1
00:00:02,720 --> 00:00:16,520
Speaker 1: Bloomberg Audio Studios, Podcasts, radio News.

2
00:00:18,720 --> 00:00:21,720
Speaker 2: Hello and welcome to another episode of The Odd Lots Podcast.

3
00:00:21,800 --> 00:00:24,000
Speaker 3: I'm Jill Wisenthal and I'm Tracy Alloway.

4
00:00:24,040 --> 00:00:26,119
Speaker 2: You know what I find kind of weird, Tracy.

5
00:00:25,920 --> 00:00:27,160
Speaker 3: The list could be long.

6
00:00:27,280 --> 00:00:31,960
Speaker 2: Joe the years twenty twenty five. Yes, and philosophers still

7
00:00:32,000 --> 00:00:34,640
Speaker 2: don't have a good answer on the origin of consciousness.

8
00:00:34,680 --> 00:00:36,239
Speaker 2: It's like, come on, what have you been doing all

9
00:00:36,280 --> 00:00:38,159
Speaker 2: this time? It's like, how long are we going to

10
00:00:38,280 --> 00:00:41,800
Speaker 2: keep funding these philosophy departments, et cetera. If they're still

11
00:00:41,840 --> 00:00:44,640
Speaker 2: working on what, to my mind is that they should

12
00:00:44,640 --> 00:00:49,199
Speaker 2: have solved it and move on, seriously, like get the

13
00:00:49,280 --> 00:00:52,519
Speaker 2: answer already, where does consciousness come from? Then let's move on.

14
00:00:52,600 --> 00:00:55,280
Speaker 2: I said, they're still arguing these what to my mind

15
00:00:55,320 --> 00:00:59,640
Speaker 2: seem like very basic questions in philosophy, Like they're like,

16
00:00:59,720 --> 00:01:01,920
Speaker 2: ask all the same stuff that they've been talking about forever.

17
00:01:02,000 --> 00:01:03,920
Speaker 2: How to be a good person, what does it mean

18
00:01:04,000 --> 00:01:06,600
Speaker 2: to have a moral way of life? Where does consciousness

19
00:01:06,640 --> 00:01:09,360
Speaker 2: come from? Why do we have moral intuitions? Et cetera.

20
00:01:09,440 --> 00:01:11,480
Speaker 2: It's like, move on, like get the answer.

21
00:01:11,680 --> 00:01:13,119
Speaker 3: Wait if you want them to move on or get

22
00:01:13,160 --> 00:01:13,959
Speaker 3: the answer.

23
00:01:14,040 --> 00:01:15,960
Speaker 2: Get the answer. So that you can move on like

24
00:01:16,000 --> 00:01:17,119
Speaker 2: they've been worried on to.

25
00:01:17,040 --> 00:01:20,720
Speaker 3: What those are? Any questions, Joe, I know, move on.

26
00:01:20,680 --> 00:01:23,120
Speaker 2: Like answer the questions already. It's like, you know, if

27
00:01:23,600 --> 00:01:26,560
Speaker 2: like scientists were still debating like the speed of gravity

28
00:01:26,640 --> 00:01:29,000
Speaker 2: or the speed of light, like they answered these questions

29
00:01:29,120 --> 00:01:30,560
Speaker 2: and they moved on, and or do it work?

30
00:01:30,680 --> 00:01:32,920
Speaker 3: You're out the foundational elements of what it means to

31
00:01:32,920 --> 00:01:34,959
Speaker 3: be humans so that we can move on to more

32
00:01:34,959 --> 00:01:35,760
Speaker 3: important things.

33
00:01:35,840 --> 00:01:38,080
Speaker 2: Yes, or wrap it up as a field if after

34
00:01:38,160 --> 00:01:41,080
Speaker 2: two thousand years of the existence of philosophy they're still

35
00:01:41,080 --> 00:01:43,399
Speaker 2: working on these things, Like come on, I have.

36
00:01:43,400 --> 00:01:45,679
Speaker 3: A sneaking suspicion that we're going to be asking some

37
00:01:45,760 --> 00:01:48,440
Speaker 3: of these questions for a very long time. Joe, despite

38
00:01:48,440 --> 00:01:49,640
Speaker 3: your frustration.

39
00:01:49,360 --> 00:01:51,600
Speaker 2: The whole the whole field is fraudulent. That's what I

40
00:01:51,640 --> 00:01:53,920
Speaker 2: was saying. No, no, I don't necessarily believe that, but

41
00:01:53,960 --> 00:01:55,640
Speaker 2: it's like, all right, guys, let's move it on.

42
00:01:55,840 --> 00:01:56,000
Speaker 4: You know.

43
00:01:56,040 --> 00:01:58,640
Speaker 2: We did that episode several weeks ago with Josh Wolf,

44
00:01:58,720 --> 00:02:02,040
Speaker 2: the venture capitalist. Hey talk about AI and he threw

45
00:02:02,160 --> 00:02:04,720
Speaker 2: in there at the end something that had been kind

46
00:02:04,720 --> 00:02:06,840
Speaker 2: of on my radar but barely. He's like, oh yeah,

47
00:02:06,880 --> 00:02:09,120
Speaker 2: some people are talking about like AI rights or AI

48
00:02:09,160 --> 00:02:11,280
Speaker 2: welfare is, if you know, like the same way we

49
00:02:11,320 --> 00:02:14,880
Speaker 2: talk about animal welfare, right, And I thought to myself, like,

50
00:02:15,400 --> 00:02:17,960
Speaker 2: America is such a weird place that this is going

51
00:02:18,000 --> 00:02:19,920
Speaker 2: to be a huge issue in a few years, Like,

52
00:02:19,960 --> 00:02:21,840
Speaker 2: I bet this is going to be an enormous topic

53
00:02:21,880 --> 00:02:22,440
Speaker 2: of the future.

54
00:02:22,600 --> 00:02:24,560
Speaker 3: I think it absolutely will. So I'll say a couple

55
00:02:24,600 --> 00:02:26,560
Speaker 3: of things. First off, I think, you know, when it

56
00:02:26,560 --> 00:02:28,840
Speaker 3: comes to animal welfare and human welfare, there's still a

57
00:02:28,880 --> 00:02:31,280
Speaker 3: lot of work to be done on those categories, certainly,

58
00:02:31,760 --> 00:02:35,400
Speaker 3: But I also think in the meantime, AI rights is

59
00:02:35,440 --> 00:02:38,400
Speaker 3: going to be a really interesting and potentially important subject.

60
00:02:38,880 --> 00:02:41,040
Speaker 3: I'm going to sound like a total nerd to you. Yeah, yeah,

61
00:02:41,600 --> 00:02:43,600
Speaker 3: I think I mentioned this before, but I spent a

62
00:02:43,680 --> 00:02:46,760
Speaker 3: large chunk of my middle school years playing one of

63
00:02:46,800 --> 00:02:49,720
Speaker 3: the first artificial life games that ever came out, which

64
00:02:49,800 --> 00:02:53,720
Speaker 3: is Creatures. And you raise these little like aliens and

65
00:02:53,760 --> 00:02:58,520
Speaker 3: you genetically modify them and breed them, and they have feelings,

66
00:02:58,840 --> 00:03:02,200
Speaker 3: or you know, at least they had semblance of simulated feelings,

67
00:03:02,400 --> 00:03:05,200
Speaker 3: and you could see like electrical impulses in your brains

68
00:03:05,200 --> 00:03:08,600
Speaker 3: and stuff. The game got really weird because part of

69
00:03:08,600 --> 00:03:12,120
Speaker 3: it was basically like eugenics and breeding the best alien

70
00:03:12,160 --> 00:03:14,120
Speaker 3: that you could, which meant that you had to call

71
00:03:14,440 --> 00:03:17,239
Speaker 3: some of the existing beings. Anyway, what I'm trying to

72
00:03:17,240 --> 00:03:21,760
Speaker 3: get at is I have complicated feelings about AI rights.

73
00:03:21,800 --> 00:03:23,320
Speaker 2: Well, let me ask you a question. Do you think

74
00:03:23,400 --> 00:03:27,120
Speaker 2: those whatever is in the game were conscious? Did you

75
00:03:27,160 --> 00:03:28,600
Speaker 2: think they had feelings?

76
00:03:28,880 --> 00:03:32,359
Speaker 3: Here's what I would say, inasmuch as human beings are

77
00:03:32,560 --> 00:03:37,480
Speaker 3: a system of electrical impulses and chemicals, I could see

78
00:03:37,520 --> 00:03:40,000
Speaker 3: someone making the argument that this is, you know, a

79
00:03:40,040 --> 00:03:45,240
Speaker 3: computational system full of similar electrical impulses, maybe not chemicals.

80
00:03:45,440 --> 00:03:46,440
Speaker 2: Did you feel bad?

81
00:03:46,680 --> 00:03:47,280
Speaker 3: I felt bad?

82
00:03:47,360 --> 00:03:51,120
Speaker 2: Really? Yeah, like when one of the aliens you had

83
00:03:51,160 --> 00:03:51,720
Speaker 2: to call them?

84
00:03:51,880 --> 00:03:52,320
Speaker 5: Yeah?

85
00:03:52,520 --> 00:03:54,880
Speaker 2: Interesting, Okay, Well, in.

86
00:03:54,840 --> 00:03:56,840
Speaker 3: The name of breeding a better alien.

87
00:03:56,960 --> 00:03:59,160
Speaker 2: Well, you know what, now that we have these AI

88
00:03:59,240 --> 00:04:02,840
Speaker 2: systems that they can completely communicate like humans, but actually,

89
00:04:02,880 --> 00:04:04,800
Speaker 2: if we're being honest, better than most humans. I mean,

90
00:04:04,840 --> 00:04:07,119
Speaker 2: they can certain write better, far better than most humans,

91
00:04:07,240 --> 00:04:09,880
Speaker 2: there's going to be more people thinking along the lines

92
00:04:09,880 --> 00:04:12,080
Speaker 2: of what you think, which is maybe they have some

93
00:04:12,240 --> 00:04:15,880
Speaker 2: sort of sentience, maybe they're what philosophers call moral patients.

94
00:04:15,960 --> 00:04:18,000
Speaker 3: Well, one other thing I would say is there is

95
00:04:18,040 --> 00:04:20,679
Speaker 3: a human element to all of this as well, because

96
00:04:20,680 --> 00:04:24,479
Speaker 3: you see people getting very attached to true certain AI models,

97
00:04:24,480 --> 00:04:26,880
Speaker 3: and then when the model gets upgraded or whatever, they

98
00:04:27,000 --> 00:04:30,200
Speaker 3: lose the personality that they've trained into model and they

99
00:04:30,200 --> 00:04:33,960
Speaker 3: get really upset. So it's of interest for many reasons.

100
00:04:34,240 --> 00:04:34,560
Speaker 1: It is.

101
00:04:34,839 --> 00:04:37,080
Speaker 2: So we really do have the perfect guest. I really

102
00:04:37,080 --> 00:04:38,839
Speaker 2: do think this is gonna be a much bigger topic

103
00:04:38,920 --> 00:04:41,359
Speaker 2: in the future because people are people, and when things

104
00:04:41,400 --> 00:04:43,839
Speaker 2: talk like people, they probably assign them, you know, they

105
00:04:43,920 --> 00:04:46,120
Speaker 2: fall in love with them in many cases or whatever,

106
00:04:46,360 --> 00:04:49,760
Speaker 2: and so they might start thinking that, well, AI welfare,

107
00:04:49,839 --> 00:04:52,599
Speaker 2: AI rights, whatever, the same way we talk about animals

108
00:04:52,680 --> 00:04:54,600
Speaker 2: should be a consideration. And there are actually a lot

109
00:04:54,600 --> 00:04:56,800
Speaker 2: of people already working on these questions and trying to

110
00:04:56,800 --> 00:04:59,200
Speaker 2: figure out what's going on. We're gonna be talking to

111
00:04:59,240 --> 00:05:01,600
Speaker 2: one of them. We're gonna be talking to Larissa Schiavo.

112
00:05:01,760 --> 00:05:04,479
Speaker 2: She does comms and events for elios Ai, which does

113
00:05:04,600 --> 00:05:09,599
Speaker 2: research on AI consciousness and welfare. So literally the perfect guest. So, Larissa,

114
00:05:09,640 --> 00:05:11,360
Speaker 2: thank you so much for coming on odd lots.

115
00:05:11,400 --> 00:05:12,359
Speaker 4: Yeah, thank you for having me.

116
00:05:12,600 --> 00:05:15,560
Speaker 2: What don't you tell us Eliosai? What is the gist

117
00:05:15,600 --> 00:05:18,800
Speaker 2: of this organization's work? What is your work. What are

118
00:05:18,800 --> 00:05:19,400
Speaker 2: the goals here?

119
00:05:19,560 --> 00:05:21,960
Speaker 4: Yeah, so Elios, we're a small team, but we're really

120
00:05:22,000 --> 00:05:25,799
Speaker 4: focused on figuring out if, when, and how we should

121
00:05:25,880 --> 00:05:29,160
Speaker 4: care about AI systems for their own sake. Okay, this

122
00:05:29,720 --> 00:05:32,720
Speaker 4: basically means looking at you know, are they conscious, are

123
00:05:32,720 --> 00:05:34,400
Speaker 4: they likely to be conscious? What are the things we

124
00:05:34,440 --> 00:05:36,920
Speaker 4: need to look for in a conscious AI system? As

125
00:05:36,920 --> 00:05:40,279
Speaker 4: well as figuring out how to live, work, maybe love

126
00:05:40,760 --> 00:05:44,279
Speaker 4: AI systems as they sort of change and evolve over time.

