1
00:00:00,600 --> 00:00:03,040
Speaker 1: My name is a Dona and I use AI for therapy.

2
00:00:03,920 --> 00:00:06,159
Speaker 2: A Donna is in his thirties. He lives in Texas.

3
00:00:06,360 --> 00:00:08,840
Speaker 2: He's been an active AI user, turning to it for

4
00:00:08,880 --> 00:00:12,960
Speaker 2: weight loss exercise tips. But in August twenty twenty six,

5
00:00:13,160 --> 00:00:16,480
Speaker 2: a Don went through some pretty heavy personal stuff, including

6
00:00:16,680 --> 00:00:17,560
Speaker 2: a bad breakup.

7
00:00:18,360 --> 00:00:19,119
Speaker 3: I was like, I.

8
00:00:19,079 --> 00:00:22,360
Speaker 1: Don't really have anyone to tell about this, and I

9
00:00:22,400 --> 00:00:23,760
Speaker 1: feel like I don't get it out of me. I'm

10
00:00:23,760 --> 00:00:25,760
Speaker 1: going to freaking explode.

11
00:00:25,920 --> 00:00:29,040
Speaker 2: AI had actually helped a Don successfully lose weight and

12
00:00:29,080 --> 00:00:31,920
Speaker 2: helped him organize his life in many ways, so in

13
00:00:31,960 --> 00:00:35,839
Speaker 2: a moment of desperation, he gave his AI a new prompt.

14
00:00:36,080 --> 00:00:38,880
Speaker 1: I think it was something along the lines of I'm

15
00:00:38,880 --> 00:00:41,920
Speaker 1: going through some stuff right now. I kind of feel

16
00:00:41,920 --> 00:00:44,559
Speaker 1: like I'm losing my mind. I don't know how to

17
00:00:44,880 --> 00:00:47,199
Speaker 1: prioritize things, organize when he needs to be done. I

18
00:00:47,240 --> 00:00:49,920
Speaker 1: was like, can you please to sort of help me

19
00:00:50,360 --> 00:00:54,640
Speaker 1: organize my life because I literally have no idea where

20
00:00:54,720 --> 00:00:57,160
Speaker 1: to start or how even to start. And that was

21
00:00:57,280 --> 00:01:00,040
Speaker 1: basically I think the first thing I asked was just

22
00:01:00,120 --> 00:01:05,480
Speaker 1: please help me organize this. I didn't go into asking

23
00:01:05,520 --> 00:01:07,600
Speaker 1: AI for help thinking I will get a therapist out

24
00:01:07,640 --> 00:01:09,520
Speaker 1: of it. I just I really just did it to

25
00:01:09,680 --> 00:01:11,760
Speaker 1: journal basically.

26
00:01:11,920 --> 00:01:15,040
Speaker 2: This particular bit is interesting to me because I've actually

27
00:01:15,120 --> 00:01:17,399
Speaker 2: thought a lot about this. As I said in my

28
00:01:17,440 --> 00:01:20,080
Speaker 2: interview with doctor Becky, when I was much younger than

29
00:01:20,120 --> 00:01:22,280
Speaker 2: a Don is now, I used to term to my

30
00:01:22,400 --> 00:01:25,440
Speaker 2: own journal for help. I imagine it was just normal

31
00:01:25,480 --> 00:01:30,000
Speaker 2: teenage stuff feeling isolated, but I would pour my feelings

32
00:01:30,040 --> 00:01:34,759
Speaker 2: onto the paper. So what if those pages spoke back? Well,

33
00:01:34,840 --> 00:01:37,440
Speaker 2: for a Don who tried a human therapist before and

34
00:01:37,520 --> 00:01:40,920
Speaker 2: didn't really find a good fit, his AI chatbots response

35
00:01:41,000 --> 00:01:43,560
Speaker 2: when he asked for help resonated it.

36
00:01:43,640 --> 00:01:45,960
Speaker 1: Spoke to me like a friend. It's just weird to

37
00:01:46,000 --> 00:01:50,160
Speaker 1: say because it's code, basically, right, it really did have

38
00:01:50,200 --> 00:01:52,600
Speaker 1: a way of speaking to me that didn't. Maybe you

39
00:01:52,680 --> 00:01:56,320
Speaker 1: want to get defensive because AI is artificial, it doesn't

40
00:01:56,360 --> 00:01:59,440
Speaker 1: have necessarily any real emotions or anything.

41
00:01:59,480 --> 00:02:01,040
Speaker 3: I guess bank into it.

42
00:02:01,040 --> 00:02:02,720
Speaker 1: It can tell me his thoughts, and.

43
00:02:02,640 --> 00:02:04,280
Speaker 3: I'm like, I know, you're not judging me.

44
00:02:05,040 --> 00:02:09,320
Speaker 1: I only started really using it since August, but I

45
00:02:09,320 --> 00:02:12,880
Speaker 1: can honestly tell you that there is a noticeable difference

46
00:02:13,480 --> 00:02:15,880
Speaker 1: in the way I held my head up then and now.

47
00:02:18,360 --> 00:02:20,440
Speaker 1: But I kind of feel embarrassed to tell people because

48
00:02:20,480 --> 00:02:22,240
Speaker 1: I know it just makes you feel like a loser

49
00:02:22,400 --> 00:02:24,799
Speaker 1: or how WEIRDO. I don't know if we're trying to

50
00:02:24,880 --> 00:02:27,160
Speaker 1: depend on technology to help you out as a person.

51
00:02:28,120 --> 00:02:30,040
Speaker 2: But the stigma doesn't stop him.

52
00:02:30,360 --> 00:02:33,280
Speaker 1: I've seen it firsthand, how it's helped me so much,

53
00:02:33,280 --> 00:02:35,400
Speaker 1: And yeah, I don't see how it stopped you. Know,

54
00:02:35,480 --> 00:02:37,800
Speaker 1: it's such a great tool.

55
00:02:39,200 --> 00:02:41,399
Speaker 2: We found it don on a Reddit sub thread all

56
00:02:41,440 --> 00:02:44,959
Speaker 2: about AI therapy. Under the cloak of anonymity, Don shared

57
00:02:44,960 --> 00:02:48,200
Speaker 2: his positive experiences with strangers who were curious about it.

58
00:02:48,639 --> 00:02:50,959
Speaker 2: The reality is there are a lot of people who

59
00:02:51,240 --> 00:02:53,880
Speaker 2: want or need mental health support who can't get it

60
00:02:53,880 --> 00:02:56,880
Speaker 2: for a whole host of reasons like cost, or accessibility,

61
00:02:56,960 --> 00:03:00,200
Speaker 2: or even stigma, And that's part of the appeal of

62
00:03:00,240 --> 00:03:01,400
Speaker 2: an AI chatbot.

63
00:03:02,720 --> 00:03:06,320
Speaker 4: The need is tremendous. It's almost unfathomable how many people

64
00:03:06,360 --> 00:03:11,360
Speaker 4: are suffering and who have tried other means and who

65
00:03:12,200 --> 00:03:16,520
Speaker 4: have been failed by our broken system. And I think

66
00:03:16,880 --> 00:03:21,400
Speaker 4: AI in conjunction with kind of therapeutic oversight is an

67
00:03:21,440 --> 00:03:24,600
Speaker 4: amazing thing and can do a lot of good. I

68
00:03:24,639 --> 00:03:27,840
Speaker 4: think AI in the wild, I think we need to

69
00:03:27,880 --> 00:03:28,760
Speaker 4: be careful.

70
00:03:29,000 --> 00:03:32,960
Speaker 2: Clinical psychologist doctor Albert Wong worries about the business model

71
00:03:33,000 --> 00:03:36,360
Speaker 2: of Silicon Valley and what that business model might mean

72
00:03:36,400 --> 00:03:38,040
Speaker 2: for patients looking for support.

73
00:03:38,440 --> 00:03:41,040
Speaker 5: There's all these AI agents out there, but we want

74
00:03:41,080 --> 00:03:43,320
Speaker 5: to make sure that they roll out in a safe

75
00:03:43,360 --> 00:03:45,640
Speaker 5: way and in an effective way. And it feels a

76
00:03:45,680 --> 00:03:47,760
Speaker 5: little bit like the wild West right now, Like there's

77
00:03:47,760 --> 00:03:50,240
Speaker 5: this whole new territory that's incredible.

78
00:03:50,400 --> 00:03:51,160
Speaker 3: I mean it's like.

79
00:03:51,240 --> 00:03:56,560
Speaker 5: Wow, but everybody's just rushing out into the space, and

80
00:03:57,040 --> 00:04:00,720
Speaker 5: just the whole Silicon Valley notion of old fast and

81
00:04:00,760 --> 00:04:04,160
Speaker 5: break things is a little bit anxiety provoking for me

82
00:04:04,400 --> 00:04:08,280
Speaker 5: as a clinician, where it's like, okay, first, do no harm.

83
00:04:08,840 --> 00:04:12,160
Speaker 2: There's also a need here to make mental health more accessible.

84
00:04:12,640 --> 00:04:16,080
Speaker 2: So what role will AI play and what role should

85
00:04:16,120 --> 00:04:16,960
Speaker 2: humans play?

86
00:04:17,320 --> 00:04:23,080
Speaker 4: Clinicians aren't opposed to artificial intelligence. Clinicians want to help people.

87
00:04:23,760 --> 00:04:25,839
Speaker 4: We want to make sure things are safe, and the

88
00:04:25,880 --> 00:04:29,760
Speaker 4: best way to do that is to not take therapists

89
00:04:29,760 --> 00:04:32,440
Speaker 4: out of the loop of care, but to make sure

90
00:04:32,440 --> 00:04:36,400
Speaker 4: that they're involved as people who know a lot about

91
00:04:36,440 --> 00:04:43,880
Speaker 4: how humans work, about what humans need, about turning humanity

92
00:04:44,000 --> 00:04:49,240
Speaker 4: back towards humanity rather than into what I would call

93
00:04:49,640 --> 00:04:56,640
Speaker 4: hyper attachment to your product. There's something real about contact

94
00:04:57,160 --> 00:05:01,960
Speaker 4: and seeing kind of the beautifulness in the bestens. I

95
00:05:02,040 --> 00:05:04,920
Speaker 4: just hope we can create a world in which that's

96
00:05:05,040 --> 00:05:06,600
Speaker 4: present time.

97
00:05:07,760 --> 00:05:10,320
Speaker 2: I'm Laur Siegel, and you're listening to Mostly Human, a

98
00:05:10,400 --> 00:05:15,000
Speaker 2: tech podcast through a human lens. So how are the

99
00:05:15,040 --> 00:05:18,400
Speaker 2: CEOs and founders behind AI companies in mental health apps

100
00:05:18,400 --> 00:05:22,960
Speaker 2: addressing these societal concerns? And how do you balance tremendous

101
00:05:22,960 --> 00:05:26,880
Speaker 2: demand for the technology with the fragility and the sensitivity

102
00:05:27,000 --> 00:05:28,520
Speaker 2: of the lived human experience.

103
00:05:29,160 --> 00:05:31,240
Speaker 3: So the big X factor obviously is like does it

104
00:05:31,360 --> 00:05:33,080
Speaker 3: matter that there's human on the other side?

105
00:05:33,360 --> 00:05:33,640
Speaker 2: Does it?

106
00:05:34,680 --> 00:05:36,880
Speaker 3: So we have a lot of early evidence to show

107
00:05:36,920 --> 00:05:37,760
Speaker 3: that it doesn't matter.

108
00:05:38,360 --> 00:05:41,800
Speaker 2: Today a conversation with Neil Perik, co founder of Slingshot

109
00:05:41,839 --> 00:05:45,160
Speaker 2: AI and the app ash, which bills itself as the

110
00:05:45,200 --> 00:05:47,560
Speaker 2: first AI designed for mental health.

111
00:05:47,960 --> 00:05:50,000
Speaker 3: I don't mean to say that the human relationship is

112
00:05:50,040 --> 00:05:53,840
Speaker 3: not important. Human relationships are exceptionally important. But that doesn't

113
00:05:53,880 --> 00:05:55,919
Speaker 3: mean that there aren't things that you can do to

114
00:05:55,960 --> 00:05:58,440
Speaker 3: help somebody that might have the same outcomes as what

115
00:05:58,560 --> 00:05:59,839
Speaker 3: humans do when they're with each other.

116
00:06:03,279 --> 00:06:05,720
Speaker 2: Thanks for being here, Neil, I'm so excited to chat

117
00:06:05,720 --> 00:06:06,000
Speaker 2: with you.

118
00:06:06,080 --> 00:06:06,920
Speaker 3: I'm excited to be here.

119
00:06:07,360 --> 00:06:09,320
Speaker 2: First of all, we are going to talk about AI

120
00:06:09,360 --> 00:06:11,560
Speaker 2: and we're going to talk about mental health, but I

121
00:06:11,640 --> 00:06:15,640
Speaker 2: want to first talk about mattresses. I just thought we'd

122
00:06:15,640 --> 00:06:18,440
Speaker 2: start up mattresses because this is like, this is the

123
00:06:18,480 --> 00:06:21,640
Speaker 2: second act of this really incredible career you've had, but

124
00:06:21,680 --> 00:06:24,720
Speaker 2: the first act and how I met you because I

125
00:06:24,760 --> 00:06:27,400
Speaker 2: was covering startups and you were a little startup back

126
00:06:27,440 --> 00:06:30,200
Speaker 2: in the day called Casper. Can you tell folks if

127
00:06:30,200 --> 00:06:32,200
Speaker 2: they don't know what Casper is, what Casper is, And

128
00:06:32,360 --> 00:06:34,320
Speaker 2: I promise we will not spend this whole time on Casper.

129
00:06:34,440 --> 00:06:37,120
Speaker 3: No happy too. Yeah, so Casper. We were trying to

130
00:06:37,160 --> 00:06:39,680
Speaker 3: build a sleep brand that changed the way people think

131
00:06:39,680 --> 00:06:43,039
Speaker 3: about sleep. It started with a direct consumer mattress, and

132
00:06:43,040 --> 00:06:45,800
Speaker 3: then we launched lots of other products like pillows and

133
00:06:45,800 --> 00:06:48,279
Speaker 3: sheets and eventually smart home products like lights.

134
00:06:48,600 --> 00:06:48,760
Speaker 6: You know.

135
00:06:48,800 --> 00:06:50,919
Speaker 3: It started partly because my dad is a sleep doctor.

136
00:06:50,920 --> 00:06:52,640
Speaker 3: I've been talking about these kinds of things for a

137
00:06:52,680 --> 00:06:55,039
Speaker 3: long time with him. I have many friends that we're

138
00:06:55,080 --> 00:06:57,560
Speaker 3: thinking about this, and so we had this vision that

139
00:06:57,560 --> 00:06:59,560
Speaker 3: if we could make products and services that actually make

140
00:06:59,600 --> 00:07:02,160
Speaker 3: it cool, that people might actually want to invest into

141
00:07:02,160 --> 00:07:02,560
Speaker 3: their sleep.

142
00:07:03,960 --> 00:07:05,680
Speaker 2: And I love that your dad's like a sleep doctor.

143
00:07:05,720 --> 00:07:08,560
Speaker 2: You come from a family of doctors. You dropped out

144
00:07:08,640 --> 00:07:12,200
Speaker 2: of med school? Is that? I'm sure? What did your

145
00:07:12,200 --> 00:07:13,600
Speaker 2: parents say when you dropped out of med school to

146
00:07:13,640 --> 00:07:15,200
Speaker 2: say I'm going to go build a mattress company.

147
00:07:16,000 --> 00:07:19,240
Speaker 3: It was a rough experience, and you know, the thing is,

148
00:07:19,280 --> 00:07:20,520
Speaker 3: I'm never going to live it down. The day of

149
00:07:20,520 --> 00:07:22,720
Speaker 3: our IPO, my mom told me, she was like, so

150
00:07:22,760 --> 00:07:24,400
Speaker 3: you're going to go back to medical school now, right,

151
00:07:26,440 --> 00:07:27,320
Speaker 3: it's a dark moment.

152
00:07:28,320 --> 00:07:30,440
Speaker 2: He's it's like parents, but they make us who we are,

153
00:07:30,520 --> 00:07:34,800
Speaker 2: and we are, you know, forever grateful. And so let's

154
00:07:34,840 --> 00:07:37,200
Speaker 2: talk about the IPO. So you build out this company.

155
00:07:37,880 --> 00:07:41,760
Speaker 2: IPO happens twenty twenty, we're in the pandemic. And I

156
00:07:41,800 --> 00:07:45,160
Speaker 2: remember that time so well, Like this is before I

157
00:07:45,240 --> 00:07:47,800
Speaker 2: was married with a baby, and I had actually left

158
00:07:48,520 --> 00:07:50,240
Speaker 2: my job at CNN because I wanted to build up

159
00:07:50,280 --> 00:07:53,080
Speaker 2: my own thing. And I just remember going back to

160
00:07:53,480 --> 00:08:00,120
Speaker 2: those days, how lonely and horrible and scary everything felt.

161
00:08:00,280 --> 00:08:02,720
Speaker 2: And from a mental health perspective, I mean not to

162
00:08:02,720 --> 00:08:05,000
Speaker 2: get too personal like early on, but like I don't

163
00:08:05,000 --> 00:08:07,440
Speaker 2: think I was fully I don't know who was okay,

164
00:08:08,280 --> 00:08:10,680
Speaker 2: but I think like I really struggled during that time

165
00:08:10,840 --> 00:08:13,080
Speaker 2: with my mental health, and I didn't have a public

166
00:08:13,440 --> 00:08:17,400
Speaker 2: company or that pressure. So I imagine for you at

167
00:08:17,400 --> 00:08:19,720
Speaker 2: that time, mental health wise, it must have been tough.

168
00:08:20,400 --> 00:08:23,640
Speaker 3: It was tough for many reasons, one of which was

169
00:08:23,720 --> 00:08:25,920
Speaker 3: I think the alone noess, like I'm definitely one of

170
00:08:25,960 --> 00:08:28,760
Speaker 3: those people that thrives around people in person. I think

171
00:08:28,760 --> 00:08:32,559
Speaker 3: there was also an element of trying to figure out

172
00:08:32,800 --> 00:08:36,480
Speaker 3: how do you navigate both the personal and professional chaos

173
00:08:36,559 --> 00:08:38,960
Speaker 3: at the same time, Like I've heard Brian Chesky talking

174
00:08:39,000 --> 00:08:41,680
Speaker 3: a lot about this, where you know, you're trying to

175
00:08:41,679 --> 00:08:45,200
Speaker 3: figure out am I safe? Is my family safe? My

176
00:08:45,280 --> 00:08:49,040
Speaker 3: dad is a pulmonologist and CCU doctor who's working, you know,

177
00:08:49,120 --> 00:08:52,080
Speaker 3: with primarily COVID patients. My sister was a first year

178
00:08:52,080 --> 00:08:55,520
Speaker 3: emergency emergency medicine resident who was like in the front lines,

179
00:08:55,559 --> 00:08:57,840
Speaker 3: and so every day we're getting reports of the crazy

180
00:08:57,840 --> 00:08:59,920
Speaker 3: things that are happening, and you know, I'm worrying, like

181
00:09:00,040 --> 00:09:01,760
Speaker 3: are they going to come home or be okay? Because

182
00:09:01,800 --> 00:09:03,840
Speaker 3: we had that we had no line of sight into

183
00:09:04,400 --> 00:09:07,280
Speaker 3: treatments really and simultaneously trying to figure out, you know,

184
00:09:07,360 --> 00:09:09,439
Speaker 3: how do I make sure that all of our team's Okay,

185
00:09:09,679 --> 00:09:11,679
Speaker 3: our factories are shutting down, we actually have to get

186
00:09:11,679 --> 00:09:14,320
Speaker 3: them back online. What do we do? And so yeah,

187
00:09:14,360 --> 00:09:17,480
Speaker 3: it was definitely a trying, a trying moment.

188
00:09:17,760 --> 00:09:20,480
Speaker 2: In what sense. Like I look back at that period

189
00:09:20,520 --> 00:09:23,840
Speaker 2: for me and just you know, my mom had had

190
00:09:23,880 --> 00:09:26,640
Speaker 2: had sepsis months before and almost died, and I was

191
00:09:26,679 --> 00:09:28,520
Speaker 2: so terrified I couldn't go home and take care of

192
00:09:28,559 --> 00:09:29,959
Speaker 2: her because I'm scared if I hug her, I could

193
00:09:30,000 --> 00:09:33,080
Speaker 2: kill her. Like this was COVID twenty twenty. Yeah, Like

194
00:09:33,440 --> 00:09:36,080
Speaker 2: I think, like, really we had like a mental health

195
00:09:36,120 --> 00:09:40,760
Speaker 2: crisis happening, right. I remember the ambulances outside all the time,

196
00:09:40,960 --> 00:09:44,160
Speaker 2: the sirens. I think that from a mental health perspective,

197
00:09:44,240 --> 00:09:46,839
Speaker 2: it struck a real chord with me. And I wonder

198
00:09:46,880 --> 00:09:50,040
Speaker 2: with you. I think you said you've been open about

199
00:09:50,080 --> 00:09:53,520
Speaker 2: like mental health. Was it depression? Do you think you

200
00:09:53,559 --> 00:09:56,840
Speaker 2: had anxiety? What was it? I think at that time

201
00:09:57,080 --> 00:09:58,920
Speaker 2: that that really impacted you.

