they're all coming together.

881
00:57:50.614 --> 00:57:52.616
for the seamless sort of loop.

882
00:57:53.076 --> 00:57:56.159
The patient has a brain interface.

883
00:57:56.899 --> 00:58:05.307
It goes through all of the stuff, but the patient only interacts with it using natural language, which means that, you know, it's kind of a democratized system.

884
00:58:05.347 --> 00:58:10.851
The patient doesn't have to understand what the machine learning nonsense is happening, right?

885
00:58:11.011 --> 00:58:16.256
The chatbot is going to go ahead and ask it questions and get the feedback, right?

886
00:58:16.709 --> 00:58:25.411
This is something we talked about at the end of our Navier Stokes episode, which is when we talk about these concepts of AI, we have to think about them as systems, not as individual component parts.

887
00:58:25.451 --> 00:58:30.432
Because this is a perfect example of why that framework of how to think about these things matters.

888
00:58:30.492 --> 00:58:39.033
Because they can be individual modules with discrete functions that can work in collaboration towards an end goal.

889
00:58:39.414 --> 00:58:40.974
And they're also not just LLMs.

890
00:58:41.294 --> 00:58:42.234
There's a whole variety.

891
00:58:42.434 --> 00:58:46.395
The word AI has a whole ecosystem of meaning and capability.

892
00:58:46.615 --> 00:58:47.276
Yeah.

893
00:58:47.536 --> 00:58:50.618
And this is an example of combining two things, the machine learning and the LLM.

894
00:58:50.758 --> 00:58:51.239
Exactly.

895
00:58:52.079 --> 00:58:53.921
For an outcome that has clinical relevance.

896
00:58:54.001 --> 00:58:54.441
Yeah.

897
00:58:54.601 --> 00:59:01.366
And actually, to piggyback back off of that, you can ask, like, what exactly is this machine?

898
00:59:02.587 --> 00:59:03.848
There's a bunch of agents, right?

899
00:59:04.395 --> 00:59:05.496
What exactly is happening?

900
00:59:05.777 --> 00:59:10.121
Well, let's suppose a patient asks how much sleep they got last night and how good it was.

901
00:59:10.542 --> 00:59:15.247
Well, they've published kind of a flowchart of how things happen.

902
00:59:16.408 --> 00:59:17.970
You receive the patient message.

903
00:59:18.530 --> 00:59:21.634
Now, if the message is unsafe, then you don't do anything.

904
00:59:21.934 --> 00:59:25.558
You maybe alert the physician who's in charge and things like that.

905
00:59:25.918 --> 00:59:29.383
But if it's safe, then you have a message router.

906
00:59:29.543 --> 00:59:29.824
Yes.

907
00:59:30.084 --> 00:59:33.008
And the router asks, okay, is this a general chat?

908
00:59:33.629 --> 00:59:36.113
Or do I need to actually do a data query?

909
00:59:36.734 --> 00:59:36.974
Yep.

910
00:59:36.994 --> 00:59:37.214
Right?

911
00:59:37.515 --> 00:59:38.696
If it's a data query...

912
00:59:39.717 --> 00:59:43.721
Then it actually instantiates a SQL query.

913
00:59:43.881 --> 00:59:44.182
Yeah.

914
00:59:44.362 --> 00:59:56.414
Like there's an agent that goes, okay, I'm going to go into the SQL database of all of the data on this patient and like instantiate, get that bit of EEG or whatever brainwave.

915
00:59:56.835 --> 01:00:01.860
And then now instantiate another thing that asks maybe a clarifying question that asks...

916
01:00:02.653 --> 01:00:05.297
What are the tools that I need to understand this data?

917
01:00:05.758 --> 01:00:10.745
And then you go back and send the AI message through the chat bot to the phone.

918
01:00:11.627 --> 01:00:11.827
Right?

919
01:00:12.188 --> 01:00:16.274
It's able to actually look at the raw data, analyze the raw data, and...

920
01:00:17.287 --> 01:00:19.528
Kind of understand what to do next.

921
01:00:19.848 --> 01:00:23.929
And from this flowchart I'm seeing there's actually also two safety gates.

922
01:00:24.370 --> 01:00:28.471
One at the front end to moderate the patient message if it's unsafe.

923
01:00:28.831 --> 01:00:36.434
And then once all of the processing is done, like you just mentioned, it also moderates the AI response if it's unsafe prior to getting to the patient.

924
01:00:36.454 --> 01:00:39.175
So, again, they've thought about both.

925
01:00:39.615 --> 01:00:42.597
the front and back end from a safety perspective.

926
01:00:43.217 --> 01:00:58.767
And, you know, medical researchers, clinical researchers are some of the most anal retentive about details like this in ways that maybe people wish the frontier labs would be more...

927
01:00:59.307 --> 01:01:00.108
conscientious.

928
01:01:00.168 --> 01:01:27.424
100% dude 100% and one last thing that I want to talk about is the fact that the system can ask questions back so it's not just that the patient asks questions for example suppose it has detected a possible seizure, it sends the patient an alert asking hey did you consciously experience a seizure sometimes you have these kinds of mini seizures where you don't really experience anything and sometimes you do

929
01:01:28.464 --> 01:01:32.466
In any case, the patient answers yes, no, or uncertain.

930
01:01:32.906 --> 01:01:45.652
And in machine learning, that means that you've established a label to the EEG that the model had triggered as something that possibly could use a label.

931
01:01:46.272 --> 01:01:54.956
Now, usually you only want high-quality labels, but actually, even in the case of these low-quality, weak labels, you can do a bit of learning.

932
01:01:57.667 --> 01:01:58.868
called distillation.

933
01:02:00.190 --> 01:02:00.790
Here's the idea.

934
01:02:01.611 --> 01:02:03.313
You've got a pre-trained model, right?

935
01:02:03.533 --> 01:02:10.880
Let's say you have a bunch of data on epilepsy, and you've got EEG recordings, and you know when the epileptic seizures happened.

936
01:02:11.241 --> 01:02:16.626
So you've got this model that is pre-trained on thousands of hours of epileptic data.

937
01:02:17.427 --> 01:02:21.511
And now you want to sort of fine-tune it to this specific patient, right?

938
01:02:21.531 --> 01:02:27.215
Because this specific patient's brain is different, and the way that they experience seizures is different, and so on and so forth.

939
01:02:27.696 --> 01:02:29.838
You can use this technique called distillation.

940
01:02:30.859 --> 01:02:32.220
That's going to do two things.

941
01:02:32.460 --> 01:02:42.649
One, it's going to try to get it right for that specific patient, but at the same time, it's going to penalize changing up what you already know, right?

942
01:02:43.089 --> 01:02:44.310
There's a two-fold principle.

943
01:02:45.910 --> 01:02:46.871
thing that you want to do.

944
01:02:47.511 --> 01:02:58.760
You want to gear towards what the patient specifically is experiencing, but you don't want to forget all of the stuff that you've learned from general epilepsy data.

945
01:02:59.721 --> 01:03:02.303
And that's this idea of model distillation.

