1
00:00:17,242 --> 00:00:17,722
Speaker 1: Pushkin.

2
00:00:23,922 --> 00:00:27,562
Speaker 2: Hello and welcome back to Drilled. I'm Amy Westerbelt. Today

3
00:00:27,642 --> 00:00:32,122
Speaker 2: we're bringing you another installment of our ongoing series Drilling Deep,

4
00:00:32,642 --> 00:00:35,922
Speaker 2: where we speak to authors of recent books that are

5
00:00:35,962 --> 00:00:39,362
Speaker 2: either about climate or about things that intersect with climate

6
00:00:39,442 --> 00:00:43,162
Speaker 2: in a big way. Today is a super timely one.

7
00:00:43,682 --> 00:00:47,882
Speaker 2: We're going to talk about AI. Specifically, we're going to

8
00:00:47,882 --> 00:00:52,282
Speaker 2: talk about Karen Howe's new book Empire of AI, Dreams

9
00:00:52,322 --> 00:00:57,962
Speaker 2: and Nightmares in Sam Altman's Open AI. In this episode,

10
00:00:58,122 --> 00:01:03,962
Speaker 2: Adam Lowenstein interviews journalist Howe, who argues that AI and

11
00:01:04,002 --> 00:01:09,922
Speaker 2: the profit driven infrastructure that surrounds it is a colonial project.

12
00:01:10,242 --> 00:01:14,082
Speaker 2: What openI boss Altman and his fellow ideologues in Silicon

13
00:01:14,122 --> 00:01:18,441
Speaker 2: Valley are pursuing, how says, is not just corporate power,

14
00:01:18,521 --> 00:01:22,722
Speaker 2: but imperial power. It makes sense when you think about

15
00:01:22,762 --> 00:01:27,842
Speaker 2: the fact that AI, at its core, its most basic

16
00:01:27,961 --> 00:01:32,401
Speaker 2: and ongoing purpose is about twenty four hour surveillance to

17
00:01:32,521 --> 00:01:36,802
Speaker 2: keep all us plates in line. These guys are building empires,

18
00:01:36,842 --> 00:01:40,602
Speaker 2: and as history shows, empires are built on resource extraction,

19
00:01:40,961 --> 00:01:47,202
Speaker 2: particularly the old fashioned kind labor, energy, minerals, land, and water.

20
00:01:48,082 --> 00:01:51,882
Speaker 2: It seems like almost overnight, big techts feel good climate

21
00:01:51,922 --> 00:01:58,002
Speaker 2: promises have evaporated. They've been swapped seamlessly for slippery promises

22
00:01:58,082 --> 00:02:03,602
Speaker 2: that so called artificial general intelligence will solve climate, never

23
00:02:03,722 --> 00:02:07,962
Speaker 2: mind that it's a fantastic concept that has no agreed

24
00:02:08,042 --> 00:02:12,802
Speaker 2: upon definition, or more fundamentally, that appears nowhere close to

25
00:02:12,922 --> 00:02:17,922
Speaker 2: actually existing. In big tech's frenzied pursuit the quote unquote

26
00:02:18,162 --> 00:02:23,642
Speaker 2: hyper scale AI dominance that evangelists claim will unlock AGI,

27
00:02:24,122 --> 00:02:28,562
Speaker 2: as well as its expanding alliances with fossil fuel backed

28
00:02:28,602 --> 00:02:33,802
Speaker 2: petro states, fossil fuel companies, and authoritarian political movements, it's

29
00:02:33,842 --> 00:02:38,482
Speaker 2: become an increasingly central contributor to the climate crisis. In

30
00:02:38,522 --> 00:02:42,402
Speaker 2: an October conversation with Drilled, How discussed how Silicon Valley

31
00:02:42,402 --> 00:02:45,402
Speaker 2: giants appear to be following the oil and gas industry's

32
00:02:45,442 --> 00:02:50,882
Speaker 2: playbook of disinformation and deceit, how Altman and Open AI's

33
00:02:50,882 --> 00:02:55,242
Speaker 2: secrecy and disingenuous rhetoric transformed the field of AI research

34
00:02:55,362 --> 00:02:59,642
Speaker 2: into corporate pr and why the destructive trajectory of AI

35
00:02:59,802 --> 00:03:04,722
Speaker 2: scale and commercialization is not inevitable, no matter what its

36
00:03:04,762 --> 00:03:08,842
Speaker 2: power hungry proponents would have You. Believe that conversation is

37
00:03:08,842 --> 00:03:12,482
Speaker 2: coming up after this quick break and find a condensed

38
00:03:12,802 --> 00:03:15,802
Speaker 2: written version of this interview on our website at drill

39
00:03:15,882 --> 00:03:22,842
Speaker 2: dot Media.

40
00:03:31,002 --> 00:03:32,962
Speaker 1: How does it feel having the book out? Because it's

41
00:03:33,122 --> 00:03:36,562
Speaker 1: it's a monumental achievement. I know you've gotten lots of

42
00:03:36,602 --> 00:03:39,482
Speaker 1: praise for them already, but it's just it's just epic

43
00:03:39,802 --> 00:03:42,402
Speaker 1: and I can't imagine the amount of work that went

44
00:03:42,442 --> 00:03:44,522
Speaker 1: into producing it.

45
00:03:45,402 --> 00:03:48,922
Speaker 3: Yeah, I was the most intense and painful thing that

46
00:03:48,962 --> 00:03:53,162
Speaker 3: I've ever worked on me, like emotionally and physically painful

47
00:03:53,202 --> 00:03:56,522
Speaker 3: because I was typing so much that I developed tendonitis

48
00:03:56,602 --> 00:04:02,202
Speaker 3: on my hands. But yeah, I know, it's I mean,

49
00:04:02,242 --> 00:04:06,802
Speaker 3: it's I think it's been interesting. Have you ever written

50
00:04:06,802 --> 00:04:07,642
Speaker 3: a book before?

51
00:04:08,442 --> 00:04:10,642
Speaker 1: It was not as long as Empire of Ai, but

52
00:04:10,722 --> 00:04:14,322
Speaker 1: I have some understanding of the like the long term

53
00:04:14,402 --> 00:04:16,681
Speaker 1: nature of a writing project like this, but not to

54
00:04:16,722 --> 00:04:19,082
Speaker 1: the same extent with the amount of reporting in the

55
00:04:19,242 --> 00:04:21,402
Speaker 1: yashion that you were wrestling with here.

56
00:04:22,282 --> 00:04:25,882
Speaker 3: The thing that I didn't really prepare for was like

57
00:04:26,162 --> 00:04:30,642
Speaker 3: the one to eighty flip from being completely alone and

58
00:04:30,882 --> 00:04:34,322
Speaker 3: isolated working on the book to like suddenly talking with

59
00:04:34,362 --> 00:04:38,322
Speaker 3: like thousands of people about the same thing once the

60
00:04:38,322 --> 00:04:42,482
Speaker 3: book was launched, And that has been both like a

61
00:04:42,602 --> 00:04:47,042
Speaker 3: very fun and new experience, and one that comes with

62
00:04:47,082 --> 00:04:48,962
Speaker 3: like a lot of relief that like people are resonating

63
00:04:48,962 --> 00:04:53,122
Speaker 3: with the thing, and also like a very uh like

64
00:04:53,202 --> 00:04:57,722
Speaker 3: slightly not slightly extremely chaotic transition, because like you go

65
00:04:57,882 --> 00:05:02,042
Speaker 3: from like mode of like total silence and not having

66
00:05:02,202 --> 00:05:05,722
Speaker 3: really feedback from anyone about anything that you're doing, to

67
00:05:05,802 --> 00:05:10,922
Speaker 3: suddenly getting bombarded with everyone's thoughts about what you've been

68
00:05:11,082 --> 00:05:14,601
Speaker 3: working on. So yeah, so it's like it's like a

69
00:05:14,842 --> 00:05:17,322
Speaker 3: it's like a strange mix of relief and and like

70
00:05:17,442 --> 00:05:20,002
Speaker 3: new sources of stress since the price.

71
00:05:21,122 --> 00:05:25,802
Speaker 1: Have you heard from either sources or industry players with

72
00:05:25,882 --> 00:05:27,042
Speaker 1: feedback about the book.

73
00:05:28,082 --> 00:05:32,762
Speaker 3: I've heard from a few sources who really loved the book,

74
00:05:32,762 --> 00:05:34,362
Speaker 3: which is always a relief. M M.

75
00:05:34,842 --> 00:05:37,362
Speaker 1: Yeah, you want to do them justice for sharing their stories.

76
00:05:37,602 --> 00:05:41,962
Speaker 3: Yeah, And I've I've heard from some like people within

77
00:05:42,082 --> 00:05:45,802
Speaker 3: industry that are not like exacts, but you know, employees

78
00:05:45,802 --> 00:05:49,082
Speaker 3: at these various companies that I've mentioned that were not sources,

79
00:05:49,122 --> 00:05:53,362
Speaker 3: but that have mentioned that they really appreciated the book.

80
00:05:53,562 --> 00:05:56,442
Speaker 3: But in terms of like formal responses from Open AI

81
00:05:56,922 --> 00:06:01,601
Speaker 3: for other companies, there's been nothing, which I much prefer

82
00:06:02,082 --> 00:06:06,681
Speaker 3: to like a lawsuit, So I'll take silence any day. Yeah,

83
00:06:06,722 --> 00:06:07,162
Speaker 3: for sure.

84
00:06:07,802 --> 00:06:12,282
Speaker 1: One of the things I found most troubling eye opening

85
00:06:12,722 --> 00:06:17,242
Speaker 1: the way that you convey how artificial intelligence or the

86
00:06:17,322 --> 00:06:21,722
Speaker 1: concept of it is an ideology in Silicon Valley and

87
00:06:21,762 --> 00:06:24,802
Speaker 1: in these companies. And one of the things that the

88
00:06:24,842 --> 00:06:28,562
Speaker 1: book does such a brilliant job communicating, and that I

89
00:06:28,602 --> 00:06:32,002
Speaker 1: wanted to convey in this conversation is the fact that

90
00:06:32,402 --> 00:06:35,921
Speaker 1: these are not like normal companies, and this is not

91
00:06:36,042 --> 00:06:39,882
Speaker 1: a normal technology, whatever normal means. Yeah, can you talk

92
00:06:39,922 --> 00:06:44,041
Speaker 1: a bit about how AI is an ideology and then

93
00:06:44,442 --> 00:06:50,162
Speaker 1: how these companies are already or are pursuing this status

94
00:06:50,242 --> 00:06:52,242
Speaker 1: as what you describe as empires.