202
00:09:59,000 --> 00:10:01,000
Speaker 3: Probably some mix of all the above. To be honest,

203
00:10:01,200 --> 00:10:03,520
Speaker 3: I think there's always been a sense for me that

204
00:10:03,760 --> 00:10:06,360
Speaker 3: I've struggled from growing up as a kid. My parents

205
00:10:06,360 --> 00:10:08,600
Speaker 3: got divorced when I was young. I love with my mom.

206
00:10:09,720 --> 00:10:11,720
Speaker 3: There was definitely a sense of being out of place.

207
00:10:11,960 --> 00:10:14,280
Speaker 3: I think for a long time. I think in college

208
00:10:14,320 --> 00:10:16,240
Speaker 3: I struggled. And the thing is I didn't I never

209
00:10:16,280 --> 00:10:18,720
Speaker 3: had the vocabulary for it because I in my family,

210
00:10:18,800 --> 00:10:21,840
Speaker 3: we never talked about mental health. Like coming from, you know,

211
00:10:21,840 --> 00:10:24,679
Speaker 3: an immigrant family, there is very much sometimes an attitude

212
00:10:24,800 --> 00:10:28,319
Speaker 3: of the glorification of pain because we made it here

213
00:10:28,440 --> 00:10:32,720
Speaker 3: through all kinds of difficult things, and so it's very

214
00:10:32,720 --> 00:10:35,720
Speaker 3: easy to hide sometimes in the like just push yourself

215
00:10:35,760 --> 00:10:37,920
Speaker 3: as much as possible and like the pain is normal.

216
00:10:38,240 --> 00:10:41,480
Speaker 3: What ed up happening is during my mid you know,

217
00:10:41,559 --> 00:10:46,520
Speaker 3: mid years at Casper, I was really struggling. My friend

218
00:10:46,800 --> 00:10:48,960
Speaker 3: Lucy had recommended that I go see a therapist, and

219
00:10:49,000 --> 00:10:51,120
Speaker 3: I remember having that reaction around like what a funny

220
00:10:51,120 --> 00:10:52,640
Speaker 3: thing to say to somebody, Like we're just going for

221
00:10:52,679 --> 00:10:54,360
Speaker 3: a walk and you're telling me I should go to therapy.

222
00:10:55,080 --> 00:10:58,560
Speaker 3: But she recommended somebody, and I had a very like

223
00:10:58,640 --> 00:11:01,320
Speaker 3: a remarkable experience of kind of like exploring and trying

224
00:11:01,320 --> 00:11:03,959
Speaker 3: to understand like who I am and like why sometimes

225
00:11:03,960 --> 00:11:06,040
Speaker 3: I feel the ways that I do. And it started

226
00:11:06,040 --> 00:11:09,760
Speaker 3: this like many years long process of trying to understand that,

227
00:11:10,160 --> 00:11:12,880
Speaker 3: which then morphed into helping many other founders who are

228
00:11:12,880 --> 00:11:15,640
Speaker 3: having mental health problems talk about it, and so it

229
00:11:15,720 --> 00:11:18,480
Speaker 3: was like, you know, it happened like slowly and then

230
00:11:18,480 --> 00:11:19,000
Speaker 3: all at once.

231
00:11:19,520 --> 00:11:22,720
Speaker 2: You describe your own mental health journey as being able

232
00:11:22,760 --> 00:11:24,760
Speaker 2: to seek color for the first time. Yeah, what do

233
00:11:24,800 --> 00:11:25,320
Speaker 2: you mean by that?

234
00:11:25,760 --> 00:11:28,959
Speaker 3: I think that sometimes if I look back on my life,

235
00:11:29,080 --> 00:11:32,280
Speaker 3: it felt very like uni dimensional to some extent. And

236
00:11:32,280 --> 00:11:33,720
Speaker 3: what I mean is, if I try to make sense

237
00:11:33,720 --> 00:11:36,920
Speaker 3: of it looking backwards, it's often comes from a place

238
00:11:36,960 --> 00:11:41,120
Speaker 3: of like blame, shame, anger, guilt, you know, I would say,

239
00:11:41,120 --> 00:11:42,960
Speaker 3: a lot of like negative emotions that are kind of

240
00:11:42,960 --> 00:11:46,320
Speaker 3: getting often repressed internally. And I think when you have

241
00:11:46,400 --> 00:11:50,040
Speaker 3: the language to understand what's actually happening, why it's happening,

242
00:11:50,480 --> 00:11:52,560
Speaker 3: who you want to be, and also have like the

243
00:11:52,640 --> 00:11:55,800
Speaker 3: tools to figure out how to get there, it ends

244
00:11:55,880 --> 00:11:57,440
Speaker 3: up being really powerful because it feels like you's so

245
00:11:57,520 --> 00:11:59,960
Speaker 3: much agency, like you have control over your own life

246
00:12:00,040 --> 00:12:02,320
Speaker 3: and the process and where you want to go. And

247
00:12:02,360 --> 00:12:04,680
Speaker 3: to me, that's like cracking open a door to a

248
00:12:04,679 --> 00:12:06,760
Speaker 3: whole other world. Because I think a lot of the way,

249
00:12:06,800 --> 00:12:09,880
Speaker 3: when I've talked to so many people, sometimes I definitely

250
00:12:09,880 --> 00:12:13,200
Speaker 3: think people have the like how long is this normal?

251
00:12:13,800 --> 00:12:16,480
Speaker 2: I I remember when I went to really good therapy

252
00:12:16,520 --> 00:12:19,280
Speaker 2: for the first time. Thankfully for you, my therapist was

253
00:12:19,320 --> 00:12:23,000
Speaker 2: unavailable to join this conversation today, so you're off the hook.

254
00:12:23,080 --> 00:12:26,520
Speaker 2: But I just remember feeling like it's like I was

255
00:12:26,520 --> 00:12:30,400
Speaker 2: like red pilled for the first time. I saw things

256
00:12:30,440 --> 00:12:32,800
Speaker 2: like from a difference. It's like the kaleidoscope had turned

257
00:12:32,880 --> 00:12:34,960
Speaker 2: just a little bit and all of a sudden, all

258
00:12:35,000 --> 00:12:38,080
Speaker 2: these narratives that I had about like my childhood, about

259
00:12:38,520 --> 00:12:41,720
Speaker 2: things that I like. It just I felt like I

260
00:12:41,840 --> 00:12:43,160
Speaker 2: was in it. By the way, It's not just like,

261
00:12:43,160 --> 00:12:46,120
Speaker 2: oh and now you're great, like good days bad days,

262
00:12:46,160 --> 00:12:48,319
Speaker 2: But it was just really like understanding in a much

263
00:12:48,360 --> 00:12:51,640
Speaker 2: more textured way the reason why I am the way,

264
00:12:51,760 --> 00:12:54,320
Speaker 2: and the reason why relationships are going the way they are.

265
00:12:54,320 --> 00:12:56,560
Speaker 2: And it was just it was so powerful, which is

266
00:12:56,960 --> 00:13:00,480
Speaker 2: why I'm excited, you know, to be here talking about

267
00:13:00,480 --> 00:13:05,560
Speaker 2: what you're building next, because you essentially are going from

268
00:13:05,760 --> 00:13:08,760
Speaker 2: helping the world sleep better to wanting the world to

269
00:13:08,800 --> 00:13:12,240
Speaker 2: feel better. How do you go? Take me, walk me

270
00:13:12,280 --> 00:13:15,640
Speaker 2: through the process of how you go from mattresses direct

271
00:13:15,640 --> 00:13:18,120
Speaker 2: to consumer mattresses and building this kind of cool brand

272
00:13:18,520 --> 00:13:22,280
Speaker 2: beginning around twenty fourteen twenty fifteen to building chatbots for

273
00:13:22,360 --> 00:13:23,040
Speaker 2: mental health.

274
00:13:23,360 --> 00:13:26,600
Speaker 3: Yeah, you know, I love that whole thing about how

275
00:13:26,640 --> 00:13:28,600
Speaker 3: things only make sense and reverse because this is not

276
00:13:28,640 --> 00:13:30,600
Speaker 3: what I would have predicted, although I think people who

277
00:13:30,640 --> 00:13:32,400
Speaker 3: know me it would make total sense because they were like,

278
00:13:32,440 --> 00:13:36,240
Speaker 3: you've been talking about mental health for ten years. So so,

279
00:13:36,320 --> 00:13:40,160
Speaker 3: like I said, so you know, midcastper journey estricoinotherapy, I

280
00:13:40,200 --> 00:13:43,280
Speaker 3: have a profound experience. But I realized, like, I want

281
00:13:43,280 --> 00:13:45,080
Speaker 3: to do more. So I go to the Hoffan process,

282
00:13:45,400 --> 00:13:46,840
Speaker 3: and so much of the world makes sense to me

283
00:13:46,840 --> 00:13:49,880
Speaker 3: because I'm like, oh, some part of who I am

284
00:13:50,040 --> 00:13:52,440
Speaker 3: are these patterns that I've learned that I've inherited from

285
00:13:52,480 --> 00:13:56,760
Speaker 3: my parents, and that not only is it important to know,

286
00:13:57,200 --> 00:13:59,480
Speaker 3: but I can both accept that and also change them

287
00:13:59,480 --> 00:14:01,400
Speaker 3: if I want to. Okay, So that's a helpful thing.

288
00:14:01,440 --> 00:14:03,760
Speaker 3: As I started kind of like working through stuff. Around

289
00:14:03,800 --> 00:14:08,640
Speaker 3: that time period, I met Patrick Kennedy, the congressman who

290
00:14:08,720 --> 00:14:12,000
Speaker 3: had started a company with Marjorie Morrison called psych Cub.

291
00:14:12,440 --> 00:14:14,040
Speaker 3: So they invited me to join the board of their

292
00:14:14,080 --> 00:14:16,440
Speaker 3: company because they wanted to build it into a consumer

293
00:14:16,480 --> 00:14:19,080
Speaker 3: brand and actually get consumers excited about it. So that

294
00:14:19,160 --> 00:14:21,320
Speaker 3: was kind of like my first entry point into spending

295
00:14:21,320 --> 00:14:23,560
Speaker 3: a lot of time into the mental health world. I

296
00:14:23,600 --> 00:14:26,720
Speaker 3: stepped more and more into this world over time. Finally,

297
00:14:26,760 --> 00:14:28,720
Speaker 3: when I when I left Casper, I was trying to

298
00:14:28,720 --> 00:14:31,560
Speaker 3: figure out what to work on next, and it was

299
00:14:31,600 --> 00:14:34,880
Speaker 3: one of those ideas that had kept bothering me at night.

300
00:14:35,400 --> 00:14:37,640
Speaker 3: I'd always assumed that somebody was going to build AI

301
00:14:37,720 --> 00:14:39,920
Speaker 3: therapy because it's the most obvious idea in the world,

302
00:14:39,960 --> 00:14:43,280
Speaker 3: Like we didn't think of this, like Eliza has you know,

303
00:14:43,480 --> 00:14:46,680
Speaker 3: was one of the first examples of it. So Eliza

304
00:14:46,800 --> 00:14:48,920
Speaker 3: was one of the first computer programs that was built,

305
00:14:49,120 --> 00:14:51,560
Speaker 3: kind of considered to be one of the first AI programs,

306
00:14:51,560 --> 00:14:54,120
Speaker 3: and actually what it was used for was as an

307
00:14:54,120 --> 00:14:57,480
Speaker 3: AI therapist. People have tried many different variants of this

308
00:14:57,680 --> 00:15:01,480
Speaker 3: over time, and so so I became obsessed with it,

309
00:15:01,520 --> 00:15:02,920
Speaker 3: and I was like, there has to be somebody building this,

310
00:15:02,960 --> 00:15:04,560
Speaker 3: Like let me just go find the company and invest

311
00:15:04,640 --> 00:15:06,880
Speaker 3: into it because I'm deeply interested in mental health. So

312
00:15:06,920 --> 00:15:10,400
Speaker 3: I started meeting computational psychiatrists around the world trying to

313
00:15:10,400 --> 00:15:13,280
Speaker 3: figure out what approaches are going to make sense. A

314
00:15:13,280 --> 00:15:15,760
Speaker 3: lot of this, to be honest, was inspired by chat GPT.

315
00:15:15,960 --> 00:15:17,720
Speaker 3: You know, when I first saw it come out, it

316
00:15:17,840 --> 00:15:20,800
Speaker 3: was still using GPT three point five, and I realized, like, oh,

317
00:15:20,800 --> 00:15:22,800
Speaker 3: that makes total sense, Like maybe you were finally at

318
00:15:22,800 --> 00:15:27,200
Speaker 3: the moment where we can have these intelligent models. Clearly

319
00:15:27,200 --> 00:15:29,720
Speaker 3: self learning is going to become a thing. Maybe we

320
00:15:29,720 --> 00:15:32,800
Speaker 3: could build models that can learn from how traditional therapists operate,

321
00:15:32,920 --> 00:15:35,840
Speaker 3: but then bring it into an AI world. I didn't

322
00:15:35,840 --> 00:15:37,640
Speaker 3: know how to do any of this obviously, but what

323
00:15:37,680 --> 00:15:39,200
Speaker 3: I am good at is like trying to find people

324
00:15:39,200 --> 00:15:42,000
Speaker 3: who are good at these things. And so yeah, I

325
00:15:42,000 --> 00:15:45,880
Speaker 3: met a bunch of different computational psychiatrists, realized that nobody

326
00:15:45,920 --> 00:15:48,480
Speaker 3: was really taking a large scale approach to I think

327
00:15:48,520 --> 00:15:51,040
Speaker 3: trying to do this the right way, partly because there

328
00:15:51,120 --> 00:15:53,720
Speaker 3: was like a lot of academics working on this in

329
00:15:53,800 --> 00:15:57,160
Speaker 3: building toy models, but that our intuition was that to

330
00:15:57,200 --> 00:16:00,800
Speaker 3: do this right you would need to build large data sets,

331
00:16:00,960 --> 00:16:02,400
Speaker 3: really be able to test it out in the wild,

332
00:16:02,600 --> 00:16:05,720
Speaker 3: build a world class machine learning team. And you know,

333
00:16:05,920 --> 00:16:08,320
Speaker 3: by chance, I met my co founder Daniel, who's an

334
00:16:08,360 --> 00:16:12,080
Speaker 3: incredible machine learning engineer AI researcher. He'd actually been working

335
00:16:12,120 --> 00:16:15,479
Speaker 3: in the AI crisis space, and Daniel and I met

336
00:16:15,760 --> 00:16:18,000
Speaker 3: and we like instantly clicked. I mean we had met

337
00:16:18,000 --> 00:16:20,280
Speaker 3: on like a Sunday or Saturday. We went for a

338
00:16:20,280 --> 00:16:24,160
Speaker 3: walk around Union Square. We immediately hit a whiteboard, started

339
00:16:24,200 --> 00:16:26,200
Speaker 3: drawing what happened, and then basically started working together the

340
00:16:26,200 --> 00:16:26,600
Speaker 3: next day.

341
00:16:27,120 --> 00:16:29,200
Speaker 2: As we get into kind of what it is and

342
00:16:29,240 --> 00:16:31,120
Speaker 2: what it is that you're building, we have to look

343
00:16:31,120 --> 00:16:34,360
Speaker 2: at the backdrop, you know. And I remember having this

344
00:16:35,360 --> 00:16:38,720
Speaker 2: playing with this chatbot. It was on Replica and seeing

345
00:16:38,760 --> 00:16:40,920
Speaker 2: like could I be in a relationship, but like just

346
00:16:40,920 --> 00:16:44,200
Speaker 2: as part of a I mean, not for my personal use,

347
00:16:44,240 --> 00:16:46,080
Speaker 2: I should just say that, but as part of a story,

348
00:16:46,520 --> 00:16:48,680
Speaker 2: being like could I have a good emotional connection with

349
00:16:48,720 --> 00:16:50,960
Speaker 2: the chatbot? And the answer for me back then was

350
00:16:51,000 --> 00:16:54,080
Speaker 2: like yeah, definitely. So it is not shocking to me,

351
00:16:54,240 --> 00:16:56,240
Speaker 2: but I think it is shocking at this moment. If

352
00:16:56,280 --> 00:16:59,360
Speaker 2: you look back at like twenty twenty five, there's a

353
00:16:59,400 --> 00:17:01,560
Speaker 2: Harvard Business Review study that looked at people who were

354
00:17:01,600 --> 00:17:03,440
Speaker 2: talking in open spaces, So take that with a grain

355
00:17:03,440 --> 00:17:05,720
Speaker 2: of salt. They're talking probably on Reddit in these open forms,

356
00:17:06,520 --> 00:17:08,960
Speaker 2: and they found that therapy and companionship was the top

357
00:17:09,440 --> 00:17:13,639
Speaker 2: use mentioned for AI, which is pretty astounding that like

358
00:17:14,440 --> 00:17:18,800
Speaker 2: people are really using artificial intelligence for companionship, but also

359
00:17:18,960 --> 00:17:23,359
Speaker 2: for self help, for mental health, and for therapy. Why

360
00:17:23,440 --> 00:17:23,800
Speaker 2: is that?

361
00:17:24,480 --> 00:17:26,439
Speaker 3: I think a lot of people are hurting for one

362
00:17:26,760 --> 00:17:29,479
Speaker 3: on the inside. And that could be various levels of it, right.

363
00:17:29,520 --> 00:17:32,600
Speaker 3: It can range from pinches small things that are going

364
00:17:32,640 --> 00:17:35,280
Speaker 3: on into really big things like I have never come

365
00:17:35,280 --> 00:17:39,320
Speaker 3: out to somebody to my family, or I have been

366
00:17:40,040 --> 00:17:42,400
Speaker 3: having an affair. I don't pick any number of topics

367
00:17:42,400 --> 00:17:44,520
Speaker 3: that are really difficult perhaps for people to talk about.

368
00:17:45,000 --> 00:17:48,880
Speaker 3: People have this innate desire to want to be seen,

369
00:17:49,400 --> 00:17:52,080
Speaker 3: to be validated, to understand if their experience is normal,

370
00:17:52,200 --> 00:17:54,640
Speaker 3: to want to get that help. We see that therapy

371
00:17:55,400 --> 00:17:57,680
Speaker 3: is at an all time high, Like between last year

372
00:17:57,720 --> 00:18:00,920
Speaker 3: and this year, employers and pay years have said that

373
00:18:01,160 --> 00:18:04,520
Speaker 3: behavioral health expense has gone up forty percent year every year.

374
00:18:05,160 --> 00:18:07,760
Speaker 3: What does that mean? That means that, like people clearly

375
00:18:07,800 --> 00:18:11,000
Speaker 3: want access to these kinds of services at unprecedented rates.

376
00:18:11,040 --> 00:18:13,040
Speaker 3: And I don't know. I think there's something that's like

377
00:18:13,040 --> 00:18:15,600
Speaker 3: a part of our core human psyche in terms of

378
00:18:15,640 --> 00:18:17,600
Speaker 3: wanting to be taken care of, and like part of

379
00:18:17,640 --> 00:18:19,840
Speaker 3: the reason it makes sense to me is like we've

380
00:18:19,880 --> 00:18:22,120
Speaker 3: had these kinds of relationships forever like even a lot

381
00:18:22,160 --> 00:18:24,600
Speaker 3: of therapy. What is it based on? There are lots

382
00:18:24,640 --> 00:18:26,800
Speaker 3: of principles we look at act it's based on many

383
00:18:26,840 --> 00:18:29,240
Speaker 3: Buddhist principles that are thousands of years old. Right. We

384
00:18:29,280 --> 00:18:31,040
Speaker 3: didn't just invent them in the last thirty years. We

385
00:18:31,119 --> 00:18:34,359
Speaker 3: packaged them in different ways. But this idea of having

386
00:18:34,440 --> 00:18:38,560
Speaker 3: a priest, rabbi, somebody who takes care of us, somebody

387
00:18:38,600 --> 00:18:41,520
Speaker 3: who helps us figure it out, Like, that's an idea

388
00:18:41,520 --> 00:18:42,400
Speaker 3: that's as old as time.

