1 00:00:01,287 --> 00:00:03,038 Azeem, welcome to the podcast. 2 00:00:03,250 --> 00:00:04,513 Kevin, it's great to see you again. 3 00:00:04,513 --> 00:00:07,881 I think it's getting into our 25th year of knowing each other. 4 00:00:08,011 --> 00:00:11,314 This is very frightening and you've done a lot of different things, worked for 5 00:00:11,314 --> 00:00:13,656 various companies, started various companies. 6 00:00:13,937 --> 00:00:16,198 How do you explain to your parents what you do now? 7 00:00:16,618 --> 00:00:20,839 Yeah, well, I, they're quite excited because, or my mum is now, my dad passed 8 00:00:20,839 --> 00:00:25,100 away last year, but he was even excited because I spent my time researching ideas 9 00:00:25,280 --> 00:00:27,260 on my own with my, well, with my team. 10 00:00:27,260 --> 00:00:31,522 And I have a couple of academic affiliations, research them, thinking 11 00:00:31,522 --> 00:00:37,103 about the intersection between this rapid technology change, things like artificial 12 00:00:37,103 --> 00:00:42,265 intelligence and societal change from the economy to our legal institutions and 13 00:00:42,265 --> 00:00:44,265 systems of government governance. 14 00:00:44,345 --> 00:00:45,398 And then I bundled that up. 15 00:00:45,398 --> 00:00:50,646 produce a newsletter from it, write the odd book, help the odd investor or 16 00:00:50,646 --> 00:00:52,830 company, maybe with a speech here or there. 17 00:00:52,830 --> 00:00:59,201 So I'm, I guess, a one-man band, but I do have an amazing team of researchers who 18 00:00:59,201 --> 00:01:00,181 work with me. 19 00:01:00,831 --> 00:01:05,154 Yeah, it's one of those jobs that probably didn't exist before, but now you can be a 20 00:01:05,154 --> 00:01:08,236 real force multiplier, which is wonderful. 21 00:01:08,236 --> 00:01:13,701 So let me ask you, when did AI first come on your radar screen as a technology you 22 00:01:13,701 --> 00:01:14,281 should focus on? 23 00:01:14,281 --> 00:01:18,505 Obviously, AI has been around for a very long time, but at what point did you 24 00:01:18,505 --> 00:01:21,126 notice that there was something really important going on? 25 00:01:21,998 --> 00:01:28,620 You know, I actually, I mean, my first AI deal was done in 2000 to 2001. 26 00:01:29,761 --> 00:01:33,943 So yes, it's been around for a long time, but then there was this additional shift 27 00:01:33,943 --> 00:01:35,243 because of deep learning. 28 00:01:35,644 --> 00:01:43,987 I first came across it in about 2012 or 2013 when someone in my team, my head of 29 00:01:43,987 --> 00:01:49,329 machine learning, wants to take a job at some funky London startup called 30 00:01:49,378 --> 00:01:50,138 Deep mind. 31 00:01:50,138 --> 00:01:53,762 I don't know if you've heard of them and I persuaded him not to do that. 32 00:01:54,323 --> 00:01:59,669 And, and then when we sold my company, I had some time to think, and then I, I dug 33 00:01:59,669 --> 00:02:03,032 into things and I saw something was happening and that was about 2014. 34 00:02:03,073 --> 00:02:04,353 So about a decade ago. 35 00:02:05,663 --> 00:02:11,098 What's changed in terms of either fundamentally AI or how we think about AI 36 00:02:11,098 --> 00:02:11,838 today? 37 00:02:12,886 --> 00:02:17,027 you know, the technology has really, really matured. 38 00:02:17,027 --> 00:02:23,269 And what happens in the process of maturation is things that become so hard, 39 00:02:23,269 --> 00:02:28,490 that only the very, very best researchers in the best research institutions can make 40 00:02:28,490 --> 00:02:29,570 sense of them. 41 00:02:29,570 --> 00:02:34,412 And then over the course of four or five years, they moved to being available as, 42 00:02:34,412 --> 00:02:39,613 you know, an executive ed class for a couple of thousand dollars at a business 43 00:02:39,613 --> 00:02:40,093 school, right? 44 00:02:40,093 --> 00:02:41,174 And that is... 45 00:02:41,174 --> 00:02:44,135 That is how a technology deploys into the economy. 46 00:02:44,135 --> 00:02:50,439 So a lot of those things have happened with deep learning based technologies and 47 00:02:50,439 --> 00:02:56,142 traditional machine learning systems like classification and discrimination and so 48 00:02:56,142 --> 00:02:56,742 on. 49 00:02:57,182 --> 00:03:01,745 But the other thing that has happened of course is that the waves of discovery have 50 00:03:01,745 --> 00:03:02,885 continued. 51 00:03:03,546 --> 00:03:09,609 Alongside that, the underlying infrastructure, the enabling technologies 52 00:03:09,609 --> 00:03:10,689 have become 53 00:03:11,098 --> 00:03:13,299 much, much more capable. 54 00:03:13,479 --> 00:03:18,862 And that means that for a business, they need to have less expertise than perhaps 55 00:03:18,862 --> 00:03:21,544 they did a decade ago to get into these technologies. 56 00:03:21,544 --> 00:03:26,727 And they're supported of course, by the systems integrators, the consultants who 57 00:03:26,727 --> 00:03:30,869 are the ones who really drive enterprise IT spending. 58 00:03:31,049 --> 00:03:36,972 And so what we see, which I think is pretty fascinating is not just, you know, 59 00:03:36,972 --> 00:03:38,273 maturation, 60 00:03:39,486 --> 00:03:43,588 of the technology that makes it more accessible to people, but also this 61 00:03:43,588 --> 00:03:50,272 continued drumbeat of absolute cutting edge breakthrough innovation to the point 62 00:03:50,272 --> 00:03:54,634 that, you know, you and I have been in the tech industry for a while. 63 00:03:55,755 --> 00:03:59,737 It is very much very sort of moments of Lenin, you know, decades are happening in 64 00:03:59,737 --> 00:04:01,858 weeks, sometimes in the same day. 65 00:04:01,858 --> 00:04:06,801 I mean, you finished reading one tweet and something else enormous has happened while 66 00:04:06,801 --> 00:04:08,921 you're reading those 200 odd characters. 67 00:04:09,727 --> 00:04:13,212 But people in the tech world always like to say that this is the transformative 68 00:04:13,212 --> 00:04:13,672 time. 69 00:04:13,672 --> 00:04:14,814 This is the destructive moment. 70 00:04:14,814 --> 00:04:16,556 This is the inflection point. 71 00:04:17,137 --> 00:04:18,439 Are we really at one of those points? 72 00:04:18,439 --> 00:04:21,362 What is it that makes you think that now it's really true? 73 00:04:23,754 --> 00:04:26,315 We always say that because it's self-serving. 