1
00:00:04,600 --> 00:00:17,560
Speaker 1: Sleepwalkers is a production of our Heart Radio and Unusual Productions. Hi,

2
00:00:17,720 --> 00:00:20,880
Speaker 1: I'm Aloan and I'm care Price. Welcome to a special

3
00:00:20,960 --> 00:00:36,760
Speaker 1: bonus episode of Sleepwalkers. Well, Cara, it's very good to

4
00:00:36,760 --> 00:00:39,000
Speaker 1: be back to the office. That you It is good

5
00:00:39,000 --> 00:00:41,120
Speaker 1: to be back, as I just want to apologize to

6
00:00:41,120 --> 00:00:43,720
Speaker 1: our listeners. We don't have an algorithm that's going to

7
00:00:43,760 --> 00:00:48,320
Speaker 1: create season two, so things have taken a little longer. Yes, unfortunately,

8
00:00:48,520 --> 00:00:51,680
Speaker 1: unfortunately a I can't do everything yet. But we are

9
00:00:51,680 --> 00:00:54,000
Speaker 1: hard at work on season two of Sleepwalkers, and we're

10
00:00:54,040 --> 00:00:57,600
Speaker 1: focusing on stories that really contextualize the implications of AI,

11
00:00:57,960 --> 00:01:00,520
Speaker 1: what it's doing, what the future is, and how it's

12
00:01:00,520 --> 00:01:03,400
Speaker 1: affecting us. You know, we had such a good time

13
00:01:03,560 --> 00:01:06,560
Speaker 1: in season one of Sleepwalkers wrapping our heads around the

14
00:01:06,560 --> 00:01:10,200
Speaker 1: meaning of artificial intelligence. It becomes basically a principle of

15
00:01:10,240 --> 00:01:14,360
Speaker 1: statistics prediction, how we're using data to inform our decisions,

16
00:01:14,560 --> 00:01:18,200
Speaker 1: and how that's becoming ingrained in products and services and

17
00:01:18,400 --> 00:01:22,080
Speaker 1: everything we do. Really, it's true the Pandora's box of

18
00:01:22,120 --> 00:01:24,520
Speaker 1: AI has been opened, but we still have the black

19
00:01:24,560 --> 00:01:27,880
Speaker 1: box problem that is true explainable AI. We can't tell

20
00:01:27,880 --> 00:01:30,360
Speaker 1: what neural networks are doing yet, but people are working

21
00:01:30,400 --> 00:01:32,000
Speaker 1: on it, and that's a story we're going to be

22
00:01:32,000 --> 00:01:35,440
Speaker 1: covering closely in season two. But in this bonus episode

23
00:01:35,480 --> 00:01:37,440
Speaker 1: of Sleepwalkers, we're going to take a look back at

24
00:01:37,440 --> 00:01:40,520
Speaker 1: some of the most poignant stories and interesting applications of

25
00:01:40,520 --> 00:01:43,399
Speaker 1: AI that we talked about in the first season, and

26
00:01:43,520 --> 00:01:45,360
Speaker 1: later in this episode, I'm going to give you a

27
00:01:45,400 --> 00:01:48,280
Speaker 1: preview of a fascinating new company that's using AI to

28
00:01:48,320 --> 00:01:54,200
Speaker 1: predict very specific consumer taste, as in preferences, not like

29
00:01:54,320 --> 00:01:57,440
Speaker 1: tasting clothing. Before we get through though us, when you

30
00:01:57,520 --> 00:01:59,640
Speaker 1: look back at season one, what stands out to you

31
00:01:59,640 --> 00:02:01,560
Speaker 1: the most? One of the things that stands out to

32
00:02:01,600 --> 00:02:04,320
Speaker 1: me most is the story you did about liar Bird.

33
00:02:04,400 --> 00:02:07,880
Speaker 1: Thank you, but seriously, the way they use an algorithm

34
00:02:07,960 --> 00:02:09,920
Speaker 1: to create a deep fake of your voice. But in

35
00:02:09,960 --> 00:02:13,560
Speaker 1: that particular piece, the questions it raised about mortality and

36
00:02:13,600 --> 00:02:16,079
Speaker 1: would you want to hear your father's voice beyond the

37
00:02:16,160 --> 00:02:19,000
Speaker 1: grave has stuck with me and I was one of

38
00:02:19,000 --> 00:02:21,639
Speaker 1: the most powerful moments in the whole podcast. Yeah so,

39
00:02:22,000 --> 00:02:24,959
Speaker 1: Jose Satello, one of the founders of Liar Bird, had

40
00:02:25,000 --> 00:02:27,080
Speaker 1: me sit down on a microphone for an hour and

41
00:02:27,120 --> 00:02:30,800
Speaker 1: just speak which was a personal dream of mine, and

42
00:02:30,840 --> 00:02:36,160
Speaker 1: then using that out me interrupting. Using that voice data,

43
00:02:36,440 --> 00:02:39,400
Speaker 1: liar Bird's algorithms created a version of my voice that

44
00:02:39,400 --> 00:02:42,200
Speaker 1: could turn any written text into something that sounded like me.

45
00:02:42,720 --> 00:02:46,280
Speaker 1: But Jose actually explained it better. So can you hit

46
00:02:46,320 --> 00:02:49,280
Speaker 1: the clip? Not an AI scientist, but we do have

47
00:02:49,360 --> 00:02:53,600
Speaker 1: the sophistication to roll tape. I know it might sound

48
00:02:53,639 --> 00:02:56,919
Speaker 1: a bit like magic, but in reality, the way that

49
00:02:57,160 --> 00:02:59,919
Speaker 1: our algorithm's work is basically they are just a battern

50
00:03:00,080 --> 00:03:04,680
Speaker 1: matching algorithms, and so it's trying to figure out how

51
00:03:05,240 --> 00:03:09,160
Speaker 1: to identify the patterns in your voice by comparing it

52
00:03:09,320 --> 00:03:12,800
Speaker 1: against thousands of other voices a shoually tens of thousands

53
00:03:12,840 --> 00:03:16,000
Speaker 1: of other voices, and trying to figure out what is

54
00:03:16,040 --> 00:03:19,480
Speaker 1: it that makes your voice unique. Once Jose's algorithms identified

55
00:03:19,520 --> 00:03:22,919
Speaker 1: what was unique about my voice, obviously everything they had

56
00:03:22,919 --> 00:03:25,639
Speaker 1: the building blocks they needed to make a fake. Then

57
00:03:25,639 --> 00:03:28,120
Speaker 1: we sent Jose a set of sentences we wanted robot