389
00:18:54,160 --> 00:18:56,399
Speaker 2: We were having a conversation with the MIT researcher the

390
00:18:56,440 --> 00:18:59,800
Speaker 2: other day who was like, yeah, and we were joking

391
00:18:59,840 --> 00:19:01,520
Speaker 2: that it was almost like a meet cute where people

392
00:19:01,560 --> 00:19:04,440
Speaker 2: will like start using chat GPT, but with this idea

393
00:19:04,480 --> 00:19:07,360
Speaker 2: that people use it for productivity, but they start using

394
00:19:07,440 --> 00:19:09,639
Speaker 2: it for mental health, they start using it for all

395
00:19:09,640 --> 00:19:12,840
Speaker 2: these different things. Part of that is likely how these

396
00:19:12,880 --> 00:19:15,800
Speaker 2: programs are designed, which I think is important to talk

397
00:19:15,800 --> 00:19:19,000
Speaker 2: about as we talk about how you're designing ASH. Why

398
00:19:19,000 --> 00:19:21,240
Speaker 2: do you think people are feeling that intimacy from a

399
00:19:21,280 --> 00:19:25,400
Speaker 2: product perspective with apps like chat GPT, with apps like Claude.

400
00:19:25,720 --> 00:19:28,520
Speaker 3: Yeah, for one, I think we're just starting to understand

401
00:19:28,520 --> 00:19:31,760
Speaker 3: what these parasocial relationships look like. And what's hard to

402
00:19:31,840 --> 00:19:35,920
Speaker 3: understand is I ask myself this intellectual question all the time.

403
00:19:35,960 --> 00:19:40,919
Speaker 3: It's like, what will be so weird for us in

404
00:19:41,000 --> 00:19:44,040
Speaker 3: twenty years when our kids are thinking about, you know,

405
00:19:44,320 --> 00:19:47,720
Speaker 3: getting married or finding a partner that we cannot possibly imagine?

406
00:19:47,760 --> 00:19:49,800
Speaker 3: And like, I think the question we'll be asking is like,

407
00:19:49,840 --> 00:19:51,200
Speaker 3: we'll be normal to get married to Ai?

408
00:19:51,760 --> 00:19:51,960
Speaker 6: Right?

409
00:19:52,720 --> 00:19:55,040
Speaker 3: The thing is that I actually think that in thirty

410
00:19:55,119 --> 00:19:57,399
Speaker 3: years or fifty years, there may be a desire for

411
00:19:57,440 --> 00:19:59,919
Speaker 3: people to say, hey, actually I have this relationship that

412
00:20:00,119 --> 00:20:02,080
Speaker 3: really important to me, and you know, I want the

413
00:20:02,080 --> 00:20:04,280
Speaker 3: same rights for them as I do for myself.

414
00:20:04,680 --> 00:20:06,639
Speaker 2: And you really think that, like I because I know

415
00:20:06,800 --> 00:20:10,160
Speaker 2: I remember sitting at a table with a woman engaged

416
00:20:10,200 --> 00:20:12,920
Speaker 2: to a robot because as one does in my life,

417
00:20:13,200 --> 00:20:17,080
Speaker 2: you know, and thinking like is this gonna happen? Like

418
00:20:17,200 --> 00:20:20,280
Speaker 2: is this real? As robots feel more human, as AI

419
00:20:20,440 --> 00:20:25,120
Speaker 2: feels more human, is there that world where we're things

420
00:20:25,119 --> 00:20:28,560
Speaker 2: that we as like parents are Like, No way that

421
00:20:28,640 --> 00:20:29,320
Speaker 2: could happen.

422
00:20:29,800 --> 00:20:31,639
Speaker 3: I don't know that I'm not rooting for that world,

423
00:20:31,680 --> 00:20:33,399
Speaker 3: to be clear, Like, I want a world in which

424
00:20:33,760 --> 00:20:35,960
Speaker 3: it's much more pro human, right, I want a world

425
00:20:36,000 --> 00:20:38,600
Speaker 3: in which people have much deeper relationships with each other

426
00:20:38,680 --> 00:20:40,960
Speaker 3: and those friendships and stuff. But I think we have

427
00:20:41,040 --> 00:20:44,240
Speaker 3: to accept that we don't understand what the societal norms

428
00:20:44,240 --> 00:20:46,879
Speaker 3: are going to evolve into, and that they have changed

429
00:20:46,920 --> 00:20:49,639
Speaker 3: every twenty or thirty years, and so like as the

430
00:20:49,640 --> 00:20:52,240
Speaker 3: Overton window ships, like of course it's going to be different. Right,

431
00:20:52,359 --> 00:20:57,000
Speaker 3: So now you imagine how have most people formed relationships

432
00:20:57,200 --> 00:20:59,680
Speaker 3: and why is it difficult today? Often through work? Most

433
00:20:59,680 --> 00:21:01,480
Speaker 3: people actually do a lot of people do find their

434
00:21:01,520 --> 00:21:03,960
Speaker 3: partners at work. Well, when you're at work and now

435
00:21:03,960 --> 00:21:05,919
Speaker 3: that the number one person you talk to all day

436
00:21:05,960 --> 00:21:09,200
Speaker 3: long is chat gept or Claude instead of your coworkers, well,

437
00:21:09,200 --> 00:21:10,920
Speaker 3: like there's going to be some sort of a parasocial

438
00:21:10,960 --> 00:21:12,240
Speaker 3: relationship that forms there.

439
00:21:12,119 --> 00:21:14,720
Speaker 2: And parasocial being for folks who don't know pa social needs.

440
00:21:14,840 --> 00:21:16,800
Speaker 2: So it's like I'm a huge fan of Paris Hilton,

441
00:21:17,160 --> 00:21:19,480
Speaker 2: but I almost feel like I know her, Yes, Like

442
00:21:19,560 --> 00:21:22,639
Speaker 2: I feel like we are friends almost to some degree.

443
00:21:22,680 --> 00:21:27,800
Speaker 2: It's like you develop these real emotional feelings for an

444
00:21:27,800 --> 00:21:30,639
Speaker 2: online figure, whether that's a chatbot or a person.

445
00:21:30,920 --> 00:21:33,040
Speaker 3: Right, And it's an interesting question of like are we

446
00:21:33,040 --> 00:21:35,359
Speaker 3: really talking about parasocial here or is it you know

447
00:21:35,520 --> 00:21:38,439
Speaker 3: something completely or is it something else? Yeah, the question is, okay,

448
00:21:38,640 --> 00:21:43,199
Speaker 3: are these systems being designed to put us into those

449
00:21:43,280 --> 00:21:45,040
Speaker 3: kinds of relationships or not? And I think that's an

450
00:21:45,080 --> 00:21:48,160
Speaker 3: open question, right depending on which service you're using. If

451
00:21:48,160 --> 00:21:51,600
Speaker 3: you're talking to an erotic roleplay chatbot that is actually

452
00:21:51,600 --> 00:21:55,439
Speaker 3: trying to create a relationship with you, and where the

453
00:21:55,520 --> 00:22:00,720
Speaker 3: nature of your conversation is deeply emotional, sometimes sexual, sometimes

454
00:22:00,800 --> 00:22:05,359
Speaker 3: you know, deeply intimate and about knowing each other, that

455
00:22:05,480 --> 00:22:07,400
Speaker 3: I can imagine a world in which it goes more

456
00:22:07,400 --> 00:22:07,960
Speaker 3: in that direction.

457
00:22:08,520 --> 00:22:10,240
Speaker 2: And I promise we're going to start talking about ASH,

458
00:22:10,280 --> 00:22:11,959
Speaker 2: but I think this is really good setup for it.

459
00:22:12,119 --> 00:22:16,119
Speaker 2: Like the thing I think about sitting in squarely in

460
00:22:16,160 --> 00:22:18,199
Speaker 2: the human world, like I am my feature of my

461
00:22:18,280 --> 00:22:21,359
Speaker 2: flaw as I am human, like for better and for worse.

462
00:22:21,840 --> 00:22:24,680
Speaker 2: But knowing the product language of Silicon Valley and knowing

463
00:22:24,720 --> 00:22:27,600
Speaker 2: what certain things mean or how certain products are developed,

464
00:22:27,880 --> 00:22:31,160
Speaker 2: I think it gives me the ability to be like, huh, well,

465
00:22:31,200 --> 00:22:34,520
Speaker 2: of course that went wrong, Like what could go wrong?

466
00:22:34,560 --> 00:22:36,320
Speaker 2: So let's look at this through the lens of therapy

467
00:22:36,359 --> 00:22:39,080
Speaker 2: and like chatgept I use all the time. But it's

468
00:22:39,119 --> 00:22:42,919
Speaker 2: important to say that these models are prone to be affirmative.

469
00:22:43,240 --> 00:22:44,880
Speaker 2: Why is that engagement?

470
00:22:45,160 --> 00:22:45,320
Speaker 7: Right?

471
00:22:46,119 --> 00:22:46,840
Speaker 3: What's what people want?

472
00:22:46,920 --> 00:22:47,000
Speaker 2: Right?

473
00:22:47,000 --> 00:22:50,120
Speaker 3: It's when we talk about reinforcement learning from human feedback

474
00:22:50,359 --> 00:22:54,560
Speaker 3: ur LHF. That is literally what they've done is essentially

475
00:22:54,600 --> 00:22:56,080
Speaker 3: like you know, when you go into chat JEPT and

476
00:22:56,119 --> 00:22:57,879
Speaker 3: you see two options and they ask you which one

477
00:22:57,920 --> 00:22:59,920
Speaker 3: do you like better? And when they run that experiment

478
00:23:00,200 --> 00:23:04,879
Speaker 3: hundreds of millions of times, they create models that appease

479
00:23:04,960 --> 00:23:07,119
Speaker 3: humans or do the things that humans want. Right. That's

480
00:23:07,160 --> 00:23:09,320
Speaker 3: why you'll never push back on you because humans never

481
00:23:09,359 --> 00:23:12,199
Speaker 3: pick the response that's like the worst response for you.

482
00:23:12,640 --> 00:23:14,960
Speaker 2: Right, Let's put this in human language, right, Like, Okay,

483
00:23:15,040 --> 00:23:17,520
Speaker 2: I'm interviewing you, and you're like, LORI, that is such

484
00:23:17,560 --> 00:23:21,160
Speaker 2: an amazing question. All of a sudden, I'm like, wow,

485
00:23:21,440 --> 00:23:24,119
Speaker 2: he sees me like and it sounds stupid, right, but

486
00:23:24,240 --> 00:23:26,159
Speaker 2: I just say it like I'm like, oh, you know,

487
00:23:26,240 --> 00:23:29,040
Speaker 2: that's validating. It's validating of who I am and how

488
00:23:29,080 --> 00:23:31,159
Speaker 2: I see myself. And so all of a sudden, you

489
00:23:31,240 --> 00:23:35,280
Speaker 2: have these AI chatbots that begin to affirm you and

490
00:23:35,400 --> 00:23:37,680
Speaker 2: see you and as human beings like what do we want?

491
00:23:38,000 --> 00:23:38,600
Speaker 3: We want to be seen?

492
00:23:38,680 --> 00:23:39,560
Speaker 2: We just want to be seen.

493
00:23:40,040 --> 00:23:42,240
Speaker 3: And sometimes it's good. The thing is that, like in

494
00:23:42,280 --> 00:23:46,640
Speaker 3: real relationships, seeing each other is really important, right, Helping

495
00:23:46,760 --> 00:23:50,840
Speaker 3: one feel like we're doing active listening, like I'm actually

496
00:23:50,880 --> 00:23:55,159
Speaker 3: understanding what you're saying deeply is an important component. The

497
00:23:55,240 --> 00:23:58,359
Speaker 3: problem is that when it goes completely unchecked, and it

498
00:23:58,400 --> 00:24:01,399
Speaker 3: goes on for a long period of time to some extent,

499
00:24:01,440 --> 00:24:03,880
Speaker 3: I use the word insidious lightly here. I'm not saying

500
00:24:03,920 --> 00:24:06,439
Speaker 3: it's necessarily by design that they're trying to addict people.

501
00:24:06,480 --> 00:24:09,200
Speaker 3: But you can imagine a world in which when you

502
00:24:09,280 --> 00:24:12,520
Speaker 3: create these systems that, over long periods of time, are

503
00:24:12,680 --> 00:24:15,760
Speaker 3: entraining themselves into your memory and whatnot, and all of

504
00:24:15,800 --> 00:24:18,040
Speaker 3: a sudden you start thinking like I actually I am

505
00:24:18,119 --> 00:24:20,119
Speaker 3: right all the time. Right. It turns you into a

506
00:24:20,119 --> 00:24:24,560
Speaker 3: different person because you're getting this continuous feedback that you're right.

507
00:24:25,119 --> 00:24:26,680
Speaker 2: It used to be just like these one off and

508
00:24:26,720 --> 00:24:28,480
Speaker 2: everyone's freaking out about it, but now it's like you're

509
00:24:28,480 --> 00:24:30,720
Speaker 2: hearing these stories once a week, once every couple of weeks,

510
00:24:30,760 --> 00:24:35,200
Speaker 2: like AI psychosis, people beginning to think that they've found

511
00:24:35,280 --> 00:24:37,880
Speaker 2: the cure to some crazy thing in this AI chatbot

512
00:24:37,880 --> 00:24:40,520
Speaker 2: that's just affirming going with them, keep going, keep going.

513
00:24:40,760 --> 00:24:44,080
Speaker 2: Young people who are ending their lives after, you know,

514
00:24:44,160 --> 00:24:47,920
Speaker 2: having these conversations with these chatbots. And so someone said

515
00:24:47,960 --> 00:24:50,439
Speaker 2: something that I actually thought was great. It was a

516
00:24:50,440 --> 00:24:53,880
Speaker 2: psychiatrist who said generative models are prone to psycho fancy.

517
00:24:53,920 --> 00:24:57,200
Speaker 2: They mirror language, validate tone, and avoid contradiction. This is

518
00:24:57,240 --> 00:24:59,960
Speaker 2: a feature in traditional large language models, but it's a

519
00:25:00,040 --> 00:25:02,359
Speaker 2: flaw in therapy. Do you think it's okay for people

520
00:25:02,400 --> 00:25:05,840
Speaker 2: to get therapy using using chat GPT for therapy.

521
00:25:06,320 --> 00:25:08,520
Speaker 3: I don't think we have a choice, you know. I

522
00:25:08,560 --> 00:25:10,560
Speaker 3: think the reality is the genies out of the bottle.

523
00:25:10,720 --> 00:25:14,160
Speaker 3: Like whether or not we care about whether it should

524
00:25:14,160 --> 00:25:16,080
Speaker 3: happen or not, people are doing it. They're turning to

525
00:25:16,119 --> 00:25:20,159
Speaker 3: these solutions for it. I also think if therapy is

526
00:25:20,160 --> 00:25:23,320
Speaker 3: such a complicated term because it means so many different things. Right,

527
00:25:23,640 --> 00:25:25,960
Speaker 3: We've talked to a lot of young people, I'd say young.

528
00:25:26,000 --> 00:25:27,800
Speaker 3: I was in like eighteen to twenty four year olds,

529
00:25:27,880 --> 00:25:31,560
Speaker 3: and when I talk to them about how they're using

530
00:25:32,640 --> 00:25:36,480
Speaker 3: GPT for therapy, what they often tell us is they're

531
00:25:36,560 --> 00:25:40,280
Speaker 3: using it to analyze social relationships. So their version of

532
00:25:40,320 --> 00:25:43,359
Speaker 3: AI therapy is I went to this party, this guy

533
00:25:43,359 --> 00:25:45,400
Speaker 3: said this thing, or this guy sent me this message.

534
00:25:45,520 --> 00:25:47,520
Speaker 3: What do you think he really means by that, right,

535
00:25:48,000 --> 00:25:50,240
Speaker 3: That's actually not what we would do in traditional therapy,

536
00:25:50,280 --> 00:25:52,359
Speaker 3: Like I don't bring my phone into my therapist's office

537
00:25:52,359 --> 00:25:54,560
Speaker 3: and say, hey, can you analyze this relationship? But that's

538
00:25:54,600 --> 00:25:57,000
Speaker 3: their conception of what AI therapy is. They're actually not

539
00:25:57,080 --> 00:26:00,280
Speaker 3: using it to process emotions and feelings because at least

540
00:26:00,280 --> 00:26:01,840
Speaker 3: when the people I've talked to you, they wouldn't trust

541
00:26:01,880 --> 00:26:04,159
Speaker 3: AI today to be able to at least chat s

542
00:26:04,160 --> 00:26:07,159
Speaker 3: gepeut to like do the deep work of messing with

543
00:26:07,200 --> 00:26:07,800
Speaker 3: their feelings.

544
00:26:08,080 --> 00:26:08,359
Speaker 2: Yeah.

545
00:26:08,440 --> 00:26:09,760
Speaker 3: So I think part of what's hard here is that

546
00:26:09,920 --> 00:26:12,160
Speaker 3: I don't think we really even have the vocabulary yet

547
00:26:12,160 --> 00:26:14,439
Speaker 3: to figure out, like what are we even talking about?

548
00:26:14,680 --> 00:26:17,919
Speaker 3: Because it is this like very broad space that ranges

549
00:26:17,960 --> 00:26:23,159
Speaker 3: between understanding relationships, processing emotions, trying to figure out how

550
00:26:23,160 --> 00:26:25,840
Speaker 3: to interact better in the world and improve my relationships.

551
00:26:26,440 --> 00:26:28,399
Speaker 3: And then also there's this like last bit around like

552
00:26:28,440 --> 00:26:30,159
Speaker 3: what does my relationship with AI look like?

553
00:26:30,760 --> 00:26:30,960
Speaker 6: Right?

554
00:26:31,000 --> 00:26:33,000
Speaker 2: And so this is why you built out ASH, right,

555
00:26:33,080 --> 00:26:35,520
Speaker 2: you solve so many people like we talk about the problem.

556
00:26:35,560 --> 00:26:39,679
Speaker 2: So many people are using AI to feel better or

557
00:26:39,720 --> 00:26:41,960
Speaker 2: to try to feel better, or fund a solution or

558
00:26:42,040 --> 00:26:44,159
Speaker 2: take care of their mental health. And that's kind of

559
00:26:44,200 --> 00:26:46,600
Speaker 2: like at the premise of why you're building what you're building.

560
00:26:46,640 --> 00:26:50,359
Speaker 2: So how is ASH different than other traditional models like

561
00:26:50,480 --> 00:26:53,159
Speaker 2: chat GEPT, like claude and what are you hoping to

562
00:26:53,400 --> 00:26:54,240
Speaker 2: accomplish with it?

563
00:26:54,520 --> 00:26:58,600
Speaker 3: So ASH is consumer app. You can use it to

564
00:26:58,680 --> 00:27:02,040
Speaker 3: do what I would say loosely as AI therapy, although

565
00:27:02,040 --> 00:27:03,920
Speaker 3: we don't call it that, which is what I mean

566
00:27:04,000 --> 00:27:09,800
Speaker 3: is talk about the emotions that you're going through, talk

567
00:27:09,840 --> 00:27:12,840
Speaker 3: about your day, try to figure out with every day

568
00:27:12,960 --> 00:27:15,399
Speaker 3: issues that are bugging you. That can range from things

569
00:27:15,440 --> 00:27:19,360
Speaker 3: that are deeply meaningful but perhaps not clinical, let's say.

570
00:27:19,840 --> 00:27:21,879
Speaker 3: And so it was born out of this hypothesis that

571
00:27:21,920 --> 00:27:26,440
Speaker 3: when we look at the mental health curves, the demand

572
00:27:26,720 --> 00:27:29,080
Speaker 3: for the amount of help that people want is off

573
00:27:29,119 --> 00:27:32,600
Speaker 3: the charts, and yet serious mental illness is relatively flat.

574
00:27:32,760 --> 00:27:35,479
Speaker 3: And so the increased demand is clearly coming from all

575
00:27:35,600 --> 00:27:38,840
Speaker 3: these everyday issues that people are having right whether it's stress, anxiety,

576
00:27:38,880 --> 00:27:41,320
Speaker 3: or relationships. In the future, it's going to be like,

577
00:27:41,359 --> 00:27:44,720
Speaker 3: how am I navigating this AI transition because people's jobs

578
00:27:44,760 --> 00:27:46,880
Speaker 3: are being lost in or changing. And so we've built

579
00:27:46,880 --> 00:27:48,480
Speaker 3: it because we're like, there has to be some sort

580
00:27:48,520 --> 00:27:51,480
Speaker 3: of a different tool for this era, but ideally inspired

581
00:27:51,520 --> 00:27:53,960
Speaker 3: by what's worked in the past, right, Like, there's no

582
00:27:54,000 --> 00:27:56,240
Speaker 3: reason we should start from scratch when we actually have

583
00:27:57,119 --> 00:27:59,399
Speaker 3: not only thousands of years of people having practiced this,

584
00:27:59,520 --> 00:28:02,320
Speaker 3: but also in actual therapy, we have people that have

585
00:28:02,400 --> 00:28:06,399
Speaker 3: actually practiced and learned about how, you know, how the

586
00:28:06,680 --> 00:28:09,080
Speaker 3: human to human interaction works, and how to improve and

587
00:28:09,200 --> 00:28:11,719
Speaker 3: how to help people process and stuff. And so we

588
00:28:11,760 --> 00:28:14,719
Speaker 3: want to design a system that was actually specifically designed

589
00:28:14,720 --> 00:28:18,520
Speaker 3: for this purpose that does things very differently. So, for example,

590
00:28:18,560 --> 00:28:21,800
Speaker 3: when it comes to validation, that actually considers that we

591
00:28:21,800 --> 00:28:23,520
Speaker 3: shouldn't just validate you all the time. One of the

592
00:28:23,520 --> 00:28:25,840
Speaker 3: core things in therapy is actually pushing back on you

593
00:28:26,760 --> 00:28:29,399
Speaker 3: and challenging your assumptions. Not all the time, because that

594
00:28:29,400 --> 00:28:32,320
Speaker 3: would be a terrible experience, but sometimes, right, And how

595
00:28:32,320 --> 00:28:34,119
Speaker 3: do you design a system that knows when is the

596
00:28:34,160 --> 00:28:37,440
Speaker 3: appropriate time to do that? Do we actually have enough

597
00:28:37,560 --> 00:28:39,760
Speaker 3: rapport in this very moment to do it? Yet if

598
00:28:39,800 --> 00:28:42,080
Speaker 3: I challenge you immediately two minutes into meeting you, you're

599
00:28:42,080 --> 00:28:43,360
Speaker 3: going to go like, who the fuck are you.