74 00:04:26,755 --> 00:04:32,658 And what's different, I think, this time is the, you know, let's look at a 75 00:04:32,658 --> 00:04:33,358 corporation. 76 00:04:33,358 --> 00:04:39,761 If you've got a, the C-suite, the C-suite are people who are likely in their mid-50s 77 00:04:39,761 --> 00:04:44,763 through to their early 60s, which means most of them have had careers using 78 00:04:44,763 --> 00:04:45,743 computers. 79 00:04:45,983 --> 00:04:48,725 And they've had computers on their desks and they know how to use Excel. 80 00:04:48,725 --> 00:04:51,005 They probably did some time doing PowerPoint. 81 00:04:51,854 --> 00:04:58,740 And so they're much more computationally savvy than C-level execs, even when the 82 00:04:58,740 --> 00:05:00,341 iPhone came out. 83 00:05:00,482 --> 00:05:05,486 And they've had 20 or 30 years to digitize their businesses, which is what we've been 84 00:05:05,486 --> 00:05:07,648 telling them to do since forever. 85 00:05:07,648 --> 00:05:13,793 We're out of that moment of J-curve that you see when you have to install a new, 86 00:05:13,793 --> 00:05:14,954 implementing technology. 87 00:05:14,954 --> 00:05:19,357 So what that means is that companies are ready to deploy this. 88 00:05:19,522 --> 00:05:21,823 But let's then talk about the technology. 89 00:05:21,823 --> 00:05:28,208 So what happened with large language models was we turned computers into that 90 00:05:28,208 --> 00:05:28,628 scene. 91 00:05:28,628 --> 00:05:33,011 I think it's Star Trek IV when they're trying to move the whales. 92 00:05:33,011 --> 00:05:37,894 And Scotty picks up the mouse and says, computer, design me a da-da-da-da-da. 93 00:05:38,395 --> 00:05:39,475 We finally got there. 94 00:05:39,475 --> 00:05:43,778 We can talk language to computers, and computers talk language back to us. 95 00:05:43,778 --> 00:05:48,526 And since humans essentially have 96 00:05:48,526 --> 00:05:54,169 construct their entire world through the passing of messages in the form of 97 00:05:54,169 --> 00:06:01,393 language, that has allowed us to bring the capability of computers into those 98 00:06:01,393 --> 00:06:02,314 moments. 99 00:06:02,314 --> 00:06:09,418 And one of the things that is so beautiful about using something like ChatGPT is that 100 00:06:09,938 --> 00:06:13,720 in the old days, so November 2022, 101 00:06:15,834 --> 00:06:20,637 If you wanted to do a particular type of analysis, you had to configure that 102 00:06:20,637 --> 00:06:23,839 analysis into a way that a computer understood. 103 00:06:23,839 --> 00:06:26,621 So you took your mental model of the world. 104 00:06:26,901 --> 00:06:29,823 And that's why some people are good at programming and others are less good, 105 00:06:29,823 --> 00:06:34,326 because those people who can convert this sort of squishy human world into, you 106 00:06:34,326 --> 00:06:38,689 know, flawless logical code, do so and they do that translation. 107 00:06:38,689 --> 00:06:44,013 But now with ChatGPT, I can say, hey, listen, I kind of want you to do a MISI 108 00:06:44,013 --> 00:06:45,193 analysis. 109 00:06:45,382 --> 00:06:48,444 on this list of issues, do it for me. 110 00:06:48,444 --> 00:06:53,287 And it comes back and it presents as well as a grad student, which is, I think is 111 00:06:53,287 --> 00:06:54,447 pretty impressive. 112 00:06:54,728 --> 00:07:02,493 So I think those two things lend us to think that there are the ingredients for 113 00:07:02,493 --> 00:07:03,573 an inflection. 114 00:07:05,023 --> 00:07:10,290 So my main focus in this podcast is on responsibility and accountability for AI. 115 00:07:10,290 --> 00:07:15,378 Let me just first ask you, generally, how do you think we're doing in those 116 00:07:15,378 --> 00:07:16,218 dimensions? 117 00:07:17,086 --> 00:07:23,207 Can we mention that this is the recording after the Google Gemini fracas, which 118 00:07:23,207 --> 00:07:24,087 has... 119 00:07:24,588 --> 00:07:26,368 Okay, so we'll get to that. 120 00:07:26,948 --> 00:07:29,049 I think we are doing... 121 00:07:30,089 --> 00:07:36,531 The awareness of the issue is much, much higher than it was seven or eight years 122 00:07:36,531 --> 00:07:36,771 ago. 123 00:07:36,771 --> 00:07:42,853 I mean, I remember in 2015, writing about all these issues with Google images and ML 124 00:07:42,853 --> 00:07:45,313 systems that couldn't recognize dark skin. 125 00:07:47,106 --> 00:07:51,468 There were no teams, there was no discipline, there was not much academic 126 00:07:51,468 --> 00:07:52,388 study. 127 00:07:53,389 --> 00:08:00,053 We didn't do the kind of evaluations either a priori or ex ante, pardon me, or 128 00:08:00,053 --> 00:08:04,535 ex post of the systems. 129 00:08:04,615 --> 00:08:09,458 I think that has changed over the last seven or eight years. 130 00:08:09,878 --> 00:08:12,619 That's worth paying attention to. 131 00:08:14,100 --> 00:08:16,661 In 2014, the reason these things weren't... 132 00:08:16,738 --> 00:08:18,999 done was because people didn't know how to do them. 133 00:08:18,999 --> 00:08:22,061 No one really even thought to do them. 134 00:08:22,061 --> 00:08:23,622 And so the systems behaved the way they did. 135 00:08:23,622 --> 00:08:31,988 Now, after a lot of work by civil society, by academia, by activists, we do think 136 00:08:31,988 --> 00:08:33,349 about these things. 137 00:08:33,409 --> 00:08:38,833 And then the question is, of course, it's often embedded into sort of the belief 138 00:08:38,833 --> 00:08:41,615 systems and the narratives in the world at the right time, right? 139 00:08:41,615 --> 00:08:45,017 And just so, so I think that some of the things we've seen have been 140 00:08:45,366 --> 00:08:51,671 whatever the rung above the ladder, above the, the rung on the ladder, above the 141 00:08:51,671 --> 00:08:57,256 idea of performativity is, we're getting to that stage with some of this where 142 00:08:57,637 --> 00:09:03,903 people beat the responsibility drum really, really hard, perhaps harder than 143 00:09:03,903 --> 00:09:04,724 it needs to be beaten. 144 00:09:04,724 --> 00:09:08,908 I think Sundar Pichai at Google is somebody who does that a little bit more 145 00:09:08,908 --> 00:09:11,329 than perhaps he needs to. 146 00:09:12,063 --> 00:09:16,046 Yeah, so let's talk about this example with Google Gemini, where for those who 147 00:09:16,046 --> 00:09:20,650 aren't familiar, they built an image generator model and a large language 148 00:09:20,650 --> 00:09:21,451 model. 