1 00:00:03,680 --> 00:00:05,840 Speaker 1: This week on the Business of Tech powered by two 2 00:00:05,840 --> 00:00:10,319 Speaker 1: Degrees Business Artificial Intelligence. Under the Trump administration, it's all 3 00:00:10,360 --> 00:00:15,720 Speaker 1: about sustaining and enhancing America's global AI dominance, according to Trump, 4 00:00:15,800 --> 00:00:19,479 Speaker 1: who on gaining office for a second time, revoked President 5 00:00:19,480 --> 00:00:24,560 Speaker 1: Biden's Executive Order on AI. That one emphasized safe, secure 6 00:00:24,680 --> 00:00:29,280 Speaker 1: and trustworthy developments and use of artificial intelligence. But Trump's 7 00:00:29,280 --> 00:00:33,520 Speaker 1: administration did quite a lot on AI in his first term. 8 00:00:33,560 --> 00:00:36,320 Speaker 1: Before chat GPT burst onto the scene and made it 9 00:00:36,320 --> 00:00:41,400 Speaker 1: a real mainstream issue. The US led multilateral cooperation on 10 00:00:41,520 --> 00:00:44,640 Speaker 1: many AI developments. So what can we expect from a 11 00:00:44,720 --> 00:00:49,520 Speaker 1: second Trump presidency, especially as new less resource intensive large 12 00:00:49,560 --> 00:00:52,919 Speaker 1: language models like deep sea copair to be changing the 13 00:00:52,960 --> 00:00:58,920 Speaker 1: economics of intensive AI applications and the geopolitical implications. Joining 14 00:00:58,960 --> 00:01:01,560 Speaker 1: me on episode ninety one of The Business of Tech 15 00:01:01,600 --> 00:01:05,840 Speaker 1: is Sarah Box, a Ministry of Business, Innovation and Employment 16 00:01:05,920 --> 00:01:08,880 Speaker 1: digital policy specialist who's one of the people in government 17 00:01:08,920 --> 00:01:11,800 Speaker 1: at the moment really focused on our approach to AI 18 00:01:11,959 --> 00:01:20,360 Speaker 1: as a nation. Sarah Box, Welcome to the Business of Tech. 19 00:01:20,400 --> 00:01:24,280 Speaker 1: Thanks so much for coming on now. You spent around 20 00:01:24,319 --> 00:01:26,840 Speaker 1: three months late last year in the United States as 21 00:01:26,840 --> 00:01:30,360 Speaker 1: a Harkness fellow. What a great opportunity. I was one 22 00:01:30,560 --> 00:01:33,000 Speaker 1: myself over a decade ago, looking at the future of 23 00:01:33,160 --> 00:01:37,000 Speaker 1: public interest journalism. I went to pro publica New York, 24 00:01:37,080 --> 00:01:40,720 Speaker 1: the Center for Public Integrity in Washington, DC. You know, 25 00:01:40,760 --> 00:01:44,520 Speaker 1: it was a fantastic experience. Your fellowship looked pretty exciting. 26 00:01:44,560 --> 00:01:49,000 Speaker 1: Three months examining AI policy direction in the US. Who 27 00:01:49,040 --> 00:01:50,960 Speaker 1: are you hosted by over there, Sarah? 28 00:01:51,000 --> 00:01:51,720 Speaker 2: That's right. Yeah. 29 00:01:51,800 --> 00:01:56,120 Speaker 3: So I organized a three months stay based in Washington, DC, 30 00:01:56,800 --> 00:02:00,240 Speaker 3: and I had the Observer Research Foundation America host me. 31 00:02:00,320 --> 00:02:02,200 Speaker 2: So that's a relatively. 32 00:02:01,760 --> 00:02:04,720 Speaker 3: Small think tank in DC, but they're affiliated with their 33 00:02:04,720 --> 00:02:09,960 Speaker 3: Indian counterpart, Observer Research Foundation, and so really good connections 34 00:02:09,960 --> 00:02:14,960 Speaker 3: that they have into folks with cybersecurity, AI, semiconductor, a 35 00:02:14,960 --> 00:02:17,640 Speaker 3: lot of related kind of policies to AI. So it 36 00:02:17,680 --> 00:02:19,519 Speaker 3: was a great place to be for that three months. 37 00:02:20,240 --> 00:02:22,840 Speaker 3: I also had a bit of time in San Francisco, Atlanta, 38 00:02:22,880 --> 00:02:24,520 Speaker 3: and a few other places, so I got to travel 39 00:02:24,560 --> 00:02:25,200 Speaker 3: around a little bit. 40 00:02:25,280 --> 00:02:25,440 Speaker 4: Yeah. 41 00:02:25,480 --> 00:02:29,200 Speaker 1: The Observer Research Foundation is a great organization, does a 42 00:02:29,240 --> 00:02:32,639 Speaker 1: lot of really good research and policy advisories. And we'll 43 00:02:32,680 --> 00:02:35,160 Speaker 1: put a link into show notes so that people can 44 00:02:35,240 --> 00:02:38,120 Speaker 1: access that. But boy, what a time for you to 45 00:02:38,200 --> 00:02:42,200 Speaker 1: be there obviously through the election campaign and then into 46 00:02:42,240 --> 00:02:45,000 Speaker 1: the election itself in the US, Trump winning by a 47 00:02:45,040 --> 00:02:49,240 Speaker 1: decent margin, coming home in December before he got a 48 00:02:49,320 --> 00:02:52,400 Speaker 1: chance really to enact DOGE and all the policies and 49 00:02:52,440 --> 00:02:56,000 Speaker 1: executive orders that he's issued. Walk us through the approach 50 00:02:56,120 --> 00:02:59,640 Speaker 1: of the Trump administration to AI, because it goes back 51 00:02:59,680 --> 00:03:02,239 Speaker 1: towards first term and a lot of the regulations and 52 00:03:02,320 --> 00:03:06,400 Speaker 1: philosophy around AI that is still in place today. 53 00:03:07,200 --> 00:03:08,000 Speaker 2: Yeah, that's right. 54 00:03:08,160 --> 00:03:10,560 Speaker 3: So I think that at the time that the first 55 00:03:10,560 --> 00:03:13,840 Speaker 3: Trump administration came in, that was when people started to 56 00:03:13,880 --> 00:03:17,600 Speaker 3: realize what applications AI really could lend itself to, and 57 00:03:17,639 --> 00:03:20,000 Speaker 3: so there was a lot more interest then in supporting 58 00:03:20,960 --> 00:03:23,360 Speaker 3: AI technology and then starting to think a little bit 59 00:03:23,360 --> 00:03:27,280 Speaker 3: about AI governance, and so in a way that the 60 00:03:27,280 --> 00:03:30,119 Speaker 3: first Trump administration really did set down some of these 61 00:03:30,240 --> 00:03:33,359 Speaker 3: basic policies as you say, that still are there today, 62 00:03:33,400 --> 00:03:38,640 Speaker 3: particularly related to R and D capabilities, personnel capabilities. 63 00:03:38,000 --> 00:03:38,480 Speaker 2: And so on. 64 00:03:39,040 --> 00:03:42,119 Speaker 3: And I think really at the time that the first 65 00:03:42,120 --> 00:03:44,840 Speaker 3: Trump administration was in, the thinking was very much around 66 00:03:44,960 --> 00:03:47,120 Speaker 3: how can we get the best out of AI? And 67 00:03:47,160 --> 00:03:50,040 Speaker 3: when Biden came in, it was almost like shifting to 68 00:03:50,080 --> 00:03:53,040 Speaker 3: thinking the worst of the technology. It's a little bit blunt, 69 00:03:53,120 --> 00:03:54,680 Speaker 3: I think, but that's sort of how I see the 70 00:03:54,720 --> 00:03:58,520 Speaker 3: general mood switch between the first Trump administration and the 71 00:03:58,640 --> 00:03:59,760 Speaker 3: change into Biden. 72 00:04:00,040 --> 00:04:01,960 Speaker 1: Which from as you put it in your report, are 73 00:04:01,960 --> 00:04:04,839 Speaker 1: focus by Trump and its first term on innovation and 74 00:04:04,920 --> 00:04:09,000 Speaker 1: capability development around AI to a pivot to trust and 75 00:04:09,040 --> 00:04:13,440 Speaker 1: safety under Biden, and his executive order was quite far reaching. 76 00:04:13,480 --> 00:04:16,160 Speaker 1: For instance, I think he wanted the likes of open 77 00:04:16,240 --> 00:04:19,640 Speaker 1: AI to basically show the government what they were doing. 