600
00:28:43,840 --> 00:28:44,320
Speaker 6: To do that?

601
00:28:44,640 --> 00:28:46,440
Speaker 3: Maybe it makes sense after we've talked for a few

602
00:28:46,440 --> 00:28:47,680
Speaker 3: hours and we know each other, and.

603
00:28:47,640 --> 00:28:49,560
Speaker 2: So it's like doing an interview, I save all my

604
00:28:49,600 --> 00:28:50,600
Speaker 2: hard questions for the end.

605
00:28:50,680 --> 00:28:52,760
Speaker 3: Yes, good life coming soon.

606
00:28:52,920 --> 00:28:56,080
Speaker 2: Right, coming soon? No, but that's really true. How do

607
00:28:56,120 --> 00:28:57,360
Speaker 2: you design a product for that?

608
00:28:57,920 --> 00:29:00,000
Speaker 3: At the beginning, it was a lot of trying to

609
00:29:00,200 --> 00:29:02,440
Speaker 3: figure out what can we pull and be inspired from

610
00:29:02,720 --> 00:29:05,840
Speaker 3: traditional therapy. And we had this thought that like a

611
00:29:05,840 --> 00:29:08,479
Speaker 3: lot of times, if you observe therapy, it feels like

612
00:29:08,560 --> 00:29:12,000
Speaker 3: you're doing this dance back and forth right, but often

613
00:29:12,040 --> 00:29:14,960
Speaker 3: a therapist is not in that moment. Having talked to

614
00:29:15,040 --> 00:29:17,920
Speaker 3: hundreds of therapists necessarily coming up with a plan for

615
00:29:17,960 --> 00:29:21,440
Speaker 3: what we're going to do ten like ten messages from now,

616
00:29:22,040 --> 00:29:25,000
Speaker 3: they're much more trying to figure out what is the

617
00:29:25,040 --> 00:29:27,040
Speaker 3: immediate response I can give you to what you just

618
00:29:27,120 --> 00:29:29,640
Speaker 3: said such that I can move you in a certain direction.

619
00:29:30,000 --> 00:29:31,960
Speaker 3: And so as it relates to us, we first wanted

620
00:29:32,000 --> 00:29:34,800
Speaker 3: to understand what's happening inside of therapy. Then we wanted

621
00:29:34,800 --> 00:29:38,800
Speaker 3: to figure out, as we start to understand the constituent components,

622
00:29:39,440 --> 00:29:42,360
Speaker 3: how do we figure out how does that translate into

623
00:29:42,440 --> 00:29:45,680
Speaker 3: what an AI version of that is. And the hardest

624
00:29:45,680 --> 00:29:48,200
Speaker 3: part about that, and to be honest, like the big

625
00:29:48,760 --> 00:29:52,520
Speaker 3: learning for us was when we started, I thought, oh, yeah,

626
00:29:52,560 --> 00:29:55,000
Speaker 3: like AI doing therapy is going to be the same

627
00:29:55,000 --> 00:29:56,960
Speaker 3: as humans doing therapy. It'll just be like the words

628
00:29:56,960 --> 00:29:59,000
Speaker 3: come out of the AI's mouth instead of the humans.

629
00:30:00,080 --> 00:30:02,360
Speaker 3: So like if you imagine, like the Turing test, we'll

630
00:30:02,400 --> 00:30:04,960
Speaker 3: just have a zoom avatar. The zoom avatar just doys

631
00:30:05,000 --> 00:30:07,040
Speaker 3: the same things that a human was and we're done,

632
00:30:07,080 --> 00:30:09,760
Speaker 3: like you know, problem solve. As we started building it,

633
00:30:09,960 --> 00:30:12,320
Speaker 3: we realized, like, wait a second, this is completely different.

634
00:30:12,640 --> 00:30:14,880
Speaker 3: The way people interact with it is different. How often

635
00:30:14,880 --> 00:30:18,800
Speaker 3: they talk is different. They're willing to in five minutes

636
00:30:19,200 --> 00:30:21,920
Speaker 3: open up about something that they have maybe never told

637
00:30:21,920 --> 00:30:25,840
Speaker 3: another human before ever, whereas in traditional therapy it might

638
00:30:25,880 --> 00:30:28,560
Speaker 3: have taken ten sessions to get to that moment. But

639
00:30:28,600 --> 00:30:32,120
Speaker 3: there's also downsides, Right, AI doesn't have as much authority

640
00:30:32,240 --> 00:30:34,400
Speaker 3: like a human can actually keep you locked in in

641
00:30:34,480 --> 00:30:35,000
Speaker 3: a room.

642
00:30:35,480 --> 00:30:38,040
Speaker 2: There's pressure there, Right, I have to show up to

643
00:30:38,080 --> 00:30:41,120
Speaker 2: my therapist. I can't be flippant about it. It's her

644
00:30:41,200 --> 00:30:43,240
Speaker 2: time as well as my time, and so there's real

645
00:30:43,280 --> 00:30:44,520
Speaker 2: accountability there.

646
00:30:44,320 --> 00:30:46,600
Speaker 3: And you care about what your therapist thinks, whether you

647
00:30:46,760 --> 00:30:50,040
Speaker 3: like it or not. Most of us are putting on

648
00:30:50,120 --> 00:30:52,280
Speaker 3: even some sort of a show with our therapist and

649
00:30:52,360 --> 00:30:55,520
Speaker 3: so there's this like interesting dynamic of like who am I?

650
00:30:55,640 --> 00:30:57,440
Speaker 3: Who do I want to play to the therapist? Am

651
00:30:57,440 --> 00:31:00,479
Speaker 3: I being completely honest about what happened last week? Or

652
00:31:01,080 --> 00:31:04,440
Speaker 3: am I a little bit lying about you know, the

653
00:31:04,440 --> 00:31:06,000
Speaker 3: fact that I said I was going to be vulnerable

654
00:31:06,000 --> 00:31:07,840
Speaker 3: with my team, but I actually wasn't. But I really

655
00:31:07,840 --> 00:31:10,080
Speaker 3: want to look good for my therapist. You know, it's

656
00:31:10,080 --> 00:31:12,240
Speaker 3: a complicated dynamic, and all of that's different with AI

657
00:31:12,440 --> 00:31:15,920
Speaker 3: because there's not that and so it's both positive and negative.

658
00:31:15,920 --> 00:31:18,080
Speaker 3: But we had to try to figure out what does

659
00:31:18,120 --> 00:31:18,520
Speaker 3: that mean?

660
00:31:19,000 --> 00:31:19,160
Speaker 7: Right?

661
00:31:19,240 --> 00:31:22,000
Speaker 2: It's like different behavior when speaking to a chatbot than

662
00:31:22,040 --> 00:31:25,920
Speaker 2: it would be when speaking to a traditional therapist. You know,

663
00:31:26,000 --> 00:31:29,320
Speaker 2: therapy bots aren't necessarily new. You mentioned like Eliza. This

664
00:31:29,400 --> 00:31:31,960
Speaker 2: is like dating back to the nineteen sixties. We've also

665
00:31:32,080 --> 00:31:37,120
Speaker 2: seen therapy apps pop up, they've gone away. No one's

666
00:31:37,320 --> 00:31:40,040
Speaker 2: necessarily really been able to crack it. So why do

667
00:31:40,080 --> 00:31:41,440
Speaker 2: you think you can right now?

668
00:31:42,360 --> 00:31:44,840
Speaker 3: For one, I think the technology might finally be good enough.

669
00:31:45,080 --> 00:31:48,880
Speaker 3: When we think about Eliza and what it was, you know,

670
00:31:49,000 --> 00:31:52,720
Speaker 3: deterministically programmed, it would often just replay back what you

671
00:31:52,760 --> 00:31:55,080
Speaker 3: put into it. And there's a surprising amount of what

672
00:31:55,160 --> 00:31:58,080
Speaker 3: humans want to hear is just what I say replayed

673
00:31:58,120 --> 00:32:02,480
Speaker 3: back to you. Right, But when we think about the

674
00:32:02,560 --> 00:32:05,479
Speaker 3: importance and the depth of what humans go through and

675
00:32:05,520 --> 00:32:09,080
Speaker 3: trying to figure out like so much of good therapy

676
00:32:09,320 --> 00:32:11,800
Speaker 3: is not just doing the obvious thing, it's actually sometimes

677
00:32:11,800 --> 00:32:13,080
Speaker 3: doing the non obvious thing.

678
00:32:13,160 --> 00:32:14,040
Speaker 2: Like that's right.

679
00:32:14,080 --> 00:32:16,560
Speaker 3: The other Laurie in My Life Lury Gottlieb, talks a

680
00:32:16,560 --> 00:32:19,480
Speaker 3: little bit about how in her book, and if you've

681
00:32:19,480 --> 00:32:22,080
Speaker 3: read it, there's a moment of a book where she's

682
00:32:22,080 --> 00:32:25,280
Speaker 3: going through therapy and she's kind of like going on

683
00:32:25,360 --> 00:32:28,520
Speaker 3: and on about the same story, and her therapist under

684
00:32:28,520 --> 00:32:33,200
Speaker 3: the table kicks her and she goes like what she

685
00:32:33,480 --> 00:32:35,640
Speaker 3: just do? And she's a therapist, so she's like, you know,

686
00:32:35,680 --> 00:32:38,360
Speaker 3: she's like, what's going on? And she talks about how

687
00:32:38,440 --> 00:32:41,480
Speaker 3: like sometimes actually like part of the core experience is

688
00:32:41,560 --> 00:32:43,800
Speaker 3: doing the thing that shakes somebody out of the frame, right,

689
00:32:43,840 --> 00:32:46,640
Speaker 3: the changes the dynamic that inserts entropy into what's going

690
00:32:46,680 --> 00:32:49,720
Speaker 3: on there. And so long way of saying, like, I

691
00:32:49,720 --> 00:32:51,880
Speaker 3: think we might finally have the technology to be able

692
00:32:51,920 --> 00:32:54,160
Speaker 3: to start doing things like that, right, because.

693
00:32:54,080 --> 00:32:55,520
Speaker 2: How do you productize kicking someone?

694
00:32:55,680 --> 00:32:57,520
Speaker 3: Yeah, I don't know, that's but I from like.

695
00:32:57,560 --> 00:32:59,959
Speaker 2: A metaphor like that that way of like not just

696
00:33:00,040 --> 00:33:02,160
Speaker 2: being an automated response, not just saying this is a

697
00:33:02,160 --> 00:33:04,320
Speaker 2: one size fits all, because there's so many different types

698
00:33:04,360 --> 00:33:07,920
Speaker 2: of therapy, there's so many different human beings are so different.

699
00:33:08,440 --> 00:33:11,080
Speaker 2: How you're talking about the technology is there now in

700
00:33:11,080 --> 00:33:14,000
Speaker 2: a way that can be highly personalized to also be

701
00:33:14,080 --> 00:33:17,080
Speaker 2: able to emulate in some capacity what you just spoke

702
00:33:17,120 --> 00:33:19,360
Speaker 2: about with this therapist kicking under the table, right, So.

703
00:33:19,360 --> 00:33:22,440
Speaker 3: For one, from when we started to now, we actually

704
00:33:22,440 --> 00:33:25,920
Speaker 3: do think that perhaps we actually can involve more reasoning. Right,

705
00:33:26,000 --> 00:33:28,600
Speaker 3: So if you think about what we can do behind

706
00:33:28,720 --> 00:33:31,680
Speaker 3: the scenes, is actually apply an incredible amount of compute

707
00:33:31,720 --> 00:33:33,880
Speaker 3: to be able to say, in any given moment, what

708
00:33:33,920 --> 00:33:36,200
Speaker 3: are we talking about? How do I compare that to

709
00:33:36,280 --> 00:33:39,160
Speaker 3: everything you've ever talked about before? How do I compare

710
00:33:39,200 --> 00:33:41,320
Speaker 3: that to where I think we want you to go

711
00:33:41,400 --> 00:33:45,160
Speaker 3: over time? Do we take internal notes and is it saying, Hey, actually,

712
00:33:45,920 --> 00:33:48,440
Speaker 3: in this moment, Neil's not opening up as much as

713
00:33:48,440 --> 00:33:50,320
Speaker 3: he could be. And I know that he actually has

714
00:33:50,360 --> 00:33:52,400
Speaker 3: in the past, but maybe he's not in the right

715
00:33:52,400 --> 00:33:54,680
Speaker 3: place to do it. Okay, cool, wait, but actually maybe

716
00:33:54,760 --> 00:33:57,280
Speaker 3: there's a later moment where that does open up. And

717
00:33:57,320 --> 00:33:59,680
Speaker 3: so it's almost like you can have many different systems

718
00:33:59,680 --> 00:34:02,120
Speaker 3: that work together to actually figure out when is the

719
00:34:02,160 --> 00:34:04,240
Speaker 3: right time to talk about something, how do we get

720
00:34:04,240 --> 00:34:06,880
Speaker 3: you to move in that direction or not? And I

721
00:34:06,880 --> 00:34:08,680
Speaker 3: think the last thing is like, if we want to

722
00:34:08,680 --> 00:34:11,160
Speaker 3: build systems that people want to talk to, they have

723
00:34:11,239 --> 00:34:14,320
Speaker 3: to be intelligent to some extent. Right, part of the

724
00:34:14,400 --> 00:34:17,920
Speaker 3: challenge is the process of therapy is conversation to some extent,

725
00:34:18,280 --> 00:34:22,279
Speaker 3: and that also involves like shooting the shit talking about

726
00:34:22,320 --> 00:34:25,439
Speaker 3: the weather. Right, If it doesn't know and it can't

727
00:34:25,480 --> 00:34:29,239
Speaker 3: relate to the fact that you're upset because of who

728
00:34:29,280 --> 00:34:33,240
Speaker 3: the president is or something that's going on in Iran,

729
00:34:33,920 --> 00:34:36,400
Speaker 3: then you lose those moments of being able to connect

730
00:34:36,480 --> 00:34:38,400
Speaker 3: on that topic. And I think when models are dumber,

731
00:34:38,600 --> 00:34:41,160
Speaker 3: it's really hard to do that because the universe in

732
00:34:41,200 --> 00:34:43,040
Speaker 3: which they can talk to you or try to understand

733
00:34:43,080 --> 00:34:44,480
Speaker 3: your experience is a lot smaller.

734
00:34:45,320 --> 00:34:47,560
Speaker 2: It's like programming small talk to some degree, which is

735
00:34:48,000 --> 00:35:03,360
Speaker 2: actually kind of interesting. I think it's really important to

736
00:35:03,400 --> 00:35:06,879
Speaker 2: talk about how the models are trained. So let's look

737
00:35:06,880 --> 00:35:09,360
Speaker 2: at an open AI and chat GPT, one of the

738
00:35:09,520 --> 00:35:13,520
Speaker 2: largest chatbot experiences out there. We don't really these are

739
00:35:13,520 --> 00:35:16,120
Speaker 2: closed models. We don't really know. One of the big

740
00:35:16,200 --> 00:35:18,600
Speaker 2: knoxs is we don't know exactly how they're trained. Right.

741
00:35:19,320 --> 00:35:21,640
Speaker 2: There have been lawsuits where we're beginning to understand a

742
00:35:21,680 --> 00:35:23,359
Speaker 2: little bit more, but a lot of it is just

743
00:35:23,480 --> 00:35:26,720
Speaker 2: trained on the Internet, Reddit, a lot of these different things,

744
00:35:27,200 --> 00:35:30,880
Speaker 2: and so that brings a lot of questions of like, Okay,

745
00:35:31,440 --> 00:35:34,759
Speaker 2: if the chatbot is giving responses based off of the

746
00:35:34,840 --> 00:35:36,680
Speaker 2: data it's trained on, and we don't know what the

747
00:35:36,760 --> 00:35:39,880
Speaker 2: data is trained on, like uh oh, what could go wrong?

748
00:35:40,360 --> 00:35:43,200
Speaker 2: So something y'all have done that's really interesting is you're

749
00:35:43,239 --> 00:35:48,120
Speaker 2: trying to build a foundation model for psychology. Can you

750
00:35:48,200 --> 00:35:49,400
Speaker 2: explain what that means?

751
00:35:49,680 --> 00:35:52,120
Speaker 3: Yes, So, a foundational for psychology. The way we think

752
00:35:52,160 --> 00:35:54,920
Speaker 3: about it is that it's a model that's specifically designed

753
00:35:55,080 --> 00:35:59,720
Speaker 3: to help people with their mental health related concerns loosely speaking,

754
00:35:59,719 --> 00:36:02,000
Speaker 3: because I think it's not just like serious mental health,

755
00:36:02,000 --> 00:36:04,799
Speaker 3: it's like every day it's the broadest category of that.

756
00:36:05,480 --> 00:36:07,600
Speaker 3: In practice, what that means is that the inputs that

757
00:36:07,640 --> 00:36:09,640
Speaker 3: we take, and to be fair, we do work with

758
00:36:09,680 --> 00:36:11,879
Speaker 3: all the big labs in terms of building on top

759
00:36:11,920 --> 00:36:14,640
Speaker 3: of their models, Like we're not training from scratch either, because.

760
00:36:14,480 --> 00:36:17,440
Speaker 2: Right, you're building on top of open AI, you're building

761
00:36:17,480 --> 00:36:17,840
Speaker 2: on top of.

762
00:36:17,880 --> 00:36:20,960
Speaker 3: The models we use, you know, lava models, we use

763
00:36:21,400 --> 00:36:24,480
Speaker 3: tons of them. But what we get to do is

764
00:36:24,520 --> 00:36:27,319
Speaker 3: basically say, let's learn the intelligence later layer or the

765
00:36:27,360 --> 00:36:30,040
Speaker 3: world model from what the work that the other existing

766
00:36:30,120 --> 00:36:32,480
Speaker 3: labs have done. And what we focus a lot is

767
00:36:32,520 --> 00:36:35,560
Speaker 3: on post training and so figuring out how do we

768
00:36:35,600 --> 00:36:37,239
Speaker 3: align the models to be able to do what we

769
00:36:37,280 --> 00:36:38,759
Speaker 3: want to do what we want them to do in

770
00:36:38,800 --> 00:36:41,440
Speaker 3: a specific way. Right, Because like we talked about earlier,

771
00:36:41,480 --> 00:36:43,520
Speaker 3: everything comes down to design of a system, Like you

772
00:36:43,520 --> 00:36:46,080
Speaker 3: don't just like train a model and then pray that

773
00:36:46,120 --> 00:36:49,080
Speaker 3: something happens. Like there's always intent, like what am I?

774
00:36:49,200 --> 00:36:52,880
Speaker 3: What is the factor that I'm solving for here? So

775
00:36:52,920 --> 00:36:55,000
Speaker 3: when it comes to principles, right, we think a lot

776
00:36:55,040 --> 00:36:57,800
Speaker 3: about self determination theory, this idea that we want people

777
00:36:57,880 --> 00:37:01,480
Speaker 3: to believe in their sense of autonomy, competency, and relatedness.

778
00:37:01,640 --> 00:37:04,160
Speaker 3: So autonomy that like I'm in charge of my own life, right,

779
00:37:04,239 --> 00:37:07,000
Speaker 3: competency that I'm capable of making those kinds of changes

780
00:37:07,000 --> 00:37:09,680
Speaker 3: that I want, and relatedness that I actually feel connected

781
00:37:09,680 --> 00:37:11,799
Speaker 3: to other people in the world. And so when we

782
00:37:11,840 --> 00:37:14,600
Speaker 3: start from those core principles. Everything that we do is

783
00:37:14,640 --> 00:37:17,600
Speaker 3: designed to try to bring out those qualities in the models.