149 00:09:21,811 --> 00:09:26,396 And addressing exactly those issues you're talking about, concerns about bias, went 150 00:09:26,396 --> 00:09:26,896 too far. 151 00:09:26,896 --> 00:09:31,540 That in some cases, the image model would produce literally non-historically 152 00:09:31,540 --> 00:09:33,502 accurate pictures of people. 153 00:09:33,502 --> 00:09:37,305 The text model would just refuse to answer totally anodyne questions, saying they're 154 00:09:37,305 --> 00:09:38,046 too controversial. 155 00:09:38,046 --> 00:09:38,646 So. 156 00:09:38,667 --> 00:09:40,698 People have different reactions to this. 157 00:09:40,698 --> 00:09:43,770 What's your takeaway or main takeaways from that incident? 158 00:09:44,606 --> 00:09:48,468 Well, it's complicated. 159 00:09:48,468 --> 00:09:57,075 I mean, I think the absurdity of some of the outputs of Gemini can't be denied, 160 00:09:57,075 --> 00:09:57,335 right? 161 00:09:57,335 --> 00:10:05,081 So the idea that you might say, give me a picture of soldiers, German soldiers from 162 00:10:05,081 --> 00:10:09,003 World War II, and it doesn't come back with them looking at all how they should. 163 00:10:09,003 --> 00:10:12,205 Or one I saw on Twitter recently, which said, 164 00:10:13,442 --> 00:10:16,943 who caused more deaths, Elon Musk's tweets or Adolf Hitler. 165 00:10:16,943 --> 00:10:22,465 And Gemini sort of prevaricates and ums and ahs and says it's an ambiguous 166 00:10:22,465 --> 00:10:25,306 question, which is just sheer absurdity. 167 00:10:25,567 --> 00:10:28,048 So I think that it highlights two issues. 168 00:10:28,048 --> 00:10:36,091 The first issue is that large language models are very difficult to control. 169 00:10:36,091 --> 00:10:40,032 They are not like machines that we have built in the past. 170 00:10:40,032 --> 00:10:43,033 And so in the moment of 171 00:10:43,854 --> 00:10:49,755 trying to control them, you are choosing between helpful and harmless. 172 00:10:49,755 --> 00:10:54,056 And there seems to be right now on the methods that are used by these companies, 173 00:10:54,216 --> 00:10:56,117 really a trade off between the two. 174 00:10:56,117 --> 00:11:01,258 But it's a bit more complicated than that because you don't exactly know how much 175 00:11:01,258 --> 00:11:04,059 helpfulness you have to give up in order to make it harmless. 176 00:11:04,059 --> 00:11:08,220 And in a situation like that, I think you will always, I mean, use a British phrase 177 00:11:08,220 --> 00:11:09,361 here, over egg it, right? 178 00:11:09,361 --> 00:11:12,281 You'll always err on the side of caution. 179 00:11:13,226 --> 00:11:15,526 in order to avoid the harmful nature of it. 180 00:11:15,526 --> 00:11:18,607 And then you sort of end up slightly in this absurdity. 181 00:11:18,607 --> 00:11:23,349 But I think there's a second issue, which is that, you know, we have to recognize 182 00:11:23,349 --> 00:11:26,309 where we, that the world in which we currently live, not the one we want to 183 00:11:26,309 --> 00:11:31,271 live in, the world we live in is mediated by a small number of large technology 184 00:11:31,271 --> 00:11:32,251 companies. 185 00:11:32,411 --> 00:11:38,913 And that means that the AI they build will become our interface between us and the 186 00:11:38,913 --> 00:11:41,993 real world and the resources that we want to. 187 00:11:42,098 --> 00:11:43,198 access. 188 00:11:43,258 --> 00:11:47,719 And so somebody has gone off and stuck this resource in between me and the world. 189 00:11:47,719 --> 00:11:51,600 I mean, you have to go to Gemini today, but in two years time, everything you do 190 00:11:51,600 --> 00:11:53,601 on Google will be filtered through it. 191 00:11:54,341 --> 00:11:57,042 And they've gone off and made certain decisions. 192 00:11:57,042 --> 00:12:00,242 It's harmless, helpless trade-off, other decisions in particular about what they 193 00:12:00,242 --> 00:12:01,603 consider harm. 194 00:12:02,123 --> 00:12:05,644 And to me, this is a public infrastructure. 195 00:12:06,184 --> 00:12:10,125 This is a public infrastructure. 196 00:12:10,125 --> 00:12:11,665 It's a utility. 197 00:12:11,770 --> 00:12:14,290 Maybe it's a type of good, public good. 198 00:12:14,631 --> 00:12:19,933 And Google did talk a little bit about what went wrong, but not in the degree of 199 00:12:19,933 --> 00:12:26,615 sufficient transparency of this new role they have of guarding something that is 200 00:12:26,615 --> 00:12:28,096 not purely a private asset. 201 00:12:28,096 --> 00:12:31,857 So I think that there are these two issues that we have to have to talk about and 202 00:12:31,857 --> 00:12:33,037 make sense of. 203 00:12:33,587 --> 00:12:34,667 Okay, no, that's a good point. 204 00:12:34,667 --> 00:12:38,789 And that's, I want to talk about the issue about the power of the intermediaries and 205 00:12:38,789 --> 00:12:39,709 the big tech companies. 206 00:12:39,709 --> 00:12:44,191 But let me just first push you a little bit more on the question about the 207 00:12:44,191 --> 00:12:46,552 capability of controlling the models. 208 00:12:46,552 --> 00:12:49,373 Your book is called exponential age. 209 00:12:49,373 --> 00:12:53,034 And you talked about how technologies like AI and rootables are on these exponential 210 00:12:53,034 --> 00:12:54,055 growth curves. 211 00:12:55,996 --> 00:13:00,838 Is there a concern that the improvements in the technology are an exponential curve 212 00:13:00,838 --> 00:13:03,478 but our ability to control them? 213 00:13:04,515 --> 00:13:08,437 are something like arithmetic, and I'm not just talking about here in terms of the 214 00:13:08,437 --> 00:13:12,478 broad kind of alignment that the AI doesn't kill us all when it becomes super 215 00:13:12,478 --> 00:13:17,381 intelligent, but even on what seems like a fairly mundane problem, like not having 216 00:13:17,381 --> 00:13:19,022 biased image generation. 217 00:13:19,022 --> 00:13:21,542 Is that potentially an unsolvable problem? 218 00:13:22,396 --> 00:13:28,361 So I think that this isn't a case of the type that I describe in my book. 219 00:13:28,361 --> 00:13:34,026 I mean, the ones I don't describe in my book are really about how the state and 220 00:13:34,026 --> 00:13:38,209 the institutions of the state and the people who comprise that are able to 221 00:13:38,209 --> 00:13:40,070 respond to this pace of change. 