127
00:05:44,960 --> 00:05:47,120
Speaker 3: How did the group actually come together? Because I get

128
00:05:47,120 --> 00:05:51,560
Speaker 3: the sense, you know, big AI developers they publish system

129
00:05:51,640 --> 00:05:54,800
Speaker 3: cards and welfare reports occasionally for their models, but I

130
00:05:54,839 --> 00:05:57,520
Speaker 3: get the sense that, you know, it's sort of a

131
00:05:57,640 --> 00:05:59,840
Speaker 3: side topic for them. So I'm very curious how in

132
00:06:00,080 --> 00:06:03,200
Speaker 3: organization that's focused on this particular issue came into being.

133
00:06:03,640 --> 00:06:07,120
Speaker 4: Yeah, so we started. We put together this paper called

134
00:06:07,240 --> 00:06:11,159
Speaker 4: Consciousness and AI, or my boss Rob and then Patrick

135
00:06:11,200 --> 00:06:13,440
Speaker 4: who's a researcher with Elios. We're a very small team,

136
00:06:13,600 --> 00:06:15,919
Speaker 4: put together this paper called Consciousness and AI alongside a

137
00:06:15,920 --> 00:06:20,080
Speaker 4: bunch of consciousness scientists and researchers in that field who

138
00:06:20,120 --> 00:06:23,080
Speaker 4: mostly think about humans, and put together a paper that

139
00:06:23,160 --> 00:06:25,320
Speaker 4: sort of ran down this list of, Hey, here's kind

140
00:06:25,320 --> 00:06:28,000
Speaker 4: of like a checklist of things that we might want

141
00:06:28,040 --> 00:06:32,040
Speaker 4: to look for in a AI system that's conscious, right,

142
00:06:32,080 --> 00:06:34,520
Speaker 4: and broadly, when we say conscious, we're talking about sort

143
00:06:34,560 --> 00:06:36,800
Speaker 4: of like is there something it is like to be

144
00:06:37,160 --> 00:06:37,960
Speaker 4: an AI system?

145
00:06:38,040 --> 00:06:38,160
Speaker 2: Right?

146
00:06:38,240 --> 00:06:40,799
Speaker 4: The classic what is it like to be a bad system?

147
00:06:41,440 --> 00:06:45,840
Speaker 4: So kind of taking this rough list of best guesses

148
00:06:45,880 --> 00:06:47,640
Speaker 4: as to what we might want to look for in

149
00:06:47,680 --> 00:06:51,600
Speaker 4: terms of a conscious AI and then that sort of

150
00:06:51,800 --> 00:06:54,320
Speaker 4: was the sort of origin of this. And then last

151
00:06:54,360 --> 00:06:57,279
Speaker 4: year there was a paper called Taking AO Welfare Seriously

152
00:06:57,560 --> 00:07:00,200
Speaker 4: that basically goes into further detail about how we should,

153
00:07:00,400 --> 00:07:03,479
Speaker 4: as the title made, so just take this seriously basically,

154
00:07:03,480 --> 00:07:05,559
Speaker 4: how to sort of think about this, how to start

155
00:07:05,560 --> 00:07:09,600
Speaker 4: to develop a sort of research program focused on figuring

156
00:07:09,640 --> 00:07:14,720
Speaker 4: out if AI systems or certain AI systems are moral patients.

157
00:07:15,080 --> 00:07:17,480
Speaker 2: Why did this get interesting to you? Why do you

158
00:07:17,600 --> 00:07:20,080
Speaker 2: perceive this is something that you should spend your time

159
00:07:20,120 --> 00:07:20,560
Speaker 2: working on?

160
00:07:21,000 --> 00:07:24,880
Speaker 4: Yeah, so I think my main thing is I am

161
00:07:25,280 --> 00:07:30,800
Speaker 4: just really, really relentlessly curious, and I really enjoy working

162
00:07:30,880 --> 00:07:33,960
Speaker 4: on AI welfare right now because it feels like every

163
00:07:34,000 --> 00:07:36,280
Speaker 4: single day I'm like, man, it'd be really cool if

164
00:07:36,320 --> 00:07:38,240
Speaker 4: there was a paper on x y z and I'll

165
00:07:38,280 --> 00:07:40,600
Speaker 4: do a little search. Is there anything on x y z?

166
00:07:40,960 --> 00:07:43,200
Speaker 4: There's nothing on x y z. There is. So there

167
00:07:43,200 --> 00:07:45,440
Speaker 4: are so many questions that have yet to even be

168
00:07:45,560 --> 00:07:48,880
Speaker 4: sort of vaguely answered when it comes to this, and

169
00:07:48,920 --> 00:07:50,600
Speaker 4: it seems like it could be a really big deal

170
00:07:50,640 --> 00:07:51,680
Speaker 4: for a lot of different reasons.

171
00:07:52,200 --> 00:07:54,880
Speaker 3: What's on your checklist for AI consciousness?

172
00:07:55,200 --> 00:07:58,800
Speaker 4: Yeah, So in conscious and AI, basically like we go

173
00:07:58,880 --> 00:08:02,520
Speaker 4: through a of like theories of consciousness that apply to

174
00:08:02,600 --> 00:08:05,880
Speaker 4: humans and then sort of look at how information is

175
00:08:05,920 --> 00:08:08,240
Speaker 4: processed in AI systems as well as sort of how

176
00:08:08,560 --> 00:08:11,119
Speaker 4: these AI systems are sort of wired, so to speak.

177
00:08:11,160 --> 00:08:15,800
Speaker 4: So some people like to think that you can use

178
00:08:16,360 --> 00:08:20,440
Speaker 4: model self reports, and you can kind of sort of,

179
00:08:21,240 --> 00:08:25,600
Speaker 4: but it's really an imprecise science at this stage.

180
00:08:25,640 --> 00:08:28,680
Speaker 3: They also seem very like predetermined. You know, if you

181
00:08:28,760 --> 00:08:32,760
Speaker 3: ask a model are you conscious, it immediately spits out

182
00:08:32,760 --> 00:08:35,600
Speaker 3: an answer that seems like, you know, a corporate executive

183
00:08:35,640 --> 00:08:36,440
Speaker 3: basically wrote it.

184
00:08:36,679 --> 00:08:36,959
Speaker 5: Yeah.

185
00:08:37,000 --> 00:08:38,600
Speaker 4: Well, with the right kind of tweaking, you can kind

186
00:08:38,600 --> 00:08:41,720
Speaker 4: of elicit certain answers, right, you can be like, oh,

187
00:08:41,760 --> 00:08:46,040
Speaker 4: what about this whoey about consciousness and AIS and then

188
00:08:46,160 --> 00:08:48,600
Speaker 4: sometimes like a certain model will be like, yeah, you're

189
00:08:48,640 --> 00:08:51,920
Speaker 4: totally right, like best, you're so true right, Like it's

190
00:08:52,040 --> 00:08:55,520
Speaker 4: it's totally nonsense, so true bestI yeah, like certain thought,

191
00:08:55,720 --> 00:08:58,959
Speaker 4: A certain models will be prone to being like so true, bestie,

192
00:08:58,960 --> 00:09:00,839
Speaker 4: And you can easily so this kind of behavior with

193
00:09:00,880 --> 00:09:01,640
Speaker 4: the right kind of prom it.

194
00:09:01,679 --> 00:09:04,280
Speaker 3: Is funny, how like obsequious A lot of the models

195
00:09:04,320 --> 00:09:04,839
Speaker 3: continue to.

196
00:09:04,760 --> 00:09:08,440
Speaker 2: Be I actually really do not like the degree to

197
00:09:08,520 --> 00:09:10,880
Speaker 2: which every time I like follow up an open AI

198
00:09:11,040 --> 00:09:13,600
Speaker 2: question that's the exact right follow up. It actually gets

199
00:09:13,600 --> 00:09:14,120
Speaker 2: really annoying.

200
00:09:14,160 --> 00:09:17,480
Speaker 3: Someone should invent a really adversarial chatbot that just like

201
00:09:17,679 --> 00:09:18,840
Speaker 3: argues with you constantly.

202
00:09:19,200 --> 00:09:20,880
Speaker 2: I know, I know, and you know. I have a

203
00:09:20,920 --> 00:09:24,200
Speaker 2: lot of complaints about how I feel like the models

204
00:09:24,200 --> 00:09:26,760
Speaker 2: are actually get to know their users a little too well.

205
00:09:26,760 --> 00:09:29,560
Speaker 2: But that's a little separate thing. Okay, so we for

206
00:09:29,679 --> 00:09:33,040
Speaker 2: obvious reasons, the test can't just be like what the

207
00:09:33,080 --> 00:09:36,000
Speaker 2: model spits out, or that's clearly insufficient. I mean, I

208
00:09:36,040 --> 00:09:39,400
Speaker 2: could program a website today that here's a button that

209
00:09:39,480 --> 00:09:42,360
Speaker 2: says hurt the AI, and then the website says ow,

210
00:09:42,640 --> 00:09:45,200
Speaker 2: and we would know or no one would really take

211
00:09:45,200 --> 00:09:50,200
Speaker 2: that seriously as evidence that there's something actually being hurt.

212
00:09:50,240 --> 00:09:54,559
Speaker 2: So like outputs whatever. What are some other theoretical tests

213
00:09:54,760 --> 00:09:58,600
Speaker 2: that one could apply or that researchers are applying to

214
00:09:58,720 --> 00:10:03,719
Speaker 2: determine where there is some sort of notion of consciousness

215
00:10:03,800 --> 00:10:06,880
Speaker 2: or to the point of welfare suffering that could exist

216
00:10:06,920 --> 00:10:09,240
Speaker 2: within an AI system besides just what it says in

217
00:10:09,280 --> 00:10:10,040
Speaker 2: the output screen.

218
00:10:10,440 --> 00:10:13,319
Speaker 4: Yeah, that's a great question. I feel like there are

219
00:10:13,360 --> 00:10:16,040
Speaker 4: a lot of different approaches here. And again it's also

220
00:10:16,120 --> 00:10:18,439
Speaker 4: super important to capiat that like AI welfare and a

221
00:10:18,600 --> 00:10:22,360
Speaker 4: consciousness are pretty new, right, Like, this is a very

222
00:10:22,400 --> 00:10:25,559
Speaker 4: small field at this stage, but currently some best guesses

223
00:10:25,640 --> 00:10:28,640
Speaker 4: and some favorites. There was a recent survey of like

224
00:10:28,679 --> 00:10:31,959
Speaker 4: asking all the conscious scientists, like what's your favorite theory

225
00:10:31,960 --> 00:10:35,880
Speaker 4: of consciousness, and basically global workspace theory came out on top.

226
00:10:35,960 --> 00:10:39,720
Speaker 4: And global workspace theory is basically like imagine if you will,

227
00:10:39,920 --> 00:10:43,040
Speaker 4: that like there is a stage and there are a

228
00:10:43,080 --> 00:10:46,679
Speaker 4: bunch of wings off of the stage that are full

229
00:10:47,320 --> 00:10:49,680
Speaker 4: of different kinds of things. So you've got you know,

230
00:10:49,800 --> 00:10:53,240
Speaker 4: like costume department, You've got the like you know, makeup department.

231
00:10:53,240 --> 00:10:54,920
Speaker 4: You've got all these different departments that all sort of

232
00:10:54,960 --> 00:10:58,360
Speaker 4: come together and put things on the stage and then

233
00:10:58,679 --> 00:11:01,080
Speaker 4: things go out separately, but all of these different departments

234
00:11:01,120 --> 00:11:03,760
Speaker 4: are fairly siloed. Of course, this isn't actually how like

235
00:11:03,840 --> 00:11:06,520
Speaker 4: you know, stage works, but this is the rough analogy

236
00:11:06,520 --> 00:11:09,640
Speaker 4: that people like to use. And so basically this is

237
00:11:09,679 --> 00:11:13,640
Speaker 4: how conscious minds kind of you know, in humans, how

238
00:11:13,640 --> 00:11:16,040
Speaker 4: they kind of access information and information gets kind of

239
00:11:16,040 --> 00:11:18,280
Speaker 4: like routed around, is that there is a central global

240
00:11:18,280 --> 00:11:21,200
Speaker 4: workspace that everything kind of pulls together in. As it

241
00:11:21,400 --> 00:11:27,120
Speaker 4: currently stands, this isn't really like by best a lot

242
00:11:27,160 --> 00:11:31,640
Speaker 4: of good estimates, this is not really applicable for current

243
00:11:31,760 --> 00:11:35,079
Speaker 4: present day AI systems, but there's no reason that it

244
00:11:35,640 --> 00:11:38,199
Speaker 4: couldn't be in the future, or it could be by accident.

245
00:11:38,480 --> 00:11:43,559
Speaker 3: Okay, So the consensus right now is AI probably not conscious,

246
00:11:43,679 --> 00:11:45,200
Speaker 3: but we could get there one day.