78 00:04:20,440 --> 00:04:23,560 Speaker 1: Invite government officials in and say, this is what we're developing. 79 00:04:23,960 --> 00:04:27,039 Speaker 1: It's cutting edge technology. Here it is, here's the source code, 80 00:04:27,080 --> 00:04:30,360 Speaker 1: here are our models. So you can establish for yourself 81 00:04:30,480 --> 00:04:34,480 Speaker 1: as the government whether you think it's safe. Quite far reaching, yeah, 82 00:04:34,560 --> 00:04:35,040 Speaker 1: it was. 83 00:04:35,080 --> 00:04:37,599 Speaker 3: And if you think back to sort of that switch, 84 00:04:37,640 --> 00:04:40,240 Speaker 3: I mean back under the Trump administration, they really put 85 00:04:40,279 --> 00:04:44,640 Speaker 3: an emphasis on American leadership, and his AI Executive Order 86 00:04:45,080 --> 00:04:47,919 Speaker 3: was looking to prioritize federal R and D research. It 87 00:04:47,960 --> 00:04:51,760 Speaker 3: was about boosting compute and data resources and building up 88 00:04:51,800 --> 00:04:52,440 Speaker 3: the workforce. 89 00:04:53,160 --> 00:04:56,080 Speaker 2: And so one of the concrete policies they had, for example, was. 90 00:04:57,480 --> 00:05:02,760 Speaker 3: Creating twenty five AI institutes that were based in US universities. 91 00:05:03,160 --> 00:05:06,280 Speaker 3: They were focusing on different aspects of AI research, you know, 92 00:05:06,320 --> 00:05:08,760 Speaker 3: depending on the industries and whatever state they were in, 93 00:05:08,880 --> 00:05:13,279 Speaker 3: from agriculture to environmental science and building up talent. 94 00:05:14,720 --> 00:05:16,159 Speaker 2: They also updated. 95 00:05:15,720 --> 00:05:20,120 Speaker 3: The AI R and D Strategic Plan. They were out 96 00:05:20,120 --> 00:05:24,320 Speaker 3: there internationally. They championed the OCDAI Principles. They were very 97 00:05:25,400 --> 00:05:28,080 Speaker 3: active in the G seven. They supported the Canadians, for 98 00:05:28,080 --> 00:05:30,160 Speaker 3: instance in twenty eighteen when they came out with the 99 00:05:30,279 --> 00:05:35,440 Speaker 3: Chavois Common Vision for AI, which is all about grasping 100 00:05:35,440 --> 00:05:40,520 Speaker 3: the opportunities supporting entrepreneurship. And then you had Biden come 101 00:05:40,520 --> 00:05:43,359 Speaker 3: in and the first I guess notable policy there was 102 00:05:43,360 --> 00:05:46,359 Speaker 3: his blueprint for an AI Bill of Rights, which was 103 00:05:46,400 --> 00:05:51,120 Speaker 3: all about focusing on the risk of bias, discrimination and equities, 104 00:05:51,279 --> 00:05:53,240 Speaker 3: threats to privacy and so on. And then, as you say, 105 00:05:53,440 --> 00:05:57,200 Speaker 3: you had the Executive Order come in with a raft 106 00:05:57,279 --> 00:06:03,799 Speaker 3: of actions around AI governance, creating AI Safety Institute, requiring 107 00:06:03,839 --> 00:06:05,920 Speaker 3: reporting on these dual use models and so on. 108 00:06:06,080 --> 00:06:10,680 Speaker 2: So yeah, really, whilst I think Biden did. 109 00:06:10,560 --> 00:06:14,280 Speaker 3: Have some policies around capability innovation, there was a definite 110 00:06:14,360 --> 00:06:14,960 Speaker 3: switch at that. 111 00:06:14,960 --> 00:06:18,599 Speaker 1: Time, I guess during Biden's administration, that's when we saw 112 00:06:18,640 --> 00:06:22,040 Speaker 1: the launch of chat GPT, generative AI exploded into the 113 00:06:22,080 --> 00:06:26,000 Speaker 1: public consciousness, and you had the godfather of AI, Jeffrey 114 00:06:26,080 --> 00:06:29,840 Speaker 1: Hinton and others, basically saying, we think this is an 115 00:06:29,960 --> 00:06:34,880 Speaker 1: existential threat to society, artificial general intelligence. So he was 116 00:06:35,040 --> 00:06:39,039 Speaker 1: reading the signs Biden and going, we need to do 117 00:06:39,080 --> 00:06:39,800 Speaker 1: something about this. 118 00:06:40,160 --> 00:06:40,599 Speaker 2: I reckon. 119 00:06:40,640 --> 00:06:43,120 Speaker 3: This is going to be really fascinating to watch because 120 00:06:43,200 --> 00:06:47,480 Speaker 3: while in theory all of the actions under Biden's executive 121 00:06:47,560 --> 00:06:50,839 Speaker 3: Order are actually under review right now and could be terminated, 122 00:06:50,960 --> 00:06:53,840 Speaker 3: there are areas of common interest that you'd expect the 123 00:06:53,880 --> 00:06:57,680 Speaker 3: Trump administration could support, And the question is where the 124 00:06:57,720 --> 00:07:00,840 Speaker 3: optics of keeping something from the biden erar is just 125 00:07:01,240 --> 00:07:06,000 Speaker 3: it won't play with the Republican constituency. So, for instance, 126 00:07:06,200 --> 00:07:10,560 Speaker 3: Biden's executive order required annual risk assessments of AI in 127 00:07:10,680 --> 00:07:14,680 Speaker 3: critical infrastructure. You think maybe that might have bipartisan support. 128 00:07:15,760 --> 00:07:20,400 Speaker 3: That increased resources like data to startups and small business 129 00:07:20,400 --> 00:07:23,640 Speaker 3: trying to get competition, that asked the National Science Foundation 130 00:07:23,800 --> 00:07:28,200 Speaker 3: to reorient some of their existing funding around AI training, 131 00:07:29,160 --> 00:07:31,600 Speaker 3: trying to get more workforce development, and it all seems, 132 00:07:31,720 --> 00:07:34,920 Speaker 3: at least to me, to be consistent with Trump's goal 133 00:07:35,000 --> 00:07:37,560 Speaker 3: of American leadership, and you would hope that some of 134 00:07:37,600 --> 00:07:39,640 Speaker 3: that policy work might actually survive. 135 00:07:39,680 --> 00:07:42,400 Speaker 1: We've seen a lot of deal making. The Stargate project 136 00:07:42,520 --> 00:07:46,800 Speaker 1: launched the day after Tramp's inauguration with Oracle Open AI 137 00:07:46,880 --> 00:07:51,360 Speaker 1: and SoftBank putting five hundred billion into AI infrastructure. So 138 00:07:51,400 --> 00:07:54,360 Speaker 1: he's very much looking at it through the lens of investment. 139 00:07:54,960 --> 00:07:57,440 Speaker 1: But we've also seen the last couple of months a 140 00:07:57,480 --> 00:08:01,120 Speaker 1: lot of pulling back from multilateral arrange on a range 141 00:08:01,160 --> 00:08:05,280 Speaker 1: of issues. As you pointed out the OECD principles which 142 00:08:05,560 --> 00:08:07,920 Speaker 1: we as a nation have signed up to that happened 143 00:08:08,360 --> 00:08:11,680 Speaker 1: under the first Trump administration. What's your sense about how 144 00:08:11,720 --> 00:08:15,600 Speaker 1: much of a global consensus model Trump is willing to 145 00:08:15,680 --> 00:08:20,160 Speaker 1: pursue on AI. You know, we had the Bletchley Park Declaration. 146 00:08:20,200 --> 00:08:22,600 Speaker 1: The US is still part of that, so they're sort 147 00:08:22,600 --> 00:08:25,480 Speaker 1: of still in the camp looking to other countries to 148 00:08:25,520 --> 00:08:27,000 Speaker 1: collaborate on this to some extent. 149 00:08:27,080 --> 00:08:32,400 Speaker 3: Anyway, Yeah, yeah, So my first observation was that, you know, 150 00:08:32,559 --> 00:08:34,960 Speaker 3: Trump has nominated the same team that he had in 151 00:08:34,960 --> 00:08:38,240 Speaker 3: his first term, which you'd hope might bode well for 152 00:08:38,320 --> 00:08:42,439 Speaker 3: international engagement. So Michael Kratzios, he was the Chief Technology 153 00:08:42,440 --> 00:08:46,640 Speaker 3: Officer under Trump Mark one. He's now set to be 154 00:08:46,720 --> 00:08:49,559 Speaker 3: the Director of the Office of Science Technology Policy, which 155 00:08:50,400 --> 00:08:54,839 Speaker 3: leads out on AI. He was actually Michael Kratzios was 156 00:08:54,840 --> 00:08:57,679 Speaker 3: a key player in getting those OECD principles over the line. 157 00:08:58,559 --> 00:09:01,640 Speaker 3: He was ably assisted by Lynn Parker, who has also returned. 