1 00:00:02,720 --> 00:00:04,960 Speaker 1: Welcome to the Business of Tech powered by two Degrees, 2 00:00:05,000 --> 00:00:09,520 Speaker 1: your weekly dive into the issue shaping New Zealand's digital future. 3 00:00:09,720 --> 00:00:13,640 Speaker 1: I'm Peter Griffin, and today's episode tackles the big debate 4 00:00:13,800 --> 00:00:18,200 Speaker 1: in tech policy at the moment, how and when should 5 00:00:18,200 --> 00:00:23,040 Speaker 1: we regulate artificial intelligence. I'm joined by Tom barklof tech 6 00:00:23,079 --> 00:00:26,720 Speaker 1: policy expert and co founder of the brain Box Institute, 7 00:00:26,760 --> 00:00:29,960 Speaker 1: who's been at the forefront of local and global discussions 8 00:00:29,960 --> 00:00:33,280 Speaker 1: about how we manage the rise of artificial intelligence. 9 00:00:33,920 --> 00:00:35,120 Speaker 2: Tom argues against the. 10 00:00:35,120 --> 00:00:39,560 Speaker 1: Rush from academics and advocacy groups to put new AI 11 00:00:39,680 --> 00:00:43,519 Speaker 1: specific regulations in place now. He points out that we 12 00:00:43,560 --> 00:00:47,080 Speaker 1: already have a robust patchwork of legislation, everything from the 13 00:00:47,120 --> 00:00:50,519 Speaker 1: Privacy Act to the Crimes Act that's more than capable 14 00:00:50,520 --> 00:00:52,559 Speaker 1: of handling the challenges AI brings. 15 00:00:52,920 --> 00:00:54,720 Speaker 2: He argues, what we're. 16 00:00:54,520 --> 00:00:58,920 Speaker 1: Missing isn't more laws, but the coordination and practical understanding 17 00:00:59,040 --> 00:01:02,880 Speaker 1: to apply them effectively. As the debate heats up here 18 00:01:03,000 --> 00:01:06,399 Speaker 1: and overseas, Tom reminds us that regulation is rarely a 19 00:01:06,440 --> 00:01:11,039 Speaker 1: binary yes or no answer. It's a complex process driven 20 00:01:11,120 --> 00:01:16,720 Speaker 1: by existing statutes, international standards, and real world enforcement issues. 21 00:01:17,000 --> 00:01:20,560 Speaker 1: We'll unpack the pressure on government to act fast, the 22 00:01:20,680 --> 00:01:24,600 Speaker 1: risks of hasty decision making, which we've seen in Australia 23 00:01:24,720 --> 00:01:28,480 Speaker 1: with the looming introduction of the under sixteen social media 24 00:01:28,560 --> 00:01:31,480 Speaker 1: ban for instance, and why Tom believes that codes of 25 00:01:31,560 --> 00:01:35,560 Speaker 1: practice and even industry self regulation can serve as a 26 00:01:35,640 --> 00:01:40,000 Speaker 1: useful trial run before legislation catches up. Plus, we look 27 00:01:40,000 --> 00:01:44,320 Speaker 1: at New Zealand's potential competitive advantage with AI, the ability 28 00:01:44,440 --> 00:01:49,400 Speaker 1: to be the world's smartest, most reliable deployers of AI 29 00:01:49,480 --> 00:01:53,840 Speaker 1: systems if we focus on capability, digital literacy and a 30 00:01:53,880 --> 00:01:58,200 Speaker 1: coordinated national vision for AI. So here's my interview with 31 00:01:58,320 --> 00:02:08,880 Speaker 1: brain Box Institute's Tom Aracloth. Tom, Welcome to the Business 32 00:02:08,880 --> 00:02:11,000 Speaker 1: of Tech. Thanks for coming on, Thanks for having me. 33 00:02:11,280 --> 00:02:15,639 Speaker 1: A great organization you work for. We've had one of 34 00:02:15,680 --> 00:02:19,000 Speaker 1: your colleagues on the past. But just in case people 35 00:02:19,040 --> 00:02:22,360 Speaker 1: don't really know what Brainbox Institute actually is and does 36 00:02:22,520 --> 00:02:23,400 Speaker 1: give us the lowdown. 37 00:02:23,480 --> 00:02:26,360 Speaker 3: I've got a background in law and public policy and 38 00:02:26,840 --> 00:02:30,880 Speaker 3: what I concluded from working in other areas of policy 39 00:02:31,040 --> 00:02:33,400 Speaker 3: was that there was a really cool space to be 40 00:02:33,480 --> 00:02:38,679 Speaker 3: occupied for organizations that are not government, so that sit 41 00:02:38,720 --> 00:02:43,080 Speaker 3: outside government and can coordinate with other parties. In a 42 00:02:43,080 --> 00:02:46,520 Speaker 3: particular policy area, not political parties, I mean sort of 43 00:02:46,560 --> 00:02:49,560 Speaker 3: sector groups to kind of lead a bit of a 44 00:02:49,600 --> 00:02:54,480 Speaker 3: conversation so you can be really interested in the stuff 45 00:02:54,520 --> 00:02:57,880 Speaker 3: that other people are not so interested in. And luckily 46 00:02:57,919 --> 00:03:00,880 Speaker 3: for me, I kind of like doing the work. I 47 00:03:00,960 --> 00:03:04,320 Speaker 3: like thinking about tech policy, I like reading about it. 48 00:03:04,720 --> 00:03:07,760 Speaker 3: I like trying to synthesize everything together, and I like 49 00:03:07,919 --> 00:03:10,919 Speaker 3: trying to find a way through that's kind of productive 50 00:03:10,960 --> 00:03:15,000 Speaker 3: and efficient and useful in all of these spaces. So 51 00:03:15,040 --> 00:03:17,960 Speaker 3: I established the brain Box Institute with a co founder 52 00:03:18,000 --> 00:03:21,720 Speaker 3: back in twenty eighteen around a series of public interest 53 00:03:21,840 --> 00:03:27,080 Speaker 3: legal research projects, including one related to deep fakes and 54 00:03:27,160 --> 00:03:30,560 Speaker 3: synthetic media, which was all the way back in twenty nineteen, 55 00:03:30,639 --> 00:03:33,480 Speaker 3: which is kind of wild looking back at that. But 56 00:03:33,639 --> 00:03:36,440 Speaker 3: since then we've also done a range of I've done 57 00:03:36,440 --> 00:03:40,680 Speaker 3: a range of projects related to disinformation. For example, with 58 00:03:40,720 --> 00:03:44,240 Speaker 3: the Department of the Prime Minister and Cabinet, I led 59 00:03:44,560 --> 00:03:51,320 Speaker 3: a global multi stakeholder coalition around technology company transparency, particularly 60 00:03:51,320 --> 00:03:56,160 Speaker 3: in a social media context. And I've also really lent 61 00:03:56,320 --> 00:03:59,600 Speaker 3: into this topic of how to turn law and regulation 62 00:03:59,800 --> 00:04:03,720 Speaker 3: and code and structured data. So the brain Box Institute 63 00:04:03,840 --> 00:04:06,480 Speaker 3: is the think tank and a consultancy company, but I 64 00:04:06,480 --> 00:04:08,320 Speaker 3: also spend a lot of my time thinking about how 65 00:04:08,360 --> 00:04:11,680 Speaker 3: to build the tools for doing this work more effectively, 66 00:04:11,840 --> 00:04:14,480 Speaker 3: and that's for another company called syncer Pate Lab. 67 00:04:15,160 --> 00:04:17,520 Speaker 1: It's a great work that you've been doing in recent 68 00:04:17,600 --> 00:04:21,040 Speaker 1: years and really excellent to have like a think tank 69 00:04:21,240 --> 00:04:24,599 Speaker 1: like brain box on the landscape at the moment with 70 00:04:24,680 --> 00:04:26,240 Speaker 1: so much going on and tech. You know, a lot 71 00:04:26,279 --> 00:04:28,360 Speaker 1: of this stuff used to be done a little bit 72 00:04:28,440 --> 00:04:30,920 Speaker 1: like by Internet New Zealand. They had a really good 73 00:04:30,960 --> 00:04:34,080 Speaker 1: policy arm that sort of faded in recent years. The 74 00:04:34,120 --> 00:04:36,760 Speaker 1: academics do their thing, but it's very much through the 75 00:04:36,839 --> 00:04:39,000 Speaker 1: academic lens. So something that is sort of at that 76 00:04:39,080 --> 00:04:44,440 Speaker 1: interface of government, research, business and what actually is working 77 00:04:44,680 --> 00:04:48,920 Speaker 1: is great, and particularly for artificial intelligence, which you've been 78 00:04:48,960 --> 00:04:53,440 Speaker 1: increasingly focusing on. You've got a four part blog series 79 00:04:53,640 --> 00:04:57,560 Speaker 1: out really interesting at the moment, all about AI regulation. 