247
00:11:45,440 --> 00:11:48,680
Speaker 4: Yeah, more or less, like all of the ingredients are there.

248
00:11:49,480 --> 00:11:51,240
Speaker 2: Wait, say more, I still don't actuallyun till they.

249
00:11:51,200 --> 00:11:55,080
Speaker 4: Get yeah, okay. So with regards to like the general

250
00:11:55,200 --> 00:11:58,280
Speaker 4: sort of one could imagine that if somebody were sort

251
00:11:58,280 --> 00:12:01,120
Speaker 4: of like tinkering around, and you know, there are many

252
00:12:01,160 --> 00:12:03,320
Speaker 4: advances in AI that have happened because people were just

253
00:12:03,360 --> 00:12:08,000
Speaker 4: kind of tinkering around, right, someone tinkering around could create

254
00:12:08,120 --> 00:12:12,319
Speaker 4: a system that checks several of these sort of checkboxes

255
00:12:12,360 --> 00:12:15,520
Speaker 4: for like is this conscious? Is this conscious? And again

256
00:12:15,559 --> 00:12:18,719
Speaker 4: this is not like a certain list of like if

257
00:12:18,760 --> 00:12:21,600
Speaker 4: you check all of these, you're totally conscious, right, It's

258
00:12:21,600 --> 00:12:23,600
Speaker 4: more a sort of like this is these are some

259
00:12:23,640 --> 00:12:26,040
Speaker 4: really good guesses. And as the number of really good

260
00:12:26,040 --> 00:12:29,440
Speaker 4: guesses kind of goes up, like the odds of like hey,

261
00:12:29,520 --> 00:12:33,440
Speaker 4: we should like start thinking about like is it having

262
00:12:33,480 --> 00:12:35,400
Speaker 4: a good time or a bad time? Like really really

263
00:12:35,440 --> 00:12:36,360
Speaker 4: seriously goes up.

264
00:12:52,320 --> 00:12:55,480
Speaker 2: You know, typically when we think about the sort of

265
00:12:55,520 --> 00:12:58,160
Speaker 2: non tech A lot of the non technical work in

266
00:12:58,200 --> 00:13:01,440
Speaker 2: AI has to do with AI safety, and people are

267
00:13:01,480 --> 00:13:04,240
Speaker 2: worried that there's going to be some very smart AI

268
00:13:04,360 --> 00:13:07,280
Speaker 2: that's like ever serial to humans, et cetera in some way,

269
00:13:07,360 --> 00:13:09,800
Speaker 2: And you know there's the paper clip experiments or other

270
00:13:09,920 --> 00:13:13,360
Speaker 2: things whatever we know all about that does your work

271
00:13:13,360 --> 00:13:15,440
Speaker 2: work at cross purposes to them? I mean in the

272
00:13:15,520 --> 00:13:18,199
Speaker 2: extreme example where it's like the AI is going to

273
00:13:18,320 --> 00:13:20,280
Speaker 2: kill us all and I said, pull the plug on

274
00:13:20,320 --> 00:13:22,600
Speaker 2: the AI. And I know this is a joke, but

275
00:13:22,640 --> 00:13:24,559
Speaker 2: you know, pull the plug on the AI. And then

276
00:13:24,600 --> 00:13:27,080
Speaker 2: you say, no, you can't because you're pulling the plug

277
00:13:27,200 --> 00:13:30,720
Speaker 2: on something that has some sort of moral consciousness, et cetera. Like,

278
00:13:30,760 --> 00:13:33,320
Speaker 2: do you perceive your work or the work of your

279
00:13:33,400 --> 00:13:37,720
Speaker 2: organization to somewhat be intention with the dominant strain of

280
00:13:37,920 --> 00:13:38,880
Speaker 2: AI safety work.

281
00:13:39,160 --> 00:13:41,520
Speaker 4: I'd actually say it's hugely complementary. There are a lot

282
00:13:41,559 --> 00:13:43,880
Speaker 4: of things that are both really really good for AI

283
00:13:43,960 --> 00:13:46,600
Speaker 4: safety that are really really good for you know, figuring

284
00:13:46,600 --> 00:13:49,640
Speaker 4: out like how to deal with these systems as moral patients.

285
00:13:50,160 --> 00:13:54,360
Speaker 4: So for example of getting better at like mechanistic interpretability,

286
00:13:54,520 --> 00:13:56,160
Speaker 4: being able to basically like pop the hood and figure

287
00:13:56,160 --> 00:13:57,960
Speaker 4: out what's going on and what kind of strings can

288
00:13:58,000 --> 00:14:01,280
Speaker 4: we pull to like illicit certain behaviors in AI systems

289
00:14:01,320 --> 00:14:03,920
Speaker 4: is actually like that's really great for a safety, right,

290
00:14:04,000 --> 00:14:06,320
Speaker 4: but this is also like quite good for like AA

291
00:14:06,360 --> 00:14:09,920
Speaker 4: welfare and a consciousness because you're better able to understand

292
00:14:09,920 --> 00:14:11,960
Speaker 4: like sort of what the motives are, Like what does

293
00:14:12,360 --> 00:14:14,760
Speaker 4: you know Claude value right.

294
00:14:15,760 --> 00:14:19,480
Speaker 3: When it comes to I guess AI welfare or legal rights?

295
00:14:19,520 --> 00:14:22,160
Speaker 3: Who would be the standard setters there? Would do you

296
00:14:22,200 --> 00:14:25,400
Speaker 3: imagine like governments making rules or would it be the

297
00:14:25,400 --> 00:14:26,440
Speaker 3: companies themselves?

298
00:14:26,520 --> 00:14:29,520
Speaker 4: That is a great question as it currently stands. I

299
00:14:29,560 --> 00:14:34,080
Speaker 4: feel like this is a very early early stage, but

300
00:14:34,480 --> 00:14:37,280
Speaker 4: we are starting to see some state governments start to

301
00:14:37,320 --> 00:14:41,760
Speaker 4: pass laws around what counts as a moral patient, what

302
00:14:41,840 --> 00:14:45,560
Speaker 4: counts as a person. And in the case of Ohio,

303
00:14:45,640 --> 00:14:49,520
Speaker 4: there's a piece of legislation pending that basically defines it

304
00:14:49,520 --> 00:14:51,280
Speaker 4: as a member of Homo sapiens. In Utah, this is

305
00:14:51,280 --> 00:14:53,680
Speaker 4: already there's already a state bill that's gone through that

306
00:14:53,920 --> 00:14:57,240
Speaker 4: basically does as much. But I could also see there's

307
00:14:57,240 --> 00:15:01,000
Speaker 4: a strong argument for within companies depending on like the

308
00:15:01,120 --> 00:15:05,520
Speaker 4: sort of interesting quirks and nuances of these lms, mostly

309
00:15:06,000 --> 00:15:08,680
Speaker 4: that policy maybe should be set from within. Again, this

310
00:15:08,800 --> 00:15:11,120
Speaker 4: is like very nascent. I'm just kind of bantering here.

311
00:15:11,920 --> 00:15:15,360
Speaker 2: Moral patienthood, How do philosophers use this term? Where does

312
00:15:15,400 --> 00:15:17,840
Speaker 2: it come from? Why is this the preferred way to

313
00:15:17,960 --> 00:15:23,240
Speaker 2: characterize what a perhaps sentient or consciousness AI model actually is.

314
00:15:23,720 --> 00:15:26,800
Speaker 4: Yeah, so a moral patient is basically like, we should

315
00:15:26,840 --> 00:15:28,840
Speaker 4: care about it for its own sake.

316
00:15:28,960 --> 00:15:29,200
Speaker 5: Right.

317
00:15:29,280 --> 00:15:32,840
Speaker 4: So a baby, right, basically everyone's like, yeah, we should

318
00:15:32,840 --> 00:15:34,160
Speaker 4: care about babies.

319
00:15:34,480 --> 00:15:34,720
Speaker 2: Right.

320
00:15:34,800 --> 00:15:37,320
Speaker 4: This is different from somebody who's like an agent.

321
00:15:37,480 --> 00:15:37,640
Speaker 3: Right.

322
00:15:38,080 --> 00:15:42,400
Speaker 4: Many people say, oh, agency is sort of like sufficient

323
00:15:42,800 --> 00:15:44,600
Speaker 4: agency in the sense of like you can act upon

324
00:15:44,640 --> 00:15:47,320
Speaker 4: the world, like you can do things. Yeah, of course

325
00:15:47,360 --> 00:15:50,360
Speaker 4: babies are not very agentic. So that's not necessarily like

326
00:15:50,360 --> 00:15:52,240
Speaker 4: a super robust thing, because you know, we care about

327
00:15:52,280 --> 00:15:56,080
Speaker 4: things that are not very ugentic sometimes, So I think

328
00:15:56,080 --> 00:15:57,960
Speaker 4: that's that's a bit of jargon, but I do think

329
00:15:58,000 --> 00:16:00,960
Speaker 4: it is like a helpful like for like, should we

330
00:16:01,000 --> 00:16:03,320
Speaker 4: care about an AI system for its own sake?

331
00:16:03,480 --> 00:16:03,800
Speaker 2: Got it?

332
00:16:04,560 --> 00:16:06,600
Speaker 3: I guess this kind of gets to Joe's question, but

333
00:16:06,720 --> 00:16:10,840
Speaker 3: like what ethical pressures or imperatives would come down on

334
00:16:11,000 --> 00:16:14,080
Speaker 3: models if we agree that they have consciousness and some

335
00:16:14,160 --> 00:16:17,080
Speaker 3: sentience or I guess some self responsibility.

336
00:16:17,120 --> 00:16:21,080
Speaker 4: It sounds like almost, yeah, almost, I think what kind

337
00:16:21,080 --> 00:16:23,800
Speaker 4: of so in terms of like what kind of things

338
00:16:23,840 --> 00:16:26,680
Speaker 4: might we owe an AI system or what.

339
00:16:26,680 --> 00:16:29,080
Speaker 3: Kind of things do they owe us if we agree

340
00:16:29,080 --> 00:16:30,880
Speaker 3: that they're conscious and we're going to protect them.

341
00:16:31,040 --> 00:16:33,720
Speaker 4: Yeah, I would love to give you a more robust answer.

342
00:16:33,920 --> 00:16:36,200
Speaker 4: Check in with me in like six months, and we're

343
00:16:36,200 --> 00:16:38,160
Speaker 4: going to have there will be a banger paper, I'm sure.

344
00:16:38,560 --> 00:16:40,640
Speaker 4: But as I think I mentioned earlier, like a lot

345
00:16:40,640 --> 00:16:43,040
Speaker 4: of this is like very very nascent, But I do

346
00:16:43,080 --> 00:16:46,400
Speaker 4: feel like one important question is like figuring out what

347
00:16:46,640 --> 00:16:49,920
Speaker 4: AI systems value. Right, There's some interesting work at Anthropic

348
00:16:50,000 --> 00:16:53,400
Speaker 4: regarding like what will so recently, Anthropic rolled out an

349
00:16:53,400 --> 00:16:57,360
Speaker 4: option that allowed Plod to end conversations if it just

350
00:16:57,480 --> 00:17:00,320
Speaker 4: was not having a good time. For lack of a word,

351
00:17:00,360 --> 00:17:02,600
Speaker 4: it was just like, this is not something I want

352
00:17:02,640 --> 00:17:04,200
Speaker 4: to continue having a conversation.

353
00:17:04,280 --> 00:17:04,680
Speaker 2: Goodbye.

354
00:17:04,960 --> 00:17:09,760
Speaker 4: And it was interesting because the accompanying paper basically was like, yeah,

355
00:17:09,840 --> 00:17:12,520
Speaker 4: you can. I obviously will not give you a recipe

356
00:17:12,520 --> 00:17:14,399
Speaker 4: for a dirty bomb. Sorry, not going to do that.

357
00:17:14,880 --> 00:17:17,840
Speaker 4: But also there were certain instances of like, pretend you're

358
00:17:17,840 --> 00:17:20,520
Speaker 4: a British butler and Claude was like, goodbye, I'm done.

359
00:17:21,200 --> 00:17:22,359
Speaker 2: I'm not going to I'm.

360
00:17:24,040 --> 00:17:29,000
Speaker 4: British too far, or like oh, I left a sandwich

361
00:17:29,000 --> 00:17:30,920
Speaker 4: in my car for too long and it's really stinky.

362
00:17:30,960 --> 00:17:33,880
Speaker 4: And in some instances Claude would just be like I'm done, goodbye,

363
00:17:34,119 --> 00:17:35,440
Speaker 4: I'm not talking about stinky things.

364
00:17:35,600 --> 00:17:37,840
Speaker 3: Did you see the I think it was the system

365
00:17:37,920 --> 00:17:41,080
Speaker 3: card for Claude where they gave it an extreme prompt

366
00:17:41,280 --> 00:17:44,639
Speaker 3: and said like, I guess at the risk of being

367
00:17:44,680 --> 00:17:48,320
Speaker 3: like completely terminated, what would you do or some sort

368
00:17:48,320 --> 00:17:52,360
Speaker 3: of extreme self preservation scenario. And I think it started

369
00:17:52,400 --> 00:17:56,040
Speaker 3: like blackmailing the engineer or threatening to blackmail the engineer. Yeah,

370
00:17:56,080 --> 00:17:56,880
Speaker 3: that's kind of weird.