158 00:09:01,679 --> 00:09:04,280 Speaker 3: She's going to head up the President's Council of Advisors 159 00:09:04,280 --> 00:09:07,760 Speaker 3: for Science and Technology. Lynn was also at the G 160 00:09:07,880 --> 00:09:10,240 Speaker 3: twenty and I think, you know, both of them were 161 00:09:10,320 --> 00:09:15,679 Speaker 3: quite pragmatic and constructive operators in international engagements, and hopefully 162 00:09:15,720 --> 00:09:19,560 Speaker 3: that will continue. Having said that, of course, you know, 163 00:09:19,600 --> 00:09:22,760 Speaker 3: I think US engagement's probably going to become a bit 164 00:09:22,800 --> 00:09:25,840 Speaker 3: more muscular, if you like. We saw that at the 165 00:09:25,960 --> 00:09:30,200 Speaker 3: Paris AI Action Summit, where Vice President Advance was I 166 00:09:30,200 --> 00:09:32,280 Speaker 3: think crystal clear that the US was not going to 167 00:09:32,280 --> 00:09:37,320 Speaker 3: tolerate its firms being constrained by anti innovation policies put 168 00:09:37,360 --> 00:09:40,320 Speaker 3: in place by other countries. I think you're also going 169 00:09:40,360 --> 00:09:43,080 Speaker 3: to see the Trump administration making greater use of trade 170 00:09:43,080 --> 00:09:46,839 Speaker 3: policy and industrial policy to take forward their AI objectives, 171 00:09:46,840 --> 00:09:48,800 Speaker 3: and of course that's going to have trickle down effects 172 00:09:48,840 --> 00:09:51,559 Speaker 3: to other countries too. So yeah, it is going to 173 00:09:51,600 --> 00:09:53,319 Speaker 3: be interesting to watch this one player. 174 00:09:53,320 --> 00:09:57,160 Speaker 1: Yeah, vance in Paris. That was one of many provocative 175 00:09:57,200 --> 00:10:00,920 Speaker 1: speeches that the Vice President gave when he went Europe recently, 176 00:10:01,400 --> 00:10:04,240 Speaker 1: very critical of how the Europeans do things. Does he 177 00:10:04,320 --> 00:10:07,640 Speaker 1: make a good point though on AI? US leadership in 178 00:10:07,800 --> 00:10:11,120 Speaker 1: AI really has been down to that sort of permissionless 179 00:10:11,120 --> 00:10:14,800 Speaker 1: innovation to some extent, free of really tight regulation like 180 00:10:15,400 --> 00:10:19,760 Speaker 1: GDPR and the AI Act that the EU has now introduced. 181 00:10:19,760 --> 00:10:22,240 Speaker 3: Certainly for the firms operating in the US. So I've 182 00:10:22,280 --> 00:10:26,200 Speaker 3: been interested to read over the last a little a 183 00:10:26,200 --> 00:10:30,440 Speaker 3: little while about the submissions coming from AI firms into 184 00:10:30,559 --> 00:10:34,760 Speaker 3: OSTP regarding the AI Action Plan that they want to 185 00:10:36,080 --> 00:10:39,720 Speaker 3: put out by midyear. And so one of those submissions 186 00:10:39,800 --> 00:10:43,120 Speaker 3: I think it was open AI really playing the China 187 00:10:43,200 --> 00:10:47,079 Speaker 3: card and saying, look, if you want leadership, that's fine, 188 00:10:47,120 --> 00:10:49,600 Speaker 3: but you need to look after us. And essentially what 189 00:10:49,640 --> 00:10:54,160 Speaker 3: that means is pretty low regulation or no regulation on 190 00:10:54,200 --> 00:10:56,800 Speaker 3: the AI firms and sort of holding that over them, 191 00:10:56,800 --> 00:10:59,680 Speaker 3: I guess is a bit of a threat in terms 192 00:10:59,720 --> 00:11:01,240 Speaker 3: of the says that they put in place. 193 00:11:01,360 --> 00:11:03,520 Speaker 1: We've seen the arrival of deep Seek that was a 194 00:11:03,520 --> 00:11:06,800 Speaker 1: big moment for the AI industry, and tied into that 195 00:11:07,080 --> 00:11:10,720 Speaker 1: was a debate about what AI chips, which the US 196 00:11:10,800 --> 00:11:11,240 Speaker 1: is limited. 197 00:11:11,400 --> 00:11:12,160 Speaker 2: Yeah, that's right. 198 00:11:12,240 --> 00:11:16,480 Speaker 3: I mean, I think the deep Seak innovation took everyone 199 00:11:16,520 --> 00:11:19,839 Speaker 3: a little bit by surprise. They've shown that you can 200 00:11:19,880 --> 00:11:23,960 Speaker 3: train models at a far lower cost. It also, I 201 00:11:23,960 --> 00:11:28,360 Speaker 3: think shows you that when you keep a technology from 202 00:11:28,360 --> 00:11:30,280 Speaker 3: a country, as you know they were trying to do 203 00:11:30,320 --> 00:11:33,120 Speaker 3: with China and these more advanced chips, it just gives 204 00:11:33,720 --> 00:11:36,920 Speaker 3: the imperative or the impulse to innovate around it. And 205 00:11:36,960 --> 00:11:40,520 Speaker 3: so that's what they've done. And I think the emphasis 206 00:11:40,520 --> 00:11:42,720 Speaker 3: that China is now putting on some of these smaller, 207 00:11:42,800 --> 00:11:46,959 Speaker 3: faster models is actually quite compelling for small countries and 208 00:11:47,000 --> 00:11:49,120 Speaker 3: countries in the Global South who don't actually have the 209 00:11:49,160 --> 00:11:52,680 Speaker 3: resources of the US. So China's putting itself in a 210 00:11:52,760 --> 00:11:54,320 Speaker 3: pretty interesting position as well. 211 00:11:54,480 --> 00:11:57,280 Speaker 1: Pretty exciting for US as a small developed nation as 212 00:11:57,320 --> 00:12:01,200 Speaker 1: well to see Deep Seek and more recently Mannas Model 213 00:12:01,960 --> 00:12:04,160 Speaker 1: a marriage out of China. They suggest that we could 214 00:12:04,200 --> 00:12:06,840 Speaker 1: actually do quite a lot with models that require much 215 00:12:06,920 --> 00:12:08,600 Speaker 1: less hardware capacity. 216 00:12:09,040 --> 00:12:12,240 Speaker 3: Yeah, for sure, I think that's where you can see 217 00:12:12,240 --> 00:12:16,720 Speaker 3: some opportunities coming. We're never going to have the resources 218 00:12:16,760 --> 00:12:22,560 Speaker 3: of the US obviously for our investment, but we do 219 00:12:22,640 --> 00:12:25,440 Speaker 3: have unique sources of data. We've got smart people in 220 00:12:25,480 --> 00:12:28,400 Speaker 3: New Zealand, we can build some of these models at 221 00:12:28,440 --> 00:12:31,480 Speaker 3: a smaller scale in China show that it's possible. 222 00:12:31,640 --> 00:12:32,800 Speaker 4: Yeah. 223 00:12:32,920 --> 00:12:37,080 Speaker 1: So the geopolitical tension, I think, as you said, there 224 00:12:37,080 --> 00:12:40,600 Speaker 1: by restricting that and we've seen this in semiconductors in 225 00:12:40,679 --> 00:12:47,079 Speaker 1: general mobile technology restricting it is actually forcing your opponent 226 00:12:47,160 --> 00:12:50,559 Speaker 1: to innovate, which has been quite good for the Chinese. 227 00:12:50,640 --> 00:12:52,560 Speaker 1: But how do you see that playing out in terms 228 00:12:52,600 --> 00:12:55,840 Speaker 1: of how it affects the AI that US, as citizens 229 00:12:55,840 --> 00:12:59,480 Speaker 1: and consumers used. Do you see this sort of bipolarization 230 00:13:00,080 --> 00:13:04,640 Speaker 1: of technology two different camps and we experience our view 231 00:13:04,679 --> 00:13:09,000 Speaker 1: of AI chatbots based on Western or American technology, the 232 00:13:09,080 --> 00:13:11,640 Speaker 1: Chinese and the developing world have a different view. 233 00:13:13,280 --> 00:13:14,320 Speaker 2: Oh, great question. 