1
00:00:02,520 --> 00:00:07,040
Speaker 1: Bloomberg Audio Studios, podcasts, radio news.

2
00:00:07,880 --> 00:00:11,680
Speaker 2: President Trump signed an executive order aimed at limiting state

3
00:00:11,760 --> 00:00:15,040
Speaker 2: level regulation of AI. The move is supported by tech

4
00:00:15,160 --> 00:00:19,360
Speaker 2: leaders who have argued local rules could stifle innovation. We're

5
00:00:19,440 --> 00:00:22,479
Speaker 2: joined by David Sachs, the White House AI and Crypto

6
00:00:22,640 --> 00:00:25,800
Speaker 2: zah Or, Senior Advisor, and David, I think it is

7
00:00:25,840 --> 00:00:28,520
Speaker 2: a really good place to start in your work with

8
00:00:28,560 --> 00:00:32,960
Speaker 2: the President in consulting and advising on the formulation of

9
00:00:33,000 --> 00:00:37,159
Speaker 2: this executive order. What was the problem that you were

10
00:00:37,200 --> 00:00:40,440
Speaker 2: trying to solve for and what is it that you

11
00:00:40,560 --> 00:00:43,879
Speaker 2: said to the President about why this EO was the

12
00:00:43,960 --> 00:00:46,880
Speaker 2: right approach to focus on state level laws?

13
00:00:48,240 --> 00:00:49,599
Speaker 3: Well, thanks for having me.

14
00:00:49,840 --> 00:00:51,960
Speaker 4: The problem that we see is that you've got one

15
00:00:52,000 --> 00:00:55,640
Speaker 4: thousand different bills going through state legislatures right now to

16
00:00:55,720 --> 00:00:59,000
Speaker 4: regulate AI, and over one hundred measures already passed. Some

17
00:00:59,040 --> 00:01:01,440
Speaker 4: of these bills are country victory, and you've got fifty

18
00:01:01,480 --> 00:01:03,400
Speaker 4: different states running in fifty different directions.

19
00:01:03,760 --> 00:01:05,240
Speaker 3: That type of compliance.

20
00:01:04,800 --> 00:01:07,560
Speaker 4: Regime is going to be very hard for small companies

21
00:01:07,560 --> 00:01:11,640
Speaker 4: and startups, especially innovators, to comply with. And so what

22
00:01:11,680 --> 00:01:16,319
Speaker 4: we need is a single federal or national framework for

23
00:01:16,400 --> 00:01:18,800
Speaker 4: AI regulation and that's what the President has supported, and

24
00:01:18,800 --> 00:01:20,839
Speaker 4: by the way, he supported this for a long time.

25
00:01:20,880 --> 00:01:23,959
Speaker 4: If you go back to his July speech on AI

26
00:01:24,200 --> 00:01:27,240
Speaker 4: he called for a single national framework then. And what

27
00:01:27,240 --> 00:01:29,800
Speaker 4: we've done with the SEO now is to make clear

28
00:01:29,840 --> 00:01:33,040
Speaker 4: that is the administration's policy and to task members of

29
00:01:33,040 --> 00:01:35,760
Speaker 4: the administration to work with Congress to try and enact

30
00:01:36,160 --> 00:01:39,520
Speaker 4: that framework through legislation, because ultimately this needs to be

31
00:01:39,520 --> 00:01:42,400
Speaker 4: a law, and in the meantime create tools that the

32
00:01:42,400 --> 00:01:46,000
Speaker 4: administration can use to push back on examples of the

33
00:01:46,040 --> 00:01:48,480
Speaker 4: most onerous and excessive state regulations.

34
00:01:49,600 --> 00:01:52,160
Speaker 2: David, there is of course some pushback you know, on

35
00:01:52,200 --> 00:01:55,960
Speaker 2: the executive order from the states themselves, from other Republicans,

36
00:01:56,720 --> 00:01:59,840
Speaker 2: you know, as you know, like I studied the July

37
00:02:01,560 --> 00:02:05,040
Speaker 2: speech and strategy closely, a big part of it, you know,

38
00:02:05,160 --> 00:02:09,800
Speaker 2: was infrastructure related and about deregulation. The concern about this

39
00:02:10,000 --> 00:02:13,520
Speaker 2: latest executive order is that while it addresses your concerns

40
00:02:13,600 --> 00:02:17,359
Speaker 2: about many different pieces of state regulation, it does not

41
00:02:17,480 --> 00:02:20,200
Speaker 2: provide for a single federal framework.