95
00:06:53,042 --> 00:06:55,642
Speaker 3: Yeah, So if you think about just the fact that,

96
00:06:57,402 --> 00:06:58,962
Speaker 3: like if we look at the history of AI and

97
00:06:59,842 --> 00:07:02,762
Speaker 3: the founding of the discipline, the scientific discipline in nineteen

98
00:07:02,802 --> 00:07:06,202
Speaker 3: fifty six, even back then, it was already an ideological

99
00:07:06,802 --> 00:07:11,242
Speaker 3: project because it was a group of scientists deciding they

100
00:07:11,242 --> 00:07:15,242
Speaker 3: wanted to try and replicate human intelligence and computers. And

101
00:07:15,482 --> 00:07:21,162
Speaker 3: that's a very political choice, Like why do that? You know,

102
00:07:21,642 --> 00:07:24,842
Speaker 3: from a it's not really clear what the scientific reason is.

103
00:07:24,842 --> 00:07:28,802
Speaker 3: It is more of a of just like a there

104
00:07:28,882 --> 00:07:35,122
Speaker 3: was an intention behind that choosing that path and setting

105
00:07:35,202 --> 00:07:40,562
Speaker 3: up a field that defined its goal as attempting to

106
00:07:40,802 --> 00:07:48,002
Speaker 3: mimic and therefore ultimately potentially replace humans. And Joseph Weisenbauman,

107
00:07:48,402 --> 00:07:51,122
Speaker 3: MIT professor, who was kind of part of the initial

108
00:07:51,882 --> 00:07:54,642
Speaker 3: co word of scientists who did start researching AI, then

109
00:07:54,842 --> 00:07:58,682
Speaker 3: later quickly became a critic of the entire endeavor, still

110
00:07:58,762 --> 00:08:01,402
Speaker 3: back in the nineteen sixties, and he said at the time,

111
00:08:01,442 --> 00:08:06,122
Speaker 3: in a very prophetic way, there is something inherently disturbing

112
00:08:06,282 --> 00:08:10,522
Speaker 3: about a goal that intends to replace a thing that

113
00:08:10,522 --> 00:08:15,362
Speaker 3: we already have, because ultimately, all this is going to

114
00:08:15,362 --> 00:08:21,602
Speaker 3: do is allow people empower politicians and corporate CEOs to

115
00:08:21,762 --> 00:08:29,122
Speaker 3: impose their will through algorithms and use the AI system

116
00:08:29,442 --> 00:08:34,322
Speaker 3: as the load bearer of responsibility, essentially, like it allows

117
00:08:34,362 --> 00:08:36,122
Speaker 3: them to do whatever they want, but to say that

118
00:08:36,242 --> 00:08:39,842
Speaker 3: something else did it. And so he was already recognizing

119
00:08:39,881 --> 00:08:43,722
Speaker 3: back then that there's an ideological project behind the creation

120
00:08:43,921 --> 00:08:47,321
Speaker 3: of AI, and so fast forward all the way to today,

121
00:08:47,962 --> 00:08:51,761
Speaker 3: that ideology has not only continued to permeate and drive

122
00:08:52,202 --> 00:08:58,362
Speaker 3: the industry and its goals, it has now become much

123
00:08:58,401 --> 00:09:02,401
Speaker 3: more extreme because they're layered on top of this fundamental

124
00:09:02,401 --> 00:09:05,122
Speaker 3: assumption that it is somehow inherently good to recreate human

125
00:09:05,122 --> 00:09:08,522
Speaker 3: intelligence and computers. We now have a particular approach to

126
00:09:08,641 --> 00:09:13,041
Speaker 3: doing that that these companies index on, which is scaling

127
00:09:13,082 --> 00:09:16,441
Speaker 3: at these models at all costs, which is also an ideology.

128
00:09:16,482 --> 00:09:21,761
Speaker 3: There is no scientific basis for choosing that approach, and

129
00:09:21,801 --> 00:09:25,161
Speaker 3: there's no scientific reason for why these companies are trying

130
00:09:25,202 --> 00:09:30,042
Speaker 3: to dominate as monopolies. This is also based on these

131
00:09:30,681 --> 00:09:34,242
Speaker 3: worldviews and values that a certain small group of people

132
00:09:34,641 --> 00:09:39,241
Speaker 3: have chosen, and that group of people have an extraordinary

133
00:09:39,281 --> 00:09:44,081
Speaker 3: amount of power to make that choice affect everyone. I'm

134
00:09:44,122 --> 00:09:46,482
Speaker 3: really glad that you're asking about the ideological aspect of it,

135
00:09:46,521 --> 00:09:50,201
Speaker 3: because I think this is a really huge part of

136
00:09:50,242 --> 00:09:53,602
Speaker 3: the story of AI that has often missed. People keep

137
00:09:53,641 --> 00:09:56,282
Speaker 3: thinking that this is just a business story. It's you know,

138
00:09:56,362 --> 00:10:01,002
Speaker 3: Silicon value doing commercial things and ultimately trying to drive

139
00:10:01,362 --> 00:10:05,081
Speaker 3: for profit maximization. But that is only half the story

140
00:10:05,082 --> 00:10:09,321
Speaker 3: when it comes to AI. There is a deep ideological

141
00:10:09,401 --> 00:10:14,562
Speaker 3: drive to recreate human intelligence too, and a belief in

142
00:10:14,842 --> 00:10:18,202
Speaker 3: the idea that this is somehow going to create to

143
00:10:18,281 --> 00:10:21,921
Speaker 3: solve all of our problems or potentially destroy us, and

144
00:10:21,962 --> 00:10:23,562
Speaker 3: that in order to do all these things that we

145
00:10:23,681 --> 00:10:28,041
Speaker 3: have to consume the entire planet's resources and all of

146
00:10:28,082 --> 00:10:30,522
Speaker 3: those that need to be scrutinized for what they are,

147
00:10:30,602 --> 00:10:34,362
Speaker 3: which is actually just an extremely narrow view about how

148
00:10:34,401 --> 00:10:36,761
Speaker 3: the world is and how it should be.

149
00:10:38,482 --> 00:10:42,802
Speaker 1: The idea of artificial general intelligence or AGI, I think,

150
00:10:42,921 --> 00:10:46,201
Speaker 1: is something people hear about a lot these days because

151
00:10:46,561 --> 00:10:49,722
Speaker 1: obviously it's great for headlines, and it's also great for

152
00:10:50,641 --> 00:10:53,921
Speaker 1: CEOs of these companies to throw out as an idea,

153
00:10:54,161 --> 00:10:56,682
Speaker 1: you know, implying that it's already here or it's just

154
00:10:56,722 --> 00:10:59,482
Speaker 1: around the corner is a great way to boost a

155
00:10:59,482 --> 00:11:02,362
Speaker 1: stock price. But they also, as you show in the book,

156
00:11:02,401 --> 00:11:05,402
Speaker 1: they also use the idea of AGI is basically a

157
00:11:05,442 --> 00:11:09,201
Speaker 1: pretext for whatever they want to do. Can you talk

158
00:11:09,242 --> 00:11:12,722
Speaker 1: a little bit about what AGE is or as close

159
00:11:12,761 --> 00:11:15,482
Speaker 1: as we can come to a definition, and then how

160
00:11:15,602 --> 00:11:22,602
Speaker 1: they use this hypothetical concept to basically justify whatever it

161
00:11:22,641 --> 00:11:23,322
Speaker 1: is they want to do.

162
00:11:24,041 --> 00:11:29,082
Speaker 3: Yeah. So, the loosest definition of AGI is the point

163
00:11:29,122 --> 00:11:33,002
Speaker 3: at which an AI system fully encapsulates all the dimensions

164
00:11:33,002 --> 00:11:37,402
Speaker 3: of human intelligence, which was the original definition of AI.

165
00:11:37,561 --> 00:11:39,881
Speaker 3: And the reason why we now use the term AGI

166
00:11:39,921 --> 00:11:43,362
Speaker 3: instead of AI is because over the decades, the term

167
00:11:43,362 --> 00:11:45,881
Speaker 3: AI has sort of been cheapened by the fact that

168
00:11:45,921 --> 00:11:49,322
Speaker 3: companies keep using it to market existing products and services,

169
00:11:49,401 --> 00:11:52,202
Speaker 3: and so now AI has come to mean what we

170
00:11:52,281 --> 00:11:57,161
Speaker 3: have currently and AGI is supposed to indicate a return

171
00:11:57,281 --> 00:12:02,802
Speaker 3: to that original premise of this scientific discipline. But the

172
00:12:02,881 --> 00:12:07,962
Speaker 3: problem why AGI is so ill defined is because we

173
00:12:08,002 --> 00:12:12,362
Speaker 3: don't have scientific cosensus around what human intelligence is. There's

174
00:12:12,442 --> 00:12:19,681
Speaker 3: no signs, there's no neurological, biological, psychological definition for this

175
00:12:20,242 --> 00:12:23,041
Speaker 3: special characteristic that humans seem to have more of than

176
00:12:23,082 --> 00:12:27,242
Speaker 3: other animal species. And in fact, like when you think

177
00:12:27,242 --> 00:12:32,962
Speaker 3: about the history of trying to measure, quantify rank human intelligence,

178
00:12:33,122 --> 00:12:37,722
Speaker 3: it has always been driven by extremely dark, dark motives eugenics,

179
00:12:39,202 --> 00:12:42,921
Speaker 3: the justification of oppressing different groups of people. And so

180
00:12:43,362 --> 00:12:47,602
Speaker 3: this is why AGI remains an extremely nebulous milestone because

181
00:12:47,881 --> 00:12:51,682
Speaker 3: different people have totally different ideas of what does it

182
00:12:51,761 --> 00:12:55,961
Speaker 3: mean that we've finally achieved human level intelligence and open

183
00:12:56,002 --> 00:12:58,641
Speaker 3: eye jokes about like this has been a long standing

184
00:12:58,681 --> 00:13:02,041
Speaker 3: joke within the company. If you ask thirteen AI researchers

185
00:13:02,722 --> 00:13:06,242
Speaker 3: what AGI is, you'll get fifteen definitions, and so you know,

186
00:13:06,281 --> 00:13:09,042
Speaker 3: they're kind of self aware about the fact that this

187
00:13:09,242 --> 00:13:14,921
Speaker 3: lacks consensus, but it becomes a tool that the executives

188
00:13:15,002 --> 00:13:17,722
Speaker 3: use to their advantage, and Altman in particular in that

189
00:13:17,881 --> 00:13:21,442
Speaker 3: because there is no neither a definition within the company

190
00:13:21,482 --> 00:13:24,641
Speaker 3: that has full consensus nor externals of the company that

191
00:13:24,641 --> 00:13:29,002
Speaker 3: has full consensus, they can just paint these visions of

192
00:13:29,082 --> 00:13:35,322
Speaker 3: AGI based on whatever is necessary to overcome a particular

193
00:13:35,401 --> 00:13:38,801
Speaker 3: hurdle that they're facing in that moment. So when Altman

194
00:13:38,881 --> 00:13:41,122
Speaker 3: is speaking in front of Congress, you'll paint AGI as

195
00:13:41,161 --> 00:13:45,362
Speaker 3: this magical system that is like a magical want that

196
00:13:45,362 --> 00:13:48,161
Speaker 3: you can waive that's going to solve climate change, cure cancer,

197
00:13:48,322 --> 00:13:50,881
Speaker 3: do all these things that people have wanted for a

198
00:13:50,962 --> 00:13:53,682
Speaker 3: very very long time. And then when you're talking to

199
00:13:53,801 --> 00:13:59,242
Speaker 3: like CEOs, AGI suddenly an employee like something that they

200
00:13:59,242 --> 00:14:01,881
Speaker 3: can buy as a service so that they don't have

201
00:14:01,921 --> 00:14:04,562
Speaker 3: to hire real humans. And then when you're talking to

202
00:14:04,801 --> 00:14:07,921
Speaker 3: you know, an average consumer, AGI suddenly a friend or

203
00:14:08,041 --> 00:14:11,281
Speaker 3: it's an assistant or whatever it is that you know

204
00:14:11,401 --> 00:14:14,321
Speaker 3: compels that person to put down money or compels the

205
00:14:14,921 --> 00:14:20,322
Speaker 3: regulators to not put in regulation. And so AGI just

206
00:14:20,402 --> 00:14:23,362
Speaker 3: shape shifts and morphs in these ways that when you

207
00:14:23,442 --> 00:14:28,042
Speaker 3: actually list out all of the possible things that supposedly

208
00:14:28,082 --> 00:14:31,321
Speaker 3: AGI is that Altman or any other company executive has

209
00:14:31,322 --> 00:14:36,802
Speaker 3: ever said it's a completely incoherent vision of a technology,

210
00:14:37,082 --> 00:14:40,602
Speaker 3: like it just doesn't make sense, and like opening eye uses.