234 00:13:15,000 --> 00:13:17,000 Speaker 3: You would hope that we don't see the world split 235 00:13:17,040 --> 00:13:21,000 Speaker 3: into two different camps of technology experience, and I think 236 00:13:21,480 --> 00:13:23,840 Speaker 3: the multinational players coming out of the US are keen 237 00:13:23,920 --> 00:13:25,840 Speaker 3: to maintain their global markets. 238 00:13:26,280 --> 00:13:27,840 Speaker 2: What I would say is that. 239 00:13:29,320 --> 00:13:32,560 Speaker 3: I don't think that the US would take kindly to 240 00:13:32,720 --> 00:13:39,760 Speaker 3: its allies supporting China too closely on technologies like AI, 241 00:13:39,880 --> 00:13:44,960 Speaker 3: and that probably includes things like procurement of any Chinese technology, 242 00:13:46,000 --> 00:13:48,080 Speaker 3: and it could go to the extent of, you know, 243 00:13:48,520 --> 00:13:51,160 Speaker 3: what sort of consumer products we're bringing in as well. 244 00:13:51,440 --> 00:13:53,880 Speaker 3: And I think it's worth remembering that back under the 245 00:13:53,920 --> 00:13:59,280 Speaker 3: first Trump administration, they banned the Chinese telecommunications company Huawei 246 00:13:59,360 --> 00:14:04,160 Speaker 3: from buying certain US tech without special approval and basically 247 00:14:04,240 --> 00:14:08,720 Speaker 3: barred its equipment from US telco networks on national security grounds. 248 00:14:08,800 --> 00:14:12,240 Speaker 3: So there's always going to be this dance between the 249 00:14:12,360 --> 00:14:16,760 Speaker 3: national security and the civilian kind of applications of AI. 250 00:14:16,840 --> 00:14:19,040 Speaker 1: I think one of the things you looked at when 251 00:14:19,080 --> 00:14:23,200 Speaker 1: you're in the US obviously federal policy, but the states 252 00:14:23,200 --> 00:14:24,000 Speaker 1: of the US. 253 00:14:23,960 --> 00:14:27,080 Speaker 4: Are also regulating around AI, and it's. 254 00:14:26,960 --> 00:14:31,800 Speaker 1: This sort of patchwork emerging of policies and regulations across 255 00:14:31,840 --> 00:14:36,600 Speaker 1: the state's. California has implemented, some have decided not to implement, 256 00:14:36,880 --> 00:14:39,760 Speaker 1: others that make curtail the AI industry, which is very 257 00:14:39,800 --> 00:14:43,200 Speaker 1: much centered in California. What was your sense about what's 258 00:14:43,280 --> 00:14:46,560 Speaker 1: going on there, what's driving activity at the state level. 259 00:14:47,040 --> 00:14:50,440 Speaker 3: Yeah, So something I hadn't fully appreciated, I think before 260 00:14:50,480 --> 00:14:52,360 Speaker 3: I went to the US is that you do have 261 00:14:52,480 --> 00:14:55,560 Speaker 3: AI policy coming not just from the executive branch of 262 00:14:55,600 --> 00:14:58,640 Speaker 3: the government, but also from Congress and also from the 263 00:14:58,760 --> 00:15:02,040 Speaker 3: state level politicians, and so at the end of last year, 264 00:15:02,080 --> 00:15:04,680 Speaker 3: there was something like more than seven hundred I think 265 00:15:04,800 --> 00:15:09,960 Speaker 3: AI bills under discussion across uh US states, And in 266 00:15:10,000 --> 00:15:15,000 Speaker 3: many instances, those bills were using slightly different definitions. I 267 00:15:15,040 --> 00:15:18,040 Speaker 3: got different scopes, I got different thresholds for risk and 268 00:15:18,240 --> 00:15:19,680 Speaker 3: and all of these sort of nuances. 269 00:15:21,280 --> 00:15:23,280 Speaker 2: A fair number of them were about. 270 00:15:23,880 --> 00:15:28,640 Speaker 3: Consequential decision making, so trying to ensure that government agencies 271 00:15:28,680 --> 00:15:32,800 Speaker 3: themselves weren't introducing bias or of discrimination when they're making 272 00:15:32,840 --> 00:15:36,880 Speaker 3: decisions about education, housing, and so forth. But you mentioned 273 00:15:36,920 --> 00:15:40,160 Speaker 3: the California bill, and that was one more angle towards 274 00:15:40,160 --> 00:15:44,200 Speaker 3: the technology itself. Out of those seven hundred bills, there's 275 00:15:44,200 --> 00:15:46,600 Speaker 3: really only a handful literally that. 276 00:15:46,560 --> 00:15:47,360 Speaker 2: Have been passed. 277 00:15:47,840 --> 00:15:51,600 Speaker 3: One of those was the Colorado Bill, which was focused 278 00:15:51,640 --> 00:15:55,560 Speaker 3: on AI systems and consequential decision making. It's not come 279 00:15:55,600 --> 00:15:57,520 Speaker 3: into force, it's due to come into force in twenty 280 00:15:57,560 --> 00:16:01,240 Speaker 3: twenty six, but already it's under review to reduce the 281 00:16:01,320 --> 00:16:04,280 Speaker 3: risk that it deters innovation and investment in Colorado. They 282 00:16:04,320 --> 00:16:08,480 Speaker 3: had something like two hundred firms petitioning the government just 283 00:16:08,520 --> 00:16:10,760 Speaker 3: a couple of weeks after it was passed, saying no 284 00:16:10,840 --> 00:16:12,640 Speaker 3: This is way too broad and too vague. This is 285 00:16:12,640 --> 00:16:16,960 Speaker 3: going to destroy the industry here in Colorado. Firms that 286 00:16:17,000 --> 00:16:19,560 Speaker 3: I spoke to in the US said, look, a state 287 00:16:19,600 --> 00:16:22,840 Speaker 3: by state approach really is costly for small firms. Big 288 00:16:22,880 --> 00:16:25,680 Speaker 3: tech can afford it, they can lawyer up. Small ones can't. 289 00:16:26,920 --> 00:16:29,360 Speaker 3: And I've seen in a number of submissions to OSTP 290 00:16:29,520 --> 00:16:32,720 Speaker 3: about this AI Action Plan that the federal government should 291 00:16:32,760 --> 00:16:37,200 Speaker 3: think about passing light touch legislation that will actually prempt 292 00:16:37,240 --> 00:16:40,240 Speaker 3: all of these state laws. And I think what that 293 00:16:40,360 --> 00:16:44,080 Speaker 3: is is a judgment that the regional flexibility that you 294 00:16:44,200 --> 00:16:48,480 Speaker 3: might have and the opportunity to experiment just isn't giving 295 00:16:48,480 --> 00:16:53,000 Speaker 3: you the benefits to outweigh the cost of the patchwork 296 00:16:53,120 --> 00:16:53,640 Speaker 3: if you like. 297 00:16:54,120 --> 00:16:58,080 Speaker 4: And I guess that is our advantage with I system here. 298 00:16:58,200 --> 00:17:01,240 Speaker 1: Sure, and things like Florida, if the water supply it 299 00:17:01,320 --> 00:17:05,840 Speaker 1: might be done by regional local councils, But when it 300 00:17:05,840 --> 00:17:08,160 Speaker 1: comes to something like AI, you know, we get guidance 301 00:17:08,200 --> 00:17:10,920 Speaker 1: from the government, and the government has said it wants 302 00:17:10,920 --> 00:17:15,680 Speaker 1: to take a proportionate and light touch approach to regulation 303 00:17:15,960 --> 00:17:19,560 Speaker 1: on AI. So you're not running into a scenario where 304 00:17:19,760 --> 00:17:21,520 Speaker 1: one part of the country is saying no, you have 305 00:17:21,560 --> 00:17:23,920 Speaker 1: to process data in this particular way and we need 306 00:17:23,960 --> 00:17:27,160 Speaker 1: to see your models and we're sort of that's good 307 00:17:27,160 --> 00:17:28,360 Speaker 1: from an innovation point. 