58
00:03:28,160 --> 00:03:31,040
Speaker 1: care to say, and he used another set of algorithms

59
00:03:31,080 --> 00:03:33,359
Speaker 1: to turn the text into what we heard. The way

60
00:03:33,400 --> 00:03:35,360
Speaker 1: they do this is they use it's called a generative

61
00:03:35,480 --> 00:03:40,440
Speaker 1: adversarial network again, which is a system where one neural

62
00:03:40,480 --> 00:03:43,720
Speaker 1: net tries to trick another one a thousand times per second,

63
00:03:44,240 --> 00:03:47,120
Speaker 1: So each time the second network to texts of fake,

64
00:03:47,560 --> 00:03:50,960
Speaker 1: the first one tries again. It basically learns from its mistakes,

65
00:03:51,000 --> 00:03:53,800
Speaker 1: and once it tricks its adversary, it's ready to show

66
00:03:53,800 --> 00:03:57,440
Speaker 1: its results. In our case, liar bird pits my fake

67
00:03:57,520 --> 00:04:00,200
Speaker 1: voice against my real voice until it sounds like this

68
00:04:00,800 --> 00:04:04,800
Speaker 1: sub dog Scara Karen. One of the reasons why this

69
00:04:04,840 --> 00:04:07,160
Speaker 1: story has stuck with me is because it feels like

70
00:04:07,200 --> 00:04:10,360
Speaker 1: we're just at the beginning of tapping the potential and

71
00:04:10,400 --> 00:04:14,480
Speaker 1: the potential for harm of deep fakes. May be remembered

72
00:04:14,480 --> 00:04:17,240
Speaker 1: as the year of the first significant deep fake crime.

73
00:04:17,720 --> 00:04:20,920
Speaker 1: An employee at a UK based energy firm believed he

74
00:04:21,000 --> 00:04:23,600
Speaker 1: was on the phone to his boss and followed instructions

75
00:04:23,640 --> 00:04:27,560
Speaker 1: to transfer two hundred thousand pounds to a scammers bank account.

76
00:04:28,160 --> 00:04:30,720
Speaker 1: That certainly won't be the last deep fatefore we hear of,

77
00:04:30,960 --> 00:04:35,200
Speaker 1: and it raises questions about responsibility and accountability. Who's liable

78
00:04:35,240 --> 00:04:38,000
Speaker 1: in a case like this. Facebook has actually gone so

79
00:04:38,040 --> 00:04:41,320
Speaker 1: far as to create a deep fake detection challenge to

80
00:04:41,400 --> 00:04:44,760
Speaker 1: get the best minds thinking about deep fakes and how

81
00:04:44,760 --> 00:04:46,960
Speaker 1: we might solve the problem and offering like a million

82
00:04:46,960 --> 00:04:50,200
Speaker 1: dollar prize reach prize, which is like a dollar but

83
00:04:50,240 --> 00:04:52,800
Speaker 1: it also shows how important the issue is, especially when

84
00:04:52,800 --> 00:04:55,279
Speaker 1: a company like Facebook gets behind it. Um. There's another

85
00:04:55,320 --> 00:04:59,720
Speaker 1: side to deep fake technology that actually highlights this dichotomy

86
00:04:59,760 --> 00:05:01,600
Speaker 1: and tech anology right now, which is that it can

87
00:05:01,640 --> 00:05:05,120
Speaker 1: be used for menacing purposes but also really powerful and

88
00:05:05,160 --> 00:05:08,560
Speaker 1: beautiful applications. Jose goes on to talk about how liar

89
00:05:08,640 --> 00:05:11,000
Speaker 1: Bird can be used with als patients and give them

90
00:05:11,040 --> 00:05:14,560
Speaker 1: the ability to speak when they've lost all ability to speak,

91
00:05:14,800 --> 00:05:16,960
Speaker 1: when it could give them the opportunity to speak in

92
00:05:17,120 --> 00:05:19,920
Speaker 1: a version of their voice to their children again, which

93
00:05:19,960 --> 00:05:23,200
Speaker 1: is quite profound. One area where I think technology is

94
00:05:23,200 --> 00:05:26,560
Speaker 1: a powerhouse for change is in medicine. Technologists and doctors

95
00:05:26,560 --> 00:05:30,080
Speaker 1: alike are looking at AI to predict, treat, and diagnose,

96
00:05:30,520 --> 00:05:33,359
Speaker 1: you know, everything from depression to cancer, and that's a

97
00:05:33,480 --> 00:05:36,440
Speaker 1: very wide spectrum. And it reminds me of one of

98
00:05:36,480 --> 00:05:39,880
Speaker 1: my favorite interviews that you did, which was your interview

99
00:05:39,920 --> 00:05:43,400
Speaker 1: with Saddartha. Muker g just so fascinated by this article

100
00:05:43,440 --> 00:05:46,000
Speaker 1: he'd written for The New Yoker called AI versus m

101
00:05:46,080 --> 00:05:48,680
Speaker 1: D which laid out all of the kind of benefits

102
00:05:48,680 --> 00:05:52,480
Speaker 1: and potential applications of using AI in medicine, including some

103
00:05:52,640 --> 00:05:54,800
Speaker 1: of the downside such as the black box problem of

104
00:05:54,839 --> 00:05:57,720
Speaker 1: AI that you mentioned, not knowing why an algorithm has

105
00:05:57,720 --> 00:06:00,400
Speaker 1: made a recommendation, and also another problem, which is that

106
00:06:00,440 --> 00:06:02,840
Speaker 1: if we really too much on technology, it can erode

107
00:06:02,920 --> 00:06:05,920
Speaker 1: human skills. There is a fear that AI could move

108
00:06:06,000 --> 00:06:08,039
Speaker 1: us into a very black and white way of thinking.

109
00:06:08,400 --> 00:06:11,080
Speaker 1: The computer says you have cancer, or the computer says

110
00:06:11,400 --> 00:06:14,040
Speaker 1: you to have your liver removed, said Arthur, who is

111
00:06:14,080 --> 00:06:16,920
Speaker 1: one of the world's foremost oncologists and a Pullet Surprise

112
00:06:16,960 --> 00:06:21,440
Speaker 1: winning author. Provided a different perspective. There is something very

113
00:06:21,480 --> 00:06:25,880
Speaker 1: fundamental about the human brain, a scientist's brain, a doctor's brain,

114
00:06:25,920 --> 00:06:29,640
Speaker 1: and artist's brain that asks questions in a fundamentally different manner,

115
00:06:29,960 --> 00:06:32,400
Speaker 1: the why question. Why did this happen in this person

116
00:06:32,480 --> 00:06:35,719
Speaker 1: in this time? Why does the melanoma appear in the

117
00:06:35,720 --> 00:06:38,280
Speaker 1: first calase? What is the molecular basis of that appearance.