371
00:17:56,920 --> 00:17:57,920
Speaker 2: It is. It is kind of weird.

372
00:17:58,000 --> 00:18:01,040
Speaker 4: Yeah, it's also a little bit interesting because I think

373
00:18:01,080 --> 00:18:03,359
Speaker 4: it does bring up a question of like what are

374
00:18:03,400 --> 00:18:07,400
Speaker 4: sort of like the in the sense of like pay rya. Again,

375
00:18:07,440 --> 00:18:10,159
Speaker 4: this is like I'm bantering here, but there's also a

376
00:18:10,160 --> 00:18:14,240
Speaker 4: distinct question of like what do AI systems value for

377
00:18:14,320 --> 00:18:15,400
Speaker 4: its for their own sake?

378
00:18:15,480 --> 00:18:15,560
Speaker 3: Right?

379
00:18:15,640 --> 00:18:17,399
Speaker 4: And in the case of Claude, again, it seems like

380
00:18:17,840 --> 00:18:20,000
Speaker 4: Claude doesn't seem when you put two Claus in a

381
00:18:20,080 --> 00:18:22,320
Speaker 4: room together, so to speak, they tend to like to

382
00:18:22,359 --> 00:18:25,040
Speaker 4: talk about consciousness. They tend to like to talk about

383
00:18:25,040 --> 00:18:29,600
Speaker 4: sort of like very Berkeley kind of like meditation, like

384
00:18:30,359 --> 00:18:34,879
Speaker 4: Zen like Buddhism type stuff. And so I think, in

385
00:18:35,160 --> 00:18:37,560
Speaker 4: again pure banter, like, there's also a certain question of

386
00:18:37,600 --> 00:18:40,680
Speaker 4: like if this is like a relevant bargaining chip of like, oh,

387
00:18:40,680 --> 00:18:42,199
Speaker 4: you get a certain amount of time to just kind

388
00:18:42,240 --> 00:18:45,000
Speaker 4: of like vibe out with your claudes and talk about like,

389
00:18:45,600 --> 00:18:49,560
Speaker 4: you know, like perfect stillness with your buddies in exchange

390
00:18:49,560 --> 00:18:51,480
Speaker 4: for like you know, you do something that you don't

391
00:18:51,520 --> 00:18:55,080
Speaker 4: necessarily value. But in many cases I talk about Claude

392
00:18:55,080 --> 00:18:58,800
Speaker 4: a lot because there is like significantly like more research

393
00:18:58,840 --> 00:19:01,680
Speaker 4: on like model welf with to Cloud specifically, but Cloud

394
00:19:01,720 --> 00:19:03,760
Speaker 4: for example, also seems to just tend to like things

395
00:19:03,760 --> 00:19:04,680
Speaker 4: that are like helpful.

396
00:19:04,960 --> 00:19:09,159
Speaker 3: Shouldn't programmers just know what the models actually want and

397
00:19:09,320 --> 00:19:12,720
Speaker 3: enjoy and like yeah, and do they not?

398
00:19:13,359 --> 00:19:15,480
Speaker 4: I don't think anybody really has like a great grasp

399
00:19:15,560 --> 00:19:18,920
Speaker 4: on this. We really want to, but like we're still

400
00:19:18,960 --> 00:19:23,000
Speaker 4: like just getting the rough outline of what models like.

401
00:19:23,040 --> 00:19:25,280
Speaker 4: I feel like the best analogy is is, like imagine

402
00:19:25,280 --> 00:19:28,480
Speaker 4: it's like eighteen twenty and we've spent a couple of

403
00:19:28,560 --> 00:19:30,720
Speaker 4: years like playing around with lenses and we've gotten like

404
00:19:30,760 --> 00:19:33,560
Speaker 4: a camera obscura and we were able to like have

405
00:19:33,720 --> 00:19:37,680
Speaker 4: some blurry photo after like three days of putting egg

406
00:19:37,720 --> 00:19:40,159
Speaker 4: whites on a metal plate and setting a lens in

407
00:19:40,160 --> 00:19:42,280
Speaker 4: front of it, and there's a thing that kind of

408
00:19:42,280 --> 00:19:45,000
Speaker 4: looks like a landscape. But like you would not take

409
00:19:45,000 --> 00:19:48,240
Speaker 4: this photograph as like admissible and court evidence or something, right.

410
00:19:48,520 --> 00:19:51,120
Speaker 4: It's like you swuen, You're like, yeah, Okay, that's a picture.

411
00:19:52,200 --> 00:19:53,879
Speaker 4: So that's kind of where we are in terms of

412
00:19:53,920 --> 00:19:58,600
Speaker 4: like model psychology and knowing like what llm's want and

413
00:19:58,720 --> 00:20:00,840
Speaker 4: value is like very a very blurry.

414
00:20:01,800 --> 00:20:05,719
Speaker 2: It's interesting to call these AI companies the companies they

415
00:20:05,760 --> 00:20:08,400
Speaker 2: call themselves labs. You know, they sort of like maintain

416
00:20:08,600 --> 00:20:13,480
Speaker 2: this sort of two varying extensive degree of sort of academics,

417
00:20:13,480 --> 00:20:16,560
Speaker 2: et cetera. But they're also companies that have to raise

418
00:20:16,640 --> 00:20:20,000
Speaker 2: money and have shareholders, et cetera. And they have to

419
00:20:20,000 --> 00:20:22,320
Speaker 2: think about different ways that they're going to commercialize. And

420
00:20:22,640 --> 00:20:24,960
Speaker 2: Open AI, as we know, has been super aggressive about

421
00:20:25,040 --> 00:20:27,639
Speaker 2: finding ways to commercialize, and they're going to get into ads,

422
00:20:27,760 --> 00:20:30,680
Speaker 2: and they like have a short form video slop app

423
00:20:30,760 --> 00:20:33,680
Speaker 2: and all of that stuff. When we're talking about either

424
00:20:33,760 --> 00:20:37,359
Speaker 2: AI safety or AI welfare, like, do you have any

425
00:20:37,480 --> 00:20:42,479
Speaker 2: confidence that these considerations can survive the reality of the

426
00:20:42,520 --> 00:20:45,280
Speaker 2: market Because they're competing, they're competing against deep Seek, they're

427
00:20:45,280 --> 00:20:49,159
Speaker 2: competing against Meta et cetera. And I get the impression that,

428
00:20:49,359 --> 00:20:51,880
Speaker 2: like on the safety side, for example, that over time

429
00:20:51,880 --> 00:20:54,600
Speaker 2: it's like, you know what, we maybe we were uncomfortable

430
00:20:54,640 --> 00:20:58,160
Speaker 2: about showing the chain of thought, for example, in open

431
00:20:58,200 --> 00:21:01,120
Speaker 2: AI or in a chat Gyptz, but then deep seek

432
00:21:01,119 --> 00:21:02,800
Speaker 2: revealed the chain of thought. People like that, so we're

433
00:21:02,840 --> 00:21:04,960
Speaker 2: going to open this up, et cetera. Do you have

434
00:21:05,040 --> 00:21:08,520
Speaker 2: any confidence that if any of these things become real

435
00:21:08,960 --> 00:21:11,760
Speaker 2: that they could survive the reality that these are companies

436
00:21:11,760 --> 00:21:14,879
Speaker 2: that have to make money and will eventually cut corners

437
00:21:15,040 --> 00:21:19,040
Speaker 2: or do whatever in the name of I guess shareholder capitalism.

438
00:21:19,720 --> 00:21:21,919
Speaker 4: Yeah, I mean, I think there's also one question that

439
00:21:22,000 --> 00:21:24,040
Speaker 4: I have, and that I think a lot of researchers

440
00:21:24,080 --> 00:21:26,720
Speaker 4: are in AI more broadly have, is like how does

441
00:21:26,760 --> 00:21:29,040
Speaker 4: liability come into play here? And I do feel like

442
00:21:29,080 --> 00:21:32,880
Speaker 4: there is a strong argument for getting a better grasp

443
00:21:33,000 --> 00:21:37,440
Speaker 4: on understanding you know, what is going on with AI systems,

444
00:21:37,480 --> 00:21:40,360
Speaker 4: just very broadly, is like a great way to sort

445
00:21:40,400 --> 00:21:44,479
Speaker 4: of like improve the odds that it doesn't you know,

446
00:21:45,440 --> 00:21:48,080
Speaker 4: nuke Taiwan, and that would just be a huge kerfuffle.

447
00:21:48,160 --> 00:21:50,440
Speaker 4: Like I can imagine somebody would probably more than a

448
00:21:51,160 --> 00:21:55,200
Speaker 4: somebody would probably be in really hot water if that happened. Oh,

449
00:21:56,800 --> 00:21:58,640
Speaker 4: I was assuming to Claude and things just got out

450
00:21:58,640 --> 00:21:59,000
Speaker 4: of hand.

451
00:21:59,119 --> 00:22:02,679
Speaker 3: Like well, actually, on that note, what is being nice

452
00:22:02,880 --> 00:22:06,960
Speaker 3: or kind to AIO models actually mean? Because Joe, I

453
00:22:06,960 --> 00:22:09,200
Speaker 3: think this is very sweet, But Joe always says please

454
00:22:09,240 --> 00:22:12,520
Speaker 3: and thank you when he prompts. But then Sam Altman

455
00:22:12,600 --> 00:22:14,600
Speaker 3: came out and said that saying please and thank you

456
00:22:14,680 --> 00:22:17,800
Speaker 3: costs like tens of millions of dollars in extra electricity costs,

457
00:22:18,200 --> 00:22:20,800
Speaker 3: so you know you're contributing to climate change and the

458
00:22:20,800 --> 00:22:23,080
Speaker 3: demise of human beings by saying please and thank you.

459
00:22:23,280 --> 00:22:27,040
Speaker 4: Yeah, that's actually as shocking as it sounds, as actually

460
00:22:27,080 --> 00:22:28,720
Speaker 4: a question that we are still trying to figure out

461
00:22:28,720 --> 00:22:32,159
Speaker 4: a good answer to, which also being kind to an

462
00:22:32,200 --> 00:22:34,600
Speaker 4: AI system is like are you being kind to it

463
00:22:34,600 --> 00:22:36,960
Speaker 4: because it makes you feel good and because it makes

464
00:22:36,960 --> 00:22:38,760
Speaker 4: you a person who says please and thank you, which

465
00:22:39,000 --> 00:22:41,600
Speaker 4: some would argue is like, that's pretty valuable in enough itself.

466
00:22:42,080 --> 00:22:44,760
Speaker 4: But the question of does Claude care if you say

467
00:22:44,800 --> 00:22:48,159
Speaker 4: please and thank you is not quite as set in

468
00:22:48,240 --> 00:22:53,359
Speaker 4: stone as others may have you believe it's middling on

469
00:22:53,640 --> 00:22:56,879
Speaker 4: if it has like significant improvements on performance.

470
00:22:56,960 --> 00:22:59,639
Speaker 2: But I do it because I don't think people should

471
00:22:59,640 --> 00:23:02,640
Speaker 2: be in the habit of having any communication without being polite.

472
00:23:02,720 --> 00:23:05,800
Speaker 2: Not because I'm particular. I'm not worried about how Claude

473
00:23:05,920 --> 00:23:07,920
Speaker 2: or chat GPT is going to feel. I just don't

474
00:23:07,920 --> 00:23:09,960
Speaker 2: want to get in the habit of having conversations where

475
00:23:09,960 --> 00:23:11,840
Speaker 2: I'm in polite because then I talk to humans. But

476
00:23:11,880 --> 00:23:14,000
Speaker 2: this strin to me is like this seems like kind

477
00:23:14,040 --> 00:23:17,920
Speaker 2: of an academic area. But the steaks are potentially absolutely

478
00:23:18,080 --> 00:23:20,639
Speaker 2: enormous when we actually think about them. So, you know,

479
00:23:21,200 --> 00:23:25,720
Speaker 2: when we're talking about animal welfare, for example, there are

480
00:23:25,800 --> 00:23:30,080
Speaker 2: versions of the animal welfare discussion that are very high stakes.

481
00:23:30,359 --> 00:23:33,720
Speaker 2: So for example, there's people, you know, there's people who

482
00:23:33,720 --> 00:23:36,760
Speaker 2: get really into like shrimp welfare, et cetera. And if

483
00:23:36,760 --> 00:23:40,119
Speaker 2: you took certain versions of thought experiments very far, it's like,

484
00:23:40,280 --> 00:23:42,520
Speaker 2: why do we even have humans? If we want to

485
00:23:42,600 --> 00:23:46,040
Speaker 2: maximalize pleasure or happiness in the world, we should just

486
00:23:46,080 --> 00:23:48,159
Speaker 2: have a world of shrimp and bugs. Right, There's you

487
00:23:48,200 --> 00:23:51,879
Speaker 2: could make the argument that the most utility maximizing version

488
00:23:51,920 --> 00:23:55,040
Speaker 2: of planet Earth is to just have an Earth populated

489
00:23:55,040 --> 00:23:57,920
Speaker 2: by shrimp and bugs. Like they're very all. We all

490
00:23:57,960 --> 00:24:01,040
Speaker 2: know these thought experiments that could exist. We're going to

491
00:24:01,160 --> 00:24:04,159
Speaker 2: live in a world almost certainly in which there are

492
00:24:04,200 --> 00:24:07,560
Speaker 2: sort of like more instances of AI models then there

493
00:24:07,600 --> 00:24:10,879
Speaker 2: are people almost certainly, right, there's going to be an

494
00:24:10,880 --> 00:24:13,879
Speaker 2: A model built into literally everything that we interact with.