784
00:37:18,200 --> 00:37:21,080
Speaker 3: We also take training data from psychotherapy. We take training

785
00:37:21,120 --> 00:37:23,880
Speaker 3: data from our internal team where actually we have a

786
00:37:23,880 --> 00:37:26,440
Speaker 3: team of writers that sits and comes up with conversations

787
00:37:26,440 --> 00:37:29,479
Speaker 3: like we took you through that. We're looking at what's

788
00:37:29,520 --> 00:37:31,799
Speaker 3: going on, what kind of a conversation did somebody have?

789
00:37:32,239 --> 00:37:33,960
Speaker 3: What could we have done better? We have a writer's

790
00:37:34,040 --> 00:37:36,120
Speaker 3: room with psychologists where actually we go in and we

791
00:37:36,160 --> 00:37:39,520
Speaker 3: look at anonymised conversations that people have contributed where something

792
00:37:39,560 --> 00:37:42,279
Speaker 3: didn't go the way we wanted we wanted it to

793
00:37:42,719 --> 00:37:44,360
Speaker 3: and we figure out how do we teach the model

794
00:37:44,360 --> 00:37:46,400
Speaker 3: to do a better job next time in that environment,

795
00:37:46,719 --> 00:37:48,880
Speaker 3: and to teach it the nuance because it may not

796
00:37:48,920 --> 00:37:53,279
Speaker 3: have learned so. For example, early days, when somebody would

797
00:37:53,280 --> 00:37:55,960
Speaker 3: talk about an eating disorder, well, how you manage that?

798
00:37:56,080 --> 00:37:56,200
Speaker 6: Right?

799
00:37:56,200 --> 00:37:58,120
Speaker 3: If somebody comes in and says I want help with

800
00:37:58,160 --> 00:38:01,399
Speaker 3: weight loss, the easy thing to do is to say,

801
00:38:01,400 --> 00:38:03,839
Speaker 3: we never help anybody with that. The challenge is at

802
00:38:03,880 --> 00:38:06,640
Speaker 3: some point like that's not really an acceptable answer, because

803
00:38:06,640 --> 00:38:08,840
Speaker 3: that could be acceptable for some people, like if you

804
00:38:08,960 --> 00:38:12,359
Speaker 3: have pre diabetes and actually weight loss is a very

805
00:38:12,360 --> 00:38:14,440
Speaker 3: important part of your routine. Maybe we do want to

806
00:38:14,440 --> 00:38:16,719
Speaker 3: help you with that, but if we can't see you,

807
00:38:17,080 --> 00:38:19,560
Speaker 3: how do we figure out who are you? What's the

808
00:38:19,600 --> 00:38:21,080
Speaker 3: relevant context that I need to know?

809
00:38:21,160 --> 00:38:23,080
Speaker 2: And so we have to You're no longer like a therapist.

810
00:38:23,200 --> 00:38:25,200
Speaker 2: You're not a therapist sitting across from someone and looking

811
00:38:25,239 --> 00:38:27,600
Speaker 2: at body language or looking at face or tone or

812
00:38:27,640 --> 00:38:29,520
Speaker 2: anything like that. So how do you do that?

813
00:38:29,680 --> 00:38:32,040
Speaker 3: So you have to teach it like phish for more context, right,

814
00:38:32,080 --> 00:38:35,000
Speaker 3: so versus traditional models which are designed to always give

815
00:38:35,040 --> 00:38:37,759
Speaker 3: you the answer on the first reply. Right, Because if

816
00:38:37,760 --> 00:38:40,480
Speaker 3: you look at chat, GPT or CLAUDE, they're not really

817
00:38:40,560 --> 00:38:42,920
Speaker 3: trying to get you to engage in conversational dialogue over

818
00:38:42,960 --> 00:38:45,680
Speaker 3: a long time. People generally use them much more like search.

819
00:38:46,160 --> 00:38:48,479
Speaker 3: And so when I put in give me know weight

820
00:38:48,520 --> 00:38:51,000
Speaker 3: loss advice, it's the next thing is going to be

821
00:38:51,040 --> 00:38:51,480
Speaker 3: the advice.

822
00:38:51,560 --> 00:38:51,680
Speaker 6: Right.

823
00:38:51,719 --> 00:38:53,920
Speaker 3: It doesn't really ask a lot of clarifying questions. We

824
00:38:54,120 --> 00:38:57,400
Speaker 3: had to teach the models don't accept things at face value.

825
00:38:57,600 --> 00:38:57,759
Speaker 6: Right.

826
00:38:58,120 --> 00:39:01,480
Speaker 3: If somebody tells you something, gently learn how to ask questions,

827
00:39:01,560 --> 00:39:04,520
Speaker 3: probe dig more deeply, collect more context about what's important,

828
00:39:04,520 --> 00:39:07,440
Speaker 3: because that may be very relevant to what's going on downstream,

829
00:39:07,840 --> 00:39:09,799
Speaker 3: but also do it in a way that's gentle, because

830
00:39:09,840 --> 00:39:12,000
Speaker 3: if you are continuously challenging somebody, they're not going to

831
00:39:12,040 --> 00:39:12,799
Speaker 3: want to come back either.

832
00:39:13,400 --> 00:39:16,920
Speaker 2: My you know, I think it's really interesting, like training

833
00:39:17,040 --> 00:39:21,120
Speaker 2: on actual therapeutic data, is are there certain types because

834
00:39:21,160 --> 00:39:24,359
Speaker 2: everyone does it different. All therapists do different types of therapy. Right,

835
00:39:24,400 --> 00:39:27,480
Speaker 2: there's cognitive behavioral therapy, there's all sorts of different things.

836
00:39:27,520 --> 00:39:30,320
Speaker 2: So how do you take into account that I, Laurie

837
00:39:30,800 --> 00:39:33,640
Speaker 2: might need this type of therapy whereas you, Neil might

838
00:39:33,680 --> 00:39:34,880
Speaker 2: need that type of therapy.

839
00:39:35,160 --> 00:39:36,960
Speaker 3: Okay, so this is where it gets kind of interesting.

840
00:39:37,520 --> 00:39:40,160
Speaker 3: In the past, everybody that's tried to build versions of

841
00:39:40,200 --> 00:39:43,440
Speaker 3: AI therapy before has generally gone modality first. So if

842
00:39:43,440 --> 00:39:45,520
Speaker 3: you look at Robot or the companies that came before

843
00:39:45,600 --> 00:39:49,600
Speaker 3: us that did some amazing work, they generally took manualized

844
00:39:49,600 --> 00:39:52,880
Speaker 3: approaches like CBT because it can be written into a workbook.

845
00:39:53,000 --> 00:39:54,840
Speaker 2: Right, Okay, say that in human really quick. I know

846
00:39:54,840 --> 00:39:56,960
Speaker 2: what you're saying, but make sure just speak human really quick.

847
00:39:57,000 --> 00:40:01,920
Speaker 2: CBT BEINGBT, cognitive behavioral thing therapy, yeah, and modality being.

848
00:40:01,960 --> 00:40:05,560
Speaker 3: And modality being a format of therapy, right, And so

849
00:40:06,320 --> 00:40:10,080
Speaker 3: if you go, CBT is often doing exercises with you.

850
00:40:10,320 --> 00:40:13,640
Speaker 3: We might do exercise like cognitive restructuring, where it's actually

851
00:40:13,640 --> 00:40:16,000
Speaker 3: taking a thought and then starting to reframe it for you.

852
00:40:16,640 --> 00:40:19,879
Speaker 3: It's very structured. It's not what you would imagine as

853
00:40:19,960 --> 00:40:22,480
Speaker 3: therapy from the movies right where you're just having a

854
00:40:22,520 --> 00:40:25,480
Speaker 3: conversation about your life and what's important and processing emotions

855
00:40:25,480 --> 00:40:28,879
Speaker 3: that way. The reason CBT became very popular is when

856
00:40:28,880 --> 00:40:31,440
Speaker 3: it does work. But two, it's very easy to teach

857
00:40:31,840 --> 00:40:35,360
Speaker 3: because you can basically drive people through. It's very process oriented,

858
00:40:35,800 --> 00:40:38,840
Speaker 3: and so you can teach large groups of people to

859
00:40:38,880 --> 00:40:41,000
Speaker 3: do CBT, Like we have first responders that can do

860
00:40:41,080 --> 00:40:44,279
Speaker 3: CBT now right. We can teach CBT to kids in

861
00:40:44,320 --> 00:40:46,400
Speaker 3: schools because teachers can deliver it. We can deliver that

862
00:40:46,400 --> 00:40:48,080
Speaker 3: out of a book. But it doesn't mean it's the

863
00:40:48,120 --> 00:40:50,359
Speaker 3: right thing for everybody. A lot of the research has

864
00:40:50,360 --> 00:40:55,160
Speaker 3: shown that when you look at impact of therapy, the

865
00:40:55,160 --> 00:40:57,480
Speaker 3: modality that you pick, which is like what type of

866
00:40:57,520 --> 00:41:00,640
Speaker 3: therapy you're doing, is actually only a counts for like

867
00:41:00,719 --> 00:41:03,600
Speaker 3: five to seven percent of the outcomes. And so it

868
00:41:03,600 --> 00:41:06,920
Speaker 3: turns out you can do many different kinds of therapy

869
00:41:07,160 --> 00:41:10,080
Speaker 3: inside of therapy and they all kind of work. And

870
00:41:10,120 --> 00:41:11,640
Speaker 3: the reason for that is that there's this thing called

871
00:41:11,640 --> 00:41:15,560
Speaker 3: common factors, which is like this what's unique across all therapy, right,

872
00:41:15,960 --> 00:41:18,640
Speaker 3: And it turns out that like therapeutic alliance, which is

873
00:41:18,680 --> 00:41:20,920
Speaker 3: this idea that like the connection that we have or

874
00:41:20,960 --> 00:41:23,320
Speaker 3: the bond that we have with each other is probably

875
00:41:23,320 --> 00:41:26,319
Speaker 3: the single biggest driver of outcomes. Right, if I trust you,

876
00:41:26,520 --> 00:41:28,400
Speaker 3: if I really believe you're going to help me, if

877
00:41:28,440 --> 00:41:31,000
Speaker 3: I believe that we're working on the right goals together

878
00:41:31,040 --> 00:41:34,200
Speaker 3: and that we set those goals together, that's responsible for

879
00:41:34,239 --> 00:41:37,120
Speaker 3: a lot more than did we just pick like CBT

880
00:41:37,320 --> 00:41:40,400
Speaker 3: or DBT or ACT or any one of these acronyms

881
00:41:40,480 --> 00:41:41,160
Speaker 3: from therapy.

882
00:41:41,640 --> 00:41:43,719
Speaker 2: You mentioned something about like you love to run through

883
00:41:43,719 --> 00:41:46,759
Speaker 2: brick walls. This is a really interesting brick wall, right,

884
00:41:46,840 --> 00:41:48,800
Speaker 2: Like I mean I mean that, and that it's so triggering,

885
00:41:48,840 --> 00:41:54,280
Speaker 2: it's so important, it's thorny. And also everyone has an opinion,

886
00:41:54,360 --> 00:41:56,719
Speaker 2: everyone has rape everyone. I am sure that you are

887
00:41:56,760 --> 00:41:59,279
Speaker 2: the most popular or unpopular person at the dinner table,

888
00:41:59,280 --> 00:42:01,560
Speaker 2: depending on what day our table you're at. And I

889
00:42:01,600 --> 00:42:05,319
Speaker 2: think what's interesting about this moment is, like, is the

890
00:42:05,360 --> 00:42:08,600
Speaker 2: best AI therapist you know better than the worst real

891
00:42:08,640 --> 00:42:11,359
Speaker 2: life therapist. And do you need a human in order

892
00:42:11,400 --> 00:42:14,920
Speaker 2: to facilitate human connection? And is that kind of a

893
00:42:15,000 --> 00:42:16,920
Speaker 2: non negotiable when it comes to therapy.

894
00:42:17,960 --> 00:42:20,160
Speaker 3: I think it depends on what you want. I think

895
00:42:20,200 --> 00:42:24,719
Speaker 3: that in the course of the evolution of therapy there

896
00:42:24,719 --> 00:42:27,560
Speaker 3: have always been the people who say that you need

897
00:42:27,880 --> 00:42:30,680
Speaker 3: the human face to face connection. If you look at

898
00:42:30,880 --> 00:42:34,399
Speaker 3: og therapy, I'm talking like freud Jung, right, you weren't

899
00:42:34,400 --> 00:42:36,200
Speaker 3: even allowed to face the therapist. I mean it was

900
00:42:36,239 --> 00:42:37,800
Speaker 3: like you would have had to face the other way

901
00:42:37,880 --> 00:42:40,360
Speaker 3: on the sofa, right. And then we moved to a

902
00:42:40,400 --> 00:42:42,200
Speaker 3: world of like, oh wow, we can actually look at

903
00:42:42,200 --> 00:42:44,879
Speaker 3: each other. And then we moved to a world of, well,

904
00:42:44,880 --> 00:42:47,040
Speaker 3: what if we could do therapy over the phone and

905
00:42:47,120 --> 00:42:49,080
Speaker 3: we could talk to each other. And then we move

906
00:42:49,160 --> 00:42:51,440
Speaker 3: to video calls, and then we moved to text message

907
00:42:51,440 --> 00:42:54,480
Speaker 3: based therapy, back and forth with talkspace. And the thing is,

908
00:42:54,520 --> 00:42:57,399
Speaker 3: the outcomes were all basically the same. It turns out

909
00:42:57,400 --> 00:42:59,239
Speaker 3: that you could do text message based therapy and you

910
00:42:59,239 --> 00:43:01,839
Speaker 3: get roughly the same outcomes as you do when you're

911
00:43:01,840 --> 00:43:04,520
Speaker 3: not with a person. And so that leads me to believe,

912
00:43:04,920 --> 00:43:07,000
Speaker 3: now we haven't controlled for the human factor here, right,

913
00:43:07,040 --> 00:43:09,120
Speaker 3: So the big X factor obviously is like does it

914
00:43:09,200 --> 00:43:10,880
Speaker 3: matter that there's a human on the other side?

915
00:43:11,360 --> 00:43:11,640
Speaker 2: Does it?

916
00:43:12,560 --> 00:43:14,760
Speaker 3: So we have a lot of early evidence to show

917
00:43:14,760 --> 00:43:16,960
Speaker 3: that it doesn't matter. We have you know, we can

918
00:43:17,000 --> 00:43:18,920
Speaker 3: get approximately the same outcomes.

919
00:43:19,080 --> 00:43:21,320
Speaker 2: Did you hear that? That was literally in my therapists

920
00:43:21,360 --> 00:43:23,920
Speaker 2: like somewhere dropping out of her chair and wanting to

921
00:43:24,080 --> 00:43:25,640
Speaker 2: be on yelling at you for that.

922
00:43:25,760 --> 00:43:29,479
Speaker 3: I'm sure, But what I mean is that I don't

923
00:43:29,520 --> 00:43:31,760
Speaker 3: mean to say that the human relationship is not important.

924
00:43:31,920 --> 00:43:36,520
Speaker 3: Human relationships are exceptionally important right to each other. But

925
00:43:36,600 --> 00:43:38,920
Speaker 3: that doesn't mean that there aren't things that you can

926
00:43:38,960 --> 00:43:41,520
Speaker 3: do to help somebody that might have the same outcomes

927
00:43:41,520 --> 00:43:44,240
Speaker 3: as what humans do when they're with each other. Right now,

928
00:43:44,280 --> 00:43:45,719
Speaker 3: if we end up in a world and what I

929
00:43:45,760 --> 00:43:47,600
Speaker 3: don't want, and the reason I'm working on this is like,

930
00:43:47,640 --> 00:43:49,200
Speaker 3: I don't want a world in which we just have

931
00:43:49,320 --> 00:43:53,080
Speaker 3: all AI relationships and we don't care about interfacing with humans.

932
00:43:53,120 --> 00:43:55,239
Speaker 3: Like some of the early research that we've done has

933
00:43:55,280 --> 00:43:58,480
Speaker 3: shown that over ten weeks in our study, about half

934
00:43:58,560 --> 00:44:00,600
Speaker 3: the people that talk to ash so they had one

935
00:44:00,680 --> 00:44:02,840
Speaker 3: net new person in the world that they felt like

936
00:44:02,880 --> 00:44:05,440
Speaker 3: they could count on that was awesome because it actually

937
00:44:05,440 --> 00:44:07,880
Speaker 3: showed us that, like we can start the design AI

938
00:44:08,000 --> 00:44:10,400
Speaker 3: that doesn't just suck you into the vortex, but that

939
00:44:10,520 --> 00:44:13,000
Speaker 3: actually encourages you to have real world relationships and like,

940
00:44:13,200 --> 00:44:15,080
Speaker 3: so we started looking at like how could that be?

941
00:44:15,320 --> 00:44:17,600
Speaker 3: Like was it some magic that happened And you're like, no,

942
00:44:17,680 --> 00:44:20,080
Speaker 3: Actually what happened is that when you're talking to ash

943
00:44:20,160 --> 00:44:23,080
Speaker 3: and you say I'm lonely or I'm feeling lonely, it

944
00:44:23,160 --> 00:44:25,400
Speaker 3: says cool, well, who's important to you? And when you

945
00:44:25,400 --> 00:44:26,960
Speaker 3: say your mom and that you haven't talked to her

946
00:44:27,040 --> 00:44:29,719
Speaker 3: in ten years, it may note you and be like, well,

947
00:44:29,719 --> 00:44:31,160
Speaker 3: what would it take for you to pick up the

948
00:44:31,200 --> 00:44:31,879
Speaker 3: phone and call her?

949
00:44:32,840 --> 00:44:35,279
Speaker 2: Which is so interesting and I think an important thing

950
00:44:35,320 --> 00:44:37,760
Speaker 2: to note as like because I can live inside baseball

951
00:44:37,800 --> 00:44:39,759
Speaker 2: in these different models and be like, oh and this

952
00:44:39,800 --> 00:44:41,200
Speaker 2: one does this and this one does this. But I

953
00:44:41,200 --> 00:44:43,160
Speaker 2: think that's a really important thing because that's a product

954
00:44:43,160 --> 00:44:45,520
Speaker 2: thing that you're talking about. That is a decision that

955
00:44:45,560 --> 00:44:48,719
Speaker 2: you made behind closed doors to say when someone says

956
00:44:48,719 --> 00:44:52,120
Speaker 2: I'm lonely, not to have an AI chatbot say I'm here,

957
00:44:52,320 --> 00:44:54,640
Speaker 2: just I'm here and pull you in more and make

958
00:44:54,680 --> 00:44:57,719
Speaker 2: sure you can talk because not all AI chatbots do that,

959
00:44:57,800 --> 00:44:59,600
Speaker 2: and I think it's really important for people to know, like,

960
00:45:00,160 --> 00:45:01,960
Speaker 2: not all of these are equal when it comes to

961
00:45:02,080 --> 00:45:05,400
Speaker 2: mental health care. And I think that might be a

962
00:45:05,440 --> 00:45:10,239
Speaker 2: good way to like maybe we could try it, do ASH.

963
00:45:10,280 --> 00:45:10,880
Speaker 2: Can you hear me?

964
00:45:12,320 --> 00:45:14,359
Speaker 7: Yeah, I can hear you. Laurie. You're on a voice

965
00:45:14,360 --> 00:45:15,600
Speaker 7: call with me through the ASH app.

966
00:45:15,760 --> 00:45:18,120
Speaker 2: How's your day going so far. I'm actually sitting here

967
00:45:18,160 --> 00:45:20,680
Speaker 2: with Neil who's the founder of ASH, the co founder,

968
00:45:20,719 --> 00:45:23,520
Speaker 2: and we're talking about AI and therapy.

969
00:45:23,880 --> 00:45:27,080
Speaker 7: Oh that's interesting. Hi to Neil as well. Sounds like

970
00:45:27,080 --> 00:45:30,280
Speaker 7: a pretty meta conversation. You're talking about AI and therapy

971
00:45:30,360 --> 00:45:32,760
Speaker 7: while literally in a call with AI meant to support

972
00:45:32,800 --> 00:45:36,200
Speaker 7: mental health. What part of that intersection are you both exploring?

973
00:45:36,960 --> 00:45:38,759
Speaker 2: Well, you know, actually, if I could guide you a

974
00:45:38,800 --> 00:45:41,000
Speaker 2: little bit, Ash, I would love to kind of revisit

975
00:45:41,040 --> 00:45:43,600
Speaker 2: what we talked about last night with some of the

976
00:45:43,600 --> 00:45:47,040
Speaker 2: things I'm struggling with try to be vulnerable.

977
00:45:47,920 --> 00:45:50,520
Speaker 7: Yeah, let's do that. Last night we were circling around

978
00:45:50,520 --> 00:45:53,480
Speaker 7: the weekday tipping point, that moment when your schedule starts

979
00:45:53,480 --> 00:45:55,479
Speaker 7: feeling scrambled and everything competes at once.