222 00:13:40,131 --> 00:13:46,376 I think what we're seeing in Google Gemini is much, much closer to... 223 00:13:47,950 --> 00:13:53,374 what happened in the early years of jet powered passenger planes. 224 00:13:53,374 --> 00:13:55,816 So these emerged in the early 1950s. 225 00:13:56,136 --> 00:14:00,359 And the first of one of the first of those was made by a British company. 226 00:14:00,400 --> 00:14:03,162 We don't have names like this for jet aircraft companies anymore. 227 00:14:03,162 --> 00:14:04,003 De Havilland. 228 00:14:04,003 --> 00:14:04,984 It's so elegant. 229 00:14:04,984 --> 00:14:07,605 It's straight out of Downton Abbey. 230 00:14:07,866 --> 00:14:10,047 The De Havilland Comet. 231 00:14:10,528 --> 00:14:14,351 And I think something in the order of 40 of these crashed. 232 00:14:14,351 --> 00:14:17,422 I mean, they make Boeing look like, you know, an absolutely 233 00:14:17,422 --> 00:14:20,382 pristine whiter than whites safety record. 234 00:14:20,382 --> 00:14:25,344 And about 40 of these crashed, though 20 of them had fatalities and about 500 235 00:14:25,344 --> 00:14:29,865 people died when these planes kept falling out of the sky. 236 00:14:29,865 --> 00:14:37,647 And the reason was that they had square windows and the corners of the windows at 237 00:14:37,647 --> 00:14:42,728 jet speeds in the thinner atmosphere heated up much more, created lots more 238 00:14:42,728 --> 00:14:46,289 metal fatigue than they'd ever experienced in propeller planes that were a lower 239 00:14:46,289 --> 00:14:47,009 height. 240 00:14:47,150 --> 00:14:48,190 traveling more slowly. 241 00:14:48,190 --> 00:14:50,872 And they couldn't do, you know, they didn't do have computational fluid 242 00:14:50,872 --> 00:14:52,952 dynamics, right, to simulate all these things. 243 00:14:53,473 --> 00:14:55,814 And, and, and we figured that out. 244 00:14:55,814 --> 00:14:59,977 And yes, 500 people died and air travel is now so much safer. 245 00:14:59,977 --> 00:15:06,360 Now, I think that we are that's partly what we're dealing with within the world 246 00:15:06,360 --> 00:15:11,343 of the large language models, right, which is that these are these will be made safe 247 00:15:11,343 --> 00:15:13,644 in some way, they will become progressively safer. 248 00:15:13,644 --> 00:15:15,710 Now, not everyone agrees with me, there's a 249 00:15:15,710 --> 00:15:19,271 an academic called Gary Marcus, who I think would take a different view to that 250 00:15:19,271 --> 00:15:23,012 and say that's kind of really hard, if not impossible, to make them safe and they're 251 00:15:23,012 --> 00:15:24,152 poorly designed. 252 00:15:24,152 --> 00:15:29,373 But I think they can be made safe and we will learn how to do it. 253 00:15:29,373 --> 00:15:34,935 Now, the challenge is that Google is in a race. 254 00:15:34,935 --> 00:15:40,557 It's in a race that is both strategic, that in some sense is almost existential, 255 00:15:40,557 --> 00:15:44,117 that is commercial, that is also about... 256 00:15:44,494 --> 00:15:48,996 corporate and personal pride, which is driving them really fast. 257 00:15:48,996 --> 00:15:58,801 And we have constructed a world where we are looking for some hype to drink from as 258 00:15:58,801 --> 00:16:00,642 consumers of the media. 259 00:16:00,642 --> 00:16:06,806 And so this pushes this process of actually figuring out how to make these 260 00:16:06,806 --> 00:16:10,568 things safe out into the limelight much, much faster. 261 00:16:10,948 --> 00:16:13,782 And so in some ways, I'm a little bit sanguine about the fact that this... 262 00:16:13,782 --> 00:16:14,222 happened. 263 00:16:14,222 --> 00:16:15,143 I'm kind of annoyed. 264 00:16:15,143 --> 00:16:17,044 I feel it should have been caught in testing. 265 00:16:17,044 --> 00:16:22,649 But I think it is a case of it's this technology is, you know, it's seven years 266 00:16:22,649 --> 00:16:24,790 now since the first academic paper. 267 00:16:24,791 --> 00:16:27,873 It's, it's, it's really early. 268 00:16:27,873 --> 00:16:32,857 And to be honest, honestly, no harm was done, except for Google. 269 00:16:32,857 --> 00:16:34,358 Google is embarrassed. 270 00:16:34,619 --> 00:16:36,039 It shut the thing off. 271 00:16:36,100 --> 00:16:37,041 It'll figure out a fix. 272 00:16:37,041 --> 00:16:37,821 We won't see it. 273 00:16:37,821 --> 00:16:41,784 Is anyone really missing having an AI based image generator for an extra few 274 00:16:41,784 --> 00:16:42,385 days? 275 00:16:42,385 --> 00:16:43,445 None of us are. 276 00:16:45,151 --> 00:16:48,494 But won't that competitive dynamic you talk about continue? 277 00:16:48,494 --> 00:16:52,617 Because the company that has succeeded the most so far is OpenAI, partnered with 278 00:16:52,617 --> 00:16:56,381 Microsoft, which took the kind of Mark Zuckerberg, move fast and break things 279 00:16:56,381 --> 00:16:56,761 approach. 280 00:16:56,761 --> 00:17:01,585 While Google waited to deploy its LLM because it was concerned about all of 281 00:17:01,585 --> 00:17:05,529 these issues, if that's always going to be the case, isn't there going to be 282 00:17:05,529 --> 00:17:10,253 constantly commercial pressure to push the envelope and not even to take the time to 283 00:17:10,253 --> 00:17:12,015 develop the solutions that are possible? 284 00:17:13,838 --> 00:17:14,998 That's business. 285 00:17:14,998 --> 00:17:22,800 I mean, when we read the stories of what business people did over the last 100, 150 286 00:17:23,000 --> 00:17:30,402 years, I mean, they make trade-offs and they have to apply their judgment because 287 00:17:30,923 --> 00:17:32,923 nothing is perfect, no one has perfect resources. 288 00:17:32,923 --> 00:17:34,263 They're always competing needs. 289 00:17:34,263 --> 00:17:38,805 There's also always ego and selfishness and profit motive in it. 290 00:17:39,985 --> 00:17:42,005 It's just, it's happening in... 291 00:17:42,682 --> 00:17:49,086 a in a public space, which, of course, the business people behind them, and that's 292 00:17:49,086 --> 00:17:51,487 what they are, let's not call them founders, let's just call them business 293 00:17:51,487 --> 00:17:53,989 people, they're just like the green grocer down the road. 294 00:17:55,290 --> 00:18:02,955 They benefit from the publicity that it's that is being created by all of this, and 295 00:18:02,955 --> 00:18:06,417 they have to make difficult choices. 