42
00:02:22,280 --> 00:02:24,480
Speaker 4: Well, at the end of the day, that single federal

43
00:02:24,520 --> 00:02:27,600
Speaker 4: framework has be enacted through law, and we need Congress

44
00:02:27,639 --> 00:02:30,000
Speaker 4: to do that. And so the President has asked Congress

45
00:02:30,040 --> 00:02:32,000
Speaker 4: to do that, and these task members of the administration

46
00:02:32,320 --> 00:02:35,880
Speaker 4: to work with Congress to produce that framework. In the meantime,

47
00:02:35,880 --> 00:02:37,959
Speaker 4: what we've done here is articulated set of principles. We

48
00:02:38,080 --> 00:02:41,560
Speaker 4: said what values are important to us. We said that

49
00:02:42,160 --> 00:02:45,280
Speaker 4: we want to protect child safety, that's important. We want

50
00:02:45,280 --> 00:02:48,560
Speaker 4: to respect copyright, we want to preserve the ability of

51
00:02:48,600 --> 00:02:53,320
Speaker 4: local communities to choose what infrastructures and their communities. We're

52
00:02:53,360 --> 00:02:56,960
Speaker 4: not seeking to preempt the states in any of those areas.

53
00:02:57,000 --> 00:02:59,840
Speaker 4: So this is an important set of principles that we

54
00:03:00,120 --> 00:03:03,080
Speaker 4: have put forth. And at the same time, the EO

55
00:03:03,160 --> 00:03:05,839
Speaker 4: provides for a number of tools that can be used

56
00:03:05,840 --> 00:03:09,600
Speaker 4: to push back on excessive state regulation. And let me

57
00:03:09,639 --> 00:03:12,840
Speaker 4: just illustrate why I think this is so necessary. What

58
00:03:12,840 --> 00:03:16,040
Speaker 4: we're really talking about here is regulation of AI models

59
00:03:16,080 --> 00:03:20,560
Speaker 4: and algorithms. Well, think about how an AI model is developed.

60
00:03:20,600 --> 00:03:23,840
Speaker 4: You can have developers in one state of multiple states

61
00:03:23,919 --> 00:03:26,640
Speaker 4: writing the code. It can then be trained in a

62
00:03:26,720 --> 00:03:29,679
Speaker 4: data center in another state. You then can have inference

63
00:03:29,760 --> 00:03:33,200
Speaker 4: happen in another state, and the entire service is provided

64
00:03:33,240 --> 00:03:37,560
Speaker 4: over the Internet using national telecommunications infrastructure. So you're dealing

65
00:03:37,600 --> 00:03:39,440
Speaker 4: there with at least four different states and all of

66
00:03:39,520 --> 00:03:43,600
Speaker 4: them can lay claim to regulating those AI models, and

67
00:03:43,680 --> 00:03:46,720
Speaker 4: those regulations can be in contradiction with each other. Even

68
00:03:46,760 --> 00:03:49,520
Speaker 4: Democrat governors have admitted this as a problem. So just

69
00:03:49,600 --> 00:03:51,800
Speaker 4: the other day, Kathy hoklel the governor of New York

70
00:03:52,120 --> 00:03:56,520
Speaker 4: It basically said that she might prefer to enact California's

71
00:03:56,600 --> 00:03:59,120
Speaker 4: SB fifty three, which is a regulation that they just

72
00:03:59,160 --> 00:04:01,960
Speaker 4: passed in California, rather than the bill that her own

73
00:04:01,960 --> 00:04:04,840
Speaker 4: assembly gave her, the Raise Act, because she sees that, wait,

74
00:04:04,840 --> 00:04:07,280
Speaker 4: do we really want to create this patchwork in different regulations.