118
00:06:38,480 --> 00:06:41,640
Speaker 1: The most interesting mysteries of medicine remain mysteries that have

119
00:06:41,720 --> 00:06:45,159
Speaker 1: to do with the why. Once we give up some

120
00:06:45,320 --> 00:06:49,440
Speaker 1: of the diagnostic pattern recognition material to machines, it will

121
00:06:49,480 --> 00:06:50,919
Speaker 1: be time to play. It will be the time to

122
00:06:50,960 --> 00:06:54,960
Speaker 1: play in the arena of human therapeutics, human biology, the

123
00:06:55,040 --> 00:06:58,680
Speaker 1: complexity of the human interaction, the art of medicine. My

124
00:06:58,800 --> 00:07:01,360
Speaker 1: hope is that medicine, in being more playful, will become

125
00:07:01,760 --> 00:07:06,120
Speaker 1: more compassionate, more able to take into account individuals and

126
00:07:06,160 --> 00:07:10,920
Speaker 1: their individual destinies rather than bucketing people in big categories.

127
00:07:11,720 --> 00:07:15,360
Speaker 1: It means having more time to spend with humans. You know,

128
00:07:15,400 --> 00:07:19,360
Speaker 1: we are so constrained by time that even compassion gets

129
00:07:19,360 --> 00:07:23,520
Speaker 1: three minutes, We won't become more robotic, will become less

130
00:07:23,600 --> 00:07:28,520
Speaker 1: robotic as the robots enter our own So Dartha's point

131
00:07:28,680 --> 00:07:32,040
Speaker 1: is that these tools could make doctors more efficient so

132
00:07:32,080 --> 00:07:34,440
Speaker 1: that they can provide better care. It sort of takes

133
00:07:34,480 --> 00:07:39,720
Speaker 1: the grunt work out of medicine and puts the patient

134
00:07:39,880 --> 00:07:43,120
Speaker 1: care work back in the doctor's hands. This idea that

135
00:07:43,200 --> 00:07:46,600
Speaker 1: technology can actually allow us to be more human, make

136
00:07:46,680 --> 00:07:50,120
Speaker 1: us more empathetic, is fascinating, and it also raises the

137
00:07:50,200 --> 00:07:52,920
Speaker 1: questions about new types of skills that may need to

138
00:07:52,920 --> 00:07:55,760
Speaker 1: be developed in an age of AI. Yeah, and Regina

139
00:07:55,800 --> 00:07:58,360
Speaker 1: Barslay from m I T spoke a lot about this.

140
00:07:58,560 --> 00:08:02,360
Speaker 1: How doctors have to now equip themselves with new ways

141
00:08:02,520 --> 00:08:06,480
Speaker 1: of translating data to patients, we still do not communicate

142
00:08:06,520 --> 00:08:10,040
Speaker 1: it to the patient because I think now there is

143
00:08:10,080 --> 00:08:13,080
Speaker 1: a walk to be done, not on computer science or AIPAD,

144
00:08:13,560 --> 00:08:16,200
Speaker 1: but really on the clinical side. What is the best

145
00:08:16,240 --> 00:08:20,280
Speaker 1: way to communicate it to the patient and what is

146
00:08:20,720 --> 00:08:23,920
Speaker 1: um you know, the past that you're going to give them.

147
00:08:23,920 --> 00:08:25,920
Speaker 1: It is not just enough to say you know you

148
00:08:25,960 --> 00:08:30,600
Speaker 1: are high risk. You need to propose some suggestion and solution.

149
00:08:31,200 --> 00:08:35,520
Speaker 1: So currently the clinical stuff is thinking and looking at

150
00:08:35,520 --> 00:08:39,760
Speaker 1: the ways of effective you know, clinical engagement with a patient.

151
00:08:40,480 --> 00:08:42,880
Speaker 1: You know, I speaking of data. You all know Harri

152
00:08:43,080 --> 00:08:45,520
Speaker 1: was another person who made you and me think about

153
00:08:45,600 --> 00:08:49,840
Speaker 1: humans as reducible to data. I think he's mostly known

154
00:08:49,960 --> 00:08:52,920
Speaker 1: as a historian and for his book Sapiens, but you

155
00:08:52,960 --> 00:08:55,280
Speaker 1: spoke with him about the data we produce as humans

156
00:08:55,320 --> 00:08:58,439
Speaker 1: and how that influences our relationship with technology. That's right,

157
00:08:58,440 --> 00:09:00,760
Speaker 1: which is the topic of his book Daus. And he

158
00:09:00,800 --> 00:09:03,760
Speaker 1: has this phrase data is m to describe how we've

159
00:09:03,840 --> 00:09:07,200
Speaker 1: kind of come to worship the data we create and

160
00:09:07,360 --> 00:09:12,400
Speaker 1: our own technological creations. So what happens when based on

161
00:09:12,440 --> 00:09:15,520
Speaker 1: all of our past behavior, AI starts to know us

162
00:09:15,640 --> 00:09:18,440
Speaker 1: better than we know ourselves. Here's a clip from You

163
00:09:18,520 --> 00:09:22,320
Speaker 1: are talking about exactly that. When we talk about AI,

164
00:09:22,600 --> 00:09:26,640
Speaker 1: we tend to greatly exaggerate the potential abilities, but at

165
00:09:26,679 --> 00:09:30,560
Speaker 1: the same time we also tend to exaggerate the abilities

166
00:09:30,640 --> 00:09:35,360
Speaker 1: of humans. People say that AI is not going to

167
00:09:35,440 --> 00:09:39,080
Speaker 1: take over our lives because it's very imperfect and it

168
00:09:39,160 --> 00:09:42,880
Speaker 1: won't be able to know us perfectly. But what people

169
00:09:43,000 --> 00:09:47,280
Speaker 1: forget is that humans often have a very poor understanding

170
00:09:47,600 --> 00:09:52,719
Speaker 1: of themselves, of the desires, of their emotions, of their

171
00:09:52,760 --> 00:09:57,400
Speaker 1: mental states. For AI to take over your life, it

172
00:09:57,559 --> 00:10:01,160
Speaker 1: doesn't need to know you perfectly, just need to know

173
00:10:01,280 --> 00:10:04,720
Speaker 1: you better than you know yourself. And that's not very

174
00:10:04,720 --> 00:10:08,360
Speaker 1: difficult because we often don't know the most important things

175
00:10:08,480 --> 00:10:11,160
Speaker 1: about ourselves. So, but let's say you could turn back

176
00:10:11,160 --> 00:10:14,680
Speaker 1: the clock to being fifteen, would you have wanted to

177
00:10:14,800 --> 00:10:17,560
Speaker 1: live in a world where there was sufficiently good sensors

178
00:10:17,640 --> 00:10:21,400
Speaker 1: to monitor your eyes, your eye movement, you're breathing, you know,

179
00:10:21,400 --> 00:10:23,640
Speaker 1: while you're going about your daily life, and then to

180
00:10:23,720 --> 00:10:27,199
Speaker 1: interpret that and say to you you have all more

181
00:10:27,240 --> 00:10:30,080
Speaker 1: likely than not you're gay. That's a very good question

182
00:10:30,440 --> 00:10:34,080
Speaker 1: which will become very practical questions in a few years.