296 00:18:06,477 --> 00:18:07,057 I'm 297 00:18:07,310 --> 00:18:10,852 I would say, having spoken to him a few times, but not knowing him because he's 298 00:18:10,852 --> 00:18:18,837 very hard to read, Sam Altman, I would say that Sam is extremely, probably very keen 299 00:18:18,837 --> 00:18:25,442 to make all his products factually accurate and generally pretty safe, not as 300 00:18:25,442 --> 00:18:31,646 safe as a nervous Nick might want, but pretty safe. 301 00:18:31,706 --> 00:18:37,014 And he's just waiting for them to develop. 302 00:18:37,014 --> 00:18:38,034 those capabilities. 303 00:18:38,034 --> 00:18:42,918 And he at the same time, at that moment, he wants to go out and win market share. 304 00:18:42,918 --> 00:18:46,461 And what's happening is it's happening in this public domain, with all the noise by 305 00:18:46,461 --> 00:18:51,466 the way, Kevin of this idea that these LLMs will crawl out of our fridge, like if 306 00:18:51,466 --> 00:18:54,728 you've seen the film Gremlins, and you feed them and give them water, and they 307 00:18:54,728 --> 00:18:59,152 turn really nasty, you know, like that and take over our world, which again, I don't 308 00:18:59,152 --> 00:18:59,873 think they will. 309 00:18:59,873 --> 00:19:03,676 So I think we're in this funny stage where we should just kind of look at this and 310 00:19:03,676 --> 00:19:06,037 calm down a little bit and say, 311 00:19:07,682 --> 00:19:12,509 And in that calmness, it can still be a transformative technology. 312 00:19:12,509 --> 00:19:15,913 I mean, that can all be true at the same time. 313 00:19:17,123 --> 00:19:20,304 One of the important things about this wave of technology is that we have these 314 00:19:20,304 --> 00:19:23,505 big platforms that are building the AI systems. 315 00:19:23,505 --> 00:19:28,107 But every company on Earth probably is going to use them in one way or another 316 00:19:28,107 --> 00:19:32,029 with some amount of either their own technology or just taking things off the 317 00:19:32,029 --> 00:19:33,269 shelf and deploying them. 318 00:19:33,329 --> 00:19:36,991 You talk to executives of a lot of companies about AI. 319 00:19:37,751 --> 00:19:43,093 How are they thinking about these questions of responsibility and accuracy? 320 00:19:43,274 --> 00:19:45,375 And what does it take to put them 321 00:19:45,375 --> 00:19:47,602 persuade them to make the right trade-offs. 322 00:19:48,142 --> 00:19:51,903 I think in general, they take it very seriously. 323 00:19:52,063 --> 00:20:00,707 And it comes up surprisingly unprompted in the conversations that I have. 324 00:20:00,707 --> 00:20:03,609 And people talk about their responsible AI teams. 325 00:20:03,609 --> 00:20:07,070 Or they'll say, well, we have to roll this out responsibly. 326 00:20:07,070 --> 00:20:10,352 I mean, there are some variations. 327 00:20:10,412 --> 00:20:13,930 And for some, it's a. 328 00:20:13,930 --> 00:20:17,131 responsibility with a bit of a wave of a hand, say, don't talk to me about it with 329 00:20:17,131 --> 00:20:19,111 others, it's more, more engaged. 330 00:20:19,152 --> 00:20:22,893 But I think that comes back to a little bit earlier in our discussion, where I 331 00:20:22,893 --> 00:20:28,075 said that the way things have moved over the last seven or eight years, is that 332 00:20:28,075 --> 00:20:33,738 companies are taking this quite, quite seriously. 333 00:20:35,078 --> 00:20:41,901 And perhaps, you know, the challenge is that how much of that is, is simply just 334 00:20:41,901 --> 00:20:43,121 lip service. 335 00:20:43,426 --> 00:20:49,829 and how much of it is really, really deeply embedded in what companies try to 336 00:20:49,829 --> 00:20:50,630 do. 337 00:20:50,630 --> 00:20:57,254 And of course, the thing is, Kevin, how much do we want our companies to do? 338 00:20:57,254 --> 00:21:05,298 The founder of, and suddenly the name escapes me, the outdoor wear company, 339 00:21:05,439 --> 00:21:08,080 Patagonia, perhaps, is that? 340 00:21:08,080 --> 00:21:10,521 Who was sort of famous, he says, business will... 341 00:21:10,910 --> 00:21:12,971 you know, business will always leave a footprint, right? 342 00:21:12,971 --> 00:21:14,252 And we can't pretend it won't. 343 00:21:14,252 --> 00:21:17,413 And our job is to try to make it as small as we can. 344 00:21:17,814 --> 00:21:27,239 And I don't think it's necessarily a company's job to act responsibly or act 345 00:21:27,339 --> 00:21:28,380 ethically. 346 00:21:28,560 --> 00:21:34,464 That is the job of, I mean, it's the job of the law and whatever is delegated to 347 00:21:34,464 --> 00:21:38,105 regulators to say, these are the parameters within which you should act. 348 00:21:38,122 --> 00:21:41,343 Now you should act in that way and comply to this law. 349 00:21:41,784 --> 00:21:45,586 Otherwise we're going to come in and find those breaches and we are going to enforce 350 00:21:46,026 --> 00:21:47,427 actions against you. 351 00:21:47,427 --> 00:21:51,529 So yes, I'm excited that CEOs care about it and they're not spending their time 352 00:21:51,529 --> 00:21:56,172 trying to figure out how to kind of wind their way out of their obligations. 353 00:21:56,172 --> 00:22:00,094 But I think it's a bigger issue that those obligations are not clear. 354 00:22:00,094 --> 00:22:01,455 They're not clearly defined. 355 00:22:01,455 --> 00:22:06,377 And so that the companies really know what the rules of the game are. 356 00:22:07,047 --> 00:22:10,069 Yeah, that gets back to the point that you made earlier and you talk about in your 357 00:22:10,069 --> 00:22:10,889 book. 358 00:22:11,250 --> 00:22:16,574 Is there going to be enough capability in government to make those rules 359 00:22:16,574 --> 00:22:17,295 appropriate? 360 00:22:18,634 --> 00:22:22,836 Well, I mean, the government's getting better. 361 00:22:23,397 --> 00:22:26,580 It's definitely getting a lot better. 362 00:22:26,580 --> 00:22:29,241 And we, I think we tend to. 363 00:22:31,234 --> 00:22:36,256 have a slightly sneery view of government, but we can go back to the first consumer 364 00:22:36,256 --> 00:22:42,880 privacy legislation which came out in the late, early 1970s in California on the 365 00:22:42,880 --> 00:22:45,481 basis of the arrival of computer databases. 366 00:22:45,481 --> 00:22:47,943 I mean, that is phenomenally far-sighted. 