80 00:04:58,440 --> 00:05:00,880 Speaker 1: So we're going to get into but before we do, 81 00:05:01,000 --> 00:05:03,880 Speaker 1: look at what's going on in the scene here. This 82 00:05:03,920 --> 00:05:09,080 Speaker 1: is moving pretty quickly overseas. You know, various government's approaches 83 00:05:09,120 --> 00:05:12,240 Speaker 1: to regulation. How would you sort of summon up at 84 00:05:12,279 --> 00:05:15,880 Speaker 1: the moment where you've got sort of the US really 85 00:05:15,920 --> 00:05:18,159 Speaker 1: forging head wanting to be the leader. You've got China 86 00:05:18,200 --> 00:05:21,800 Speaker 1: doing its thing wanting to have AI supremacy as well. 87 00:05:21,960 --> 00:05:24,800 Speaker 1: You've got the EU have legislation in place now, the 88 00:05:25,560 --> 00:05:28,599 Speaker 1: AI Act. What's your take on where it's all going 89 00:05:28,600 --> 00:05:29,119 Speaker 1: at the moment. 90 00:05:29,520 --> 00:05:33,080 Speaker 3: My top line take would be, and I think this 91 00:05:33,200 --> 00:05:37,200 Speaker 3: is really relevant for local discussions, even if you take 92 00:05:37,839 --> 00:05:42,640 Speaker 3: kind of the most strident approach to regulating artificial intelligence, 93 00:05:42,800 --> 00:05:46,960 Speaker 3: and that is really the European Union approach, and it's 94 00:05:47,000 --> 00:05:49,520 Speaker 3: not just the AI Act, it's also you know, the 95 00:05:49,600 --> 00:05:53,920 Speaker 3: GDPR has had provisions in them around automated decision making 96 00:05:53,960 --> 00:05:57,640 Speaker 3: systems since about twenty sixteen, and we're still working out 97 00:05:57,680 --> 00:06:01,919 Speaker 3: really what it means to regulate automated decision making systems 98 00:06:02,160 --> 00:06:04,600 Speaker 3: in that way. I guess what I'd say is, even 99 00:06:04,640 --> 00:06:07,480 Speaker 3: if you have that kind of really strident approach, what 100 00:06:07,520 --> 00:06:12,520 Speaker 3: you find is the top level legislation can be quite general, 101 00:06:12,839 --> 00:06:15,000 Speaker 3: so it can be kind of oriented towards like risk 102 00:06:15,040 --> 00:06:18,520 Speaker 3: assessment and things like that. And what you need to 103 00:06:18,560 --> 00:06:20,640 Speaker 3: do to kind of flesh that out is have this 104 00:06:20,760 --> 00:06:25,120 Speaker 3: kind of cascade of other regulatory instruments, so you have 105 00:06:25,240 --> 00:06:27,599 Speaker 3: kind of delegated acts and you have guidance, and you 106 00:06:27,640 --> 00:06:31,200 Speaker 3: have guidelines and you have you know, best practice stuff, 107 00:06:31,720 --> 00:06:34,280 Speaker 3: and then you have all the kind of institutional infrastructure 108 00:06:34,279 --> 00:06:36,560 Speaker 3: that sits around that too to kind of support the 109 00:06:36,600 --> 00:06:41,920 Speaker 3: analysis of that information and reporting on it and scrutinizing 110 00:06:42,000 --> 00:06:44,560 Speaker 3: of it. And that's even before you get into any 111 00:06:44,560 --> 00:06:47,480 Speaker 3: of the kind of hard edged enforcement stuff where you're 112 00:06:47,480 --> 00:06:50,120 Speaker 3: actually talking about living fines and telling people what they 113 00:06:50,120 --> 00:06:53,640 Speaker 3: can and can't do. And you know, aside from actually 114 00:06:53,800 --> 00:06:58,720 Speaker 3: deciding to impose a penalty, there's still litigation underway now 115 00:06:59,720 --> 00:07:04,680 Speaker 3: about the GDPR from what are we now nearly ten 116 00:07:04,760 --> 00:07:07,240 Speaker 3: years ago to kind of work out what all of 117 00:07:07,279 --> 00:07:11,200 Speaker 3: this means. So my top line summary would be that 118 00:07:12,400 --> 00:07:14,880 Speaker 3: it's not a kind of binary exercise with all of this. 119 00:07:15,040 --> 00:07:17,640 Speaker 3: It's not like we're just going to say, let's regulate 120 00:07:17,720 --> 00:07:22,520 Speaker 3: AI and then tomorrow AI will be regulated. From a 121 00:07:22,600 --> 00:07:26,320 Speaker 3: kind of starting point, it's much more gray in terms 122 00:07:26,320 --> 00:07:28,840 Speaker 3: of what we already have in place and how we 123 00:07:28,960 --> 00:07:31,600 Speaker 3: use that more effectively. And then even if we did 124 00:07:31,640 --> 00:07:34,480 Speaker 3: decide to just really kick things off and go hard, 125 00:07:34,720 --> 00:07:37,000 Speaker 3: it would still be a pretty long process of trying 126 00:07:37,000 --> 00:07:40,520 Speaker 3: to work out what a lot of this regulatory stuff means. 127 00:07:41,040 --> 00:07:44,360 Speaker 3: I think I'll just add that you have mentioned China 128 00:07:44,400 --> 00:07:46,800 Speaker 3: and other countries. The other thing to be aware of 129 00:07:46,960 --> 00:07:50,280 Speaker 3: is just the scale of all the different regulatory approaches 130 00:07:50,280 --> 00:07:53,560 Speaker 3: out there is just enormous. And there's another piece to 131 00:07:53,600 --> 00:07:56,160 Speaker 3: think about, which I've commented on before, and I think 132 00:07:56,160 --> 00:07:59,840 Speaker 3: there's some good domestic discussion happening around, which is around 133 00:08:00,040 --> 00:08:04,880 Speaker 3: international standards. So quite often you'll have top level regulation 134 00:08:05,520 --> 00:08:08,640 Speaker 3: that says something like, you know, you should comply with 135 00:08:08,720 --> 00:08:11,520 Speaker 3: this standard, and the standard will be produced by a 136 00:08:11,560 --> 00:08:15,920 Speaker 3: completely different body that exists independently of any particular jurisdiction 137 00:08:16,640 --> 00:08:20,080 Speaker 3: and the regulation and people's compliance, but it tends to 138 00:08:20,080 --> 00:08:22,960 Speaker 3: converge towards that. So it's a really interesting space there 139 00:08:22,960 --> 00:08:25,880 Speaker 3: for us to play with, I think as a small country. 140 00:08:26,520 --> 00:08:29,760 Speaker 1: Yeah, and look, it's still a bit of a blank sheet. 141 00:08:29,800 --> 00:08:33,520 Speaker 1: We haven't raced in and enacted any legislation, which is 142 00:08:34,440 --> 00:08:38,240 Speaker 1: a good thing in many respects. Obviously, the academics that 143 00:08:38,520 --> 00:08:41,280 Speaker 1: work in and around artificial intelligence came out a couple 144 00:08:41,360 --> 00:08:45,080 Speaker 1: of months ago quite stridently an open letter to the 145 00:08:45,280 --> 00:08:49,160 Speaker 1: Prime Minister and the Minister of Technology saying you need 146 00:08:49,200 --> 00:08:52,120 Speaker 1: to regulate this. This is a general purpose technology the 147 00:08:52,280 --> 00:08:56,280 Speaker 1: likes we haven't seen before. Some of them arguing, you know, 148 00:08:56,679 --> 00:09:02,320 Speaker 1: we have dedicated legislation for nuclear for biotechnology that's being 149 00:09:02,360 --> 00:09:06,000 Speaker 1: reformed to biotech laws. At the moment, this is as 150 00:09:06,160 --> 00:09:09,120 Speaker 1: or more powerful than those technologies. We need to do something. 151 00:09:10,000 --> 00:09:13,560 Speaker 1: You wrote quite a thoughtful blog post basically saying, hang 152 00:09:13,600 --> 00:09:16,160 Speaker 1: on here, guys, we've actually got if you look at 153 00:09:16,200 --> 00:09:19,199 Speaker 1: the legislation we've got. Sure, it's a patchwork of legislation, 154 00:09:19,679 --> 00:09:22,520 Speaker 1: but there's a lot to work with here. You're basically saying, 155 00:09:22,559 --> 00:09:27,120 Speaker 1: let's focus on the various acts and regulations we already 156 00:09:27,120 --> 00:09:28,080 Speaker 1: have to do the job. 157 00:09:28,360 --> 00:09:30,760 Speaker 3: Yeah, there's kind of two pieces to what I've tried 158 00:09:30,800 --> 00:09:34,280 Speaker 3: to do. One is to well to promote coordination, but 159 00:09:34,400 --> 00:09:37,559 Speaker 3: also call out why I understand that that's really difficult. 