211
00:14:41,522 --> 00:14:46,162
Speaker 3: They've also written down in documents completely different definitions of

212
00:14:46,202 --> 00:14:50,122
Speaker 3: this as well, Like they just they officially say on

213
00:14:50,162 --> 00:14:54,282
Speaker 3: their website that AGI is highly autonomous systems that outperform

214
00:14:54,402 --> 00:14:57,962
Speaker 3: humans and most economically valuable work, So they are defining

215
00:14:58,002 --> 00:15:01,802
Speaker 3: it as a labor automating technology. But then in their

216
00:15:01,842 --> 00:15:05,642
Speaker 3: contracts with Microsoft, as reported by the Information, AGI is

217
00:15:05,642 --> 00:15:07,962
Speaker 3: apparently an AI system that can generate one hundred billion

218
00:15:08,002 --> 00:15:12,681
Speaker 3: dollars in revenue, which is like a completely different definition.

219
00:15:14,322 --> 00:15:18,802
Speaker 3: So yeah, it's just it's just a rhetorical tool forgetting

220
00:15:18,802 --> 00:15:19,321
Speaker 3: what they want.

221
00:15:19,602 --> 00:15:22,202
Speaker 1: On a side note, and I think this is probably

222
00:15:22,242 --> 00:15:23,961
Speaker 1: obvious to a lot of people, but I think still

223
00:15:23,962 --> 00:15:30,082
Speaker 1: bears emphasizing is the hunger that corporate CEOs have to

224
00:15:30,162 --> 00:15:32,802
Speaker 1: get rid of their needy employees. And by needy, I

225
00:15:32,802 --> 00:15:40,042
Speaker 1: mean like healthcare benefits, vacation, parental leave, potential union organizing.

226
00:15:40,442 --> 00:15:44,442
Speaker 1: They salivate at the idea of replacing workers with machines,

227
00:15:44,442 --> 00:15:47,522
Speaker 1: and so you can see how the concept of AGI

228
00:15:47,602 --> 00:15:49,961
Speaker 1: would be so appealing to a CEO.

229
00:15:50,681 --> 00:15:57,482
Speaker 3: Absolutely yeah, and this is why this has continued, in

230
00:15:57,522 --> 00:16:00,682
Speaker 3: part to be a self perpetuating myth. I guess you

231
00:16:00,681 --> 00:16:04,682
Speaker 3: could say that it is possible to build AGI is

232
00:16:04,722 --> 00:16:10,002
Speaker 3: because there is such a high demand for that magical

233
00:16:10,242 --> 00:16:14,922
Speaker 3: thing that a rep sense that it is just all

234
00:16:15,002 --> 00:16:19,042
Speaker 3: of the market incentives and all of the political incentives

235
00:16:20,002 --> 00:16:24,321
Speaker 3: push Silicon Valley to continue propping that myth up and

236
00:16:25,282 --> 00:16:27,522
Speaker 3: using it to dominate the narrative.

237
00:16:28,561 --> 00:16:32,122
Speaker 1: Something you mentioned earlier and that you trace throughout the

238
00:16:32,122 --> 00:16:36,602
Speaker 1: book is the way that the promise of AGI solving

239
00:16:36,642 --> 00:16:42,962
Speaker 1: the climate crisis has been used to rationalize really unfathomable

240
00:16:43,482 --> 00:16:46,362
Speaker 1: environmental costs. And you describe it at one point in

241
00:16:46,402 --> 00:16:48,842
Speaker 1: the book as a gamble that are basically saying, like,

242
00:16:49,282 --> 00:16:54,602
Speaker 1: it doesn't really matter resource consumption, the land grabbing, not

243
00:16:54,681 --> 00:16:58,441
Speaker 1: to mention the human exploitation that it requires right now,

244
00:16:59,042 --> 00:17:03,762
Speaker 1: because this magical mystery technology will solve this crisis that

245
00:17:03,802 --> 00:17:06,522
Speaker 1: we're contributing to in a massive way at some point

246
00:17:06,561 --> 00:17:08,922
Speaker 1: in the future. Can you talk a bit about that

247
00:17:09,042 --> 00:17:12,562
Speaker 1: gamble that they are making the planet's chips, even though

248
00:17:12,922 --> 00:17:15,081
Speaker 1: no one actually gave them those chips to play with,

249
00:17:15,121 --> 00:17:18,042
Speaker 1: but they're basically the extent to which they do not

250
00:17:18,121 --> 00:17:21,042
Speaker 1: seem to care about any of the impacts of the

251
00:17:21,081 --> 00:17:26,321
Speaker 1: technologies they're building because of this future promise to solve

252
00:17:26,482 --> 00:17:30,321
Speaker 1: this existential crisis. It's really pretty astonishing.

253
00:17:31,202 --> 00:17:35,522
Speaker 3: It is incredibly astonishing. Yeah, I think the first to

254
00:17:35,561 --> 00:17:39,401
Speaker 3: just quantify what level of environmental impact we're talking about,

255
00:17:39,482 --> 00:17:42,121
Speaker 3: there was this McKenzie report that came out earlier this

256
00:17:42,202 --> 00:17:46,202
Speaker 3: year that did an estimation on how much energy would

257
00:17:46,242 --> 00:17:50,882
Speaker 3: actually be required to sustain the current projections of AI

258
00:17:50,922 --> 00:17:54,962
Speaker 3: industry growth, and we would need to add two to

259
00:17:55,042 --> 00:17:57,842
Speaker 3: six times the amount of energy consumed by California onto

260
00:17:57,882 --> 00:18:01,282
Speaker 3: the global grid in five years just to sustain the

261
00:18:01,321 --> 00:18:04,042
Speaker 3: AA industry and its state of center of demands. That

262
00:18:04,242 --> 00:18:08,922
Speaker 3: is insane. That is six possibly six times the fifth

263
00:18:09,002 --> 00:18:11,282
Speaker 3: largest economy in the world, and there have been, you know,

264
00:18:11,321 --> 00:18:14,522
Speaker 3: other projections that have said that's more energy than all

265
00:18:14,561 --> 00:18:18,802
Speaker 3: of India, the largest energy consumer at this point globally,

266
00:18:19,561 --> 00:18:21,961
Speaker 3: and most of that will come from fossil fuels. So

267
00:18:22,002 --> 00:18:25,962
Speaker 3: we are not only talking about an acceleration of climate change,

268
00:18:26,121 --> 00:18:29,962
Speaker 3: we are also talking about an acceleration of air pollution.

269
00:18:30,242 --> 00:18:34,642
Speaker 3: We are already seeing reports of methane gas turbines, for example,

270
00:18:34,642 --> 00:18:38,641
Speaker 3: being used to power Colossus, the supercomputer that Elon Musk

271
00:18:38,682 --> 00:18:42,682
Speaker 3: built for XAI and for training Rock in Memphis, Tennessee.

272
00:18:43,081 --> 00:18:47,882
Speaker 3: That's pumping thousands of tons of pollutants into working class

273
00:18:48,402 --> 00:18:52,682
Speaker 3: communities that have already had a long history of being

274
00:18:52,722 --> 00:18:57,442
Speaker 3: denied the fundamental right to clean air for decades. And

275
00:18:57,962 --> 00:19:02,361
Speaker 3: we also see the exacerbation of the water the clean

276
00:19:02,442 --> 00:19:07,321
Speaker 3: water crisis. There is a huge crisis of people getting

277
00:19:07,361 --> 00:19:10,601
Speaker 3: access to clean drinking water around the world, in part

278
00:19:10,682 --> 00:19:15,601
Speaker 3: due to the acceleration of climate change, and AI also

279
00:19:15,682 --> 00:19:19,282
Speaker 3: exacerbates that. There was this investigation from Bloomberg that found

280
00:19:19,321 --> 00:19:21,841
Speaker 3: that two thirds of the data center is being built

281
00:19:21,841 --> 00:19:25,602
Speaker 3: for the AI industry are going into places that are

282
00:19:25,601 --> 00:19:32,242
Speaker 3: already scarce on freshwater resources, and data centers require freshwater

283
00:19:32,282 --> 00:19:35,841
Speaker 3: resources to cool in order to make sure that these

284
00:19:36,361 --> 00:19:40,921
Speaker 3: extremely expensive computer chips don't overheat and bust. And so

285
00:19:41,442 --> 00:19:48,122
Speaker 3: you have this multi layered environmental public health crisis kind

286
00:19:48,121 --> 00:19:53,842
Speaker 3: of being pushed to the brank and undermining people's fundamental

287
00:19:53,882 --> 00:20:01,042
Speaker 3: ability to live a decent quality life. And that's happening now.