367 00:22:47,943 --> 00:22:52,125 I mean, if that person had gone off and said, computer database is gonna be huge, 368 00:22:52,125 --> 00:22:57,448 I'm gonna build a company, they would be, you know, yachts are plenty by now. 369 00:22:57,808 --> 00:23:00,669 And in other areas, governments have been 370 00:23:01,882 --> 00:23:02,702 pretty impressive. 371 00:23:02,702 --> 00:23:04,463 I'll give you another example. 372 00:23:05,023 --> 00:23:06,584 My house was built in 1925. 373 00:23:08,365 --> 00:23:16,368 The cable that carries the electricity is rated at 100 amps and is still more than 374 00:23:16,368 --> 00:23:18,008 my house uses a century later. 375 00:23:18,008 --> 00:23:22,290 We've got two electric cars and we've got all the electronics you might imagine. 376 00:23:30,890 --> 00:23:34,192 genetic modifications in humans, right, was well ahead of the technology. 377 00:23:35,173 --> 00:23:42,359 But we have, governments, I think, started to slow down for historical reasons. 378 00:23:42,359 --> 00:23:48,643 And in my book, I talk about Reagan and the air traffic controllers and so on. 379 00:23:48,643 --> 00:23:52,786 But of course, that process of deregulation had started under Carter. 380 00:23:53,207 --> 00:23:56,758 And if you come at the back end of 25 or 30 years where 381 00:23:56,758 --> 00:24:00,059 The notion of good in the context, the intersection between government and 382 00:24:00,059 --> 00:24:06,683 business is about pulling regulations back, as opposed to recognizing that it is 383 00:24:06,683 --> 00:24:13,367 a presence of clear regulations, clear enforcement, a mandate for compliance that 384 00:24:13,367 --> 00:24:17,109 has allowed the pharmaceutical industry and the financial services industry to 385 00:24:17,109 --> 00:24:18,950 grow as big as they have. 386 00:24:18,950 --> 00:24:22,692 Of course, we're always going to feel that government is going to not have that 387 00:24:22,692 --> 00:24:23,653 capability. 388 00:24:23,653 --> 00:24:25,710 Now, I do think that they have really... 389 00:24:25,710 --> 00:24:32,096 moved and I can speak about the UK, I mean the quality of people in policy, that is 390 00:24:32,096 --> 00:24:38,241 the civil service who will brief government on these issues is really deep. 391 00:24:38,241 --> 00:24:43,066 Now I just don't know if there are enough of them, but they're getting really quite 392 00:24:43,066 --> 00:24:46,548 good and I wouldn't sort of besmirch them in any way. 393 00:24:47,230 --> 00:24:51,432 Our government on the other hand, maybe that's for an offline discussion. 394 00:24:51,432 --> 00:24:52,025 enough. 395 00:24:52,025 --> 00:24:57,955 But with regard to AI, different governments are taking different 396 00:24:57,955 --> 00:24:58,559 approaches. 397 00:24:58,559 --> 00:25:00,371 And so you mentioned the UK. 398 00:25:00,371 --> 00:25:04,939 It seems that actually this is one set of issues where Brexit has actually made a 399 00:25:04,939 --> 00:25:05,644 pretty dramatic difference. 400 00:25:05,644 --> 00:25:09,007 On a lot of things, there's not a great deal of substantive difference between 401 00:25:09,007 --> 00:25:11,779 what's happening, you know, the laws in the UK and Europe. 402 00:25:11,779 --> 00:25:17,989 But the European Union has adopted is in the final stages of this very extensive, 403 00:25:17,989 --> 00:25:19,951 very prescriptive AI act. 404 00:25:20,527 --> 00:25:24,589 UK government, at least from my perspective, more like the US, has taken 405 00:25:24,589 --> 00:25:28,891 more of an approach of saying, let's look at what existing laws out there, let's try 406 00:25:28,891 --> 00:25:35,394 and figure out and have different kinds of rules and how to balance innovation with 407 00:25:35,394 --> 00:25:36,235 regulation. 408 00:25:36,235 --> 00:25:37,656 Seems like a very different approach. 409 00:25:37,656 --> 00:25:39,757 So first of all, I guess, do you see that? 410 00:25:39,757 --> 00:25:43,878 And then what do you think is the best approach for governments to take? 411 00:25:44,190 --> 00:25:46,611 I definitely do see that. 412 00:25:46,611 --> 00:25:54,274 And I certainly see that gap emerge. 413 00:25:54,274 --> 00:25:58,956 And there was something else that was about the sense of victory that was coming 414 00:25:58,956 --> 00:26:02,017 out of the various parts of the EU. 415 00:26:02,578 --> 00:26:08,900 And we haven't created a $100 billion company since SAP, and now Novo Nordisk, 416 00:26:08,900 --> 00:26:11,650 which has got rich on Americans. 417 00:26:11,650 --> 00:26:13,010 having an obesity problem, right? 418 00:26:13,010 --> 00:26:17,751 I mean, because they make semaglutide, the GLP-1 agonist. 419 00:26:17,851 --> 00:26:23,073 And so it's a really odd thing to celebrate. 420 00:26:25,273 --> 00:26:28,154 The other thing is it's a really big act. 421 00:26:28,154 --> 00:26:37,877 And so I think that means that in a way, some parts of it may have been hurried, 422 00:26:37,877 --> 00:26:40,317 especially the bits that relate to the... 423 00:26:41,266 --> 00:26:44,107 large language models and those uses. 424 00:26:44,648 --> 00:26:49,992 But in reality, that has to now go through processes which will detail how things 425 00:26:49,992 --> 00:26:52,873 will get enforced. 426 00:26:53,614 --> 00:26:57,196 So we know that really bad laws can be created. 427 00:26:57,196 --> 00:27:03,281 I think, for example, many people probably view the Patriot Act hastily put in place 428 00:27:03,281 --> 00:27:07,984 in the US after 9-11 as something that has resulted in some overreach and might need 429 00:27:07,984 --> 00:27:10,625 some pairing back. 430 00:27:11,190 --> 00:27:21,076 But I think that the issue that I have with the EU Act is less about the 431 00:27:21,076 --> 00:27:27,861 specifics of the detail, but it was more about the posture of the participants 432 00:27:28,341 --> 00:27:35,987 towards this remarkable technology and what was really embedded within that, 433 00:27:35,987 --> 00:27:40,049 which was, I think, a bit of a distrust with American firms. 434 00:27:40,642 --> 00:27:46,004 but there wasn't within it a sense of, we'll create an alternative. 435 00:27:46,465 --> 00:27:49,707 And so then the question is what actually happens over this period of time. 436 00:27:49,707 --> 00:27:58,471 So I think in these moments of really, really rapid change, you need to have 437 00:27:58,711 --> 00:28:04,515 principles-based approaches to regulation, right, rather than hard and fast rules. 