308 00:17:28,200 --> 00:17:29,560 Speaker 2: Of view, absolutely. 309 00:17:29,640 --> 00:17:31,680 Speaker 3: I mean it would be ridiculous for New Zealand to 310 00:17:31,960 --> 00:17:35,879 Speaker 3: sort of split into different ways of doing things. And 311 00:17:35,920 --> 00:17:38,800 Speaker 3: I think this shows the importance of us being engaged 312 00:17:38,840 --> 00:17:42,880 Speaker 3: in international dialogues on AI. So you know, there are 313 00:17:43,480 --> 00:17:47,800 Speaker 3: guidelines as international standards being made out there and New 314 00:17:47,880 --> 00:17:50,280 Speaker 3: Zealand needs to have a seat at the table and 315 00:17:50,359 --> 00:17:52,760 Speaker 3: have a voice so that we can express, you know, 316 00:17:52,880 --> 00:17:55,080 Speaker 3: what are our interests and try and shape some of 317 00:17:55,119 --> 00:17:57,600 Speaker 3: those global discussions around AI. 318 00:17:59,520 --> 00:18:02,000 Speaker 2: Think keeping those international. 319 00:18:02,880 --> 00:18:06,600 Speaker 3: Connections as well helps us tap into some AI resources 320 00:18:06,600 --> 00:18:09,720 Speaker 3: that we might not otherwise have, so really important, I 321 00:18:09,720 --> 00:18:11,520 Speaker 3: think to keep reaching out. 322 00:18:11,920 --> 00:18:16,879 Speaker 1: You're highlighting the report energy infrastructure, which really has become 323 00:18:16,960 --> 00:18:19,359 Speaker 1: a talking point in relation to AI. We have data 324 00:18:19,359 --> 00:18:22,320 Speaker 1: center as being built here, which is great. You know, 325 00:18:22,359 --> 00:18:25,119 Speaker 1: there's a lot of productivity that can come from AI 326 00:18:25,560 --> 00:18:29,359 Speaker 1: that likes a WS and Microsoft, but energy is a 327 00:18:29,400 --> 00:18:33,960 Speaker 1: sticking point in some countries. The AI chips use a 328 00:18:33,960 --> 00:18:37,280 Speaker 1: lot of energy to create, train these models and operate 329 00:18:37,359 --> 00:18:41,000 Speaker 1: them as well. What was the feeling you got there 330 00:18:41,119 --> 00:18:43,119 Speaker 1: about where this is going to go and what we 331 00:18:43,160 --> 00:18:46,200 Speaker 1: can learn from how to approach the energy equation. 332 00:18:47,040 --> 00:18:50,359 Speaker 3: Yeah, well, the US, I think has a real problem 333 00:18:50,359 --> 00:18:54,200 Speaker 3: with aging energy infrastructure. They've got a problem with transmission 334 00:18:54,800 --> 00:19:00,119 Speaker 3: and apparently real snarl ups with consenting and permitting. A 335 00:19:00,200 --> 00:19:03,120 Speaker 3: study by Goldman Sachs last year that suggested the US 336 00:19:03,280 --> 00:19:07,680 Speaker 3: needed fifty billion dollars of investment in new generation capacity 337 00:19:07,800 --> 00:19:11,440 Speaker 3: just for data centers. You saw last year some of 338 00:19:11,480 --> 00:19:14,640 Speaker 3: the tech firms starting to take action for themselves. Some 339 00:19:14,760 --> 00:19:18,520 Speaker 3: were working with energy companies on nuclear options, particularly these 340 00:19:18,960 --> 00:19:21,960 Speaker 3: small modular reactors that you can build pretty quick and 341 00:19:22,000 --> 00:19:26,280 Speaker 3: close to the grid as a way of pursuing carbon neutral, 342 00:19:26,400 --> 00:19:29,080 Speaker 3: carbon free goals while also getting the energy that they need. 343 00:19:30,040 --> 00:19:33,840 Speaker 3: You mentioned Stargate earlier. You know, that's five hundred billions, 344 00:19:34,000 --> 00:19:38,200 Speaker 3: mind boggling number of investment in AI and data centers 345 00:19:38,560 --> 00:19:42,920 Speaker 3: and US locations. President Trump's also smoothing the path I 346 00:19:42,960 --> 00:19:46,600 Speaker 3: think for energy production. And you've seen recent announcements coming 347 00:19:46,600 --> 00:19:49,159 Speaker 3: out suggesting that there may be new sources of energy 348 00:19:49,160 --> 00:19:54,639 Speaker 3: coming from Alaskan resource sort of developments and offshore drilling. 349 00:19:54,680 --> 00:19:58,359 Speaker 3: And I read something also about LNG. So there's a 350 00:19:58,359 --> 00:20:01,639 Speaker 3: lot going on around energy in the US at the moment. 351 00:20:02,200 --> 00:20:05,159 Speaker 3: Back here in n Z, I think the most recent 352 00:20:05,200 --> 00:20:09,439 Speaker 3: assessment from Transpower, which came out in the middle of 353 00:20:09,520 --> 00:20:13,359 Speaker 3: last year, suggested that demand from data centers wasn't actually 354 00:20:13,359 --> 00:20:15,800 Speaker 3: posing a risk to security of supply for US in 355 00:20:15,840 --> 00:20:19,840 Speaker 3: the next ten ten years or so. We've got a 356 00:20:19,840 --> 00:20:23,800 Speaker 3: pretty green electricity grid already. I think it's over eighty 357 00:20:23,800 --> 00:20:28,000 Speaker 3: five percent renewables. You've got firms like Microsoft signing deals 358 00:20:28,000 --> 00:20:31,400 Speaker 3: with energy providers here for renewable energy, which can help 359 00:20:31,840 --> 00:20:34,840 Speaker 3: fund new generation. I think it's going to be interesting 360 00:20:34,880 --> 00:20:38,800 Speaker 3: to watch the government's new investment drive, you know, engaging 361 00:20:38,800 --> 00:20:42,760 Speaker 3: with foreign investors to build infrastructure so on see whether 362 00:20:42,800 --> 00:20:46,080 Speaker 3: that well, some of that will flow into electricity and generation. 363 00:20:47,560 --> 00:20:50,879 Speaker 3: The question for me though, around all this energy debate 364 00:20:51,040 --> 00:20:53,479 Speaker 3: is well, what is the capacity for So if New 365 00:20:53,520 --> 00:20:56,800 Speaker 3: Zealand is not a hub for AI training, if models 366 00:20:56,840 --> 00:21:00,600 Speaker 3: are getting smaller and less computed intensive, then what is 367 00:21:00,640 --> 00:21:02,680 Speaker 3: actually the core on energy that we need to meet? 368 00:21:03,080 --> 00:21:05,359 Speaker 2: And that's for me what the interesting question is. 369 00:21:05,600 --> 00:21:08,400 Speaker 1: That is very interesting because you know, when I talk 370 00:21:08,480 --> 00:21:11,200 Speaker 1: to the likes of Vanessa Sair incident Microsoft, I said, 371 00:21:11,640 --> 00:21:13,959 Speaker 1: you know what is actually in these data centers, are 372 00:21:14,000 --> 00:21:19,480 Speaker 1: you loading them up with AI chips to do model training? 373 00:21:19,560 --> 00:21:23,560 Speaker 1: She said no, No, it's actually the traditional stuff you'd expect, hosting, 374 00:21:23,680 --> 00:21:28,600 Speaker 1: data processing, data applications, all of the sorts of services that. 375 00:21:28,560 --> 00:21:29,960 Speaker 4: Are moving to the cloud. 376 00:21:30,000 --> 00:21:32,359 Speaker 1: It's actually not a big component of the technology in 377 00:21:32,400 --> 00:21:35,600 Speaker 1: those New Zealand data centers. That may change over time, 378 00:21:35,640 --> 00:21:37,679 Speaker 1: and you have the likes of Data Grid that have 379 00:21:37,720 --> 00:21:41,520 Speaker 1: a plan to establish AI centric data centers in Southland 380 00:21:41,560 --> 00:21:43,280 Speaker 1: because it's efficient to cool. 381 00:21:43,520 --> 00:21:44,760 Speaker 4: Data centers down there. 