288
00:20:01,242 --> 00:20:06,282
Speaker 3: This is not speculative, like this is literally playing out currently,

289
00:20:07,361 --> 00:20:11,361
Speaker 3: and the AA industry justifies all of this base on

290
00:20:11,722 --> 00:20:16,441
Speaker 3: a speculative assumption that maybe we will get to a

291
00:20:16,482 --> 00:20:19,802
Speaker 3: point where these AI systems can then fix all of

292
00:20:19,841 --> 00:20:24,201
Speaker 3: those problems. And that is the thing that is so

293
00:20:24,482 --> 00:20:28,522
Speaker 3: mind blowing to me, is that in this AI era,

294
00:20:29,282 --> 00:20:35,802
Speaker 3: somehow future speculation can be used to justify present day,

295
00:20:35,962 --> 00:20:41,682
Speaker 3: real world substantial harm. And the question that I always

296
00:20:41,922 --> 00:20:45,042
Speaker 3: ask people is like, how long do we wait for

297
00:20:45,081 --> 00:20:49,522
Speaker 3: this potential speculative future. How much harm do we suffer

298
00:20:50,002 --> 00:20:53,522
Speaker 3: before we realize that future might not arrive and we

299
00:20:53,642 --> 00:20:57,881
Speaker 3: might not survive these harms. And of course, like the

300
00:20:57,882 --> 00:21:00,242
Speaker 3: people who are ultimately making the decisions, they're not the

301
00:21:00,242 --> 00:21:03,081
Speaker 3: ones that they're the brunt of these harms. They're not

302
00:21:03,121 --> 00:21:05,802
Speaker 3: the ones that are bearing the brunt of climate having

303
00:21:06,002 --> 00:21:10,722
Speaker 3: changed already, and they're not the ones breathing in this

304
00:21:10,841 --> 00:21:15,721
Speaker 3: toxic air from fossil fuels burning in their communities. And

305
00:21:15,762 --> 00:21:18,882
Speaker 3: so for THEMB the question of how long can we

306
00:21:19,121 --> 00:21:23,161
Speaker 3: kind of bear to do all these things before we

307
00:21:23,361 --> 00:21:27,601
Speaker 3: potentially reach euphoria is really really long compared to most

308
00:21:27,642 --> 00:21:29,481
Speaker 3: of the rest of the world.

309
00:21:30,242 --> 00:21:32,881
Speaker 1: There's a lot of ways in which the tech industry

310
00:21:32,922 --> 00:21:36,322
Speaker 1: broadly but especially AI, if there is still any separation

311
00:21:36,402 --> 00:21:38,881
Speaker 1: between the two seems to be following the fossil fuel

312
00:21:38,962 --> 00:21:42,161
Speaker 1: industry playbook, and one of those ways seems to be

313
00:21:43,162 --> 00:21:46,402
Speaker 1: this narrative of you know, A, this is inevitable, It's

314
00:21:46,442 --> 00:21:48,962
Speaker 1: happening whether we like it or not, and so we

315
00:21:49,081 --> 00:21:51,522
Speaker 1: might as well might as well be the ones doing it.

316
00:21:52,081 --> 00:21:54,762
Speaker 1: And b this is the you know, the price of progress,

317
00:21:54,802 --> 00:21:57,682
Speaker 1: and if you are against this, then you're against human progress.

318
00:21:58,202 --> 00:22:02,441
Speaker 1: These really powerful narratives of inevitability and progress really struck

319
00:22:02,442 --> 00:22:05,561
Speaker 1: me because it's the exact same rhetoric that the fossil

320
00:22:05,561 --> 00:22:08,081
Speaker 1: fuel industry has been deploying in some form or another

321
00:22:08,162 --> 00:22:08,922
Speaker 1: for decades.

322
00:22:09,561 --> 00:22:11,482
Speaker 3: Yeah. That's such a great point, I mean, and it

323
00:22:12,002 --> 00:22:16,482
Speaker 3: is the trump card for any industry that knows that

324
00:22:16,522 --> 00:22:23,522
Speaker 3: they're engaging in something that's deeply incorrect. And yeah, I mean,

325
00:22:24,002 --> 00:22:27,522
Speaker 3: these technologies are not inevitable. Like the thing that I

326
00:22:27,561 --> 00:22:31,282
Speaker 3: spend a lot of time trying to do with my

327
00:22:31,361 --> 00:22:36,602
Speaker 3: book is document the moments in which open AI employees

328
00:22:37,042 --> 00:22:42,201
Speaker 3: made decisions that fundamentally shape the trajectory of how the

329
00:22:42,202 --> 00:22:46,201
Speaker 3: technology was introduced into society based purely on you know,

330
00:22:46,361 --> 00:22:49,802
Speaker 3: whims of judgment, Like technology is always a product of

331
00:22:49,841 --> 00:22:54,802
Speaker 3: human choices and sometimes those choices are actually seem really

332
00:22:54,882 --> 00:22:59,122
Speaker 3: minor in the moment. Yeah, So, like, how can it

333
00:22:59,162 --> 00:23:02,802
Speaker 3: be inevitable when the shape of a technology can be

334
00:23:03,002 --> 00:23:05,561
Speaker 3: swayed one way or the other based on just like

335
00:23:05,682 --> 00:23:09,361
Speaker 3: a person sitting in a room deciding on the fly

336
00:23:09,882 --> 00:23:13,162
Speaker 3: to do one thing or another and the other thing

337
00:23:13,202 --> 00:23:16,361
Speaker 3: with progress. So yeah, this is like the thing that

338
00:23:16,442 --> 00:23:20,401
Speaker 3: I hope we can have a more nuanced conversation about

339
00:23:20,482 --> 00:23:25,522
Speaker 3: moving forward. Is there's always this idea that the tech

340
00:23:25,561 --> 00:23:31,121
Speaker 3: industry perpetuates that AI is there's only one form of AI,

341
00:23:31,601 --> 00:23:34,481
Speaker 3: and so if we want progress from AI, we just

342
00:23:34,561 --> 00:23:38,522
Speaker 3: have to accept the costs of this progress, meaning the

343
00:23:38,561 --> 00:23:41,722
Speaker 3: colossal environmental impact, the colossal public health impacts, and so

344
00:23:41,802 --> 00:23:45,242
Speaker 3: on and so forth. AI is actually a collection of

345
00:23:45,361 --> 00:23:49,282
Speaker 3: many different technologies that work in very different ways, that

346
00:23:49,321 --> 00:23:53,242
Speaker 3: have very different costs benefit trade offs. And I often

347
00:23:53,282 --> 00:23:56,522
Speaker 3: make the analogy that it's like transportation transportation. There's many

348
00:23:56,561 --> 00:23:59,442
Speaker 3: different types of transportation, and when we talk about how

349
00:23:59,482 --> 00:24:03,601
Speaker 3: we need to improve our transportation options as part of

350
00:24:04,121 --> 00:24:07,962
Speaker 3: client resiliency and fighting climate change, we're not saying like

351
00:24:08,042 --> 00:24:11,522
Speaker 3: we just need more transportation. We're saying we need more bikes,

352
00:24:11,561 --> 00:24:15,881
Speaker 3: we need more public transit. We need less gas guzzling trucks.

353
00:24:15,882 --> 00:24:18,081
Speaker 3: We need to electrify cars. There's like a much more

354
00:24:18,162 --> 00:24:20,921
Speaker 3: nuanced conversation about like which mode of transportation are we

355
00:24:20,962 --> 00:24:25,802
Speaker 3: talking about, and how to tweak and tune and adjust

356
00:24:25,922 --> 00:24:28,442
Speaker 3: the supply and demand of each of these different types

357
00:24:28,482 --> 00:24:32,601
Speaker 3: of transportation in order to get that progress. And we

358
00:24:32,682 --> 00:24:36,081
Speaker 3: need to have a similar nuanced conversation about AI. The

359
00:24:36,321 --> 00:24:39,522
Speaker 3: large scale AI models that Silicon Valley has imposed on

360
00:24:39,601 --> 00:24:44,561
Speaker 3: everyone that I describe as an imperial consolidation project. That

361
00:24:44,841 --> 00:24:47,682
Speaker 3: is the form of AI that is the most costly

362
00:24:48,361 --> 00:24:53,841
Speaker 3: and has very unclear benefits that just do not justify

363
00:24:54,402 --> 00:24:58,442
Speaker 3: those costs. But when talking about the kinds of progress

364
00:24:58,442 --> 00:25:01,841
Speaker 3: that we would actually want in general, putting technology aside,

365
00:25:01,882 --> 00:25:05,482
Speaker 3: like things like wanting client change to be resolved, wanting

366
00:25:05,841 --> 00:25:10,442
Speaker 3: better drugs, better healthcare for treating different types of diseases

367
00:25:10,482 --> 00:25:13,282
Speaker 3: like these are there are actually types of progress that

368
00:25:13,361 --> 00:25:16,841
Speaker 3: we already know how to build AI systems for that

369
00:25:16,882 --> 00:25:21,282
Speaker 3: look fundamentally different from these large SCALEI systems. Those types

370
00:25:21,321 --> 00:25:25,321
Speaker 3: of AI systems they're task specific, they're often very very

371
00:25:25,361 --> 00:25:29,202
Speaker 3: small and cost extremely small amounts of energy to develop.

372
00:25:29,682 --> 00:25:32,641
Speaker 3: They're trained on highly curated data sets that do not

373
00:25:32,762 --> 00:25:36,961
Speaker 3: require the type of labor exploitation and content moderation that

374
00:25:37,002 --> 00:25:42,202
Speaker 3: goes into something like Tatgybut and they have already documented.

375
00:25:42,922 --> 00:25:46,202
Speaker 3: We know that when we build these systems, we get

376
00:25:46,442 --> 00:25:49,601
Speaker 3: benefits on the end, rather than this speculative maybe in

377
00:25:49,642 --> 00:25:52,402
Speaker 3: the future will reach AGI one day and it will

378
00:25:52,442 --> 00:25:57,121
Speaker 3: solve everything. And so once you kind of realize that,

379
00:25:57,162 --> 00:26:00,282
Speaker 3: once you realize there's actually a portfolio of different options

380
00:26:00,882 --> 00:26:04,722
Speaker 3: for the shape of AI and what it can do,

381
00:26:05,361 --> 00:26:09,161
Speaker 3: it suddenly becomes like blatantly obvious that we're actually in

382
00:26:09,202 --> 00:26:13,482
Speaker 3: this colossal capital misallocation problem where we're putting all of

383
00:26:13,482 --> 00:26:17,802
Speaker 3: our capital in the one type of AI that has

384
00:26:17,841 --> 00:26:21,641
Speaker 3: the absolute worst trade offs, and we should be rapidly

385
00:26:22,002 --> 00:26:24,922
Speaker 3: urgently shifting that capital to all of the other types

386
00:26:24,962 --> 00:26:28,002
Speaker 3: of AI that have the correct trade offs very little

387
00:26:28,002 --> 00:26:29,601
Speaker 3: cost for maximal benefit.

388
00:26:30,682 --> 00:26:34,081
Speaker 1: The question of scale, or the idea or the obsession

389
00:26:34,561 --> 00:26:37,841
Speaker 1: with scale, I guess you describe as a doctrine in

390
00:26:37,922 --> 00:26:40,722
Speaker 1: Silicon Valley and in these companies that the only way

391
00:26:40,722 --> 00:26:43,441
Speaker 1: to do this, as you're just referencing, is to scale

392
00:26:43,482 --> 00:26:46,681
Speaker 1: as big and as fast as possible, And something I

393
00:26:46,682 --> 00:26:49,202
Speaker 1: see a lot in headlines these days is in press

394
00:26:49,242 --> 00:26:53,681
Speaker 1: releases is talk of hyper scaling and mega campuses and

395
00:26:53,722 --> 00:26:56,282
Speaker 1: all of this stuff, which is obviously touted is a

396
00:26:56,321 --> 00:26:58,882
Speaker 1: good thing. When you see hyper scale and mega campus

397
00:26:58,962 --> 00:27:01,202
Speaker 1: and where bigger is better, so we must be doing

398
00:27:01,242 --> 00:27:07,361
Speaker 1: something revolutionary and profitable. But it seems like the biggest

399
00:27:07,561 --> 00:27:10,841
Speaker 1: certainly from an environmental perspective, the biggest costs come from

400
00:27:11,162 --> 00:27:11,722
Speaker 1: the scale.