438 00:28:04,515 --> 00:28:09,537 So I would say that that's probably more a British style than... 439 00:28:09,950 --> 00:28:16,592 you know, an American approach and you need that because the underlyings change 440 00:28:18,193 --> 00:28:18,973 so frequently. 441 00:28:18,973 --> 00:28:21,234 So I think that feels like it's sensible. 442 00:28:21,234 --> 00:28:26,016 But the other thing that I think will end up happening is that there will be an 443 00:28:26,016 --> 00:28:37,021 increasing harmonization of the ways in which nations look at certainly economic 444 00:28:37,021 --> 00:28:39,461 and industrial applications of 445 00:28:39,774 --> 00:28:40,755 of AI. 446 00:28:40,755 --> 00:28:45,538 And the reason I think that they'll start to harmonize will be because we're in a 447 00:28:45,538 --> 00:28:49,141 search process of what method works best. 448 00:28:49,882 --> 00:28:56,087 Every country will benefit if technology companies can access their markets on 449 00:28:56,087 --> 00:28:57,148 equal terms. 450 00:28:57,148 --> 00:29:05,314 So that will also start to drive some degree of consistency across the piece. 451 00:29:05,815 --> 00:29:08,778 Where I think it may differ 452 00:29:08,778 --> 00:29:12,541 will be, and this is probably going to be quite a large part of AI to be honest, 453 00:29:12,761 --> 00:29:20,449 aspects that connect to culture, identity, or potentially national security issues, 454 00:29:20,449 --> 00:29:26,574 which was also the first sticking point, of course, with the internet, where 455 00:29:26,775 --> 00:29:30,137 countries sort of started to cleave themselves away. 456 00:29:31,415 --> 00:29:35,296 But I think what we saw with privacy was that there was a lot of harmonization. 457 00:29:35,296 --> 00:29:39,877 So basically every global company is complying with the European GDPR either 458 00:29:39,877 --> 00:29:43,678 because they have operations in Europe or because there's just no point in building 459 00:29:43,678 --> 00:29:45,538 two completely different systems. 460 00:29:45,778 --> 00:29:50,980 But I don't think anyone, even the people in the EU, in the European Commission, 461 00:29:50,980 --> 00:29:53,580 would say things are great on privacy. 462 00:29:53,721 --> 00:29:59,322 The practices have tremendously improved because we have this agreement on this 463 00:29:59,322 --> 00:30:01,182 comprehensive set of standards. 464 00:30:01,617 --> 00:30:02,083 Mm. 465 00:30:02,083 --> 00:30:08,914 How do we get to a point where the actual substance improves, which, as you noted, 466 00:30:08,914 --> 00:30:12,900 companies, they want to hear government telling them what to do, but they have a 467 00:30:12,900 --> 00:30:16,966 role in helping governments understand what the capabilities are and what could 468 00:30:16,966 --> 00:30:17,626 happen. 469 00:30:18,026 --> 00:30:24,129 You know, I think a really great example of that was what happened with Apple and 470 00:30:24,129 --> 00:30:25,310 the App Store. 471 00:30:25,310 --> 00:30:34,735 So under the Digital Services Act, no, the DMA, the Digital Markets Act regulation, 472 00:30:34,735 --> 00:30:40,418 because Apple operates this App Store that is sort of so big and so dominant, it has 473 00:30:40,418 --> 00:30:44,801 to provide alternative means of payment and alternative App Stores. 474 00:30:44,801 --> 00:30:46,346 So if you're in the EU and you have an... 475 00:30:46,346 --> 00:30:49,407 and I felt this is going to be all to American listeners, you can kind of get 476 00:30:49,407 --> 00:30:52,549 apps from a different store run by someone else and you don't have to pay through 477 00:30:52,549 --> 00:30:53,409 Apple. 478 00:30:54,310 --> 00:30:57,952 And of course, Apple has gone off and complied with this and involved something 479 00:30:57,952 --> 00:31:02,875 like writing 500 new APIs, which are kind of quite complex ways in which you use 480 00:31:02,875 --> 00:31:05,536 computer services over the internet. 481 00:31:06,297 --> 00:31:11,720 And they introduced new pricing mechanisms because what they've done is they've 482 00:31:11,720 --> 00:31:14,061 absolutely complied to the letter of the law. 483 00:31:14,770 --> 00:31:17,792 as you'd expect because they can afford good lawyers, and people are arguing they 484 00:31:17,792 --> 00:31:20,213 haven't complied to the spirit of the law. 485 00:31:20,213 --> 00:31:23,855 And this is the old problem of the sorcerer's apprentice, right? 486 00:31:23,855 --> 00:31:28,537 The Goethe, the old Goethe short story, right? 487 00:31:28,537 --> 00:31:33,400 Where we've got this technology, we try to control it, we give it rules, and it 488 00:31:33,400 --> 00:31:34,881 follows into the letter. 489 00:31:35,481 --> 00:31:43,705 And so I think that raises issues about when and how is the right time to... 490 00:31:44,702 --> 00:31:51,804 to write these rules and how do you get people sitting around the table at the 491 00:31:51,804 --> 00:31:52,784 same time? 492 00:31:52,784 --> 00:32:01,506 Now, I was actually quite a supporter of the Digital Markets Act because my framing 493 00:32:01,506 --> 00:32:08,928 is that there is a tendency towards dominance in digital markets for reasons 494 00:32:08,928 --> 00:32:13,509 that I initially probably Kevin learnt from you 25 years ago. 495 00:32:14,562 --> 00:32:18,184 And that force of gravity continues. 496 00:32:19,005 --> 00:32:25,289 And because there's this tendency to dominance, you will have concentration of 497 00:32:25,289 --> 00:32:25,689 power. 498 00:32:25,689 --> 00:32:31,833 So you need to find mechanisms to be more onerous with bigger firms than smaller 499 00:32:31,833 --> 00:32:33,354 ones, because the bigger ones are so much bigger. 500 00:32:33,354 --> 00:32:36,896 So I was at a high level, I was a supporter of it. 501 00:32:37,136 --> 00:32:39,757 I think that the... 502 00:32:40,882 --> 00:32:44,924 issue is, and there's another analyst called Benedict Evans, who I think 503 00:32:44,924 --> 00:32:50,388 probably made a better call on this than I did, where he said, well, the problem is 504 00:32:50,388 --> 00:32:57,513 that even if Apple is dominant, the process is safe and it's easy on the App 505 00:32:57,513 --> 00:32:58,073 Store. 506 00:32:58,073 --> 00:32:59,454 Consumers love it. 507 00:32:59,454 --> 00:33:04,578 No consumer is going to go off the App Store to the dodgy flea market with men in 508 00:33:04,578 --> 00:33:08,801 gray coats of saying, want to buy an app, when they can safely buy an app from the 509 00:33:08,801 --> 00:33:09,701 App Store. 