75
00:04:07,440 --> 00:04:10,200
Speaker 4: So even democratic governors are realizing this is a problem,

76
00:04:10,280 --> 00:04:12,800
Speaker 4: and if they all run in different directions, then we're

77
00:04:12,840 --> 00:04:15,280
Speaker 4: going to end up with a patchwork or a mismash

78
00:04:15,680 --> 00:04:18,120
Speaker 4: of regulations that are impossible for companies to comply with.

79
00:04:18,160 --> 00:04:21,360
Speaker 4: What the President is calling for here is just common sense.

80
00:04:21,600 --> 00:04:25,000
Speaker 4: We want to get to a single national framework of

81
00:04:25,080 --> 00:04:28,240
Speaker 4: compliance as opposed to fifty states running in different directions.

82
00:04:28,560 --> 00:04:31,200
Speaker 1: Meanwhile, Kathy Hokle actually is getting a bit of criticism,

83
00:04:31,240 --> 00:04:34,920
Speaker 1: perhaps for narrowing and what some are saying is bowing

84
00:04:34,960 --> 00:04:37,840
Speaker 1: down to business. David, I'm really interested in how you

85
00:04:37,920 --> 00:04:41,360
Speaker 1: oppose that view, because there is anxiety in the population,

86
00:04:41,839 --> 00:04:45,680
Speaker 1: AI versus jobs, AI versus energy bills. How are you

87
00:04:45,720 --> 00:04:48,960
Speaker 1: giving them the sense that we haven't seen federal government

88
00:04:49,000 --> 00:04:51,560
Speaker 1: and indeed now state governments just handing over the reins

89
00:04:51,600 --> 00:04:54,000
Speaker 1: to big tech billionaires as people call them.

90
00:04:54,839 --> 00:04:55,039
Speaker 3: Right.

91
00:04:55,080 --> 00:04:56,960
Speaker 4: No, I understand there's a lot of fear out there

92
00:04:57,040 --> 00:04:59,920
Speaker 4: about AI and job los specifically, and a lot of

93
00:05:00,040 --> 00:05:02,640
Speaker 4: those fears have been drummed up. Let me just say

94
00:05:02,640 --> 00:05:04,320
Speaker 4: on the job loss question, because I think this is

95
00:05:04,400 --> 00:05:07,120
Speaker 4: really important that Yale just released a study and it

96
00:05:07,240 --> 00:05:09,880
Speaker 4: showed that in the thirty three months after the launch

97
00:05:09,920 --> 00:05:13,080
Speaker 4: of chat GPT, there was no discernible disruption to the

98
00:05:13,160 --> 00:05:16,640
Speaker 4: US job market none. They said, no discernible disruption. And

99
00:05:16,680 --> 00:05:18,960
Speaker 4: in fact, if you look right now, more jobs are

100
00:05:19,000 --> 00:05:21,200
Speaker 4: being created than being lost. So this whole idea of

101
00:05:21,320 --> 00:05:23,599
Speaker 4: job losses just isn't true. There was an article on

102
00:05:23,600 --> 00:05:26,280
Speaker 4: the Wall Street Journal just last week talking about the

103
00:05:26,880 --> 00:05:31,160
Speaker 4: construction boom that's happening that's benefiting construction workers like electricians,

104
00:05:31,200 --> 00:05:34,800
Speaker 4: like plumbers, like workers who pour concrete or hang drywall.

105
00:05:35,000 --> 00:05:37,159
Speaker 3: Their wages are up thirty percent.