401
00:27:11,841 --> 00:27:12,081
Speaker 3: Right.

402
00:27:12,442 --> 00:27:16,921
Speaker 1: It's not necessarily the technology itself, but the size at

403
00:27:16,922 --> 00:27:19,162
Speaker 1: which that they are obsessed with doing this stuff.

404
00:27:19,321 --> 00:27:19,561
Speaker 2: Is that?

405
00:27:19,802 --> 00:27:20,282
Speaker 1: Is that right?

406
00:27:20,682 --> 00:27:24,402
Speaker 3: Yeah? Exactly. That is my fundamental critique about silicon valleys

407
00:27:24,402 --> 00:27:29,762
Speaker 3: a Perrhai. They've created this idea that you can no

408
00:27:29,841 --> 00:27:33,401
Speaker 3: longer challenge within the valley, that in order to reach

409
00:27:33,802 --> 00:27:38,602
Speaker 3: some kind of utopic state through the creation of AGI,

410
00:27:38,722 --> 00:27:42,242
Speaker 3: you just have to pour ever more data into these

411
00:27:42,282 --> 00:27:45,282
Speaker 3: models and train them on ever larger supercomputers. And we

412
00:27:45,321 --> 00:27:50,081
Speaker 3: are now talking about supercomputers the size of Manhaeian. Meta

413
00:27:50,242 --> 00:27:54,882
Speaker 3: and Opening Eye have both drafted plans to build supercommuters

414
00:27:55,081 --> 00:28:00,482
Speaker 3: the size of and that consume the same power demands

415
00:28:00,522 --> 00:28:06,401
Speaker 3: as Manhaian. This is not actually scientifically sound, you know,

416
00:28:06,522 --> 00:28:09,802
Speaker 3: like when I had the privileges of starting to cover

417
00:28:09,882 --> 00:28:12,642
Speaker 3: AI research. Before chat GBT came out, and well before

418
00:28:12,682 --> 00:28:16,121
Speaker 3: large language models became the obsession, AI research was actually

419
00:28:16,121 --> 00:28:18,922
Speaker 3: heading in the exact opposite direction. At the time. People

420
00:28:19,002 --> 00:28:22,841
Speaker 3: were looking to build smaller and smaller models, and it

421
00:28:22,882 --> 00:28:25,361
Speaker 3: was all about tiny AI because they were all of

422
00:28:25,361 --> 00:28:29,482
Speaker 3: these different techniques that researchers were experimenting with that found

423
00:28:29,601 --> 00:28:32,522
Speaker 3: that you can build really powerful AI systems with minimal

424
00:28:32,561 --> 00:28:37,082
Speaker 3: amounts of data, and also on pocket like a phone,

425
00:28:37,121 --> 00:28:41,802
Speaker 3: like your pocket computer, and that kind of understanding that

426
00:28:42,002 --> 00:28:46,522
Speaker 3: we can actually create new algorithms, create new training techniques,

427
00:28:47,082 --> 00:28:50,522
Speaker 3: just redesign AI systems from the ground up to use

428
00:28:50,762 --> 00:28:53,402
Speaker 3: lesson less resources, something that's been totally thrown out the

429
00:28:53,402 --> 00:28:57,922
Speaker 3: window by Silicon Valley. And it does not make sense

430
00:28:57,962 --> 00:29:01,682
Speaker 3: to me why we have got and stuck in Well,

431
00:29:01,802 --> 00:29:03,681
Speaker 3: I mean it does because some of the people, the

432
00:29:03,682 --> 00:29:05,562
Speaker 3: people that are putting us on this map have an

433
00:29:05,602 --> 00:29:08,961
Speaker 3: extraordinary out of capital, extraordinary out of narrative power and

434
00:29:09,002 --> 00:29:13,762
Speaker 3: so on. But yeah, it's not scientifically technically justified, and

435
00:29:14,122 --> 00:29:17,922
Speaker 3: it has gotten us into this absolutely twisted state where

436
00:29:18,042 --> 00:29:21,201
Speaker 3: we are burning down the earth in order to do

437
00:29:21,242 --> 00:29:22,962
Speaker 3: something that's wholly unnecessary.

438
00:29:23,882 --> 00:29:26,802
Speaker 1: And I keep coming back to the psychology and the

439
00:29:26,842 --> 00:29:33,002
Speaker 1: worldview of the people making these decisions, because talking about social, environmental, political,

440
00:29:33,082 --> 00:29:35,962
Speaker 1: cultural trends, you know, we often talk for good reason

441
00:29:36,002 --> 00:29:40,201
Speaker 1: about problems that are systemic and structural rather than individual.

442
00:29:40,722 --> 00:29:42,681
Speaker 1: But I feel like this is one of those cases

443
00:29:42,722 --> 00:29:45,082
Speaker 1: where it is all of those things, but it's also

444
00:29:45,362 --> 00:29:48,642
Speaker 1: very individual in the sense that so few people are

445
00:29:48,642 --> 00:29:52,602
Speaker 1: making these decisions that are impacting so many of us.

446
00:29:53,002 --> 00:29:55,002
Speaker 1: And I'm just wondering, because you've spent so much time

447
00:29:55,682 --> 00:29:57,962
Speaker 1: with folks at you know, current former employees at open

448
00:29:58,002 --> 00:30:01,681
Speaker 1: Ai as well as the other companies, and there's obviously

449
00:30:01,682 --> 00:30:04,402
Speaker 1: a lot of churn between them, Like, how much do

450
00:30:04,482 --> 00:30:09,282
Speaker 1: you think this trajectory that we're on from you know,

451
00:30:09,322 --> 00:30:13,122
Speaker 1: pocket sized ai have a decade ago to Manhattan sized

452
00:30:13,122 --> 00:30:16,681
Speaker 1: supercomputers potentially in the near future. How much of that

453
00:30:16,722 --> 00:30:20,562
Speaker 1: do you think is driven by just individual desire for

454
00:30:21,442 --> 00:30:27,002
Speaker 1: wealth and more specifically power and domination, versus actually wanting

455
00:30:27,042 --> 00:30:31,962
Speaker 1: to believing in their hearts that these technologies will solve

456
00:30:32,002 --> 00:30:34,681
Speaker 1: the problems that they claim to be able to solve.

457
00:30:35,242 --> 00:30:37,842
Speaker 1: I'm just wondering how you think about the psychology at

458
00:30:37,842 --> 00:30:40,522
Speaker 1: play there and how much of our future is being

459
00:30:40,562 --> 00:30:43,962
Speaker 1: dictated by a really small number of people's desire for

460
00:30:44,602 --> 00:30:46,442
Speaker 1: domination over the rest of us.

461
00:30:48,282 --> 00:30:50,762
Speaker 3: Yeah, I think the best way to answer this question is,

462
00:30:51,042 --> 00:30:55,602
Speaker 3: I think in order to be one of these people,

463
00:30:55,762 --> 00:30:59,562
Speaker 3: you have to be self delusional, and so I'm sure

464
00:30:59,602 --> 00:31:03,962
Speaker 3: that they believe that they believe that they're doing this

465
00:31:04,642 --> 00:31:09,722
Speaker 3: because of out of goodness. I don't think you can

466
00:31:09,842 --> 00:31:15,602
Speaker 3: sustain waking up every morning and orchestrating these deals and

467
00:31:15,642 --> 00:31:19,802
Speaker 3: decisions that are terraforming her than reshaping geopolitics and rewiring

468
00:31:19,882 --> 00:31:25,802
Speaker 3: everything in these ways without deluding yourself into believing that

469
00:31:25,842 --> 00:31:29,441
Speaker 3: you are on a mission for the benefit of all

470
00:31:29,442 --> 00:31:32,082
Speaker 3: of humanity. And I think that is that was not

471
00:31:32,242 --> 00:31:35,642
Speaker 3: something I fully understood until I reported my book and

472
00:31:35,922 --> 00:31:39,441
Speaker 3: spoke with so many people in this world. Was I

473
00:31:39,482 --> 00:31:42,722
Speaker 3: did think that there was more distance between the rhetoric

474
00:31:42,762 --> 00:31:47,842
Speaker 3: that they used publicly to leverage whatever public opinion or

475
00:31:48,362 --> 00:31:52,962
Speaker 3: their political capital or narratives to get what they wanted

476
00:31:53,082 --> 00:31:56,481
Speaker 3: and their actual true beliefs. And what I realized the

477
00:31:56,482 --> 00:32:00,882
Speaker 3: more that I reported is actually the rhetoric and what

478
00:32:00,962 --> 00:32:04,602
Speaker 3: they believe. The distance collapses over time because you just

479
00:32:04,682 --> 00:32:09,522
Speaker 3: cannot continue to say all of these things and do

480
00:32:09,642 --> 00:32:14,642
Speaker 3: all of these things without giving telling yourself that that

481
00:32:14,802 --> 00:32:18,922
Speaker 3: is true, that is your truth. People's beliefs become molded

482
00:32:19,082 --> 00:32:24,482
Speaker 3: by what they need to accumulate more power and you

483
00:32:24,522 --> 00:32:27,602
Speaker 3: accumulate more wealth. You know, I had the because I

484
00:32:27,682 --> 00:32:31,802
Speaker 3: started reporting on Opening Eye several years before chat Gibt

485
00:32:32,002 --> 00:32:35,002
Speaker 3: came out. Like, I interviewed people back then that I've

486
00:32:35,002 --> 00:32:37,282
Speaker 3: been and reinterviewed for the book, So I sort of

487
00:32:37,482 --> 00:32:42,642
Speaker 3: have seen people transition in their beliefs over time as

488
00:32:42,682 --> 00:32:46,602
Speaker 3: they've become more and more part of the company and

489
00:32:47,402 --> 00:32:52,082
Speaker 3: this world, and it really does completely transform them and

490
00:32:52,322 --> 00:32:54,642
Speaker 3: who they think they are and what their worldviews are.

491
00:32:54,682 --> 00:32:56,921
Speaker 3: Like I was talking to people back then who were like,

492
00:32:57,082 --> 00:32:59,082
Speaker 3: I don't really believe in the AGI think, but you know,

493
00:32:59,482 --> 00:33:01,602
Speaker 3: I just get I get a good salary, I get

494
00:33:01,642 --> 00:33:05,242
Speaker 3: a lot of resources to play around with cutting edge research.