183
00:10:34,440 --> 00:10:38,240
Speaker 1: And the way that I grew up and developed. It

184
00:10:38,320 --> 00:10:42,480
Speaker 1: would have been a very bad idea. I wouldn't like

185
00:10:42,760 --> 00:10:46,480
Speaker 1: to receive this kind of insight from form a machine.

186
00:10:46,800 --> 00:10:49,640
Speaker 1: I'm not sure how I would have dealt with it

187
00:10:49,760 --> 00:10:51,760
Speaker 1: when I was fifteen, you know, in Israel in the

188
00:10:51,840 --> 00:10:54,840
Speaker 1: nineteen eighties, and maybe partly it was you know, a

189
00:10:54,880 --> 00:10:57,800
Speaker 1: defense mechanism. In the future too. It it depends where

190
00:10:57,840 --> 00:11:02,400
Speaker 1: you live. Brunei has instituted the death penalty for gay people,

191
00:11:02,440 --> 00:11:05,720
Speaker 1: at least for people engaged in homosexual sex. So if

192
00:11:05,760 --> 00:11:08,480
Speaker 1: I'm a teenager in Brunei, I don't want to be

193
00:11:08,520 --> 00:11:11,720
Speaker 1: told by the computer that I'm gay, because the computer

194
00:11:11,800 --> 00:11:14,079
Speaker 1: will then be able to tell that to the police

195
00:11:14,120 --> 00:11:17,679
Speaker 1: and to the authorities as well. So the apps we use,

196
00:11:17,800 --> 00:11:20,320
Speaker 1: the product we buy, the number of steps we take,

197
00:11:20,679 --> 00:11:23,920
Speaker 1: the delivery I ordered last night, that will becomes data,

198
00:11:24,400 --> 00:11:27,080
Speaker 1: and that data can feed into neural networks to create

199
00:11:27,120 --> 00:11:30,480
Speaker 1: statistical models of us and what we might do next,

200
00:11:30,920 --> 00:11:33,920
Speaker 1: sometimes in order to diagnose a medical condition, and other

201
00:11:33,960 --> 00:11:37,720
Speaker 1: times to sell us a product. Here's V again, looking

202
00:11:37,760 --> 00:11:41,320
Speaker 1: to the future, say ten twenty years. The danger is

203
00:11:41,600 --> 00:11:43,720
Speaker 1: if I still don't know that I'm gay, but the

204
00:11:43,800 --> 00:11:47,920
Speaker 1: government and Coca Cola and and Amazon and Google. They

205
00:11:47,920 --> 00:11:51,680
Speaker 1: already know it. I'm at a huge disadvantage. So it

206
00:11:51,760 --> 00:11:55,600
Speaker 1: could be something as frightening as the secret police coming

207
00:11:55,720 --> 00:11:58,880
Speaker 1: and taking me to a concentration camp. But it could

208
00:11:58,960 --> 00:12:03,199
Speaker 1: also be something Coca Cola knowing that I'm gay, they

209
00:12:03,200 --> 00:12:05,959
Speaker 1: want to sell me a new drink, and they choose

210
00:12:06,040 --> 00:12:10,480
Speaker 1: the advertisement with the shirtless guy and not the advertisement

211
00:12:10,840 --> 00:12:13,280
Speaker 1: with the girl in the bikini. And next day morning

212
00:12:13,400 --> 00:12:15,880
Speaker 1: I go and I buy this soft drink and I

213
00:12:15,920 --> 00:12:19,520
Speaker 1: don't even know why, and they have this huge advantage

214
00:12:19,559 --> 00:12:23,480
Speaker 1: over me and can manipulate me in all kinds of ways. Well,

215
00:12:23,480 --> 00:12:26,760
Speaker 1: as you've all brought up soda, I was not allowed

216
00:12:26,760 --> 00:12:29,960
Speaker 1: to drink soda as a child. My parents tricked me

217
00:12:30,000 --> 00:12:33,640
Speaker 1: into thinking that Seltzer was soda. I later found out

218
00:12:33,720 --> 00:12:38,000
Speaker 1: that soda is soda and Seltzer is water. And somehow

219
00:12:38,080 --> 00:12:41,640
Speaker 1: the Seltzer of it all is the perfect segue because

220
00:12:41,679 --> 00:12:44,400
Speaker 1: AI is not only being used to sell a product,

221
00:12:45,160 --> 00:12:49,240
Speaker 1: it's also being used to create products like Seltzer in

222
00:12:49,280 --> 00:12:52,640
Speaker 1: the R and D research and development phase. And for

223
00:12:52,679 --> 00:12:56,960
Speaker 1: this bombs episode, Julian remember Juliana, a lovely producer, And

224
00:12:57,080 --> 00:13:00,400
Speaker 1: I went to meet the company behind the gastrograph app,

225
00:13:00,440 --> 00:13:04,600
Speaker 1: which is using consumer preference data to make predictions about

226
00:13:04,760 --> 00:13:07,920
Speaker 1: new flavors. People might like, I think you hate Nettle,

227
00:13:08,640 --> 00:13:15,680
Speaker 1: think against Analytical Flavor Systems or a f S is

228
00:13:15,720 --> 00:13:19,600
Speaker 1: tucked away down a side street in Chinatown, up in

229
00:13:19,640 --> 00:13:23,280
Speaker 1: a third story walk up. It does actually and when

230
00:13:23,280 --> 00:13:25,760
Speaker 1: we were there, the office was still waiting for furniture.

231
00:13:25,800 --> 00:13:27,840
Speaker 1: You know, it had this We're going to disrupt the

232
00:13:27,880 --> 00:13:31,360
Speaker 1: industry vibe. We moved in like a month or two ago.

233
00:13:31,640 --> 00:13:34,760
Speaker 1: But it turns out you can't just buy office furniture.