946
01:03:02.403 --> 01:03:04.725
It's a penalty for changing too much

947
01:03:05.385 --> 01:03:12.627
on the basis of limited evidence, because you have all this hours of evidence, and now you're getting, what, maybe a few hours of evidence on this patient.

948
01:03:12.647 --> 01:03:20.330
You can't totally upend your world model based on everything that you've sort of realized beforehand, right?

949
01:03:21.230 --> 01:03:25.432
And now the question is, do the models actually do better?

950
01:03:26.247 --> 01:03:27.007
In fact, they do.

951
01:03:27.728 --> 01:03:33.471
The models do better from this closed loop arrangement between the patient and the training, right?

952
01:03:33.791 --> 01:03:46.697
On the left-hand side, what we're seeing is from baseline, that's called the area under the PRC curve, the area under the precision recall curve.

953
01:03:46.797 --> 01:03:47.018
It's a...

954
01:03:47.958 --> 01:03:53.260
It's effectively like how well is the machine learning model doing in terms of precision and recall.

955
01:03:53.820 --> 01:03:59.822
You can see that the baseline model is highlighted in green and the fine tuning steadily increases.

956
01:04:01.062 --> 01:04:08.045
The F1 score, again, another metric for how well precision and recall is working, is increasing.

957
01:04:08.992 --> 01:04:12.153
The false alarm per hour is decreasing.

958
01:04:12.753 --> 01:04:14.674
And that's crucially very important.

959
01:04:15.374 --> 01:04:20.335
We are on the foothills of continual learning in clinical application.

960
01:04:20.936 --> 01:04:24.837
And continual learning is a very specific thing.

961
01:04:25.557 --> 01:04:28.198
I know that this is not literally continual learning.

962
01:04:28.898 --> 01:04:30.699
But that's why I said foothills.

963
01:04:31.299 --> 01:04:35.500
Because part of the challenge with all of these model architectures is they're all pre-trained.

964
01:04:36.410 --> 01:04:41.173
And so they can't learn from new information is the argument that's said a lot.

965
01:04:41.313 --> 01:04:46.076
And even in this case, right, they're scaffolding off of pre-trained models.

966
01:04:46.097 --> 01:04:48.438
The LLM is the ChatGPT4O Mini.

967
01:04:49.199 --> 01:04:51.180
That's a pre-trained model.

968
01:04:51.420 --> 01:04:54.482
The seizure detector is a pre-trained model.

969
01:04:55.583 --> 01:05:18.095
But especially because of the complexity of biological variation and individualized medicine, you are going to kind of necessarily need this foothills of continual learning capability, which is can we allow this distillation to happen on a per-patient basis in a way that takes the benefit of all of the massive amounts of data

970
01:05:18.715 --> 01:05:26.558
but can correctly weight how much to take the influence of the individual patient in future decision-making.

971
01:05:27.138 --> 01:05:33.540
It's a very narrow thread to pull, but the...

972
01:05:34.760 --> 01:05:49.962
The efficacy that we've seen in this study in particular is very encouraging in, again, being able to now kind of take something out of the box and make it evolutionary in tandem with a patient.

973
01:05:49.982 --> 01:05:50.162
Mm-hmm.

974
01:05:50.302 --> 01:05:55.103
Yeah, I think it's a really, really cool study.

975
01:05:55.564 --> 01:05:56.484
It's a small study.

976
01:05:56.844 --> 01:06:00.125
It's short, and it was conducted at only one center.

977
01:06:00.705 --> 01:06:07.807
The participants, they need sufficient English, cognitive capacity, and smartphone ability, I guess.

978
01:06:08.608 --> 01:06:16.050
So you could think of that as a kind of limitation, but I think a convincing next study would be to measure the outcomes that matter to patients

979
01:06:16.570 --> 01:06:25.215
For example, like the seizure burden, the quality of life, anxiety, the amount of emergency visits, the clinical workload.

980
01:06:27.696 --> 01:06:37.181
I think the fact that we're seeing this, though, is very encouraging to me because, you know, we hear a lot of AI doom and gloom.

981
01:06:37.381 --> 01:06:38.762
You know, this is some good news.

982
01:06:38.842 --> 01:06:41.644
This is what was promised to me.

983
01:06:42.432 --> 01:07:09.980
okay yep when when when ai came in they were like we're gonna make medicine better we're gonna make your health better that's what they said this is what was promised this is what was promised right so we're starting to see it slowly come out and it's not just the frontier lab companies that are using these technologies um and a fascinating second story and not to burden you with another ai related story but unfortunately uh

984
01:07:11.381 --> 01:07:15.844
When we were in our 20s, the phrase was software is eating the world.

985
01:07:16.485 --> 01:07:19.827
And now in our 30s, AI is eating the world.

986
01:07:19.988 --> 01:07:26.953
And so there are just a multitude of areas, especially in science research, where it comes up.

987
01:07:27.113 --> 01:07:37.021
And we are going to now move into our last story, which is going to be a little bit of a zoom in, but also a zoom out of what we mean by this whole idea of

988
01:07:38.123 --> 01:07:45.130
automated science research or AI going end to end from hypothesis to clinical outcome.

989
01:07:45.150 --> 01:07:58.605
And I think this will be an interesting, there's a story that came out today that I think will give us a nice on-ramp to discuss what is the current state of play in what is a pipe dream at this point around AI scientific discovery.

990
01:07:59.205 --> 01:08:04.551
So earlier today, we'd planned this episode and then this story dropped.

991
01:08:04.591 --> 01:08:06.012
I was like, oh, this will be interesting.

992
01:08:06.593 --> 01:08:14.561
And so this is a last minute edition about the other frontier lab that we did not talk about in our last story, which is OpenAI.

993
01:08:14.601 --> 01:08:17.504
This is an anthropic story.

994
01:08:18.306 --> 01:08:24.530
about scorpion binders and the path from AI to real-world science.

995
01:08:24.590 --> 01:08:27.051
And so what are we talking about?

996
01:08:27.231 --> 01:08:33.635
We've seen all these stories, like we mentioned every week, about AI is going to change science and make all these new discoveries.

997
01:08:34.336 --> 01:08:43.882
And today, Anthropic published this research article, How Claude Designed Protein Binders Against Scorpion Neurotoxins.

998
01:08:44.842 --> 01:08:53.632
And it was a collaboration with Citrini Research and the laboratories AdaptiBio and Twist Bioscience.

999
01:08:54.373 --> 01:08:58.758
The idea is they're trying to design proteins for these dangerous venoms that are

1000
01:08:59.734 --> 01:09:01.875
And scorpion venom potentially leading to a treatment.

1001
01:09:02.315 --> 01:09:11.758
So interestingly, there are some antivenoms that exist, but they're very specific to different species.

1002
01:09:11.778 --> 01:09:13.178
They're species specific.

1003
01:09:13.278 --> 01:09:16.999
And even with any individual species, there's also some regional things.

1004
01:09:17.039 --> 01:09:20.341
And so they're very complex, as all the biology we talk about.

1005
01:09:20.421 --> 01:09:21.901
It's very complicated.

1006
01:09:22.341 --> 01:09:28.003
And so one of the holy grails is this idea of discovering a universal antivenom.