495
00:33:05,762 --> 00:33:08,842
Speaker 3: It's basically like working in academia. But just like way

496
00:33:08,842 --> 00:33:14,522
Speaker 3: better funded through have then now become complete AGI believers,

497
00:33:14,602 --> 00:33:18,882
Speaker 3: like do not even question for a second that they

498
00:33:18,922 --> 00:33:22,522
Speaker 3: are building something that could be akin to a god

499
00:33:22,802 --> 00:33:30,241
Speaker 3: and will have utopic qualities, and that this is the

500
00:33:30,282 --> 00:33:32,002
Speaker 3: reason that they're brought to this earth.

501
00:33:32,802 --> 00:33:36,082
Speaker 1: I'm sure it's intoxicating to feel that. I can see

502
00:33:36,082 --> 00:33:38,722
Speaker 1: how people might get addicted to that feeling of being

503
00:33:39,442 --> 00:33:41,322
Speaker 1: on some sort of divine mission.

504
00:33:42,562 --> 00:33:45,362
Speaker 3: Yeah, absolutely, and not just a divine mission, Like people

505
00:33:45,362 --> 00:33:48,882
Speaker 3: have to get addicted to just the sheer. They get

506
00:33:48,922 --> 00:33:50,762
Speaker 3: addicted to a lot of things. They get addicted to

507
00:33:50,762 --> 00:33:54,802
Speaker 3: the sheer size of the resources that they can have

508
00:33:54,882 --> 00:33:58,442
Speaker 3: access to. They get addicted to the amount of influence

509
00:33:58,442 --> 00:34:00,602
Speaker 3: that they can have. You know, like I've had people

510
00:34:00,642 --> 00:34:04,522
Speaker 3: say to me, well, there's really very few other companies

511
00:34:04,562 --> 00:34:09,882
Speaker 3: that you can work to have ripple effects on billions

512
00:34:09,922 --> 00:34:12,761
Speaker 3: of people around the world. And for them, that is

513
00:34:13,082 --> 00:34:17,082
Speaker 3: They're not saying that in a machi Abelian way. They're

514
00:34:17,122 --> 00:34:19,722
Speaker 3: saying that in a I want to change the world

515
00:34:20,082 --> 00:34:23,801
Speaker 3: to be a better place, So what better leverage can

516
00:34:23,842 --> 00:34:27,202
Speaker 3: I get than being at a company that touches billions

517
00:34:27,201 --> 00:34:29,922
Speaker 3: of people's lives, Like they're saying it from a place

518
00:34:29,962 --> 00:34:34,562
Speaker 3: of this is, how could you not let take that opportunity?

519
00:34:34,602 --> 00:34:37,162
Speaker 3: If you, like, if you wanted to change the world

520
00:34:37,161 --> 00:34:39,362
Speaker 3: and you stepped into my shoes, wouldn't you do the

521
00:34:39,402 --> 00:34:39,921
Speaker 3: same thing.

522
00:34:40,802 --> 00:34:42,962
Speaker 1: It's really unsettling to think about.

523
00:34:43,482 --> 00:34:49,802
Speaker 3: Yeah, it's it totally the realization that the self delusion

524
00:34:49,842 --> 00:34:54,722
Speaker 3: is that strong. Breading the book really made me feel

525
00:34:54,882 --> 00:35:01,082
Speaker 3: a lot more freaked out about Yeah, like where we

526
00:35:01,282 --> 00:35:04,482
Speaker 3: where we could be heading if we don't as a

527
00:35:04,522 --> 00:35:08,002
Speaker 3: society rise up to contain the empire of AI.

528
00:35:08,642 --> 00:35:11,522
Speaker 1: Yeah, I mean, it just shows why empire is really

529
00:35:11,522 --> 00:35:16,281
Speaker 1: the perfect description for what is happening. Was there a

530
00:35:16,282 --> 00:35:18,761
Speaker 1: point when you were reporting the book or when you

531
00:35:18,802 --> 00:35:22,321
Speaker 1: were writing it, when that became clear to you that

532
00:35:22,442 --> 00:35:25,442
Speaker 1: this was the project they were undertaking, was the construction

533
00:35:25,522 --> 00:35:28,482
Speaker 1: of not just a company or a conglomerate, but an empire.

534
00:35:29,282 --> 00:35:33,842
Speaker 3: So interestingly, I actually when I first started working on

535
00:35:33,882 --> 00:35:36,202
Speaker 3: this book, it had very little to do with OPENINGI

536
00:35:36,282 --> 00:35:40,002
Speaker 3: and only to do with the AI industry as an empire.

537
00:35:40,842 --> 00:35:43,281
Speaker 3: And the reason is because I had been working on

538
00:35:43,322 --> 00:35:47,602
Speaker 3: a project called AI Colonialism for MIT Technology Review just

539
00:35:48,122 --> 00:35:51,482
Speaker 3: a year before Chatbut came out, and I had already

540
00:35:51,522 --> 00:35:55,321
Speaker 3: been thinking about the parallels between the colonialism and the

541
00:35:55,322 --> 00:35:58,322
Speaker 3: way that the AI industry was operating around the world.

542
00:35:58,882 --> 00:36:01,602
Speaker 3: And this was based in part on me reading a

543
00:36:01,642 --> 00:36:04,761
Speaker 3: bunch of scholarship that I identified these parallels and then

544
00:36:04,842 --> 00:36:08,681
Speaker 3: growing really fascinated by it and looking to see whether

545
00:36:08,762 --> 00:36:12,962
Speaker 3: or not that was actually happening the real world and

546
00:36:13,042 --> 00:36:18,242
Speaker 3: traveling to different countries to document those parallels and finding

547
00:36:18,562 --> 00:36:22,882
Speaker 3: extremely striking parallels in the process. And so I'd originally

548
00:36:22,882 --> 00:36:24,882
Speaker 3: cold see that the book is just let me just

549
00:36:25,042 --> 00:36:28,841
Speaker 3: write more about that, about the argument that there is

550
00:36:28,882 --> 00:36:34,042
Speaker 3: something very neo colonial happening here with silicon values push

551
00:36:34,282 --> 00:36:38,761
Speaker 3: to build and deploy this technology. And then when chat

552
00:36:38,802 --> 00:36:42,962
Speaker 3: GBT came out, my agent was actually the one that said, well,

553
00:36:43,002 --> 00:36:45,442
Speaker 3: how does this change things? Do you think that's fixed

554
00:36:45,562 --> 00:36:48,922
Speaker 3: the problems? And I was like, no, it's made it

555
00:36:49,681 --> 00:36:53,962
Speaker 3: way worse, like orders of magnitude worse, because now they've

556
00:36:54,322 --> 00:36:59,761
Speaker 3: fired the first shot where every single company is now

557
00:36:59,802 --> 00:37:02,161
Speaker 3: going to go after scale the way that Opening Eye

558
00:37:02,642 --> 00:37:07,241
Speaker 3: has religiously done, which was not at all the norm

559
00:37:07,681 --> 00:37:10,922
Speaker 3: before then, and that's going to accelerate the resource extraction,

560
00:37:11,082 --> 00:37:15,801
Speaker 3: accelerate the labor exploitation, accelerate the environmental degradation, and in

561
00:37:15,922 --> 00:37:18,602
Speaker 3: all of the ways that I was already seeing those

562
00:37:18,642 --> 00:37:22,762
Speaker 3: parallels to the Empire, and so yeah, so the revelation

563
00:37:22,882 --> 00:37:25,762
Speaker 3: went the other way around. It was starting with Empire

564
00:37:25,762 --> 00:37:28,442
Speaker 3: and then realizing that Open AI was the center of

565
00:37:28,602 --> 00:37:31,482
Speaker 3: so much of this story. Rather than starting with Open

566
00:37:31,522 --> 00:37:33,362
Speaker 3: AI and then realizing they were an empire.

567
00:37:34,402 --> 00:37:36,482
Speaker 1: Something you point out throughout the book is the way

568
00:37:36,482 --> 00:37:39,761
Speaker 1: that open AI kind of sets the precedent for the

569
00:37:39,802 --> 00:37:43,442
Speaker 1: way the rest of the industry goes, certainly like the

570
00:37:43,442 --> 00:37:46,681
Speaker 1: obsessional scale as you mentioned. Another one is the kind

571
00:37:46,721 --> 00:37:52,002
Speaker 1: of rapid evolution or devolution from these being essentially research

572
00:37:52,042 --> 00:37:55,241
Speaker 1: hubs doing peer reviewed research. And yes they're funded by

573
00:37:55,762 --> 00:37:59,281
Speaker 1: private industry, but everything's very open, very academic, lots of

574
00:37:59,322 --> 00:38:04,482
Speaker 1: collaboration with universities and scholars, to in another parallel to

575
00:38:04,482 --> 00:38:08,602
Speaker 1: the fossil fuel industry, very secretive. They have the information

576
00:38:08,681 --> 00:38:11,162
Speaker 1: about what these things do as best as anyone does,

577
00:38:11,241 --> 00:38:14,882
Speaker 1: but they're not publishing it, they're not sharing it. Critical

578
00:38:14,882 --> 00:38:19,322
Speaker 1: research gets suffocated, people get fired for criticizing the regime.

579
00:38:20,161 --> 00:38:22,522
Speaker 1: Do you talk about the kind of the suppression of

580
00:38:22,721 --> 00:38:27,042
Speaker 1: inconvenient information and how fast that that shift occurred from

581
00:38:27,161 --> 00:38:30,562
Speaker 1: we're essentially you know, private research hubs to this is

582
00:38:30,602 --> 00:38:34,002
Speaker 1: proprietary information and you know, do not deviate from the narrative.

583
00:38:34,762 --> 00:38:39,002
Speaker 3: Yeah. So I think people who start becoming aware of

584
00:38:39,042 --> 00:38:42,522
Speaker 3: the air industry and as practice post Chatjubut would think

585
00:38:42,562 --> 00:38:47,481
Speaker 3: it's perfectly normal that these companies are very secretive about

586
00:38:47,522 --> 00:38:50,681
Speaker 3: their quote unquote intellectual property, and any kind of AI

587
00:38:50,762 --> 00:38:53,801
Speaker 3: research would be considered intellectual property, and therefore it's not

588
00:38:54,642 --> 00:38:59,161
Speaker 3: weird to them that these companies are are typefisted about

589
00:38:59,161 --> 00:39:02,882
Speaker 3: that kind of information and not transparent For people who

590
00:39:03,282 --> 00:39:07,602
Speaker 3: knew what the norms were in AI research before chat GBT,

591
00:39:07,882 --> 00:39:12,962
Speaker 3: this is a dramatic one eighty shift. Before even though

592
00:39:14,002 --> 00:39:18,002
Speaker 3: much of the tech innutry was bankrolling AI research, the

593
00:39:18,562 --> 00:39:23,761
Speaker 3: understanding was that in order to be competitive in attracting

594
00:39:23,882 --> 00:39:29,482
Speaker 3: talent you had to entice AI researchers by promising them

595
00:39:29,522 --> 00:39:31,321
Speaker 3: that they would ultimately get to publish all of their

596
00:39:31,402 --> 00:39:37,482
Speaker 3: work in public. Because AI research was still very academically driven,

597
00:39:37,562 --> 00:39:40,161
Speaker 3: and so even within it, even if you were a

598
00:39:40,161 --> 00:39:44,482
Speaker 3: Google researcher or a Meta researcher or whatever, your street

599
00:39:44,522 --> 00:39:46,681
Speaker 3: cred was still based on how many papers are you

600
00:39:46,721 --> 00:39:50,202
Speaker 3: publishing in top publications, how many citations are they getting.