106
00:05:36,800 --> 00:05:39,279
Speaker 4: Because this infrastructure boom that's happening right now, and there's

107
00:05:39,320 --> 00:05:42,719
Speaker 4: actually a job shortage in many of those trades, meaning

108
00:05:42,920 --> 00:05:45,720
Speaker 4: we need more workers going into those trades. So what

109
00:05:45,720 --> 00:05:48,160
Speaker 4: we're seeing right now is an overall AI boom that's

110
00:05:48,160 --> 00:05:52,360
Speaker 4: benefiting the economy. You know, the GDP growth rate was

111
00:05:52,400 --> 00:05:55,560
Speaker 4: tracking about four percent, and half of that up to

112
00:05:55,560 --> 00:05:58,479
Speaker 4: half of it's been attributed to AI. So I just

113
00:05:58,520 --> 00:06:01,720
Speaker 4: think that this narrow about job loss has been blown

114
00:06:01,760 --> 00:06:05,120
Speaker 4: out of proportion. Certainly there could be job displacement in

115
00:06:05,160 --> 00:06:07,440
Speaker 4: the future, but we haven't seen any of that so far.

116
00:06:07,640 --> 00:06:09,560
Speaker 3: It's been quite the opposite. It has been job gains.

117
00:06:10,360 --> 00:06:13,400
Speaker 2: David final one on the EO, if I may, you

118
00:06:13,440 --> 00:06:17,080
Speaker 2: know what this EO allows for. Is it the sort

119
00:06:17,080 --> 00:06:20,560
Speaker 2: of hope that it will lead to the DOJ sewing

120
00:06:20,760 --> 00:06:24,599
Speaker 2: states like New York and California. And if that's the case,

121
00:06:25,080 --> 00:06:28,039
Speaker 2: you know the President and the administration's confidence that you'd

122
00:06:28,080 --> 00:06:29,480
Speaker 2: win them.

123
00:06:29,960 --> 00:06:31,800
Speaker 4: Well, that is one of the tools that is in

124
00:06:31,839 --> 00:06:34,719
Speaker 4: the EO is that the DOJ has been tasked to

125
00:06:34,760 --> 00:06:38,600
Speaker 4: form a litigation task Force that would have the ability

126
00:06:38,640 --> 00:06:43,000
Speaker 4: to push back on excessively burn some state laws, laws

127
00:06:43,040 --> 00:06:46,440
Speaker 4: that may be unconstitutional, violate the First Amendment, things like that.

128
00:06:46,440 --> 00:06:48,919
Speaker 4: By the way, the DOJ already had that power, so

129
00:06:49,000 --> 00:06:51,680
Speaker 4: this is not a novel power. But what's being done

130
00:06:51,720 --> 00:06:53,960
Speaker 4: here in the CEO is we're marshaling all the resources

131
00:06:53,960 --> 00:06:56,599
Speaker 4: of the federal government behind the strategy of the President

132
00:06:56,600 --> 00:06:58,760
Speaker 4: to create a national framework. Now, in terms of what

133
00:06:58,880 --> 00:07:01,160
Speaker 4: laws would go after, that's a decision that has been made.

134
00:07:02,360 --> 00:07:05,440
Speaker 4: We haven't decided whether California and New York should be

135
00:07:05,480 --> 00:07:08,600
Speaker 4: targets in that way. The one that I think is

136
00:07:08,600 --> 00:07:12,400
Speaker 4: probably the most successive is this Colorado law that seeks

137
00:07:12,400 --> 00:07:15,960
Speaker 4: to prohibit algorithmic discrimination. What that basically says is that

138
00:07:15,960 --> 00:07:18,280
Speaker 4: if an AI model has a disparate impact on a

139
00:07:18,320 --> 00:07:22,760
Speaker 4: protected group, then that model is violating the law. Model developers,

140
00:07:22,760 --> 00:07:24,160
Speaker 4: by the way, I have no idea how to comply

141
00:07:24,280 --> 00:07:26,960
Speaker 4: with this because they're not aware of all the downstream

142
00:07:27,080 --> 00:07:29,320
Speaker 4: uses of their model. I mean, if a business decides

143
00:07:29,320 --> 00:07:32,240
Speaker 4: to use an AI model in a hiring decision, for example,

144
00:07:32,760 --> 00:07:34,840
Speaker 4: that business is already on the hook for discrimination. So

145
00:07:34,840 --> 00:07:37,920
Speaker 4: how would the model developer know that it was being

146
00:07:38,000 --> 00:07:39,920
Speaker 4: used in that way? But what Colorado was trying to

147
00:07:39,960 --> 00:07:43,840
Speaker 4: do there is get their ideology inserted into the model.