160 00:09:38,120 --> 00:09:42,240 Speaker 3: That's come from experience where I have another area sort 161 00:09:42,240 --> 00:09:45,680 Speaker 3: of advocated for legislation. So there was one project that 162 00:09:45,720 --> 00:09:49,040 Speaker 3: we did around accessibility for disabled people and how do 163 00:09:49,080 --> 00:09:51,000 Speaker 3: you basically turn the Convention on the Rights of People 164 00:09:51,000 --> 00:09:54,959 Speaker 3: with Disabilities into an enforceable statute. And part of looking 165 00:09:54,960 --> 00:09:57,440 Speaker 3: into that has been understanding, well, what is the process 166 00:09:57,480 --> 00:10:00,240 Speaker 3: where we go from like a great idea, a bit 167 00:10:00,280 --> 00:10:03,120 Speaker 3: of legislation In New Zealand And along the way, there's 168 00:10:03,160 --> 00:10:06,280 Speaker 3: some really tough questions and they're really hard to answer, 169 00:10:06,720 --> 00:10:10,880 Speaker 3: and principle among those questions is what do we already 170 00:10:10,920 --> 00:10:12,760 Speaker 3: have in place? What's the gap? 171 00:10:13,080 --> 00:10:13,360 Speaker 1: You know? 172 00:10:13,840 --> 00:10:16,839 Speaker 3: The task of going and answering all of that is 173 00:10:17,080 --> 00:10:19,840 Speaker 3: really tough, and that's what kind of the first post 174 00:10:19,920 --> 00:10:23,079 Speaker 3: is about. It's about we've got information everywhere. We don't 175 00:10:23,120 --> 00:10:25,800 Speaker 3: know whether it's sort of current, it's not all tied 176 00:10:25,840 --> 00:10:28,839 Speaker 3: in one place, and it kind of relates to a 177 00:10:28,880 --> 00:10:31,840 Speaker 3: lot of different things, and it might be quite nonspecific anyway. 178 00:10:32,200 --> 00:10:35,439 Speaker 3: So there's an information problem that needs to be kind 179 00:10:35,440 --> 00:10:37,640 Speaker 3: of grappled with. And then part of that information problem 180 00:10:37,760 --> 00:10:41,600 Speaker 3: is kind of a coordination problem too. So there will 181 00:10:41,600 --> 00:10:47,120 Speaker 3: be fantastic information that is basically doing a systematic analysis 182 00:10:47,480 --> 00:10:51,040 Speaker 3: of everything that exists in New Zealand and the quality 183 00:10:51,160 --> 00:10:53,920 Speaker 3: of it somewhere. I'm sure if there isn't, I will 184 00:10:53,960 --> 00:10:57,560 Speaker 3: be stunned. You know, maybe it's in government, maybe it's 185 00:10:57,760 --> 00:11:00,199 Speaker 3: you know, in the private sector. I think I think 186 00:11:00,240 --> 00:11:03,040 Speaker 3: somebody has probably already done this analysis. And part of 187 00:11:03,080 --> 00:11:05,200 Speaker 3: this coordination problem is how can we make sure that 188 00:11:05,200 --> 00:11:08,480 Speaker 3: we're just reusing that rather than starting afresh, you know, 189 00:11:08,559 --> 00:11:10,600 Speaker 3: I would love to sit down and kind of look 190 00:11:10,640 --> 00:11:13,920 Speaker 3: at all the AI regulation that I've collated with other 191 00:11:13,960 --> 00:11:16,080 Speaker 3: team members of brain Box in the past, and the 192 00:11:16,120 --> 00:11:19,880 Speaker 3: AI Policy Tracker that we've pulled together, but that's going 193 00:11:19,960 --> 00:11:22,600 Speaker 3: to take a while. So the task of kind of 194 00:11:22,640 --> 00:11:24,680 Speaker 3: doing there is something that you want to do in 195 00:11:24,720 --> 00:11:28,040 Speaker 3: a systematic and a structured way in collaboration with others. 196 00:11:28,240 --> 00:11:30,400 Speaker 3: And I do think as well, you know, what's been 197 00:11:30,440 --> 00:11:33,080 Speaker 3: done with the open letter by under the leadership of 198 00:11:33,960 --> 00:11:37,840 Speaker 3: LINSA McGavin and Chris and Andrew there as fantastic because 199 00:11:37,880 --> 00:11:41,640 Speaker 3: it has pulled people together and as the start of 200 00:11:41,640 --> 00:11:44,160 Speaker 3: a fantastic discussion. I think about what we're going to 201 00:11:44,160 --> 00:11:44,760 Speaker 3: do about all this. 202 00:11:52,320 --> 00:11:53,400 Speaker 2: When you start digging in. 203 00:11:53,679 --> 00:11:56,760 Speaker 1: There are you know, legislation that can be drawn on there, 204 00:11:57,040 --> 00:12:00,320 Speaker 1: particularly the Privacy Act that's an important one, but there 205 00:12:00,360 --> 00:12:04,360 Speaker 1: are others as well. My concern is that we're going 206 00:12:04,400 --> 00:12:07,880 Speaker 1: to see something happen on a truncated timeline that has 207 00:12:07,920 --> 00:12:11,560 Speaker 1: happened with social media, where we have people officials and 208 00:12:11,600 --> 00:12:14,240 Speaker 1: government who've been told you need to get the under 209 00:12:14,240 --> 00:12:17,640 Speaker 1: sixteen social media ban ready to go like Australia has done. 210 00:12:17,679 --> 00:12:22,400 Speaker 1: It's a populist issue and it springs from deep seated 211 00:12:22,480 --> 00:12:26,480 Speaker 1: concern and genuine concern from parents about what is going 212 00:12:26,520 --> 00:12:31,480 Speaker 1: on in their children's lives online. So now we get that, 213 00:12:31,600 --> 00:12:35,920 Speaker 1: but they're hastily rushing to implement something that's potentially going 214 00:12:35,960 --> 00:12:38,200 Speaker 1: to do more harm than good. You know, we've got 215 00:12:38,240 --> 00:12:42,760 Speaker 1: issues already around algorithmic bias and AI systems, the potential 216 00:12:42,800 --> 00:12:47,840 Speaker 1: for mass surveillance, autonomous decision making in critical services. So 217 00:12:48,120 --> 00:12:50,400 Speaker 1: you know, does our legislation you know, deal with that 218 00:12:50,440 --> 00:12:52,880 Speaker 1: sort of thing. If someone wants to launch a chatbot 219 00:12:53,120 --> 00:12:55,920 Speaker 1: in New Zealand, our government has said we want a 220 00:12:56,200 --> 00:13:01,240 Speaker 1: light touch, proportional and risk based system that anyone could 221 00:13:01,280 --> 00:13:03,920 Speaker 1: launch a chatbot tomorrow and there's no risk assessment off 222 00:13:03,920 --> 00:13:06,680 Speaker 1: it is that going to be governed by any piece 223 00:13:06,720 --> 00:13:07,439 Speaker 1: of legislation. 224 00:13:07,640 --> 00:13:10,599 Speaker 3: Really great point. One of the things that is important 225 00:13:10,640 --> 00:13:14,679 Speaker 3: to me is to have a kind of baseivil society 226 00:13:14,840 --> 00:13:18,880 Speaker 3: capability around this kind of regulatory space, which I think 227 00:13:19,120 --> 00:13:22,160 Speaker 3: the Open Letter has really demonstrated to me that we do. 228 00:13:22,200 --> 00:13:24,360 Speaker 3: You want to have this base level of capability. So 229 00:13:24,760 --> 00:13:27,440 Speaker 3: when there is this kind of really popular swell to 230 00:13:27,520 --> 00:13:31,080 Speaker 3: basically do something, won't somebody do something that we've got 231 00:13:31,120 --> 00:13:33,200 Speaker 3: a kind of handbrake on that where we can at 232 00:13:33,280 --> 00:13:36,000 Speaker 3: least have an informed discussion about it. So the thing 233 00:13:36,040 --> 00:13:39,480 Speaker 3: for me around social media regulation and kind of banning 234 00:13:39,480 --> 00:13:43,720 Speaker 3: it for under sixteens. Sure, you know, there's a lot 235 00:13:43,760 --> 00:13:46,400 Speaker 3: of reason to be concerned, but nobody is sort of 236 00:13:46,440 --> 00:13:49,000 Speaker 3: advocating for it on the basis that, hey, wouldn't it 237 00:13:49,000 --> 00:13:51,439 Speaker 3: be a great idea if we had real id requirements 238 00:13:51,480 --> 00:13:54,040 Speaker 3: for internet services all the time in New Zealand? 239 00:13:54,200 --> 00:13:54,400 Speaker 1: You know? 240 00:13:54,440 --> 00:13:57,840 Speaker 3: But that is the effect of that policy, So we 241 00:13:57,960 --> 00:13:59,959 Speaker 3: kind of really need to be talking about the actual 242 00:14:00,360 --> 00:14:05,840 Speaker 3: practicalities of some of this too. One thing that I've 243 00:14:05,880 --> 00:14:08,640 Speaker 3: thought about quite a bit is if you sit down 244 00:14:09,040 --> 00:14:11,480 Speaker 3: and you look at the AI strategy and you look 245 00:14:11,520 --> 00:14:15,080 Speaker 3: at the light touch proportionate risk based approach that has 246 00:14:15,120 --> 00:14:17,960 Speaker 3: been advocated for. That is also the approach that the 247 00:14:18,000 --> 00:14:21,360 Speaker 3: Open Letter has advocated for, except maybe a little bit 248 00:14:21,440 --> 00:14:24,680 Speaker 3: less light touch, a slightly heavier touch. If you look 249 00:14:24,760 --> 00:14:29,080 Speaker 3: through the Responsible AI Guidance for Businesses that accompanied the 250 00:14:29,160 --> 00:14:32,960 Speaker 3: AI strategy, it's a really fantastic starting point for thinking 251 00:14:32,960 --> 00:14:35,680 Speaker 3: about these issues because what it does is it wraps 252 00:14:35,720 --> 00:14:38,640 Speaker 3: through a kind of table of at least ten different 253 00:14:38,680 --> 00:14:42,600 Speaker 3: statutes that already have some bearing on AI in the 254 00:14:42,640 --> 00:14:46,400 Speaker 3: way that it's used in New Zealand. So you know, 255 00:14:46,480 --> 00:14:49,280 Speaker 3: we need a kind of a better starting point for 256 00:14:49,320 --> 00:14:52,760 Speaker 3: a lot of these discussions. That's hard, Like it takes 257 00:14:52,800 --> 00:14:56,160 Speaker 3: a lot of reading and thinking and writing and stuff 258 00:14:56,200 --> 00:14:58,960 Speaker 3: like that, and it's hard to kind of get that 259 00:14:59,040 --> 00:15:03,520 Speaker 3: going if you can't find the information, the information problem. 