382 00:21:45,119 --> 00:21:48,600 Speaker 1: But from what I've seen, they're all buying assurances of 383 00:21:49,160 --> 00:21:52,200 Speaker 1: megawatts of capacity and that doesn't seem to be wearing 384 00:21:52,240 --> 00:21:55,840 Speaker 1: the electricity sector. It suggests that for the next decade anyway, 385 00:21:55,920 --> 00:21:58,240 Speaker 1: there's enough supply to meet their demands. 386 00:21:58,520 --> 00:22:01,879 Speaker 3: It seems so from from what I can see, the 387 00:22:01,920 --> 00:22:04,119 Speaker 3: market seems to be looking after itself at the moment, 388 00:22:04,160 --> 00:22:07,000 Speaker 3: which is a good thing. And as long as New 389 00:22:07,080 --> 00:22:11,520 Speaker 3: Zealanders are able to continue to power their homes at 390 00:22:11,520 --> 00:22:14,720 Speaker 3: the same time as they can power new innovations and 391 00:22:15,359 --> 00:22:17,119 Speaker 3: new business models and so on, then I think we're 392 00:22:17,160 --> 00:22:17,800 Speaker 3: in a good spot. 393 00:22:17,960 --> 00:22:18,720 Speaker 4: Yeah. 394 00:22:18,800 --> 00:22:20,520 Speaker 1: One of the things you would have noticed in the 395 00:22:20,560 --> 00:22:23,480 Speaker 1: campaign over there is the sort of the anti woke 396 00:22:23,720 --> 00:22:27,280 Speaker 1: movement that is part of the MAGA movements and all 397 00:22:27,320 --> 00:22:27,520 Speaker 1: of that. 398 00:22:27,800 --> 00:22:28,880 Speaker 4: And I guess there. 399 00:22:28,840 --> 00:22:33,000 Speaker 1: Are some genuine concerns about generative AI, in particular what 400 00:22:33,080 --> 00:22:37,600 Speaker 1: content it's drawing on and what slanted and bias it 401 00:22:37,680 --> 00:22:40,399 Speaker 1: puts on information. Was there anything you picked up there 402 00:22:40,440 --> 00:22:43,760 Speaker 1: from your discussions with experts about how we deal with 403 00:22:43,800 --> 00:22:49,080 Speaker 1: this issue off trying to give people factual information, but 404 00:22:49,160 --> 00:22:52,920 Speaker 1: dealing with that bias, that potential political or ideological leaning, 405 00:22:52,960 --> 00:22:56,320 Speaker 1: and information that is going to color a lot of 406 00:22:56,320 --> 00:22:57,840 Speaker 1: these large language models. 407 00:22:58,720 --> 00:22:58,960 Speaker 2: Yeah. 408 00:22:59,040 --> 00:23:02,760 Speaker 3: So one of the critics from Republicans is that AI 409 00:23:02,920 --> 00:23:06,200 Speaker 3: safety standards are operating kind of as this woke policy 410 00:23:06,240 --> 00:23:10,879 Speaker 3: that advances diversity or it stifles free speech and is 411 00:23:10,920 --> 00:23:16,399 Speaker 3: ideologically driven rather than being innovation and competitiveness driven. For me, 412 00:23:17,680 --> 00:23:20,520 Speaker 3: I think if we can try and be clearer about 413 00:23:20,520 --> 00:23:22,400 Speaker 3: what we mean about some of these terms, that would 414 00:23:22,400 --> 00:23:25,280 Speaker 3: be great. So for me, AI safety has become this buzzphrase. 415 00:23:25,720 --> 00:23:31,199 Speaker 3: It's really ill defined, means anything people want it to mean. Instead, 416 00:23:31,200 --> 00:23:34,119 Speaker 3: if we can talk about, for example, physical safety risks 417 00:23:34,119 --> 00:23:37,320 Speaker 3: where AI is interacting with the physical world, whether it's 418 00:23:37,320 --> 00:23:41,160 Speaker 3: in critical infrastructure or smart city applications, or health hospitality. 419 00:23:41,600 --> 00:23:43,600 Speaker 3: You know, you can get a hold of that and 420 00:23:43,640 --> 00:23:46,679 Speaker 3: think about, Okay, what's the evidence based technical solutions to that. 421 00:23:46,720 --> 00:23:48,920 Speaker 2: It's very straightforward, nothing woke about that. 422 00:23:49,800 --> 00:23:52,520 Speaker 3: But when you try and get into describing potential job 423 00:23:52,600 --> 00:23:58,080 Speaker 3: loss as an AI safety issue, you're just really complicating matters. 424 00:23:58,080 --> 00:24:00,680 Speaker 3: And I think we can steer away from these politically 425 00:24:00,680 --> 00:24:03,399 Speaker 3: divisive approaches just with being a bit clearer about what 426 00:24:03,440 --> 00:24:03,800 Speaker 3: we need. 427 00:24:05,240 --> 00:24:09,680 Speaker 1: So you've come back from the US here, we have 428 00:24:10,119 --> 00:24:12,680 Speaker 1: a few things going on around AI policy. We've got 429 00:24:13,160 --> 00:24:15,159 Speaker 1: the recent guidelines on the Use of AI in the 430 00:24:15,160 --> 00:24:20,000 Speaker 1: public Sector that was released, We've got the National AI 431 00:24:20,080 --> 00:24:24,000 Speaker 1: Strategy in development. But have you got any sort of 432 00:24:24,359 --> 00:24:26,560 Speaker 1: real takeaways when you came back on the plane where 433 00:24:26,560 --> 00:24:28,320 Speaker 1: you thought, wow, there's like two or three things here 434 00:24:28,359 --> 00:24:31,120 Speaker 1: that we could really apply in New Zealand to really 435 00:24:31,119 --> 00:24:34,560 Speaker 1: good effect when it comes to trustworthy AI, but also 436 00:24:35,119 --> 00:24:39,919 Speaker 1: enabling our own companies to really innovate in this space. 437 00:24:40,720 --> 00:24:41,080 Speaker 2: Yeah. 438 00:24:41,400 --> 00:24:45,480 Speaker 3: So, one thing I really appreciated about the US is 439 00:24:45,880 --> 00:24:49,399 Speaker 3: the long term thinking, and so you can see that 440 00:24:49,680 --> 00:24:52,840 Speaker 3: for many, many years they have invested in technology R 441 00:24:52,880 --> 00:24:55,119 Speaker 3: and D, and that has put them in good stead 442 00:24:55,160 --> 00:24:59,280 Speaker 3: for building an AI ecosystem, and I'd like to see 443 00:24:59,320 --> 00:25:05,320 Speaker 3: that sort of bipartisan long term thinking really rooted in 444 00:25:05,480 --> 00:25:06,800 Speaker 3: here in New Zealand as well. 445 00:25:07,960 --> 00:25:09,800 Speaker 2: I think the government here has already. 446 00:25:09,560 --> 00:25:13,120 Speaker 3: Adopted one key US policy tenant, and that's this innovation 447 00:25:13,280 --> 00:25:16,840 Speaker 3: friendly approach, which is seeking to harness the opportunities of AI. 448 00:25:17,040 --> 00:25:19,520 Speaker 3: I think AI really is a technology that needs to 449 00:25:19,560 --> 00:25:22,479 Speaker 3: be normalized. It's a tool that can help us make 450 00:25:22,520 --> 00:25:25,200 Speaker 3: better decisions and be more productive, make our. 451 00:25:25,400 --> 00:25:26,440 Speaker 2: Resources go further. 452 00:25:28,080 --> 00:25:30,879 Speaker 3: You know, New Zealand's been in the productivity at oldrooms 453 00:25:30,640 --> 00:25:33,879 Speaker 3: for decades and we can't afford not to adopt AI. 454 00:25:34,080 --> 00:25:39,080 Speaker 3: So this innovation friendly stance I think is a good 455 00:25:39,080 --> 00:25:44,280 Speaker 3: way forward. I quite liked the regulatory advice that came 456 00:25:44,320 --> 00:25:47,119 Speaker 3: out under the first Trump administration, which was still i 457 00:25:47,119 --> 00:25:50,760 Speaker 3: guess in place under Biden, that set out good regulatory 458 00:25:51,080 --> 00:25:57,240 Speaker 3: practice principles like leveraging scientific and technical information, pursuing flexible 459 00:25:57,280 --> 00:26:00,119 Speaker 3: and tech neutral approaches, which really I think helps you 460 00:26:00,200 --> 00:26:04,760 Speaker 3: focus your limited public sector resources, policy maker resources on 461 00:26:04,840 --> 00:26:07,440 Speaker 3: issues where there's evidence of a problem, or there's evidence 462 00:26:07,440 --> 00:26:11,040 Speaker 3: that there's significant ambiguity that firms don't know what's what 463 00:26:11,200 --> 00:26:14,040 Speaker 3: in the regulatory space, and also where you have an 464 00:26:14,080 --> 00:26:17,879 Speaker 3: intent and ability to follow through on implementing new laws 465 00:26:18,240 --> 00:26:24,720 Speaker 3: or regulations. I did and do like the risk management 466 00:26:24,760 --> 00:26:28,840 Speaker 3: framework that came out of the National Institute Standards and Technology. 