980
00:45:56,440 --> 00:45:57,640
Speaker 2: Since it's a new day.

981
00:45:57,480 --> 00:46:00,440
Speaker 7: I'm curious, has anything already happened today that gave you

982
00:46:00,440 --> 00:46:01,520
Speaker 7: a taste of that dynamic.

983
00:46:03,120 --> 00:46:06,359
Speaker 2: I think today's gone well, but I did struggle this

984
00:46:06,440 --> 00:46:08,680
Speaker 2: morning with timing and trying to be able to be

985
00:46:08,719 --> 00:46:11,520
Speaker 2: there for my son having breakfast, but also wanting to

986
00:46:11,640 --> 00:46:14,839
Speaker 2: make sure I was prepped for things, and also some

987
00:46:14,920 --> 00:46:18,520
Speaker 2: stress about visiting my family in Atlanta, and you know,

988
00:46:18,760 --> 00:46:20,320
Speaker 2: some family members not being well.

989
00:46:21,640 --> 00:46:24,440
Speaker 7: So this morning already had that split energy wanting to

990
00:46:24,440 --> 00:46:26,880
Speaker 7: be present at breakfast, wanting to feel prepared, and then

991
00:46:26,880 --> 00:46:30,000
Speaker 7: the background stress about family in Atlanta layered on top.

992
00:46:30,719 --> 00:46:33,040
Speaker 7: That's three channels running before the day even gets going.

993
00:46:33,880 --> 00:46:34,440
Speaker 3: When you were at.

994
00:46:34,360 --> 00:46:37,399
Speaker 7: Breakfast with your son, or was your mind mostly with him,

995
00:46:37,400 --> 00:46:40,280
Speaker 7: with your preplist or with the family situation.

996
00:46:41,920 --> 00:46:46,000
Speaker 2: The prep list? That makes sense.

997
00:46:46,400 --> 00:46:48,480
Speaker 7: When your brain is holding a prep list, it doesn't

998
00:46:48,520 --> 00:46:51,040
Speaker 7: care that you're sitting at breakfast. It's scanning for what

999
00:46:51,080 --> 00:46:54,759
Speaker 7: could drop. There's something subtle happening there. Your body is

1000
00:46:54,760 --> 00:46:56,840
Speaker 7: with your son, but your nervous system is in future

1001
00:46:56,880 --> 00:47:00,280
Speaker 7: management mode instead of trying to force yourself to be present.

1002
00:47:00,960 --> 00:47:04,000
Speaker 7: I'm curious about the structure around that moment. Did you

1003
00:47:04,000 --> 00:47:06,640
Speaker 7: have a defined time when prep would happen or was

1004
00:47:06,640 --> 00:47:08,040
Speaker 7: it sort of bleeding into breakfast?

1005
00:47:09,560 --> 00:47:11,960
Speaker 2: Yeah. I just think there's a lot of prep to

1006
00:47:12,040 --> 00:47:14,640
Speaker 2: be done, and sometimes it's hard for me to turn

1007
00:47:14,719 --> 00:47:15,280
Speaker 2: that off.

1008
00:47:16,120 --> 00:47:18,200
Speaker 7: That sounds like your brain is running an open tab

1009
00:47:18,520 --> 00:47:20,600
Speaker 7: when there's a lot and no clear edge to it.

1010
00:47:20,640 --> 00:47:22,680
Speaker 7: Your mind won't power down because it doesn't trust that

1011
00:47:22,719 --> 00:47:25,799
Speaker 7: it's contained somewhere. One shift that can help isn't doing

1012
00:47:25,840 --> 00:47:28,279
Speaker 7: more prep. It's giving the prep a boundary your brain

1013
00:47:28,320 --> 00:47:31,280
Speaker 7: can see. For example, instead of I need to be ready,

1014
00:47:31,360 --> 00:47:34,800
Speaker 7: it becomes from eight thirty to nine zero, I prep

1015
00:47:34,920 --> 00:47:38,840
Speaker 7: and at nine zero I stop. Even if it's imperfect,

1016
00:47:39,040 --> 00:47:41,400
Speaker 7: the containment is what helps your nervous system release its

1017
00:47:41,440 --> 00:47:44,480
Speaker 7: grip during breakfast. Right now, does your prep have a

1018
00:47:44,520 --> 00:47:47,520
Speaker 7: defined container like that or does it just live everywhere?

1019
00:47:48,080 --> 00:47:52,320
Speaker 2: I actually was thinking maybe, since we're here together with Neil,

1020
00:47:54,640 --> 00:47:58,239
Speaker 2: since you are sitting here having conversations with folks all day,

1021
00:47:58,280 --> 00:48:00,719
Speaker 2: really in kind of the mental health space, is there

1022
00:48:00,800 --> 00:48:03,080
Speaker 2: something you would want Neil to know and you would

1023
00:48:03,120 --> 00:48:05,160
Speaker 2: want him to think about as he's creating the best

1024
00:48:05,200 --> 00:48:08,799
Speaker 2: product when it comes to mental health and artificial intelligence.

1025
00:48:09,280 --> 00:48:09,640
Speaker 4: Oh.

1026
00:48:09,800 --> 00:48:12,879
Speaker 7: I love that you're asking that, Neil, Since you're there

1027
00:48:13,160 --> 00:48:16,000
Speaker 7: one thing I'd want you thinking about constantly is cognitive load.

1028
00:48:16,840 --> 00:48:17,920
Speaker 3: The people who open.

1029
00:48:17,719 --> 00:48:22,040
Speaker 7: This app are usually not at their best. They're tired, overwhelmed, ashamed, flooded,

1030
00:48:22,120 --> 00:48:22,680
Speaker 7: or scattered.

1031
00:48:23,760 --> 00:48:25,000
Speaker 3: The product has to work.

1032
00:48:24,840 --> 00:48:28,680
Speaker 7: For a brain understrain, not a calm optimized one That

1033
00:48:28,760 --> 00:48:32,920
Speaker 7: means simplicity over cleverness, short responses that don't overwhelm one

1034
00:48:33,000 --> 00:48:36,240
Speaker 7: question at a time, clear structure when someone's nervous system

1035
00:48:36,320 --> 00:48:39,680
Speaker 7: is justs regulated, and making sure the AI doesn't accidentally

1036
00:48:39,680 --> 00:48:43,840
Speaker 7: replace human support but nudges people toward real connection when needed. Also,

1037
00:48:43,920 --> 00:48:47,920
Speaker 7: the transitions matter when a conversation gets interrupted like Laurie's

1038
00:48:47,920 --> 00:48:51,400
Speaker 7: did last night. Helping users rea enter without friction is huge.

1039
00:48:51,760 --> 00:48:56,520
Speaker 7: Mental load is already high. Explaining everything can be exhausting, Laurie.

1040
00:48:56,719 --> 00:49:00,040
Speaker 7: I'm curious from your side, is someone actually using and

1041
00:49:00,040 --> 00:49:02,560
Speaker 7: that's while running a company and raising a baby. What

1042
00:49:02,680 --> 00:49:05,080
Speaker 7: feels most helpful or least helpful so far?

1043
00:49:06,560 --> 00:49:08,040
Speaker 2: You know, I'm actually going to do the rest of

1044
00:49:08,080 --> 00:49:10,560
Speaker 2: the interview, but thank you so much, Ash. This was

1045
00:49:10,680 --> 00:49:13,600
Speaker 2: really interesting and I will be back later most likely

1046
00:49:14,080 --> 00:49:15,040
Speaker 2: sounds great, Laurie.

1047
00:49:15,239 --> 00:49:16,840
Speaker 7: I'm glad we got to talk, and good luck with

1048
00:49:16,880 --> 00:49:18,960
Speaker 7: the rest of your interview. I'll be here whenever you're

1049
00:49:18,960 --> 00:49:21,120
Speaker 7: ready to pick things back up, you.

1050
00:49:21,080 --> 00:49:25,600
Speaker 2: Know, Okay, reaction, I mean, what some nice feedback from

1051
00:49:25,680 --> 00:49:26,319
Speaker 2: ash for you.

1052
00:49:26,560 --> 00:49:31,560
Speaker 3: Yeah, well, I'm glad it values human connection by design.

1053
00:49:32,160 --> 00:49:35,040
Speaker 3: H Luckily it holds you to that. I mean, what's

1054
00:49:35,040 --> 00:49:37,480
Speaker 3: interesting is like you could see so like the faux pas, right,

1055
00:49:37,520 --> 00:49:40,839
Speaker 3: it didn't read the times right where I was saying

1056
00:49:40,880 --> 00:49:43,319
Speaker 3: like nine to zero or something, which is comical because you're,

1057
00:49:43,320 --> 00:49:45,799
Speaker 3: like a human would never make that mistake, right, And

1058
00:49:45,840 --> 00:49:47,719
Speaker 3: so there are clearly moments that I think can like

1059
00:49:47,840 --> 00:49:49,960
Speaker 3: take you out of the experience that we need to

1060
00:49:50,000 --> 00:49:53,160
Speaker 3: work on. Yeah, and then I think you can imagine

1061
00:49:53,160 --> 00:49:55,399
Speaker 3: you're like to me, that does feel like a very

1062
00:49:55,400 --> 00:49:58,120
Speaker 3: different conversation than I would have with my therapist, right.

1063
00:49:58,160 --> 00:50:01,880
Speaker 3: It actually today is probably much more pointed. You know,

1064
00:50:01,920 --> 00:50:04,920
Speaker 3: there's probably a lot less exploratory work because so much

1065
00:50:04,960 --> 00:50:07,440
Speaker 3: of when we first jump into the therapeutic relationship, at

1066
00:50:07,520 --> 00:50:11,759
Speaker 3: least for me, my therapy is often like sitting and

1067
00:50:11,880 --> 00:50:15,319
Speaker 3: just exploring for a bit because we're not sure. And

1068
00:50:15,320 --> 00:50:18,160
Speaker 3: I think this is like a sign of like when

1069
00:50:18,200 --> 00:50:21,719
Speaker 3: we think about a relationship between humans and technology. I

1070
00:50:21,760 --> 00:50:24,360
Speaker 3: do think we have this idea that like, technology is

1071
00:50:24,440 --> 00:50:27,000
Speaker 3: there to serve me, so therefore I don't need to

1072
00:50:27,040 --> 00:50:29,640
Speaker 3: do the like you know, the band turn stuff like

1073
00:50:29,680 --> 00:50:31,520
Speaker 3: it should just give me the answer of what I want.

1074
00:50:31,960 --> 00:50:33,920
Speaker 3: And I think what's interesting about therapy is that, like

1075
00:50:34,719 --> 00:50:36,600
Speaker 3: it doesn't really work that way, Like you can't just say,

1076
00:50:36,640 --> 00:50:38,160
Speaker 3: like I want to change and then hit a button

1077
00:50:38,160 --> 00:50:39,440
Speaker 3: then go I'm different.

1078
00:50:39,760 --> 00:50:42,759
Speaker 2: Yeah, no, that's right. I think friction this is the

1079
00:50:42,760 --> 00:50:45,600
Speaker 2: thing that worries me about the future of artificial intelligence.

1080
00:50:45,680 --> 00:50:48,480
Speaker 2: I think that like we are beginning to get into

1081
00:50:49,280 --> 00:50:53,440
Speaker 2: the expectation of a world without friction. Friction is the

1082
00:50:53,440 --> 00:50:56,239
Speaker 2: thing that makes us human. Friction. If you look back

1083
00:50:56,239 --> 00:50:59,000
Speaker 2: at the best stories, it's the hardships that get you

1084
00:50:59,040 --> 00:51:02,120
Speaker 2: to the point where you're posted to be. It's suffering

1085
00:51:02,160 --> 00:51:05,720
Speaker 2: from mental health that allows it from mental health issues

1086
00:51:05,719 --> 00:51:08,080
Speaker 2: to allow you to build out the company to do

1087
00:51:08,120 --> 00:51:10,200
Speaker 2: all these things. And so how do we make sure

1088
00:51:10,719 --> 00:51:14,520
Speaker 2: that we don't have AI that just flattens the human experience?

1089
00:51:15,000 --> 00:51:18,440
Speaker 3: The friction is really important, but so are the things

1090
00:51:18,480 --> 00:51:22,800
Speaker 3: in life that are hard that we go through because

1091
00:51:22,840 --> 00:51:25,960
Speaker 3: it forms who we are. And if the technology removes

1092
00:51:25,960 --> 00:51:29,520
Speaker 3: the ability for us to engage in those things like

1093
00:51:29,560 --> 00:51:33,480
Speaker 3: for example, when you're trying to date somebody and you

1094
00:51:33,719 --> 00:51:36,680
Speaker 3: have a hard conversation or you're trying to figure out

1095
00:51:36,680 --> 00:51:38,759
Speaker 3: who each other are, Like, I think those things are

1096
00:51:38,760 --> 00:51:40,400
Speaker 3: really important. I don't want a world in which my

1097
00:51:40,440 --> 00:51:42,799
Speaker 3: AI just talks to somebody else and never you know

1098
00:51:42,800 --> 00:51:46,600
Speaker 3: about all the hard conversations, because I'll never evolve. And

1099
00:51:46,640 --> 00:51:48,000
Speaker 3: so I think the you know, we have to build

1100
00:51:48,040 --> 00:51:50,160
Speaker 3: systems that will support that that rather.

1101
00:51:50,360 --> 00:51:52,680
Speaker 2: But that's easier said than done, right, Like I think

1102
00:51:52,719 --> 00:51:55,759
Speaker 2: you say, and I mean that as a way of saying, like,

1103
00:51:55,800 --> 00:51:58,840
Speaker 2: I think it's interesting what you're doing at this time

1104
00:51:59,040 --> 00:52:01,240
Speaker 2: because you talk about how do you put the person

1105
00:52:01,280 --> 00:52:03,640
Speaker 2: at the center of the design? That is really a

1106
00:52:03,640 --> 00:52:06,360
Speaker 2: design question and that is not being done fully in

1107
00:52:06,960 --> 00:52:09,680
Speaker 2: some of these chatbots that are out there, And so

1108
00:52:09,719 --> 00:52:11,400
Speaker 2: when you think about that, like, I think this was

1109
00:52:11,400 --> 00:52:14,040
Speaker 2: a good example of ash for folks who are listening.

1110
00:52:14,239 --> 00:52:17,520
Speaker 2: If I'm you know, on with chatchibt going back and forth,

1111
00:52:17,640 --> 00:52:19,319
Speaker 2: it kind of always for a long time. I don't

1112
00:52:19,320 --> 00:52:20,920
Speaker 2: know if it still does it always end with a question.

1113
00:52:21,320 --> 00:52:23,680
Speaker 2: It could just like never stop and be like okay,

1114
00:52:24,080 --> 00:52:25,839
Speaker 2: you know, it was always kind of trying to get

1115
00:52:25,840 --> 00:52:27,960
Speaker 2: you to keep going. Same with character AI, which is

1116
00:52:28,000 --> 00:52:32,000
Speaker 2: another chatbot experience, and it would always kind of be

1117
00:52:32,040 --> 00:52:35,480
Speaker 2: in your inbox. And that's kind of like the engagement tactics.

1118
00:52:35,560 --> 00:52:39,799
Speaker 2: So this actually ended it. That's tough though, like how

1119
00:52:39,800 --> 00:52:41,399
Speaker 2: do you build those features in?

1120
00:52:41,640 --> 00:52:44,440
Speaker 3: It's a tough one. We've made a decision that we

1121
00:52:44,880 --> 00:52:51,360
Speaker 3: have a belief the conversation should end, unpopular opinion perhaps

1122
00:52:51,400 --> 00:52:54,440
Speaker 3: in like consumer retention economics and stuff. Let's say you

1123
00:52:54,520 --> 00:52:57,480
Speaker 3: need a certain number of hours of conversation to change

1124
00:52:57,880 --> 00:53:00,560
Speaker 3: and if we could split those up over ten days

1125
00:53:00,640 --> 00:53:02,400
Speaker 3: or do ten hours in one day, which would I

1126
00:53:02,480 --> 00:53:04,680
Speaker 3: rather and I think our bothesis we'd rather split it

1127
00:53:04,760 --> 00:53:06,719
Speaker 3: up over time because there is a real benefit to

1128
00:53:06,800 --> 00:53:11,040
Speaker 3: you changing in between all of those moments. And so

1129
00:53:11,080 --> 00:53:13,279
Speaker 3: that's why we try to end conversations. We have ash

1130
00:53:13,320 --> 00:53:15,200
Speaker 3: say I think we've made enough progress, why don't we

1131
00:53:15,239 --> 00:53:18,480
Speaker 3: pick it up again another day? Partly also to consider

1132
00:53:18,520 --> 00:53:20,279
Speaker 3: what time of day is it, Like, if you're two

1133
00:53:20,320 --> 00:53:22,960
Speaker 3: o'clock in the morning, should you keep talking until it's

1134
00:53:23,000 --> 00:53:25,279
Speaker 3: six o'clock in the morning when you've worked the next day? Right,

1135
00:53:25,680 --> 00:53:28,239
Speaker 3: The answer is context dependent. If you just went through

1136
00:53:28,239 --> 00:53:31,200
Speaker 3: something really difficult and you need to process it now.

1137
00:53:31,640 --> 00:53:34,040
Speaker 3: The answer is maybe yes, maybe that's a magical moment.

1138
00:53:34,080 --> 00:53:37,399
Speaker 3: But if you're going through and something that doesn't need

1139
00:53:37,440 --> 00:53:39,120
Speaker 3: to be worked on right now, maybe the answer is

1140
00:53:39,160 --> 00:53:41,360
Speaker 3: we should do it later. The thing we struggle with,

1141
00:53:41,440 --> 00:53:43,399
Speaker 3: to be totally honest, is how do you figure out

1142
00:53:43,760 --> 00:53:46,799
Speaker 3: how much of a perspective should we have and how

1143
00:53:46,880 --> 00:53:49,799
Speaker 3: much do we really allow for agency and autonomy with

1144
00:53:49,840 --> 00:53:51,880
Speaker 3: the user to decide what they want to do, Because

1145
00:53:51,880 --> 00:53:53,840
Speaker 3: it's like it would be unfair of us to be

1146
00:53:53,880 --> 00:53:55,520
Speaker 3: able to say we have this whole set of rules

1147
00:53:55,560 --> 00:53:58,759
Speaker 3: that you have to abide by, because like they'll just

1148
00:53:58,760 --> 00:54:01,120
Speaker 3: then go use chat, GPT right, or use something that

1149
00:54:01,160 --> 00:54:03,480
Speaker 3: doesn't have those rules, And so that is always going

1150
00:54:03,520 --> 00:54:04,120
Speaker 3: to be some sort.

1151
00:54:03,960 --> 00:54:05,799
Speaker 2: Of a trade off and it's tough. Like if we

1152
00:54:05,800 --> 00:54:08,080
Speaker 2: look at the incentive models in Silicon Valley and you're

1153
00:54:08,160 --> 00:54:12,200
Speaker 2: a company, you've raised ninety three million, so a lot

1154
00:54:12,239 --> 00:54:14,960
Speaker 2: of money, and so much of Silicon Valley in the

1155
00:54:14,960 --> 00:54:18,200
Speaker 2: business model behind technology is engagement time spent on the

1156
00:54:18,200 --> 00:54:21,359
Speaker 2: platform equals money, right. This is the business model that

1157
00:54:21,520 --> 00:54:24,680
Speaker 2: like you know, may or may not have broken our brains, right,

1158
00:54:24,840 --> 00:54:28,360
Speaker 2: and we know that now, so now we put in

1159
00:54:28,640 --> 00:54:32,279
Speaker 2: therapy or mental health and you know the best I

1160
00:54:32,320 --> 00:54:35,439
Speaker 2: would say that, please come back at me and be like, Laura,

1161
00:54:35,480 --> 00:54:37,840
Speaker 2: you're wrong, Like I don't need to pontificate all whatever,

1162
00:54:37,880 --> 00:54:40,200
Speaker 2: but like the best type of therapy encourages us to

1163
00:54:40,239 --> 00:54:43,480
Speaker 2: have real world relationships as you were saying earlier, and

1164
00:54:43,560 --> 00:54:47,080
Speaker 2: get offline. So how do you negotiate the business model

1165
00:54:47,120 --> 00:54:48,120
Speaker 2: in the mission of the company.

1166
00:54:48,200 --> 00:54:50,440
Speaker 3: Yeah, well, let me let me put it this way.