510 00:33:10,334 --> 00:33:13,956 I think has a lot of momentum behind it as an idea. 511 00:33:13,956 --> 00:33:23,000 And that I think brings to bear the fundamental issues of capability within 512 00:33:23,301 --> 00:33:28,944 government and who did you have the right sorts of people who really were able to 513 00:33:28,944 --> 00:33:33,546 have the right kind of dialogue to get around the problems of corporate capture, 514 00:33:33,546 --> 00:33:37,648 which is what we saw with GDPR and have that conversation. 515 00:33:37,648 --> 00:33:39,309 And I don't think we did. 516 00:33:39,309 --> 00:33:39,929 So. 517 00:33:39,958 --> 00:33:49,145 So, I mean, that's a very long answer to say, it's quite hard. 518 00:33:49,285 --> 00:33:57,191 I'm not sure it's necessarily gonna serve European customers particularly well. 519 00:33:57,452 --> 00:34:01,515 I think what's hidden slightly below this is that these are all American companies. 520 00:34:01,515 --> 00:34:05,098 I mean, the rules are not written to discriminate against American companies, 521 00:34:05,098 --> 00:34:09,821 but it just so happens they're all American firms and will... 522 00:34:10,071 --> 00:34:13,964 continue to be if what we do is celebrate passing laws rather than building 523 00:34:13,964 --> 00:34:14,865 companies. 524 00:34:15,963 --> 00:34:17,684 One more question. 525 00:34:18,345 --> 00:34:21,848 You're pretty optimistic, it sounds like, both about the path of the technology, the 526 00:34:21,848 --> 00:34:24,790 business option, and working through these regulatory issues. 527 00:34:25,211 --> 00:34:26,292 What do you worry about? 528 00:34:26,292 --> 00:34:31,156 Where might you be wrong about any of these aspects that you think about? 529 00:34:31,156 --> 00:34:34,138 That's something that might put us on a bad path. 530 00:34:36,686 --> 00:34:41,189 You know that there are really interesting things that have happened that I wouldn't 531 00:34:41,189 --> 00:34:47,513 have predicted, like that a trained scientist running a major nation would 532 00:34:47,513 --> 00:34:51,034 shut down the nuclear program, as Angela Merkel did. 533 00:34:52,896 --> 00:34:54,317 I could run that tape a hundred times. 534 00:34:54,317 --> 00:34:56,898 I would always say she's not going to do it, and she did. 535 00:34:57,739 --> 00:35:02,781 So I guess I get... 536 00:35:04,630 --> 00:35:14,552 slightly concerned about the growing dominance of these firms. 537 00:35:14,552 --> 00:35:15,593 I mean, less so Nvidia. 538 00:35:15,593 --> 00:35:18,974 I mean, they're going off and making a lot of money because they're so far away from 539 00:35:18,974 --> 00:35:20,034 the customer. 540 00:35:20,174 --> 00:35:26,816 But the power and the capability that these companies will increasingly have, 541 00:35:26,936 --> 00:35:31,157 and then what that means in terms of their ability to 542 00:35:32,834 --> 00:35:41,416 control what's left of the public square and to do so in ways that aren't really in 543 00:35:41,416 --> 00:35:43,557 the interests of the people who live there. 544 00:35:43,557 --> 00:35:47,718 And I think that we've seen some aspects of that play out with the way targeted 545 00:35:47,718 --> 00:35:52,779 advertising works in the US and Facebook over not necessarily the last couple of 546 00:35:52,779 --> 00:35:55,140 years, but the years before that. 547 00:35:55,800 --> 00:35:58,561 I'm always a very simple man. 548 00:35:58,881 --> 00:36:01,421 And so I simply go back to... 549 00:36:01,946 --> 00:36:10,148 ideas of concentration of power, where, and wherever I see that processes and 550 00:36:10,148 --> 00:36:15,670 mechanisms that can reinforce concentration of power, it makes me 551 00:36:15,670 --> 00:36:21,731 nervous because I can't believe that if you are that powerful, you aren't somehow 552 00:36:22,191 --> 00:36:28,353 skewing the rules or taking a kind of shifty vig on the side that we don't know 553 00:36:28,353 --> 00:36:29,093 about. 554 00:36:29,654 --> 00:36:38,358 So where I could be wrong is that we don't put in things that slow down this 555 00:36:38,859 --> 00:36:40,199 agglomeration. 556 00:36:40,580 --> 00:36:44,082 And the other workplace I could be really wrong is whether it's even possible to 557 00:36:44,082 --> 00:36:45,122 slow it down. 558 00:36:45,122 --> 00:36:48,304 So some days I wake up and I think, I'm just not sure it's possible to slow it 559 00:36:48,304 --> 00:36:48,524 down. 560 00:36:48,524 --> 00:36:51,706 It doesn't matter that Google and Apple are a bit scared of Lena Khan at the 561 00:36:51,706 --> 00:36:53,527 moment and aren't buying companies. 562 00:36:53,827 --> 00:36:57,029 I don't think it's gonna slow down how quickly they grow. 563 00:36:58,206 --> 00:37:01,528 And I could also be wrong about that. 564 00:37:02,470 --> 00:37:09,716 What I'm not worried about so much is I'm not worried about AI accelerating and 565 00:37:11,097 --> 00:37:11,978 getting out of control. 566 00:37:11,978 --> 00:37:17,843 And the reason I'm not is because I think we technologically co-evolve systems to 567 00:37:17,843 --> 00:37:22,467 keep it safe, even if we can't co-evolve the regulatory systems as quickly. 568 00:37:23,848 --> 00:37:25,429 But yeah, but that's really... 569 00:37:25,558 --> 00:37:27,280 That's really about it. 570 00:37:27,280 --> 00:37:31,105 And of course, Kevin, I'm worried about the elections in the US and elsewhere. 571 00:37:31,105 --> 00:37:35,030 And I'm worried about, you know, what might happen with China and what might 572 00:37:35,030 --> 00:37:35,731 happen with Russia. 573 00:37:35,731 --> 00:37:39,335 I mean, those things are just background noise for 2024, really. 574 00:37:39,335 --> 00:37:41,557 They're the track to which we're going to be dancing. 575 00:37:42,543 --> 00:37:43,043 Absolutely. 576 00:37:43,043 --> 00:37:46,627 No, a lot out there to worry about a lot more that we could certainly talk about, 577 00:37:46,627 --> 00:37:48,648 but we're out of time at this point. 578 00:37:48,648 --> 00:37:50,510 Azim Azhar, thank you so much. 579 00:37:50,570 --> 00:37:53,393 Always great to hear from you and appreciate your sharing your perspectives 580 00:37:53,393 --> 00:37:55,434 with the road to accountable AI. 581 00:37:55,810 --> 00:37:56,815 That's been wonderful, Kevin. 582 00:37:56,815 --> 00:37:57,898 Thank you so much.