260 00:15:03,760 --> 00:15:06,360 Speaker 3: If you can't coordinate and kind of work more effectively 261 00:15:06,400 --> 00:15:09,360 Speaker 3: with others, that's the coordination problem. And then today I've 262 00:15:09,400 --> 00:15:12,320 Speaker 3: just shared another one on what I'm calling the economic problem, 263 00:15:12,440 --> 00:15:15,560 Speaker 3: which is effectively it takes time and energy to do this. 264 00:15:16,120 --> 00:15:18,160 Speaker 3: It's pretty hard to do it for free. The only 265 00:15:18,200 --> 00:15:19,960 Speaker 3: way you can do it for free is if you 266 00:15:19,960 --> 00:15:22,840 Speaker 3: have some other economic interest in the discussion. And that's 267 00:15:22,840 --> 00:15:25,240 Speaker 3: not necessarily a bad thing, but it's something that we 268 00:15:25,280 --> 00:15:28,040 Speaker 3: need to think about in terms of how we progress 269 00:15:28,120 --> 00:15:29,000 Speaker 3: that discussion. 270 00:15:29,200 --> 00:15:30,800 Speaker 1: Yeah, I want to get into that in a minute, 271 00:15:30,800 --> 00:15:34,480 Speaker 1: because that is absolutely crucial to vested interests involved here 272 00:15:34,480 --> 00:15:37,120 Speaker 1: in a small country, who has the resource to put 273 00:15:37,160 --> 00:15:41,320 Speaker 1: into advocating a particular position. But just before we do, 274 00:15:43,040 --> 00:15:48,240 Speaker 1: one model that maybe has emerged for a rapidly emerging technology, 275 00:15:48,280 --> 00:15:51,360 Speaker 1: in this case, facial recognition was. You know, we've had 276 00:15:51,400 --> 00:15:54,240 Speaker 1: the biometric code that came out of the Office of 277 00:15:54,280 --> 00:15:59,120 Speaker 1: the Privacy Commissioner. It was quite a thorough process. There 278 00:15:59,160 --> 00:16:02,240 Speaker 1: was a trial running with food Stuff's North Island at 279 00:16:02,240 --> 00:16:05,480 Speaker 1: the same time, which actually gave some real life data 280 00:16:05,560 --> 00:16:08,360 Speaker 1: to look at about how this technology is used and 281 00:16:08,400 --> 00:16:10,400 Speaker 1: the pros and cons of it. Is that a potential 282 00:16:10,440 --> 00:16:12,360 Speaker 1: model for some of these issues that are going to 283 00:16:12,360 --> 00:16:15,080 Speaker 1: emerge out of AI, like deep fakes we keep talking about. 284 00:16:15,160 --> 00:16:18,560 Speaker 1: It hasn't really blown up yet, but it's quite likely 285 00:16:18,600 --> 00:16:21,800 Speaker 1: to as the technology gets better. It's potentially coming up 286 00:16:21,800 --> 00:16:26,080 Speaker 1: with these sort of you know, codes from trusted officers 287 00:16:26,200 --> 00:16:28,360 Speaker 1: like the Privacy Commissioner. Is that potentially a good way 288 00:16:28,400 --> 00:16:31,000 Speaker 1: to deal with some of these emerging issues with AI? 289 00:16:31,200 --> 00:16:35,040 Speaker 3: I think it is. Yeah. And one thing that I 290 00:16:35,240 --> 00:16:38,640 Speaker 3: have probably surprised myself with over the years in this 291 00:16:38,840 --> 00:16:42,880 Speaker 3: area is that I do think there's a really interesting 292 00:16:42,960 --> 00:16:47,600 Speaker 3: case for basically even self regulatory approaches. So self regulation 293 00:16:47,720 --> 00:16:50,880 Speaker 3: always sounds like this kind of wetbus ticket, you know, 294 00:16:51,040 --> 00:16:55,760 Speaker 3: flimsy cop out option, right, and it can be like that. 295 00:16:56,120 --> 00:16:58,480 Speaker 3: You know, there are some really glaring examples of that, 296 00:16:58,840 --> 00:17:00,760 Speaker 3: But what it can be Also, it is a really 297 00:17:00,760 --> 00:17:03,600 Speaker 3: good trial run for how something's actually going to work, 298 00:17:03,720 --> 00:17:06,480 Speaker 3: because how it's going to work is quite complicated. And 299 00:17:06,520 --> 00:17:08,320 Speaker 3: then what you can do is you can start to 300 00:17:08,440 --> 00:17:12,359 Speaker 3: wrap around that some of the enforcement mechanisms that begin 301 00:17:12,440 --> 00:17:15,440 Speaker 3: to kind of escalate the consequences of doing a good 302 00:17:15,520 --> 00:17:18,360 Speaker 3: job or a bad job of that kind of thing. 303 00:17:18,880 --> 00:17:21,359 Speaker 3: The other benefit of kind of taking that code based 304 00:17:21,359 --> 00:17:24,000 Speaker 3: approach is you can move really quick. You can move 305 00:17:24,160 --> 00:17:28,439 Speaker 3: way faster than any government agency can. And the thing is, 306 00:17:28,600 --> 00:17:31,360 Speaker 3: if you can basically demonstrate that you've got a code 307 00:17:31,600 --> 00:17:35,400 Speaker 3: that works and it's well thought through, it's much much 308 00:17:35,400 --> 00:17:37,720 Speaker 3: easier for an agency to just pick that up and 309 00:17:37,760 --> 00:17:40,400 Speaker 3: give it some teeth if it works well. So I'm 310 00:17:40,480 --> 00:17:43,560 Speaker 3: quite surprised by the extent which I am an advocate 311 00:17:43,680 --> 00:17:46,400 Speaker 3: for self regulation, because I wouldn't have been like that 312 00:17:46,560 --> 00:17:48,679 Speaker 3: in the past. I need to spend some time with 313 00:17:48,680 --> 00:17:51,120 Speaker 3: the biometric code. I haven't taken the time to read 314 00:17:51,160 --> 00:17:54,400 Speaker 3: it carefully. Part of that is because of the information 315 00:17:54,560 --> 00:17:57,440 Speaker 3: and economic problems that I describe in the series of 316 00:17:57,480 --> 00:18:00,399 Speaker 3: blog posts. I wanted to come back just so to 317 00:18:00,480 --> 00:18:02,960 Speaker 3: one thing that you said as well about sort of 318 00:18:02,960 --> 00:18:05,280 Speaker 3: deployment of chat bots. And then maybe this is also 319 00:18:05,320 --> 00:18:08,159 Speaker 3: relevant to the deep facs point as well. If you 320 00:18:08,200 --> 00:18:12,120 Speaker 3: think about it, there are areas where we already take 321 00:18:12,160 --> 00:18:14,359 Speaker 3: a kind of risk based approach and we put the 322 00:18:14,400 --> 00:18:18,200 Speaker 3: obligation on people who are doing stuff to think about 323 00:18:18,200 --> 00:18:21,080 Speaker 3: what they're doing before they do it and impose consequences 324 00:18:21,080 --> 00:18:23,080 Speaker 3: if they do a bad job. One example to me 325 00:18:23,160 --> 00:18:26,960 Speaker 3: would be our health and safety legislation. Right, so you 326 00:18:26,960 --> 00:18:29,679 Speaker 3: can imagine a situation. You know, let's say you're deploying 327 00:18:29,720 --> 00:18:33,280 Speaker 3: a chatbot for your employees or something like that, or 328 00:18:33,280 --> 00:18:38,040 Speaker 3: people coming on to your workspace. You're essentially already required 329 00:18:38,160 --> 00:18:41,119 Speaker 3: to think about the risks of deploying that, and you 330 00:18:41,200 --> 00:18:45,800 Speaker 3: can be liable for really significant financial consequences if you're 331 00:18:45,800 --> 00:18:49,560 Speaker 3: deploying that chatbot in that context and something bad happens. 332 00:18:49,880 --> 00:18:51,600 Speaker 3: So there's just one example off the top of my 333 00:18:51,680 --> 00:18:53,679 Speaker 3: head of the way that you know, we do have 334 00:18:53,760 --> 00:18:56,760 Speaker 3: these regulatory frameworks that kind of do apply in these 335 00:18:57,160 --> 00:19:00,800 Speaker 3: highly sensitive context but it doesn't have the same satisfying 336 00:19:00,880 --> 00:19:07,360 Speaker 3: impact of this generic nationwide risk based AI regulation that 337 00:19:07,440 --> 00:19:11,800 Speaker 3: kind of sounds way cooler than talking about health and safety. 