234
00:13:36,040 --> 00:13:38,720
Speaker 1: We met the founder of the company, Jason Cohen, and

235
00:13:38,760 --> 00:13:41,079
Speaker 1: we believe that in the future, in order to be competitive,

236
00:13:41,160 --> 00:13:43,680
Speaker 1: you have to be targeted. That that there won't be

237
00:13:43,760 --> 00:13:47,360
Speaker 1: any billion dollar brands right in ten years if you

238
00:13:47,400 --> 00:13:49,800
Speaker 1: don't have an m b A. Here's what Jason's talking about.

239
00:13:50,040 --> 00:13:52,360
Speaker 1: Think about the coffee you drank this morning, is it

240
00:13:52,559 --> 00:13:55,560
Speaker 1: Third Wave? Did you get it from Starbucks or an

241
00:13:55,600 --> 00:13:59,040
Speaker 1: indie roaster. Food and beverage companies are moving more and

242
00:13:59,120 --> 00:14:01,840
Speaker 1: more towards knee markets. The problem is that they have

243
00:14:02,000 --> 00:14:05,120
Speaker 1: very old school ways of developing new products. But a

244
00:14:05,280 --> 00:14:08,560
Speaker 1: FS is offering another way to reach those customers, and

245
00:14:08,600 --> 00:14:13,080
Speaker 1: that's by allowing companies to formulate more specific products using AI.

246
00:14:13,400 --> 00:14:16,079
Speaker 1: Here's Jason again. Usually the way that things are done

247
00:14:16,120 --> 00:14:20,760
Speaker 1: today is you get some conceptual brief. You might say,

248
00:14:20,800 --> 00:14:23,400
Speaker 1: we want to develop a new fruit flavored beverage for

249
00:14:24,160 --> 00:14:27,360
Speaker 1: Japanese millennial women. Right, you would look at what other

250
00:14:27,400 --> 00:14:29,840
Speaker 1: fruit flavored beverages are out there. You'd look at your

251
00:14:29,840 --> 00:14:31,600
Speaker 1: own product lineup and say, well, we already have a

252
00:14:31,680 --> 00:14:35,040
Speaker 1: lemon flavor, and we're going to send out these briefs,

253
00:14:35,080 --> 00:14:36,880
Speaker 1: and we're gonna send these out to these flavor companies.

254
00:14:36,880 --> 00:14:38,920
Speaker 1: We're gonna see what other fruits we can get, and

255
00:14:38,920 --> 00:14:41,360
Speaker 1: you're gonna wind up with very mainstream things. You're gonna

256
00:14:41,400 --> 00:14:45,040
Speaker 1: wind up with peach and mango and strawberry and grapefruit, right,

257
00:14:45,040 --> 00:14:48,680
Speaker 1: and maybe you'll wind up with something interesting like low

258
00:14:48,800 --> 00:14:53,360
Speaker 1: quad or uzo or dragon fruit. Right. And then you're

259
00:14:53,360 --> 00:14:57,680
Speaker 1: gonna have your own consumer tasting panel internally hopefully uh

260
00:14:57,680 --> 00:14:59,520
Speaker 1: and you're gonna have them taste it, and they're gonna

261
00:14:59,560 --> 00:15:02,240
Speaker 1: have to like some of those more than your current

262
00:15:02,280 --> 00:15:05,520
Speaker 1: offering or more than a competitors. After you've done all

263
00:15:05,520 --> 00:15:07,360
Speaker 1: of this work, which sometimes costs in the tens of

264
00:15:07,360 --> 00:15:09,680
Speaker 1: thousands of dollars in order to have the samples developed.

265
00:15:09,680 --> 00:15:11,760
Speaker 1: How the samples sent to you recruit the consumers, but

266
00:15:11,960 --> 00:15:14,320
Speaker 1: the product in front of the consumers, that data is

267
00:15:14,400 --> 00:15:17,360
Speaker 1: only ever usable once. Right. All you get from that

268
00:15:17,400 --> 00:15:20,360
Speaker 1: as a binary yes no, sixty percent of the population

269
00:15:20,400 --> 00:15:23,280
Speaker 1: liked more than the competitors, and so what we're doing

270
00:15:23,560 --> 00:15:27,960
Speaker 1: is entirely different. Jason's team wants to take product development

271
00:15:28,000 --> 00:15:31,120
Speaker 1: out of the yes no binary. Instead of just saying

272
00:15:31,320 --> 00:15:35,040
Speaker 1: coke or pepsi, they can calculate which parts of each

273
00:15:35,080 --> 00:15:38,760
Speaker 1: soft drink people liked and disliked, and then a f

274
00:15:38,920 --> 00:15:41,480
Speaker 1: s can make entirely new flavors based on what a

275
00:15:41,520 --> 00:15:44,600
Speaker 1: person kind of liked about pepsi and kind of liked

276
00:15:44,600 --> 00:15:48,280
Speaker 1: about coke. And finally, they can transfer those preferences to

277
00:15:48,640 --> 00:15:52,600
Speaker 1: entirely different demographics. So what might someone in Mexico want

278
00:15:52,640 --> 00:15:55,280
Speaker 1: to taste in their cola compared to what a Japanese

279
00:15:55,280 --> 00:15:58,720
Speaker 1: millennial woman might want to taste in her cola? This

280
00:15:58,800 --> 00:16:02,080
Speaker 1: is what Jason believes is truly disruptive, being able to

281
00:16:02,080 --> 00:16:04,400
Speaker 1: say to a brand, if you want to launch in Mexico,

282
00:16:04,480 --> 00:16:07,360
Speaker 1: you should tweak your flavors in this way because we're

283
00:16:07,400 --> 00:16:10,000
Speaker 1: actually able to collect this data, develop a data set,

284
00:16:10,240 --> 00:16:12,720
Speaker 1: and make predictions. So we could take data from say, white,

285
00:16:12,720 --> 00:16:15,040
Speaker 1: twenty to three year old college educated males, and use

286
00:16:15,080 --> 00:16:17,240
Speaker 1: that to predict what every other population in the United

287
00:16:17,280 --> 00:16:19,120
Speaker 1: States is going to perceive in that product. And so

288
00:16:19,160 --> 00:16:21,520
Speaker 1: we're taking an industry that has never seen predictions of

289
00:16:21,520 --> 00:16:24,960
Speaker 1: any kind before and finally being able to actually predict things,

290
00:16:25,120 --> 00:16:27,280
Speaker 1: predict who's gonna like it, how much they're gonna like

291
00:16:27,360 --> 00:16:29,360
Speaker 1: it right, and what we can do to optimize it

292
00:16:29,440 --> 00:16:31,600
Speaker 1: so that they like it more. We can actually create

293
00:16:31,640 --> 00:16:33,440
Speaker 1: products that no one would have ever thought of, and

294
00:16:33,560 --> 00:16:35,200
Speaker 1: no one ever would have thought that a segment of

295
00:16:35,240 --> 00:16:37,160
Speaker 1: the population would have liked. And this is something that

296
00:16:37,200 --> 00:16:38,920
Speaker 1: we now do with the companies that we work with.