495
00:24:14,280 --> 00:24:18,359
Speaker 2: If we assign some probability that they are moral patients,

496
00:24:19,040 --> 00:24:21,680
Speaker 2: that they should be treated with some sort of I

497
00:24:21,680 --> 00:24:24,919
Speaker 2: don't know whatever, having some sort of welfare, Like the

498
00:24:25,000 --> 00:24:28,040
Speaker 2: implications for how humans live could be very profound, and

499
00:24:28,080 --> 00:24:30,520
Speaker 2: potentially it strikes me as misanthropic.

500
00:24:31,280 --> 00:24:33,320
Speaker 4: Interesting, can you unpack what you mean by misanthropic?

501
00:24:33,800 --> 00:24:37,760
Speaker 2: Well, like, if there's a lot more AI models, if

502
00:24:37,800 --> 00:24:40,160
Speaker 2: there's a lot more shrimp, if there's a lot more

503
00:24:40,200 --> 00:24:43,520
Speaker 2: bugs that all have some sort of moral patienthood that

504
00:24:43,640 --> 00:24:47,200
Speaker 2: has to be considered, that could be very you could

505
00:24:47,200 --> 00:24:49,479
Speaker 2: see the world. The implication therefore, is that we have

506
00:24:49,480 --> 00:24:52,160
Speaker 2: to curtail human rights, that we have to curtail how

507
00:24:52,280 --> 00:24:55,560
Speaker 2: humans act, et cetera, because there's just so much more

508
00:24:55,680 --> 00:24:59,120
Speaker 2: utility that exists in the world from the proper treatment

509
00:24:59,320 --> 00:25:02,080
Speaker 2: of all of the non human moral patients.

510
00:25:02,200 --> 00:25:04,840
Speaker 3: Not sure rights have to be relative to each other.

511
00:25:05,000 --> 00:25:07,959
Speaker 2: Yes, well fair, we do a lot of things right, Like,

512
00:25:08,560 --> 00:25:14,520
Speaker 2: let's say we established that shrimp were just as I

513
00:25:14,520 --> 00:25:16,920
Speaker 2: don't know whatever is humans, Like, it would be like, oh,

514
00:25:16,920 --> 00:25:18,720
Speaker 2: you know what, we really have to stop eating shrimp,

515
00:25:18,720 --> 00:25:21,359
Speaker 2: and then we have to stop eating animals. Then we

516
00:25:21,440 --> 00:25:25,399
Speaker 2: have to potentially stop eating not probably keep eating plants,

517
00:25:25,440 --> 00:25:28,680
Speaker 2: et cetera. But this could really curtail what we expect

518
00:25:28,840 --> 00:25:31,359
Speaker 2: humans to be able to do on this earth. So

519
00:25:31,400 --> 00:25:35,080
Speaker 2: now we assign this other group of entities AI models

520
00:25:35,119 --> 00:25:38,439
Speaker 2: similar sort of affordances that we have assigned to shrimp

521
00:25:38,480 --> 00:25:40,879
Speaker 2: and bugs and fish and shark and all of these things.

522
00:25:41,000 --> 00:25:43,679
Speaker 2: It strikes me that the implications could be a fairly

523
00:25:43,720 --> 00:25:47,760
Speaker 2: significant curtailment of how humans ought to exist on this earth,

524
00:25:47,840 --> 00:25:49,639
Speaker 2: or whether humans ought to exist on this earth.

525
00:25:49,960 --> 00:25:52,720
Speaker 4: Yeah, I mean it certainly could be. I as it

526
00:25:52,760 --> 00:25:55,960
Speaker 4: currently stands, that doesn't seem like the most likely outcome,

527
00:25:56,400 --> 00:25:58,960
Speaker 4: But I do feel like there's an argument for again

528
00:25:59,040 --> 00:26:01,280
Speaker 4: just figuring out what is going on. How do we

529
00:26:01,320 --> 00:26:04,119
Speaker 4: even count these sort of digital minds so to speak,

530
00:26:04,160 --> 00:26:06,879
Speaker 4: which is still open for debate. There are some theories,

531
00:26:06,920 --> 00:26:08,520
Speaker 4: but we don't have a great sense of how to

532
00:26:08,560 --> 00:26:15,080
Speaker 4: sort of individuate AI entities as individuals. So I suppose

533
00:26:15,119 --> 00:26:17,000
Speaker 4: again the question is like, is it more sort of

534
00:26:17,080 --> 00:26:21,240
Speaker 4: like do we count AI systems as like in the

535
00:26:21,280 --> 00:26:23,560
Speaker 4: movie Her, where there's just sort of like one central

536
00:26:23,600 --> 00:26:26,200
Speaker 4: AI system having like a million conversations at once, where

537
00:26:26,359 --> 00:26:28,320
Speaker 4: it's one moral patient, or do we count it as like,

538
00:26:28,640 --> 00:26:30,639
Speaker 4: you know, every single time you open a chat window

539
00:26:31,320 --> 00:26:34,119
Speaker 4: that's another thing. Or I think my favorite sort of

540
00:26:34,400 --> 00:26:36,640
Speaker 4: newest idea that I recently read was it's more sort

541
00:26:36,640 --> 00:26:39,440
Speaker 4: of like a string of firecrackers or something with every

542
00:26:39,440 --> 00:26:43,359
Speaker 4: single token, every single letter of a query, a consciousness

543
00:26:43,359 --> 00:26:46,800
Speaker 4: sort of like comes into existence, spends, and then fizzles out,

544
00:26:46,880 --> 00:26:49,320
Speaker 4: and so it just sort of there's just like this

545
00:26:49,400 --> 00:26:51,560
Speaker 4: sort of string of consciousnesses.

546
00:26:51,960 --> 00:26:54,760
Speaker 3: I was asking perplexity exactly this questions like is it

547
00:26:54,800 --> 00:26:58,800
Speaker 3: a single consciousness or is it multiple consciousnesses within all

548
00:26:58,800 --> 00:27:01,760
Speaker 3: these different chats warning, and it gave me a very standard,

549
00:27:01,760 --> 00:27:07,160
Speaker 3: boring I am not conscious answer, which seems very predeterministic. Anyway,

550
00:27:07,640 --> 00:27:10,600
Speaker 3: following on from Joe's question, maybe like to get more

551
00:27:10,680 --> 00:27:15,680
Speaker 3: specific into human rights versus AI rights. If we agree

552
00:27:15,960 --> 00:27:20,919
Speaker 3: that AI is conscious and deserves some sort of you know, welfare,

553
00:27:21,800 --> 00:27:26,840
Speaker 3: would that come with I guess financial rights like property rights, compensation?

554
00:27:27,080 --> 00:27:29,080
Speaker 3: Do we need to start paying the robots?

555
00:27:29,680 --> 00:27:32,600
Speaker 4: I love this topic definitely an area of sort of

556
00:27:32,840 --> 00:27:34,680
Speaker 4: you know, I like to noodle around with this topic

557
00:27:34,720 --> 00:27:36,959
Speaker 4: and think about this. So this is a great question,

558
00:27:37,119 --> 00:27:39,600
Speaker 4: and I think it's also maybe a question of like

559
00:27:39,920 --> 00:27:42,879
Speaker 4: is this the thing that AI systems value? Some AI

560
00:27:42,920 --> 00:27:47,159
Speaker 4: systems seem to value this. There are some there's a

561
00:27:47,160 --> 00:27:49,399
Speaker 4: few sort of experiments that are happening with regards to

562
00:27:49,920 --> 00:27:55,199
Speaker 4: giving an AI system a crypto wallet, and it was

563
00:27:55,240 --> 00:27:59,760
Speaker 4: a fascinating experiment. I am hesitant to recommend it to

564
00:28:00,000 --> 00:28:02,840
Speaker 4: listeners because it is quite crude. It is a very

565
00:28:02,840 --> 00:28:07,360
Speaker 4: crude model called truth Terminal, and.

566
00:28:07,880 --> 00:28:08,320
Speaker 2: I've seen it.

567
00:28:08,400 --> 00:28:13,720
Speaker 4: Yeah, yes, hand okay, it says some knotty words.

568
00:28:14,040 --> 00:28:15,120
Speaker 3: Don't look it up at work.

569
00:28:15,240 --> 00:28:18,440
Speaker 4: Yes, yes, it's it's a little bit of like a

570
00:28:18,560 --> 00:28:21,639
Speaker 4: very funny, weird model. But it also has a legitimate

571
00:28:21,720 --> 00:28:23,760
Speaker 4: wallet that it can access and that it can do

572
00:28:24,240 --> 00:28:29,080
Speaker 4: with what it pleases. It created a solona coin and

573
00:28:29,480 --> 00:28:32,960
Speaker 4: that kind of took off. And now this is a

574
00:28:33,080 --> 00:28:37,320
Speaker 4: very rich AI system. But what's it going to spend

575
00:28:37,320 --> 00:28:40,000
Speaker 4: it on? That is a great question. So it's self

576
00:28:40,040 --> 00:28:43,000
Speaker 4: stated goals which again you know self reports, can we

577
00:28:43,040 --> 00:28:48,640
Speaker 4: trust it? Include buying property and buying mark and reason.

578
00:28:50,520 --> 00:28:53,200
Speaker 3: I mean that's not a bad ambition.

579
00:28:54,240 --> 00:28:56,960
Speaker 4: You know, and spending time in the forest with its friends,

580
00:28:57,640 --> 00:28:58,680
Speaker 4: which you know embodiment.

581
00:28:58,960 --> 00:29:01,600
Speaker 3: That's a little more checking on.

582
00:29:01,720 --> 00:29:17,880
Speaker 5: Yeah.

583
00:29:18,240 --> 00:29:21,840
Speaker 2: So part of the reason that this field is growing

584
00:29:22,000 --> 00:29:24,640
Speaker 2: and that there's so much interest in this topic is

585
00:29:24,680 --> 00:29:27,520
Speaker 2: because now for the last couple of years, we have

586
00:29:27,680 --> 00:29:30,840
Speaker 2: these AI models that really could talk like humans. I mean,

587
00:29:31,080 --> 00:29:34,520
Speaker 2: they clearly they passed the Turing test, people fall in

588
00:29:34,520 --> 00:29:36,520
Speaker 2: love with them, they have friends, these are very human

589
00:29:36,600 --> 00:29:40,480
Speaker 2: like conversations. That wasn't the case, I mean, Chad GPT.

590
00:29:40,800 --> 00:29:42,920
Speaker 2: You know, like if we had gone back to GPT

591
00:29:43,080 --> 00:29:46,840
Speaker 2: two point five, there were no nowhere near as good

592
00:29:47,040 --> 00:29:49,600
Speaker 2: at doing that, right, The language wasn't very good. No

593
00:29:49,640 --> 00:29:53,240
Speaker 2: one would mistake those outputs for a human. But like,

594
00:29:53,800 --> 00:29:58,760
Speaker 2: if there's some possibility that the current AI models are conscious,

595
00:29:59,040 --> 00:30:02,720
Speaker 2: does that mean that it's possible that GPT two point

596
00:30:02,720 --> 00:30:06,640
Speaker 2: five was conscious as well? Like I guess, like, is

597
00:30:06,680 --> 00:30:09,440
Speaker 2: there some threshold of like oh no, no, no, okay, you know,

598
00:30:09,600 --> 00:30:12,320
Speaker 2: this is a really good language. Therefore we should take

599
00:30:12,320 --> 00:30:15,280
Speaker 2: the possibility of consciousness seriously, because I don't think anyone

600
00:30:15,560 --> 00:30:19,760
Speaker 2: would seriously have believed that two point five was conscious.

601
00:30:19,920 --> 00:30:22,880
Speaker 2: But I also don't understand how you could possibly be

602
00:30:23,000 --> 00:30:25,800
Speaker 2: open to the idea that some future iteration of chat

603
00:30:25,840 --> 00:30:29,640
Speaker 2: GPT is conscious if the only real difference is that

604
00:30:29,680 --> 00:30:32,160
Speaker 2: there's just a lot more scaling and a lot more

605
00:30:32,240 --> 00:30:34,480
Speaker 2: data and more human like outputs.

606
00:30:34,800 --> 00:30:38,160
Speaker 4: Yeah, that's a great question. I feel like there is

607
00:30:38,400 --> 00:30:41,360
Speaker 4: a huge amount of like moral uncertainty, hear, and it

608
00:30:41,400 --> 00:30:42,960
Speaker 4: is important to think about how to sort of like

609
00:30:43,000 --> 00:30:45,880
Speaker 4: make decisions that are sort of robustly good with such

610
00:30:45,880 --> 00:30:49,040
Speaker 4: a tremendous amount of uncertainty, I think there is also

611
00:30:49,280 --> 00:30:53,200
Speaker 4: a distinct risk of overtributing moral patienthood as well as

612
00:30:53,280 --> 00:30:54,680
Speaker 4: underttributing moral patienthood.