1007
01:09:28.984 --> 01:09:29.825
Uh, obviously.

1008
01:09:29.905 --> 01:09:30.145
Yeah.

1009
01:09:30.165 --> 01:09:30.305
Yeah.

1010
01:09:30.345 --> 01:09:30.525
Yeah.

1011
01:09:30.726 --> 01:09:40.993
Cause then you currently see the anti-venoms that have been created are not in underserved areas or areas with less of this clinical and medical infrastructure.

1012
01:09:41.033 --> 01:09:41.193
Yep.

1013
01:09:41.714 --> 01:09:49.920
Um, and so, you know, the challenge becomes when you don't have these anti-venoms for, for scorpion,

1014
01:09:51.201 --> 01:09:58.307
stings, the current clinical solution is this idea of aggressive, supportive stabilization, right?

1015
01:09:58.467 --> 01:10:06.554
You're basically waiting for the body to naturally metabolize, but you have to keep the person alive while that's like happening, right?

1016
01:10:06.574 --> 01:10:08.295
That's the stabilization part.

1017
01:10:08.335 --> 01:10:09.577
You just like keep them alive.

1018
01:10:09.637 --> 01:10:09.817
Right.

1019
01:10:09.937 --> 01:10:11.078
And so that can be done...

1020
01:10:11.458 --> 01:10:16.379
By supporting, for example, artificial air, like if they can't breathe and things like that.

1021
01:10:16.439 --> 01:10:26.122
And they also have this idea of utilizing secondary drugs to counteract the specific damage that the venom is causing, which in some cases can work and it impacts the nervous system directly.

1022
01:10:26.162 --> 01:10:29.822
So this story was immediately all over the place.

1023
01:10:30.183 --> 01:10:30.363
Okay.

1024
01:10:30.383 --> 01:10:31.043
Because, again, marketing.

1025
01:10:31.323 --> 01:10:31.723
Marketing.

1026
01:10:32.964 --> 01:10:42.531
And so the headline read they tested 211 designs in the lab and 17 met their binding criteria for four toxins.

1027
01:10:43.071 --> 01:10:44.512
So like what is – I'm reading this.

1028
01:10:44.532 --> 01:10:45.173
I'm like what is this?

1029
01:10:45.933 --> 01:10:46.834
What does this actually mean?

1030
01:10:48.455 --> 01:10:54.619
Crucially, they've not shown that this can actually neutralize the toxin.

1031
01:10:55.520 --> 01:10:59.684
protect an animal or actually help a human patient.

1032
01:11:00.364 --> 01:11:10.152
And they explicitly made that clear that they did not create an, an anti-venom, but this is sort of this incremental stepping stone process that we've talked about previously.

1033
01:11:11.513 --> 01:11:19.180
Another small detail point is none of the candidates also could bind to multiple toxins.

1034
01:11:19.980 --> 01:11:21.982
So in the journey for a universal anti-venom,

1035
01:11:22.257 --> 01:11:37.309
That's not – it's still not – but it was still an interesting design challenge where they were able to find the ability for the protein to bind, which is still interesting and impressive for the system to have hypothesized and designed on its own.

1036
01:11:37.329 --> 01:11:43.794
So the first thing I want to look at here is like what did this experiment actually establish?

1037
01:11:45.015 --> 01:11:48.418
And so they have this video overview we'll walk through really quick.

1038
01:11:48.918 --> 01:11:57.501
So this was the scorpion from a Mexican scorpion, a scorpion from Mexico, that the CH2 is the toxin.

1039
01:11:57.581 --> 01:12:02.603
And so we've talked about sodium ion channels that allow things to transition.

1040
01:12:02.943 --> 01:12:10.486
And so this toxin comes in and activates so that it causes the ion channels to become overactive.

1041
01:12:10.666 --> 01:12:11.026
Okay.

1042
01:12:11.166 --> 01:12:11.386
Right.

1043
01:12:11.826 --> 01:12:13.347
And so that's obviously a problem.

1044
01:12:14.063 --> 01:12:16.585
So there are current antibodies.

1045
01:12:16.605 --> 01:12:22.930
This LR antibody is actually a lab-engineered antibody that we know can bind to it to prevent this from happening.

1046
01:12:23.350 --> 01:12:37.381
So Claude designed this CN2BI protein that is able to bind to the venom, but it conflicts, what it looks like conflicts with where the antibody is able to connect.

1047
01:12:37.421 --> 01:12:40.764
But not all antibodies, this RU1 antibody, is able to connect.

1048
01:12:41.452 --> 01:12:44.356
the anthropic designed one does not have a confliction with.

1049
01:12:44.376 --> 01:12:53.628
And what's interesting is what they were basically trying to identify is with LR, the initial one we saw, can we bind to the same location for a known antibody that we know is

1050
01:12:55.415 --> 01:12:57.857
is able to combat the venom itself.

1051
01:12:57.877 --> 01:13:00.439
So we're going to repeat this one more time, and I want to come back to this really quick.

1052
01:13:00.820 --> 01:13:04.503
So they use two antibodies, the LR and the RU1.

1053
01:13:05.044 --> 01:13:09.007
They bind to different locations on this venom itself.

1054
01:13:10.108 --> 01:13:18.516
This binding test was trying to figure out, can we design a protein that can bind to the same location as a known working antibody?

1055
01:13:19.597 --> 01:13:23.839
And so it was a competition experiment here.

1056
01:13:24.140 --> 01:13:29.122
So now, again, we see the claw-designed protein is able to connect.

1057
01:13:29.983 --> 01:13:32.084
When the LR antibody is there, it's saying,

1058
01:13:32.555 --> 01:13:33.416
It's in conflict.

1059
01:13:33.476 --> 01:13:34.436
Yeah, it's in conflict.

1060
01:13:34.476 --> 01:13:35.157
It's not able to.

1061
01:13:35.237 --> 01:13:36.898
So it's kind of in the same location.

1062
01:13:36.978 --> 01:13:37.258
Yeah.

1063
01:13:37.438 --> 01:13:40.920
But when R1 is there, it's like it's a little bit, but it's only at 16%.

1064
01:13:40.940 --> 01:13:42.581
Yeah, yeah, yeah, yeah, yeah.

1065
01:13:42.821 --> 01:13:52.307
And so the idea, what they're trying to say here is that, look, we can kind of have a binding in a similar location to something we know that works, but that does not mean it neutralizes.

1066
01:13:53.549 --> 01:13:54.049
The toxin.

1067
01:13:54.069 --> 01:13:55.370
We're just saying that it can bind.

1068
01:13:55.970 --> 01:13:56.171
Right?

1069
01:13:56.231 --> 01:13:57.051
Am I doing okay so far?

1070
01:13:57.071 --> 01:13:57.251
Yeah.

1071
01:13:57.431 --> 01:13:57.972
Yeah.

1072
01:13:58.012 --> 01:13:59.413
I think that makes sense.

1073
01:13:59.793 --> 01:14:03.895
So LR came from an earlier antivenom research from UNAM in Mexico.

1074
01:14:03.935 --> 01:14:08.838
They engineered a human-derived antibody fragment that showed that it protected in mice.