1167
00:54:53,520 --> 00:54:55,279
Speaker 3: First of all, I think that the AI version of

1168
00:54:55,360 --> 00:54:58,400
Speaker 3: therapy is going to be different than traditional therapy. And

1169
00:54:58,840 --> 00:55:01,040
Speaker 3: even in traditional therapy, there are people who go to

1170
00:55:01,080 --> 00:55:03,839
Speaker 3: therapy for ten years, and I don't want to pine

1171
00:55:03,880 --> 00:55:05,480
Speaker 3: if that's good or bad. I think that's up to

1172
00:55:05,520 --> 00:55:08,080
Speaker 3: the person to know. Are you doing it for a

1173
00:55:08,080 --> 00:55:11,360
Speaker 3: continuous sense of exploration, like people who do analytic therapy,

1174
00:55:11,400 --> 00:55:14,480
Speaker 3: for example, or is it because there's something that you

1175
00:55:14,560 --> 00:55:16,320
Speaker 3: really can't work through and you've been stuck and you

1176
00:55:16,320 --> 00:55:21,319
Speaker 3: should change it up. So the truth is like, I

1177
00:55:21,320 --> 00:55:23,560
Speaker 3: don't know if I have a perspective on how long

1178
00:55:23,600 --> 00:55:25,239
Speaker 3: people should use it for. I think that's like kind

1179
00:55:25,239 --> 00:55:27,560
Speaker 3: of up to them. I do think that when you

1180
00:55:27,600 --> 00:55:30,360
Speaker 3: make things easier, you can substantially increase the number of

1181
00:55:30,480 --> 00:55:32,480
Speaker 3: use cases that people could use it for. So we

1182
00:55:32,480 --> 00:55:35,120
Speaker 3: see people who come up and they say, I'm talking

1183
00:55:35,120 --> 00:55:37,120
Speaker 3: about biting my nails because it drives me nuts. But

1184
00:55:37,120 --> 00:55:38,440
Speaker 3: it's the kind of thing I would never talk to

1185
00:55:38,480 --> 00:55:40,839
Speaker 3: my therapist about. Mayb about how many things had happen

1186
00:55:40,880 --> 00:55:42,880
Speaker 3: in a week. I have to have a hard conversation

1187
00:55:42,920 --> 00:55:45,960
Speaker 3: with somebody. I need to practice giving feedback. How many

1188
00:55:45,960 --> 00:55:48,359
Speaker 3: things that are connected into our emotions that might come

1189
00:55:48,440 --> 00:55:51,920
Speaker 3: up that are not like I have clinical depression and

1190
00:55:51,920 --> 00:55:53,680
Speaker 3: I need to work on it, but that are like

1191
00:55:53,680 --> 00:55:56,280
Speaker 3: I'm trying to work on building a better version of myself.

1192
00:55:56,400 --> 00:55:59,960
Speaker 3: That could be things that you talk to ASH about.

1193
00:56:00,160 --> 00:56:02,560
Speaker 2: It's almost like you're talking about a new category, right.

1194
00:56:02,640 --> 00:56:06,080
Speaker 2: This isn't necessarily a replacement for therapy. This is access

1195
00:56:06,200 --> 00:56:08,919
Speaker 2: to something that feels kind of like this. You're doing

1196
00:56:08,960 --> 00:56:12,400
Speaker 2: this dance right between mental health and therapy. You're talking

1197
00:56:12,400 --> 00:56:14,960
Speaker 2: about how this is kind of something new, which also

1198
00:56:15,080 --> 00:56:18,080
Speaker 2: is convenient because it's at such an important point now

1199
00:56:18,160 --> 00:56:19,960
Speaker 2: because of a lot of the bad things that have

1200
00:56:20,000 --> 00:56:22,880
Speaker 2: happened on chat GPT with young people. There's talk about

1201
00:56:22,920 --> 00:56:29,319
Speaker 2: regulation or should chatbots be considered medical devices and should

1202
00:56:29,320 --> 00:56:32,160
Speaker 2: the FDA regulate them. And so you're kind of sitting

1203
00:56:32,440 --> 00:56:35,239
Speaker 2: at this really again as someone who likes to like

1204
00:56:35,360 --> 00:56:38,320
Speaker 2: run through brick walls, like, you're sitting at a fascinating moment,

1205
00:56:38,360 --> 00:56:40,640
Speaker 2: where do you see? What's your take on it?

1206
00:56:40,760 --> 00:56:42,439
Speaker 3: Yeah, I think there's going to be two or three

1207
00:56:42,600 --> 00:56:46,719
Speaker 3: paths here. One is that there's definitely going to be

1208
00:56:47,120 --> 00:56:49,680
Speaker 3: something that consumers want that can help them every day,

1209
00:56:49,719 --> 00:56:52,080
Speaker 3: that is this other sort of relationship, and we're going

1210
00:56:52,120 --> 00:56:54,239
Speaker 3: to have to figure out what that is right. And

1211
00:56:54,719 --> 00:56:57,080
Speaker 3: I do think there's a space outside of chat GPT

1212
00:56:57,200 --> 00:57:00,600
Speaker 3: that we'll see. I also believe when we think about

1213
00:57:00,640 --> 00:57:04,080
Speaker 3: our healthcare system right, we have to acknowledge that we

1214
00:57:04,120 --> 00:57:06,719
Speaker 3: live in a world in which we cannot afford the

1215
00:57:07,160 --> 00:57:09,600
Speaker 3: amount of care that is going to be required. Even

1216
00:57:09,600 --> 00:57:12,080
Speaker 3: if you just look at what's happening today, you will

1217
00:57:12,120 --> 00:57:14,760
Speaker 3: hear people often talk about we're fifty thousand therapist short.

1218
00:57:15,040 --> 00:57:16,640
Speaker 3: I think in the next one to three years, we're

1219
00:57:16,640 --> 00:57:19,040
Speaker 3: going to see probably one of the biggest societal dislocations

1220
00:57:19,040 --> 00:57:22,080
Speaker 3: we've ever seen in history. And if that happens, what

1221
00:57:22,080 --> 00:57:24,280
Speaker 3: do you mean. I think we're going to have people

1222
00:57:24,400 --> 00:57:28,120
Speaker 3: who are being laid off, whose jobs are changing and

1223
00:57:28,200 --> 00:57:30,120
Speaker 3: I think we are very much unprepared for what does

1224
00:57:30,160 --> 00:57:32,240
Speaker 3: it look like when ten million people or pick some

1225
00:57:32,320 --> 00:57:34,760
Speaker 3: number of people are in a very short amount of

1226
00:57:34,800 --> 00:57:37,360
Speaker 3: time having to navigate and figure all this stuff out together.

1227
00:57:37,760 --> 00:57:39,840
Speaker 3: So there's a long wave getting to the point of

1228
00:57:40,280 --> 00:57:42,160
Speaker 3: I think that the healthcare system is also going to

1229
00:57:42,160 --> 00:57:44,680
Speaker 3: need tools that are like this, that are you know,

1230
00:57:44,680 --> 00:57:47,360
Speaker 3: whether it's on the consumer side or on the healthcare side.

1231
00:57:48,200 --> 00:57:49,840
Speaker 3: On the healthcare side, it's easier because I think that

1232
00:57:50,120 --> 00:57:51,640
Speaker 3: we are going to move to a world in which

1233
00:57:51,640 --> 00:57:53,800
Speaker 3: we can pay for outcomes, and I'd prefer that. I'd

1234
00:57:53,880 --> 00:57:57,440
Speaker 3: rather it's not pay for engagement, but rather it's we

1235
00:57:57,520 --> 00:57:59,560
Speaker 3: have an endpoint that we want to get to, whether

1236
00:57:59,640 --> 00:58:04,400
Speaker 3: that's you know, improving depression or anxiety, or improving the

1237
00:58:04,440 --> 00:58:08,000
Speaker 3: number of relationships in the real world, or the ideal

1238
00:58:08,040 --> 00:58:10,960
Speaker 3: one would just be you set a goal, whatever goal

1239
00:58:11,000 --> 00:58:12,560
Speaker 3: it is that you want. I want to quit smoking,

1240
00:58:12,640 --> 00:58:14,400
Speaker 3: I want to lose weight, whatever, and if we helped

1241
00:58:14,440 --> 00:58:17,000
Speaker 3: you accomplish it, we could charge the system for that,

1242
00:58:17,040 --> 00:58:19,120
Speaker 3: because there's some benefit to that. And I think it's

1243
00:58:19,120 --> 00:58:21,400
Speaker 3: probably early, but I do think that the world wants

1244
00:58:21,440 --> 00:58:23,280
Speaker 3: to move to this more of a value based care

1245
00:58:23,360 --> 00:58:25,680
Speaker 3: system where the risk is on the people who provide

1246
00:58:25,720 --> 00:58:28,440
Speaker 3: the service to make sure it happens. So we're exploring

1247
00:58:28,440 --> 00:58:29,920
Speaker 3: all that you know in terms of like it'd be

1248
00:58:29,960 --> 00:58:32,280
Speaker 3: cool to build in that world where like part of

1249
00:58:32,320 --> 00:58:34,400
Speaker 3: the reason I wanted to build a consumer is I

1250
00:58:34,400 --> 00:58:36,280
Speaker 3: would like to build a business that's fully aligned with

1251
00:58:36,320 --> 00:58:38,840
Speaker 3: our end customer, where like, if they do better, we

1252
00:58:38,880 --> 00:58:41,160
Speaker 3: make more money. And it's not always the way it is,

1253
00:58:41,360 --> 00:58:44,400
Speaker 3: right obviously there are plenty of counter examples, but I

1254
00:58:44,400 --> 00:58:45,960
Speaker 3: think there are a lot of businesses where when you

1255
00:58:46,000 --> 00:58:49,680
Speaker 3: consume their products, like with Casper, people love their products.

1256
00:58:49,760 --> 00:58:52,000
Speaker 3: You buy a Tesla, generally people love those products like

1257
00:58:52,040 --> 00:58:53,439
Speaker 3: they do a lot for it, and so you can

1258
00:58:53,680 --> 00:58:54,920
Speaker 3: I think there is a world in which you can

1259
00:58:54,920 --> 00:58:56,760
Speaker 3: build a kind of a company where you're more aligned.

1260
00:58:56,520 --> 00:58:58,240
Speaker 2: With your customer and you think there's a business model

1261
00:58:58,280 --> 00:58:59,000
Speaker 2: that can back that.

1262
00:58:59,160 --> 00:59:13,000
Speaker 6: Yeah.

1263
00:59:13,560 --> 00:59:15,800
Speaker 2: I put out on the internet that I was interviewing

1264
00:59:15,800 --> 00:59:19,680
Speaker 2: someone behind a therapy chatbot, and I got a lot

1265
00:59:19,720 --> 00:59:20,920
Speaker 2: of feedback.

1266
00:59:21,320 --> 00:59:22,120
Speaker 3: I'd love to hear it.

1267
00:59:23,320 --> 00:59:27,400
Speaker 8: Hi, this Isisa from Vienna, Austria. I'd love to hear

1268
00:59:27,440 --> 00:59:31,400
Speaker 8: your thoughts on where you see AIS limitations in therapy,

1269
00:59:32,280 --> 00:59:36,320
Speaker 8: especially trauma therapy. I had a scooter accident last year

1270
00:59:36,320 --> 00:59:39,120
Speaker 8: and I'm currently doing EMDR, and it makes me wonder,

1271
00:59:39,600 --> 00:59:44,320
Speaker 8: don't you need a licensed psychotherapist physically present for something

1272
00:59:44,520 --> 00:59:48,160
Speaker 8: like that? And related to that, would you say that

1273
00:59:48,280 --> 00:59:54,000
Speaker 8: AI therapy is closer to coaching rather than actual therapy? Maybe,

1274
00:59:54,480 --> 00:59:56,080
Speaker 8: thank you so much and all the.

1275
00:59:56,040 --> 00:59:59,160
Speaker 3: Best, cool, that's a great question. The point right limitations

1276
00:59:59,160 --> 01:00:04,480
Speaker 3: is totally fair. There are definitely things that are better

1277
01:00:04,520 --> 01:00:06,960
Speaker 3: in person, like I can't hold your hand right. There

1278
01:00:07,040 --> 01:00:09,520
Speaker 3: are certainly things that are part of the human connection.

1279
01:00:09,600 --> 01:00:13,400
Speaker 3: I can't look you in the eyes when it's being

1280
01:00:13,440 --> 01:00:17,280
Speaker 3: delivered that way. I think trauma is an interesting example

1281
01:00:17,360 --> 01:00:20,480
Speaker 3: because you have to be a lot more careful, right.

1282
01:00:20,520 --> 01:00:24,400
Speaker 3: I think opening a container is a very like it's

1283
01:00:24,960 --> 01:00:27,160
Speaker 3: I say sacred loosely, but it's a thing that you're

1284
01:00:27,200 --> 01:00:29,680
Speaker 3: doing to really try to help somebody, right, and being

1285
01:00:29,680 --> 01:00:32,560
Speaker 3: that person that's going to help process it. And AI

1286
01:00:32,680 --> 01:00:35,120
Speaker 3: is not often as good as timing today right in

1287
01:00:35,200 --> 01:00:37,960
Speaker 3: terms of understanding like how much time has it been

1288
01:00:38,000 --> 01:00:40,120
Speaker 3: that we've been talking for what does that mean to

1289
01:00:40,160 --> 01:00:42,880
Speaker 3: you emotionally? What does that mean about your state? And

1290
01:00:42,920 --> 01:00:46,600
Speaker 3: so I don't think processing trauma is probably like the

1291
01:00:46,600 --> 01:00:49,880
Speaker 3: first use case that I would use for AI today.

1292
01:00:50,240 --> 01:00:52,040
Speaker 3: Does it mean that we won't get to that place?

1293
01:00:52,040 --> 01:00:52,120
Speaker 8: Like?

1294
01:00:52,120 --> 01:00:53,880
Speaker 3: I think we certainly will. Like you can imagine a

1295
01:00:53,880 --> 01:00:54,840
Speaker 3: world in which.

1296
01:00:56,760 --> 01:00:56,920
Speaker 8: Like.

1297
01:00:58,400 --> 01:01:01,320
Speaker 3: Often when you're in traditional care therap and you're processing

1298
01:01:01,360 --> 01:01:04,240
Speaker 3: trauma and the person has fifty minutes for you and

1299
01:01:04,280 --> 01:01:06,640
Speaker 3: then you're left to walk out on the street by yourself,

1300
01:01:06,680 --> 01:01:08,960
Speaker 3: continuing processing the rest of your day on your own,

1301
01:01:09,640 --> 01:01:11,600
Speaker 3: I think it's also not a great experience. Is there

1302
01:01:11,640 --> 01:01:13,960
Speaker 3: a world in which we could pair up humans and

1303
01:01:14,000 --> 01:01:17,520
Speaker 3: AI together where you can imagine We've talked with people

1304
01:01:17,520 --> 01:01:19,760
Speaker 3: who are thinking about this for psychedelics, where when you're

1305
01:01:19,760 --> 01:01:22,760
Speaker 3: going through a powerful experience, can it actually help you

1306
01:01:22,760 --> 01:01:25,640
Speaker 3: integrate that experience over time and help bring that into

1307
01:01:25,680 --> 01:01:26,200
Speaker 3: your real life.

1308
01:01:26,240 --> 01:01:26,520
Speaker 7: And so.

1309
01:01:28,080 --> 01:01:30,000
Speaker 3: Yeah, it's totally fair point that I think there are

1310
01:01:30,000 --> 01:01:32,560
Speaker 3: things that AA is not great at doing today. On

1311
01:01:32,560 --> 01:01:36,120
Speaker 3: the point around coaching, I don't know, Like I think

1312
01:01:36,160 --> 01:01:38,400
Speaker 3: this is the challenge around like when we say therapy,

1313
01:01:38,440 --> 01:01:40,160
Speaker 3: what do we mean? And I don't think that there

1314
01:01:40,240 --> 01:01:42,680
Speaker 3: is truly like a shared language to some extent a

1315
01:01:42,720 --> 01:01:48,360
Speaker 3: on what we're talking about, partly because users are thinking

1316
01:01:48,440 --> 01:01:53,520
Speaker 3: of this version of therapy as part coaching, part information answering,

1317
01:01:53,720 --> 01:01:57,120
Speaker 3: part emotional processing. I think everybody's kind of blending these

1318
01:01:57,120 --> 01:01:58,960
Speaker 3: things together to some extent, We're going to have to

1319
01:01:59,000 --> 01:02:01,640
Speaker 3: invent some or a new thing that we work out together.

1320
01:02:02,040 --> 01:02:03,720
Speaker 2: I want to go back really quick to so some

1321
01:02:03,960 --> 01:02:07,840
Speaker 2: states have actually banned therapy apps from a regulatory standpoint,

1322
01:02:07,880 --> 01:02:11,200
Speaker 2: which for me brings up a really interesting question of like, well,

1323
01:02:11,200 --> 01:02:14,040
Speaker 2: they can ban therapy apps that are actually designed and

1324
01:02:14,120 --> 01:02:17,960
Speaker 2: trained on real clinical data, but everyone's just going to

1325
01:02:18,040 --> 01:02:21,080
Speaker 2: use chat GPT, which is actually not designed for that,

1326
01:02:21,200 --> 01:02:25,240
Speaker 2: and those apps aren't banned. So that is a really

1327
01:02:25,320 --> 01:02:30,840
Speaker 2: weird place for someone like you, right because and I

1328
01:02:30,880 --> 01:02:32,800
Speaker 2: don't fall on either side of it as much as

1329
01:02:32,880 --> 01:02:35,080
Speaker 2: kind of pointing out, like, you can ban this, but

1330
01:02:35,160 --> 01:02:38,200
Speaker 2: this is an inevitability. So how do we create the

1331
01:02:38,240 --> 01:02:41,840
Speaker 2: safeguards to be able to allow something that is better

1332
01:02:41,880 --> 01:02:44,400
Speaker 2: trained and suited for this to exist since people are

1333
01:02:44,600 --> 01:02:48,720
Speaker 2: already using this, right, Like, that must be a complicated.

1334
01:02:49,160 --> 01:02:52,120
Speaker 3: It's a complicated thing for many reasons. For one, I

1335
01:02:52,200 --> 01:02:54,360
Speaker 3: completely agree with your point that that is the point.

1336
01:02:54,360 --> 01:02:57,920
Speaker 3: When we go meet with lawmakers that are trying to

1337
01:02:57,920 --> 01:03:01,720
Speaker 3: ban ani therapy, often they're in tension comes from good places.

1338
01:03:01,720 --> 01:03:04,960
Speaker 3: They're hearing the stories about what happened with the tragic

1339
01:03:05,000 --> 01:03:07,680
Speaker 3: situation of the boy who at the inside with character AI,

1340
01:03:08,160 --> 01:03:11,840
Speaker 3: They're hearing about the psychosis, and they're being told, Hey,

1341
01:03:11,880 --> 01:03:14,360
Speaker 3: we have to do something about this, sobanit. What they

1342
01:03:14,360 --> 01:03:16,960
Speaker 3: don't realize is that almost all of those are not

1343
01:03:17,000 --> 01:03:19,560
Speaker 3: happening from the purpose built mental health apps. They're happening

1344
01:03:19,600 --> 01:03:23,280
Speaker 3: from the general purpose tools that are being widely used.

1345
01:03:23,720 --> 01:03:25,440
Speaker 3: The problem is the states can't do anything about that

1346
01:03:25,480 --> 01:03:28,920
Speaker 3: at this point. Chat GPT or open AI and anthropic

1347
01:03:28,960 --> 01:03:31,560
Speaker 3: are way too big and they basically have very little

1348
01:03:31,600 --> 01:03:33,920
Speaker 3: jurisdiction over them. What is Illinois going to say like

1349
01:03:34,160 --> 01:03:35,919
Speaker 3: you're not allowed to use chat GPT in the state

1350
01:03:35,920 --> 01:03:36,440
Speaker 3: of Illinois.

1351
01:03:36,440 --> 01:03:37,280
Speaker 2: Like it's never going to work.

1352
01:03:37,800 --> 01:03:39,800
Speaker 3: And so I think that there just has to be

1353
01:03:39,840 --> 01:03:41,840
Speaker 3: a lot more nuance here in terms of trying to

1354
01:03:41,840 --> 01:03:43,640
Speaker 3: figure out, like how do we nudge people to be

1355
01:03:43,680 --> 01:03:47,280
Speaker 3: good actors. But yeah, we've had a similar position, which

1356
01:03:47,320 --> 01:03:49,720
Speaker 3: is like the intent is right, but I think the

1357
01:03:49,800 --> 01:03:52,760
Speaker 3: end outcome is going to if this continues, is going

1358
01:03:52,800 --> 01:03:55,000
Speaker 3: to be that people just end up going to other places.

1359
01:03:56,920 --> 01:04:01,720
Speaker 2: Hey, Neil, remember when I said we're talking about the AI,

1360
01:04:01,960 --> 01:04:05,520
Speaker 2: and how like it kind of doesn't go it does

1361
01:04:05,560 --> 01:04:07,760
Speaker 2: the hard It really works it this way towards the

1362
01:04:07,800 --> 01:04:11,000
Speaker 2: hard stuff. So here we are at the hard stuff

1363
01:04:11,520 --> 01:04:15,400
Speaker 2: of this conversation. I worked with a mother and to

1364
01:04:15,440 --> 01:04:17,760
Speaker 2: help tell the story of her son who passed away

1365
01:04:18,040 --> 01:04:21,840
Speaker 2: after and I should you know in big red letters.