148
00:07:43,880 --> 00:07:45,000
Speaker 3: That's very concerning to us.

149
00:07:45,120 --> 00:07:47,160
Speaker 4: We think there's a first amendment issue there, But look,

150
00:07:47,200 --> 00:07:49,360
Speaker 4: we haven't made any decisions in terms of how that

151
00:07:49,600 --> 00:07:51,040
Speaker 4: litigation task for US to be used.

152
00:07:51,120 --> 00:07:53,680
Speaker 1: David Briefly, all of this is set in the context

153
00:07:53,680 --> 00:07:57,440
Speaker 1: of US versus China and a deemed to run forward

154
00:07:57,680 --> 00:08:00,960
Speaker 1: on AI development. Meanwhile, a busy week and eight two

155
00:08:01,000 --> 00:08:03,880
Speaker 1: hundreds might indeed be able to get to China. How

156
00:08:03,920 --> 00:08:05,600
Speaker 1: many do you think you'll do in volumes and what

157
00:08:05,680 --> 00:08:07,280
Speaker 1: do you think the appetite is of China to buy

158
00:08:07,280 --> 00:08:08,960
Speaker 1: in videos more sophisticated chips.

159
00:08:10,160 --> 00:08:11,000
Speaker 3: Well, it's interesting.

160
00:08:11,040 --> 00:08:13,080
Speaker 4: I just saw an article that said that China was

161
00:08:13,120 --> 00:08:15,640
Speaker 4: rejecting the H two hundred, So apparently they don't want them,

162
00:08:15,880 --> 00:08:17,680
Speaker 4: and I think the reason for that is they want

163
00:08:17,680 --> 00:08:20,840
Speaker 4: semiconductor independence. The same way that the United States wanted

164
00:08:21,000 --> 00:08:23,960
Speaker 4: to be energy independent. They want to be semiconductor independent.

165
00:08:24,200 --> 00:08:26,800
Speaker 4: So they're rejecting our chips. And that's part of the

166
00:08:26,840 --> 00:08:30,240
Speaker 4: calculation that goes into the decision of what we authorized

167
00:08:30,280 --> 00:08:32,679
Speaker 4: to be sold to China. The US policy has always

168
00:08:32,720 --> 00:08:35,240
Speaker 4: been that we don't allow the leading edge chips, and

169
00:08:35,320 --> 00:08:35,640
Speaker 4: we're not.

170
00:08:36,160 --> 00:08:37,640
Speaker 3: This is this H two hundred chip.

171
00:08:37,880 --> 00:08:39,640
Speaker 4: It was state of the art a couple of years ago,

172
00:08:39,720 --> 00:08:42,320
Speaker 4: but now it's been superseded by the new or Blackwell

173
00:08:42,400 --> 00:08:45,280
Speaker 4: architecture and the Ruben architecture that's coming out next year.

174
00:08:45,600 --> 00:08:48,040
Speaker 4: So this is now a lagging chip, not a leading chip.

175
00:08:48,080 --> 00:08:50,440
Speaker 4: But what you see is China is not taking them

176
00:08:50,720 --> 00:08:53,600
Speaker 4: because they want to prop up and subsidize Wilwey.

177
00:08:53,640 --> 00:08:55,480
Speaker 3: They want to create a national champion.

178
00:08:55,840 --> 00:08:58,400
Speaker 4: And that was part of our calculation of selling not

179
00:08:58,559 --> 00:09:00,719
Speaker 4: the best but lagging chips tochch is you can take

180
00:09:00,800 --> 00:09:03,199
Speaker 4: market share away from Walwig. But I think the Chinese

181
00:09:03,240 --> 00:09:06,280
Speaker 4: government's figured that out and that's why they're not allowing them.

182
00:09:06,440 --> 00:09:09,240
Speaker 1: David Sachs, we always wish we had more time White House,

183
00:09:09,280 --> 00:09:11,720
Speaker 1: AI and cryptos Are We thank you for joining us

184
00:09:11,760 --> 00:09:12,880
Speaker 1: today on the executive Order