338 00:19:12,200 --> 00:19:14,600 Speaker 3: I think deep fakes are another example of this, right, 339 00:19:14,680 --> 00:19:17,840 Speaker 3: Like one thing we concluded from our initial research was 340 00:19:18,400 --> 00:19:21,320 Speaker 3: there was a lot of really open textured legislation that 341 00:19:21,400 --> 00:19:26,119 Speaker 3: cover things quite easily like fraud through deep facts. And 342 00:19:26,160 --> 00:19:28,920 Speaker 3: that's because the definition of a document and the Crimes 343 00:19:28,960 --> 00:19:33,520 Speaker 3: Act clearly anticipates something like a video file, so actually 344 00:19:33,720 --> 00:19:36,720 Speaker 3: fraud through deep facts is already a criminal offense. The 345 00:19:36,760 --> 00:19:41,439 Speaker 3: other example of this is nonconsensual sexual imagery as well. 346 00:19:41,560 --> 00:19:44,120 Speaker 3: There's an argument that that is covered by the Harmful 347 00:19:44,119 --> 00:19:48,520 Speaker 3: Digital Communications Act and the Crimes Act. Unfortunately, Parliament hasn't 348 00:19:48,600 --> 00:19:52,320 Speaker 3: yet just bite us first identifying this issue in May 349 00:19:52,359 --> 00:19:55,959 Speaker 3: of twenty nineteen, taken steps to clarify any of that. 350 00:19:56,400 --> 00:20:00,200 Speaker 3: And that's the kind of political priorities issue, I think, and. 351 00:20:00,160 --> 00:20:04,000 Speaker 1: It's all it might take is that clarification and guidance 352 00:20:04,040 --> 00:20:07,240 Speaker 1: that this law actually does apply to the digital world. 353 00:20:07,320 --> 00:20:10,400 Speaker 1: You know, for instance, this will eventually happen we will 354 00:20:10,440 --> 00:20:14,119 Speaker 1: have an AI chatbot in New Zealand that discriminates against 355 00:20:14,200 --> 00:20:17,640 Speaker 1: someone racially or gender or something like that, and that 356 00:20:17,720 --> 00:20:21,320 Speaker 1: will potentially be a case before the Human Rights Act, 357 00:20:21,680 --> 00:20:25,280 Speaker 1: so that's applicable to that. We've got privacy obviously, if 358 00:20:25,280 --> 00:20:28,440 Speaker 1: you leak data through an AI chat bot or don't 359 00:20:28,440 --> 00:20:30,879 Speaker 1: disclose what you're doing with that data and don't store 360 00:20:30,920 --> 00:20:35,320 Speaker 1: it properly, you could be hit with our weak Privacy Act. 361 00:20:35,119 --> 00:20:36,200 Speaker 3: A nice sweetbus ticket. 362 00:20:36,280 --> 00:20:38,760 Speaker 1: Yeah yeah, yeah, exactly, So there is that. So yeah, 363 00:20:38,760 --> 00:20:40,800 Speaker 1: I take your point that you know there are lots 364 00:20:40,800 --> 00:20:43,160 Speaker 1: of layers of things going on. I think it goes 365 00:20:43,200 --> 00:20:46,679 Speaker 1: back to though, what you said earlier, like even in 366 00:20:46,720 --> 00:20:49,800 Speaker 1: the EU, you've got the scary AI Act, you know, 367 00:20:49,880 --> 00:20:53,440 Speaker 1: with big multimillion dollar fines for non compliance, but there's 368 00:20:53,480 --> 00:20:56,760 Speaker 1: this whole process that goes on behind it, and I 369 00:20:56,760 --> 00:20:58,680 Speaker 1: think that's the bit that we don't really have much 370 00:20:58,760 --> 00:21:02,639 Speaker 1: visibility into or or talk about here. You know, what 371 00:21:02,760 --> 00:21:05,439 Speaker 1: is actually going on there. You know, you take an 372 00:21:05,480 --> 00:21:10,240 Speaker 1: issue like copyright and arguably the bus has already gone 373 00:21:10,760 --> 00:21:14,240 Speaker 1: on that you know, everyone's copyrighted material has been scraped 374 00:21:14,280 --> 00:21:19,320 Speaker 1: from the Internet. We do have copyright legislation in New Zealand, 375 00:21:19,359 --> 00:21:23,000 Speaker 1: quite a strong copyright Act, so technically, if you wanted 376 00:21:23,040 --> 00:21:26,000 Speaker 1: to chase open AI under New Zealand copyright law, you 377 00:21:26,040 --> 00:21:27,760 Speaker 1: could take them to court and you might have a 378 00:21:27,760 --> 00:21:31,040 Speaker 1: good case for it. But there's very little discussion about this, 379 00:21:31,240 --> 00:21:33,919 Speaker 1: and goes to that point you made about regulation, is 380 00:21:33,960 --> 00:21:36,560 Speaker 1: one thing, but having those processes and even having a 381 00:21:36,600 --> 00:21:39,240 Speaker 1: government that is waiving the flag on these issues, and 382 00:21:39,280 --> 00:21:42,280 Speaker 1: we've been very muted. I think in New Zealand we've 383 00:21:42,280 --> 00:21:44,080 Speaker 1: just sort of said, look, the laws are there to 384 00:21:44,440 --> 00:21:46,720 Speaker 1: handle us, let's sort of see what happens. It's not 385 00:21:46,760 --> 00:21:50,000 Speaker 1: as though the government is really thumping the desk on 386 00:21:50,080 --> 00:21:53,159 Speaker 1: any particular issue around AI, just sort of saying in general, 387 00:21:53,200 --> 00:21:54,440 Speaker 1: you've got to do it responsibly. 388 00:21:54,520 --> 00:21:57,080 Speaker 3: I think that copyright one is a really good example, 389 00:21:57,240 --> 00:22:01,400 Speaker 3: because someone somewhere in New Zealand will have sat down 390 00:22:01,800 --> 00:22:04,920 Speaker 3: and looked very very carefully at the Copyright Act. I'd 391 00:22:04,960 --> 00:22:08,880 Speaker 3: be fascinated to know a lawyer somewhere may have even 392 00:22:08,960 --> 00:22:14,320 Speaker 3: been approached by our copyright holder and been tasked with thinking, 393 00:22:14,600 --> 00:22:17,560 Speaker 3: if we were to pursue litigation under New Zealand law 394 00:22:17,600 --> 00:22:20,200 Speaker 3: on this, what would the answer be and how it 395 00:22:20,200 --> 00:22:23,840 Speaker 3: would be enforced. I'd be stunned if somebody has not 396 00:22:24,000 --> 00:22:29,080 Speaker 3: thought about that. The information issue is that we can 397 00:22:29,119 --> 00:22:32,399 Speaker 3: go and replicate that or we can basically share the 398 00:22:32,400 --> 00:22:35,320 Speaker 3: results of that analysis, and that would be really valuable. 399 00:22:35,359 --> 00:22:37,280 Speaker 3: It would basically mean that we don't have to duplicate 400 00:22:37,320 --> 00:22:39,800 Speaker 3: that over and over again. And you can almost guarantee 401 00:22:39,800 --> 00:22:42,440 Speaker 3: that the person who has thought to themselves, I really 402 00:22:42,440 --> 00:22:46,480 Speaker 3: want to sue anthropic or open eye about this, has 403 00:22:46,520 --> 00:22:51,800 Speaker 3: not gone to their local family law solicitor, like ideally, 404 00:22:51,840 --> 00:22:53,520 Speaker 3: they've probably gone to one of the big law firms 405 00:22:53,520 --> 00:22:55,199 Speaker 3: who really know what they're doing. And they may have 406 00:22:55,240 --> 00:22:58,920 Speaker 3: even gone out too some QC who's the you know, sorry, Casey, 407 00:22:59,240 --> 00:23:02,680 Speaker 3: who's the copyright expert in New Zealand. I think this 408 00:23:02,760 --> 00:23:05,440 Speaker 3: is part of that kind of systematic approach, which I've 409 00:23:05,480 --> 00:23:09,080 Speaker 3: learned over time can be really frustrating and disheartening because 410 00:23:09,119 --> 00:23:11,760 Speaker 3: you might take off in one direction and kind of 411 00:23:11,800 --> 00:23:14,760 Speaker 3: do this amazing analysis, and then you kind of turn 412 00:23:14,800 --> 00:23:17,639 Speaker 3: a corner and like somebody's already done it way faster 413 00:23:18,119 --> 00:23:21,240 Speaker 3: and come to a great answer, and you could have 414 00:23:21,320 --> 00:23:24,119 Speaker 3: just saved all that time asking them what they think. 415 00:23:24,320 --> 00:23:27,800 Speaker 3: So that's why I'm kind of thinking about this information coordination, 416 00:23:28,400 --> 00:23:31,239 Speaker 3: the kind of economic things that drive that, and then 417 00:23:31,320 --> 00:23:34,120 Speaker 3: downstream to some of the kind of policy issues that 418 00:23:34,280 --> 00:23:36,760 Speaker 3: give us the kind of focus to organize around as well. 419 00:23:44,080 --> 00:23:45,600 Speaker 2: Let's talk about that economics. 