601
00:39:50,282 --> 00:39:53,962
Speaker 3: It was just like the credentialing was just very very

602
00:39:54,322 --> 00:39:58,482
Speaker 3: academically coded. I guess you could say Opening Eye. Not

603
00:39:58,522 --> 00:40:02,482
Speaker 3: only did they close themselves off in a great irony

604
00:40:02,522 --> 00:40:05,642
Speaker 3: to their name, they closed the entire industry off by

605
00:40:06,122 --> 00:40:10,801
Speaker 3: doing a lot of work early on to see this

606
00:40:11,082 --> 00:40:15,161
Speaker 3: idea that maybe opening up a lot of this AI

607
00:40:15,282 --> 00:40:20,082
Speaker 3: research could be dangerous, and therefore it was actually the

608
00:40:20,122 --> 00:40:26,042
Speaker 3: responsible thing to shutter this research so that quote unquote

609
00:40:26,082 --> 00:40:29,921
Speaker 3: bad actors could not get access to it. And that

610
00:40:30,002 --> 00:40:33,921
Speaker 3: created a cascading effect across the industry where suddenly when

611
00:40:34,322 --> 00:40:37,522
Speaker 3: it became clear to companies that they could still retain

612
00:40:38,442 --> 00:40:43,042
Speaker 3: the top researchers and be competitive in the talent race

613
00:40:43,522 --> 00:40:47,161
Speaker 3: while being closed off, then it became sort of like

614
00:40:47,322 --> 00:40:49,161
Speaker 3: dead obvious to all of them that was the best

615
00:40:49,201 --> 00:40:53,442
Speaker 3: business decision is you want to keep your competitive advantage

616
00:40:53,482 --> 00:40:56,321
Speaker 3: close to your chest and not share that information. And

617
00:40:56,402 --> 00:41:01,522
Speaker 3: so what we've seen happen is very quickly, because structurally

618
00:41:02,042 --> 00:41:07,362
Speaker 3: most AIR researchers were ready employed by big tech, Suddenly

619
00:41:07,681 --> 00:41:13,882
Speaker 3: the majority of the research discipline now becomes distorted based

620
00:41:13,922 --> 00:41:17,721
Speaker 3: on what companies think is appropriate or not appropriate for

621
00:41:17,802 --> 00:41:20,201
Speaker 3: the public to know. And so it's not actually a

622
00:41:20,282 --> 00:41:24,442
Speaker 3: science anymore. I mean, it's just pr And we have

623
00:41:24,642 --> 00:41:31,602
Speaker 3: an absolute, completely inaccurate understanding now of what the true

624
00:41:32,122 --> 00:41:36,962
Speaker 3: capabilities and limitations of these AI systems really are because

625
00:41:37,642 --> 00:41:41,922
Speaker 3: you cannot audit what these companies claim the ANA systems

626
00:41:41,922 --> 00:41:45,962
Speaker 3: can do without any of the transparency on how they

627
00:41:45,962 --> 00:41:48,721
Speaker 3: were built, what data were they trained on, what kind

628
00:41:48,762 --> 00:41:52,122
Speaker 3: of tests are being run on them, what contexts have

629
00:41:52,201 --> 00:41:54,082
Speaker 3: they been stress tested in, and so on.

630
00:41:54,762 --> 00:41:57,201
Speaker 1: And that's true for some of the environmental impacts and

631
00:41:57,241 --> 00:42:00,882
Speaker 1: carbon emissions too, right, it's hard to understand the true

632
00:42:00,922 --> 00:42:03,882
Speaker 1: impact without them revealing some of this information.

633
00:42:04,762 --> 00:42:08,522
Speaker 3: Yeah, exactly. So we don't know how much energy it

634
00:42:08,562 --> 00:42:11,361
Speaker 3: takes to train a model, I mean, And of course

635
00:42:11,402 --> 00:42:14,361
Speaker 3: there are open source models, and there's been plenty of well,

636
00:42:14,402 --> 00:42:17,122
Speaker 3: I shouldn't say plenty, there has been some really fantastic

637
00:42:17,161 --> 00:42:20,681
Speaker 3: research on using open source models as a proxy for

638
00:42:20,842 --> 00:42:25,002
Speaker 3: understanding the energy and environmental impacts of training these large

639
00:42:25,042 --> 00:42:30,522
Speaker 3: SCALEI systems. But we only know what the real impacts

640
00:42:30,562 --> 00:42:34,882
Speaker 3: of the commercial closed source systems are when companies deem

641
00:42:34,922 --> 00:42:38,482
Speaker 3: it okay to release that information, which has been very

642
00:42:38,642 --> 00:42:42,602
Speaker 3: rare but has happened when they've come under extreme public

643
00:42:42,642 --> 00:42:44,282
Speaker 3: pressure to do so.

644
00:42:44,282 --> 00:42:44,362
Speaker 2: So.

645
00:42:44,482 --> 00:42:50,522
Speaker 3: Google has over time slowly dribbled out some information about

646
00:42:50,562 --> 00:42:54,242
Speaker 3: the environmental impacts of its models, and Meta has tried

647
00:42:54,282 --> 00:42:57,122
Speaker 3: to also do a little bit of that to try

648
00:42:57,161 --> 00:43:01,922
Speaker 3: and generate some like good pr around their transparency and

649
00:43:02,042 --> 00:43:03,881
Speaker 3: trying to position themselves as sort of like an open

650
00:43:03,922 --> 00:43:09,402
Speaker 3: source leader, but it is so limited and the transparency

651
00:43:09,442 --> 00:43:14,442
Speaker 3: is completely disproportional to what is actually needed based on

652
00:43:14,922 --> 00:43:18,082
Speaker 3: the size of the environmental impact and energy consumption.

653
00:43:19,322 --> 00:43:21,922
Speaker 1: There's something that you noted at some point in the

654
00:43:21,962 --> 00:43:24,882
Speaker 1: book you made the what When I read it, it

655
00:43:24,922 --> 00:43:27,042
Speaker 1: seemed very obvious, but I hadn't thought about it at all,

656
00:43:27,241 --> 00:43:31,522
Speaker 1: The point that to solve a global challenge like the

657
00:43:31,562 --> 00:43:35,322
Speaker 1: climate crisis whatever, again, whatever solve means well, we'll also

658
00:43:35,442 --> 00:43:40,002
Speaker 1: need more social cohesion and global cooperation, and of course

659
00:43:40,282 --> 00:43:45,522
Speaker 1: that is precisely what these empires are undermining. I guess

660
00:43:45,522 --> 00:43:47,562
Speaker 1: I had been thinking about it, you know, as the

661
00:43:47,602 --> 00:43:49,522
Speaker 1: way they wanted me to think about it in some ways,

662
00:43:49,562 --> 00:43:53,321
Speaker 1: as a technology. Could you just unpack that a little bit,

663
00:43:53,681 --> 00:43:57,681
Speaker 1: the way that these empires are undermining the very cohesion

664
00:43:57,681 --> 00:44:00,602
Speaker 1: in cooperation that you would need to actually make progress.

665
00:44:01,482 --> 00:44:04,801
Speaker 3: Yeah, this is something that I have has been sort

666
00:44:04,802 --> 00:44:07,281
Speaker 3: of a revolution that I've had again and again and

667
00:44:07,282 --> 00:44:10,801
Speaker 3: again and reporting on technology, is that so often as

668
00:44:10,842 --> 00:44:14,402
Speaker 3: a society we are vulnerable to believing that technology will

669
00:44:14,442 --> 00:44:18,201
Speaker 3: solve social problems, and actually only social solutions will solve

670
00:44:18,201 --> 00:44:23,402
Speaker 3: social problems. And ultimately, climate change at this point is

671
00:44:23,442 --> 00:44:26,882
Speaker 3: not a technology problem. In fact, we have an abundance

672
00:44:26,882 --> 00:44:29,082
Speaker 3: of solutions that we could be deploying, and it's just

673
00:44:29,122 --> 00:44:32,482
Speaker 3: a lack of political will. It's a lack of global cooperation,

674
00:44:32,562 --> 00:44:35,082
Speaker 3: a lack of all of these soft skills that have

675
00:44:35,282 --> 00:44:38,562
Speaker 3: caused us to fail at actually getting there. There's this

676
00:44:38,681 --> 00:44:41,241
Speaker 3: organization called Climate Change AI that I really love. It's

677
00:44:41,282 --> 00:44:43,201
Speaker 3: a nonprofit that was set up by a bunch of

678
00:44:43,241 --> 00:44:47,761
Speaker 3: AI researchers around the world to think deeply about what

679
00:44:47,922 --> 00:44:51,801
Speaker 3: is how do we actually what can AI actually do

680
00:44:51,922 --> 00:44:56,241
Speaker 3: to try and mitigate climate change? And they document this

681
00:44:56,642 --> 00:44:59,201
Speaker 3: long list in a white paper of all of these

682
00:44:59,201 --> 00:45:03,801
Speaker 3: different computational challenges because ultimately, like AI is a computational

683
00:45:03,962 --> 00:45:07,322
Speaker 3: technology that lends itself to computational challenges. So it documents

684
00:45:07,322 --> 00:45:13,122
Speaker 3: all these computational challenges that could be helpful for solving

685
00:45:13,562 --> 00:45:16,281
Speaker 3: when it comes to climate change medication, like integrating more

686
00:45:16,282 --> 00:45:19,201
Speaker 3: renewable energy into the grid, more accurate weather prediction, more

687
00:45:19,201 --> 00:45:24,882
Speaker 3: accurate climate disaster prediction, reducing the energy demands of buildings

688
00:45:24,882 --> 00:45:28,922
Speaker 3: and cities. And then in the white paper it says, ultimately,

689
00:45:29,241 --> 00:45:33,882
Speaker 3: it's not just these things, these technologies that are going

690
00:45:33,922 --> 00:45:36,802
Speaker 3: to get us there because these technolog like they list

691
00:45:37,002 --> 00:45:40,442
Speaker 3: all the AI tech techniques that they list have absolutely

692
00:45:40,482 --> 00:45:43,882
Speaker 3: nothing to do with large scale AI generative AI systems.

693
00:45:44,122 --> 00:45:48,922
Speaker 3: They're all techniques that have been around for basically a decade.