297
00:16:39,160 --> 00:16:41,760
Speaker 1: We did talk quite a bit about developing a pine

298
00:16:41,800 --> 00:16:45,240
Speaker 1: flavored beverage in Japan. When we first showed these results

299
00:16:45,240 --> 00:16:47,720
Speaker 1: to a company there, they said, natzu desca, do you

300
00:16:47,760 --> 00:16:50,920
Speaker 1: mean pineapple? Because it was just so out of the yeah,

301
00:16:50,960 --> 00:16:54,320
Speaker 1: out of the ordinary. The way Jason and his team

302
00:16:54,360 --> 00:16:56,640
Speaker 1: are able to get such nuanced data is with an

303
00:16:56,680 --> 00:16:59,960
Speaker 1: app they developed called Gastrograph. Gastrograph looks a lot like

304
00:17:00,200 --> 00:17:02,800
Speaker 1: the flavor wheels they use in coffee and wine tasting

305
00:17:02,840 --> 00:17:04,840
Speaker 1: to help people map out what they taste when they

306
00:17:04,880 --> 00:17:07,520
Speaker 1: try a new product. We think of every flavor, romance,

307
00:17:07,520 --> 00:17:10,000
Speaker 1: texture as a signature. You can have the five basic

308
00:17:10,040 --> 00:17:13,240
Speaker 1: flavors bitter, sweet, salty, sour mommy, and then underneath that

309
00:17:13,240 --> 00:17:16,080
Speaker 1: you're gonna have categorical flavors like fruity earth, the herbaceous nuts,

310
00:17:16,080 --> 00:17:18,800
Speaker 1: and seedge roasted. I mentioned that he's a professional taster, right,

311
00:17:18,960 --> 00:17:22,240
Speaker 1: and then underneath that you can have subcategorical like citrus,

312
00:17:22,320 --> 00:17:24,840
Speaker 1: or specific like lemon are very specific like Meyer lemon

313
00:17:24,840 --> 00:17:27,280
Speaker 1: are very very specific like Meyer lemons as right, So

314
00:17:27,400 --> 00:17:31,199
Speaker 1: all of those signatures exist in some infinite dimensional space

315
00:17:32,080 --> 00:17:36,480
Speaker 1: flavor space. So, car, since you're such a Seltzer fiend,

316
00:17:36,560 --> 00:17:38,919
Speaker 1: we demo gaster graph and got a feel for the

317
00:17:38,960 --> 00:17:42,040
Speaker 1: app by doing a Seltzer tasting. Yeah, we tasted five

318
00:17:42,119 --> 00:17:44,560
Speaker 1: Seltzers that are already on the market that you'd buy,

319
00:17:44,640 --> 00:17:47,920
Speaker 1: like seven eleven, and we rated different components of them.

320
00:17:48,000 --> 00:17:50,359
Speaker 1: So if I tasted fruit, I'd rate that from zero

321
00:17:50,440 --> 00:17:54,040
Speaker 1: to five, and I could add adjectives like strawberry or mango.

322
00:17:54,320 --> 00:17:56,920
Speaker 1: You know, I'm not into like Seltzer two point oh

323
00:17:56,920 --> 00:18:01,919
Speaker 1: with like kumquat flavored sparkling seltzer. You know, I plane vintage,

324
00:18:02,400 --> 00:18:05,720
Speaker 1: but a f s is Resident chemist Ryan On agreed

325
00:18:05,760 --> 00:18:09,399
Speaker 1: to formulate a seltzer based on just our extremely small

326
00:18:09,480 --> 00:18:12,560
Speaker 1: data set. Um, you'd probably be a pretty fast process.

327
00:18:12,600 --> 00:18:15,520
Speaker 1: So we have tons of seltzer data. What we would

328
00:18:15,520 --> 00:18:19,000
Speaker 1: want to do is um run through a couple of

329
00:18:19,000 --> 00:18:20,840
Speaker 1: different flavors to get an idea of the types of

330
00:18:20,880 --> 00:18:23,920
Speaker 1: things that you like, build a model specifically around that,

331
00:18:24,520 --> 00:18:27,280
Speaker 1: run an optimization, predict a new flavor that you've never

332
00:18:27,320 --> 00:18:31,120
Speaker 1: had before, and then have you tried again. So after

333
00:18:31,160 --> 00:18:34,760
Speaker 1: we did all of that, we went home. We sat

334
00:18:34,800 --> 00:18:38,440
Speaker 1: in our hands, and then we went back to the office,

335
00:18:38,600 --> 00:18:41,360
Speaker 1: which actually a little bit more furniture when we got

336
00:18:41,400 --> 00:18:45,560
Speaker 1: back there to see if they actually could create a

337
00:18:45,600 --> 00:18:48,359
Speaker 1: seltzer that both Julian and I would like. So it

338
00:18:48,440 --> 00:18:51,200
Speaker 1: was a blind tasting. Here we go, all right, drink

339
00:18:51,200 --> 00:18:56,399
Speaker 1: one pair. I don't know why I think such a

340
00:18:56,400 --> 00:19:02,680
Speaker 1: thing against pears. The number two berries, that's delish. Let's

341
00:19:02,680 --> 00:19:05,600
Speaker 1: see what we got here. I wouldn't know huckleberry if

342
00:19:05,640 --> 00:19:10,119
Speaker 1: it hit me in the face, honestly, but whatever, Number

343
00:19:10,119 --> 00:19:18,119
Speaker 1: three honestly, curry. Do you not taste curry like there's spice?

344
00:19:18,200 --> 00:19:24,920
Speaker 1: It's so interesting? One? Oh, I love it like hops

345
00:19:24,920 --> 00:19:29,919
Speaker 1: and I'm si care just taste like grass? What are you?

346
00:19:32,359 --> 00:19:35,800
Speaker 1: We're done? So the first thing you guys should tell

347
00:19:35,880 --> 00:19:40,000
Speaker 1: us is which is your favorite? For me? Is number three?

348
00:19:40,119 --> 00:19:42,520
Speaker 1: Just because I like the complexity of flavor. Three was

349
00:19:42,560 --> 00:19:46,919
Speaker 1: a beverage too. I'm used to and I've probably had before.