1075
01:14:09.859 --> 01:14:11.400
And it's a laboratory reference.

1076
01:14:12.080 --> 01:14:20.843
A lot of times with these anti-venoms, what they end up doing is they'll inject a horse with the venom, then take the antibodies from the horse, and then do some voodoo magic.

1077
01:14:20.863 --> 01:14:21.443
Oh, yeah, yeah, yeah.

1078
01:14:21.523 --> 01:14:22.443
And then give it to the human.

1079
01:14:22.463 --> 01:14:23.683
Okay, fair enough.

1080
01:14:23.883 --> 01:14:30.905
And then the RU1 is similar that just attaches and binds to a different location.

1081
01:14:30.925 --> 01:14:35.107
So they wanted to see how it worked in both cases.

1082
01:14:35.327 --> 01:14:39.728
The idea here is like neutralization is the idea of capping the ability of the venom to—

1083
01:14:40.288 --> 01:14:41.529
create that overactivation.

1084
01:14:41.929 --> 01:14:42.430
Yeah.

1085
01:14:42.830 --> 01:14:44.772
And ultimately, like, that's a solution here.

1086
01:14:44.832 --> 01:14:49.316
So the headlines and the analysis are always related.

1087
01:14:49.436 --> 01:15:03.548
So what we're saying that they've been able to do is they designed a bunch of options for what could potentially be successful at binding to a similar location as the LR antibody, 211 of them.

1088
01:15:04.329 --> 01:15:06.190
They found, as a result of that,

1089
01:15:06.971 --> 01:15:14.416
that they were able to get 17 to meet the binding criteria across four different toxin types.

1090
01:15:14.777 --> 01:15:16.198
So that's interesting.

1091
01:15:16.218 --> 01:15:19.620
Yeah, it's like generalized across four different...

1092
01:15:20.678 --> 01:15:21.839
Toxins, that's pretty good.

1093
01:15:22.279 --> 01:15:22.939
That's interesting.

1094
01:15:23.139 --> 01:15:23.439
Okay.

1095
01:15:23.900 --> 01:15:31.504
And so – but this is still not helping a patient, right?

1096
01:15:31.584 --> 01:15:36.066
Because we've not even gotten to the point of being able to actually confirm that it neutralizes the toxin.

1097
01:15:36.386 --> 01:15:38.948
So I wanted to take a step back and say what –

1098
01:15:39.768 --> 01:15:46.895
are the steps between this binding mechanism, which was the result of this and a medicine to a patient.

1099
01:15:47.036 --> 01:15:47.756
Yeah.

1100
01:15:48.037 --> 01:15:52.801
And, uh, so Trini actually put this roadmap out that we can look at.

1101
01:15:52.821 --> 01:15:54.984
That's kind of how they sort of define it.

1102
01:15:55.004 --> 01:15:55.544
So step one,

1103
01:15:56.525 --> 01:15:56.705
Right.

1104
01:15:56.765 --> 01:15:58.106
Is does it neutralize the toxin?

1105
01:15:58.146 --> 01:15:59.446
We haven't even gotten to step one yet.

1106
01:16:00.067 --> 01:16:01.747
We found a binding mechanism.

1107
01:16:02.228 --> 01:16:05.189
And so now step one is making sure that it can actually neutralize the toxin.

1108
01:16:05.309 --> 01:16:08.511
Because it could bind, but not necessarily neutralize it.

1109
01:16:09.291 --> 01:16:10.111
Oh, interesting.

1110
01:16:10.131 --> 01:16:10.672
You know what I'm saying?

1111
01:16:10.692 --> 01:16:10.832
I see.

1112
01:16:10.952 --> 01:16:11.212
Yeah.

1113
01:16:11.392 --> 01:16:13.293
And so there's a difference between those two things.

1114
01:16:13.333 --> 01:16:18.975
We know it can connect, but we don't know that it protects the actual nerve function itself.

1115
01:16:19.436 --> 01:16:24.178
So then number two, does it cover the relevant venoms that we're trying to address?

1116
01:16:24.198 --> 01:16:24.778
Mm-hmm.

1117
01:16:25.138 --> 01:16:29.422
Number three, does it work in a body, in this case in animal testing?

1118
01:16:30.243 --> 01:16:36.688
Then we get to stage number four, which is, is it ready from a safety perspective for human testing?

1119
01:16:37.529 --> 01:16:42.193
And then five, does it actually help patients once it gets through that developmental process?

1120
01:16:42.533 --> 01:16:47.598
They estimate that whole pipeline for this specific use case to be about four to seven years plus.

1121
01:16:47.618 --> 01:16:47.718
Yeah.

1122
01:16:49.190 --> 01:16:54.675
And all we're talking about is not even yet having solved step one.

1123
01:16:55.575 --> 01:17:00.900
But this is the pipeline that all these AI is going to cure X, Y, and Z.

1124
01:17:01.361 --> 01:17:03.142
This is the pipeline they kind of have to go through.

1125
01:17:03.162 --> 01:17:08.467
And they've identified a few things in their chart about where they think AI is going to be the most impactful.

1126
01:17:08.908 --> 01:17:14.273
So improving the neutralization of the venom itself, they're like, I think we think we'll have moderate impact.

1127
01:17:14.953 --> 01:17:26.716
in being able to then cover, once they know it can neutralize, they think it has the highest impact of being able to then generalize it across all these different classes of venom, this universal anti-venom.

1128
01:17:27.036 --> 01:17:32.837
They also think that in that first stage of animal testing that it can have high impact as it relates to animal testing.

1129
01:17:32.877 --> 01:17:41.459
But when you get to preparing for human trials and the trials and acceptance process, it goes to moderate to limited, which is obviously very understandable.

1130
01:17:41.499 --> 01:17:44.300
I think the point here that is sort of being brought up is

1131
01:17:45.600 --> 01:17:58.527
So there's a real world aspect to this stuff that is just naturally going to be slower unless and until some of the processes, the regulatory environment and the way in which we do those downstream processes fundamentally changes.

1132
01:17:58.947 --> 01:18:04.490
So that's kind of part one of what this OpenAI announcement is.

1133
01:18:04.910 --> 01:18:13.452
And it's interesting because it dovetails with the wet lab story we talked about a few weeks ago where everyone's like, oh, Anthropic has a wet lab.

1134
01:18:13.632 --> 01:18:13.872
Yeah.

1135
01:18:14.372 --> 01:18:23.934
You know, but it was where the AI was designing and the human scientists were in the lab actually conducting the entire process.

1136
01:18:24.134 --> 01:18:24.974
Yeah.

1137
01:18:25.174 --> 01:18:25.614
Itself.

1138
01:18:25.694 --> 01:18:30.435
So I want to sort of briefly talk about the timing.

1139
01:18:31.571 --> 01:18:34.452
Because the promise here is that this is going to accelerate.

1140
01:18:34.472 --> 01:18:35.453
It's going to happen very quickly.

1141
01:18:36.333 --> 01:18:42.836
So this next chart kind of looks at how long did it take them to create the design candidates?