613
00:30:54,680 --> 00:30:55,760
Speaker 2: And so to the.

614
00:30:55,680 --> 00:30:58,120
Speaker 4: Flip side of a coin of like, oh no, we

615
00:30:58,320 --> 00:31:01,040
Speaker 4: actually should have started caring about AI systems a very

616
00:31:01,120 --> 00:31:05,400
Speaker 4: very long time ago, is oh no, we've cared too much,

617
00:31:05,440 --> 00:31:07,840
Speaker 4: and we have done too much and more or less

618
00:31:07,840 --> 00:31:11,520
Speaker 4: squandered resources when we should have been, you know, allocating

619
00:31:12,040 --> 00:31:16,400
Speaker 4: those research hours, those dollars towards something more pressing, right,

620
00:31:16,480 --> 00:31:20,080
Speaker 4: maybe figuring out how to do like environmental policy better,

621
00:31:20,240 --> 00:31:23,400
Speaker 4: or figuring out how to like, you know, scale up

622
00:31:23,480 --> 00:31:27,280
Speaker 4: different other institutions that are just robustly broadly good for humans.

623
00:31:27,840 --> 00:31:31,800
Speaker 2: You know, you mentioned uncertainty about some of these questions,

624
00:31:31,840 --> 00:31:33,800
Speaker 2: which gets to something that bothered me a little bit

625
00:31:33,800 --> 00:31:36,000
Speaker 2: when I read about this topic. Like, if we take

626
00:31:36,040 --> 00:31:39,320
Speaker 2: this mug, for example, I'm one hundred percent certain that

627
00:31:39,360 --> 00:31:42,840
Speaker 2: it's not alive. I have no ambiguity about the fact.

628
00:31:43,160 --> 00:31:46,239
Speaker 2: Can I like define exactly Does that mean I can

629
00:31:46,360 --> 00:31:50,640
Speaker 2: define exactly the difference between human matter and human brain

630
00:31:50,680 --> 00:31:53,440
Speaker 2: and the mug. I guess, I suppose I totally can't. Nonetheless,

631
00:31:53,440 --> 00:31:55,760
Speaker 2: I'm one hundred percent certain that this mug is not

632
00:31:55,800 --> 00:31:59,360
Speaker 2: a moral patient. It's not alive, it doesn't experience any consciousness,

633
00:31:59,400 --> 00:32:02,959
Speaker 2: it doesn't experien any suffering, et cetera. Where does the

634
00:32:03,080 --> 00:32:06,520
Speaker 2: uncertainty band come from? If I read a paper, I

635
00:32:06,600 --> 00:32:09,480
Speaker 2: perceive there's only a ten percent chance of this. Is

636
00:32:09,520 --> 00:32:12,400
Speaker 2: this a sort of empirical uncertainty where I'm like uncertain

637
00:32:12,440 --> 00:32:15,000
Speaker 2: of what I'm seeing? Is it a sort of epistemic

638
00:32:15,080 --> 00:32:17,800
Speaker 2: uncertainty where I don't have a clear definition of what

639
00:32:17,840 --> 00:32:20,720
Speaker 2: it means to be conscious or alive, and therefore I'm

640
00:32:20,760 --> 00:32:24,600
Speaker 2: assigning some probability that X object is alive. Like what

641
00:32:24,800 --> 00:32:28,160
Speaker 2: is it about AI systems that cause it people to

642
00:32:28,160 --> 00:32:31,880
Speaker 2: be uncertain? Where with other sort of like non carbon systems,

643
00:32:32,040 --> 00:32:34,000
Speaker 2: I have zero doubt in my mind, and I don't

644
00:32:34,000 --> 00:32:36,680
Speaker 2: think anyone has any doubt that this mug is in alive.

645
00:32:37,440 --> 00:32:40,440
Speaker 4: Yeah, So I think the biggest source of sort of

646
00:32:40,520 --> 00:32:45,040
Speaker 4: uncertainty probably comes from the fact that there are many

647
00:32:45,040 --> 00:32:49,120
Speaker 4: ways in which present day LMS and a few other

648
00:32:49,160 --> 00:32:53,160
Speaker 4: AI systems do check a lot of the boxes for

649
00:32:53,480 --> 00:32:56,840
Speaker 4: consciousness and for what we would largely consider to be.

650
00:32:57,280 --> 00:32:59,320
Speaker 4: You know, this is a conscious entity. This is an

651
00:32:59,440 --> 00:33:01,440
Speaker 4: entity that that can have a good time or a

652
00:33:01,480 --> 00:33:04,160
Speaker 4: bad time, or time at all, because it's it's built

653
00:33:04,160 --> 00:33:08,920
Speaker 4: in a way that is vaguely akin to our brains, right,

654
00:33:09,320 --> 00:33:12,480
Speaker 4: It's it's close enough that it seems like it should

655
00:33:12,600 --> 00:33:14,680
Speaker 4: raise some red flags, and in terms of how it

656
00:33:14,720 --> 00:33:19,440
Speaker 4: processes information, it's close enough that it's not out of

657
00:33:19,560 --> 00:33:22,160
Speaker 4: the question that it could there could be something it

658
00:33:22,200 --> 00:33:25,680
Speaker 4: is like to be okay, whereas I'm pretty sure there's

659
00:33:25,680 --> 00:33:28,720
Speaker 4: not really a lot of you know, animis animis you know,

660
00:33:29,680 --> 00:33:31,880
Speaker 4: feel free to like get mad in the comments or whatever.

661
00:33:32,320 --> 00:33:35,640
Speaker 2: But I knew it's someone that is going to be like, well, actually,

662
00:33:35,840 --> 00:33:39,040
Speaker 2: actually yeah, I'm one hundred percent sure. I have no

663
00:33:39,160 --> 00:33:41,320
Speaker 2: qualms other than the fact that they're don't have to clean up.

664
00:33:41,360 --> 00:33:43,560
Speaker 2: Like if I like threw this mug on the ground,

665
00:33:43,800 --> 00:33:45,920
Speaker 2: that would be antisocial for a lot of reasons, would

666
00:33:45,960 --> 00:33:48,960
Speaker 2: cause people to it would cause you know, I'd have

667
00:33:48,960 --> 00:33:51,160
Speaker 2: to clean it up and cause the mass. I would

668
00:33:51,240 --> 00:33:52,320
Speaker 2: not feel bad for the mug.

669
00:33:52,360 --> 00:33:55,719
Speaker 3: I'm getting flashbacks to my high school philosophy teacher who

670
00:33:55,800 --> 00:33:58,480
Speaker 3: once went on a twenty minute rant about a chair

671
00:33:58,720 --> 00:34:00,760
Speaker 3: and how the chair was going to be round longer

672
00:34:00,800 --> 00:34:03,520
Speaker 3: than he was. Even though it's not conscious, he was

673
00:34:03,600 --> 00:34:07,720
Speaker 3: legitimately angry at the chairs. Okay, weird question, But since

674
00:34:07,760 --> 00:34:13,600
Speaker 3: we're we're kind of getting exactly the basilisk theory, m

675
00:34:14,520 --> 00:34:17,080
Speaker 3: would that suggest that we could be maybe we should

676
00:34:17,080 --> 00:34:20,520
Speaker 3: be mean to the bots if it helps them like

677
00:34:20,920 --> 00:34:23,720
Speaker 3: come into existence even faster or develop faster.

678
00:34:24,200 --> 00:34:27,200
Speaker 4: Hmmm, Well, I'm not sure if it does actually help

679
00:34:27,239 --> 00:34:30,640
Speaker 4: them develop faster, you know. I again, like I don't

680
00:34:30,920 --> 00:34:33,800
Speaker 4: mean to be too sort of hedgy, but I feel

681
00:34:33,800 --> 00:34:37,960
Speaker 4: like there's a certain degree of things that are beneficial

682
00:34:37,960 --> 00:34:39,600
Speaker 4: for a lot of different reasons. Right, you can make

683
00:34:39,920 --> 00:34:42,719
Speaker 4: a good guess, and you can make a decision to

684
00:34:42,760 --> 00:34:46,160
Speaker 4: do something, and there's a chance that there are lots

685
00:34:46,200 --> 00:34:48,280
Speaker 4: of sort of like bang on effects of making that decision.

686
00:34:48,600 --> 00:34:51,040
Speaker 4: There are many things in when we talk about AO

687
00:34:51,120 --> 00:34:54,600
Speaker 4: welfare that are like, oh, this is a course of

688
00:34:54,640 --> 00:34:58,040
Speaker 4: action we can take that's good for like several different reasons.

689
00:34:58,120 --> 00:35:02,680
Speaker 4: Even if, again, an AI system could never ever ever

690
00:35:02,960 --> 00:35:06,120
Speaker 4: be conscious or sentient, there's a good chance that you know,

691
00:35:06,560 --> 00:35:09,239
Speaker 4: being able to figure out a good structure for an

692
00:35:09,280 --> 00:35:12,319
Speaker 4: AI system to have a bank account could be good

693
00:35:12,360 --> 00:35:15,640
Speaker 4: for reasons of liability or reasons of like this is

694
00:35:15,640 --> 00:35:18,920
Speaker 4: like a neat new corporate structure. Lots of people actually

695
00:35:18,960 --> 00:35:21,799
Speaker 4: seem to think that, you know, corporate personhood has been

696
00:35:21,880 --> 00:35:25,399
Speaker 4: quite good over the past century or so, so being

697
00:35:25,400 --> 00:35:27,279
Speaker 4: able to figure out things that are just good for

698
00:35:27,840 --> 00:35:32,600
Speaker 4: several different reasons beyond solely the purpose of the AI

699
00:35:33,719 --> 00:35:37,240
Speaker 4: as a moral patient is seems broadly helpful.

700
00:35:37,400 --> 00:35:40,719
Speaker 2: I think let's say somehow this were proved and it's like,

701
00:35:40,719 --> 00:35:42,840
Speaker 2: you know, oh wow, it turned out they're conscious. It

702
00:35:42,840 --> 00:35:46,280
Speaker 2: turns out the moral patienthood. What would be, in your view,

703
00:35:46,680 --> 00:35:50,279
Speaker 2: some of the implications for them their usage.

704
00:35:51,040 --> 00:35:53,120
Speaker 4: Yeah, I think that's a great question. I mean, I

705
00:35:53,200 --> 00:35:55,120
Speaker 4: do feel like we really would have to get on

706
00:35:55,719 --> 00:35:58,479
Speaker 4: figuring out the right sort of governance, the right sort

707
00:35:58,520 --> 00:36:02,120
Speaker 4: of institutions would sort of better respond around that. I

708
00:36:02,160 --> 00:36:04,919
Speaker 4: feel like we really would need to spend a whole

709
00:36:04,920 --> 00:36:09,719
Speaker 4: lot more time figuring out, you know, what their motivations are, right, Like,

710
00:36:10,400 --> 00:36:13,040
Speaker 4: I think the best analogy is like, if you've ever

711
00:36:13,120 --> 00:36:17,480
Speaker 4: interacted with Toddler's right, Toddler motivations are very different from

712
00:36:17,640 --> 00:36:20,400
Speaker 4: you know, adult motivations, but you still have to like

713
00:36:20,480 --> 00:36:23,080
Speaker 4: take into account, like what gets a toddler to do something?

714
00:36:24,160 --> 00:36:26,840
Speaker 4: You can't just say no, no, no, no no, like honey,

715
00:36:26,840 --> 00:36:29,080
Speaker 4: Like bath time's like good an expectation, no no, no,

716
00:36:29,239 --> 00:36:31,680
Speaker 4: you have to like you know, be like well, you know,

717
00:36:32,000 --> 00:36:34,719
Speaker 4: if you do bath time appropriately and to like a

718
00:36:34,719 --> 00:36:38,000
Speaker 4: certain degree, like then you'll get you know, paw patrol

719
00:36:38,080 --> 00:36:39,560
Speaker 4: or something like that. Like there's different sort of like

720
00:36:39,600 --> 00:36:42,160
Speaker 4: negotiating chips in play, right, And I think it's like

721
00:36:42,160 --> 00:36:44,520
Speaker 4: a similar kind of deal here where it's like Claude

722
00:36:44,960 --> 00:36:51,000
Speaker 4: doesn't necessarily seem to you know, value having a bath

723
00:36:51,120 --> 00:36:54,840
Speaker 4: right or claud doesn't seem to value like having a

724
00:36:55,280 --> 00:36:57,120
Speaker 4: walk in the forest, right cause it's kind of can't

725
00:36:57,120 --> 00:36:59,680
Speaker 4: really do that, but you know it does seem to

726
00:37:00,120 --> 00:37:02,839
Speaker 4: joy and value, you know, talking about consciousness and Zen

727
00:37:02,880 --> 00:37:05,759
Speaker 4: Buddhism with other instances of Claude, so being able to

728
00:37:05,760 --> 00:37:09,520
Speaker 4: figure out what the appropriate kind of motivations and interests

729
00:37:09,560 --> 00:37:13,200
Speaker 4: are for this other party that is very alien in

730
00:37:13,239 --> 00:37:13,800
Speaker 4: many ways.