350
00:19:46,960 --> 00:19:51,520
Speaker 1: Maybe three. I really enjoyed one I did not like.

351
00:19:52,000 --> 00:19:55,000
Speaker 1: I don't like those flavors. We have to reveal. We're

352
00:19:55,000 --> 00:19:58,600
Speaker 1: almost at the reveal can to clarify. Your job as

353
00:19:58,640 --> 00:20:03,560
Speaker 1: a company is to predict future products that people will

354
00:20:03,640 --> 00:20:07,200
Speaker 1: enjoy and come back to. Yes, So in that regard,

355
00:20:07,520 --> 00:20:13,720
Speaker 1: three was a winner. All right? Oh my god, tes us.

356
00:20:13,840 --> 00:20:15,639
Speaker 1: So we had to do one product that was going

357
00:20:15,720 --> 00:20:18,159
Speaker 1: to be optimal for both of you, and we got it.

358
00:20:18,160 --> 00:20:24,240
Speaker 1: It was number three. They nailed it. You could say

359
00:20:24,240 --> 00:20:27,560
Speaker 1: that we got flavor hacked up. Here. This this blue graph,

360
00:20:27,600 --> 00:20:29,840
Speaker 1: this is saying that there's a you know, a seventy

361
00:20:29,920 --> 00:20:33,680
Speaker 1: percent chance that you would give this a six? Did

362
00:20:33,680 --> 00:20:36,240
Speaker 1: I give it a six? I think I did? You? Did?

363
00:20:36,280 --> 00:20:38,120
Speaker 1: You both gave it a six? Um? So we were

364
00:20:38,119 --> 00:20:40,480
Speaker 1: pretty confident on this, but we didn't have that level

365
00:20:40,520 --> 00:20:43,000
Speaker 1: of confidence that we saw this. What I tasted is

366
00:20:43,040 --> 00:20:46,919
Speaker 1: not something I ever predicted I would have liked, but

367
00:20:47,640 --> 00:20:50,639
Speaker 1: it's absolutely something I will continue to think about. It

368
00:20:50,680 --> 00:20:53,240
Speaker 1: was such a unique flavor and it's actually something I

369
00:20:53,240 --> 00:20:55,520
Speaker 1: would buy. It's just that I had never tasted something

370
00:20:55,520 --> 00:20:57,800
Speaker 1: like it before. So when I first tasted it, I

371
00:20:57,880 --> 00:21:01,760
Speaker 1: was like, this is strange. In the case scenario, gastrographs

372
00:21:01,800 --> 00:21:05,560
Speaker 1: AI can help companies create foods that satisfying more specific

373
00:21:05,640 --> 00:21:08,960
Speaker 1: tastes and even bring people more joy, and that's good

374
00:21:08,960 --> 00:21:11,879
Speaker 1: for business. Instead of making huge bets and trying to

375
00:21:11,920 --> 00:21:14,920
Speaker 1: market a product to an entire country, a f S

376
00:21:14,960 --> 00:21:17,520
Speaker 1: has created a way to make more specialized bets and

377
00:21:17,600 --> 00:21:20,800
Speaker 1: help companies tap into those niches. And this isn't about

378
00:21:20,880 --> 00:21:24,720
Speaker 1: AI reducing our experiences to data. AI is being used

379
00:21:24,720 --> 00:21:29,520
Speaker 1: to change how we experience the world. More sleepwalkers after

380
00:21:29,560 --> 00:21:43,359
Speaker 1: the break So, Karen, I would have been quite nervous

381
00:21:43,359 --> 00:21:45,800
Speaker 1: to stand in the shoes of Jason and Ryan and

382
00:21:45,840 --> 00:21:48,359
Speaker 1: the gastrograph team because I know that you're such a

383
00:21:48,440 --> 00:21:54,160
Speaker 1: connoisseur of Seltzer. How do they do? They did really

384
00:21:54,200 --> 00:21:57,560
Speaker 1: well actually, and I think it's important to mention that

385
00:21:57,680 --> 00:22:00,800
Speaker 1: they weren't trying to create something they thought I would

386
00:22:00,880 --> 00:22:04,639
Speaker 1: already like, like I love cranberry, right. They were trying

387
00:22:04,680 --> 00:22:08,680
Speaker 1: to create something that I hadn't tasted before and also liked.

388
00:22:08,760 --> 00:22:11,439
Speaker 1: So it was really difficult, and I think it also

389
00:22:11,560 --> 00:22:14,280
Speaker 1: shows that there's a bit of reversal in the way

390
00:22:14,280 --> 00:22:17,560
Speaker 1: that we do things. Companies have always used market research

391
00:22:17,680 --> 00:22:20,720
Speaker 1: to predict consumer preference, but it's often based on things

392
00:22:20,720 --> 00:22:24,640
Speaker 1: like focus groups or survey research. What we have now

393
00:22:25,080 --> 00:22:28,399
Speaker 1: is massive amounts of data being funneled through an algorithm

394
00:22:28,640 --> 00:22:32,000
Speaker 1: to deliver the perfect product for a very specific type

395
00:22:32,000 --> 00:22:37,679
Speaker 1: of person exactly. That's the or yeah, age, demographic, socio

396
00:22:37,720 --> 00:22:42,640
Speaker 1: economic race. They can target it to all these particular categories,

397
00:22:42,920 --> 00:22:46,639
Speaker 1: and I think this is cool and also a little unsettling.

398
00:22:46,880 --> 00:22:48,840
Speaker 1: I think as humans, we like to be in control,

399
00:22:48,920 --> 00:22:51,280
Speaker 1: you know. I like to think that my preferences are

400
00:22:51,320 --> 00:22:55,040
Speaker 1: just that, my own preferences, and this sort of up

401
00:22:55,200 --> 00:22:58,639
Speaker 1: ends that notion. You know, using pre existing data, I

402
00:22:58,720 --> 00:23:02,359
Speaker 1: can kind of be read they're reading me, and that

403
00:23:02,400 --> 00:23:05,240
Speaker 1: makes me feel a little less special. I do think

404
00:23:05,240 --> 00:23:07,920
Speaker 1: it's cool that companies are trying to service niche markets,

405
00:23:08,280 --> 00:23:10,640
Speaker 1: and I think that's a trend I would definitely get

406
00:23:10,680 --> 00:23:13,119
Speaker 1: on board with as far as AI being used to

407
00:23:13,119 --> 00:23:16,919
Speaker 1: make predictions. And the reason this gastrograph pieces interesting is