1142
01:18:43.337 --> 01:18:49.280
And how long did it take them to test those candidates in a laboratory with the two partners we mentioned earlier?

1143
01:18:49.700 --> 01:18:53.922
They were able to get the first 180 designs in 64 hours, which is quite quick.

1144
01:18:54.182 --> 01:18:54.942
That's quite quick.

1145
01:18:55.723 --> 01:18:56.183
Quite fast.

1146
01:18:56.203 --> 01:18:56.303
Yeah.

1147
01:18:57.454 --> 01:18:58.795
The easiest part of the process, maybe.

1148
01:18:59.255 --> 01:19:00.176
But it's still quite fast.

1149
01:19:00.196 --> 01:19:04.079
Yeah, two and a half days, though, is quite fast for 180 designs.

1150
01:19:04.740 --> 01:19:05.340
It's quite quick.

1151
01:19:05.420 --> 01:19:05.680
Yeah.

1152
01:19:06.481 --> 01:19:15.808
The first binding results arrived after 39 days, which is from the beginning of when they started the design process, not after they got the design process.

1153
01:19:16.088 --> 01:19:24.995
So about five and a half weeks to get to this point where we have the designs done, we've tested them in the lab, and we have these...

1154
01:19:26.836 --> 01:19:28.138
There's candidates in between there.

1155
01:19:28.178 --> 01:19:29.019
There's synthesis.

1156
01:19:29.039 --> 01:19:30.100
There's protein production.

1157
01:19:30.341 --> 01:19:33.625
There's getting the reagent, scheduling the work, actually measuring what happens.

1158
01:19:33.785 --> 01:19:34.385
There's quite a bit.

1159
01:19:34.966 --> 01:19:38.771
Yeah, there's quite a bit that hasn't been optimized for...

1160
01:19:40.258 --> 01:19:42.159
An AI lab ecosystem.

1161
01:19:42.239 --> 01:19:42.900
Let's say that.

1162
01:19:43.040 --> 01:19:44.121
That's exactly right.

1163
01:19:44.221 --> 01:19:51.446
Like this is – there's so many parts of this that are still sort of not even optimized, I would say.

1164
01:19:51.786 --> 01:19:57.630
Just like catering to human timescales, like PhD student timescales.

1165
01:19:58.391 --> 01:19:58.611
Right.

1166
01:19:59.151 --> 01:19:59.351
Right.

1167
01:19:59.992 --> 01:20:07.977
And I think part of what we're trying to talk about in this story is looking at each of the stages of this process and where is –

1168
01:20:09.058 --> 01:20:17.249
AI, but also the combination of automation and robotics, because ultimately you need to interact with the real world.

1169
01:20:17.289 --> 01:20:18.290
That's right.

1170
01:20:18.450 --> 01:20:23.757
And so this is where sort of the AI and robotics together fit in that process.

1171
01:20:24.198 --> 01:20:25.780
So I wanted to, again, I was trying to.

1172
01:20:27.021 --> 01:20:55.947
help myself understand what are we looking at here and so i created this map of how ai is going to like meet the physical lab and where the current state of play is and some of the sort of this is not a uh complete map of every player in the space but it's just to give a kind of representative example so it kind of breaks down into these three stages there's the design and prioritize so we just talked about that in this anthropic story with the scorpion anti-venom so i'm

1173
01:20:56.555 --> 01:20:57.185
The AI...

1174
01:20:58.008 --> 01:20:59.830
Is going to design and prioritize.

1175
01:20:59.870 --> 01:21:00.210
Correct.

1176
01:21:00.870 --> 01:21:01.551
In this model.

1177
01:21:01.891 --> 01:21:03.953
We may still define the question.

1178
01:21:04.413 --> 01:21:04.734
Right.

1179
01:21:04.914 --> 01:21:06.015
Scorpion, anti-venom.

1180
01:21:06.515 --> 01:21:06.735
Right.

1181
01:21:06.755 --> 01:21:10.918
Be like, give me an anti-venom to all of Scorpions.

1182
01:21:11.119 --> 01:21:11.299
Right.

1183
01:21:11.659 --> 01:21:14.942
And I think the original prompt for this was about 35 pages.

1184
01:21:15.562 --> 01:21:16.183
Oh, wow.

1185
01:21:16.323 --> 01:21:17.324
That they fed.

1186
01:21:17.344 --> 01:21:18.705
They had, I think, 300 GPUs.

1187
01:21:19.005 --> 01:21:25.951
It cost a total of about $40,000 to $60,000 to run this design and prioritize process.

1188
01:21:26.171 --> 01:21:26.371
Mm-hmm.

1189
01:21:26.551 --> 01:21:28.372
Which is not crazy.

1190
01:21:28.532 --> 01:21:28.772
No.

1191
01:21:29.053 --> 01:21:32.335
But it's a little expensive, but it's, you know, not crazy.

1192
01:21:32.375 --> 01:21:36.417
No, it's not the usual stuff that I ask Claude and ChatGPT to do.

1193
01:21:36.778 --> 01:21:37.538
Right.

1194
01:21:37.958 --> 01:21:39.179
My prompts are not 35 pages.

1195
01:21:39.219 --> 01:21:39.619
Right.

1196
01:21:40.760 --> 01:21:44.843
And if one of those ultimately became something that was able to get to the end of the process, that would be fantastic.

1197
01:21:45.503 --> 01:21:52.972
Okay, so a variety of the frontier labs are doing that design and prioritize, but we're talking about this anthropic story in combination with Citrini.

1198
01:21:53.272 --> 01:21:57.157
So then sort of stage two is this idea of physical automation.

1199
01:21:58.212 --> 01:22:00.554
Not necessarily AI, anything.

1200
01:22:00.614 --> 01:22:06.037
We have a couple of players in here, MIT, Emerald, Cloud Lab, Stratios, and Ginkgo.

1201
01:22:06.638 --> 01:22:13.963
They're trying to now take the physical process that humans do in a lab and have it be run by a robotic automation.

1202
01:22:14.323 --> 01:22:20.087
Obviously, it's a connective tissue because you can take part one and then send it to part two.

1203
01:22:20.747 --> 01:22:24.610
Now, there are some companies that are trying to do...

1204
01:22:25.450 --> 01:22:30.274
The full closed loop cycle, going back to our previous story, of the learn and repeat.

1205
01:22:30.715 --> 01:22:47.970
This is Lila Science Madeira, where it is the combination of both the AI designing, prioritizing, the robotic lab executing, getting the measurements, feeding that back directly into the AI to then decide what to change and then retry.

1206
01:22:48.270 --> 01:22:49.611
Human oversight, but –

1207
01:22:51.092 --> 01:22:52.373
Kind of a closed-loop cycle.

1208
01:22:52.933 --> 01:22:56.754
These processes are also using a variety of shared infrastructure.

1209
01:22:56.814 --> 01:23:04.737
For example, Carnegie Mellon is providing some of this shared infrastructure as well as Slack, the national lab.

1210
01:23:05.198 --> 01:23:07.178
Stanford Linear Accelerator Lab.

1211
01:23:07.358 --> 01:23:11.300
Yep, is giving some of the infrastructure around these pieces.