694
00:45:49,082 --> 00:45:53,082
Speaker 3: And they were like, that is evidence enough to show

695
00:45:53,122 --> 00:45:56,281
Speaker 3: you that it's not actually a technical problem at the

696
00:45:56,362 --> 00:45:59,002
Speaker 3: end of the day. And so that was kind of

697
00:45:59,002 --> 00:46:01,082
Speaker 3: what I was getting at in that line, is like,

698
00:46:02,042 --> 00:46:09,161
Speaker 3: these companies are creating AI systems currently that are straining resources,

699
00:46:09,201 --> 00:46:13,562
Speaker 3: which heats up geopolitical competition over those resources. They are

700
00:46:13,842 --> 00:46:19,522
Speaker 3: perpetuating an arms race narrative that also erodes the willingness

701
00:46:19,562 --> 00:46:23,801
Speaker 3: of different geopolitical powers to cooperate with one another. They

702
00:46:23,842 --> 00:46:28,962
Speaker 3: are creating machines of misinformation and disinformation that's eroding the

703
00:46:28,962 --> 00:46:32,082
Speaker 3: fabric of trust in various societies that is also the

704
00:46:32,082 --> 00:46:36,562
Speaker 3: building block for cohesion and cooperation. And so they're kind

705
00:46:36,602 --> 00:46:39,881
Speaker 3: of basically like, when you start to realize it that

706
00:46:40,282 --> 00:46:42,842
Speaker 3: client change isn't really a technology problem and it is

707
00:46:42,842 --> 00:46:46,002
Speaker 3: a social problem, all of the things that we need

708
00:46:47,201 --> 00:46:53,522
Speaker 3: to fortify are actually being chipped away at maybe not

709
00:46:53,602 --> 00:46:57,361
Speaker 3: chipped away, like halfed away by the industry with the

710
00:46:57,402 --> 00:46:59,362
Speaker 3: current systems that they're building.

711
00:47:00,042 --> 00:47:02,402
Speaker 1: There's a very hopeful point that you make in a

712
00:47:02,442 --> 00:47:06,042
Speaker 1: couple of places throughout the book, which is that despite

713
00:47:06,082 --> 00:47:10,322
Speaker 1: the power that these empires have, they're not entirely invulnerable.

714
00:47:11,082 --> 00:47:13,281
Speaker 1: Could you share one of the stories of some of

715
00:47:13,282 --> 00:47:17,042
Speaker 1: the communities that have successfully fought back. I was particularly

716
00:47:17,602 --> 00:47:20,922
Speaker 1: moved by the way, you know, in response to some

717
00:47:21,002 --> 00:47:24,522
Speaker 1: of these companies trying to come into communities in Chile

718
00:47:24,802 --> 00:47:28,962
Speaker 1: or elsewhere. Yeah, under the secrecy of shell companies trying

719
00:47:28,962 --> 00:47:31,602
Speaker 1: to sneak in a massive data center that's going to

720
00:47:31,642 --> 00:47:33,802
Speaker 1: use up all the water in an already drought stricken

721
00:47:33,962 --> 00:47:36,842
Speaker 1: part of the world. People see what's going on, and

722
00:47:36,882 --> 00:47:40,642
Speaker 1: they mobilize, and they have had some really inspiring successes.

723
00:47:41,562 --> 00:47:45,282
Speaker 3: Yeah. So one reason I also think the empire analogy

724
00:47:45,362 --> 00:47:48,962
Speaker 3: is extremely pertinent is because empires are made to feel inevitable,

725
00:47:49,002 --> 00:47:52,762
Speaker 3: and yet they've always fallen in history, and it's because

726
00:47:52,802 --> 00:47:57,201
Speaker 3: their foundations are extremely weak. You cannot actually ultimately sustain

727
00:47:57,882 --> 00:48:01,401
Speaker 3: this degree of extractivism and exploitation has such a large

728
00:48:01,442 --> 00:48:05,722
Speaker 3: scale without people revolting, And that is what I document

729
00:48:05,762 --> 00:48:09,042
Speaker 3: happening in Chile, where Chile has a very long history

730
00:48:09,042 --> 00:48:13,482
Speaker 3: of extractivism and being a concept that came out of

731
00:48:13,602 --> 00:48:17,241
Speaker 3: decolonial scholars from Latin America talking about the harvesting of

732
00:48:17,282 --> 00:48:21,441
Speaker 3: their resources for the benefit of people far away rather

733
00:48:21,482 --> 00:48:27,122
Speaker 3: than the local community. And they essentially see the AI

734
00:48:27,241 --> 00:48:30,562
Speaker 3: industry and all the data centers and the mining that

735
00:48:30,602 --> 00:48:33,602
Speaker 3: the industry is doing to try and build these colossal

736
00:48:34,602 --> 00:48:38,721
Speaker 3: silicon monstrosities as just an extension of this history, like

737
00:48:38,882 --> 00:48:42,802
Speaker 3: yet again, we are dealing with a greedy industry trying

738
00:48:42,842 --> 00:48:48,721
Speaker 3: to extract at scale our precious earth. And so they

739
00:48:48,762 --> 00:48:53,602
Speaker 3: immediately became activated when they discovered that this was happening,

740
00:48:53,842 --> 00:48:58,122
Speaker 3: and I spoke with these activists in this community right

741
00:48:58,201 --> 00:49:03,241
Speaker 3: on the outskirts of Santiago called Serrijos. That was they

742
00:49:03,322 --> 00:49:06,482
Speaker 3: learned that Google was trying to build a data center

743
00:49:07,362 --> 00:49:12,042
Speaker 3: in the one town that actually has access to a

744
00:49:12,042 --> 00:49:15,602
Speaker 3: public drinking water source. Because in Chile, due to its

745
00:49:16,201 --> 00:49:20,161
Speaker 3: history of dictatorship, most things were privatized, including water, but

746
00:49:20,201 --> 00:49:23,962
Speaker 3: the one municipality that had a free public drinking water

747
00:49:24,002 --> 00:49:27,522
Speaker 3: resource was the municipality that Google then was like, yep,

748
00:49:27,562 --> 00:49:29,321
Speaker 3: We're going to put our data center in there and

749
00:49:29,362 --> 00:49:32,442
Speaker 3: then tap into this freshwater resource and use tool our facilities,

750
00:49:32,962 --> 00:49:37,721
Speaker 3: and this community lit up and was like absolutely not.

751
00:49:38,802 --> 00:49:42,642
Speaker 3: These activists started knocking on all of the neighbors doors,

752
00:49:43,201 --> 00:49:48,281
Speaker 3: posting flyers, creating you like memes and art, political art

753
00:49:48,442 --> 00:49:53,241
Speaker 3: and all this stuff to educate their neighbors that this

754
00:49:53,402 --> 00:49:57,082
Speaker 3: project is not at all going to benefit our community

755
00:49:57,082 --> 00:50:01,282
Speaker 3: and is in fact taking the one extremely precious resource

756
00:50:01,442 --> 00:50:04,842
Speaker 3: that we have to your point in a drought stricken time.

757
00:50:05,681 --> 00:50:08,642
Speaker 3: And they made so much noise that escalated all the

758
00:50:08,642 --> 00:50:12,281
Speaker 3: way to both Google's had quarters in Mountain View and

759
00:50:12,962 --> 00:50:18,761
Speaker 3: to the national Chilean government. And basically after this extraordinary

760
00:50:18,842 --> 00:50:22,442
Speaker 3: pressure where they've stalled this project for five years, the

761
00:50:22,681 --> 00:50:28,761
Speaker 3: Chilean government has now created a roundtable for consulting residents

762
00:50:28,962 --> 00:50:33,442
Speaker 3: and environmental activists on their data center plans and putting

763
00:50:33,482 --> 00:50:37,962
Speaker 3: them in conversation with companies like Google and Microsoft. It's

764
00:50:37,962 --> 00:50:40,282
Speaker 3: not a perfect solution, and that the activists have said,

765
00:50:41,161 --> 00:50:43,082
Speaker 3: the moment that they blank, everything can fall apart, and

766
00:50:43,122 --> 00:50:44,922
Speaker 3: so they have to continue being vigilant, they have to

767
00:50:44,962 --> 00:50:49,281
Speaker 3: continue protesting, resisting being on the streets. But it is

768
00:50:49,322 --> 00:50:53,441
Speaker 3: a remarkable step that they got the government to make

769
00:50:53,562 --> 00:50:56,201
Speaker 3: to even bring them to the table, and is just

770
00:50:56,241 --> 00:50:59,042
Speaker 3: a lesson to be learned by everyone around the world

771
00:50:59,042 --> 00:51:04,562
Speaker 3: that if you remember that you actually still have agency,

772
00:51:05,322 --> 00:51:13,642
Speaker 3: you can absolutely shape the trajectory of AI development by loudly,

773
00:51:14,082 --> 00:51:20,002
Speaker 3: forcefully making that kind of trouble when these companies are

774
00:51:20,042 --> 00:51:21,681
Speaker 3: engaging in this imperial activity.

775
00:51:22,522 --> 00:51:24,362
Speaker 1: So such a good note to end on. Thanks so

776
00:51:24,442 --> 00:51:28,002
Speaker 1: much for this conversation, Karen, and for what is truly

777
00:51:28,721 --> 00:51:32,842
Speaker 1: such an important piece of work in scholarship. So I

778
00:51:32,842 --> 00:51:35,401
Speaker 1: am grateful for the enormous amount of time that went

779
00:51:35,402 --> 00:51:35,841
Speaker 1: into it.

780
00:51:36,762 --> 00:51:42,922
Speaker 3: Thank you so much for having me Adam.

781
00:51:43,002 --> 00:51:47,602
Speaker 2: Drilled is an original Critical Frequency production. This episode was

782
00:51:47,642 --> 00:51:52,241
Speaker 2: reported and written by Adam Lowenstein and produced by Peter duff.

783
00:51:52,922 --> 00:51:56,201
Speaker 2: Artwork for Drill is by Matt Fleming, fact checking by

784
00:51:56,322 --> 00:52:00,281
Speaker 2: Naomi barr Our. First Amendment attorney is James Wheaton with

785
00:52:00,322 --> 00:52:05,122
Speaker 2: the First Amendment project. Drilled is distributed by Pushkin Industries.

786
00:52:05,522 --> 00:52:10,322
Speaker 2: Huge thanks to the team there, including Greta Cohen, Eric Sandler, Grease, Ross,

787
00:52:10,642 --> 00:52:16,522
Speaker 2: Morgan Rattner, Owen Miller, Kira Posey, Jordan McMillan, Brian Schreberneck,

788
00:52:16,721 --> 00:52:20,402
Speaker 2: and Jake Flanagan. You can find a written version of

789
00:52:20,442 --> 00:52:24,761
Speaker 2: this interview and lots of other author interviews, stories, and

790
00:52:24,802 --> 00:52:27,762
Speaker 2: all kinds of content on our website at drilled dot media.

791
00:52:28,201 --> 00:52:30,922
Speaker 2: You can also sign up for our newsletter there and

792
00:52:31,082 --> 00:52:34,482
Speaker 2: follow us on all the social media sites at drilled media.

793
00:52:34,562 --> 00:52:36,522
Speaker 2: Thanks for listening and we'll see you next time.