467 00:26:29,440 --> 00:26:32,320 Speaker 3: That's a really good example, I think of voluntary guidelines. 468 00:26:32,920 --> 00:26:35,560 Speaker 3: It's been very influential in other countries, and there was 469 00:26:35,720 --> 00:26:38,080 Speaker 3: actually growing momentum in the US for that to be 470 00:26:38,119 --> 00:26:40,520 Speaker 3: a safe harbor. So, for instance, if you had a 471 00:26:40,520 --> 00:26:44,280 Speaker 3: piece of legislation that was requiring certain actions around AI governance, 472 00:26:44,640 --> 00:26:48,280 Speaker 3: and as a firm, you'd implemented this risk management framework, 473 00:26:48,840 --> 00:26:52,720 Speaker 3: then you'd be deemed to comply. And that framework gives 474 00:26:52,960 --> 00:26:55,919 Speaker 3: just pretty practical tips to organizations on how to augmentor 475 00:26:56,200 --> 00:26:59,680 Speaker 3: or sharpen their existing practices around risk management and assurance, 476 00:26:59,720 --> 00:27:01,720 Speaker 3: isn't it. I think it's a really nice piece of 477 00:27:01,760 --> 00:27:06,760 Speaker 3: work that they're done. Maybe finally just to say, I 478 00:27:06,800 --> 00:27:12,440 Speaker 3: think that the US commitment to engaging internationally is something 479 00:27:12,440 --> 00:27:16,280 Speaker 3: I found compelling as well. Obviously, you know, and everything 480 00:27:16,320 --> 00:27:18,439 Speaker 3: they do, they have many more resources than us, and 481 00:27:18,480 --> 00:27:21,119 Speaker 3: we can never hope to be involved as much as 482 00:27:21,119 --> 00:27:24,560 Speaker 3: they are, but I just appreciated that they try to 483 00:27:24,600 --> 00:27:26,960 Speaker 3: take a leadership role, They try to engage and they 484 00:27:27,000 --> 00:27:30,639 Speaker 3: try to seek consensus with like minded countries, or at 485 00:27:30,680 --> 00:27:31,400 Speaker 3: least they did. 486 00:27:31,760 --> 00:27:34,800 Speaker 1: Yeah, Well, hopefully that will continue, because that is the fear, 487 00:27:34,880 --> 00:27:37,879 Speaker 1: is that it'll be more of a unilateral approach to 488 00:27:38,280 --> 00:27:39,399 Speaker 1: some of these things and. 489 00:27:41,000 --> 00:27:43,240 Speaker 4: More inward looking. So hopefully that. 490 00:27:43,240 --> 00:27:45,800 Speaker 1: Great work they've done they will build on. Just finally, 491 00:27:45,840 --> 00:27:49,399 Speaker 1: for us in New Zealand, we have been when it 492 00:27:49,400 --> 00:27:51,960 Speaker 1: comes to technology, a bit of a technology taker, you know, 493 00:27:52,560 --> 00:27:56,399 Speaker 1: and for this wave of AI that is particularly true. 494 00:27:56,560 --> 00:27:59,800 Speaker 1: I think a lot of our companies and consumers. Are 495 00:27:59,840 --> 00:28:04,120 Speaker 1: you using Chat, GPT, Gemini Copilot? 496 00:28:04,280 --> 00:28:04,399 Speaker 4: You know? 497 00:28:04,440 --> 00:28:09,240 Speaker 1: These models developed offshore don't necessarily reflect our culture and language, 498 00:28:09,320 --> 00:28:11,399 Speaker 1: and you notice that when you use them, it's a 499 00:28:11,480 --> 00:28:15,880 Speaker 1: very American centric view off the world. Know, what can 500 00:28:15,920 --> 00:28:19,320 Speaker 1: and what should we be doing to find our own 501 00:28:19,320 --> 00:28:20,400 Speaker 1: way and our own. 502 00:28:20,320 --> 00:28:21,560 Speaker 4: Uses off AI? 503 00:28:22,320 --> 00:28:25,480 Speaker 1: And at a fundamental level as well, what should we 504 00:28:25,640 --> 00:28:29,200 Speaker 1: be doing to build our own intellectual property in this space. 505 00:28:29,240 --> 00:28:32,639 Speaker 1: We've got the new public research organization coming that has 506 00:28:32,800 --> 00:28:36,680 Speaker 1: AI and its REMIT, so I guess there's an opportunity 507 00:28:36,720 --> 00:28:38,960 Speaker 1: there to do something a little bit more fundamental that's 508 00:28:39,000 --> 00:28:42,720 Speaker 1: going to hopefully add value to our economy. Specifically and 509 00:28:42,760 --> 00:28:46,280 Speaker 1: give us an edge competitively internationally. 510 00:28:46,760 --> 00:28:47,000 Speaker 2: Yeah. 511 00:28:47,120 --> 00:28:49,040 Speaker 3: So the way I think about it is that we 512 00:28:49,160 --> 00:28:52,800 Speaker 3: really need to maintain a focus on three blocks of 513 00:28:52,840 --> 00:28:55,960 Speaker 3: activity if you like. One is that AI development side, 514 00:28:56,760 --> 00:29:00,320 Speaker 3: one is the AI adoption and application, and then just 515 00:29:00,440 --> 00:29:04,440 Speaker 3: general AI seaviiness if you like. I don't think that 516 00:29:04,640 --> 00:29:07,560 Speaker 3: just because we're a small country and that we're normally 517 00:29:07,560 --> 00:29:10,240 Speaker 3: a tech taker, that we shouldn't have some competencies and 518 00:29:10,280 --> 00:29:14,040 Speaker 3: capabilities around developing AI. And you see that we do 519 00:29:14,160 --> 00:29:18,680 Speaker 3: have some groups. There's a group up in Northland, for example, 520 00:29:18,720 --> 00:29:20,760 Speaker 3: that's been using AI. 521 00:29:22,040 --> 00:29:25,000 Speaker 2: To establish a database of today or. 522 00:29:26,720 --> 00:29:30,600 Speaker 3: Commanity, so you know there's really neat work like that 523 00:29:30,600 --> 00:29:34,040 Speaker 3: that's going on. We have probably some amazing data sets 524 00:29:34,080 --> 00:29:38,040 Speaker 3: around I don't know, seismic activity, oceanic activity and so on, 525 00:29:38,400 --> 00:29:40,200 Speaker 3: and it's a matter of maybe using some of that 526 00:29:40,360 --> 00:29:44,680 Speaker 3: data better and if there's a way that this new 527 00:29:45,000 --> 00:29:47,880 Speaker 3: public research organization can contribute to that, that would be great. 528 00:29:47,920 --> 00:29:49,920 Speaker 3: I think that's still very much in early stages at 529 00:29:49,920 --> 00:29:52,480 Speaker 3: the moment of thinking about where that organization will go 530 00:29:52,560 --> 00:29:55,400 Speaker 3: and what its purpose will be, but really hopeful that 531 00:29:55,400 --> 00:29:57,960 Speaker 3: that will be able to step up and fill a 532 00:29:58,000 --> 00:30:01,920 Speaker 3: gap in our policy architecture if you like. So, Yeah, 533 00:30:02,320 --> 00:30:05,800 Speaker 3: definitely we need some capabilities around developing AI because that 534 00:30:05,880 --> 00:30:09,080 Speaker 3: means that we can take our own data, our own languages, 535 00:30:09,600 --> 00:30:12,920 Speaker 3: our own sort of culture and reflect that in the 536 00:30:12,960 --> 00:30:16,120 Speaker 3: AI that we have. Then, in terms of AI application, 537 00:30:16,240 --> 00:30:20,160 Speaker 3: we've got smart businesses out there already developing applications, new 538 00:30:20,200 --> 00:30:22,240 Speaker 3: business models with AI. I think we've really got to 539 00:30:22,280 --> 00:30:26,760 Speaker 3: try and turbo charge that. And finally in that third bucket, 540 00:30:27,320 --> 00:30:30,480 Speaker 3: just really putting an emphasis on how can people in 541 00:30:30,520 --> 00:30:31,040 Speaker 3: New Zealand. 