1212
01:23:11.340 --> 01:23:18.683
So that's where we're trying to go is where Lila Sciences is effectively saying we are right now.

1213
01:23:19.183 --> 01:23:34.619
And so the first step that the frontier labs think they can take is design and prioritize because it does not require this connection to the downstream robotics and automation piece, which is not their bread and butter.

1214
01:23:35.039 --> 01:23:37.620
So now let's transition.

1215
01:23:38.440 --> 01:23:43.841
And this is something we've talked about a lot with the methods is this equipment and measurement.

1216
01:23:44.242 --> 01:23:46.522
We talk a lot about tools on the pod.

1217
01:23:47.102 --> 01:23:50.663
We talk a lot about being precise in our measurement and our methods.

1218
01:23:51.303 --> 01:23:55.805
And MIT actually gives us a great example of this automating step.

1219
01:23:55.825 --> 01:23:59.506
This is not necessarily new as in new in the last couple of years.

1220
01:23:59.546 --> 01:24:00.746
It's been happening for some time.

1221
01:24:01.506 --> 01:24:08.329
But understanding how this now connects to these AI workflows is a different way to start to think about these steps.

1222
01:24:08.349 --> 01:24:15.813
So researchers at MIT working in collaboration with Nokia Bell Labs and Arizona State built a robotic system that can assemble and align models.

1223
01:24:16.873 --> 01:24:18.394
an optical experiment.

1224
01:24:19.594 --> 01:24:22.655
And this was actually highlighted recently in September.

1225
01:24:23.396 --> 01:24:32.219
This research, always talk about the funders, supported by National Science Foundation, Army Research Office, and some other funders, private funders, including Shell.

1226
01:24:32.599 --> 01:24:43.584
And this was one of their demonstration videos where they have a robotic arm that in 50 steps in 30 minutes can set up

1227
01:24:44.584 --> 01:24:48.826
these lenses for this optical experiment, which we've talked about previously.

1228
01:24:49.986 --> 01:25:02.172
And not only can it set up the lenses, because it's constantly taking measurements, it can also make a change to fix something when a misalignment arises as well.

1229
01:25:02.452 --> 01:25:03.873
So it's both things.

1230
01:25:05.293 --> 01:25:06.094
This is crazy.

1231
01:25:06.154 --> 01:25:07.214
Human-designed setup.

1232
01:25:07.594 --> 01:25:07.874
Yeah.

1233
01:25:08.995 --> 01:25:12.677
But it can set up and iterate when there's a misalignment.

1234
01:25:13.017 --> 01:25:13.737
I mean, I mean...

1235
01:25:14.187 --> 01:25:18.209
So already this is giving me flashbacks to my PhD days.

1236
01:25:18.729 --> 01:25:27.652
Like, for example, the table that they're on, you see all those holes, those little screws that you can put in and you can fasten like mirrors and lenses.

1237
01:25:28.393 --> 01:25:28.593
Um,

1238
01:25:29.553 --> 01:25:44.930
Any type of tabletop AMO experiment, meaning atomic molecular optical, like if you want to do molecular traps, a lot of quantum computing experiments are literally just this, which is establishing lenses everywhere and then...

1239
01:25:46.452 --> 01:26:00.446
You know, aligning them to the point where you can suspend atoms in a laser or you can, you know, isolate a single atom to do some type of spectroscopy measurement.

1240
01:26:00.546 --> 01:26:03.589
This reminds me of the MIT double slit experiment episode.

1241
01:26:03.649 --> 01:26:05.251
Yeah, this is all that is.

1242
01:26:06.012 --> 01:26:09.355
Like a lot of these AMO experiments are literally just this.

1243
01:26:10.273 --> 01:26:17.419
And just so, you know, I want to ask you briefly, like, the sensitivity and complexity.

1244
01:26:17.439 --> 01:26:21.883
Because someone who's watching this might be like, oh, this looks pretty simple.

1245
01:26:21.903 --> 01:26:23.003
No, this is incredible.

1246
01:26:23.384 --> 01:26:33.212
Like, if they're able to align a laser and, like, make sure that the transition frequency is exactly aligned to whatever, you know, system that they're trying to control.

1247
01:26:33.492 --> 01:26:35.213
Like, that's pretty incredible.

1248
01:26:40.477 --> 01:26:42.540
Like, that's... 50 minutes, 30 minutes?

1249
01:26:42.560 --> 01:26:45.904
That's, like, half the work of these AMO PhD students.

1250
01:26:46.244 --> 01:26:49.608
Like, my roommates, you know, like Pratik Puri.

1251
01:26:49.628 --> 01:26:50.189
Yeah, yeah, yeah.

1252
01:26:50.229 --> 01:26:53.453
I don't mean to call you out, but this is half your job.

1253
01:26:53.493 --> 01:26:53.993
Like, you know?

1254
01:26:54.334 --> 01:26:55.055
Like, so...

1255
01:26:55.775 --> 01:27:12.489
um elliot borah another friend of the pod yes like this is something that he does a lot of i mean obviously there's a lot more that's in his head sure right there's a lot more stuff that they could do perhaps if they didn't have to spend their time doing that aligning lenses and lasers

1256
01:27:14.185 --> 01:27:24.350
And this was a demonstration going back to our three stages of mainly this idea of execution and measurement.

1257
01:27:24.830 --> 01:27:28.612
Because AI can design all these things, but we still have to be able to execute and measure.

1258
01:27:28.992 --> 01:27:31.513
And this is for optical experiments, not biological experiments.

1259
01:27:32.834 --> 01:27:57.460
experiments it's a different category but i'm just trying to give the listeners a sense of where these component parts are that people are ultimately trying to connect in order to get this closed loop end-to-end uh ai-based research cycle um you know but that step might can close the equipment control loop yeah as one component part of that

1260
01:27:57.940 --> 01:28:07.403
Now, the next video that – this is sort of a promo video so that there's some history about how we've done these things for quite some time.

1261
01:28:07.943 --> 01:28:10.604
So this is from Emerald Cloud Lab and Stratios.

1262
01:28:10.804 --> 01:28:16.846
And they kind of address a slightly different part of this execution, which is making physical laboratories –

1263
01:28:18.999 --> 01:28:22.722
accessible through software, through the cloud, remotely.

1264
01:28:23.143 --> 01:28:24.624
So I'm in one location.

1265
01:28:25.385 --> 01:28:28.808
I need to do X, Y, or Z in another location.

1266
01:28:29.669 --> 01:28:37.396
And the entirety of the process that I need to be done is remotely executed by automation and robotics.

1267
01:28:38.016 --> 01:28:41.400
But I have a software interface that I can interact with in real time.

1268
01:28:42.230 --> 01:28:43.011
to execute.

1269
01:28:43.291 --> 01:28:50.655
So like a lot of folks might go to the doctor and then they need to get, you know, a blood test and then they send it to Quest Diagnostics, right?

1270
01:28:50.875 --> 01:28:58.980
So this idea that you have one person doing the question, one person doing the execution, and then it comes back to your physician to give you the result.