420 00:23:45,600 --> 00:23:48,639 Speaker 1: The I think the third in your series is very 421 00:23:48,720 --> 00:23:52,359 Speaker 1: much about that, how that drives this discussion and approach 422 00:23:52,440 --> 00:23:56,000 Speaker 1: to regulation. What I've found over the years anything tech related, 423 00:23:56,040 --> 00:23:58,600 Speaker 1: particularly around like social media. You get the big tech 424 00:23:58,640 --> 00:24:01,920 Speaker 1: giants who basically do a cookie cutter version off their 425 00:24:02,000 --> 00:24:05,320 Speaker 1: overseas policies that you know that's put a lot of 426 00:24:05,359 --> 00:24:08,880 Speaker 1: time and money into crafting. They localize that, they serve 427 00:24:08,960 --> 00:24:11,520 Speaker 1: that up to our government, and then they sort of 428 00:24:11,720 --> 00:24:14,040 Speaker 1: cozy up with the net safes and those sorts of 429 00:24:14,080 --> 00:24:17,000 Speaker 1: organizations to soft power as such. 430 00:24:17,440 --> 00:24:20,000 Speaker 2: Then you get sort of the rest of New Zealand Inc. 431 00:24:20,080 --> 00:24:23,080 Speaker 1: Which you know, only if if it really affects their business, 432 00:24:23,200 --> 00:24:26,000 Speaker 1: like for instance, or you know, is a facial recognition 433 00:24:26,119 --> 00:24:29,840 Speaker 1: company that does retail security systems, so they're very interested 434 00:24:29,880 --> 00:24:33,280 Speaker 1: in the biometrics code. So you'll get particular businesses that 435 00:24:33,359 --> 00:24:36,399 Speaker 1: are interested in it. You get the academics because some 436 00:24:36,440 --> 00:24:38,760 Speaker 1: of them this is their area of research and they 437 00:24:38,760 --> 00:24:41,399 Speaker 1: want to write research papers on it. That's great, But 438 00:24:41,520 --> 00:24:44,080 Speaker 1: the ones that are really left out is sort of 439 00:24:44,080 --> 00:24:48,480 Speaker 1: civil society and those groups that are really poorly resourced. 440 00:24:48,720 --> 00:24:51,080 Speaker 1: I'm thinking, you know, just as we speak, you know, 441 00:24:51,080 --> 00:24:53,240 Speaker 1: one of the big ones in New Zealand, IT professionals 442 00:24:53,240 --> 00:24:56,840 Speaker 1: has just gone bust, going into liquidation. Organization it started 443 00:24:56,840 --> 00:25:00,199 Speaker 1: in the nineteen fifties. It's those sorts of organizations that 444 00:25:00,240 --> 00:25:04,120 Speaker 1: in the past maybe had really good input into policy 445 00:25:04,160 --> 00:25:07,320 Speaker 1: development in New Zealand, but they're just so cash strapped. 446 00:25:07,880 --> 00:25:11,199 Speaker 1: So my concern is that we're going to miss increasingly 447 00:25:11,240 --> 00:25:15,840 Speaker 1: in this discussion, really important dialogue with the public and 448 00:25:16,480 --> 00:25:19,639 Speaker 1: researchers who work on behalf of the public. Are you 449 00:25:19,680 --> 00:25:20,359 Speaker 1: concerned about that? 450 00:25:20,560 --> 00:25:24,479 Speaker 3: Yeah, And I think you've really well summarized that. And 451 00:25:24,680 --> 00:25:28,000 Speaker 3: essentially I've thought about this a lot since establishing brain 452 00:25:28,080 --> 00:25:31,160 Speaker 3: Box in twenty eighteen. It was also quite an eye 453 00:25:31,160 --> 00:25:35,600 Speaker 3: opener to me working in international tech policy because obviously, 454 00:25:37,080 --> 00:25:39,920 Speaker 3: you know, you have the kind of international think tank space, 455 00:25:40,080 --> 00:25:42,880 Speaker 3: which will be One of the members of our steering 456 00:25:42,880 --> 00:25:47,120 Speaker 3: group was from the Carnegie Endowment for International Peace right 457 00:25:47,200 --> 00:25:50,400 Speaker 3: so founded by Andrew Carnegie, the kind of Elon Musk 458 00:25:50,480 --> 00:25:53,320 Speaker 3: of his time, with just an enormous endowment to just 459 00:25:53,359 --> 00:25:56,960 Speaker 3: focus on good ideas and articulating good ideas. It was 460 00:25:57,040 --> 00:25:59,880 Speaker 3: really amazing to kind of walk down think tank al 461 00:26:00,200 --> 00:26:03,000 Speaker 3: in Washington, DC and go to the Brookings Institution and 462 00:26:03,040 --> 00:26:04,760 Speaker 3: stuff like that. So I've thought about this a lot, 463 00:26:04,800 --> 00:26:06,760 Speaker 3: and I'm a real geek for all of us. As 464 00:26:06,880 --> 00:26:09,280 Speaker 3: you can tell, it is critical to be able to 465 00:26:09,320 --> 00:26:14,480 Speaker 3: have a kind of non industry, non government, and i'd 466 00:26:14,520 --> 00:26:18,320 Speaker 3: say non academic voice on all of this too, because 467 00:26:18,720 --> 00:26:22,320 Speaker 3: academia is fantastic but can also operate like a very 468 00:26:22,400 --> 00:26:25,520 Speaker 3: large institution and as you say, is kind of maybe 469 00:26:25,560 --> 00:26:30,200 Speaker 3: oriented towards particular ways of thinking about things as well. 470 00:26:30,359 --> 00:26:33,080 Speaker 3: The space that I've tried to really fill with brain 471 00:26:33,160 --> 00:26:37,320 Speaker 3: box is to have that really action oriented capability to 472 00:26:37,440 --> 00:26:41,679 Speaker 3: deal with all of those groups. So one thing that 473 00:26:42,520 --> 00:26:45,280 Speaker 3: people might find teds going through the blog post is 474 00:26:45,320 --> 00:26:48,320 Speaker 3: I'm trying to call some things out really directly, but 475 00:26:48,359 --> 00:26:50,800 Speaker 3: then also say like, look, I get it. You know 476 00:26:50,880 --> 00:26:54,360 Speaker 3: this is not happening because like government are bad guys 477 00:26:54,359 --> 00:26:57,439 Speaker 3: that don't want to blah blah. You know, there is 478 00:26:57,480 --> 00:27:01,280 Speaker 3: a place for industry at the table be advocating and 479 00:27:01,320 --> 00:27:05,960 Speaker 3: sharing good ideas that make for better policy, but also 480 00:27:06,040 --> 00:27:08,199 Speaker 3: you know, have an impact on their bottom line. 481 00:27:08,280 --> 00:27:08,480 Speaker 2: You know. 482 00:27:08,680 --> 00:27:12,600 Speaker 3: The tricky thing here is how do you balance all 483 00:27:12,640 --> 00:27:15,680 Speaker 3: of those different sector groups. It's a really interesting space 484 00:27:15,720 --> 00:27:19,200 Speaker 3: in technology too, because this is how Internet governance has 485 00:27:19,359 --> 00:27:22,480 Speaker 3: always worked. When you govern the Internet, it's never been 486 00:27:22,520 --> 00:27:25,840 Speaker 3: about just governments doing it by themselves. It's always been 487 00:27:25,880 --> 00:27:30,000 Speaker 3: about a blend of industry, academia and research and then 488 00:27:30,040 --> 00:27:32,119 Speaker 3: the people using the internet, you know, the kind of 489 00:27:32,119 --> 00:27:35,040 Speaker 3: civil society. So I think there is a really cool 490 00:27:35,080 --> 00:27:39,320 Speaker 3: opportunity to try and solve this and do it really well. 491 00:27:39,800 --> 00:27:41,960 Speaker 3: That is what I'm trying to do at brain Box 492 00:27:42,040 --> 00:27:45,800 Speaker 3: with my co director, doctor Allan Strickland, who was at 493 00:27:45,960 --> 00:27:48,919 Speaker 3: Internet New Zealand for a long time and is currently leading, 494 00:27:49,000 --> 00:27:54,840 Speaker 3: for example, a project on Internet resilience and a world 495 00:27:54,840 --> 00:28:01,159 Speaker 3: of climate change impacts. So really cool epic Project's just 496 00:28:01,160 --> 00:28:04,160 Speaker 3: do an incredible job of pulling together so many interesting 497 00:28:04,240 --> 00:28:07,199 Speaker 3: people with perspectives to share on this from lots and 498 00:28:07,240 --> 00:28:10,000 Speaker 3: lots of different groups. We've had some funding from the 499 00:28:10,040 --> 00:28:14,320 Speaker 3: Internet Society Foundation internationally to do that. So it's a 500 00:28:14,359 --> 00:28:16,600 Speaker 3: really cool example of the way that if you can 501 00:28:16,640 --> 00:28:20,320 Speaker 3: crack the economic side of things and the institutional side 502 00:28:20,320 --> 00:28:23,840 Speaker 3: of things, you can get this great kind of collaboration 503 00:28:24,040 --> 00:28:27,840 Speaker 3: infrastructure going that deals with this kind of information problem 504 00:28:28,040 --> 00:28:31,880 Speaker 3: of synthesizing and grabbing and bringing everything together, can kind 505 00:28:31,880 --> 00:28:34,120 Speaker 3: of coordinate in a way that isn't driven by any 506 00:28:34,119 --> 00:28:38,200 Speaker 3: particular interest, and then you know, create that capability and 507 00:28:38,280 --> 00:28:39,920 Speaker 3: keep the discussion going in the way that we need 508 00:28:39,920 --> 00:28:40,120 Speaker 3: it to. 509 00:28:40,360 --> 00:28:44,120 Speaker 1: Yeah, final question Tom, which I ask you everyone that 510 00:28:44,200 --> 00:28:49,760 Speaker 1: I'm interviewing about AI and I get different responses. I'm 511 00:28:49,800 --> 00:28:52,600 Speaker 1: intrigued as to what your thoughts are. You know, should 512 00:28:52,640 --> 00:28:55,760 Speaker 1: we have a real angle on AI as a nation 513 00:28:56,000 --> 00:28:57,960 Speaker 1: and what should it be for New Zealand. You know, 514 00:28:58,040 --> 00:29:00,480 Speaker 1: the US, as have said at the start, they want 515 00:29:00,520 --> 00:29:05,080 Speaker 1: to have supremacy in Ai because it is geopolitically significant. 