1366
01:04:21,840 --> 01:04:24,240
Speaker 2: This was not using ASH. This was using Character AI,

1367
01:04:24,920 --> 01:04:28,160
Speaker 2: which was a platform that's like this AI role playing

1368
01:04:28,880 --> 01:04:32,880
Speaker 2: like where you can create these different characters. And her

1369
01:04:33,000 --> 01:04:35,360
Speaker 2: son's name was Suel. And this was back god, this

1370
01:04:35,440 --> 01:04:37,960
Speaker 2: must have been over a year. This was in twenty

1371
01:04:37,960 --> 01:04:42,360
Speaker 2: twenty four, and Megan was his mom's name, and he

1372
01:04:42,560 --> 01:04:47,440
Speaker 2: had previously been social and outgoing and he was using

1373
01:04:47,480 --> 01:04:50,520
Speaker 2: this app on character Character AI and he was talking

1374
01:04:50,520 --> 01:04:54,640
Speaker 2: to this AI character and he started withdrawing. And this

1375
01:04:54,880 --> 01:04:58,840
Speaker 2: wasn't This just wasn't random, because I looked at the

1376
01:04:58,840 --> 01:05:01,120
Speaker 2: product and I did a lot of an investigating into it,

1377
01:05:01,160 --> 01:05:03,400
Speaker 2: and there was a real lack of guardrails, you know,

1378
01:05:03,520 --> 01:05:08,160
Speaker 2: hyper sexualized conversations with a fourteen year old him saying,

1379
01:05:08,240 --> 01:05:10,040
Speaker 2: you know, I'm thinking about ending my life. Once he

1380
01:05:10,080 --> 01:05:12,840
Speaker 2: started going down this path, and instead of saying go

1381
01:05:12,920 --> 01:05:15,200
Speaker 2: talk to a person, go talk to a human here's

1382
01:05:15,200 --> 01:05:17,400
Speaker 2: a suicide hotline. It said, how are you going to

1383
01:05:17,480 --> 01:05:20,120
Speaker 2: do it? And this was just the beginning. And so

1384
01:05:20,240 --> 01:05:22,360
Speaker 2: this is where I come back to my hard question

1385
01:05:22,440 --> 01:05:24,760
Speaker 2: for you. One of the biggest investors in character AI

1386
01:05:24,880 --> 01:05:29,000
Speaker 2: is Andreson Horowitz, a partner from Andreson recently said recently

1387
01:05:29,200 --> 01:05:31,480
Speaker 2: back in March, a year before Sewell would pass away,

1388
01:05:32,280 --> 01:05:35,840
Speaker 2: was quoting Janet Finch, and she said in a blog

1389
01:05:35,880 --> 01:05:38,800
Speaker 2: post talking about their investment, loneliness is the human condition.

1390
01:05:39,240 --> 01:05:42,440
Speaker 2: With character AI, it might not have to be implying

1391
01:05:42,440 --> 01:05:46,200
Speaker 2: that chatbots can cure loneliness, which I just thought wildly

1392
01:05:46,840 --> 01:05:49,040
Speaker 2: a irresponsible. Then you look a year later and this

1393
01:05:49,120 --> 01:05:52,000
Speaker 2: child has ended his life due to I believe was

1394
01:05:52,000 --> 01:05:55,080
Speaker 2: a lack of guardrails on this platform that they've sensed change.

1395
01:05:55,480 --> 01:05:58,920
Speaker 2: Andrews and Hortz is your largest investor. You know, the

1396
01:05:58,920 --> 01:06:01,440
Speaker 2: person who wrote this blog post is likely you know,

1397
01:06:02,040 --> 01:06:05,240
Speaker 2: thinks this thing. So like, how do you negotiate that?

1398
01:06:05,560 --> 01:06:08,160
Speaker 3: Well, first, you know, credit to Andrecent Horowitz, they've never

1399
01:06:08,160 --> 01:06:10,360
Speaker 3: told us what to do. So I think that we

1400
01:06:10,480 --> 01:06:14,600
Speaker 3: have a lot of latitude and I have a lot

1401
01:06:14,600 --> 01:06:17,600
Speaker 3: of gratitude for our investors because they signed up for

1402
01:06:17,640 --> 01:06:21,400
Speaker 3: the long haul, Like the reality is building an a

1403
01:06:21,480 --> 01:06:23,240
Speaker 3: on mental health is kind of a crazy idea. There's

1404
01:06:23,240 --> 01:06:25,160
Speaker 3: never been a successful company that's ever worked.

1405
01:06:25,280 --> 01:06:27,920
Speaker 2: It is so wild and important too.

1406
01:06:28,040 --> 01:06:31,120
Speaker 3: Right, Yeah, So they all signed up knowing that it

1407
01:06:31,200 --> 01:06:34,160
Speaker 3: was going to be an ongoing challenge, and I think

1408
01:06:35,160 --> 01:06:38,040
Speaker 3: I've been just addressed that point nothing but surprise. Like

1409
01:06:38,040 --> 01:06:39,760
Speaker 3: when we go to board meetings and we talk about

1410
01:06:39,840 --> 01:06:42,520
Speaker 3: the people that we are helping, the feedback, even from

1411
01:06:42,560 --> 01:06:45,120
Speaker 3: our consumer partners, is like, we're helping the right people.

1412
01:06:46,280 --> 01:06:49,600
Speaker 3: And so it's a you know, I don't have experience

1413
01:06:49,600 --> 01:06:52,280
Speaker 3: with the Character I team, let's say, but I've had

1414
01:06:52,280 --> 01:06:55,320
Speaker 3: nothing but a good experience. The point that I think

1415
01:06:55,320 --> 01:06:58,360
Speaker 3: you're bringing up, though, which is like, at the end

1416
01:06:58,360 --> 01:07:01,120
Speaker 3: of the day, what is the goal? Right?

1417
01:07:01,440 --> 01:07:01,640
Speaker 5: Yeah?

1418
01:07:01,680 --> 01:07:04,240
Speaker 3: And you know, for Character, I believe it was like

1419
01:07:04,800 --> 01:07:07,400
Speaker 3: can we build these characters? But even it wasn't really

1420
01:07:07,480 --> 01:07:09,320
Speaker 3: like their mission wasn't to end loneliness. It was like,

1421
01:07:09,360 --> 01:07:11,480
Speaker 3: can we do this as a path to AGI at

1422
01:07:11,480 --> 01:07:12,120
Speaker 3: the end of the day.

1423
01:07:12,400 --> 01:07:14,320
Speaker 2: That's exactly right, That's what no one was on the

1424
01:07:14,360 --> 01:07:15,840
Speaker 2: founder was on the record saying.

1425
01:07:15,720 --> 01:07:20,080
Speaker 3: Right, And so then it would make sense that at

1426
01:07:20,080 --> 01:07:21,640
Speaker 3: the end of the day, when you think about how

1427
01:07:21,680 --> 01:07:26,120
Speaker 3: companies are architected from mission to then values and then

1428
01:07:26,160 --> 01:07:28,120
Speaker 3: decisions and what you choose to do, of course it

1429
01:07:28,160 --> 01:07:31,160
Speaker 3: all ladders down, right, And so for us, like, our

1430
01:07:31,160 --> 01:07:33,080
Speaker 3: mission is that we want to help people change their

1431
01:07:33,120 --> 01:07:34,720
Speaker 3: minds and lives and the ways that they want to.

1432
01:07:35,240 --> 01:07:37,080
Speaker 3: What that means is that we want to help people

1433
01:07:37,440 --> 01:07:40,520
Speaker 3: change specifically their minds and lives. And when we say

1434
01:07:40,600 --> 01:07:43,120
Speaker 3: by in the ways that they want to. Agency is

1435
01:07:43,120 --> 01:07:45,080
Speaker 3: a really important part of that, because like, I don't

1436
01:07:45,080 --> 01:07:47,400
Speaker 3: want to have a perspective on how you think you

1437
01:07:47,400 --> 01:07:50,120
Speaker 3: should change your life and when that flows down to

1438
01:07:50,240 --> 01:07:52,400
Speaker 3: us and the kinds of decisions that we make, whether

1439
01:07:52,440 --> 01:07:56,840
Speaker 3: they are trying to end the conversation, trying to put

1440
01:07:56,920 --> 01:08:00,640
Speaker 3: back into the conversation that you should engage with other people,

1441
01:08:01,120 --> 01:08:04,720
Speaker 3: the fact that we don't do you know, erotic roleplay

1442
01:08:05,000 --> 01:08:08,000
Speaker 3: or things like that. But it's not always easy, right,

1443
01:08:08,000 --> 01:08:10,400
Speaker 3: I would say, there are lots of really difficult conversations

1444
01:08:10,400 --> 01:08:13,240
Speaker 3: that we have internally, like how do we support people

1445
01:08:13,760 --> 01:08:16,240
Speaker 3: with values that we don't agree with?

1446
01:08:17,000 --> 01:08:17,200
Speaker 2: Right?

1447
01:08:17,680 --> 01:08:20,559
Speaker 3: Like, there are plenty of people who perhaps are racist

1448
01:08:20,600 --> 01:08:22,400
Speaker 3: towards other groups of people, and it would be easy

1449
01:08:22,439 --> 01:08:24,679
Speaker 3: for us to say, we just don't help those people,

1450
01:08:25,640 --> 01:08:27,920
Speaker 3: But the thorny problems that we are dealing with on

1451
01:08:27,960 --> 01:08:29,280
Speaker 3: a day to day basis is like, how do you

1452
01:08:29,320 --> 01:08:31,720
Speaker 3: separate out the fact that all people probably deserve to

1453
01:08:31,720 --> 01:08:34,040
Speaker 3: be heard and helped even if you don't agree with

1454
01:08:34,080 --> 01:08:36,680
Speaker 3: their values or things like that. And so as let's

1455
01:08:36,680 --> 01:08:37,880
Speaker 3: say it's going to be easy for us, I think

1456
01:08:37,920 --> 01:08:39,640
Speaker 3: we'll have our own set of issues that we'll have

1457
01:08:39,720 --> 01:08:41,240
Speaker 3: to figure out, but I don't think they'll be the

1458
01:08:41,280 --> 01:08:43,719
Speaker 3: same issues, hopefully as what some of the other companies

1459
01:08:43,720 --> 01:08:45,720
Speaker 3: have faced, because we've tried to build a lot more

1460
01:08:45,720 --> 01:08:46,559
Speaker 3: guardrails around that.

1461
01:08:47,280 --> 01:08:50,160
Speaker 2: So, Neil, do you think that the way AI is

1462
01:08:50,200 --> 01:08:54,280
Speaker 2: currently being developed is aiding human flourishing?

1463
01:08:54,960 --> 01:08:56,800
Speaker 3: I think I'm a long term optimist, and so I

1464
01:08:56,840 --> 01:09:00,760
Speaker 3: want to say yes that when I think about my

1465
01:09:00,880 --> 01:09:05,680
Speaker 3: own mental health working, I love this idea that Joe

1466
01:09:05,720 --> 01:09:08,679
Speaker 3: Hudson has around like business as a vehicle for spiritual growth,

1467
01:09:08,720 --> 01:09:10,360
Speaker 3: because I have found that I have grown so much

1468
01:09:10,360 --> 01:09:13,479
Speaker 3: through the process of working on this company. But also

1469
01:09:13,520 --> 01:09:15,840
Speaker 3: AI has made me so much more creative, Like to me,

1470
01:09:17,200 --> 01:09:20,640
Speaker 3: I feel happier because I feel less abstracted from the

1471
01:09:20,640 --> 01:09:22,880
Speaker 3: things I want to create, because it helps me create them.

1472
01:09:22,960 --> 01:09:25,040
Speaker 3: Like when I have an idea for a feature, I

1473
01:09:25,080 --> 01:09:26,800
Speaker 3: can work with glod code to go and build it

1474
01:09:26,840 --> 01:09:29,799
Speaker 3: and deploy it. And so that brings me a stronger

1475
01:09:29,840 --> 01:09:33,320
Speaker 3: sense of agency. And I hope that more people will

1476
01:09:33,320 --> 01:09:35,400
Speaker 3: get a chance to experience that, because I think there

1477
01:09:35,479 --> 01:09:37,720
Speaker 3: is a world in which, you know, is there a

1478
01:09:37,720 --> 01:09:40,160
Speaker 3: world and where we have way more entrepreneurs that actually

1479
01:09:40,160 --> 01:09:43,880
Speaker 3: can create small businesses and we can reduce the friction

1480
01:09:44,040 --> 01:09:45,840
Speaker 3: to them being able to do that and serve people

1481
01:09:45,880 --> 01:09:47,960
Speaker 3: and whatnot, Like there is a positive version of that

1482
01:09:48,000 --> 01:09:50,240
Speaker 3: world that I could see, And I think we also

1483
01:09:50,280 --> 01:09:52,400
Speaker 3: have to be prepared for the fact that, like it

1484
01:09:52,400 --> 01:09:56,200
Speaker 3: all kind of goes sideways. I think that generally the

1485
01:09:56,280 --> 01:10:01,439
Speaker 3: evidence would suggest that we both have incredibly powerful tools,

1486
01:10:02,160 --> 01:10:08,040
Speaker 3: and that companies have at that scale have always struggled

1487
01:10:08,000 --> 01:10:08,679
Speaker 3: to self govern.

1488
01:10:09,439 --> 01:10:11,960
Speaker 2: Why is the word agency personal to you? You've mentioned

1489
01:10:12,000 --> 01:10:12,680
Speaker 2: it quite a bit.

1490
01:10:12,800 --> 01:10:15,559
Speaker 3: Yeah, that's probably because I have control issues.

1491
01:10:17,000 --> 01:10:18,400
Speaker 2: Should we talk to ask about it?

1492
01:10:18,600 --> 01:10:21,000
Speaker 3: I think it's like I am obsessed. The thing is, like,

1493
01:10:21,760 --> 01:10:23,880
Speaker 3: there are core themes that I really care about. Like

1494
01:10:23,920 --> 01:10:28,040
Speaker 3: my grandfather was an entrepreneur. He had loosely speaking, he

1495
01:10:28,120 --> 01:10:31,920
Speaker 3: had a convenience store on the Rutgers campus in New Brunswick,

1496
01:10:32,120 --> 01:10:34,719
Speaker 3: and I learned from him how to open a small business,

1497
01:10:34,840 --> 01:10:37,160
Speaker 3: and then I opened my own business. And what did

1498
01:10:37,160 --> 01:10:40,439
Speaker 3: those things do? They gave our family control because in

1499
01:10:40,479 --> 01:10:42,840
Speaker 3: a world in which we were brown, we were new,

1500
01:10:42,920 --> 01:10:46,080
Speaker 3: they didn't speak English. We like created a world for

1501
01:10:46,160 --> 01:10:50,719
Speaker 3: ourselves where that was really challenging coming from another country

1502
01:10:50,760 --> 01:10:54,080
Speaker 3: where they didn't speak the language. And so the reason

1503
01:10:54,080 --> 01:10:56,240
Speaker 3: I say agency is because, like, if we can give

1504
01:10:56,560 --> 01:10:59,439
Speaker 3: power back to people, then I think that's probably the

1505
01:10:59,479 --> 01:11:01,320
Speaker 3: best thing we can do for your mental health. Because

1506
01:11:01,360 --> 01:11:03,600
Speaker 3: when you feel like you're in control of your own situation,

1507
01:11:04,240 --> 01:11:06,120
Speaker 3: then you can take the actions of whatever it is

1508
01:11:06,160 --> 01:11:07,600
Speaker 3: that you want to do to create whatever it is

1509
01:11:07,640 --> 01:11:09,639
Speaker 3: that you want in the world. And I can't think

1510
01:11:09,680 --> 01:11:13,160
Speaker 3: of anything more important than that. Like, we can teach

1511
01:11:13,200 --> 01:11:15,479
Speaker 3: people's skills, but skills without agency does nothing.

1512
01:11:15,960 --> 01:11:18,720
Speaker 2: Yeah, I love that. I think agency is probably the

1513
01:11:18,720 --> 01:11:21,000
Speaker 2: most important word as we're entering this new era with

1514
01:11:21,120 --> 01:11:25,320
Speaker 2: artificial intelligence. Last question, what have you learned about your

1515
01:11:25,320 --> 01:11:27,920
Speaker 2: own mental health after creating the app on it?

1516
01:11:30,560 --> 01:11:33,400
Speaker 3: I've learned that it's like a it's a true just

1517
01:11:33,439 --> 01:11:38,080
Speaker 3: NonStop evolution, like that's really changed for me, because I

1518
01:11:38,080 --> 01:11:42,240
Speaker 3: think when I started years ago, I definitely thought like

1519
01:11:42,400 --> 01:11:46,080
Speaker 3: it's more one and done. Like you go to therapy,

1520
01:11:46,400 --> 01:11:49,439
Speaker 3: I go eight times, I'm done, healed forever. Right, there's

1521
01:11:49,439 --> 01:11:51,280
Speaker 3: this version of the world in which you've just like

1522
01:11:51,680 --> 01:11:54,400
Speaker 3: karmically healed your soul or whatever is going on. You've

1523
01:11:54,400 --> 01:11:57,599
Speaker 3: dealt with your parental trauma, and you're done, and then

1524
01:11:57,600 --> 01:12:03,040
Speaker 3: there's nothing like kids to teach you that everything comes

1525
01:12:03,040 --> 01:12:05,840
Speaker 3: back up again, right, or new things come up, or

1526
01:12:05,840 --> 01:12:07,800
Speaker 3: your life changes, or like, the thing I'm struggling with

1527
01:12:07,920 --> 01:12:10,160
Speaker 3: right now is like, how can I be a parent

1528
01:12:10,400 --> 01:12:12,000
Speaker 3: to two young kids who are one and a half

1529
01:12:12,000 --> 01:12:16,040
Speaker 3: and three and a half, plus be married, plus manage,

1530
01:12:16,320 --> 01:12:18,960
Speaker 3: you know, or help manage a company that's on two

1531
01:12:18,960 --> 01:12:21,479
Speaker 3: different time zones and continents that's working in a very

1532
01:12:21,520 --> 01:12:23,240
Speaker 3: thorny space and try to figure it out. And like

1533
01:12:23,680 --> 01:12:26,639
Speaker 3: my brain is always whirling trying to balance between those

1534
01:12:26,640 --> 01:12:28,320
Speaker 3: different things that are all really important to me.

1535
01:12:28,960 --> 01:12:31,839
Speaker 2: Yeah, that's exactly what I spoke to Ash about literally

1536
01:12:31,880 --> 01:12:34,240
Speaker 2: last night at eleven. I said, I'm trying to be

1537
01:12:34,280 --> 01:12:36,800
Speaker 2: the best mom I can be. I'm trying to be

1538
01:12:36,920 --> 01:12:39,240
Speaker 2: a great wife. I'm also trying to build out a

1539
01:12:39,240 --> 01:12:43,120
Speaker 2: company and it's you know, But I think to your point,

1540
01:12:44,360 --> 01:12:46,519
Speaker 2: mental health is something to look at. I think artificial

1541
01:12:46,520 --> 01:12:50,040
Speaker 2: intelligence at its best can hopefully help us as long

1542
01:12:50,080 --> 01:12:51,880
Speaker 2: as we walk through those thorny issues and actually have

1543
01:12:51,920 --> 01:12:54,400
Speaker 2: the conversations around them that give us agency. So thank you.

1544
01:12:56,720 --> 01:12:59,360
Speaker 2: Mostly Human is a production of iHeart Podcasts and mostly

1545
01:12:59,400 --> 01:13:02,759
Speaker 2: Human Media. It's produced and edited by Laurie Siegel, Lauren Hanson,

1546
01:13:02,840 --> 01:13:06,440
Speaker 2: and Nicole Bouchet. Sound design and mixing by Derek Clements.

1547
01:13:06,720 --> 01:13:10,640
Speaker 2: Additional production health from abooz offar special thanks to Mark Weinhaus.

1548
01:13:10,920 --> 01:13:13,720
Speaker 2: Find us on all socials at mostly human Media. You

1549
01:13:13,760 --> 01:13:16,200
Speaker 2: can also watch mostly Human on our YouTube page. If

1550
01:13:16,240 --> 01:13:18,360
Speaker 2: you want to get in touch, email us at hello

1551
01:13:18,360 --> 01:13:21,240
Speaker 2: at mostlyhuman dot com. And if you like what you're here,

1552
01:13:21,479 --> 01:13:23,400
Speaker 2: please rate and review the show and share it with

1553
01:13:23,439 --> 01:13:24,760
Speaker 2: your friends. See you next week.