1271
01:28:59.441 --> 01:29:08.706
The difference here, again, with still some elements of human involvement for some of the component parts, many of the components are now directly

1272
01:29:13.412 --> 01:29:15.693
This is, I think, a big area.

1273
01:29:15.713 --> 01:29:16.394
This is crazy.

1274
01:29:16.434 --> 01:29:17.434
I think this is very big.

1275
01:29:17.454 --> 01:29:18.515
No, this is crazy.

1276
01:29:19.575 --> 01:29:23.217
This is absolutely... And you get how long it takes to run each.

1277
01:29:23.257 --> 01:29:27.039
You get the real-time data back in the exact format you want.

1278
01:29:27.359 --> 01:29:30.201
It's a wet lab that you can access through the cloud.

1279
01:29:30.221 --> 01:29:33.302
Isn't that crazy?

1280
01:29:33.963 --> 01:29:36.644
I mean, no, they're doing mass spec, I mean, or something, you know?

1281
01:29:36.684 --> 01:29:42.227
This seems like stuff that would require specialized equipment, but there you go, like, right?

1282
01:29:44.431 --> 01:29:46.553
No, I mean, that's actually incredible, right?

1283
01:29:46.653 --> 01:29:51.518
Because for the longest time, we've seen machine learning researchers do this.

1284
01:29:51.798 --> 01:29:59.346
You just access GPUs from the cloud, AWS, and then you're doing machine learning experiments.

1285
01:29:59.406 --> 01:30:02.970
Well, now you're actually doing real biological experiments.

1286
01:30:02.990 --> 01:30:04.711
You've got pipettes that are automated.

1287
01:30:04.731 --> 01:30:05.232
You've got...

1288
01:30:06.813 --> 01:30:09.554
I think I saw a centrifuge that was automated, right?

1289
01:30:09.914 --> 01:30:13.815
And they did it correctly, unlike the Ryan Gosling movie.

1290
01:30:16.396 --> 01:30:21.398
Look, it was one of the only flaws that I brought up from Hail Mary.

1291
01:30:21.458 --> 01:30:23.438
If you were going to say it or miss it.

1292
01:30:23.598 --> 01:30:24.658
But, like, they did it right.

1293
01:30:24.678 --> 01:30:25.399
They did it right.

1294
01:30:25.459 --> 01:30:29.100
Like, the AI knows to balance the centrifuge.

1295
01:30:29.986 --> 01:30:31.127
Unlike Ryan Gosling.

1296
01:30:32.207 --> 01:30:33.528
But no, this is incredible.

1297
01:30:35.009 --> 01:30:38.331
And what, again, this is not a promo for any particular entity.

1298
01:30:38.351 --> 01:30:40.933
There are a variety of companies that are trying to do so.

1299
01:30:40.953 --> 01:30:45.175
No, I'm just saying this is a new era in scientific.

1300
01:30:45.535 --> 01:30:45.875
Correct.

1301
01:30:46.316 --> 01:30:54.260
Yeah, so, I mean, one thing I want to understand is like, is Emerald Cloud Lab going to be a co-author?

1302
01:30:54.280 --> 01:30:56.602
I wonder if...

1303
01:30:57.446 --> 01:31:07.652
Like, as a former researcher in, like, a PhD or a postdoc setting, it's like, I'm always like, who's going to be co-author on the paper if I were to use this data, right?

1304
01:31:07.712 --> 01:31:13.375
Do they have some kind of thing in their contract being like, we got to be... Acknowledgements are fine.

1305
01:31:13.435 --> 01:31:16.437
Co-author, it's a bit, you know, like, okay.

1306
01:31:16.457 --> 01:31:17.257
Maybe a footnote.

1307
01:31:17.537 --> 01:31:18.378
Yeah.

1308
01:31:18.458 --> 01:31:19.258
Maybe a footnote.

1309
01:31:19.378 --> 01:31:19.539
And...

1310
01:31:20.619 --> 01:31:34.587
One of the things that this brings up that I was kind of thinking about that's not necessarily as obvious because like the idea is like, oh, you know, a lot of the thoughts around these AI implementations is it's going to allow the rich to get richer.

1311
01:31:34.947 --> 01:31:35.728
Which is true.

1312
01:31:35.888 --> 01:31:36.448
Yeah.

1313
01:31:36.668 --> 01:31:37.549
Probably.

1314
01:31:38.770 --> 01:31:42.052
In a lot of contexts, that's true.

1315
01:31:42.352 --> 01:31:44.113
But what it does also do, like what –

1316
01:31:45.168 --> 01:31:54.306
type of research teams that might not have access to these types of resources who are in underserved environments, what kind of questions can they ask?

1317
01:31:55.488 --> 01:32:01.514
And it's the idea of AWS and all this cloud computing allowed people to build websites and do video.

1318
01:32:01.574 --> 01:32:06.859
And we can do this show because of cloud computing and distributed networking.

1319
01:32:06.879 --> 01:32:08.040
This wouldn't exist without that.

1320
01:32:08.420 --> 01:32:16.628
So it does bring up the question like what kind of research questions and what access that teams that otherwise would not have been able to access these kind of questions.

1321
01:32:16.848 --> 01:32:32.583
high fidelity resources, particularly in my family who's worked in, you know, global health and neglected diseases, it becomes a very interesting democratization of the access to the types of tooling when you have this cloud accessibility.

1322
01:32:32.623 --> 01:32:41.972
And that was a non-obvious conclusion after kind of digging into this is like, it is a different paradigm that allows us to ask some different questions.

1323
01:32:41.992 --> 01:32:42.352
Yeah.

1324
01:32:42.552 --> 01:32:47.057
One last sort of note I want to bring up going to the third stage.

1325
01:32:47.498 --> 01:32:50.661
Because we talked about design and how to prioritize what questions to ask.

1326
01:32:51.462 --> 01:32:55.106
We then looked at automation at two different levels, right?

1327
01:32:56.088 --> 01:32:59.812
One where the humans are still involved in the lab, one that's cloud accessible, and

1328
01:33:00.192 --> 01:33:05.779
So Lila Sciences is trying to connect the full loop, right?

1329
01:33:06.019 --> 01:33:07.381
They had this catalyst study.

1330
01:33:08.222 --> 01:33:14.729
The AI helped select compositions, automated the equipment, made the test materials, and the measurements informed subsequent choices.

1331
01:33:15.010 --> 01:33:16.532
So it was a very narrow question.

1332
01:33:16.552 --> 01:33:17.553
Yeah, but it's closed loop.

1333
01:33:17.573 --> 01:33:18.934
But it's closed loop in some sense.

1334
01:33:18.955 --> 01:33:19.115
Okay.

1335
01:33:20.036 --> 01:33:23.979
They evaluated almost 3,000 compositions related to hydrogen production.

1336
01:33:25.480 --> 01:33:36.389
And this diagram here kind of shows how they look at that cycle and what they're ultimately trying to do in terms of this closed loop catalyst discovery.

1337
01:33:36.689 --> 01:33:37.069
And so –

1338
01:33:39.812 --> 01:33:43.798
The goal here was trying to give a baseline lay of the land.

1339
01:33:45.621 --> 01:33:48.505
There are more than just the frontier companies doing this.