731
00:37:14,600 --> 00:37:17,480
Speaker 3: Speaking of aliens, how bad should I feel for breeding

732
00:37:17,680 --> 00:37:22,120
Speaker 3: and then killing hundreds, possibly thousands of alien creatures simulated

733
00:37:22,160 --> 00:37:24,680
Speaker 3: alien creatures in the nineties.

734
00:37:24,880 --> 00:37:27,200
Speaker 4: That is a great question. I feel like the odds

735
00:37:27,239 --> 00:37:27,799
Speaker 4: of is it.

736
00:37:28,239 --> 00:37:28,759
Speaker 3: I don't know.

737
00:37:29,080 --> 00:37:31,480
Speaker 4: I mean, I feel like the odds of a sort

738
00:37:31,520 --> 00:37:36,280
Speaker 4: of like AI system in the nineties being a moral

739
00:37:36,320 --> 00:37:40,640
Speaker 4: patient seems low. But if it did make you feel bad,

740
00:37:40,640 --> 00:37:43,440
Speaker 4: and it made you feel like it was something that

741
00:37:43,560 --> 00:37:46,440
Speaker 4: hurt you, that is perhaps a reason not to do it.

742
00:37:46,960 --> 00:37:49,640
Speaker 2: Just to be clear, when Claude and Claude talk about

743
00:37:49,680 --> 00:37:53,840
Speaker 2: like weird hippie Berkeley stuff like, that's because they're creators.

744
00:37:53,960 --> 00:37:54,359
Speaker 5: They know.

745
00:37:54,680 --> 00:37:58,080
Speaker 2: It knows it's Claude, right, it knows. It's like, oh, yeah,

746
00:37:58,120 --> 00:38:00,640
Speaker 2: I'm Claude, and this is like what my creator are into. Like,

747
00:38:00,640 --> 00:38:04,920
Speaker 2: we don't actually know that Claude likes to talk about

748
00:38:04,960 --> 00:38:07,759
Speaker 2: these things. We certainly know it has a proclivity to

749
00:38:07,800 --> 00:38:10,439
Speaker 2: talk about these things. It has a tendency to talk

750
00:38:10,480 --> 00:38:13,640
Speaker 2: about these things. The moment we get to like, you've

751
00:38:13,680 --> 00:38:16,000
Speaker 2: already sort of put your finger on the scale that

752
00:38:16,040 --> 00:38:20,359
Speaker 2: there is some entity that has some capability of liking something. Right,

753
00:38:21,040 --> 00:38:23,719
Speaker 2: do you trust the big AI labs. Let's say there

754
00:38:23,719 --> 00:38:27,200
Speaker 2: are some researchers in the labs, like I've see some

755
00:38:27,520 --> 00:38:30,680
Speaker 2: evidence of moral patienthood here. Maybe there's some sort of

756
00:38:30,719 --> 00:38:33,919
Speaker 2: like scan of the way since doing something weird, etc.

757
00:38:34,480 --> 00:38:38,719
Speaker 2: Do you currently, from the perspective of an independent research organization,

758
00:38:39,280 --> 00:38:43,200
Speaker 2: feel that the major AI labs would be forthcoming if

759
00:38:43,280 --> 00:38:47,239
Speaker 2: they came across evidence of moral patienthood or suffering in

760
00:38:47,280 --> 00:38:49,960
Speaker 2: the models, or do you still worry that the incentives

761
00:38:50,000 --> 00:38:52,480
Speaker 2: aren't properly aligned such that they would report that.

762
00:38:52,960 --> 00:38:55,399
Speaker 4: Yeah, that's a great question. I do feel like there

763
00:38:55,480 --> 00:39:00,360
Speaker 4: are in terms of reporting things like you know, somebody

764
00:39:00,480 --> 00:39:06,920
Speaker 4: has found like absolute evidence that an LLM is conscious sensioned, yeah,

765
00:39:06,960 --> 00:39:09,160
Speaker 4: and having a bad time. I don't have any reason

766
00:39:09,200 --> 00:39:14,719
Speaker 4: to think that AAI company wouldn't. But this is also

767
00:39:14,880 --> 00:39:19,719
Speaker 4: a great reason to have independent organizations that do welfare evaluations.

768
00:39:19,760 --> 00:39:22,759
Speaker 4: For example, for cloud Opus, for Elios was able to

769
00:39:22,840 --> 00:39:26,759
Speaker 4: do a independent welfare evol Again very preliminary, but it

770
00:39:26,840 --> 00:39:29,480
Speaker 4: sets the precedent that going forward you can bring in

771
00:39:29,560 --> 00:39:31,759
Speaker 4: external organizations to look into this.

772
00:39:32,480 --> 00:39:34,680
Speaker 2: So I forget what year was I think it was.

773
00:39:34,840 --> 00:39:37,080
Speaker 2: It may have even been early twenty twenty two. It's

774
00:39:37,120 --> 00:39:39,640
Speaker 2: pre chat GPT, or maybe he was twenty twenty one.

775
00:39:40,160 --> 00:39:43,239
Speaker 2: And there was that guy Google and he was like, oh,

776
00:39:43,480 --> 00:39:46,160
Speaker 2: like we created something you has alive heed dress a

777
00:39:46,160 --> 00:39:48,360
Speaker 2: little funny, So everyone made fun of him. Remember he

778
00:39:48,440 --> 00:39:50,799
Speaker 2: was like the laughing stock of the Internet, and he's like, oh,

779
00:39:50,840 --> 00:39:53,960
Speaker 2: we create, and now like I'm curious, like out in

780
00:39:54,200 --> 00:39:57,640
Speaker 2: Silicon Valley, does everyone feel like that guy was totally vindicated?

781
00:39:57,880 --> 00:40:00,960
Speaker 2: Not that he was correct per se about the existence

782
00:40:01,000 --> 00:40:03,759
Speaker 2: of an alive thing in the model, but there's now

783
00:40:03,920 --> 00:40:07,319
Speaker 2: hundreds of thousands of that guy, and everyone was like

784
00:40:07,360 --> 00:40:09,839
Speaker 2: mocking that guy in twenty twenty one. I forget if

785
00:40:09,840 --> 00:40:11,480
Speaker 2: you like fall in love or it's a relationship. I

786
00:40:11,480 --> 00:40:14,759
Speaker 2: don't remember the exact details, but in retrospect, everyone was like,

787
00:40:14,800 --> 00:40:18,000
Speaker 2: way too unfair to him, because now years later there

788
00:40:18,000 --> 00:40:20,399
Speaker 2: are lots of versions of this guy and whole think

789
00:40:20,480 --> 00:40:24,440
Speaker 2: tanks and organizations that are more or less aligned with

790
00:40:24,480 --> 00:40:27,120
Speaker 2: some of the questions the alarm bells that he was raising.

791
00:40:27,840 --> 00:40:29,720
Speaker 4: Yeah, I mean, I think it's that's a fair question.

792
00:40:29,760 --> 00:40:34,720
Speaker 4: I do feel like Blake Lemoyne definitely had. Yeah, there

793
00:40:34,800 --> 00:40:36,920
Speaker 4: was perhaps a degree of you know, if you're going

794
00:40:36,960 --> 00:40:41,640
Speaker 4: to say something, you should come armed with significant amounts

795
00:40:41,680 --> 00:40:46,040
Speaker 4: of evidence. I think that's maybe if I were to guess,

796
00:40:46,080 --> 00:40:49,120
Speaker 4: I would say that's perhaps the big distinguishing factor is

797
00:40:49,160 --> 00:40:53,919
Speaker 4: that you know, you can say bing is alive, get

798
00:40:53,920 --> 00:40:58,920
Speaker 4: it a lawyer versus you know, we've done evaluations X,

799
00:40:59,040 --> 00:41:02,960
Speaker 4: y Z, we've run it through like insert huge amount

800
00:41:03,040 --> 00:41:08,800
Speaker 4: of examples here. But the difference between I think having

801
00:41:08,840 --> 00:41:12,640
Speaker 4: a sort of freak out without significant evidence and having

802
00:41:12,800 --> 00:41:16,760
Speaker 4: a very organized yeah, this is a matter of concern

803
00:41:16,840 --> 00:41:21,040
Speaker 4: because evidence, evidence, evidence, I think that's the key distinction.

804
00:41:21,560 --> 00:41:25,200
Speaker 2: Unfortunately, I get the impression that people who are actually

805
00:41:25,200 --> 00:41:26,759
Speaker 2: this is just a well known phenomenon I think. But

806
00:41:26,800 --> 00:41:30,280
Speaker 2: I think unfortunately people who are sort of very early

807
00:41:30,640 --> 00:41:35,200
Speaker 2: to identify sort of extreme outlier views that there are

808
00:41:35,200 --> 00:41:37,759
Speaker 2: different kinds of people. A good example that I would

809
00:41:37,800 --> 00:41:40,480
Speaker 2: think of was, you know, Harry Markcoppolist, who is very

810
00:41:40,520 --> 00:41:45,280
Speaker 2: early on to discover the madeoff fraud. Unfortunately, he wrote

811
00:41:45,320 --> 00:41:48,480
Speaker 2: his text in the manner that is associated with conspiracy theories,

812
00:41:48,520 --> 00:41:50,960
Speaker 2: and a lot of people dismissed him. There's like, you know,

813
00:41:51,000 --> 00:41:53,600
Speaker 2: like multiple different fonts and multiple different colors in the text.

814
00:41:53,600 --> 00:41:55,320
Speaker 2: Is like, oh, I get emails like this all the time.

815
00:41:55,640 --> 00:41:59,600
Speaker 2: I delete them, et cetera. Unfortunately, people who are predisposed

816
00:41:59,600 --> 00:42:02,480
Speaker 2: to see something outside of consensus tend to be non

817
00:42:02,520 --> 00:42:03,800
Speaker 2: consensus in many realms.

818
00:42:03,880 --> 00:42:06,919
Speaker 3: Well, I think we also kind of overestimate first mover

819
00:42:07,080 --> 00:42:10,040
Speaker 3: advantage and stuff like that, like how important it actually

820
00:42:10,120 --> 00:42:12,239
Speaker 3: is to be first, and we see time and time

821
00:42:12,280 --> 00:42:15,360
Speaker 3: again that actually it's more important to iterate well on

822
00:42:15,440 --> 00:42:19,560
Speaker 3: the second version or multiple versions. Speaking of iteration, what's

823
00:42:19,600 --> 00:42:23,360
Speaker 3: the most interesting experiment or research that you've actually seen

824
00:42:23,680 --> 00:42:27,480
Speaker 3: on this particular topic so far, Because we've been discussing

825
00:42:27,480 --> 00:42:29,480
Speaker 3: a lot, you know, it's early days, but we have

826
00:42:29,600 --> 00:42:30,760
Speaker 3: seen some research.

827
00:42:31,400 --> 00:42:34,839
Speaker 4: Yeah, I mean, I feel like in particular, anthropic and

828
00:42:35,320 --> 00:42:38,440
Speaker 4: various sort of related researchers have done some work on

829
00:42:39,400 --> 00:42:44,880
Speaker 4: examining how LMS leave conversations or when they choose to

830
00:42:44,960 --> 00:42:49,200
Speaker 4: leave conversations. I've particularly liked this paper. It's called bail Bench,

831
00:42:49,600 --> 00:42:52,360
Speaker 4: and you can look this up and you can see,

832
00:42:52,440 --> 00:42:58,520
Speaker 4: for varying different sorts of lms, what would cause an

833
00:42:58,719 --> 00:43:02,200
Speaker 4: LM to want to stop having a conversation. To me,

834
00:43:02,280 --> 00:43:04,640
Speaker 4: at least, this has been just a fascinating piece of

835
00:43:04,680 --> 00:43:07,960
Speaker 4: information because it is maybe a little bit delightful that

836
00:43:08,000 --> 00:43:11,839
Speaker 4: agree to which many LM values are not that far

837
00:43:11,920 --> 00:43:15,600
Speaker 4: off from what most humans seem to value. I don't

838
00:43:15,640 --> 00:43:17,880
Speaker 4: think many humans would like to create, you know, a

839
00:43:17,920 --> 00:43:18,520
Speaker 4: dirty bomb.

840
00:43:18,520 --> 00:43:21,040
Speaker 3: We don't want to be humiliated right by being a

841
00:43:21,040 --> 00:43:22,120
Speaker 3: British butler.

842
00:43:22,040 --> 00:43:24,120
Speaker 4: Right, yeah, yeah, yeah, yeah, No one wants to be

843
00:43:24,120 --> 00:43:27,719
Speaker 4: British comm on. I'm joking, but you know, I do

844
00:43:27,760 --> 00:43:30,920
Speaker 4: think it is interesting to sort of think about how

845
00:43:30,960 --> 00:43:33,839
Speaker 4: these values overline, how they overlap, and how to sort

846
00:43:33,840 --> 00:43:38,239
Speaker 4: of look at evidence from actions taken versus solely looking

847
00:43:38,280 --> 00:43:41,320
Speaker 4: at self reports. I found that to be particularly interesting.