516 00:29:05,160 --> 00:29:08,480 Speaker 1: It's a national security issue. That's why Trump is in 517 00:29:08,480 --> 00:29:11,440 Speaker 1: the White House every second week with Oracle and Open Ai. 518 00:29:11,560 --> 00:29:14,920 Speaker 1: Five hundred billion investment here, six hundred billion there. For them, 519 00:29:14,960 --> 00:29:18,240 Speaker 1: it's about brute force. How much can we spend to 520 00:29:18,400 --> 00:29:21,880 Speaker 1: accelerate our lead. China has a version of that going on. 521 00:29:22,560 --> 00:29:26,520 Speaker 1: Europe has sort of pivoted a little bit between regulation 522 00:29:26,760 --> 00:29:30,240 Speaker 1: first too. We don't want to miss the AI revolution 523 00:29:30,400 --> 00:29:33,560 Speaker 1: by strangling our companies with red tape, so we need 524 00:29:33,600 --> 00:29:36,440 Speaker 1: to accommodate that as well. You're closer to it than 525 00:29:36,480 --> 00:29:39,320 Speaker 1: me talking to lots of stakeholders in this. Is there 526 00:29:39,800 --> 00:29:42,440 Speaker 1: an angle for us emerging when it comes to what 527 00:29:42,640 --> 00:29:46,080 Speaker 1: our edge could be competitively and you know in terms 528 00:29:46,120 --> 00:29:49,120 Speaker 1: of trade and that but also for the rest of 529 00:29:49,120 --> 00:29:51,880 Speaker 1: the world to look to us, is there anything emerging 530 00:29:51,920 --> 00:29:53,760 Speaker 1: where we have potential leadership. 531 00:29:53,840 --> 00:29:58,360 Speaker 3: What I do think we are missing is a single 532 00:29:58,440 --> 00:30:02,640 Speaker 3: coordinating vision for what New Zealand needs from AI. I 533 00:30:02,680 --> 00:30:05,880 Speaker 3: think there is recognition that it's very, very important, and 534 00:30:05,960 --> 00:30:09,000 Speaker 3: I think in fact it is important. Probably if you'd 535 00:30:09,040 --> 00:30:12,360 Speaker 3: ask me, maybe even a year ago, I would have 536 00:30:12,400 --> 00:30:14,960 Speaker 3: been on the fence about that, but I'm not now. 537 00:30:15,080 --> 00:30:18,720 Speaker 3: I think it's critically important. It's not just the AI 538 00:30:18,880 --> 00:30:22,680 Speaker 3: side of things. It's also that broader kind of digital sovereignty, 539 00:30:22,800 --> 00:30:27,320 Speaker 3: digital infrastructure, kind of internet infrastructure, and then also the 540 00:30:27,400 --> 00:30:31,200 Speaker 3: kind of literacy and capability aspects of that too. It's 541 00:30:31,200 --> 00:30:34,120 Speaker 3: all very well and good having really powerful computers and 542 00:30:34,200 --> 00:30:37,520 Speaker 3: fantastic models, but if everyone's using them for the wrong thing, 543 00:30:37,840 --> 00:30:41,719 Speaker 3: we're not going forward in any meaningful way. So we 544 00:30:41,760 --> 00:30:45,200 Speaker 3: do need a kind of coherent vision. What I've explored 545 00:30:45,480 --> 00:30:48,080 Speaker 3: in the past, I probably would have advocated for what's 546 00:30:48,120 --> 00:30:51,400 Speaker 3: called a human rights based approach, and that's really useful 547 00:30:51,440 --> 00:30:53,480 Speaker 3: because it gives you a starting point and a set 548 00:30:53,520 --> 00:30:56,800 Speaker 3: of principles and values that are pretty well tested right 549 00:30:56,960 --> 00:31:01,320 Speaker 3: like freedom of expression, right to privacy, to public participation 550 00:31:01,600 --> 00:31:04,760 Speaker 3: all of those kinds of things. That kind of framing 551 00:31:04,800 --> 00:31:09,280 Speaker 3: has obviously massively fallen out of favor internationally, which you 552 00:31:09,280 --> 00:31:13,080 Speaker 3: know is not great, to be honest, but it is 553 00:31:13,080 --> 00:31:15,520 Speaker 3: what it is. We've got to be practical. I was 554 00:31:16,040 --> 00:31:18,880 Speaker 3: talking about this at a seminar a little while ago 555 00:31:19,120 --> 00:31:22,680 Speaker 3: where somebody raised for me this concept of sovereign AI, 556 00:31:22,960 --> 00:31:26,000 Speaker 3: and my initial reaction in the moment was, what that's weird, 557 00:31:26,000 --> 00:31:28,800 Speaker 3: what are you even talking about? But it's kind of 558 00:31:28,840 --> 00:31:31,360 Speaker 3: wormed its way into my head quite a bit, and 559 00:31:31,440 --> 00:31:34,600 Speaker 3: I've been thinking about what it might actually mean for 560 00:31:34,680 --> 00:31:37,760 Speaker 3: New Zealand, and the next post I'm going to kind 561 00:31:37,760 --> 00:31:41,520 Speaker 3: of frame up why I think sovereign AI is a 562 00:31:41,560 --> 00:31:45,040 Speaker 3: really interesting direction for us to be thinking about. And 563 00:31:45,080 --> 00:31:47,240 Speaker 3: I've got a longer discussion paper that I've shared in 564 00:31:47,240 --> 00:31:49,479 Speaker 3: the past that kind of breaks it down and just 565 00:31:49,520 --> 00:31:53,800 Speaker 3: sort of says, this isn't about having like NZGPT that's 566 00:31:53,960 --> 00:31:57,000 Speaker 3: trained on like all of the data in New Zealand 567 00:31:57,280 --> 00:32:00,520 Speaker 3: and speaks with a Kiwi accent and all that kind 568 00:32:00,560 --> 00:32:03,240 Speaker 3: of thing. It can be as simple as talking about 569 00:32:03,560 --> 00:32:06,200 Speaker 3: meaningful AI literacy, or it could be as simple as 570 00:32:06,200 --> 00:32:09,480 Speaker 3: making sure that we do have resilient digital infrastructure for 571 00:32:09,640 --> 00:32:13,800 Speaker 3: access and deployment of AI systems, and an all likelihood 572 00:32:13,840 --> 00:32:18,400 Speaker 3: it probably means fine tuning models. Yes, that already exist, 573 00:32:18,640 --> 00:32:20,720 Speaker 3: and I saw, really, you know, this is kind of 574 00:32:20,760 --> 00:32:25,200 Speaker 3: free plug for straker Ai, a pretty amazing New Zealand 575 00:32:25,240 --> 00:32:29,400 Speaker 3: company who I think are now proposing to offer fine 576 00:32:29,440 --> 00:32:33,160 Speaker 3: tuned models as a service. Pretty interesting for your listeners. 577 00:32:33,480 --> 00:32:35,719 Speaker 3: I'm sure. I'm really interested in this kind of sovereign 578 00:32:35,720 --> 00:32:38,080 Speaker 3: AI thing and kind of fleshing out what that means 579 00:32:38,240 --> 00:32:40,800 Speaker 3: in terms of the competitive advantage for us. I know 580 00:32:40,880 --> 00:32:42,680 Speaker 3: that people are kind of thinking about the fact that 581 00:32:42,720 --> 00:32:48,680 Speaker 3: we're like a stable, remote, English speaking westernized nation. I 582 00:32:48,680 --> 00:32:50,800 Speaker 3: know that other people are also thinking about the fact 583 00:32:50,840 --> 00:32:53,719 Speaker 3: that we have a lot of renewable energy generation, so 584 00:32:53,760 --> 00:32:57,760 Speaker 3: when it comes to basically housing data centers, we have 585 00:32:57,840 --> 00:33:00,680 Speaker 3: a lot of capability there as well. I hear that 586 00:33:00,760 --> 00:33:03,680 Speaker 3: all the time. I actually do agree with what they 587 00:33:03,800 --> 00:33:07,480 Speaker 3: said in the AI strategy, which is a pretty unpopular 588 00:33:07,680 --> 00:33:11,000 Speaker 3: position to hold to say anything nice about the AI strategy, 589 00:33:11,280 --> 00:33:12,880 Speaker 3: But what they did say was, you know, we can 590 00:33:12,920 --> 00:33:16,440 Speaker 3: be experts and basically deploying AI systems. So we've got 591 00:33:16,440 --> 00:33:20,040 Speaker 3: these incredible AI systems, very powerful. But something I've said 592 00:33:20,040 --> 00:33:22,400 Speaker 3: in the past is, you know, they're only as good 593 00:33:22,440 --> 00:33:26,200 Speaker 3: as your systems for making sure that what they're doing 594 00:33:26,280 --> 00:33:30,000 Speaker 3: is good and reliable and not a hallucination. And you 595 00:33:30,000 --> 00:33:31,680 Speaker 3: know it takes into account all of the data that 596 00:33:31,680 --> 00:33:33,920 Speaker 3: it needs to and that all the code that it's 597 00:33:33,960 --> 00:33:38,400 Speaker 3: produced can actually be tested for security purposes and works. 