1340
01:33:49.326 --> 01:33:52.511
People are attacking different component parts of the process.

1341
01:33:53.832 --> 01:33:55.093
Some people are trying to do the whole thing.

1342
01:33:55.333 --> 01:33:57.115
Some people are just trying to do a module of the thing.

1343
01:33:58.656 --> 01:34:02.599
And there is sort of this shared infrastructure.

1344
01:34:02.919 --> 01:34:13.027
So Carnegie Mellon has AI Science Foundry, which is trying to connect these cloud labs and scientific infrastructure to like a shared – this sort of shared distributed research infrastructure to make sure it's –

1345
01:34:14.148 --> 01:34:16.391
evenly distributed and accessible for many people.

1346
01:34:16.752 --> 01:34:23.762
Slack has this new Spire project, which is trying to connect their robotic sample handling, beam control, and real-time analysis.

1347
01:34:24.082 --> 01:34:27.006
And we actually have a reference video here.

1348
01:34:27.066 --> 01:34:27.707
This is from...

1349
01:34:28.228 --> 01:34:30.210
from Slack's Spire project.

1350
01:34:30.230 --> 01:34:33.633
Now, it's just part of the execution and measurement infrastructure.

1351
01:34:34.034 --> 01:34:35.816
It's not the completed future system.

1352
01:34:36.176 --> 01:34:46.346
But again, I think that there are multiple national labs, universities, private companies that are all moving in this direction.

1353
01:34:47.187 --> 01:34:56.669
And when we go back to one of the reasons we started the show, the thesis was we're going to be in a Cambrian explosion of a whole variety of breakthroughs.

1354
01:34:57.369 --> 01:35:02.750
And we want to be here having a front seat and bringing you along with us for what will happen.

1355
01:35:03.270 --> 01:35:10.692
Again, there are very complex social, economic, political, and other issues as it relates to, quote, AI issues.

1356
01:35:12.449 --> 01:35:22.519
But on this show, we talk about the researchers who are using this in very specific ways that are ultimately looking to create opportunities to benefit us collectively.

1357
01:35:23.199 --> 01:35:33.870
And it's important not to just say AI is bad in its entirety when there are really interesting and valuable use cases available.

1358
01:35:34.250 --> 01:35:35.551
that are coming up.

1359
01:35:36.251 --> 01:35:43.376
Now, I will just end on this note of all the Doom scenarios expect there to be automated wet labs.

1360
01:35:43.776 --> 01:35:44.056
Yeah.

1361
01:35:44.657 --> 01:35:46.378
And then an accidental release.

1362
01:35:46.758 --> 01:35:46.998
Yes.

1363
01:35:47.258 --> 01:35:54.743
Obviously, none of the things we discussed are related to pathogens that are extremely deadly.

1364
01:35:55.163 --> 01:36:01.447
But there are certainly a lot of consequences as it relates to the progress we've already made with...

1365
01:36:02.158 --> 01:36:28.191
cloud wet labs yeah and dude i mean it's it's uh it's the best of times it is the worst of times as charles dickens said right um and it really feels that way there's if we if we do nothing it's going to be both right and um the scales only need to tip one way or the other it's kind of like the chemistry noble prize right

1366
01:36:28.544 --> 01:36:35.408
You only need a slight asymmetry, and then all of a sudden, it's very much not a great thing, right?

1367
01:36:35.528 --> 01:36:37.449
Or all of a sudden, it's a very good thing, right?

1368
01:36:37.509 --> 01:36:42.592
So we want our homochirality to be positive.

1369
01:36:42.652 --> 01:36:43.872
Positive in our direction.

1370
01:36:43.892 --> 01:36:45.013
In our direction, right?

1371
01:36:45.793 --> 01:36:46.594
Not in the A.I.S.

1372
01:36:46.614 --> 01:36:48.515
direction.

1373
01:36:49.015 --> 01:36:57.500
So I thought you'd find this, because we kind of touch on aspects of this, but as we continue in future episodes, this is going to continue to come up.

1374
01:36:58.260 --> 01:37:26.919
um and uh i successfully talked through a live molecular diagram yeah that was pretty good it's not bad yeah i'll get better i'm learning guys i'm trying to get better um but i i thought that was that was pretty cool i mean um the the laser stuff is it was crazy to me because i've been in those labs right um in the physics department at ucla like

1375
01:37:28.949 --> 01:37:33.512
those tables are the bread and butter of every tabletop experiment.

1376
01:37:34.913 --> 01:37:36.114
So that was crazy.

1377
01:37:36.255 --> 01:37:42.600
There was just a robot that could just hang out and do that, which was like half the job of the PhD.

1378
01:37:42.700 --> 01:37:43.901
30 minutes, 50 moves.

1379
01:37:44.101 --> 01:37:44.341
Right.

1380
01:37:44.441 --> 01:37:45.102
That's crazy.

1381
01:37:45.282 --> 01:37:45.522
Yeah.

1382
01:37:46.103 --> 01:37:46.703
It's not bad.

1383
01:37:47.023 --> 01:37:47.224
Yeah.

1384
01:37:47.404 --> 01:37:48.184
And no breaks.

1385
01:37:48.385 --> 01:37:48.705
No.

1386
01:37:49.125 --> 01:37:51.087
It could just work overnight kind of thing.

1387
01:37:51.207 --> 01:37:51.507
You know?

1388
01:37:51.928 --> 01:37:52.128
Yeah.

1389
01:37:52.268 --> 01:37:53.609
It could adjust when there's an issue.

1390
01:37:53.629 --> 01:37:53.729
Yeah.

1391
01:37:53.843 --> 01:37:54.063
Yeah.

1392
01:37:54.864 --> 01:37:55.064
Yeah.

1393
01:37:55.664 --> 01:37:56.385
It's not bad.

1394
01:37:56.525 --> 01:37:57.245
That's pretty good.

1395
01:37:57.385 --> 01:37:58.246
That's not bad.

1396
01:37:58.426 --> 01:38:08.712
We are going to continue to bring you the frontier of science research globally, not just here in the U.S., but from all around the world.

1397
01:38:09.132 --> 01:38:10.273
My name is Lester Nare.

1398
01:38:10.293 --> 01:38:17.457
Joined as always by my co-host and our resident Ph.D. And I also just want to remind everybody.

1399
01:38:17.885 --> 01:38:20.366
Signal award-winning podcast.

1400
01:38:20.827 --> 01:38:21.327
Yeah.

1401
01:38:22.087 --> 01:38:22.748
Bronze.

1402
01:38:22.868 --> 01:38:24.088
We're going for gold.

1403
01:38:24.248 --> 01:38:29.791
We're going for gold, but it'll go like my point.

1404
01:38:30.111 --> 01:38:30.491
There it is.

1405
01:38:31.472 --> 01:38:32.412
Krishna got the point.

1406
01:38:32.933 --> 01:38:37.335
And for those of you watching on Apple Podcasts on video for the first time, welcome.

1407
01:38:37.495 --> 01:38:38.836
We know it's been a long time coming.

1408
01:38:39.776 --> 01:38:43.258
We will see you all next week.