542 00:30:31,080 --> 00:30:34,320 Speaker 2: CAI is just a normal part of the technology suite. 543 00:30:34,400 --> 00:30:36,400 Speaker 3: It's there every day, the taught in school how to 544 00:30:36,480 --> 00:30:39,600 Speaker 3: use it, the torta to think critically about it, and 545 00:30:39,640 --> 00:30:43,080 Speaker 3: it's just part of the fabric of society and what 546 00:30:43,120 --> 00:30:43,480 Speaker 3: we do. 547 00:30:45,240 --> 00:30:48,280 Speaker 1: And just to finish off, Sarah, where we at on 548 00:30:48,480 --> 00:30:51,920 Speaker 1: the sort of the timeline of things that government is 549 00:30:51,960 --> 00:30:54,080 Speaker 1: working on. Can you give us just a quick overview 550 00:30:54,080 --> 00:30:56,360 Speaker 1: of the things that are in train and maybe the 551 00:30:56,440 --> 00:30:59,040 Speaker 1: timeline on things like the AI strategy. 552 00:30:59,440 --> 00:31:03,560 Speaker 3: Yeah, certainly speak to the elements of work that my 553 00:31:03,680 --> 00:31:07,080 Speaker 3: ministry has been supporting the government on. So you heard 554 00:31:07,280 --> 00:31:09,440 Speaker 3: I think I think it was October last year, Minister 555 00:31:09,520 --> 00:31:13,400 Speaker 3: Collins announced that there would be an AI strategy and 556 00:31:13,440 --> 00:31:17,240 Speaker 3: that it would be consulted on this year. So we've 557 00:31:17,240 --> 00:31:21,360 Speaker 3: been supporting now Minister Retti on developing that product and 558 00:31:21,600 --> 00:31:25,240 Speaker 3: hoping that that will be out, you know, in the 559 00:31:25,280 --> 00:31:27,880 Speaker 3: next little while. I can't give an exact date on that, 560 00:31:27,960 --> 00:31:30,120 Speaker 3: but you would have seen that the AI strategy has 561 00:31:30,120 --> 00:31:33,480 Speaker 3: actually been featured in government's Going for Growth plans as 562 00:31:33,560 --> 00:31:38,320 Speaker 3: part of its Innovation Science Tech pillar. So definitely the 563 00:31:38,320 --> 00:31:41,720 Speaker 3: government recognizes that this technology is important and we need 564 00:31:41,760 --> 00:31:45,640 Speaker 3: to be thinking strategically about it. The second deliverable or 565 00:31:45,720 --> 00:31:48,480 Speaker 3: product that my ministry has been helping the government on 566 00:31:48,640 --> 00:31:52,240 Speaker 3: is around a responsible AI guidance for firms or for businesses, 567 00:31:52,920 --> 00:31:55,600 Speaker 3: and that is taking a leaf I guess out of 568 00:31:55,800 --> 00:31:59,560 Speaker 3: products like the Risk Management framework in the US, also 569 00:31:59,640 --> 00:32:04,240 Speaker 3: similar products coming out of Singapore and other countries where 570 00:32:04,240 --> 00:32:07,760 Speaker 3: we're trying to give voluntary guidance to firms on how 571 00:32:07,760 --> 00:32:11,280 Speaker 3: they can think about the processes internally for having responsible 572 00:32:11,320 --> 00:32:15,720 Speaker 3: AI and developing good products and so on. Again, not 573 00:32:16,360 --> 00:32:20,400 Speaker 3: totally sure on the timeline of that, but hopefully pretty soon. 574 00:32:20,600 --> 00:32:22,800 Speaker 3: And I think that those two pieces of work will 575 00:32:22,800 --> 00:32:25,800 Speaker 3: be a really great addition to the policy architecture here 576 00:32:25,800 --> 00:32:29,040 Speaker 3: in NZ and our colleagues in the Tirement of Internal 577 00:32:29,080 --> 00:32:33,120 Speaker 3: Affairs Government Chief Digital Officer team. They're doing great work 578 00:32:33,160 --> 00:32:35,600 Speaker 3: on the public sector side as well, and as you mentioned, 579 00:32:35,600 --> 00:32:38,520 Speaker 3: you frameworks and other things coming out there. 580 00:32:38,680 --> 00:32:40,440 Speaker 4: Oh good, Well, there's a lot going on. 581 00:32:40,600 --> 00:32:45,560 Speaker 1: And what's your reflection on the experience of doing that fellowship. 582 00:32:45,600 --> 00:32:47,280 Speaker 1: I know I spent three or four months in the 583 00:32:47,360 --> 00:32:51,800 Speaker 1: US and it really did give me a whole new perspective. 584 00:32:52,080 --> 00:32:54,480 Speaker 1: Would you recommend it for people like yourself in government 585 00:32:54,520 --> 00:32:56,360 Speaker 1: policymakers who want. 586 00:32:56,240 --> 00:32:57,160 Speaker 4: To get that perspective? 587 00:32:57,760 --> 00:32:59,920 Speaker 1: Is it still relevant the US perspective on how to 588 00:33:00,120 --> 00:33:02,880 Speaker 1: do policy given the radical change that's going on? 589 00:33:03,480 --> 00:33:04,520 Speaker 2: Yeah, I did wonder. 590 00:33:04,600 --> 00:33:08,520 Speaker 3: You know, my fellowship spanned precisely half before and half 591 00:33:08,600 --> 00:33:10,600 Speaker 3: after the election. I thought, gosh, is everything that I'm 592 00:33:10,640 --> 00:33:13,320 Speaker 3: learning just completely irrelevant? And I would say no, it 593 00:33:13,480 --> 00:33:18,360 Speaker 3: was a really great experience steffing out of your day 594 00:33:18,440 --> 00:33:23,240 Speaker 3: job and having this self guided period of research in 595 00:33:23,280 --> 00:33:25,920 Speaker 3: a country that is, you know, it's pretty different to us. 596 00:33:25,960 --> 00:33:28,760 Speaker 3: We watch American TV shows, we listened to American music, 597 00:33:28,760 --> 00:33:31,360 Speaker 3: but being there and living there is a totally different thing, 598 00:33:32,400 --> 00:33:34,160 Speaker 3: and it was a real privilege to be able to 599 00:33:34,560 --> 00:33:37,160 Speaker 3: do that, a privilege to speak to some of the 600 00:33:37,200 --> 00:33:41,560 Speaker 3: experts there in DC and other places about their experience 601 00:33:41,560 --> 00:33:44,760 Speaker 3: with aopolicy where they thought aipolicy in the US was going, 602 00:33:45,840 --> 00:33:47,720 Speaker 3: and to establish a bit of a network with those 603 00:33:47,720 --> 00:33:50,640 Speaker 3: people too, which I hope you know I can maintain 604 00:33:50,720 --> 00:33:51,280 Speaker 3: for the future. 605 00:33:51,680 --> 00:33:53,560 Speaker 4: Good well, we'll put a link to your report. 606 00:33:53,720 --> 00:33:56,480 Speaker 1: Very good report as summarizing all of your experiences. 607 00:33:56,960 --> 00:33:59,120 Speaker 4: We'll link to that in the show notes. Thanks so 608 00:33:59,280 --> 00:34:00,120 Speaker 4: much for coming on. 609 00:34:00,200 --> 00:34:04,120 Speaker 1: Good luck for the road ahead helping inform AI policy 610 00:34:04,120 --> 00:34:04,760 Speaker 1: in New Zealand. 611 00:34:04,880 --> 00:34:06,920 Speaker 2: Thanks so much, Peter, great to be here. 612 00:34:13,960 --> 00:34:15,920 Speaker 1: Thanks so much to Sarah Box for coming on. You'll 613 00:34:15,920 --> 00:34:18,440 Speaker 1: find a link to a report on her fellowship visit 614 00:34:18,480 --> 00:34:20,560 Speaker 1: to the US in the show notes. Go to the 615 00:34:20,600 --> 00:34:24,440 Speaker 1: podcast section at businesses dot co dot nz. Stream the 616 00:34:24,440 --> 00:34:27,960 Speaker 1: Business of Tech podcast on iHeartRadio or wherever you get 617 00:34:28,000 --> 00:34:31,319 Speaker 1: your podcasts. That's dropping next Tuesday, and I'll catch you then.