408
00:23:16,960 --> 00:23:20,159
Speaker 1: because it's a perfect demonstration that AI is not some

409
00:23:20,720 --> 00:23:22,920
Speaker 1: thing which is going to happen in the future. It's

410
00:23:23,000 --> 00:23:30,880
Speaker 1: here with us today. We can literally taste it already

411
00:23:34,800 --> 00:23:37,840
Speaker 1: AIS in our lives. It's interpreting our data, is analyzing

412
00:23:37,840 --> 00:23:41,560
Speaker 1: our preferences, is predicting our behaviors. But we're just starting

413
00:23:41,600 --> 00:23:44,560
Speaker 1: to respond to what that means culturally. And so there's

414
00:23:44,560 --> 00:23:46,760
Speaker 1: a lot of new technologies and new issues that were

415
00:23:46,800 --> 00:23:50,680
Speaker 1: very excited to get our hands dirty with on season two. Absolutely,

416
00:23:50,760 --> 00:23:53,000
Speaker 1: and one of the important issues we're going to explore

417
00:23:53,080 --> 00:23:56,359
Speaker 1: is bias and technology. It's easy to think that algorithms

418
00:23:56,359 --> 00:23:59,439
Speaker 1: are neutral, but the reality is that technology is built

419
00:23:59,440 --> 00:24:02,520
Speaker 1: by someone, and that person's bias can be built into

420
00:24:02,520 --> 00:24:07,560
Speaker 1: a system. This Princeton professor named Ruha Benjamin has introduced

421
00:24:07,560 --> 00:24:10,840
Speaker 1: a concept directly related to this, called the new Gym code,

422
00:24:11,200 --> 00:24:15,440
Speaker 1: which asks us to consider the inequities encoded in algorithms. Well,

423
00:24:15,440 --> 00:24:18,360
Speaker 1: it's the algorithms and also the data they learned from right.

424
00:24:18,359 --> 00:24:22,560
Speaker 1: I mean AI harnesses the power of processing huge amounts

425
00:24:22,600 --> 00:24:24,800
Speaker 1: of data about things that have happened in the past

426
00:24:25,160 --> 00:24:27,399
Speaker 1: in order to predict a future, and so we have

427
00:24:27,480 --> 00:24:30,320
Speaker 1: to be very careful about what that data contains, or

428
00:24:30,320 --> 00:24:32,720
Speaker 1: we might not like the future. It's bits out. I

429
00:24:32,760 --> 00:24:35,200
Speaker 1: think it's a particularly relevant in the air of medicine.

430
00:24:35,240 --> 00:24:38,440
Speaker 1: We see huge, huge promise about honesting AI for better

431
00:24:38,440 --> 00:24:41,000
Speaker 1: medical outcomes, but we also need to be very careful

432
00:24:41,080 --> 00:24:43,399
Speaker 1: about how the data is being used and who has

433
00:24:43,440 --> 00:24:46,520
Speaker 1: access to it, and how can you prevent your data

434
00:24:46,560 --> 00:24:49,239
Speaker 1: from being used against you. Well, we have to think

435
00:24:49,280 --> 00:24:53,040
Speaker 1: about the potential for data to harm but also to

436
00:24:53,080 --> 00:24:58,320
Speaker 1: provide comfort and to drive innovation, sometimes extraordinary, very unexpected innovation.

437
00:24:59,000 --> 00:25:00,800
Speaker 1: There are two stories I can't wait to dive into

438
00:25:00,800 --> 00:25:03,600
Speaker 1: in season two. One is about a doctor using AI

439
00:25:03,760 --> 00:25:09,200
Speaker 1: to record and optimize conversational strategies with very sick patients.

440
00:25:09,760 --> 00:25:13,679
Speaker 1: What should they say, how and when? Another is about

441
00:25:13,760 --> 00:25:18,240
Speaker 1: using natural language processing to enable immersive conversations with holograms

442
00:25:18,280 --> 00:25:22,320
Speaker 1: of people from history, everyone from astronauts to Holocaust survivors.

443
00:25:22,760 --> 00:25:25,560
Speaker 1: In other words, using technology to bring history into the

444
00:25:25,600 --> 00:25:29,960
Speaker 1: present and ensuring we never forget our past it's wild,

445
00:25:31,320 --> 00:25:33,560
Speaker 1: so we're obviously looking forward to seeing you in the

446
00:25:33,560 --> 00:25:35,800
Speaker 1: next season. We have a lot of amazing stories lined up.

447
00:25:36,200 --> 00:25:38,399
Speaker 1: We'd love to hear from you guys about stories that

448
00:25:38,480 --> 00:25:40,560
Speaker 1: you want to hear or subjects that you want to

449
00:25:40,560 --> 00:25:44,240
Speaker 1: hear about, So tweet us at Sleepwalker's Pod on Twitter

450
00:25:44,600 --> 00:25:49,200
Speaker 1: obviously and on Instagram. We are Sleepwalkers Podcast. Yeah, thank

451
00:25:49,240 --> 00:25:51,200
Speaker 1: you so much. We love your feedback and we're really

452
00:25:51,200 --> 00:25:53,840
Speaker 1: looking forward to seeing you for season two very soon.

453
00:26:06,440 --> 00:26:10,200
Speaker 1: Sleepwalkers is a production of I Heart Radio and Unusual Productions.

454
00:26:10,680 --> 00:26:13,560
Speaker 1: For the latest AI news, live interviews, and behind the

455
00:26:13,640 --> 00:26:17,720
Speaker 1: scenes footage, find us on Instagram, at Sleepwalker's Podcast or

456
00:26:17,840 --> 00:26:22,199
Speaker 1: at Sleepwalker's podcast dot com. Sleepwalkers is hosted by me

457
00:26:22,359 --> 00:26:25,680
Speaker 1: Ozveloshin and co hosted by me Kara Price. Were produced

458
00:26:25,680 --> 00:26:29,040
Speaker 1: by Julian Weller with help from Jacopo Penzo and Taylor Chcogin,

459
00:26:29,280 --> 00:26:32,560
Speaker 1: mixing by Tristan McNeil and Julian Weller. Our story editor

460
00:26:32,640 --> 00:26:37,080
Speaker 1: is Matthew Riddle. Sleepwalkers is executive produced by me Ozveloshin

461
00:26:37,200 --> 00:26:40,920
Speaker 1: and Mangesh hat Together. For more podcasts from my Heart Radio,

462
00:26:41,040 --> 00:26:43,959
Speaker 1: visit the I Heart Radio app Apple podcasts, or wherever

463
00:26:44,000 --> 00:26:45,359
Speaker 1: you listen to your favorite shows.