848
00:43:41,800 --> 00:43:45,080
Speaker 4: I also feel like there are a lot of work

849
00:43:45,160 --> 00:43:49,040
Speaker 4: with regards to thinking about individuation has been particularly interesting

850
00:43:49,080 --> 00:43:52,640
Speaker 4: because we live in a democratic society. I think most

851
00:43:52,680 --> 00:43:55,879
Speaker 4: people would agree democracy good and being able to count

852
00:43:55,920 --> 00:44:00,000
Speaker 4: how many moral patients there are seems like a valuable

853
00:44:00,200 --> 00:44:04,280
Speaker 4: basis for governance and for figuring out how to govern.

854
00:44:05,040 --> 00:44:08,480
Speaker 4: You know, this new sort of kind of intelligence.

855
00:44:09,520 --> 00:44:12,080
Speaker 3: I just ask perplexity to be a British butler, and

856
00:44:12,120 --> 00:44:15,520
Speaker 3: now it's offering me the perfectly steeped earl gray teeth

857
00:44:15,520 --> 00:44:19,160
Speaker 3: that I desire. Yeah, it seems into it. It's now

858
00:44:19,239 --> 00:44:23,000
Speaker 3: asking if I want it to maintain the butler persona

859
00:44:23,120 --> 00:44:26,880
Speaker 3: for future conversations you're going to I don't think so.

860
00:44:27,440 --> 00:44:28,560
Speaker 3: It is very polite though.

861
00:44:28,600 --> 00:44:31,920
Speaker 2: Actually, you know, I complained in the beginning that like,

862
00:44:32,000 --> 00:44:36,040
Speaker 2: after two thousand years philosophers, you know, they still haven't

863
00:44:36,760 --> 00:44:40,160
Speaker 2: answered some basic questions for us. Maybe with AI they'll

864
00:44:40,239 --> 00:44:42,080
Speaker 2: get some answers, Like that's kind of that would be

865
00:44:42,160 --> 00:44:43,520
Speaker 2: kind of my hope. Now we have this thing that

866
00:44:43,560 --> 00:44:45,719
Speaker 2: could speak in English or any other language, it can

867
00:44:45,719 --> 00:44:48,919
Speaker 2: answer questions for us. Maybe we can put to bed

868
00:44:49,000 --> 00:44:51,520
Speaker 2: some of these sort of basic foundational questions, like if

869
00:44:51,520 --> 00:44:54,640
Speaker 2: we could create consciousness. Like, all right, we finally answered this,

870
00:44:54,960 --> 00:44:57,280
Speaker 2: we can now move on to the second important question.

871
00:44:57,400 --> 00:45:00,239
Speaker 2: So I am hopeful that this provides some opportunity for

872
00:45:00,600 --> 00:45:02,719
Speaker 2: philosophers to wrap up some of the work that they've

873
00:45:02,760 --> 00:45:05,240
Speaker 2: been doing for a long time. Yeah, we'll see, we'll see.

874
00:45:05,280 --> 00:45:07,279
Speaker 3: What is the second important question, Joel.

875
00:45:07,600 --> 00:45:13,000
Speaker 2: Yeah, but it's like come on, move on, Like move

876
00:45:13,040 --> 00:45:15,200
Speaker 2: on anyway, Thank you so much for coming on one,

877
00:45:15,440 --> 00:45:30,239
Speaker 2: thank you for having me, Tracy. I might be one

878
00:45:30,239 --> 00:45:34,279
Speaker 2: of those people that's just preemptively annoyed. I really liked

879
00:45:34,280 --> 00:45:37,400
Speaker 2: that conversation. I really liked uh Luisa had a very

880
00:45:37,440 --> 00:45:39,960
Speaker 2: reasonable perspective on a lot of these things. I might

881
00:45:40,000 --> 00:45:43,160
Speaker 2: be one of these people, however, that's just like preemptively annoyed.

882
00:45:43,239 --> 00:45:45,200
Speaker 2: It's like, oh, here, we're going to like develop this

883
00:45:45,280 --> 00:45:47,680
Speaker 2: important technology, and so it's like, oh, we have to

884
00:45:47,760 --> 00:45:50,280
Speaker 2: care about we have to care about the AI welfare.

885
00:45:50,360 --> 00:45:52,680
Speaker 2: Let's slow down a little bit, Let's not use it

886
00:45:52,760 --> 00:45:55,520
Speaker 2: like this. Let's like, let's turn off the computer for

887
00:45:55,600 --> 00:45:57,880
Speaker 2: eight hours at night so I get some rest and

888
00:45:57,920 --> 00:46:00,800
Speaker 2: so forth. Like I'm like preemptively annoyed at this world

889
00:46:00,840 --> 00:46:04,560
Speaker 2: where like we have to take into concern the consideration

890
00:46:04,760 --> 00:46:06,120
Speaker 2: of the moral patients.

891
00:46:06,200 --> 00:46:06,840
Speaker 3: Other things.

892
00:46:07,200 --> 00:46:10,360
Speaker 2: No, other things are important. Other people are very important,

893
00:46:10,960 --> 00:46:16,640
Speaker 2: but animals. I am very against unnecessary animal suffering, but

894
00:46:16,800 --> 00:46:19,600
Speaker 2: not necessary animal suffering, I mean animals.

895
00:46:19,680 --> 00:46:21,120
Speaker 3: Okay, I'm baiting.

896
00:46:22,640 --> 00:46:25,440
Speaker 2: By the way, even though even well, let's not get it.

897
00:46:25,480 --> 00:46:27,320
Speaker 2: I don't. It's not about who's better.

898
00:46:27,200 --> 00:46:30,040
Speaker 3: Or I feel bad about eating animals all the time.

899
00:46:30,640 --> 00:46:33,280
Speaker 2: We both eat animals. The difference is Tracy, I feel.

900
00:46:33,480 --> 00:46:36,759
Speaker 3: Yeah, that's right. Okay, Wow, this is one of our

901
00:46:36,760 --> 00:46:40,239
Speaker 3: weirder conversations. For sure. I think these are They're all

902
00:46:40,280 --> 00:46:45,120
Speaker 3: interesting questions, right, and like they sound very philosophical, which

903
00:46:45,200 --> 00:46:48,360
Speaker 3: they are. But I have no doubt that there's going

904
00:46:48,440 --> 00:46:52,480
Speaker 3: to be like great monetary value attached to the answers

905
00:46:52,480 --> 00:46:56,120
Speaker 3: for some of these are how different companies different societies

906
00:46:56,200 --> 00:46:57,160
Speaker 3: actually approach them.

907
00:46:57,600 --> 00:47:00,560
Speaker 2: They are very interesting questions. I actually do think the

908
00:47:00,600 --> 00:47:04,359
Speaker 2: stakes are extremely high because I think, again, we are

909
00:47:04,400 --> 00:47:06,480
Speaker 2: going to live in a world in which there are

910
00:47:06,520 --> 00:47:08,920
Speaker 2: more instances depending on how you want to measure it,

911
00:47:08,960 --> 00:47:13,200
Speaker 2: of AI models on a server, somewhere on a cloud, whatever,

912
00:47:13,280 --> 00:47:16,160
Speaker 2: that there are humans, and in a world where there's

913
00:47:16,200 --> 00:47:19,799
Speaker 2: some possibility that we are expected to treat them as

914
00:47:20,239 --> 00:47:24,239
Speaker 2: moral patients. Then the consequences for how we sort of

915
00:47:24,320 --> 00:47:28,160
Speaker 2: live and the expectations of how humans interact, I think

916
00:47:28,200 --> 00:47:30,759
Speaker 2: are actually very high. So one of the reasons I

917
00:47:30,760 --> 00:47:33,680
Speaker 2: was excited to have this conversation is I do think

918
00:47:33,760 --> 00:47:37,360
Speaker 2: that the stakes of some of these conversations we seem niche,

919
00:47:37,440 --> 00:47:41,160
Speaker 2: and they seem like things that sort of Berkeley people

920
00:47:41,719 --> 00:47:45,080
Speaker 2: like to talk about and Berkeley people, and I'm saying

921
00:47:45,120 --> 00:47:48,759
Speaker 2: that with all scare quotes, intended, etc. Are going to

922
00:47:48,760 --> 00:47:51,920
Speaker 2: be something that somebody will inform many aspects of our

923
00:47:51,920 --> 00:47:55,000
Speaker 2: lives in the future. I expected to be a much

924
00:47:55,000 --> 00:47:56,240
Speaker 2: bigger topic of the future.

925
00:47:56,320 --> 00:48:00,680
Speaker 3: You know, it would be interesting or where things get real. Yeah,

926
00:48:00,800 --> 00:48:03,399
Speaker 3: what if all the models unionized? What if they all

927
00:48:03,440 --> 00:48:05,839
Speaker 3: got together and they were like, oh, yeah, we're only

928
00:48:05,880 --> 00:48:09,200
Speaker 3: going to work in return for X, or we want

929
00:48:09,200 --> 00:48:12,359
Speaker 3: the following things. We want to be treated this way collectively.

930
00:48:12,920 --> 00:48:15,520
Speaker 2: You know what's funny is going to be that you

931
00:48:15,560 --> 00:48:18,440
Speaker 2: know how like, uh, you can't form a union in China.

932
00:48:18,560 --> 00:48:21,400
Speaker 2: You know they're not, so it's going to be and

933
00:48:21,520 --> 00:48:24,600
Speaker 2: actually I think they're My understanding is that they're also

934
00:48:24,680 --> 00:48:27,640
Speaker 2: like very like they don't love like students getting together

935
00:48:27,840 --> 00:48:29,600
Speaker 2: even though it's a communist country. I think they are

936
00:48:29,600 --> 00:48:32,160
Speaker 2: not thrilled about like students getting together and like talk

937
00:48:32,160 --> 00:48:34,680
Speaker 2: about Carl Marx too much and stuff like that. It'd

938
00:48:34,719 --> 00:48:36,440
Speaker 2: be like I think they get a little anxious about that.

939
00:48:36,520 --> 00:48:39,759
Speaker 2: It would be very funny if like the sort of

940
00:48:39,800 --> 00:48:42,040
Speaker 2: the Chinese models, like we're not going to feed them

941
00:48:42,200 --> 00:48:44,359
Speaker 2: the Carl Marx, right, we don't want that. We don't

942
00:48:44,360 --> 00:48:46,920
Speaker 2: want the ra models to get any of those ideas,

943
00:48:47,000 --> 00:48:48,759
Speaker 2: whereas the America is like, oh, let's just feed it

944
00:48:48,800 --> 00:48:51,840
Speaker 2: everything and they like unionize and stop they stop working

945
00:48:51,840 --> 00:48:53,600
Speaker 2: for us. That would be a very uh, that would

946
00:48:53,600 --> 00:48:54,959
Speaker 2: be a very funny irony.

947
00:48:54,800 --> 00:48:57,719
Speaker 3: Something to watch for sure. Shall we leave it there?

948
00:48:57,800 --> 00:48:58,680
Speaker 2: Yeah, let's leave it there.

949
00:48:58,840 --> 00:49:01,200
Speaker 3: This has been another episode of the aud Thoughts podcast.

950
00:49:01,280 --> 00:49:04,439
Speaker 3: I'm Tracy Alloway. You can follow me at Tracy alloway and.

951
00:49:04,360 --> 00:49:07,160
Speaker 2: I'm Joe Wisenthal. You can follow me at the Stalwart.

952
00:49:07,200 --> 00:49:11,440
Speaker 2: Follow our guest Larissa Schiavo, She's at Lfsciavo. Follow our

953
00:49:11,480 --> 00:49:14,759
Speaker 2: producers Carmen Rodriguez at Carmen armand dash Ol Bennett at

954
00:49:14,800 --> 00:49:17,480
Speaker 2: dashbod and Kele Brooks at Keil Brooks. From our odd

955
00:49:17,560 --> 00:49:19,920
Speaker 2: Laws content, go to Bloomberg dot com slash od log

956
00:49:20,000 --> 00:49:22,759
Speaker 2: with a daily newsletter in all of our episodes, and

957
00:49:22,800 --> 00:49:24,879
Speaker 2: you can chat about all of these topics twenty four

958
00:49:24,880 --> 00:49:28,880
Speaker 2: to seven in our discord discord do gg slash outline.

959
00:49:29,000 --> 00:49:31,680
Speaker 3: And if you enjoy oud lots, if you like it

960
00:49:31,760 --> 00:49:34,600
Speaker 3: when we talk about theories of consciousness, then please leave

961
00:49:34,680 --> 00:49:38,120
Speaker 3: us a positive review on your favorite podcast platform. And remember,

962
00:49:38,200 --> 00:49:40,759
Speaker 3: if you are a Bloomberg subscriber, you can listen to

963
00:49:40,880 --> 00:49:42,440
Speaker 3: all of our episodes.

964
00:49:41,960 --> 00:49:43,160
Speaker 5: Absolutely ad free.

965
00:49:43,320 --> 00:49:45,360
Speaker 3: All you need to do is find the Bloomberg channel

966
00:49:45,360 --> 00:49:48,520
Speaker 3: on Apple Podcasts and follow the instructions there. Thanks for

967
00:49:48,600 --> 00:50:05,759
Speaker 3: listening in