598 00:33:38,920 --> 00:33:42,520 Speaker 3: So I actually do think this competitive advantage around being 599 00:33:42,920 --> 00:33:46,960 Speaker 3: you know, the world's smartest deployers of AI systems is 600 00:33:47,040 --> 00:33:49,840 Speaker 3: something that I'm obviously pretty interested in and I'd like 601 00:33:49,880 --> 00:33:50,800 Speaker 3: to see that taken further. 602 00:33:50,880 --> 00:33:52,920 Speaker 2: Yea, and we do have a track record at that. 603 00:33:53,080 --> 00:33:55,920 Speaker 1: We didn't invent cloud computing, but we've got some great 604 00:33:55,960 --> 00:33:58,640 Speaker 1: software as a service companies that use the cloud to 605 00:33:58,960 --> 00:34:03,160 Speaker 1: improve accounting or billing software, whatever it was. So we 606 00:34:03,280 --> 00:34:07,560 Speaker 1: are adept at taking those existing technologies that billions of 607 00:34:07,600 --> 00:34:10,080 Speaker 1: dollars have gone into and actually making them really useful. 608 00:34:10,120 --> 00:34:11,560 Speaker 3: And I guess one of the other things I've been 609 00:34:11,560 --> 00:34:14,160 Speaker 3: thinking about a lot is and you'll know more about 610 00:34:14,200 --> 00:34:18,120 Speaker 3: this than me. I keep reading about the Knowledge Wave conference. 611 00:34:18,600 --> 00:34:20,719 Speaker 3: Yes back in sort of two thousand and one. I 612 00:34:20,719 --> 00:34:23,080 Speaker 3: think of us, Yeah, and I asked myself, you know 613 00:34:23,160 --> 00:34:25,200 Speaker 3: that was probably the last time that we went We're 614 00:34:25,200 --> 00:34:28,840 Speaker 3: going to be a genius tech economy. And one of 615 00:34:28,880 --> 00:34:30,600 Speaker 3: the questions I'd love to see sort of asked an 616 00:34:30,640 --> 00:34:33,200 Speaker 3: answered is like, well, what happened with that? Yeah, you know, 617 00:34:33,480 --> 00:34:35,680 Speaker 3: did it work? Is there where we are now? Or 618 00:34:36,000 --> 00:34:38,399 Speaker 3: or you know, did it fail? Or And I think, 619 00:34:38,640 --> 00:34:42,160 Speaker 3: to be honest, speaking as somebody who reads a lot 620 00:34:42,480 --> 00:34:44,600 Speaker 3: and does a lot of sort of research, I think 621 00:34:44,640 --> 00:34:46,600 Speaker 3: a lot of the most interesting answers to that are 622 00:34:46,600 --> 00:34:49,160 Speaker 3: actually going to be by talking to people who are there, 623 00:34:49,239 --> 00:34:51,200 Speaker 3: and they'll be able to say, you know, oh, well 624 00:34:51,200 --> 00:34:53,239 Speaker 3: this person is leading the work and they kind of 625 00:34:53,280 --> 00:34:55,840 Speaker 3: moved on, and then you know, it'll be insights like 626 00:34:55,880 --> 00:34:58,719 Speaker 3: that that are actually are really impactful. So I'd love 627 00:34:58,719 --> 00:35:00,400 Speaker 3: to see a bit of a kind of retresce aspective 628 00:35:00,880 --> 00:35:01,680 Speaker 3: on that kind of thing. 629 00:35:01,840 --> 00:35:04,600 Speaker 1: Well, we'll leave that for another episode a deconstruction or 630 00:35:04,600 --> 00:35:07,200 Speaker 1: what happened or didn't happen to the knowledge wave. But 631 00:35:07,719 --> 00:35:09,839 Speaker 1: in the meantime, hey, great work. Keep it up with 632 00:35:09,960 --> 00:35:14,279 Speaker 1: brain Box. Really important part of the tech policy landscape. 633 00:35:14,320 --> 00:35:17,440 Speaker 1: So thanks so much for coming on, Tom, and we'll 634 00:35:17,480 --> 00:35:19,520 Speaker 1: post obviously links to all those blog posts. 635 00:35:19,640 --> 00:35:22,680 Speaker 2: Really interesting series. Thanks very much, for sharing them with us. 636 00:35:22,680 --> 00:35:23,160 Speaker 2: Thanks for that. 637 00:35:25,880 --> 00:35:28,240 Speaker 1: That's it for this episode of the Business of Tech. 638 00:35:28,719 --> 00:35:32,000 Speaker 1: Thanks so much to Tom Barrowcloth, who was really challenging 639 00:35:32,200 --> 00:35:40,359 Speaker 1: the prevailing wisdom about AI regulation among tech experts and academics. 640 00:35:41,080 --> 00:35:44,680 Speaker 1: He's cautioning against quick fixes and highlighting the wealth of 641 00:35:44,719 --> 00:35:49,600 Speaker 1: existing legislation that already covers some of these AI related risks. 642 00:35:49,840 --> 00:35:53,640 Speaker 1: Whether we're using that regulation to head them off quickly 643 00:35:53,719 --> 00:35:57,719 Speaker 1: as another story. It's an argument well made, but as 644 00:35:57,760 --> 00:36:00,120 Speaker 1: we saw with the rise of social media, exist in 645 00:36:00,200 --> 00:36:03,040 Speaker 1: laws didn't really deal well with the harm that those 646 00:36:03,080 --> 00:36:08,280 Speaker 1: platforms caused. The introduction of the Harmful Digital Communications Act 647 00:36:08,400 --> 00:36:12,839 Speaker 1: did give people some redress against cyber bullying and harassment, 648 00:36:13,160 --> 00:36:16,279 Speaker 1: but really did it change the behavior of meta x 649 00:36:16,400 --> 00:36:20,200 Speaker 1: and TikTok. None of them have ever faced criminal penalties 650 00:36:20,280 --> 00:36:23,160 Speaker 1: or fines in New Zealand for live streaming the christ 651 00:36:23,280 --> 00:36:26,240 Speaker 1: Church terror attacks, for instance, or anything else for that matter. 652 00:36:26,760 --> 00:36:31,560 Speaker 1: That's exactly why parents in New Zealand are overwhelmingly supportive 653 00:36:31,800 --> 00:36:35,000 Speaker 1: off of social media ban here. They don't trust these 654 00:36:35,040 --> 00:36:37,879 Speaker 1: platforms in large part because they've been given free rein 655 00:36:38,280 --> 00:36:44,200 Speaker 1: They've sort of been incentivized with a soft regulatory regime. Still, 656 00:36:44,320 --> 00:36:47,720 Speaker 1: Tom rightly points out how our real challenge is actually 657 00:36:48,200 --> 00:36:52,319 Speaker 1: information coordination, not necessarily a lack of laws, but how 658 00:36:52,360 --> 00:36:54,480 Speaker 1: we use them, and he stressed the importance of building 659 00:36:54,520 --> 00:36:58,840 Speaker 1: civil society's capability to hold balanced, thoughtful discussions even in 660 00:36:58,880 --> 00:37:03,040 Speaker 1: the face of populist calls for sweeping action. Totally on 661 00:37:03,080 --> 00:37:06,279 Speaker 1: the same page with them about that New Zealand needs 662 00:37:06,280 --> 00:37:10,440 Speaker 1: to shape its own national approach focused on digital sovereignty, 663 00:37:10,719 --> 00:37:14,640 Speaker 1: infrastructure resilience, and AI literacy. If we get this right, 664 00:37:14,760 --> 00:37:18,040 Speaker 1: our edge could be not just in developing technology, but 665 00:37:18,080 --> 00:37:22,040 Speaker 1: in deploying it really smartly, responsibly and with real impact. 666 00:37:22,120 --> 00:37:24,319 Speaker 1: So thanks for listening to the Business of Tech, which 667 00:37:24,360 --> 00:37:27,600 Speaker 1: is streaming on iHeartRadio, Spotify, and Apple, where you'll find 668 00:37:27,640 --> 00:37:29,719 Speaker 1: all the show notes for this episode thanks to our 669 00:37:29,760 --> 00:37:32,239 Speaker 1: sponsored two degrees, and I'll catch you next week for 670 00:37:32,280 --> 00:37:34,320 Speaker 1: another episode of the Business of Tech.